id
stringlengths 27
33
| source
stringclasses 1
value | format
stringclasses 1
value | text
stringlengths 13
1.81M
|
|---|---|---|---|
warning/0506/hep-ph0506278.html | ar5iv | text | # 1 Introduction
## 1 Introduction
A precise knowledge of parton distribution functions (PDF) in the proton is going to be one of the cornerstones of the physics analysis of LHC data , as well as a key ingredient for studies at other high energy accelerators. In hadron collisions, in fact, all high-$`p_{}`$ final states are produced through the hard scattering of partons, thus both potential new physics signals and Standard Model backgrounds are affected in shape and normalization by parton distributions.
PDF’s are currently determined by several groups through fits to different sets of high-energy data. All current fits are performed at least at next-to-leading order (NLO) in perturbative QCD, while the remarkable recent calculation of the three-loop Altarelli–Parisi splitting functions has made it possible to perform consistent next-to-next-to-leading order (NNLO) fits, by restricting the data set to cross sections for which the theoretical calculation has been performed to that order.
It is well known, on the other hand, that finite-order QCD calculations are limited in their range of applicability by the occurrence of large logarithms near the boundaries of phase space, both at large and at small values of $`x`$. These logarithms must be resummed, in order to enlarge the region in which perturbation theory can be trusted. Large-$`x`$ logarithms, in particular, are known to be related to soft and collinear gluon emission, and their resummation (threshold resummation) is well understood , and applied to a wide range of hard QCD processes (see, for example, ).
In this paper we shall address the possibility of including the effect of threshold resummations in parton fits, and we shall gauge the impact of these effects on large-$`x`$ quark distributions, by performing a simple analysis of Deep Inelastic Scattering (DIS) data.
We believe that including resummations would be useful in several respects. From a phenomenological viewpoint, making use of resummed predictions would allow for the inclusion of more large-$`x`$ data points in parton fits. In DIS, for example, data corresponding to values of $`W^2=Q^2(1x)/x`$ smaller than about $`15`$ GeV<sup>2</sup> are typically excluded from the fits , since they cannot be accounted for by making use of NLO perturbative results. Including resummations should lower this bound considerably . Resummations are also known to reduce the theoretical uncertainty of QCD predictions , which would correspondingly decrease one of the sources of error for PDF’s. Finally, it should be emphasized that, although soft-gluon resummations modify hard cross sections only near partonic threshold, they can affect parton distributions at smaller values of $`x`$ through sum rules as well as evolution.
Taking a more formal viewpoint, including resummation effects would bring about significant progress in the process of giving a precise definition of a fitted leading twist PDF. Resummations, in fact, are inevitably entangled with power corrections, which become increasingly important near the edges of phase space. This, however, should not be understood as an extra source of ambiguity: on the contrary, the inclusion of resummations highlights an inherent ambiguity, which is always present when finite-order perturbative predictions are used to extract from data the values of operator matrix elements. In general, it is not consistent to attribute a fixed twist to quantities evaluated at finite perturbative orders in a mass-independent regularization scheme, such as dimensional regularization. In such a scheme, one must first give a precise definition of the perturbative contribution to all orders, which entails a definition of power-suppressed contributions. Such a definition can only be given when all-order contributions have been computed, at least in the region of phase space where power corrections are expected to become dominant. This somewhat formal issue could become practical when a sufficiently precise comparison between PDF’s obtained by fitting data and PDF’s obtained from the lattice becomes possible.
Finally, it should be noted that resummed predictions exist for most of the cross sections used in global PDF fits, although with a varying degree of accuracy. The gold-plated process remains inclusive DIS, where, remarkably, we now have a full next-to-next-to-next-to-leading order (NNNLO) QCD prediction , as well as next-to-next-to-leading logarithmic (NNLL) soft-gluon resummation , and refined QCD-motivated models of the leading power corrections ; even a class of non-logarithmic terms has been shown to exponentiate . The Drell–Yan cross section is understood with almost the same degree of accuracy , with the added feature of a recent NNLO computation of the vector boson rapidity distribution . This is interesting because it has recently been shown that reasonably competitive parton fits can be obtained on the basis of these two processes only. If one wishes to rely upon a wider data set, next-to-leading logarithmic (NLL) resummed predictions exist also for the prompt-photon production cross section ; there, however, phenomenological problems remain, partly associated with a possible inconsistency of different data sets, and possibly related to the need to perform a more refined resummation and to include consistently power-suppressed corrections . Jet production in hadron collisions is more problematic: in fact, although the theoretical tools to perform NLL resummation have been available for some time and a phenomenological study has been performed in , it has recently been pointed out that, for most jet definitions, jet cross sections are plagued by nonglobal logarithms , starting at NLL level. Pushing the accuracy of soft-gluon resummation beyond leading logarithms (LL) for these cross sections will thus require more work.
It seems fair to conclude that enough resummation technology exists to perform a resummed global PDF fit. Including only DIS and Drell-Yan data, such a fit could actually be consistently performed at NNLO/NNLL level; in order to further constrain combinations of partons, which are hard to determine using these data only, one might then decide to trade some logarithmic accuracy in exchange for more data coverage.
In order to assess the impact that the inclusion of resummations might have on parton distributions, and more specifically on large-$`x`$ quark distributions, in the following we shall perform a fit of large-$`x`$ DIS data, using both NLO and NLL-resummed coefficient functions. It should not be regarded as an attempt to a global fit (see, e.g., Ref. for an analysis of large-$`x`$ PDF’s in the context of a global fit), since we shall clearly be forced to make several approximations in order to extract partons from such a comparatively small data set. Rather, it should be seen as a toy model of a resummed fit, providing a rough quantitative assessment of the impact of resummations. We find that soft-gluon effects typically suppress quark distributions by amounts ranging from a few percent to about $`1520\%`$ at large but not extreme values of $`x`$, $`0.55x0.75`$, for moderate $`Q^2`$. Sum rules also force a compensating enhancement in the distribution at smaller values of $`x`$, which, however, cannot be reliably determined within our current approximations. These effects would indeed warrant a more detailed investigation, if the current goal for PDF-related uncertainties (a few percent) were to be enforced also at these relatively large values of $`x`$.
## 2 Data and parametrizations
Large-$`x`$ DIS data come predominantly from fixed-target experiments. In order to have at our disposal different linear combinations of large-$`x`$ partons, we shall consider here charged-current (CC) data from neutrino-iron DIS, collected by the NuTeV collaboration , and neutral-current (NC) data from muon scattering from the NMC and BCDMS collaborations.
For our purposes, it will be sufficient to examine data at fixed values of $`Q^2`$, which we shall pick not too small so as to minimize the impact of power corrections, which are enhanced at the boundaries of phase space. We also require good data coverage for all the three experiments considered. We shall use $`Q^2=31.62\mathrm{GeV}^2`$ and $`Q^2=12.59\mathrm{GeV}^2`$, which corresponds to a cut in $`W^2`$ between $`4`$ and $`5\mathrm{GeV}^2`$, given the measured values of $`x`$. We shall check at the end that our results at the two selected values of $`Q^2`$ are compatible with NLO perturbative evolution. Since threshold resummation naturally takes place in Mellin moment space, our procedure will be to construct parametrizations of the data at the chosen values of $`Q^2`$, compute Mellin moments of the parametrizations, and then use them to extract moments of the corresponding PDF’s, with and without resummation. The difference between resummed and unresummed moments of PDF’s is per se a useful and solid result, since any QCD analysis can in principle be reformulated in Mellin space. In any case, we will also provide a simple $`x`$-space parametrization in order to illustrate the impact of the results in a more conventional manner. Studies of DIS structure functions in moment space were also performed in , by making use of data from the CLAS detector at Jefferson Laboratory; the corresponding values of $`Q^2`$ are however too small for a perturbative study like the present one.
Let us now turn to the NMC, BCDMS and NuTeV data sets we are considering. An efficient and convenient parametrization of NMC and BCDMS data for the NC structure function $`F_2`$, for proton, deuteron, and separately for the nonsinglet combination, has been provided in Ref. , and was recently upgraded for protons with the inclusion of HERA data in Ref. . The parametrization was constructed by first generating a large set of Monte Carlo copies of the original data, including all information on errors and correlations; subsequently, a neural network was trained on each copy of the data, yielding a set of parametrizations which, taken together, give a faithful and unbiased representation of the probability distribution in the space of structure functions.
In principle, the neural parametrization can be used for any values of $`x`$ and $`Q^2`$. In practice, errors will become increasingly large when one moves away from the region of the data. We use values of $`Q^2`$ which are well inside the measured region, with data coverage up to $`x=0.75`$. Specifically, we will be interested in the nonsinglet structure function $`F_2^{\mathrm{ns}}(x,Q^2)`$, which is unaffected by the gluon contribution and provides a combination of quark distributions, essentially $`ud`$, which is linearly independent from the ones sampled by NuTeV data. The neural parametrization of $`F_2^{\mathrm{ns}}(x,Q^2)`$ was previously used in conjunction with the technique of truncated Mellin moments for a determination of $`\alpha _s`$, which is unaffected by parametrization biases.
To illustrate the quality of the data, we show in Fig. 1 the nonsinglet structure function $`F_2^{\mathrm{ns}}(x,Q^2)`$, computed with the neural parametrization at our chosen values of $`Q^2`$, and for $`x=n/40`$, $`n=1,\mathrm{},39`$. The central values are given by the averages of the results obtained with the one thousand neural networks of the NNPDF collaboration, and error bars are the corresponding standard deviations. Error bars are relatively large, because $`F_2^{\mathrm{ns}}(x,Q^2)`$ is the difference between proton and deuteron structure functions, which entails a loss of precision. Central values and errors for the moments are similarly obtained by computing the moments with each neural network, and then taking averages and standard deviations.
NuTeV provides data for the CC structure functions $`F_2`$ and $`F_3`$. Since data are taken on an iron target, they need to be rescaled to include nuclear corrections, which were computed in by fitting the ratio $`F_2^{Fe}/F_2^D`$. The required smearing factor is given by
$$N(x)=1.100.36x0.28\mathrm{exp}(21.94x)+2.77x^{14.41}.$$
(2.1)
We consider first the charged-current structure function $`F_3`$ and its parton content. One has
$$xF_3=\frac{1}{2}\left(xF_3^\nu +xF_3^{\overline{\nu }}\right)=x\left[\underset{q,q^{}}{}|V_{qq^{}}|^2\left(q\overline{q}\right)C_3^q\right],$$
(2.2)
where $`V_{qq^{}}`$ are the relevant Cabibbo–Kobayashi–Maskawa (CKM) matrix elements and $`C_3^q`$ is the appropriate coefficient function. We fit the data at our chosen values of $`Q^2`$ using the functional form
$$xF_3(x)=Cx^\rho (1x)^\sigma (1+kx).$$
(2.3)
Eq. (2.3) is quite similar to the functional form which is used as initial condition for parton densities in the global analyses . We checked the stability of our fit by modifying the last factor of Eq. (2.3) with the inclusion of further powers of $`x`$ or logarithmic terms in $`x`$. We find that the parametrization (2.3), with four tunable parameters, is reliable enough to reproduce the data with quite small errors on the best-fit parameters and reasonable values of the $`\chi ^2`$ per degree of freedom.
The best-fit values at $`Q^2=31.62`$ GeV<sup>2</sup> are $`C=0.103\pm 0.012`$, $`\rho =0.294\pm 0.034`$, $`\sigma =3.325\pm 0.089`$, $`k=42.972\pm 4.700`$, corresponding to $`\chi ^2/\mathrm{dof}=7.20/6`$. At $`Q^2=12.59`$ GeV<sup>2</sup> we find instead $`C=0.054\pm 0.005`$, $`\rho =0.245\pm 0.038`$, $`\sigma =3.374\pm 0.145`$, $`k=99.719\pm 0.247`$, corresponding to $`\chi ^2/\mathrm{dof}=2.06/6`$. The data and the best-fit curves at the relevant values of $`Q^2`$ are shown in Fig. 2.
The situation for the structure function $`F_2`$, extracted from charged-current data, is slightly more complicated since there is a singlet component, and thus gluon-initiated processes also contribute. Such processes are not logarithmically enhanced at large $`x`$, and in fact, in the region of interest for our purposes, the gluon contribution to the structure function is significantly suppressed. We will handle it by subtracting it from the data point by point, using a gluon distribution determined by a global fit. The parton content of the charged-current structure function $`F_2`$ is
$$F_2\frac{1}{2}\left(F_2^\nu +F_2^{\overline{\nu }}\right)=x\underset{q,q^{}}{}|V_{qq^{}}|^2\left[(q+\overline{q})C_2^q+gC_2^g\right]=F_2^q+F_2^g.$$
(2.4)
We will proceed by fitting only $`F_2^q`$ and computing the gluon-initiated contribution using the gluon distribution from the NLO set CTEQ6M . We have checked that our results are not affected by the specific choice of gluon density, by repeating the calculation with, e.g., the set MRST2001 . As above, we pick the parametrization
$$F_2^q(x)=F_2(x)F_2^g(x)=Ax^\alpha (1x)^\beta (1+bx).$$
(2.5)
When doing the fit, we assume that we can neglect correlations among data points, as well as the error on $`F_2^g`$ with respect to the error on $`F_2`$ quoted by NuTeV. At $`Q^2=31.62\mathrm{GeV}^2`$, the best fit values for the parameters in Eq. (2.5) are $`A=0.240\pm 0.002`$, $`\alpha =0.562\pm 0.020`$, $`\beta =3.211\pm 0.065`$, $`b=13.085\pm 0.767`$, with $`\chi ^2/\mathrm{dof}=9.99/6`$. At $`Q^2=12.59\mathrm{GeV}^2`$, on the other hand, we find $`A=0.038\pm 0.005`$, $`\alpha =0.816\pm 0.021`$, $`\beta =2.697\pm 0.050`$, $`b=66.804\pm 7.583`$, with $`\chi ^2/\mathrm{dof}=9.55/6`$. In Fig. 3 we plot the data points of $`F_2^q(x)`$ at the chosen values of $`Q^2`$, along with the curve given by Eq. (2.5), according to the central values of the best-fit parameters.
At this point we have at our disposal parametrized expressions, including errors and correlations for the parameters, within the stated approximations, for the structure functions $`xF_3`$, $`F_2^q`$ and $`F_2^{\mathrm{ns}}`$. We can thus compute moments for the specified values of $`Q^2`$, and extract the moments of the corresponding parton densities by dividing out the appropriate coefficient functions, with and without resummations.
## 3 A simple parton fit
Having subtracted the contribution of gluon-initiated processes from the charged current structure function $`F_2`$ in Eq. (2.4), the factorization
$$F_i(x,Q^2)=x_x^1\frac{d\xi }{\xi }q_i(\xi ,\mu _F^2)C_i(\frac{x}{\xi },\frac{Q^2}{\mu _F^2},\alpha _s(\mu _R^2))$$
(3.1)
applies to all the structure functions we shall be considering ($`F_i=\{F_2,xF_3,F_2^{\mathrm{ns}}\}`$). In all cases $`q_i`$ is a combination of (anti)quark distributions, while $`C_i`$ is the appropriate coefficient function.
The coefficient functions $`C_i`$ for quark-initiated DIS contain terms that become large when the Bjorken variable $`x`$ for the partonic process is close to $`x=1`$, which forces gluon radiation from the incoming quark to be soft or collinear. At $`𝒪(\alpha _s)`$, for example, the coefficient functions can be written in the form,
$$C_i^{\mathrm{NLO}}(x,\frac{Q^2}{\mu _F^2},\alpha _s(\mu _R^2))=\delta (1x)+\frac{\alpha _s(\mu _R^2)}{2\pi }H_i(x,\frac{Q^2}{\mu _F^2}).$$
(3.2)
Treating all quarks as massless, the part of $`H_i`$ which contains terms that are logarithmically enhanced as $`x1`$ reads:
$$H_{i,\mathrm{soft}}(x,\frac{Q^2}{\mu _F^2})=2C_F\left\{\left[\frac{\mathrm{ln}(1x)}{1x}\right]_++\frac{1}{(1x)_+}\left(\mathrm{ln}\frac{Q^2}{\mu _F^2}\frac{3}{4}\right)\right\}.$$
(3.3)
Taking a Mellin transform, the contributions proportional to $`\alpha _s[\mathrm{ln}(1x)/(1x)]_+`$ and to $`\alpha _s[1/(1x)]_+`$ correspond to double $`(\alpha _s\mathrm{ln}^2N)`$ and single $`(\alpha _s\mathrm{ln}N)`$ logarithms of the Mellin variable $`N`$. Retaining only terms that are singular at large $`N`$ one finds in fact
$$\widehat{H}_{i,\mathrm{soft}}(N,\frac{Q^2}{\mu _F^2})=2C_F\left\{\frac{1}{2}\mathrm{ln}^2N+\left[\gamma _E+\frac{3}{4}\mathrm{ln}\frac{Q^2}{\mu _F^2}\right]\mathrm{ln}N\right\}.$$
(3.4)
The resummation of soft-gluon effects, responsible for this singular behaviour of DIS structure functions, has been well understood for a long time : it results in exponentiation of all singular contributions to the Mellin moments of the coefficient functions $`C_i`$ at large values of the moment variable $`N`$. In the $`\overline{\mathrm{MS}}`$ factorization scheme, soft resummation was implemented in in the massless approximation, and in for heavy quark production. In the following, we shall consider values of $`Q^2`$ much larger than the relevant quark masses, so that we can safely apply the results in the massless approximation.
The pattern of exponentiation of logarithmic singularities is nontrivial: one finds that Mellin moments of the coefficient functions can be written as
$$\widehat{C}_i^{\mathrm{res}}(N,\frac{Q^2}{\mu _F^2},\alpha _s(\mu _R^2))=(N,\frac{Q^2}{\mu _F^2},\alpha _s(\mu _R^2))\mathrm{\Delta }(N,\frac{Q^2}{\mu _F^2},\alpha _s(\mu _R^2)),$$
(3.5)
where $``$ is a finite remainder, nonsingular as $`N\mathrm{}`$, while
$`\mathrm{ln}\mathrm{\Delta }(N,{\displaystyle \frac{Q^2}{\mu _F^2}},\alpha _s(\mu _R^2))`$ $`=`$ $`{\displaystyle _0^1}dx{\displaystyle \frac{x^{N1}1}{1x}}\{{\displaystyle _{\mu _F^2}^{(1x)Q^2}}{\displaystyle \frac{dk^2}{k^2}}A\left[\alpha _s(k^2)\right]`$ (3.6)
$`+`$ $`B\left[\alpha _s\left(Q^2(1x)\right)\right]\}.`$
In Eq. (3.6) the leading logarithms (LL), of the form $`\alpha _s^n\mathrm{ln}^{n+1}N`$, are generated at each order by the function $`A`$. Next-to-leading logarithms (NLL), on the other hand, of the form $`\alpha _s^n\mathrm{ln}^nN`$, require the knowledge of the function $`B`$. In general, resumming $`\mathrm{N}^k\mathrm{LL}`$ to all orders requires the knowledge of the function $`A`$ to $`k+1`$ loops, and of the function $`B`$ to $`k`$ loops. In the following, we will adopt the common standard of NLL resummation, therefore we need the expansions
$$A(\alpha _s)=\underset{n=1}{\overset{\mathrm{}}{}}\left(\frac{\alpha _s}{\pi }\right)^nA^{(n)};B(\alpha _s)=\underset{n=1}{\overset{\mathrm{}}{}}\left(\frac{\alpha _s}{\pi }\right)^nB^{(n)}$$
(3.7)
to second order for $`A`$ and to first order for $`B`$. The relevant coefficients are
$`A^{(1)}`$ $`=`$ $`C_F,`$
$`A^{(2)}`$ $`=`$ $`{\displaystyle \frac{1}{2}}C_F\left[C_A\left({\displaystyle \frac{67}{18}}{\displaystyle \frac{\pi ^2}{6}}\right){\displaystyle \frac{5}{9}}n_f\right],`$ (3.8)
$`B^{(1)}`$ $`=`$ $`{\displaystyle \frac{3}{4}}C_F.`$
Notice that in Eq. (3.6) the term containing the function $`A(\alpha _s)`$ resums the contributions of gluons that are both soft and collinear, and in fact the anomalous dimension $`A`$ can be extracted order by order from the residue of the singularity of the nonsinglet splitting function as $`x1`$. The function $`B`$, on the other hand, is related to collinear emission from the final state current jet. In the case of heavy quarks, the function $`B(\alpha _s)`$ needs to be replaced by a different function, called $`S(\alpha _s)`$ in , which is instead characteristic of processes with massive quarks, and includes effects of large-angle soft radiation.
Turning to our fit, we observe that, upon taking Mellin moments, the convolution in Eq. (3.1) turns into a product, and it becomes straightforward to extract moments of the parton combinations $`q_i(x,Q^2)`$ at NLO, or with NLL resummation. Setting $`\mu _F=\mu _R=Q`$, one simply finds
$$\widehat{q}_i^{\mathrm{NLO}}(N,Q^2)=\frac{\widehat{F}_i(N1,Q^2)}{\widehat{C}_i^{\mathrm{NLO}}(N,1,\alpha _s(Q^2))};\widehat{q}_i^{\mathrm{res}}(N,Q^2)=\frac{\widehat{F}_i(N1,Q^2)}{\widehat{C}_i^{\mathrm{res}}(N,1,\alpha _s(Q^2))},$$
(3.9)
where the resummed coefficient function has been suitably matched to NLO, in order to avoid double counting of logarithmic contributions.
Since we are considering only three measurements, we need to introduce further approximations in order to be able to extract individual parton distributions. We will use isospin symmetry of the sea, so that $`\overline{u}=\overline{d}`$ and $`s=\overline{s}`$; further, we shall take the charm quark distribution to vanish and, for simplicity, we will impose a simple proportionality relation between antiquark distributions, $`\overline{s}=\kappa \overline{u}`$. In the fit shown below, we shall assume $`\kappa =1/2`$. All of these assumptions are essentially harmless at large $`x`$, where sea quarks are negligible: they allow us, however, to solve for the valence quark distributions $`u`$, $`d`$ and, say, $`s`$. The expressions for the parton combinations $`q_i`$ become particularly simple if one approximates the elements of the CKM matrix by neglecting terms of order $`(\mathrm{sin}\theta _C)^4`$ in $`|V_{qq^{}}|^2`$. Within the stated approximations, one finds then
$`q_2(x,Q^2)`$ $`=`$ $`u(x,Q^2)+d(x,Q^2)+3s(x,Q^2),`$
$`q_3(x,Q^2)`$ $`=`$ $`u(x,Q^2)+d(x,Q^2)s(x,Q^2)`$ (3.10)
$`q_2^{\mathrm{ns}}(x,Q^2)`$ $`=`$ $`{\displaystyle \frac{1}{6}}\left(u(x,Q^2)d(x,Q^2)\right),`$
which is easily inverted to give $`u`$, $`d`$ and $`s`$.
Having extracted the moments of the parton combinations in Eq. (3.10), one easily derives moments of individual quark distributions. In order to estimate the error on the moments, we proceed as follows. Regarding $`F_2^{\mathrm{ns}}`$, the neural parametrization is designed to allow for a simple calculation of the error on any functional of the data: one simply computes the standard deviation of the desired observable over the set of neural nets. A similar procedure would also yield correlations between different observables (in our case moments). In the present case, we shall neglect correlations between moments since we do not have a sufficiently reliable method to evaluate them in the case of NuTeV data. Concerning charged-current structure functions, we mimick the neural method by generating a Monte Carlo set of parametrizations of $`F_2`$ and $`F_3`$ at the chosen values of $`Q^2`$, assuming the parameters of our fits are gaussian distributed around their mean values with the stated errors. We then compute errors on moments as standard deviations over the Monte Carlo set.
Our results for the moments of the valence up quark distribution are shown in Figs. 45. The trend is clear, as could have been anticipated by comparing the resummed coefficient function with the NLO one: in the $`\overline{\mathrm{MS}}`$ scheme, resummation enhances the moments of the coefficient function, and thus suppresses the moments of quark distributions, with an effect increasing with the moment index $`N`$. The effect is unequivocal at both values of $`Q^2`$, since resummed and NLO moments differ by more than their error, beginning with $`N6`$. At $`Q^2=31.62`$ GeV<sup>2</sup> errors are somewhat smaller, however the effect of resummation is also slightly reduced, so that errors tend to have a somewhat larger overlap. In fact, as observed, e.g., in Ref. , LL and NLL terms in the Sudakov exponent are weighted by powers of $`\alpha _s`$ and the coupling constant is larger when evaluated at a lower scale. The trend is, in any case, still clearly visible, and the shift in the central values is of comparable size.
The down quark distribution is significantly smaller than the up quark distribution at large $`x`$, thus large-$`N`$ moments are suppressed. With our current, relatively small data set, the effect of resummation on the down quark distribution is completely overshadowed by statistical errors, so that the values of the moments are compatible with $`0`$ beginning at $`N10`$. Similarly, moments of the strange quark distribution cannot be reliably determined with these data. We shall then concentrate on the up quark distribution for the remainder of our analysis. One may just note in passing that resummation does not appear to shift even the central values of the moments of $`d(x,Q^2)`$: taken seriously, this would suggest that the bulk of the effect is carried by the largest valence distribution for the chosen hadron.
In order to provide a more conventional (and possibly a more practical) parametrization of the effect of resummations, we have also performed a fit of the moments presented in Figs. 45 to a simple $`x`$-space parametrization, choosing:
$$u(x)=Dx^\gamma \left(1x\right)^\delta .$$
(3.11)
The functional form (3.11) is sufficient to fit the moments of the up quark distribution with small errors on $`D`$, $`\gamma `$ and $`\delta `$, and a low $`\chi ^2/\mathrm{dof}`$. We have checked that the inclusion of other terms, for example a further factor linear in $`x`$, as in Eqs. (2.3) and (2.5) does not significantly improve the quality of the fit. The results using the fitting function (3.11) are shown in Figs. 67. The best-fit parameters at $`Q^2=12.59`$ GeV<sup>2</sup> are $`D=3.025\pm 0.534`$, $`\gamma =0.418\pm 0.101`$ and $`\delta =3.162\pm 0.116`$, with $`\chi ^2/\mathrm{dof}=1.62/11`$, for the NLO fit, while the resummed fit yields $`D=4.647\pm 0.881`$, $`\gamma =0.247\pm 0.109`$ and $`\delta =3.614\pm 0.128`$, with $`\chi ^2/\mathrm{dof}=0.64/11`$. At $`Q^2=31.62`$ GeV<sup>2</sup> we find $`D=2.865\pm 0.420`$, $`\gamma =0.463\pm 0.086`$, $`\delta =3.301\pm 0.098`$ and $`\chi ^2/\mathrm{dof}=1.10/11`$ for the NLO fit, as well as $`D=3.794\pm 0.583`$, $`\gamma =0.351\pm 0.090`$, $`\delta =3.598\pm 0.104`$ and $`\chi ^2/\mathrm{dof}=0.53/11`$ for the resummed fit<sup>*</sup><sup>*</sup>*We note that the values of $`\chi ^2/\mathrm{dof}`$ are very small. This might be due to the fact that we are neglecting correlations between moments.. The error bands in Figs. 67 correspond to a prediction at one-standard-deviation confidence level. They are obtained, as above, by generating a Monte Carlo sample of parametrizations of the stated form, assuming a gaussian distribution for the parameters $`D`$, $`\gamma `$ and $`\delta `$ with the stated errors; thus, they reflect only statistical errors and do not take into account biases due to the simple choice of functional form.
To display more clearly the quantitative effect of the resummation, we also present in Fig. 8 the central values for the normalized deviation of the resummed prediction from the NLO distribution, $`\mathrm{\Delta }u(x)=\left(u_{\mathrm{NLO}}(x)u_{\mathrm{res}}(x)\right)/u_{\mathrm{NLO}}(x)`$, at the two chosen values of $`Q^2`$.
A few comments concerning Figs. 68 are in order. First of all, an evident feature of the result is the change in sign of the effect around the point $`x=1/2`$. This is a stable feature of all our fits, and it can be traced back to the momentum sum rule, which is essentially unaffected by the resummation. Depletion of valence quarks at large $`x`$ is thus partly compensated by an increase at smaller values of $`x`$. The further change in sign at values of $`x`$ around 0.1, on the other hand, cannot be taken too seriously, since it happens in a region which is dominated by extrapolation within our current data set, so that errors are correspondingly very large. The impact of resummation is larger, as must be expected, at the lower value of $`Q^2`$.
At $`Q^2=12.59`$ GeV<sup>2</sup> and large $`x`$, it is to be expected that power corrections will play a role too. We have chosen, however, not to introduce them explicitly in our parametrization of $`F_i(x,Q^2)`$ since, as discussed in the introduction, their effect is inevitably tied to the precise treatment of the resummation. Disentangling resummations and power corrections is best left to a more precise quantitative analysis performed in the context of a global fit. We have, in any case, checked that target mass corrections do not significantly influence our results.
We have verified that our fits, which are performed independently at two fixed values of $`Q^2`$, are consistent with perturbative evolution. To this end, we have evolved the moments of the up quark distribution, starting at $`Q^2=31.62`$ GeV<sup>2</sup>, using the NLO Altarelli–Parisi anomalous dimension, down to $`Q^2=12.59`$ GeV<sup>2</sup>, and we have compared the results of the evolution with the direct fits of Figs. 45. The comparison is shown in Figs. 910. The results are fully compatible within errors, and we believe that the agreement could be further improved if one included power corrections, which are relevant especially at $`Q^2=12.59`$ GeV<sup>2</sup>. In the evolution, we have used $`\alpha _s(12.59\mathrm{GeV}^2)=0.2394`$, corresponding to $`\alpha _s(31.62\mathrm{GeV}^2)=0.2064`$, or to $`\mathrm{\Lambda }_{\mathrm{QCD}}^{(5)}=226`$ MeV , with appropriate matching at the $`b`$-quark mass threshold, set to $`m_b=`$ 4.5 GeV. Our choices are consistent with Ref. .
From a phenomenological point of view, we see that the impact of soft-gluon resummation on quark distributions can be sizeable, albeit only at values of $`x`$ which are quite large, say $`0.55<x<0.75`$. More precisely, as one can verify from Fig. 8, the suppression of the resummed up quark distribution with respect to the NLO one reaches $`5\%`$ at $`x0.58`$, $`10\%`$ at $`x0.65`$ and $`20\%`$ at $`x0.75`$ for $`Q^2=12.59`$ GeV<sup>2</sup>, while for $`Q^2=31.62`$ GeV<sup>2</sup> the same suppression factors are reached at $`x0.61`$, $`x0.69`$ and $`x0.8`$, respectively. Such values of $`x`$ can be relevant for several high energy processes, ranging from the production of high-mass Drell–Yan pairs, to high-$`E_T`$ jets, to the exchange of heavy resonances in the $`t`$-channel of hadron collisions. Considering, for example, the eccess of high-$`E_T`$ jets seen at the Tevatron Run I by CDF , the effect of including resummations in a PDF fit would have been to actually enhance the discrepancy between theory and experiment, since resummations suppress valence quarks at large $`x`$, and thus would have lowered the QCD prediction. Of course, a fully consistent treatment would have required making use of a resummed partonic cross section as well, which might have had a balancing effect. Interestingly, resummation may well be moving valence quarks from large to medium values of $`x`$, though the evidence for that in our present fit is at best qualitative. If that were the case, one might encounter several competing effects, depending on the partonic subprocess. For example, in Drell-Yan production at high mass and high rapidity, the heavy vector boson is produced by fusion of high-$`x`$ and low-$`x`$ partons, and one could have a depletion when the high-$`x`$ parton is a quark, or a slight enhancement (or no effect at all) when the high-$`x`$ parton is a gluon and the quark has lower momentum fraction. Finally, it should be noted that, in a fully consistent treatment, the depletion of the cross section due to resummed PDF’s may well compete against the enhancement of the hard partonic cross section which is commonly found when resumming logarithms to that accuracy, resulting in a reduced impact of soft gluons at hadron level.
## 4 Outlook
We have performed a first attempt to assess the impact of soft-gluon resummation on fits of parton distribution functions. We have argued that a global fit including soft-gluon effects is both feasible and desirable, from a theoretical as well as phenomenological point of view. That being said, it is useful to gauge the size of the effect that resummation might have. To that end, we have performed a simple fit of large-$`x`$ DIS data from the NuTeV, NMC and BCDMS collaborations. Our fit is not meant to be used as a complement or a substitute for a global fit: it is based on a small set of data, concentrated at large $`x`$, and does not consistently include all the constraints arising from sum rules and evolution which are properly taken into account in global fits. Our results should instead be seen as a first semiquantitative study of the impact of soft-gluon effects, and we believe that they might be an incentive for the dedicated collaborations performing PDF fits to include these effects in their algorithms. We have shown that, in the $`\overline{\mathrm{MS}}`$ scheme, the main effect of soft resummation is to suppress valence quark distributions at large $`x`$, by an amount ranging from a few percent to as much as 20$`\%`$, in the range $`0.55<x<0.75`$. This suppression may be partly compensated by a weaker enhancement, of the order of a few percent, at medium values of $`x`$, $`0.1<x<0.5`$, an effect which however falls largely inside our current statistical errors. It should be noted that a sizeable effect of this kind at large values of $`x`$ and moderate $`Q^2`$ will feed through to smaller values of $`x`$ at large $`Q^2`$ via evolution. We expect that including resummations should help to lower the theoretical uncertainty in the determination of PDF’s, and more in general in QCD cross sections, both by reducing scale uncertainties, and by allowing for stronger constraints on large-$`x`$ partons, thanks to the inclusion of more data points. We note also that the effect of resummations on PDF’s (a suppression) may turn out to be competing with the effect on partonic cross sections (in general, an enhancement). Disentangling such competing effects to gain a precise quantitative understanding of soft gluon effects on hadron-level cross sections can only be achieved with a consistent treatment of resummations, including their effects both in PDF global fits and in hard cross sections.
Acknowledgements
The authors would like to thank S. Forte, M.L. Mangano and M.H. Seymour for useful discussions, and L. Del Debbio, J. Rojo and A. Piccione for help with the neural network code of the NNPDF collaboration. We are also grateful to the NuTeV collaboration, and especially to D. Naples and M. Tzanov, for their assistance in the use of the NuTeV structure function data. L.M. wishes to thank CERN for hospitality and support during the early phase of this work, which is also supported in part by MIUR under contract $`2004021808\mathrm{\_}009`$. |
warning/0507/cond-mat0507416.html | ar5iv | text | # Two sub-band conductivity of Si quantum well
## I Introduction
Developement of silicon-on-insulator technology has enabled fabrication of silicon heterostructure devices where thin single crystalline Si film is sandwitched between amorphous SiO<sub>2</sub> layers. This kind of SiO<sub>2</sub>-Si-SiO<sub>2</sub> quantum well provides a unique material system where electron density can be tuned in a broad range due to high potential barriers formed by the SiO<sub>2</sub> layers. Previous work on SiO<sub>2</sub>-Si-SiO<sub>2</sub> quantum wells has mainly focused on single and bi-layer magneto transport Takashina et al. (2004); Prunnila et al. (2005a, b). In this work we focus on the issue how the two sub-band or bi-layer transport affects the low temperature conductivity of double gate SiO<sub>2</sub>-Si-SiO<sub>2</sub> quantum well with 14 nm thick Si layer.
## II Experimental
The samples were fabricated on commercially available (100) silicon-on-insulator wafer as described in detail in Ref. Prunnila et al. (2005b). The cross-sectional sample structure consist of n<sup>+</sup> Si top gate, top gate oxide (OX), Si well, back gate oxide (BOX) and n<sup>+</sup> Si back gate. The top gate is polycrystalline while the back gate is crystalline silicon. All results reported here were obtained from a 100 $`\mu `$m wide Hall bar sample with 400 $`\mu `$m voltage probe distance. The layer thickness for the Si well, top gate oxide and back gate oxide were $`t_\text{W}`$ = 14 nm, $`t_{\text{OX}}`$ = 40 nm, and $`t_{\text{BOX}}`$ = 83 nm, respectively. Figure 1 shows the schematic device cross-section together with self-consistent wave functions and effective potential at two total electron density ($`n`$) values.
In the experiments the sample was mounted to a sample holder of a He-3 cryostat, which was at base temperature (270 mK). The electron density was determined from the Shubnikov- de Haas (SdH) oscillations of the diagonal resistivity $`\rho _{\text{xx}}`$ utilizing the standard methods: in the single sub-band gate bias windows $`\rho _{\text{xx}}`$ was measured as a function of the gate voltages at constant magnetic field $`B`$ and $`n`$ was determined from the minimum positions of $`\rho _{\text{xx}}`$. In the presence of two sub-bands $`\rho _{\text{xx}}`$ was recorded as a function of $`B`$. Then a Fourier transform was performed numerically to $`\rho _{\text{xx}}(1/B)`$ and a peak position multiplied by $`e/h`$ ($`e`$ fundamental charge, $`h`$ Plank’s constant) in the spectrum gave the sheet density per degeneracy of a sub-band.
## III Results and discussion
### III.1 Sub-band densities at balanced gate bias
Left vertical axis of Fig. 2 shows the different electron densities as a function of top gate voltage $`V_{\text{TG}}`$ along the balanced (or symmetric) gate bias line where the back gate voltage $`V_{\text{BG}}=V_{\text{TG}}t_{\text{BOX}}/t_{\text{OX}}`$. This choice of gate biases produce symmetric potential in the Si well. The total density $`n`$ is obtained by summing the bonding sub-band $`n_\text{B}`$ and anti-bonding sub-band densities $`n_{\text{AB}}`$. On the balanced gate line the threshold for $`n_{\text{AB}}`$ is at $`V_{\text{TG}}0.8`$ V and $`n=n_\text{B}0.7\times 10^{16}`$ m<sup>-2</sup>. Note that the SdH method does not reveal the degeneracies of the system. We have made an assumption that both sub-bands arise from the high perpendicular mass valleys (non-primed valleys), which gives the total degeneracy of four after including the spin degeneracy. Our a priori assumption is validated by noting that $`V_{\text{TG}}`$ vs. $`n=n_\text{B}+n_{\text{AB}}`$ slope corresponds accurately to the total gate capacitance. The light perpendicular mass valleys (non-primed valleys) have a total degeneracy of eight and this degeneracy would lead to incorrect total density.
Right vertical axis of Fig. 2 shows the bonding anti-bonding energy gap $`\mathrm{\Delta }_{\text{BAB}}`$ which is obtained from $`n_\text{B}`$ and $`n_{\text{AB}}`$ by assuming ideal 2D density of states together with bulk effective mass $`m^{}=m_\text{t}=0.19m_0`$ parallel to the quantum well plane. The increased gate drive and $`n`$ leads to expected reduction of $`\mathrm{\Delta }_{\text{BAB}}`$ due to the potential barrier formation in the middle of the Si well as demonstrated by the self-consistent calculations in Figs. 1(b) and (c). The calculated $`\mathrm{\Delta }_{\text{BAB}}`$ is larger than the experimental one by $`0.72`$ meV (on average) in the density range of Fig. 2 (not shown). This discrepancy can be explained by noting that we have neglected the exchange and correlation effects in the calculations.
### III.2 Conductivity and mobility
Figure 3(a) shows (zero magnetic field) conductivity $`\sigma =1/\rho _{\text{xx}}`$ measured as a function of top and back gate voltages not . The scales of the voltage axes in Fig. 3(a) are chosen in such a manner that if we move perpendicularly to the balanced bias line the electron density stays (roughly) constant, i.e., we move along $`n=`$const. line and merely shift the position of the electron gas inside the Si slab. In the gate bias regions $`V_{\text{TG}}0`$ or $`V_{\text{BG}}0`$ only single sub-band is occupied and $`\sigma `$ behaves monotonically, which is expected on the basis of simple Coulomb - surface roughness scattering picture and is consistent with mobility measurements of sub-10 nm thick Si well Prunnila et al. (2005b).
The overall asymmetry of the conductivity with respect to the symmetric bias line arises from the different quality of the top (Si-OX) and back (Si-BOX) interfaces of the Si well. The back interface has substantially larger disorder in comparison to the top interface as can be observed from Figs. 3(b) and (c), which show the back interface and top interface mobilities $`\mu =\sigma /en`$, i.e., the mobilities along the axes of Fig. 3(a). Note that the $`n`$-axis of Figs. 3(b) and (c) ,together with $`n`$ plotted in Fig. 2 , also give an idea of the magnitude of electron density. The observed disorder difference is mainly due to larger surface roughness of the Si-BOX interface Prunnila et al. (2005b) consistent with the fact that the top-back mobility ratio actually increases as a function of carrier density. At low density surface roughness plays a minor role and the mobilities almost conincide, which is also partly due to electron wave function spreading throughout the Si well. The presence of the two Si-SiO<sub>2</sub> interfaces also reduce the peak mobility and shift it towards higher electron densities Prunnila et al. (2004).
In the bias window $`V_{\text{TG,BG}}0`$ the conductivity behaves strongly non-monotonically. We can observe, e.g., that close to threshold $`\sigma (V_{\text{TG}},V_{\text{BG}})|_{n=\text{const.}}`$ has a local minimum at symmetric bias. Note that at symmetric bias the second sub-band starts to populate at $`V_{\text{TG}}0.8`$ V (as was shown in the previous Sub-section) and the minimum is particularly strong when we are below this threshold. If we increase the gate biases the minimum at $`V_{\text{BG}}=V_{\text{TG}}t_{\text{BOX}}/t_{\text{OX}}`$ for $`\sigma (V_{\text{TG}},V_{\text{BG}})|_{n=\text{const.}}`$ is clearly splitted into two local minima, which follow closely the axes of the Figure. The minimum that follows $`V_{\text{TG}}`$-axis is illustrated more clearly in Fig. 4 which shows $`\sigma (V_{\text{BG}},V_{\text{TG}}`$=const.$`).`$ Note that if we would plot $`\sigma (V_{\text{TG}},V_{\text{BG}}`$=const.$`)`$ we would obtain a similar curve (only the magnitude of $`\sigma `$ would be different due to different quality of the Si-OX and Si-BOX interfaces). By comparing the high magnetic field $`\rho _{\text{xx}}`$ data of Ref. Prunnila et al. (2005b) and Fig. 3(a) we note that the local minimum in the vicinity of $`V_{\text{BG,TG}}0`$ for $`\sigma (V_{\text{BG,TG}},V_{\text{TG,BG}}=`$const.$`)`$ occurs at the position which is the threshold where $`\rho _{\text{xx}}`$ starts to show signatures of bi-layer transport.
The above non-monotonic behavior of $`\sigma `$ has a striking similarity to the substrate bias experiments of bulk Si inversion layers, where it was found that a positive substrate bias tends to reduce the mobility Fowler (1975). This effect has been addressed to spin flip scattering from singly populated localized band tail states of higher sub-bandsFeng et al. (1999) (, whose population increases with substrate bias in bulk devices). Scattering from localized band tail electrons is the most probable origin for the behavior of $`\sigma `$ in the vicinity of the $`V_{\text{TG(BG)}}0`$ at high $`V_{\text{BG(TG)}}`$ and around $`V_{\text{BG}}=V_{\text{TG}}t_{\text{BOX}}/t_{\text{OX}}`$ below the threshold of the second sub-band. In detail, e.g., for the data in Fig. 4: the conductivity saturates and starts to decrease because the population in the localized band tail of the second sub-band increases and these localized electrons scatter the electrons in the mobile first sub-band. Then at certain threshold the Fermi level passes through ”the mobility edge” of the second sub-band and the conductivity starts to increase due to presence of two sub-bands with extended states.
If we increase the electron density by gate bias beyond the threshold of the second sub-band the local minimum in $`\sigma (V_{\text{BG}},V_{\text{TG}})|_{n=\text{const.}}`$ at $`V_{\text{BG}}=V_{\text{TG}}t_{\text{BOX}}/t_{\text{OX}}`$ disappears due to presence of (solely) extended states. Then above $`V_{\text{TG}}2`$ V another minimum appears. This is indicated by the down arrows in Fig. 4. The origin of this feature is related to the resonance effect observed in tunneling coupled double quantum wells (often referred as resistance resonance) Palevski et al. (1990); Ohno et al. (1993); Berk et al. (1994). At high electron density and gate bias (at both gates) the double sub-band system in the Si well can be described equivalently as a weak or medium coupling bi-layer, which is a direct consequence of the barrier formation in the middle of the well \[see Figs. 1(b)&(c)\]. Far away from the balanced gate bias $`V_{\text{BG}}=V_{\text{TG}}t_{\text{BOX}}/t_{\text{OX}}`$ the barrier localizes the wave functions of the different layers or sub-bands to the different sides of the well. At balanced gate bias the sub-band wave functions delocalize across the well. This localization - delocalization is illustrated in Fig. 4 (right axis) where we show the average position $`z_a=z=\mathrm{\Psi }\left|z\right|\mathrm{\Psi }`$ ($`\mathrm{\Psi }=\mathrm{\Psi }_{B,AB}`$) and deviation (position uncertainty) $`\mathrm{\Delta }z=\sqrt{z^2z^2}`$ of the sub-bands. At balanced bias $`\mathrm{\Delta }z`$ has a maximum and $`z_a=0`$, which results in mobility reduction of the sub-band that otherwise is localized in the vicinity of the less disordered interface while the mobility of the other sub-band stays practically unaffected Ohno et al. (1993). This leads to the local minimum in $`\sigma `$ at $`V_{\text{BG}}=V_{\text{TG}}t_{\text{BOX}}/t_{\text{OX}}`$ at high electron densities. The detailed form of this minimum is strongly affected by the momentum relaxation time and quantum lifetime Berk et al. (1994) and further discussion will be published elsewhere.
## IV Summary
We have reported on two sub-band transport in 14 nm thick Si quantum well at 270 mK. The conductivity of the quantum well showed non-monotonic double gate bias dependency. At symmetric well potential and high density these were addressed to sub-band wave function delocalization in the quantization direction and to different disorder of the top and bottom interfaces of the Si well. In the gate bias regimes close to 2$`^{\text{nd}}`$ sub-band / bi-layer threshold the non-monotonic behavior was interpreted to arise from spin flip scattering from singly populated localized band tail states of higher sub-bands.
###### Acknowledgements.
M. Markkanen is thanked for assistance in the sample fabrication. Academy of Finland is acknowledged for financial support through project # 205467. |
warning/0507/hep-ph0507240.html | ar5iv | text | # Contents
## Chapter 1 Introduction
Atomic nuclei consist of nucleons (protons and neutrons) which are built of quarks and gluons. Quarks and gluons can also combine to form matter called mesons that are not commonly found on Earth, but are naturally created in processes that occur in outer space. Quarks and gluons interact with each other via the strong nuclear force to form hadrons such as nucleons and mesons. This strong force confines quarks and gluons inside hadrons. Exchange of light mesons, in particular $`\pi `$, $`\rho `$, $`\sigma `$, $`\omega `$, can be used to approximate the effective nucleon-nucleon interaction that binds protons and neutrons in atomic nuclei. In the so-called constituent quark model which describes matter to a good approximation, mesons are regarded as quark-antiquark pairs. In the same model, the proton, the neutron, and other baryons are triplets of quarks.
Recently, there has been evidence for a new kind of particles dubbed “exotic mesons”. They are called exotic because they have unusual quantum numbers, which are not allowed for quark-antiquark pairs. A theoretical description of the mechanism of exotic meson decays is vital to our understanding of nature of the quark-gluon interaction. A complete model which describes the behavior of exotic mesons should be based on the theory of relativity, in which one deals with velocities close to speed of light. In this work we will attempt to construct a relativistic model of hadronic decays of exotic mesons, in particular the so-called $`\pi _1`$ exotic meson. The main goal is to determine how much relativistic description of the $`\pi _1`$ decays differ from the existing nonrelativistic predictions.
This thesis is organized as follows. A brief review of QCD and the constituent quark model will be given in the following sections. In Chapter 2 we will introduce the exotic mesons and review the experimental situation. In Chapter 3 we will discuss the foundations of our model. In Chapter 4 we will present relativistic dynamics for noninteracting and interacting particles, and prepare the spinor framework for relativistic exotic meson decays. In Chapter 5 and 6, a relativistic construction of normal and exotic meson wave function will be given, with a particular emphasis on relativistic effects, including spin-orbit coupling and phase space modification. Decays of normal mesons and of the $`\pi _1`$ exotic meson will be analyzed in Chapter 7 and 8, respectively. In Chapter 9 we will explore the effects caused by residual interactions between the decay products, the so-called final state interactions. The final conclusions will be summarized in Chapter 10.
### 1.1 QCD and the constituent quark model
The discovery of the pion in 1947 helped to understand the nature of the nucleon-nucleon force. However many other mesons and baryons were found shortly after, which implied that none of these particles were elementary, and that pions were not the quanta of the strong interaction. In order to extend the $`N+\pi `$ scheme to other hadrons and to account for certain decay patterns such as a long lifetime of the $`\mathrm{\Sigma }^{}`$, M. Gell-Mann, T. Nakano and K. Nishijima independently proposed in 1953 the concept of strangeness. Strong decays with a short lifetime on the order of $`10^{24}10^{22}s`$ had the property of conserving strangeness, whereas much longer weak decays violated this conservation.
In 1961, Gell-Mann and independently Y. Ne’eman introduced the eightfold way, i.e., the SU(3) symmetry ordering of all subatomic particles analogous to the ordering of the chemical elements in the periodic table. It was a generalization of the SU(2) isospin symmetry of the nucleon into an SU(3) with strangeness as a second additive quantum number. A success of this theory was the discovery of the $`\mathrm{\Omega }^{}`$ baryon. The lightest mesons were also organized into a nonet.
As in the periodic table, a large number of hadrons suggested the existence of substructure. SU(3) symmetry and, in particular, its breaking led Gell-Mann and G. Zweig to postulate the quark in 1963 . They suggested that mesons and baryons are composites of quarks or antiquarks having three flavors: $`u`$, $`d`$ and the heavier $`s`$. Since fractional charges have never been observed, the introduction of quarks was treated more as a mathematical explanation of flavor patterns than as a postulate of an actual physical model. In 1965 O. W. Greenberg, M. Y. Han and Y. Nambu introduced the quark property of color charge in order to remedy a statistics problem in constructing the $`\mathrm{\Delta }^{++}`$ wave function. All observed hadrons had to be neutral singlets of the color SU(3) symmetry.
In 1968-69, an experiment at SLAC in which electrons were scattered off protons (deep inelastic scattering) led J. Bjorken and R. P. Feynman to realize that the data could be explained as evidence of small hard cores inside the proton called partons. This picture, however, had several problems, for example that $`50\%`$ of the proton momentum was not in quarks.
In 1973, a quantum field theory of the strong interaction was formulated by H. Fritzsch and Gell-Mann, based on the Yang-Mills color SU(3) nonabelian gauge symmetry which is different from the approximate flavor SU(3). In this theory, called quantum chromodynamics (QCD), quarks and massless gluons (quanta of the strong-interaction field) carry a color charge. The structure of QCD is similar to quantum electrodynamics, based on the U(1) symmetry, but much richer. Because gluons carry color charge they can interact with other gluons. The gauge-invariant Lagrangian, unlike that of QED, has cubic and quartic terms in the field potential leading to nonlinear classical equations of motion and interesting topological properties of the vacuum.
In 1973, D. Politzer, D. Gross and F. Wilczek discovered that QCD has a special property called asymptotic freedom, i.e., at short distances the coupling constant is small enough for perturbation theory to be valid. Unfortunately, at larger distances the coupling constant is on the order of 1 and the QCD Hamiltonian cannot be solved perturbatively. Quantum chromodynamics exhibits at this scale another distinct feature, quark color confinement, so that we may observe only colorless particles. At present we know six quark flavors: $`u`$, $`d`$, $`s`$, $`c`$, $`b`$ and $`t`$. Together with gluons they are included in the Standard Model of fundamental particles and interactions.
The QCD Lagrangian is in practice very difficult to solve. Thus, one is forced to deal with phenomenological models. One such model, the QCD sum-rule approach, was introduced in the late 1970’s and applied to describe mesonic properties . This technique was also extended to baryons . The basic idea of QCD sum rules is to match a QCD description of an appropriate momentum-space correlation function with a phenomenological one, and establish a correspondence between hadronic and quark degrees of freedom . This approach provides a connection between the QCD Lagrangian and hadron physics.
Meson and baryon spectra are well described in the constituent quark model (CQM). The CQM Hamiltonians written in the 1960’s contained only the kinetic terms and short distance spin-spin interaction. They quite successfully predicted the magnetic moments of the ground state baryons, as long as the magnetic moments of constituent quarks had their classical values. The use of a nonrelativistic model was justified for $`c\overline{c}`$ and $`b\overline{b}`$ mesons, but did not work very well for light mesons. More sophisticated Hamiltonians treated spin-dependent interactions nonperturbatively, and based them on QCD . It was possible to describe hadrons within a unified, relativized quark model with chromodynamics, in which the $`q\overline{q}`$ interaction is a sum of the Coulomb (one-gluon-exchange) potential and a linear confining term expected from QCD .
High energy hadron-hadron and hadron-nucleus scattering at small and intermediate momentum transfers are well described by assuming that mesons and baryons are bound states of two and three constituent quarks, respectively . Moreover, exclusive processes at small and intermediate momentum transfers agree well with the constituent quark model predictions of elastic and transition form factors. Therefore, the CQM approach provides a relevant description of nonperturbative QCD at low and intermediate momentum scales. Because it is hard to derive the constituent quark model from QCD, one may search for a relation between CQM and sum rules .
### 1.2 Mesons
The most convenient formulation of QCD is a constituent representation in which hadron states are dominated by a small number of constituents. It will be assumed that in this representation, interactions that change the number of particles as well as other relativistic effects are small. A natural choice is the Coulomb gauge because it operates with simple degrees of freedom in the nonrelativistic limit. This framework works especially in QED, where for example the hydrogen atom is very well described by the Coulomb potential. The Coulomb gauge will be discussed more in Chapter 3.
Each hadron is composed of quarks and may contain valence gluons. The simplest configuration of quarks that gives a color singlet is a quark-antiquark pair (meson). The next possibility for a colorless strongly interacting particle is a bound state of three quarks (baryon). One could construct more complicated configurations, for example the so-called pentaquarks, but until now, there has been no strong evidence for objects built of more than three quarks.
Mesons can be classified based on their quantum numbers $`J^{PC}`$. Here $`J`$ is total angular momentum of a particle, $`P`$ is its parity, and $`C`$ denotes charge conjugation. The total angular momentum is given by
$$𝐉=𝐋+𝐒,$$
(1.1)
where $`L`$ is relative orbital angular momentum of a quark-antiquark pair and $`S`$ denotes total intrinsic spin of this pair,
$$𝐒=𝐒_1+𝐒_2.$$
(1.2)
Because quarks are fermions with spin 1/2, the values of $`S`$ can be either 0 or 1. Thus, the values of $`J`$ are integer and mesons are bosons. The orbital angular momentum $`L`$ can take any integer positive value or zero, and this determines all possible meson configurations.
Parity, which determines how the sign of the wave function of a particle behaves under a spatial reflection, can be obtained from
$$P=(1)^{L+1},$$
(1.3)
whereas charge conjugation, describing the particle-antiparticle symmetry (well-defined only for neutral mesons composed of a quark and an antiquark of the same flavor), is given by
$$C=(1)^{L+S}.$$
(1.4)
In the case of two-boson systems, parity would be given by a slightly different formula, $`P=(1)^L`$.
The $`u`$ flavor is given the third component of isospin $`I_3=1/2`$, whereas the $`d`$ flavor has $`I_3=1/2`$. The concept of isospin came from the idea of treating the proton and the neutron as two states of one particle, the nucleon, having two values of the $`I_3`$, like fermions have two values of spin quantized along a fixed axis. Light unflavored mesons, i.e., mesons containing only flavors $`u`$ and $`d`$, have $`I_3`$ equal to either 0 or 1. For example, the pion with $`J^{PC}=0^+`$ and $`I=1`$ has three isospin components: $`\pi ^+`$, $`\pi ^0`$ and $`\pi ^{}`$ (isospin triplet), whereas for $`I=0`$ there is only one component: $`\eta `$ (singlet). The $`s`$ quark has isospin 0; thus strange mesons have isospin $`1/2`$. From three quark flavors we can build up nine mesons grouped into an octet and a singlet. In Table 1.2 we present the classification of light unflavored mesons with respect to the above quantum numbers. In Table 1.2 we show the flavor wave functions of the nine pseudoscalar ($`J^{PC}=0^+`$) and nine vector ($`J^{PC}=1^{}`$) mesons.
In reality however, we do not observe exact configurations corresponding to $`\eta _0`$ and $`\eta _8`$ but rather their linear combinations known as $`\eta `$ and $`\eta ^{}`$. The transformation matrix between both pairs must be orthogonal and thus has one parameter, the mixing angle $`\theta `$. A similar situation occurs for $`\omega _0`$ and $`\omega _8`$, but in this case the mixing angle is such that the observed particles are given by
$$\omega =\frac{1}{\sqrt{2}}(u\overline{u}+d\overline{d}),\varphi =s\overline{s}.$$
(1.5)
For a particular combination of $`L`$ and $`S`$ we have more than one flavor multiplet due to different radial quantum numbers. Normal pions and kaons have $`n=1`$, and are the lightest, whereas radially excited mesons with higher values of $`n`$ are heavier. For example, the $`\pi (1300)`$ is a candidate for a radially excited $`\pi `$, and the $`\rho (1450)`$ is a candidate for a radially excited $`\rho `$. In this work we will deal only with radial ground state mesons because their orbital wave functions should have the same size as the well-known pions and kaons. Knowledge of a meson’s structure is the first step towards understanding its dynamics, which is responsible for the spectrum of all observed mesons and their decay widths. In fact, all mesons that can decay strongly are not bound states but resonances, and we can study QCD by analyzing how they decay. The quark model description of such decays assumes $`q\overline{q}`$ pair creation in the gluonic field of the decaying meson. A phenomenological model based on quark-antiquark pair production from the vacuum is referred to as the $`{}_{}{}^{3}P_{0}^{}`$ model . Although this decay mechanism gives quite successful values for meson widths, it is not rigorously related to QCD which allows $`q\overline{q}`$ creation only from a gluon. The recent benchmark predictions for decay widths of light mesons are given in Refs. .
## Chapter 2 Exotic mesons
In the preceding chapter we presented classification of the low-lying unflavored mesons. There are, however, certain combinations of internal meson quantum numbers: spin $`J`$, parity $`P`$, and charge conjugation $`C`$, which are missing in this classification, such as $`0^{},\mathrm{\hspace{0.17em}0}^+,\mathrm{\hspace{0.17em}1}^+`$, or $`2^{}`$. These quantum numbers cannot be obtained from adding the quantum numbers of the quark and the antiquark alone. The corresponding mesons are referred to as the exotic mesons.
From lattice QCD and model calculations it follows that the lightest exotic mesons may be obtained by adding an extra constituent gluon with $`J^{PC}=1^{}`$ to a quark-antiquark system. Such $`q\overline{q}g`$ states are referred to as hybrid mesons. In this work we will focus on mesons with the $`J^{PC}=1^+`$ quantum numbers. The isovector multiplet with $`J^{PC}=1^+`$: $`\pi _1^+,\pi _1^{},\pi _1^0`$, is predicted to be the lightest exotic . One must emphasize however, that $`q\overline{q}g`$ states can also have nonexotic quantum numbers. The hybrid components of normal mesons may be important in the mechanism of meson decays. Nonexotic hybrid mesons will be discussed in Chapter 7.
Hadrons with excited gluonic degrees of freedom may supply new insight into quantum chromodynamics at low energies, where the gluon dynamics should be responsible for phenomena such as color confinement and dynamical symmetry breaking. Therefore, the discovery of exotic mesons is of a great importance. In this chapter we will briefly review the experimental situation in the search for the $`\pi _1`$, and present theoretical predictions for its mass and width.
### 2.1 Experimental situation
A resonance can be identified by analyzing the spectrum of its decay products. If a strong, narrow resonance is present, this dependence takes the form of a sharp peak, as shown in Fig. 2.1. In this picture presenting the BNL E852 data of the $`\eta \pi ^0`$ channel in the charge exchange reaction $`\pi ^{}p\eta \pi ^0n`$ , two resonances can be clearly seen. The corresponding particles are the $`a_0(980)`$ and the $`a_2(1320)`$. If a resonance is weakly produced, an amplitude analysis may be required which identifies the resonance by a phase motion of the amplitude as a function of the invariant mass of the decay products.
Using such amplitude analyses, several candidates for the $`\pi _1`$ have been recently reported. The $`\pi _1`$(1400) with mass $`M=1370\pm 16_{30}^{+50}`$ MeV and width $`\mathrm{\Gamma }=385\pm 40_{105}^{+65}`$ MeV, was reported by the E852 Collaboration in the $`\eta \pi ^{}`$ channel of the process $`\pi ^{}p\pi ^{}\eta p`$ . This state was confirmed by the Crystal Barrel Collaboration in the $`\pi \eta `$ channel in the reactions $`\overline{p}n\pi ^{}\pi ^0\eta `$ and $`\overline{p}p\pi ^0\pi ^0\eta `$ . In the $`\eta \pi ^0`$ channel, two resonances shown in Fig. 2.1 provide benchmarks for the amplitude analysis. A possible signal on the order of 1$`\%`$ of the dominant $`a_2(1320)`$ have been extracted in both $`\eta \pi ^0`$ and $`\eta \pi ^{}`$ final states.
A Breit-Wigner (BW) parametrization of the S-wave and D-wave corresponding to the $`a_0`$ and $`a_2`$ mesons in the $`\eta \pi ^0`$ and $`\eta \pi ^{}`$ channels is confirmed by the data, but the resonance interpretation of the P-wave is problematic. First, the left panel of Fig. 2.2 which represents the $`\eta \pi ^{}`$ spectrum shows that the signal for the $`\pi _1(1400)`$ is weak. Second, it is impossible to find a selfconsistent set of the BW parameters for the P-wave. As a result, its phase as a function of the invariant mass does not increase over $`90^{}`$ which is required for a resonance .
The E852 Collaboration has also reported two $`\pi _1(1600)`$ states. One of these has $`M=1597\pm 10_{10}^{+45}`$ MeV and $`\mathrm{\Gamma }=340\pm 40\pm 50`$ MeV, and decays into $`\eta ^{}\pi `$ . In this channel, two strong amplitudes are extracted corresponding to the $`a_2(1320)`$ and $`\pi _1(1600)`$, as shown in the right panel of Fig. 2.2. The exotic signal is here much stronger, as compared to those in the $`\eta \pi ^0`$ and $`\eta \pi ^{}`$ channels.
The other $`\pi _1(1600)`$ state with $`M=1593\pm 8_{47}^{+29}`$ MeV and $`\mathrm{\Gamma }=168\pm 20_{12}^{+150}`$ MeV, was reported in the $`\rho ^0\pi ^{}`$ channel . In this case, all expected well-known states: $`a_1(1260)`$, $`a_2(1320)`$, and $`\pi _2(1670)`$ are observed, as shown in Fig. 2.3 . In addition, the amplitude analysis shows that the amplitude with exotic numbers $`J^{PC}=1^+`$ has structure which is consistent with a resonance at 1.6 GeV decaying into $`\rho \pi `$. Evidence for the $`\pi _1(1600)`$ has also been reported by the VES collaboration in three channels, $`b_1\pi `$, $`\eta ^{}\pi `$ and $`\rho \pi `$ , with $`M=1.61(2)`$ GeV and $`\mathrm{\Gamma }=0.29(3)`$ GeV. The $`\pi _1(1600)`$ signals in all these channels are somewhat different from one another, and therefore further experiments are needed to clarify the nature of these signals.
Only the $`\pi _1(1600)`$ reported in the $`\rho \pi `$ channel has a width on the order of 100$``$200 MeV, i.e., comparable to other meson resonances. The broad structures in the $`\eta \pi `$ and $`\eta ^{}\pi `$ channels can be accounted for by low-energy rescattering effects . It is possible however, that the $`1^+`$ exotic meson in the $`\eta ^{}\pi `$ channel is the same as $`\pi _1(1600)`$ meson seen through its decay into $`\rho \pi `$. However, at this point this is only speculation .
### 2.2 Theoretical predictions
The mass of the $`\pi _1`$ can be obtained from calculations based on lattice QCD . They give values in the region $`1.82.0`$ GeV. Theoretical predictions for this mass are based on various models. The QCD sum-rule predictions vary widely between 1.5 and 2.5 GeV . The MIT bag model places this mass in the region $`1.31.8`$ GeV . According to the constituent gluon model, light exotics should have masses in the $`1.82.2`$ GeV range . The diquark cluster model predicts the $`\pi _1`$ state at 1.4 GeV . Finally, the flux tube model predicts the $`1^+`$ mass similar to the lattice results .
In Tables 2.1 and 2.2 we show $`\pi _1`$ width predictions calculated using various nonrelativistic models: IKP , CP and PSS . At $`\pi _1`$ mass equal to 1.6 GeV, the dominant modes are $`\pi b_1(1235)`$ and $`\pi f_1(1285)`$. For larger values of this mass, the above modes are still dominant, together with the $`K\overline{K}_1(1400)`$.
It should be noted that in each channel, one outgoing meson has orbital angular momentum $`L=0`$ (these mesons such as $`\pi `$ or $`K`$ are called S-mesons). The other has $`L=1`$ (these mesons such as $`b_1`$, $`f_1`$ or $`K_1`$ are called P-mesons). The above models favor modes that satisfy the so-called $`S+P`$ selection rule. It states that a hybrid meson prefers to decay into one S-meson and one P-meson. There is a chance that relativistic corrections could significantly change this situation and favor the $`\eta \pi `$, $`\eta ^{}\pi `$, and $`\rho \pi `$ modes. This work aims to explore that possibility.
Theoretical predictions indicate the importance of searching for the $`\pi _1`$ in the $`\rho \pi `$, $`b_1\pi `$, and $`f_1\pi `$ channels. They also suggest a search for the $`K\overline{K}_1`$ channel. In order to compare these predictions with experiment however, more knowledge of branching ratios is necessary.
## Chapter 3 Dynamical foundations
In the preceding chapter we described exotic mesons as quark-antiquark-gluon bound states. In order to proceed to their dynamics, we need to know how to obtain the exotic meson wave functions. In the nonrelativistic case, this can be done by using the Born-Oppenheimer approximation which works quite successfully for normal mesons treated as $`q\overline{q}`$ states. Furthermore, this procedure together with lattice simulations will provide some important information about the composition of the lightest hybrid meson.
The constituent gluon plays a central role in the structure of an exotic meson. As the photon in QED, the gluon needs to be described in a particular gauge. A natural framework for introducing the constituent quark model and providing insights into calculating meson decays is the Coulomb gauge, which is free of unphysical degrees of freedom and has a good nonrelativistic quantum-mechanical limit. Relativization will be accomplished by using a relativistic phase space and transforming quark-antiquark states under Lorentz boosts.
The model presented in this work is microscopic, i.e., at the level of quarks and gluons. However, mesons interact with each other via meson exchange. This force may contribute significantly to the dynamics of exotic mesons, and the quantitative analysis of this problem will be the subject of Chapter 9.
We will begin the present chapter with the Born-Oppenheimer approximation. Then we will review the Coulomb gauge for QCD and describe the Coulomb gauge picture of normal and hybrid mesons. Finally we will proceed to relativistic effects in the Coulomb-gauge constituent quark model.
### 3.1 Heavy quarkonia and the Born-Oppenheimer approximation
Lattice simulations are quite successful in predicting mass spectra for mesons and baryons. Hadronic decays, however, provide the real difficulty for such estimates. Thus we are left with phenomenological models of the coupling between mesons and hybrids. In general, there are two approaches for describing hadronic decays of hybrid mesons. The first regards a hybrid as a quark-antiquark state with an additional constituent gluon . Such a meson would decay through gluon dissociation into a $`q\overline{q}`$ pair . The second approach assumes that a hybrid is a quark-antiquark pair moving on an adiabatic surface generated by an excited gluonic flux-tube . In this case a hybrid meson would decay because of phenomenological pair production described by the $`{}_{}{}^{3}P_{0}^{}`$ model . Recently, an extended version of the flux tube model has been introduced .
If constituent quarks composing mesons are heavy, such systems (heavy quarkonia) can be studied using the Born-Oppenheimer approximation . In this approach it is assumed that formation of gluonic field distributions decouples from the dynamics of the slowly moving quarks, and therefore hadronic decays can be described within nonrelativistic quantum mechanics. This approximation can be justified for light quarks, because dynamical chiral symmetry breaking leads to massive consituent quarks.
In the Born-Oppenheimer model, a hybrid meson is treated analogously to a diatomic molecule in which the heavy quarks correspond to the nuclei and the gluon field corresponds to the electrons. Initially, a quark and an antiquark are treated as spatially fixed color sources and this determines the glue energy levels as a function of the $`q\overline{q}`$ separation. Each energy level defines an adiabatic potential $`V_{q\overline{q}}(r)`$. The quark motion is restored by solving the radial Schrödinger equation for each of these potentials.
The lowest static potential gives a normal meson spectrum, whereas the excited potentials lead to hybrid mesons. The static potentials are determined from lattice simulations. The gluonic configurations can be classified according to symmetries of the $`q\overline{q}`$ “molecule”. The strong interaction is invariant under rotations around the $`q\overline{q}`$ axis, a reflection in a plane containing the pair, and with respect to the product $`PC`$. Each configuration can be thus labeled by the corresponding eigenvalues, denoted by $`\mathrm{\Lambda }`$ (the magnitude of the projection of the total gluon angular momentum onto the molecular axis), $`Y=\pm 1`$ (the sign of this projection), and $`PC=\pm 1`$, respectively.
States with $`\mathrm{\Lambda }=0,1,2,\mathrm{}`$ are denoted by $`\mathrm{\Sigma },\mathrm{\Pi },\mathrm{\Delta },\mathrm{}`$, respectively. States which are even (odd) under the combined $`PC`$ operation are denoted by $`g`$ ($`u`$). Lattice simulations for the ground state configuration and the lowest gluonic excitations are shown in Fig. 3.1 . The parameter $`r_0`$ is on the order of 0.5 fm. In the ground state (normal meson) $`\mathrm{\Lambda }=0`$ ($`\mathrm{\Sigma }_g^+`$) and for the first excited state $`\mathrm{\Lambda }=1`$ ($`\mathrm{\Pi }_u`$).
If the gluon is in a relative S-wave with respect to a $`q\overline{q}`$ pair, it has $`PC=+1`$. Lattice results show, however, that the lowest excited configuration has the gluon with $`PC=1`$ so the gluon orbital angular momentum with respect to a $`q\overline{q}`$ pair must be odd. The simplest choice is $`L=1`$. Therefore, in Chapter 5, in order to construct the $`\pi _1`$ spin wave function we will couple a transverse gluon to the $`q\overline{q}`$ state with the $`\rho `$ quantum numbers, in a relative P-wave.
### 3.2 The Coulomb gauge
In quantum electrodynamics, the electromagnetic field arises naturally from demanding an invariance of the action under the local gauge U(1) transformation. If $`\varphi `$ is a complex scalar field then the corresponding Lagrangian
$$L=(_\mu \varphi )(^\mu \varphi )m^2\varphi ^{}\varphi $$
(3.1)
is invariant under the transformation
$$\varphi e^{i\mathrm{\Lambda }}\varphi ,\varphi ^{}e^{i\mathrm{\Lambda }}\varphi ^{},$$
(3.2)
where $`\mathrm{\Lambda }`$ is an arbitrary real constant. If $`\mathrm{\Lambda }`$ depends on the spacetime coordinates, however, then the derivative of the field does not transform covariantly, i.e., in the same way as $`\varphi `$. In order to remedy this problem one introduces the covariant derivative (like in the general theory of relativity)
$$D_\mu =_\mu +ieA_\mu ,$$
(3.3)
where $`e`$ is a real constant (the electric charge) and $`A_\mu `$ is the electromagnetic potential. This potential must transform according to
$$A_\mu A_\mu +\frac{1}{e}_\mu \mathrm{\Lambda }.$$
(3.4)
The quantity
$$F_{\mu \nu }=_\mu A_\nu _\nu A_\mu $$
(3.5)
is the electromagnetic field tensor and its six nonzero components correspond to the fields $`𝐄`$ and $`𝐁`$. The simplest Lagrangian built up from the gauge-invariant quantities is thus
$$L=\frac{1}{4}F_{\mu \nu }F^{\mu \nu },$$
(3.6)
and its variation with respect to $`A_\mu `$ leads to the Maxwell equations in vacuum. Because of the gauge invariance we may introduce one constraint on the components of the field potential. In the Coulomb gauge this constraint is given by
$$𝐀=0.$$
(3.7)
In quantum chromodynamics, the local gauge transformations form the SU(3) group
$$\varphi _iS_{ij}\varphi _j=(e^{\frac{i}{2}\lambda ^k\mathrm{\Lambda }^k})_{ij}\varphi _j,$$
(3.8)
where $`i,j=1,2,3`$ and $`k=\mathrm{1..8}`$. The matrices $`\lambda `$ are the hermitian and traceless generators of SU(3) (Gell-Mann matrices). In this case the expression for the covariant derivative is given by
$$D_\mu =_\mu +igA_\mu ^k\frac{\lambda ^k}{2},$$
(3.9)
whereas the potential transforms according to
$$A_\mu ^k\frac{\lambda ^k}{2}SA_\mu ^k\frac{\lambda ^k}{2}S^1\frac{i}{g}(_\mu S)S^1.$$
(3.10)
The gauge-invariant field tensor is given by
$$G_{\mu \nu }^a=_\mu A_\nu ^a_\nu A_\mu ^a+gf^{abc}A_\mu ^bA_\nu ^c,$$
(3.11)
where $`f_{abc}`$ are the SU(3) structure constants, and the simplest field Lagrangian is thus
$$L=\frac{1}{4}G_{\mu \nu }^aG_a^{\mu \nu }.$$
(3.12)
The chromoelectric field corresponds to the $`G_{0\alpha }`$ components of the field tensor
$$𝐄^a=\dot{𝐀}^aA^{0a}+gf^{abc}A^{0b}𝐀^c,$$
(3.13)
and satisfies the Gauss law
$$𝐄^a+gf^{abc}𝐀^b𝐄^c=g\rho _q^a.$$
(3.14)
Here $`\rho _q^a=\psi ^{}\frac{\lambda ^a}{2}\psi `$ is the quark color charge density. Introducing the covariant derivative in the adjoint representation
$$D_\mu ^{ab}=\delta ^{ab}_\mu +igT_{ab}^cA_\mu ^c,$$
(3.15)
where $`T_{ab}^c=if^{cab}`$, leads to
$$𝐃^{ab}𝐄^b=g\rho _q^a.$$
(3.16)
If $`𝐄_{}=\varphi `$ is the longitudinal part of of the chromoelectric field then we obtain
$$(𝐃^{ab})\varphi ^b=g\rho ^a,$$
(3.17)
where $`\rho =\rho _q+\rho _g`$ is a total color charge density. Here the transverse gluon color charge density is given by
$$\rho _g^a=f^{abc}𝐄_{}^b𝐀^c,$$
(3.18)
where $`𝐄_{}`$ is the transverse part of the chromoelectric field. Combining the above equations leads to:
$`\varphi ^a={\displaystyle \frac{1}{𝐃}}g\rho ^a,`$
$`A^{0a}={\displaystyle \frac{1}{𝐃}}(^2){\displaystyle \frac{1}{𝐃}}g\rho ^a.`$ (3.19)
The last equation results in the instantaneous nonabelian Coulomb interaction Hamiltonian
$$H_C=\frac{1}{2}d^3xd^3y\rho ^a(x)K_{ab}(x,y;A)\rho ^b(y),$$
(3.20)
where
$$K_{ab}(x,y;A)=x,a|\frac{g}{𝐃}(^2)\frac{g}{𝐃}|y,b.$$
(3.21)
After quantization, the field $`𝐄_{}`$ becomes the momentum conjugate to the vector potential. The confining nonabelian Coulomb potential will be represented in the diagrams below by the dashed lines. More details related to properties of the Coulomb-gauge QCD can be found in Ref. .
A simple phenomenological picture of hadrons and their decays in terms of quantum mechanical wave functions emerges naturally in a fixed gauge approach. In the Coulomb gauge, for example, the precursor of flux tube dynamics originates from the nonabelian Coulomb potential, which also determines the quark wave functions . The string couples to a $`q\overline{q}`$ pair via transverse gluon emission and absorption and such a coupling carries the $`{}_{}{}^{3}S_{1}^{}`$ quantum numbers.
In a description of decays based on Coulomb gauge quantization it is necessary to include the hybrid quark-antiquark-gluon configurations, since they appear as intermediate states in the decay of mesons. If such hybrid states also exist as asymptotic states, they would provide insight into the dynamics of confined gluons . Fig. 3.2 shows diagrams corresponding to strong decays of $`q\overline{q}g`$ hybrid mesons (top) and $`q\overline{q}`$ normal mesons (bottom). The gluons connecting the Coulomb lines represent formation of the flux tube, e.g. the gluon string in the ground state. The overall initial state is enclosed by the solid oval. In the lower diagram the hybrid meson state appears as an intermediate state in a normal meson decay, which is assumed to proceed via mixing of a $`q\overline{q}`$ pair with a virtual excitation of a gluonic string and its subsequent decay.
In the Coulomb gauge the quantum numbers of the gluonic states can be associated with those of a transverse gluon in the presence of the static $`q\overline{q}`$ source. This is because transverse gluons are dressed , and on average behave like the constituent particles with the effective mass $`m_g500\text{ MeV}`$ . Thus low-energy excited gluonic states are expected to have a small number of transverse gluons. The flux tube itself is expected to emerge from the strong coupling of transverse gluons to the Coulomb potential. The transverse gluon wave function can be obtained by diagonalizing the net quark-antiquark-gluon interactions shown in Fig. 3.3 (in addition to the gluon kinetic energy).
The above three-body interaction plays an essential role in the dynamics of hybrid mesons. A transverse gluon has a gradient coupling to the Coulomb potential. Thus the P-wave transverse gluon receives no energy shift from this coupling and the energy of the S-wave gluon state is increased. In the Coulomb gauge picture, the shift of the S-wave state via this three-body interaction may be the cause of the inversion of the S$``$P levels seen on the lattice. Using only two-body potentials between quarks and gluons leads to the S-wave gluon in the lowest energy excited state, which disagrees with lattice data .
### 3.3 Relativistic effects
Another and very important issue is the question of relativistic effects. Even though a simple nonrelativistic description appears to be quite successful in predicting decay widths of mesons as heavy as $`12`$GeV, the presence of light quarks raises the question of validity of this description. It has been shown that relativistic effects for hadronic form factors may be significant . It is possible that they are responsible for the discrepancy between experimental data and theoretical predictions. This work will try to estimate the size of relativistic effects applied to the $`\pi _1`$ decays.
In order to calculate relativistic decay amplitudes exactly in the Coulomb gauge, one needs to find the fundamental quantities (dynamical generators of the Poincaré group) in terms of the chromodynamical fields. This problem, however, is very difficult to state and in this work we will not solve the Coulomb gauge QCD Hamiltonian to obtain meson wave functions. Instead, we will use the general transformation properties under the remaining kinematical symmetries (rotations and translations) to construct the states.
The relativistic meson and hybrid spin wave functions will be elements of irreducible, unitary representations of the Poincaré group for noninteracting particles. The interaction between particles should enter the dynamics by finding a new mass operator (the Bakamjian-Thomas model). Because an exact form of the strong potential between quarks is unknown, we will employ instead a simple parametrization of the meson orbital wave function. This is clearly an approximation which cannot be avoided without solving dynamical equations for the boost generators .
## Chapter 4 Relativistic dynamics
In this chapter we will review the Lorentz group in vector and spinor representations, which is a framework for the ten fundamental quantities describing the dynamics of a system of noninteracting particles. Then we will introduce the Bakamjian-Thomas construction of these generators for interacting particles. Finally we will discuss a Wigner rotation, which plays an essential role in constructing relativistic and covariant spin wave functions for mesons and hybrids.
### 4.1 The Lorentz group
The principle of relativity in the framework of general relativity requires that physical laws must be invariant under all transformations of the coordinates. Gravitational fields are automatically included if one deals with curvilinear coordinates, however, they are important only for large-scale phenomena. Yet in the physics of elementary particles, the curvature of the spacetime is small and can be neglected. Therefore one needs to deal only with the metric tensor of a flat spacetime. In this case the principle of relativity requires that physical laws must be invariant under transformations from one inertial frame to another. Such transformations are called inhomogeneous Lorentz transformations, and the coordinates transform linearly according to
$$x^\mu =\mathrm{\Lambda }_\nu ^\mu x^\nu +a^\mu ,$$
(4.1)
where $`\mathrm{\Lambda }_\nu ^\mu `$ is the Lorentz matrix, and $`a^\mu `$ is a constant four-vector. From the invariance of the finite interval $`x^{}_{}{}^{}\mu x_\mu ^{^{}}=x^\mu x_\mu `$ for $`a^\mu =0`$, it follows that the Lorentz matrix must be orthogonal,
$$\mathrm{\Lambda }_\rho ^\mu \mathrm{\Lambda }_\lambda ^\nu g_{\mu \nu }=g_{\rho \lambda },$$
(4.2)
or $`\mathrm{\Lambda }\mathrm{\Lambda }^T=1`$. Thus its determinant can be either 1 or -1.
All inhomogeneous Lorentz transformations can be divided into four categories, depending on the signs of the determinant of $`\mathrm{\Lambda }`$ and the component $`\mathrm{\Lambda }_{\mathrm{\hspace{0.17em}\hspace{0.17em}0}}^0`$. We will be interested in the proper Lorentz transformations, having both signs positive. They can be built up from infinitesimal transformations involving boosts, rotations and translations, but cannot involve reflections. The proper inhomogeneous Lorentz transformations are continuous and form a Lie group called the Poincaré group. If $`a^\mu =0`$ then this group is called the Lorentz group. The principle of relativity will be satisfied if physical laws are invariant under infinitesimal transformations given by (4.1) in which
$$\mathrm{\Lambda }_\nu ^\mu =\delta _\nu ^\mu +\omega _\nu ^\mu ,$$
(4.3)
where $`\omega _\nu ^\mu `$ are infinitesimal quantities that are antisymmetric $`\omega _{\mu \nu }=\omega _{\nu \mu }`$. This property results from the orthogonality of $`\mathrm{\Lambda }_\nu ^\mu `$.
Rotations are orthogonal transformations of the coordinates mixing their spatial components and form a subgroup O(3) of the Lorentz group. They are described by a Lorentz matrix with $`\mathrm{\Lambda }_{\mathrm{\hspace{0.17em}\hspace{0.17em}0}}^0=1`$, $`\mathrm{\Lambda }_i^0=0`$ and $`\mathrm{\Lambda }_{\mathrm{\hspace{0.17em}\hspace{0.17em}0}}^i=0`$. The remaining components are functions of three angles which may be chosen as the Eulerian angles of a rigid body. For example, a rotation by the angle $`\varphi `$ about the z-axis corresponds to the Lorentz matrix,
$$R_z(\varphi )=\left(\begin{array}{cccc}1& 0& 0& 0\\ 0& \mathrm{cos}\varphi & \mathrm{sin}\varphi & 0\\ 0& \mathrm{sin}\varphi & \mathrm{cos}\varphi & 0\\ 0& 0& 0& 1\end{array}\right),$$
(4.4)
and similarly for two other axes. For an infinitesimal angle of rotation we can write only linear terms in $`\varphi `$,
$$R_z(\delta \varphi )=1+iJ_z\delta \varphi ,$$
(4.5)
and the passage to a finite rotation is given by
$$R_z(\varphi )=\underset{N\mathrm{}}{lim}(1+iJ_z\frac{\varphi }{N})^N=e^{iJ_z\varphi }.$$
(4.6)
The matrix $`J_z`$ is called the generator of the rotation about the z-axis. The rotation group is nonabelian, i.e. $`[R_x,R_y][R_y,R_x]`$. The corresponding generators satisfy the Lie algebra,
$$[J_i,J_k]=iϵ_{ikl}J_l.$$
(4.7)
Boosts are described by a Lorentz matrix with $`\mathrm{\Lambda }_j^i=0`$. For example, a boost in the z-direction with velocity $`v`$ corresponds to
$$B_z(\psi )=\left(\begin{array}{cccc}\mathrm{cosh}\psi & 0& 0& \mathrm{sinh}\psi \\ 0& 1& 0& 0\\ 0& 0& 1& 0\\ \mathrm{sinh}\psi & 0& 0& \mathrm{cosh}\psi \end{array}\right),$$
(4.8)
where $`\mathrm{cosh}\psi =\gamma =(1v^2)^{1/2}`$. Writing
$$B_z(\psi )=1+iK_z(\psi )$$
(4.9)
for an infinitesimal value of $`\psi `$ leads to the Lie algebra of the homogenous Lorentz group (rotations and boosts), given by (4.7) and
$`[J_i,K_j]=iϵ_{ijk}K_k,`$
$`[K_i,K_j]=iϵ_{ijk}J_k.`$ (4.10)
An arbitrary vector $`(p^0,𝐩)`$ transforms under the boost with velocity $`𝐯`$ according to
$`p^0=\gamma (p^0+𝐯𝐩),`$
$`𝐩_{}^{}=\gamma (𝐩_{}+𝐯p^0),`$
$`𝐩_{}^{}=𝐩_{},`$ (4.11)
where $`𝐩_{}+𝐩_{}=𝐩`$ and $`𝐩_{}=(𝐯𝐩)𝐯/v^2`$. Parametrization by $`M`$, $`𝐏`$ and $`E`$ such that $`𝐯\gamma =𝐏/M`$ and $`E=\sqrt{M^2+𝐏^2}`$ leads to the following transformation laws:
$`p^0=p^0{\displaystyle \frac{E}{M}}+{\displaystyle \frac{𝐩𝐏}{M}},`$
$`𝐩^{}=𝐩+p^0{\displaystyle \frac{𝐏}{M}}+{\displaystyle \frac{(𝐩𝐏)𝐏}{M(E+M)}}.`$ (4.12)
If we introduce the four-dimensional antisymmetric generators $`J_{\mu \nu }`$ defined by
$$J_{ij}=ϵ_{ijk}J_k,J_{0i}=K_i,$$
(4.13)
then the Lie algebra may be written as one equation,
$$[J_{\mu \nu },J_{\rho \sigma }]=i(g_{\nu \rho }J_{\mu \sigma }+g_{\mu \sigma }J_{\nu \rho }g_{\mu \rho }J_{\nu \sigma }g_{\nu \sigma }J_{\mu \rho }).$$
(4.14)
The components $`J_{\mu \nu }`$ are proportional to the quantities $`\omega _{\mu \nu }`$ given in (4.3), and infinitesimal constants of proportionality are either $`\delta \varphi `$ or $`\delta \psi `$.
The above operators may be expressed as differential operators instead of matrices. This will enable us to introduce the generators of translations. For an infinitesimal rotation about the z-axis we have
$`J_zf(x,y,z)=\underset{\varphi 0}{lim}i{\displaystyle \frac{f(x^{},y^{},z^{})f(x,y,z)}{\varphi }}=`$
$`=\underset{\varphi 0}{lim}i{\displaystyle \frac{f(x+y\varphi ,yx\varphi ,z)f(x,y,z)}{\varphi }}=i(y{\displaystyle \frac{}{x}}x{\displaystyle \frac{}{y}})f.`$ (4.15)
For a boost in the z-direction we obtain similarly,
$$K_z=i(t\frac{}{z}+z\frac{}{t}).$$
(4.16)
One can easily check that such defined operators satisfy the Lie algebra of the homogenous Lorentz group, (4.7) and 4.10). For translations we can write
$$T_zf(t,x,y,z)=\underset{\zeta 0}{lim}i\frac{f(t,x,y,z+\zeta )f(t,x,y,z)}{\zeta }=i\frac{}{z}f,$$
(4.17)
and this leads to
$`[T_\mu ,T_\nu ]=0,`$
$`[T_\rho ,J_{\mu \nu }]=i(T_\nu g_{\mu \rho }T_\mu g_{\nu \rho }).`$ (4.18)
Equations (4.14) and (4.18) constitute the complete Lie algebra of the Poincaré group.
### 4.2 The ten fundamental quantities
Another requirement for a dynamical theory is that the equations of motion should be expressible in Hamiltonian form. This is necessary in order to make a transition from classical to quantum theory. The dynamics of a system is described by quantities called dynamical variables, which for particles can be taken as their coordinates and momenta, and for fields as their four-coordinates in spacetime. Any two dynamical variables $`\xi `$ and $`\eta `$ must have a Poisson bracket $`[\xi ,\eta ]`$, and its form must not change under an infinitesimal Lorentz transformation. From this it follows that each dynamical variable $`\xi `$ will change according to
$$\xi ^{}=\xi +[\xi ,F],$$
(4.19)
where $`F`$ is an infinitesimal dynamical variable independent of $`\xi `$ and depends on the change in the coordinate system. Thus it must depend linearly on the infinitesimal quantities $`\omega _{\mu \nu }`$ and $`a_\mu `$. Therefore we can write
$$F=P^\mu a_\mu +\frac{1}{2}M^{\mu \nu }\omega _{\mu \nu },$$
(4.20)
where $`P^\mu `$ and $`M^{\mu \nu }=M^{\nu \mu }`$ are finite dynamical variables called the fundamental quantities .
Each of the ten fundamental quantities is associated with an infinitesimal transformation of the Poincaré group. $`P_0`$ is the total energy of the system and is related to a translation in time, $`P_i`$ form the three-dimensional total momentum and are related to translations in space, and $`M_{ij}`$ correspond to the total angular momentum and are related to three-dimensional rotations. The quantities $`M_{0i}`$ correspond to boosts but do not form any additive constants of motion. From the commutation relations between infinitesimal transformations (4.14) and (4.18), it follows that the Poincaré algebra is given by:
$`[P_\mu ,P_\nu ]=0,`$
$`[P_\rho ,M_{\mu \nu }]=g_{\nu \rho }P_\mu g_{\mu \rho }P_\nu ,`$
$`[M_{\mu \nu },M_{\rho \sigma }]=g_{\mu \rho }M_{\nu \sigma }g_{\nu \sigma }M_{\mu \rho }+g_{\nu \rho }M_{\mu \sigma }+g_{\mu \sigma }M_{\nu \rho }.`$ (4.21)
In order to describe a dynamical system one must find a solution of these equations, i.e. $`P_\mu `$ and $`M_{\mu \nu }`$. This is the central issue in relativistic quantum mechanics.
A simple solution of (4.21) for a single point particle is given by
$$P_\mu =p_\mu ,M_{\mu \nu }=q_\mu p_\nu =q_\nu p_\mu ,$$
(4.22)
where $`q_\mu `$ are the coordinates of a point in spacetime and $`p_\mu `$ are their conjugate momenta,
$$[q_\mu ,q_\nu ]=0,[p_\mu ,p_\nu ]=0,[p_\mu ,q_\nu ]=g_{\mu \nu }.$$
(4.23)
One usually works with dynamical variables referring to a particular instant of time. The fundamental quantities associated with transformations that leave this instant invariant (spatial translations and rotations) appear to be simple, whereas the remaining $`P_0`$ and $`M_{i0}`$ called Hamiltonians are not. Without loss of generality we may take $`q_0=0`$. Therefore $`p_0`$ no longer has a meaning. But we can modify formulae (4.22) in order to eliminate $`p_0`$ from them. Let us take
$`P_\mu =p_\mu +\lambda _\mu (p^\rho p_\rho m^2),`$
$`M_{\mu \nu }=q_\mu p_\nu q_\nu p_\mu +\lambda _{\mu \nu }(p^\rho p_\rho m^2),`$ (4.24)
where $`m`$ is a constant, with an appropriate choice of $`\lambda _\mu `$ and $`\lambda _{\mu \nu }`$. This leads to
$`P_i=p_i,M_{ij}=q_ip_jq_jp_i,`$
$`P_0=\sqrt{p_jp_j+m^2},M_{i0}=q_i\sqrt{p_jp_j+m^2}.`$ (4.25)
These are the fundamental quantities for a particle with mass $`m`$ in the so-called instant form of dynamics. There are two other forms: the point form and the front form, but the quantities appearing there are not as intuitive as in the instant form .
If for a single particle we replace $`q_i`$ by the operators $`i\frac{}{p_i}`$ and $`M_{i0}`$ by the so-called velocity operators $`V_i=\frac{1}{2}(q_iH+Hq_i)`$, where $`H=P_0`$, we will transit to quantum dynamics. In vector notation we can write
$`𝐏=𝐩,𝐌=𝐪\times 𝐩,`$
$`H=\sqrt{m^2+𝐩^2},𝐕={\displaystyle \frac{1}{2}}(𝐪H+H𝐪).`$ (4.26)
For two noninteracting particles the ten operators are given by sums,
$`𝐏=𝐩_1+𝐩_2,𝐌=𝐪_1\times 𝐩_1+𝐪_2\times 𝐩_2,`$
$`H=\sqrt{m_1^2+𝐩_1^2}+\sqrt{m_2^2+𝐩_2^2},𝐕=𝐪_1\sqrt{m_1^2+𝐩_1^2}+𝐪_2\sqrt{m_2^2+𝐩_2^2}.`$ (4.27)
The expression
$$M=\sqrt{H^2𝐏^2}$$
(4.28)
is the mass operator of the system viewed as a single entity, and commutes with all ten operators (4.27).
The last topic we will discuss in this section is related to the Casimir operators of the Lorentz group, i.e., the quantities that commute with all ten fundamental quantities $`P_\mu `$ and $`M_{\mu \nu }`$. Using the equations of the Poincaré algebra (4.21) one can show that the only operators that have this property are
$$C_1=P^\mu P_\mu ,C_2=W^\mu W_\mu ,$$
(4.29)
where $`W^\mu `$ is the Pauli-Lubanski pseudovector,
$$W^\mu =\frac{1}{2}ϵ^{\mu \nu \lambda \rho }M_{\nu \lambda }P_\rho .$$
(4.30)
The first Casimir operator is just equal to $`M^2`$, and is related to the mass of a system viewed as a single entity, or to the mass of a particle. In the rest frame of a massive particle with mass $`m`$, the operator $`C_2`$ behaves like the square of the angular momentum operator $`𝐌^2`$, and for spin $`s`$ has $`2s+1`$ eigenvalues. For a massless particle, however, there exist only two eigenvalues $`\pm s`$ .
### 4.3 Spinor representation of the Lorentz group
We defined the Lorentz group via the transformation properties of the coordinates $`x^\mu `$. Quantities that transform under the Lorentz tranformations in the same way as the coordinates are called vectors, and the matrix $`\mathrm{\Lambda }_\nu ^\mu `$ is referred to as the vector representation of the Lorentz group. This representation is suitable when dealing with vector particles having integer values of spin. However, for particles with spin $`1/2`$ (fermions) it is much more useful to introduce the spinor representation of the Lorentz group . This will be the subject of the present section.
Consider the group SU(2), consisting of $`2\times 2`$ unitary matrices with unit determinant. These conditions imply
$$U=\left(\begin{array}{cc}a& b\\ b^{}& a^{}\end{array}\right),|a|^2+|b|^2=1.$$
(4.31)
It can be shown that if a matrix $`H`$ is hermitian and traceless, so is $`H^{}`$ obtained by the transformation $`H^{}=UHU^{}`$. Let us introduce the matrix $`X`$ given by
$$X=r^i\sigma _i=\left(\begin{array}{cc}z& xiy\\ x+iy& z\end{array}\right),$$
(4.32)
where $`\sigma _i`$ are the Pauli matrices. Since $`X`$ is hermitian and traceless, so is $`X^{}=UXU^{}`$. We also have $`\text{det}X^{}=\text{det}X`$ which gives
$$x^2+y^2+z^2=x^2+y^2+z^2.$$
(4.33)
This is the condition for a rotation of the position vector $`𝐫`$. Therefore, we arrive at the conclusion that the SU(2) transformation is related to the O(3) rotation.
We would like to find the explicit form of the matrix $`U`$ that corresponds to an arbitrary rotation. For a rotation about the z-axis we have
$$x^{}=x\mathrm{cos}\varphi +y\mathrm{sin}\varphi ,y^{}=x\mathrm{sin}\varphi +y\mathrm{cos}\varphi ,z^{}=z,$$
(4.34)
and substituting this into $`X^{}U=UX`$ gives $`b=0`$ and $`a=e^{i\varphi /2}`$. Thus
$$U_z(\varphi )=\left(\begin{array}{cc}e^{i\varphi /2}& 0\\ 0& e^{i\varphi /2}\end{array}\right)=e^{i\sigma _z\varphi /2}.$$
(4.35)
This result can be generalized to a rotation about the axis with the unit vector $`𝐧`$,
$$U_𝐧(\varphi )=e^{i\sigma 𝐧\varphi /2}=\mathrm{cos}\varphi /2+i\mathrm{sin}\varphi /2\sigma 𝐧.$$
(4.36)
The above relation is very similar to the corresponding expression for the Lorentz matrix for a rotation, $`R_𝐧(\varphi )=e^{i𝐉𝐧}`$, and this is related to the fact that the Pauli matrices satisfy the same commutation relations as the matrices $`J_i`$:
$$[\sigma _i,\sigma _j]=iϵ_{ijk}\sigma _k.$$
(4.37)
When a vector rotates by the full angle $`2\pi `$, a spinor rotates only by the angle $`\pi `$ and changes sign with respect to the original value. Thus both matrices $`U`$ and $`U`$ correspond to the same rotation matrix $`R`$.
The matrix $`U`$ is regarded as the transformation matrix of a two-dimensional complex object $`\xi =\left(\begin{array}{c}\xi _1\\ \xi _2\end{array}\right)`$,
$$\xi ^{}=U\xi ,\xi ^{}=U^{}\xi ^{}.$$
(4.38)
The quantities having the above transformation property are called spinors. We see that $`\xi `$ and $`\xi ^{}`$ transform in different ways, but we may show that $`\left(\begin{array}{c}\xi _1\\ \xi _2\end{array}\right)`$ and $`i\sigma _2\xi =\left(\begin{array}{c}\xi _2^{}\\ \xi _1^{}\end{array}\right)`$ transform in the same way under SU(2). We also notice that $`\xi ^{}i\sigma _2\xi `$ is a scalar under rotations, whereas $`\xi (i\sigma _2\xi )^{}`$ transforms like a vector.
Now we proceed to transformations of spinors under boosts. From the Lie algebra of the Lorentz group (4.7) and (4.10) it follows that the matrices $`K_i=\pm iJ_i`$ are its solutions. Therefore, spinors should transform under boosts according to formula (4.36) with the replacement $`\sigma ^i\pm i\sigma ^i`$. We may define two types of spinors $`\xi `$ and $`\eta `$, transforming with a plus or a minus sign, respectively. For the first one we have
$$𝐉^{(1/2)}=\sigma /2,𝐊^{(1/2)}=i\sigma /2,$$
(4.39)
and if $`(\varphi ,\psi )`$ are the parameters of a pure rotation and a pure boost this spinor transforms according to
$$\xi ^{}=e^{i\sigma /2(\varphi i\psi )}\xi =C\xi .$$
(4.40)
For the second one we have
$$𝐉^{(1/2)}=\sigma /2,𝐊^{(1/2)}=i\sigma /2,$$
(4.41)
and the Lorentz transformation is given by
$$\eta ^{}=e^{i\sigma /2(\varphi +i\psi )}\eta =D\eta .$$
(4.42)
These are inequivalent representations of the Lorentz group and there is no matrix $`S`$ such that $`D=SCS^1`$. Instead, we have $`D=\sigma _2C^{}\sigma _2`$. The matrices $`C`$ and $`D`$ are no longer unitary, but still unimodular. Such matrices build the group SL(2,C) which is related to the Lorentz group like SU(2) was related to the rotation group. The matrix $`X`$ is now given by
$$X=x^\mu \sigma _\mu ,$$
(4.43)
where $`\sigma _0`$ is the 2$`\times `$2 unit matrix, and the transformation law has the form
$$X^{}=GXG^1,$$
(4.44)
where $`G`$ belongs to the SL(2,C).
If we define the parity transformation $`𝐯𝐯`$, which changes the sign of $`𝐊`$ but leaves the sign of $`𝐉`$, then the spinors $`\xi `$ and $`\eta `$ will interchange. Therefore we may define the four-spinor $`\mathrm{\Psi }=\left(\begin{array}{c}\xi \\ \eta \end{array}\right)`$, transforming under rotations and boosts according to
$$\mathrm{\Psi }^{}=\left(\begin{array}{cc}e^{i/2\sigma (\varphi i\psi )}& 0\\ 0& e^{i/2\sigma (\varphi +i\psi )}\end{array}\right)\mathrm{\Psi },$$
(4.45)
and under parity like
$$\mathrm{\Psi }^{}=\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right)\mathrm{\Psi }.$$
(4.46)
The bispinor $`\mathrm{\Psi }`$ is an irreducible representation of the Lorentz group extended by parity.
For pure boosts we can write
$`\xi ^{}=[\mathrm{cosh}\psi /2+\sigma 𝐧\mathrm{sinh}\psi /2]\xi ,`$
$`\eta ^{}=[\mathrm{cosh}\psi /2\sigma 𝐧\mathrm{sinh}\psi /2]\eta ,`$ (4.47)
where $`𝐧`$ is a unit vector in the direction of the boost. If $`\xi `$ and $`\eta `$ refer to a particle at rest, then $`\xi ^{}=\xi (𝐩)`$ and $`\eta ^{}=\eta (𝐩)`$, where $`\mathrm{cosh}\psi =\gamma `$ and $`𝐩=m𝐯\gamma `$. One can show that in a moving frame of reference these spinors satisfy the equation
$$\left(\begin{array}{cc}m& E+\sigma 𝐩\\ E\sigma 𝐩& m\end{array}\right)\mathrm{\Psi }(𝐩)=0,$$
(4.48)
where $`E=\sqrt{m^2+𝐩^2}`$. Introducing the 4$`\times `$4 Dirac matrices in the chiral representation,
$$\gamma ^0=\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right),\gamma ^i=\left(\begin{array}{cc}0& \sigma ^i\\ \sigma ^i& 0\end{array}\right),$$
(4.49)
leads to the Dirac equation
$$(\gamma ^\mu p_\mu m)\mathrm{\Psi }(p)=0.$$
(4.50)
The matrices $`\gamma ^\mu `$ satisfy the anticommutation relation
$$\gamma ^\mu \gamma ^\nu +\gamma ^\nu \gamma ^\mu =2g^{\mu \nu },$$
(4.51)
which is actually their definition.
We will work in the standard representation of the Dirac matrices, in which $`\gamma ^0`$ is diagonal,
$$\gamma ^0=\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right).$$
(4.52)
It can be obtained from the chiral representation by
$$\gamma _s^0=T\gamma _c^0T^1,$$
(4.53)
where
$$T=\frac{1}{\sqrt{2}}\left(\begin{array}{cc}1& 1\\ 1& 1\end{array}\right).$$
(4.54)
Therefore the bispinor becomes
$$\mathrm{\Psi }=\frac{1}{\sqrt{2}}\left(\begin{array}{c}\xi +\eta \\ \xi \eta \end{array}\right),$$
(4.55)
and the spinor representation of the boost is given by the matrix
$$S=\left(\begin{array}{cc}\mathrm{cosh}\psi /2& \mathrm{sinh}\psi /2\sigma 𝐧\\ \mathrm{sinh}\psi /2\sigma 𝐧& \mathrm{cosh}\psi /2\end{array}\right),$$
(4.56)
or finally
$$S(0𝐩)=S(m,𝐩)=\frac{1}{\sqrt{2m(E(m,𝐩)+m)}}\left(\begin{array}{cc}E(m,𝐩)+m& \sigma 𝐩\\ \sigma 𝐩& E(m,𝐩)+m\end{array}\right).$$
(4.57)
The spinor representation for rotations and boosts can be also derived from the assumption that the Dirac equation in the position space
$$(i\gamma ^\mu _\mu m)\mathrm{\Psi }=0$$
(4.58)
is invariant under Lorentz transformations $`x^\mu x^\mu `$. For the bispinor we will assume that it transforms according to
$$\mathrm{\Psi }^{}=S\mathrm{\Psi },$$
(4.59)
where $`S`$ is a unimodular matrix corresponding to the Lorentz transformation. Its form can be derived from the requirement that the Dirac equation will remain unchanged,
$$(i\gamma ^\mu ^{}_\mu ^{}m)\mathrm{\Psi }^{}=0,$$
(4.60)
leading to
$$S^1\gamma ^\mu S=\mathrm{\Lambda }_\nu ^\mu \gamma ^\nu .$$
(4.61)
In order to derive the form of $`S`$, we will first consider the infinitesimal Lorentz transformation (4.3). Consequently, the matrix $`S`$ will be a linear function of the generators $`\omega _{\mu \nu }`$, and the simplest guess is
$$S=1+a\omega _{\mu \nu }\gamma ^\mu \gamma ^\nu ,$$
(4.62)
where $`a`$ is some constant. The inverse matrix $`S^1`$ in the linear approximation which suffices for our considerations is
$$S^1=1a\omega _{\mu \nu }\gamma ^\mu \gamma ^\nu .$$
(4.63)
Substitution of $`S`$ and $`S^1`$ into (4.58) gives $`a=1/4`$. The passage to finite transformations is again done by exponentiation,
$$S=e^{\frac{1}{4}\omega _{\mu \nu }\gamma ^\mu \gamma ^\nu }.$$
(4.64)
One can show that this expression is equivalent to (4.57).
The explicit form of bispinors in the frame of reference in which a particle has momentum $`𝐩`$ can be obtained by acting with (4.57) on the solutions of the Dirac equations in the rest frame, given by
$$u(\lambda )=\left(\begin{array}{c}\chi (\lambda )\\ 0\end{array}\right),v(\lambda )=\left(\begin{array}{c}0\\ i\sigma _2\chi (\lambda )\end{array}\right),$$
(4.65)
where $`\lambda =\pm 1/2`$ are the values of spin. The term $`i\sigma _2`$ in $`v`$ guarantees that spinors $`v`$ and $`u^{}`$ transform in the same way under SU(2), and
$$\chi ^T(+1/2)=(1,0),\chi ^T(1/2)=(0,1).$$
(4.66)
Thus we get
$$u(𝐩,\lambda )=\frac{1}{\sqrt{E(m,𝐩)+m}}\left(\begin{array}{c}(E(m,𝐩)+m)\chi (\lambda )\\ (\sigma 𝐩)\chi (\lambda )\end{array}\right)$$
(4.67)
and
$$v(𝐩,\lambda )=\frac{1}{\sqrt{E(m,𝐩)+m}}\left(\begin{array}{c}(\sigma 𝐩)i\sigma _2\chi (\lambda )\\ (E(m,𝐩)+m)i\sigma _2\chi (\lambda )\end{array}\right).$$
(4.68)
The nonrelativistic limit of the Dirac equation is free of paradoxical properties only in the first approximation. It is possible, however, to find a representation in which it is clear how to associate operators with classical dynamical variables so that these operators tend to their expected nonrelativistic form . In the presence of an external field the nonrelativistic reduction is most conveniently obtained by an infinite set of canonical transformations related to a free field transformation. This is referred to as the Foldy-Wouthuysen representation, and can also be extended to Klein-Gordon and Proca particles . In this work the nonrelativistic limit will be reduced to the first approximation. Therefore, the Dirac representation will be suitable for our purposes.
### 4.4 The Bakamjian-Thomas model for interacting particles
For a system with a fixed number of particles, $`P_i`$ and $`M_{ij}`$ will be sums of their values for separate particles,
$$P_i=p_i,M_{ij}=(q_ip_jq_jp_i).$$
(4.69)
For the Hamiltonians one must add the interaction terms,
$`P_0={\displaystyle \sqrt{p_jp_j+m^2}}+U,`$
$`M_{i0}={\displaystyle q_i\sqrt{p_jp_j+m^2}}+U_i.`$ (4.70)
From the commutation relations (4.21) it follows that $`U`$ is a three-dimensional scalar, $`U_i`$ is a three-dimensional vector, and
$$U_i=q_iU+b_i,$$
(4.71)
where $`b_i`$ is a constant three-dimensional vector. The remaining conditions for $`U`$ and $`U_i`$ are quadratic and therefore a construction of a complete dynamical theory of a relativistic theory is very difficult.
An interaction enters only in $`H`$ and $`𝐕`$. A practical method of constructing the generators in this case was developed in , where the set of new operators satisfying simpler commutation relations was introduced. In this set, the interaction appears only in the mass operator (4.28). Suppose we can make a transformation from $`𝐪_1,𝐪_2`$ and $`𝐩_1,𝐩_2`$ to the total momentum $`𝐏`$, the coordinates of the center-of-mass $`𝐑`$, the relative momentum $`𝐩`$ and the relative coordinate vector $`𝐫`$. The commutation relations are not disturbed if:
1. $`𝐌=𝐑\times 𝐏+𝐫\times 𝐩`$,
2. $`M`$ depends on $`𝐩`$ only,
3. $`𝐕`$ can be expressed in terms of $`M`$, $`𝐏`$, $`𝐌`$ and $`𝐑`$ only.
The interaction will be included if we replace $`M`$ by any other function of $`𝐩`$ and $`𝐫`$ which is a scalar for space rotations. The only nonzero commutators of the set $`M,𝐏,𝐌,𝐑`$ are
$$[P_i,R_j]=i\delta _{ij},[M_i,M_j]=iϵ_{ijk}M_k,$$
(4.72)
and the mass operator $`M`$ is Poincare invariant if it commutes with $`𝐏,𝐌,𝐑`$. Therefore it is only necessary to make sure that the above condition is satisfied. The macroscopic Hamiltonian of a system is given by
$$H=\sqrt{𝐏^2+M^2(𝐩,𝐫)},$$
(4.73)
and is obtained from the microscopic one (4.70) via introducing the above relative variables. The above results can be generalized to systems with more than two particles, and to particles with intrinsic spin . An explicit construction for a unitary operator that insures the free motion of the center of mass of any system is given in . However, for a given potential there is no unique way in which the relative variables may be defined . Unfortunately, an exact form of the potential is not known and one must use phenomenological forms of the mass operator. In this work we will assume a gaussian form of the orbital wave function and fit it to a few measured form factors. This approach will not allow for a deeper understanding of the relativistic quark dynamics, although it makes possible to estimate relativistic corrections to the $`\pi _1`$ decay widths.
### 4.5 Wigner rotation
In this section we will derive how spin transforms under the boost transformations. It will be necessary for a construction of covariant spin wave functions for quark-antiquark pairs (mesons). Our goal is to solve
$$\mathrm{\Lambda }(0𝐏)|𝐩,\lambda =\mathrm{\Lambda }(0𝐏)\mathrm{\Lambda }(0𝐩)|0,\lambda ,$$
(4.74)
where $`\mathrm{\Lambda }(𝐩𝐪)`$ boosts a particle with a momentum $`𝐩`$ to a frame in which the momentum is equal to $`𝐪`$, and $`|𝐩,\lambda `$ is the spinor state. Multiplying the above expression by the identity $`\mathrm{\Lambda }(0𝐩^{})\mathrm{\Lambda }(𝐩^{}0)`$, where $`𝐩^{}`$ is obtained from $`𝐩`$ according to (4.12), leads to
$$\mathrm{\Lambda }(0𝐏)|𝐩,\lambda =\mathrm{\Lambda }(0𝐩^{})R(𝐩,𝐏)|0,\lambda .$$
(4.75)
The quantity $`R`$ is given by
$$R(𝐩,𝐏)=\mathrm{\Lambda }(𝐩^{}0)\mathrm{\Lambda }(𝐩𝐩^{})\mathrm{\Lambda }(0𝐩),$$
(4.76)
and we will find its explicit form.
In spinor representation we can write
$$R(𝐩,𝐏)=S(m,𝐩^{})S(M,𝐏)S(m,𝐩),$$
(4.77)
where $`M`$ parametrizes the boost $`\mathrm{\Lambda }(0𝐏)`$ like before. Substituting the expression (4.57) in the above equation leads after somewhat lengthy calculations to
$$R(𝐩,𝐏)=\left(\begin{array}{cc}D^{(1/2)}(𝐩,𝐏)& 0\\ 0& D^{(1/2)}(𝐩,𝐏)\end{array}\right),$$
(4.78)
where
$$D_{\lambda \lambda ^{}}^{(1/2)}(𝐪,𝐏)=\left[\frac{(E(m,𝐪)+m)(E(M,𝐏)+M)+𝐏𝐪+i\sigma (𝐏\times 𝐪)}{\sqrt{2(E(m,𝐪)+m)(E(M,𝐏)+M)(E(m,𝐪)E(M,𝐏)+𝐏𝐪+mM)}}\right]_{\lambda \lambda ^{}}.$$
(4.79)
It can be shown that $`D^{(1/2)}`$ has the form $`\mathrm{cos}\varphi /2+i\mathrm{sin}\varphi /2\sigma 𝐧`$ and thus the matrix (4.78) represents a pure rotation. The matrix $`D^{(1/2)}`$ is called the Wigner rotation matrix. In nonrelativistic quantum mechanics, spin should not change under boosts and this is reflected in the large-mass limit of formula (4.79),
$$D_{\lambda \lambda ^{}}^{(1/2)}\delta _{\lambda \lambda ^{}}.$$
(4.80)
Finally, we obtain the transformation law for spinor states under boosts,
$$\mathrm{\Lambda }(0𝐏)|𝐩,\lambda =\mathrm{\Lambda }(0𝐩^{})D_{\lambda \lambda ^{}}^{(1/2)}(𝐩,𝐏)|0,\lambda ^{}=D_{\lambda \lambda ^{}}^{(1/2)}(𝐩,𝐏)|\mathrm{\Lambda }(0𝐏)𝐩,\lambda ^{}.$$
(4.81)
The state of a system having more than one spin index transforms like a spin tensor, i.e., each index transforms independently with the $`D^{(1/2)}`$ matrix according to formula (4.81).
## Chapter 5 Relativistic spin wave function for mesons and hybrids
Having described transformation laws for a single particle with spin $`1/2`$, we may proceed to systems of noninteracting particles. We will focus on quark-antiquark pairs, i.e. mesons. This is necessary in order to construct the spin wave functions for the outgoing mesons resulting from the decay of the $`\pi _1`$. Since these mesons have nonzero momenta, a relativistic model of hadronic decays will have to include a Wigner rotation of spin. A similar construction is also required for the $`\pi _1`$ spin wave function because the $`q\overline{q}`$ pair must be boosted to a moving frame before it couples with a gluon to a rest-frame hybrid. In the following section we will show how to build a relativistic spin wave function for each light unflavored meson (or a meson with equal masses of quarks). Following that we will add the gluon and build the $`\pi _1`$. Mesons with different masses of quarks will be considered later. In each case we will begin with a rest-frame function, and then use the results of the preceding chapter to obtain a general expression for any frame of reference<sup>2</sup><sup>2</sup>2This chapter is based on work by A.P.Szczepaniak.
### 5.1 Meson spin wave functions
The spin wave function for a meson is constructed as an element of an irreducible representation of the Poincare group . In the rest frame of a meson, the quark momenta are given by
$$l_q^\mu =(E(m_q,𝐪),𝐪),l_{\overline{q}}^\mu =(E(m_{\overline{q}},𝐪),𝐪),$$
(5.1)
and the normalized spin-0 and spin-1 wave function corresponding to $`J^{PC}=0^+`$ and $`J^{PC}=1^{}`$ are simply given by Clebsch-Gordan coefficients,
$$\mathrm{\Psi }_{q\overline{q}}(𝐪,𝐥_{q\overline{q}}=0,\sigma _q,\sigma _{\overline{q}})=\frac{1}{2},\sigma _q;\frac{1}{2},\sigma _{\overline{q}}|0,0=\chi ^{}(\sigma _q)\frac{i\sigma _2}{\sqrt{2}}\chi (\sigma _{\overline{q}}),$$
(5.2)
and
$$\mathrm{\Psi }_{q\overline{q}}^{\lambda _{q\overline{q}}}(𝐪,𝐥_{q\overline{q}}=0,\sigma _q,\sigma _{\overline{q}})=\frac{1}{2},\sigma _q;\frac{1}{2},\sigma _{\overline{q}}|1,\lambda _{q\overline{q}}=\chi ^{}(\sigma _q)\frac{\sigma ^ii\sigma _2}{\sqrt{2}}\chi (\sigma _{\overline{q}})ϵ^i(\lambda _{q\overline{q}}).$$
(5.3)
A factor $`i\sigma _2`$ accounts that the antiparticle spin doublet transforms under SU(2) in the same way as the particle doublet. The canonical polarization vectors
$$ϵ(\pm 1)=\frac{1}{\sqrt{2}}\left(\begin{array}{c}1\\ \pm i\\ 0\end{array}\right),ϵ(0)=\left(\begin{array}{c}0\\ 0\\ 1\end{array}\right),$$
(5.4)
correspond to spin 1 quantized along the z-axis and satisfy the orthogonality relation
$$\underset{\lambda }{}ϵ^i(\lambda )ϵ^j(\lambda )=\delta ^{ij}.$$
(5.5)
The invariant mass of the $`q\overline{q}`$ pair is
$$m_{q\overline{q}}=E(m_q,𝐪)+E(m_{\overline{q}},𝐪),$$
(5.6)
where $`E(m,𝐩)=\sqrt{m^2+𝐩^2}`$, and the total momentum of this system $`𝐥_{q\overline{q}}`$ is of course equal to zero. In the following we will assume
$$m_q=m_{\overline{q}}=m.$$
(5.7)
This condition is satisfied to a good approximation by the quarks $`u,d`$ and may be used for a construction of light unflavored meson states.
The rest frame wave functions (5.2) and (5.3) may also be expressed in terms of Dirac spinors quantized along the z-axis,
$$\mathrm{\Psi }_{q\overline{q}}(𝐪,𝐥_{q\overline{q}}=0,\sigma _q,\sigma _{\overline{q}})=\frac{1}{\sqrt{2}m_{q\overline{q}}}\overline{u}(𝐪,\sigma _q)\gamma ^5v(𝐪,\sigma _{\overline{q}}),$$
(5.8)
and
$$\mathrm{\Psi }_{q\overline{q}}^{\lambda _{q\overline{q}}}(𝐪,𝐥_{q\overline{q}}=0,\sigma _q,\sigma _{\overline{q}})=\frac{1}{\sqrt{2}m_{q\overline{q}}}\overline{u}(𝐪,\sigma _q)[\gamma ^i\frac{2q^i}{m_{q\overline{q}}+2m}]v(𝐪,\sigma _{\overline{q}})ϵ^i(𝐥_{q\overline{q}}=0,\lambda _{q\overline{q}}).$$
(5.9)
A spin wave function written in this form is manifestly covariant and thus it is straightforward to find how it transforms under Lorentz transformations. In the above $`ϵ^i(𝐥_{q\overline{q}}=0,\lambda _{q\overline{q}})`$ are the spatial components of the polarization four-vector
$$ϵ^\mu (𝐥_{q\overline{q}}=0,\lambda _{q\overline{q}})=(0,ϵ(\lambda _{q\overline{q}})),$$
(5.10)
whose time component is zero in order to satisfy the transversity condition $`ϵ^\mu (𝐤,\lambda )k_\mu =0`$.
Now we apply a boost from the rest frame of a $`q\overline{q}`$ pair to a frame of reference in which the momenta of the quark and the antiquark are $`𝐥_q`$ and $`𝐥_{\overline{q}}`$, respectively, and the total momentum is $`𝐥_{q\overline{q}}=𝐥_q+𝐥_{\overline{q}}`$, as shown in Fig. 5.1. The new momenta are given by
$$𝐥_q=𝐪+\frac{(𝐪𝐥_{q\overline{q}})𝐥_{q\overline{q}}}{E(m_{q\overline{q}},𝐥_{q\overline{q}})[m_{q\overline{q}}+E(m_{q\overline{q}},𝐥_{q\overline{q}})]}+\frac{E(m,𝐪)}{m_{q\overline{q}}}𝐥_{q\overline{q}},$$
(5.11)
and
$$𝐥_{\overline{q}}=𝐪\frac{(𝐪𝐥_{q\overline{q}})𝐥_{q\overline{q}}}{E(m_{q\overline{q}},𝐥_{q\overline{q}})[m_{q\overline{q}}+E(m_{q\overline{q}},𝐥_{q\overline{q}})]}+\frac{E(m,𝐪)}{m_{q\overline{q}}}𝐥_{q\overline{q}}.$$
(5.12)
The spin wave function of a meson in a moving frame is obtained from the rest frame wave function, as we stated at the end of Chapter 4, by acting with the Wigner rotation matrix (4.79) on each of both spin indices. This gives
$$\mathrm{\Psi }_{q\overline{q}}^{\lambda _{q\overline{q}}}(𝐪,𝐥_{q\overline{q}},\lambda _q,\lambda _{\overline{q}})=\underset{\sigma _q,\sigma _{\overline{q}}}{}\mathrm{\Psi }_{q\overline{q}}^{\lambda _{q\overline{q}}}(𝐪,𝐥_{q\overline{q}}=0,\sigma _q,\sigma _{\overline{q}})D_{\lambda _q\sigma _q}^{(1/2)}(𝐪,𝐥_{q\overline{q}})D_{\lambda _{\overline{q}}\sigma _{\overline{q}}}^{(1/2)}(𝐪,𝐥_{q\overline{q}}).$$
(5.13)
The above Wigner rotation matrix corresponds to the boost with $`𝐯\gamma =𝐏/M`$. In spinor representation Eq. (4.81) leads to the following transformation laws for spinors $`u^{}`$ and $`v`$:
$`{\displaystyle \underset{\sigma _{\overline{q}}}{}}D_{\lambda _{\overline{q}}\sigma _{\overline{q}}}^{(1/2)}(𝐪,𝐥_{q\overline{q}})v(𝐪,\sigma _{\overline{q}})=S(𝐥_{q\overline{q}}0)v(𝐥_{\overline{q}},\lambda _{\overline{q}}),`$
$`{\displaystyle \underset{\sigma _q}{}}D_{\lambda _q\sigma _q}^{(1/2)}(𝐪,𝐥_{q\overline{q}})u^{}(𝐪,\sigma _q)=u^{}(𝐥_q,\lambda _q)S^{}(𝐥_{q\overline{q}}0),`$ (5.14)
where $`S(𝐥_{q\overline{q}}0)`$ is the Dirac representation of the boost taking $`𝐥_q`$ to $`𝐪`$ and $`𝐥_{\overline{q}}`$ to $`𝐪`$, given by (4.57) with $`𝐩=𝐥_{q\overline{q}}`$ and $`M=m_{q\overline{q}}`$. From these laws we obtain the general form of the spin-0 wave function:
$`\mathrm{\Psi }_{q\overline{q}}(𝐪,𝐥_{q\overline{q}},\lambda _q,\lambda _{\overline{q}})={\displaystyle \frac{1}{\sqrt{2}m_{q\overline{q}}}}u^{}(𝐥_q,\lambda _q)S^{}(𝐥_{q\overline{q}}0)\gamma ^0\gamma ^5S(𝐥_{q\overline{q}}0)v(𝐥_{\overline{q}},\lambda _{\overline{q}})=`$
$`={\displaystyle \frac{1}{\sqrt{2}m_{q\overline{q}}}}\overline{u}(𝐥_q,\lambda _q)S^1(𝐥_{q\overline{q}}0)\gamma ^5S(𝐥_{q\overline{q}}0)v(𝐥_{\overline{q}},\lambda _{\overline{q}})=`$
$`={\displaystyle \frac{1}{\sqrt{2}m_{q\overline{q}}}}\overline{u}(𝐥_q,\lambda _q)\gamma ^5v(𝐥_{\overline{q}},\lambda _{\overline{q}}).`$ (5.15)
Similarly we derive the spin-1 wave function:
$`\mathrm{\Psi }_{q\overline{q}}^{\lambda _{q\overline{q}}}(𝐪,𝐥_{q\overline{q}},\lambda _q,\lambda _{\overline{q}})=`$
$`={\displaystyle \frac{1}{\sqrt{2}m_{q\overline{q}}}}u^{}(𝐥_q,\lambda _q)S^{}(𝐥_{q\overline{q}}0)\gamma ^0\left(\gamma ^i{\displaystyle \frac{2q^i}{m_{q\overline{q}}+2m}}\right)S(𝐥_{q\overline{q}}0)v(𝐥_{\overline{q}},\lambda _{\overline{q}})ϵ^i(\lambda _{q\overline{q}})=`$
$`={\displaystyle \frac{1}{\sqrt{2}m_{q\overline{q}}}}\overline{u}(𝐥_q,\lambda _q)S^1(𝐥_{q\overline{q}}0)\left(\gamma ^i{\displaystyle \frac{2q^i}{m_{q\overline{q}}+2m}}\right)S(𝐥_{q\overline{q}}0)v(𝐥_{\overline{q}},\lambda _{\overline{q}})ϵ^i(\lambda _{q\overline{q}})=`$
$`={\displaystyle \frac{1}{\sqrt{2}m_{q\overline{q}}}}\overline{u}(𝐥_q,\lambda _q)\left[\mathrm{\Lambda }_\nu ^i(𝐥_{q\overline{q}}0)\gamma ^\nu {\displaystyle \frac{2q^i}{m_{q\overline{q}}+2m}}\right]v(𝐥_{\overline{q}},\lambda _{\overline{q}})ϵ^i(\lambda _{q\overline{q}})=`$
$`={\displaystyle \frac{1}{\sqrt{2}m_{q\overline{q}}}}\overline{u}(𝐥_q,\lambda _q)\left[\gamma ^\nu {\displaystyle \frac{p_q^\nu p_{\overline{q}}^\nu }{m_{q\overline{q}}+2m}}\right]v(𝐥_{\overline{q}},\lambda _{\overline{q}})\mathrm{\Lambda }_\nu ^i(𝐥_{q\overline{q}}0)ϵ^i(\lambda _{q\overline{q}})=`$
$`={\displaystyle \frac{1}{\sqrt{2}m_{q\overline{q}}}}\overline{u}(𝐥_q,\lambda _q)\left[\gamma ^\mu {\displaystyle \frac{l_q^\mu l_{\overline{q}}^\mu }{m_{q\overline{q}}+2m}}\right]v(𝐥_{\overline{q}},\lambda _{\overline{q}})ϵ_\mu (𝐥_{q\overline{q}},\lambda _{q\overline{q}}),`$ (5.16)
where $`ϵ^\mu (𝐥_{q\overline{q}},\lambda _{q\overline{q}})`$ are obtained from (5.4) through the boost with $`\beta \gamma =𝐥_{q\overline{q}}/m_{q\overline{q}}`$:
$`ϵ^0(𝐥_{q\overline{q}},\lambda _{q\overline{q}})={\displaystyle \frac{𝐥_{q\overline{q}}ϵ(\lambda _{q\overline{q}})}{m_{q\overline{q}}}},`$
$`ϵ(𝐥_{q\overline{q}},\lambda _{q\overline{q}})=ϵ(\lambda _{q\overline{q}})+{\displaystyle \frac{(𝐥_{q\overline{q}}ϵ(\lambda _{q\overline{q}}))𝐥_{q\overline{q}}}{m_{q\overline{q}}(E(m_{q\overline{q}},𝐥_{q\overline{q}})+m_{q\overline{q}})}}.`$ (5.17)
The invariant mass of the $`q\overline{q}`$ system is now
$$m_{q\overline{q}}=m_{q\overline{q}}(𝐥_q,𝐥_{\overline{q}})=\sqrt{(E(m,𝐥_q)+E(m,𝐥_{\overline{q}}))^2(𝐥_q+𝐥_{\overline{q}})^2}.$$
(5.18)
The wave functions (5.15) and (5.16) are still normalized:
$`{\displaystyle \underset{\lambda _q,\lambda _{\overline{q}}}{}}\mathrm{\Psi }_{q\overline{q}}^{}(𝐪,𝐥_{q\overline{q}},\lambda _q,\lambda _{\overline{q}})\mathrm{\Psi }_{q\overline{q}}(𝐪,𝐥_{q\overline{q}},\lambda _q,\lambda _{\overline{q}})=1,`$
$`{\displaystyle \underset{\lambda _q,\lambda _{\overline{q}}}{}}\mathrm{\Psi }_{q\overline{q}}^{\lambda _{q\overline{q}}}(𝐪,𝐥_{q\overline{q}},\lambda _q,\lambda _{\overline{q}})\mathrm{\Psi }_{q\overline{q}}^{\lambda _{q\overline{q}}^{}}(𝐪,𝐥_{q\overline{q}},\lambda _q,\lambda _{\overline{q}})=\delta _{\lambda _{q\overline{q}}\lambda _{q\overline{q}}^{}}.`$ (5.19)
By coupling the spin wave function (5.8) or (5.9), respectively, with one unit of the orbital angular momentum $`L=1`$, one obtains the rest frame spin wave functions for the quark-antiquark pair with quantum numbers $`J^{PC}=1^+`$ or $`0^{++}`$, $`1^{++}`$ and $`2^{++}`$. Explicitly we have
$$\mathrm{\Psi }_{q\overline{q}}^{\lambda _{q\overline{q}}}(𝐪,𝐥_{q\overline{q}}=0,\sigma _q,\sigma _{\overline{q}})=\frac{1}{\sqrt{2}m_{q\overline{q}}(𝐪,𝐪)}\overline{u}(𝐪,\sigma _q)\gamma ^5v(𝐪,\sigma _{\overline{q}})Y_{1\lambda _{q\overline{q}}}(\overline{𝐪})$$
(5.20)
for the $`1^+`$ meson, and
$`\mathrm{\Psi }_{q\overline{q}}^{\lambda _{q\overline{q}}}(𝐪,𝐥_{q\overline{q}}=0,\sigma _q,\sigma _{\overline{q}})={\displaystyle \underset{\lambda ,l}{}}{\displaystyle \frac{1}{\sqrt{2}m_{q\overline{q}}}}\overline{u}(𝐪,\sigma _q)[\gamma ^i{\displaystyle \frac{2q^i}{m_{q\overline{q}}+2m}}]v(𝐪,\sigma _{\overline{q}})ϵ^i(0,\lambda )`$
$`\times Y_{1l}(\overline{𝐪})1,\lambda ;1,l|J,\lambda _{q\overline{q}}`$ (5.21)
for the $`J^{++}`$ ($`J=0,1,2`$). Here $`Y_{L\lambda }(\overline{𝐪})`$ is a spherical harmonic and $`\overline{𝐪}=𝐪/|𝐪|`$. Using (5.13) one can show that the wave functions for the $`q\overline{q}`$ pair moving with the total momentum $`𝐥_{q\overline{q}}`$ are given by
$$\mathrm{\Psi }_{q\overline{q}}^{\lambda _{q\overline{q}}}(𝐥_q,𝐥_{\overline{q}},\lambda _q,\lambda _{\overline{q}})=\frac{1}{\sqrt{2}m_{q\overline{q}}(𝐥_q,𝐥_{\overline{q}})}\overline{u}(𝐥_q,\lambda _q)\gamma ^5v(𝐥_{\overline{q}},\lambda _{\overline{q}})Y_{1\lambda _{q\overline{q}}}(\overline{𝐪}),$$
(5.22)
and
$`\mathrm{\Psi }_{q\overline{q}}^{\lambda _{q\overline{q}}}(𝐥_q,𝐥_{\overline{q}},\lambda _q,\lambda _{\overline{q}})={\displaystyle \underset{\lambda ,l}{}}{\displaystyle \frac{1}{\sqrt{2}m_{q\overline{q}}}}\overline{u}(𝐥_q,\lambda _q)\left[\gamma ^\mu {\displaystyle \frac{l_q^\mu l_{\overline{q}}^\mu }{m_{q\overline{q}}+2m}}\right]v(𝐥_{\overline{q}},\lambda _{\overline{q}})ϵ_\mu (𝐥_{q\overline{q}},\lambda )`$
$`\times Y_{1l}(\overline{𝐪})1,\lambda ;1,l|J,\lambda _{q\overline{q}},`$ (5.23)
respectively, where $`m_{q\overline{q}}=m_{q\overline{q}}(𝐥_q,𝐥_{\overline{q}})`$ and $`𝐪`$ remains the same but must be written in terms of new variables:
$$𝐪=\mathrm{\Lambda }(𝐥_{q\overline{q}}0)𝐥_q=𝐥_q\frac{E(m,𝐥_q)𝐥_{q\overline{q}}}{m_{q\overline{q}}}+\frac{(𝐥_{q\overline{q}}𝐥_q)𝐥_{q\overline{q}}}{m_{q\overline{q}}(E(m_{q\overline{q}},𝐥_{q\overline{q}})+m_{q\overline{q}})}.$$
(5.24)
In order to construct meson spin wave functions for higher orbital angular momenta $`L`$ one need only to replace $`Y_{1l}`$ with $`Y_{Ll}`$ in (5.22) and (5.23). In $`L=0`$ \[formulae (5.8), (5.9), (5.15) and (5.16)\] we skipped a constant factor $`Y_{00}`$ to make the spin wave function normalized to $`1`$. But from now on, for consistency, we will assume this constant being implicitly included. In the nonrelativistic limit, where Wigner rotations may be ignored, all the spin wave functions for mesons simply reduce to the Clebsch-Gordan coefficients that we started from, coupled to appropriate spherical harmonics.
### 5.2 The $`\pi _1`$ spin wave function
As we showed in Chapter 3, in the lightest hybrid meson $`\pi _1`$ wave function, a constituent gluon is expected to have one unit of orbital angular momentum with respect to a $`q\overline{q}`$ pair. Thus, the quantum numbers $`P,C`$ require a quark and an antiquark to have parallel spins ($`S=1`$)
In the rest frame of a 3-body system with a $`q\overline{q}`$ pair moving with momentum $`𝐐`$ and a transverse gluon with momentum $`𝐐`$, the total spin wave function of the hybrid is obtained by coupling the $`q\overline{q}`$ spin-1 wave function (5.16) and the gluon wave function ($`J^{PC}=1^{}`$) to a total spin $`S=0,1,2`$ and $`J^{PC}=0^{++},1^{++},2^{++}`$ states, respectively. Here, we will derive the expressions for each value of $`S`$ separately, although the physical $`\pi _1`$ state should be a superposition of all three components. The way of calculating the corresponding coefficients in this linear combination will be given in Chapter 7. The total $`J^{PC}=1^+`$ exotic meson wave function is then obtained by adding one unit of orbital angular momentum between the gluon and the $`q\overline{q}`$:
$`\mathrm{\Psi }_{q\overline{q}g(S)}^{\lambda _{ex}}(\lambda _q,\lambda _{\overline{q}},\lambda _g)={\displaystyle \underset{\lambda _{q\overline{q}},\sigma =\pm 1,M,l}{}}\mathrm{\Psi }_{q\overline{q}}^{\lambda _{q\overline{q}}}(𝐪,𝐥_{q\overline{q}}=𝐐,\lambda _q,\lambda _{\overline{q}})1,\lambda _{q\overline{q}};1,\sigma |S,MD_{\lambda _g\sigma }^{(1)}(\overline{𝐐})`$
$`\times Y_{1l}(\overline{𝐐})S,M;1,l|1,\lambda _{ex}.`$ (5.25)
The spin-1 rotation matrix $`D^{(1)}`$ relates the transverse gluon states in the helicity basis $`\sigma `$ (i.e., along its momentum) to the basis described by spin $`\lambda _g`$ quantized along a fixed z-axis, as shown in Fig. 5.2. Thus, the helicity basis is rotated so that a coupling between a gluon and a $`q\overline{q}`$ pair may be done in the same basis with Clebsch-Gordan coefficients, since we are in the rest frame of a hybrid. Explicitly we have
$$|𝐐,\lambda _g=\underset{\sigma }{}D_{\lambda _g\sigma }^{(1)}(\varphi ,\theta ,\varphi )|𝐐,\sigma ,$$
(5.26)
where $`\theta `$ and $`\varphi `$ are the polar angle and the azimuth of the direction of the gluon momentum $`𝐐`$, as shown in Fig. 5.3.
For the gluon polarization vector with spin quantized along the z-axis we can write
$$ϵ_c^i(𝐐,\lambda _g)=\underset{\sigma =\pm 1}{}D_{\lambda _g\sigma }^{(1)}(\varphi ,\theta ,\varphi )ϵ_h^i(𝐐,\sigma ),$$
(5.27)
where the helicity polarization vectors are given by
$$ϵ_h^i(𝐐,\sigma )=\underset{\lambda _g}{}D_{\lambda _g\sigma }^{(1)}(\varphi ,\theta ,\varphi )ϵ^i(\lambda _g).$$
(5.28)
Using the unitarity of the matrix $`D^{(1)}`$ one can show
$$ϵ_c^i(𝐐,\lambda _g)ϵ_h^i(𝐐,\sigma )=D_{\lambda _g\sigma }^{(1)},$$
(5.29)
and with the help of the identity $`ϵ_h^i(𝐐,\sigma )ϵ_h^j(𝐐,\sigma )=\delta ^{ij}\overline{Q}^i\overline{Q}^j`$ we finally obtain
$$ϵ_c^i(𝐐,\lambda _g)=ϵ^j(\lambda _g)(\delta ^{ij}\overline{Q}^i\overline{Q}^j),$$
(5.30)
where $`\overline{𝐐}^i=𝐐^i/|𝐐|`$.
The Clebsch-Gordan coefficients and the spherical harmonic in (5.25) can be expressed in terms of the polarization vectors (5.4). For example:
$`1,\lambda ^{};0,0|1,\lambda =ϵ^{}(\lambda ^{})ϵ(\lambda ),`$
$`1,\lambda ^{};1,\lambda |0,0={\displaystyle \frac{1}{\sqrt{3}}}ϵ^{}(\lambda ^{})ϵ^{}(\lambda ),`$
$`1,\lambda ^{};1,\lambda ^{\prime \prime }|1,\lambda ={\displaystyle \frac{i}{\sqrt{2}}}[ϵ^{}(\lambda ^{})\times ϵ^{}(\lambda ^{\prime \prime })]ϵ(\lambda ),`$
$`Y_{1l}(\overline{𝐐})=\sqrt{{\displaystyle \frac{3}{4\pi }}}ϵ(l)\overline{𝐐}.`$ (5.31)
Therefore we obtain:
$`{\displaystyle \underset{l}{}}1,\lambda _{q\overline{q}};1,\lambda _g|0,0Y_{1l}(\overline{𝐐})0,0;1,l|1,\lambda _{ex}[ϵ^{}(\lambda _{q\overline{q}})ϵ^{}(\lambda _g)][\overline{𝐐}ϵ(\lambda _{ex})],`$
$`{\displaystyle \underset{l,s}{}}1,\lambda _{q\overline{q}};1,\lambda _g|1,sY_{1l}(\overline{𝐐})1,s;1,l|1,\lambda _{ex}[ϵ^{}(\lambda _{q\overline{q}})\times ϵ^{}(\lambda _g)][\overline{𝐐}\times ϵ(\lambda _{ex})],`$
$`{\displaystyle \underset{l,s}{}}1,\lambda _{q\overline{q}};1,\lambda _g|2,sY_{1l}(\overline{𝐐})2,s;1,l|1,\lambda _{ex}\overline{𝐐}[ϵ^{}(\lambda _{q\overline{q}})ϵ^{}(\lambda _g)]ϵ(\lambda _{ex}),`$ (5.32)
and the action of the rotation matrix $`D^{(1)}`$ on the gluon states results in replacing $`ϵ^i(\lambda _g)`$ with $`ϵ_c^i(𝐐,\lambda _g)`$. The normalized hybrid wave functions are then given by:
$`\mathrm{\Psi }_{q\overline{q}g(S=0)}^{\lambda _{ex}}=\sqrt{{\displaystyle \frac{3}{8\pi }}}{\displaystyle \underset{\lambda _{q\overline{q}}}{}}\mathrm{\Psi }_{q\overline{q}}^{\lambda _{q\overline{q}}}(𝐪,𝐥_{q\overline{q}}=𝐐,\lambda _q,\lambda _{\overline{q}})[ϵ^{}(\lambda _{q\overline{q}})ϵ_c^{}(𝐐,\lambda _g)][\overline{𝐐}ϵ(\lambda _{ex})],`$
$`\mathrm{\Psi }_{q\overline{q}g(S=1)}^{\lambda _{ex}}=\sqrt{{\displaystyle \frac{3}{8\pi }}}{\displaystyle \underset{\lambda _{q\overline{q}}}{}}\mathrm{\Psi }_{q\overline{q}}^{\lambda _{q\overline{q}}}(𝐪,𝐥_{q\overline{q}}=𝐐,\lambda _q,\lambda _{\overline{q}})[ϵ^{}(\lambda _{q\overline{q}})\times ϵ_c^{}(𝐐,\lambda _g)][\overline{𝐐}\times ϵ(\lambda _{ex})],`$
$`\mathrm{\Psi }_{q\overline{q}g(S=2)}^{\lambda _{ex}}=\sqrt{{\displaystyle \frac{27}{104\pi }}}{\displaystyle \underset{\lambda _{q\overline{q}}}{}}\mathrm{\Psi }_{q\overline{q}}^{\lambda _{q\overline{q}}}(𝐪,𝐥_{q\overline{q}}=𝐐,\lambda _q,\lambda _{\overline{q}})`$
$`\times \overline{𝐐}[ϵ^{}(\lambda _{q\overline{q}})ϵ_c^{}(𝐐,\lambda _g)]ϵ(\lambda _{ex}),`$ (5.33)
where
$$(AB)_{ij}=A_iB_j+A_jB_i\frac{2}{3}\delta _{ij}(𝐀𝐁).$$
(5.34)
Writing the $`q\overline{q}`$ spin wave function more explicitly in terms of the quark momenta $`𝐩_q`$ and $`𝐩_{\overline{q}}`$ gives
$`\mathrm{\Psi }_{q\overline{q}g(S)}^{\lambda _{ex}}(𝐩_q,𝐩_{\overline{q}},\lambda _q,\lambda _{\overline{q}},\lambda _g)={\displaystyle \frac{1}{\sqrt{2}m_{q\overline{q}}}}\overline{u}(𝐩_q,\lambda _q)\left[\gamma ^\mu {\displaystyle \frac{p_q^\mu p_{\overline{q}}^\mu }{m_{q\overline{q}}+2m}}\right]v(𝐩_{\overline{q}},\lambda _{\overline{q}})`$
$`\times \psi _{\mu (S)}(𝐩_q𝐩_{\overline{q}},\lambda _g,\lambda _{ex}),`$ (5.35)
where the gluon terms are respectively:
$`\psi _{\mu (S=0)}(𝐐,\lambda _g,\lambda _{ex})=\sqrt{{\displaystyle \frac{3}{8\pi }}}ϵ_{c\mu }^{}(𝐐,\lambda _g)\overline{Q}^lϵ^l(\lambda _{ex}),`$
$`\psi _{\mu (S=1)}(𝐐,\lambda _g,\lambda _{ex})=\sqrt{{\displaystyle \frac{3}{8\pi }}}\left[g_\mu ^k{\displaystyle \frac{K_\mu K^k}{m_{q\overline{q}}(E_{q\overline{q}}+m_{q\overline{q}})}}\right]ϵ_c^l(𝐐,\lambda _g)\overline{Q}^kϵ^l(\lambda _{ex}),`$
$`\psi _{\mu (S=2)}(𝐐,\lambda _g,\lambda _{ex})={\displaystyle \frac{3}{\sqrt{13}}}\left(\psi _{\mu (S=1)}(𝐐,\lambda _g,\lambda _{ex}){\displaystyle \frac{2}{3}}\psi _{\mu (S=0)}(𝐐,\lambda _g,\lambda _{ex})\right),`$ (5.36)
and
$`m_{q\overline{q}}=m_{q\overline{q}}(𝐩_q,𝐩_{\overline{q}}),E_{q\overline{q}}=E(m_{q\overline{q}},𝐐),`$
$`𝐊=𝐐,K^0=E_{q\overline{q}}+m_{q\overline{q}},ϵ_c^0(𝐐,\lambda _g)=0.`$ (5.37)
The loss of linear independence between all three functions $`\psi _{\mu (S)}`$ came from the replacement of $`ϵ^i(\lambda _g)`$ with $`ϵ_c^i(𝐐,\lambda _g)`$, which is perpendicular to the momentum vector $`𝐐`$.
The spin wave functions for other hybrid mesons can be constructed in similar fashion. In particular, in Chapter 7 we will describe the $`\rho `$ and $`b_1`$ mesons as gluonic bound states, and construct corresponding wave functions.
## Chapter 6 Meson and hybrid states
In the preceding chapter we constructed the spin wave functions for normal and hybrid mesons assuming that quarks do not interact, so each spin index can transform separately under a Wigner rotation. The interaction between a quark and an antiquark enters through the Hamiltonian $`H=P^0`$ and the boost generators of the Poincaré group $`M^{0i}`$. It is possible to produce models of interaction for a fixed number of constituents that preserve the Poincaré algebra for noninteracting particles following the prescription of Bakamjian and Thomas, as we discussed in Chapter 4. Unfortunately, such a construction does not guarantee that physical observables such as current matrix elements or decay amplitudes will be relativistically covariant. Thus we must deal with phenomenological models of the quark dynamics, and in this chapter we will follow the common practice of employing a simple parametrization of the orbital wave functions.
### 6.1 Mesons as $`q\overline{q}`$ bound states
Unitary representations of noncompact groups are infinite-dimensional . The rotation group is compact, because rotating by the angle $`2\pi `$ (or $`4\pi `$ for spinors) returns the transformed quantity back to the original state. However, this is not the case for boosts and therefore they do not form a compact group. This is reflected in the fact that the spinor representation of the Lorentz group (4.45) is not unitary.
In quantum mechanics we are only interested in a unitary representation of a symmetry group, because the transition probabilities between states do not depend on the choice of a frame of reference. The problem of non-unitarity of the Lorentz group is solved by introducing the Fock space in which states are described by kets $`|𝐩,\lambda `$ with momentum $`𝐩`$ and spin $`\lambda `$. This representation is infinite-dimensional because the spectrum of values of $`𝐩`$ is continuous, and thus it is unitary. It is also irreducible, because the states have well-defined values of mass $`m`$ and spin $`s`$. In this section we will construct states for all mesons whose spin wave functions we have built in Chapter 5.
The $`\pi (I=1)`$ and $`\eta (I=0)`$ states ($`J^{PC}=0^+`$), characterized by momentum $`𝐏`$ and spin $`\lambda _{q\overline{q}}`$, are constructed in terms of the annihilation and creation operators:
$`|0^+(𝐏,I,I_3)={\displaystyle \underset{\lambda ,c,f}{}}{\displaystyle \frac{d^3𝐩_q}{(2\pi )^32E(m,𝐩_q)}\frac{d^3𝐩_{\overline{q}}}{(2\pi )^32E(m,𝐩_{\overline{q}})}2(E(m,𝐩_q)+E(m,𝐩_{\overline{q}}))}`$
$`\times (2\pi )^3\delta ^3(𝐩_q+𝐩_{\overline{q}}𝐏){\displaystyle \frac{1}{\sqrt{3}}}\delta _{c_qc_{\overline{q}}}{\displaystyle \frac{1}{2}},f_q;{\displaystyle \frac{1}{2}},f_{\overline{q}}|I,I_3\mathrm{\Psi }_{q\overline{q}}(𝐩_q,𝐩_{\overline{q}},\lambda _q,\lambda _{\overline{q}})`$
$`\times {\displaystyle \frac{1}{N(P)}}\psi _L(m_{q\overline{q}}(𝐩_q,𝐩_{\overline{q}})/\mu )b_{𝐩_q\lambda _qf_qc_q}^{}d_{𝐩_{\overline{q}}\lambda _{\overline{q}}f_{\overline{q}}c_{\overline{q}}}^{}|0,`$ (6.1)
where the operators satisfy the anticommutation relations:
$`\{b_{𝐩\lambda fc},b_{𝐩^{}\lambda ^{}f^{}c^{}}^{}\}=\{d_{𝐩\lambda fc},d_{𝐩^{}\lambda ^{}f^{}c^{}}^{}\}=(2\pi )^32E(m,𝐩)\delta ^3(𝐩𝐩^{})\delta _{\lambda \lambda ^{}}\delta _{ff^{}}\delta _{cc^{}},`$
$`\{b_{𝐩\lambda fc},b_{𝐩^{}\lambda ^{}f^{}c^{}}\}=\{d_{𝐩\lambda fc},d_{𝐩^{}\lambda ^{}f^{}c^{}}\}=\{b_{𝐩\lambda fc}^{},b_{𝐩^{}\lambda ^{}f^{}c^{}}^{}\}=\{d_{𝐩\lambda fc}^{},d_{𝐩^{}\lambda ^{}f^{}c^{}}^{}\}=0,`$
$`\{b_{𝐩\lambda fc},d_{𝐩^{}\lambda ^{}f^{}c^{}}\}=\{b_{𝐩\lambda fc},d_{𝐩^{}\lambda ^{}f^{}c^{}}^{}\}=\{b_{𝐩\lambda fc}^{},d_{𝐩^{}\lambda ^{}f^{}c^{}}\}=\{b_{𝐩\lambda fc}^{},d_{𝐩^{}\lambda ^{}f^{}c^{}}^{}\}=0.`$ (6.2)
In the above, $`\mathrm{\Psi }_{q\overline{q}}`$ represents the spin-0 wave function (5.15), written explicitly in terms of the momenta $`𝐩_q`$ and $`𝐩_{\overline{q}}`$ instead of the relativistic relative momentum $`𝐪`$ and the center-of-mass momentum $`𝐏=𝐥_{q\overline{q}}`$ (Eq. (5.11) with $`𝐥_q=𝐩_q`$ and Eq. (5.12) with $`𝐥_{\overline{q}}=𝐩_{\overline{q}}`$). The third component of isospin, flavor and color are respectively denoted by $`I_3`$, $`f`$ and $`c`$. The factor $`\delta _{c_qc_{\overline{q}}}`$ guarantees that the meson state is colorless.
The orbital wave function $`\psi _L`$ results from the strong and electroweak interaction between quarks that leads to a bound state (meson). Such a function depends on momenta only through the invariant mass of a quark-antiquark pair (5.18). Normalization constants are denoted by $`N`$ (with $`P=|𝐏|`$) and the $`\mu `$’s are free parameters, being scalar functions of meson quantum numbers. Finally, the isospin Clebsch-Gordan coefficient can be written as
$`{\displaystyle \frac{1}{2}},f_q;{\displaystyle \frac{1}{2}},f_{\overline{q}}|1,I_3={\displaystyle \frac{1}{\sqrt{2}}}\sigma _{f_qf_{\overline{q}}}^iϵ^i(I_3),`$
$`{\displaystyle \frac{1}{2}},f_q;{\displaystyle \frac{1}{2}},f_{\overline{q}}|0,0={\displaystyle \frac{1}{\sqrt{2}}}\delta _{f_qf_{\overline{q}}},`$ (6.3)
where $`f=1`$ for the $`u`$ quark (antiquark) and $`f=2`$ for the $`d`$. The flavor structure of the $`\eta `$ state (as well as other isospin zero mesons) was chosen as a linear combination $`\frac{1}{\sqrt{2}}(|u\overline{u}+|d\overline{d})`$, although in general these states are linear combinations $`\mathrm{cos}(\varphi )[|u\overline{u}+|d\overline{d}]/\sqrt{2}+\mathrm{sin}(\varphi )|s\overline{s}`$. The $`|s\overline{s}`$ does not contribute to the $`\pi _1`$ decay amplitude and therefore may be neglected in calculations, provided this amplitude is multiplied by a factor $`\mathrm{cos}(\varphi )`$.
Similarly the $`\rho (I=1)`$ and $`\omega ,\varphi (I=0)`$ states ($`J^{PC}=1^{}`$) are given by
$`|1^{}(𝐏,I,I_3,\lambda )={\displaystyle \underset{\lambda ,c,f}{}}{\displaystyle \frac{d^3𝐩_q}{(2\pi )^32E(m,𝐩_q)}\frac{d^3𝐩_{\overline{q}}}{(2\pi )^32E(m,𝐩_{\overline{q}})}2(E(m,𝐩_q)+E(m,𝐩_{\overline{q}}))}`$
$`\times (2\pi )^3\delta ^3(𝐩_q+𝐩_{\overline{q}}𝐏){\displaystyle \frac{1}{\sqrt{3}}}\delta _{c_qc_{\overline{q}}}{\displaystyle \frac{1}{2}},f_q;{\displaystyle \frac{1}{2}},f_{\overline{q}}|I,I_3\mathrm{\Psi }_{q\overline{q}}^\lambda (𝐩_q,𝐩_{\overline{q}},\lambda _q,\lambda _{\overline{q}})`$
$`\times {\displaystyle \frac{1}{N(P)}}\psi _L(m_{q\overline{q}}(𝐩_q,𝐩_{\overline{q}})/\mu )b_{𝐩_q\lambda _qf_qc_q}^{}d_{𝐩_{\overline{q}}\lambda _{\overline{q}}f_{\overline{q}}c_{\overline{q}}}^{}|0,`$ (6.4)
where $`\mathrm{\Psi }_{q\overline{q}}^\lambda `$ denotes the spin-1 wave function (5.16).
The $`b_1(I=1)`$ and $`h_1(I=0)`$ mesons ($`J^{PC}=1^+`$) have an additional orbital angular momentum $`L=1`$ represented by $`Y_{1l}(\overline{𝐪})`$, where $`𝐪`$ is the momentum of the constituent quark in the meson rest frame
$$𝐪(𝐩_q,𝐏)=\mathrm{\Lambda }(𝐏0)𝐩_q=𝐩_q\frac{E(m,𝐩_q)𝐏}{m_{q\overline{q}}}+\frac{(𝐏𝐩_q)𝐏}{m_{q\overline{q}}(E(m_{q\overline{q}},𝐏)+m_{q\overline{q}})},$$
(6.5)
with
$$m_{q\overline{q}}=m_{q\overline{q}}(𝐩_q,𝐏𝐩_q),\overline{𝐪}=𝐪/|𝐪|.$$
(6.6)
The corresponding states are given by (6.4), although the spin wave function $`\mathrm{\Psi }_{q\overline{q}}^\lambda `$ is given now by (5.22). Finally, the $`a(I=1)`$ and $`f(I=0)`$ states ($`J^{PC}=0,1,2^{++}`$) are described by (6.4) with the spin wave function (5.23).
The states are normalized
$$𝐏,\lambda ,I_3|𝐏^{},\lambda ^{},I_3^{}=(2\pi )^32E(m_M,𝐏)\delta ^3(𝐏𝐏^{})\delta _{\lambda \lambda ^{}}\delta _{I_3I_3^{}},$$
(6.7)
where $`m_M`$ is the meson mass. That fixes the normalization constants,
$`N_M^2(P)=(2E(m_M,𝐏))^1{\displaystyle \frac{d^3𝐤}{(2\pi )^3}\frac{(E(m,𝐤)+E(m,𝐏𝐤))^2}{E(m,𝐤)E(m,𝐏𝐤)}[Y_{L0}(\overline{𝐪}(𝐤,𝐏))]^2}`$
$`\times [\psi _L(m_{q\overline{q}}(𝐤,𝐏𝐤)/\mu _M)]^2,`$ (6.8)
with $`𝐪`$ given in (6.5) and $`L`$ being the orbital angular momentum of the meson. Without loss of generality we have taken $`𝐏=P𝐞_z`$.
The orbital angular momentum wave function for a meson depends on the potential between a quark and an antiquark. An explicit form of such a potential is not known exactly and such a function must be modeled. Because of Lorentz invariance it may depend on momenta only through the invariant mass of a $`q\overline{q}`$ pair. Moreover, it must tend to zero for large momenta fast enough to make the amplitude convergent. The simplest choice is the gaussian function
$$\psi _L(m_{q\overline{q}}(𝐩_q,𝐩_{\overline{q}})/\mu )=e^{m_{q\overline{q}}^2(𝐩_q,𝐩_{\overline{q}})/8\mu ^2}.$$
(6.9)
The integrals in the decay amplitudes are not elementary and must be computed numerically. In the nonrelativistic limit (for a large $`m`$), however, they can be expressed in terms of the error function.
The free parameters of the model presented are: the quark masses $`m`$, the size parameters of the orbital wave functions $`\mu `$ and the strong coupling $`g`$. The pion decay constant $`f_\pi `$ and the elastic form factor $`F_\pi `$, defined by
$$0|A^{\mu ,i}(\mathrm{𝟎})|\pi ^k(𝐩)=f_\pi p^\mu \delta _{ik},$$
(6.10)
and
$$\pi ^i(𝐩^{})|V^{\mu ,j}(\mathrm{𝟎})|\pi ^k(𝐩)=F_\pi (p^\mu +p^\mu )iϵ_{ijk},$$
(6.11)
are used to constrain the $`\pi `$ wave function parameters (with the $`\pi `$ state given by (6.1)). The axial and the vector currents are defined by
$$A^{\mu ,i}(\mathrm{𝟎})=\overline{\psi }_{cf}(\mathrm{𝟎})\gamma ^\mu \gamma _5\frac{\sigma ^i}{2}\psi _{cf},$$
(6.12)
and
$$V^{\mu ,j}(\mathrm{𝟎})=\overline{\psi }_{cf}(\mathrm{𝟎})\gamma ^\mu \frac{\sigma ^j}{2}\psi _{cf},$$
(6.13)
with $`\psi _{cf}(𝐱)`$ given in (7.2). By virtue of Lorentz invariance $`f_\pi `$ is a constant, whereas $`F_\pi `$ is a function of $`Q^2=(𝐩𝐩^{})^2`$.
As mentioned previously, it is not possible to construct the wave functions with a fixed number of constituents in a Lorentz covariant way. Thus the current matrix elements are expected not to be exactly Lorentz covariant. This will be reflected, for example, in different values of $`f_\pi `$ obtained from spatial and time components of the axial current (rotational symmetry is not broken). Even if we replaced the factor $`E(m_M,𝐏)`$ in (6.7) by $`1`$, it would be very difficult to find the generators of the Poincare group that satisfy the commutation relations. Thus, our model with the exponential orbital wave functions will not be exactly covariant. The resulting form factors will depend on the frame of reference. In order to obtain $`F_\pi (Q^2=0)=1`$, one typically employs the time component $`\mu =0`$ and works in the Breit frame of reference. In this case we obtain
$$f_\pi (P)=\frac{\sqrt{3}m}{N_\pi (P)E(m_\pi ,𝐏)}\frac{d^3𝐩}{(2\pi )^3}\frac{(p^0+q^0)^2}{p^0q^0m_{q\overline{q}}}e^{\frac{m_{q\overline{q}}^2}{8\mu _\pi ^2}},$$
(6.14)
and
$`F_\pi (𝐏,𝐏^{})={\displaystyle \frac{1}{N_\pi (P)N_\pi (P^{})(E(m_\pi ,𝐏)+E(m_\pi ,𝐏^{}))}}{\displaystyle \frac{d^3𝐩}{(2\pi )^3}\frac{(p^0+q^0)(p^0+r^0)}{m_{q\overline{q}}m_{q\overline{q}}^{}p^0q^0r^0}}`$
$`\times [(pq)r^0+(pr)q^0(qr)p^0+m^2(p^0+q^0+r^0)]e^{\frac{m_{q\overline{q}}^2+m_{q\overline{q}}^2}{8\mu _\pi ^2}},`$ (6.15)
where:
$`𝐪=𝐏𝐩,𝐫=𝐏^{}𝐩,p^0=E(m,𝐩),q^0=E(m,𝐪),r^0=E(m,𝐫),`$
$`m_{q\overline{q}}(𝐩,𝐪)=[(E(m,𝐩)+E(m,𝐪))^2(𝐩+𝐪)^2]^{1/2},`$
$`m_{q\overline{q}}=m_{q\overline{q}}(𝐩,𝐪),m_{q\overline{q}}^{}=m_{q\overline{q}}(𝐩,𝐫),`$ (6.16)
and the pion normalization constant is given in (6.8). For other light unflavored mesons there are not enough experimental data to constrain their parameters $`\mu `$. However, they are expected to be on the same order as $`\mu _\pi `$.
By taking $`m`$ large as compared to the $`\mu `$’s and $`P_0`$, one obtains the nonrelativistic limit in which quarks are heavy. Their motion may be described by nonrelativistic quantum mechanics and, as we demonstrated at the end of Chapter 5, spin does not change via Wigner rotations. Therefore, all spin wave functions are just described by Clebsch-Gordan coefficients and spherical harmonics, and the spin factors in the decay amplitudes reduce to traces of products of Pauli matrices. All energy terms $`E(m,𝐩)`$ tend to $`m`$, whereas the invariant masses (5.18) and (6.20) tend to $`2m`$ and $`2m+E(m_g,𝐐)`$, respectively. In the orbital wave functions, however, we must keep the next leading terms depending on momenta, otherwise the amplitude would become divergent:
$`m_{q\overline{q}}(𝐩_q,𝐩_{\overline{q}})2m+{\displaystyle \frac{(𝐩_q𝐩_{\overline{q}})^2}{4m}},`$
$`m_{q\overline{q}g}(𝐩_q,𝐩_{\overline{q}},𝐐)2m+m_g+{\displaystyle \frac{𝐩_q^2+𝐩_2^{\overline{q}}}{2m}}+E(m_g,𝐐).`$ (6.17)
In the above we have $`𝐐=𝐩_q𝐩_{\overline{q}}`$ because the $`q\overline{q}g`$ state is at rest. The normalization constants are given in this limit by
$$N_M^2(P)=2E^1(m_M,𝐏)\frac{d^3𝐤}{(2\pi )^3}[Y_{L0}(\overline{𝐪}(𝐤,𝐏))]^2[\psi _L(m_{q\overline{q}}(𝐤,𝐏𝐤)/\mu _M)]^2,$$
(6.18)
where $`L`$ is the orbital angular momentum of a meson. For the decay amplitudes we will not derive the nonrelativistic formulae from the beginning, but instead, we will go with $`m`$ to very large values and keep only the leading terms.
### 6.2 Exotic mesons as $`q\overline{q}g`$ bound states
In our model a hybrid is regarded as a bound state of a quark, an antiquark and a gluon. Therefore, we can construct it in terms of the annihilation and creation operators. The $`\pi _1`$ state $`I^G(J^{PC})=1^{}(1^+)`$ in its rest frame is given by
$`|ex(I_3,\lambda _{ex})={\displaystyle \underset{\lambda ,c,f}{}}{\displaystyle \frac{1}{N_{ex}}}{\displaystyle \frac{d^3𝐩_q}{(2\pi )^32E(m,𝐩_q)}\frac{d^3𝐩_{\overline{q}}}{(2\pi )^32E(m,𝐩_{\overline{q}})}\frac{d^3𝐐}{(2\pi )^32E(m_g,𝐐)}}`$
$`\times \mathrm{\hspace{0.17em}2}(E(m,𝐩_q)+E(m,𝐩_{\overline{q}})+E(m_g,𝐐))(2\pi )^3\delta ^3(𝐩_q+𝐩_{\overline{q}}+𝐐)`$
$`\times \mathrm{\Psi }_{q\overline{q}g}^{\lambda _{ex}}(𝐩_q,𝐩_{\overline{q}},\lambda _q,\lambda _{\overline{q}},\lambda _g)\psi _L^{}(m_{q\overline{q}}(𝐩_q,𝐩_{\overline{q}})/\mu _{ex},m_{q\overline{q}g}(𝐩_q,𝐩_{\overline{q}},𝐐)/\mu _{ex^{}})`$
$`\times {\displaystyle \frac{1}{2}}\lambda _{c_qc_{\overline{q}}}^{c_g}{\displaystyle \frac{1}{2}},f_q;{\displaystyle \frac{1}{2}},f_{\overline{q}}|I,I_3b_{𝐩_q\lambda _qf_qc_q}^{}d_{𝐩_{\overline{q}}\lambda _{\overline{q}}f_{\overline{q}}c_{\overline{q}}}^{}a_{𝐐\lambda _gc_g}^{}|0,`$ (6.19)
where the spin wave function $`\mathrm{\Psi }_{q\overline{q}g}`$ was given in (5.35) for $`S=0,1,2`$. The orbital wave function $`\psi _L^{}`$ depends only on the invariant mass of a quark-antiquark pair $`m_{q\overline{q}}`$ and the invariant mass of a 3-body system,
$$m_{q\overline{q}g}(𝐩_q,𝐩_{\overline{q}},𝐐)=E(m,𝐩_q)+E(m,𝐩_{\overline{q}})+E(m_g,𝐐).$$
(6.20)
Here $`m_g`$ is the dynamical mass of a gluon in the Coulomb gauge (arising from the strong interaction with virtual particles), and $`\lambda _{c_qc_{\overline{q}}}^{c_g}`$ are the SU(3) Gell-Mann matrices. They guarantee that a hybrid meson is colorless. The gluon operators satisfy the commutation relations:
$`[a_{𝐩\lambda c},a_{𝐩^{}\lambda ^{}c^{}}^{}]=(2\pi )^32E(m_g,𝐩)\delta ^3(𝐩𝐩^{})\delta _{\lambda \lambda ^{}}\delta _{cc^{}},`$
$`[a_{𝐩\lambda c},a_{𝐩^{}\lambda ^{}c^{}}]=[a_{𝐩\lambda c}^{},a_{𝐩^{}\lambda ^{}c^{}}^{}]=0.`$ (6.21)
The normalization (6.7) leads to
$`N_{ex}^2={\displaystyle \frac{3}{4\pi }}(2m_{ex})^1{\displaystyle \frac{d^3𝐤}{(2\pi )^3}\frac{d^3𝐥}{(2\pi )^3}\frac{(E(m,𝐤)+E(m,𝐥)+E(m_g,𝐤𝐥))^2}{2E(m,𝐤)E(m,𝐥)E(m_g,𝐤𝐥)}\frac{(k_z+l_z)^2}{(𝐤+𝐥)^2}}`$
$`\times [\psi _L^{}(m_{q\overline{q}}(𝐤,𝐥)/\mu _{ex},m_{q\overline{q}g}(𝐤,𝐥,𝐤𝐥)/\mu _{ex^{}})]^2.`$ (6.22)
Hybrid mesons with other quantum numbers can be constructed in similar fashion.
The covariant orbital wave function of the $`\pi _1`$ may depend only on the invariant masses $`m_{q\overline{q}}`$ and $`m_{q\overline{q}g}`$. A natural choice is a product of two gaussian functions,
$$\psi _L^{}(m_{q\overline{q}}(𝐩_q,𝐩_{\overline{q}})/\mu _{ex},m_{q\overline{q}g}(𝐩_q,𝐩_{\overline{q}},𝐐)/\mu _{ex}^{})=e^{m_{q\overline{q}}^2(𝐩_q,𝐩_{\overline{q}})/8\mu _{ex}^2}e^{m_{q\overline{q}g}^2(𝐩_q,𝐩_{\overline{q}},𝐐)/8\mu _{ex}^2}.$$
(6.23)
In the nonrelativistic limit the $`\pi _1`$ normalization constant is given by
$$N_{ex}^2=\frac{3}{4\pi }\frac{(2m+m_g)^2}{4m^2m_gm_{ex}}\frac{d^3𝐤}{(2\pi )^3}\frac{d^3𝐥}{(2\pi )^3}\frac{(k_z+l_z)^2}{(𝐤+𝐥)^2}[\psi _L^{}(m_{q\overline{q}}(𝐤,𝐥)/\mu _{ex},m_{q\overline{q}g}(𝐤,𝐥,𝐤𝐥)/\mu _{ex^{}})]^2.$$
(6.24)
## Chapter 7 Decay of normal mesons
The common approach to normal meson decays is based on the $`{}_{}{}^{3}P_{0}^{}`$ model where $`q\overline{q}`$ pair creation is described by an effective operator that creates this pair from the vacuum in the presence of the normal $`q\overline{q}`$ component of the decaying meson, as shown in Fig. 7.1
In the QCD-motivated Coulomb gauge, however, the decay of a normal meson is expected to proceed via mixing of the $`q\overline{q}`$ state with the $`q\overline{q}g`$ hybrid component followed by gluon dissociation to a $`q\overline{q}`$ pair, as shown in Fig. 7.2. The dashed line represents the confining non-abelian Coulomb potential. The hybrid component of the wave function is obtained by integrating the $`q\overline{q}`$ wave function over the amplitude of the transverse gluon emission from the Coulomb line . The quantum numbers $`P,C`$ of such a $`q\overline{q}`$ state are determined from the corresponding conservation laws, whereas its spin may have more than one value (denoted in this chapter by $`J`$).
We will be interested in estimating the size of relativistic effects in meson decays, and not in giving the absolute width predictions. Therefore, we can make calculations for each value of $`J`$ separately. The relative contributions from various $`J`$ and the total width may be obtained from the above gluon emission amplitude. Since the quark pair is emitted in the $`S=1`$, $`L=0`$ state, this decay mechanism is also referred to as the $`{}_{}{}^{3}S_{1}^{}`$ model.
The Hamiltonian $`H`$ of gluon dissociation is equal to
$$H=\underset{c,f}{}d^3𝐱\overline{\psi }_{c_1f_1}(𝐱)(g\gamma 𝐀^{c_g}(𝐱))\psi _{c_2f_2}(𝐱)\delta _{f_1f_2}\frac{1}{2}\lambda _{c_1c_2}^{c_g}.$$
(7.1)
In the constituent basis used here, the single-particle quark and antiquark wave functions correspond to the states of massive particles with a relativistic dispersion relation, in which the running quark mass is approximated by a constant constituent mass $`m`$,
$$\psi _{cf}(𝐱)=\underset{\lambda }{}\frac{d^3𝐤}{(2\pi )^32E(m,𝐤)}[u(𝐤,\lambda )b_{𝐤\lambda cf}+v(𝐤,\lambda )d_{𝐤\lambda cf}^{}]e^{i𝐤𝐱}.$$
(7.2)
Similarly, the gluon field $`𝐀^{c_g}`$ is expanded in a basis of transverse polarization vectors, with a single-particle wave function characterizing a state with mass $`m_g`$,
$$𝐀^{c_g}(𝐱)=\underset{\lambda }{}\frac{d^3𝐤}{(2\pi )^32E(m_g,𝐤)}[ϵ_c(𝐤,\lambda )a_{𝐤\lambda }^{c_g}+ϵ_c^{}(𝐤,\lambda )a_{𝐤\lambda }^{c_g}]e^{i𝐤𝐱}.$$
(7.3)
Here $`g`$ is the strong coupling constant. The Hamiltonian part contributing to the decay amplitude is
$`H={\displaystyle \underset{\lambda ,c,f}{}}{\displaystyle \frac{d^3𝐤_1}{(2\pi )^32E(m,𝐤_1)}\frac{d^3𝐤_2}{(2\pi )^32E(m,𝐤_2)}\frac{d^3𝐤}{(2\pi )^32E(m_g,𝐤)}(2\pi )^3\delta ^3(𝐤𝐤_1+𝐤_2)\delta _{f_1f_2}}`$
$`\times {\displaystyle \frac{1}{2}}\lambda _{c_1c_2}^{c_g}\overline{u}(𝐤_1,\lambda _1)(g\gamma ^jϵ_c^j(𝐤,\lambda _g))v(𝐤_2,\lambda _2)b_{𝐤_1\lambda _1c_1f_1}^{}d_{𝐤_2\lambda _2c_2f_2}^{}a_{𝐤\lambda _g}^{c_g}.`$ (7.4)
In this chapter we will study the decays of the $`\rho `$ and $`b_1`$ since they are dominated (almost $`100\%`$) by a single mode, and their widths are well-known from experiment. Therefore they can be used to test the model presented in this work. Because we are interested in widths, all calculations will be done in the rest frame of a decaying meson.
### 7.1 Decay $`\rho 2\pi `$
We will start from the $`{}_{}{}^{3}P_{0}^{}`$ Hamiltonian:
$$H=\mathrm{\Lambda }\underset{c,f}{}d^3𝐱\overline{\psi }_{c_1f_1}(𝐱)\psi _{c_2f_2}(𝐱)\delta _{f_1f_2}\delta _{c_1c_2},$$
(7.5)
with $`\psi `$ defined in (7.2) and $`\mathrm{\Lambda }`$ being a mass scale which can be fixed by the absolute decay width and is expected to be of the order of the average quark momentum. The amplitude of this mode is determined by the matrix element
$$\pi (𝐏),\pi (𝐏)|H|\rho .$$
(7.6)
The corresponding states are given by Eqs. (6.1) and (6.4).
In this matrix element we have two operators $`b`$ and two $`d`$ coming from the outgoing meson states, whereas the decaying meson state provides one $`b^{}`$ and one $`d^{}`$. Moreover, we get one $`b^{}`$ and one $`d^{}`$ from the Hamiltonian. Thus each pair $`b,b^{}`$ and $`d,d^{}`$ appears twice and Wick’s rearrangement leads to two nonzero terms. The anticommutation relations (6.2) give Dirac delta functions that guarantee the conservation of momentum and Kronecker deltas acting in flavor and color space. This simplifies the integration over momenta and reduces summations over flavor and color indices to calculating traces of corresponding matrix products.
The two terms in the above matrix element will be equal after integrating (up to a sign) because of symmetry, so it is enough to deal with only one and multiply the final expression for the amplitude by 2. A short proof of this statement will be given for the $`\pi _1`$ decay in the next chapter. Summing over color gives a factor $`1/\sqrt{3}`$, whereas summing over flavor leads to a trace of a product of Pauli matrices appearing in the isospin factors (6.3),
$$2^{3/2}Tr(\sigma _i^\rho \sigma _j^\pi \sigma _k^\pi )ϵ_i(I_3^\rho )ϵ_j^{}(I_3^\pi )ϵ_k^{}(I_3^\pi ).$$
(7.7)
For all possible isospin channels the flavor factor is equal to $`\pm 1/\sqrt{2}`$. For spin we obtain
$$W^{\lambda _\rho }(𝐩,𝐫,𝐤,𝐥)=\frac{Tr\left[(\mathit{}+m)(\mathit{}+m)\left(\gamma ^i\frac{p^il^i}{m_{q\overline{q}}(𝐩,𝐥)+2m}\right)(\mathit{}m)(\mathit{}m)\right]ϵ^i(\lambda _\rho )}{2^{3/2}m_{q\overline{q}}(𝐩,𝐤)m_{q\overline{q}}(𝐫,𝐥)m_{q\overline{q}}(𝐩,𝐥)},$$
(7.8)
where $`𝐩`$ and $`𝐥`$ denote respectively the momenta of a quark and an antiquark in $`\rho `$, and $`𝐫`$ and $`𝐤`$ denote respectively the momenta of a quark and an antiquark created from the vacuum, as shown in Fig. 7.1. We assume that all quarks are on-shell particles, i.e.:
$$p^0=E(m,𝐩),r^0=E(m,𝐫),k^0=E(m,𝐤),l^0=E(m,𝐥).$$
(7.9)
Integration over momenta gives $`(2\pi )^3\delta ^3(\mathrm{𝟎})A(𝐏)`$, where $`A`$ denotes the amplitude of this decay,
$`A(𝐏,\lambda _\rho )={\displaystyle \frac{2\mathrm{\Lambda }}{\sqrt{6}N_\pi ^2(P)N_\rho (0)}}{\displaystyle \frac{d^3𝐩}{(2\pi )^3}\frac{(E(m,𝐩)+E(m,𝐏𝐩))^2}{E(m,𝐩)E^2(m,𝐏𝐩)}}`$
$`\times W^{\lambda _\rho }(𝐩,𝐩𝐏,𝐏𝐩,𝐩)[\psi _L(m_{q\overline{q}}(𝐩,𝐏𝐩)/\mu _\pi )]^2`$
$`\times \psi _L((E(m,𝐩)+E(m,𝐏𝐩))/\mu _\rho ).`$ (7.10)
The corresponding width is obtained from
$$d\mathrm{\Gamma }=\frac{1}{32\pi ^2}\frac{P_0|A(𝐏_0)|^2}{m_\rho ^2}d\mathrm{\Omega },$$
(7.11)
with $`𝐏_0`$ satisfying
$$P_0=\sqrt{\frac{m_\rho ^2}{4}m_\pi ^2}$$
(7.12)
and $`P_0=|𝐏_0|`$. The amplitude must be, according to the Wigner-Eckart theorem, of the form:
$$A(𝐏,\lambda _\rho )=a_1(P)Y_{1\lambda _\rho }(𝐏/P),$$
(7.13)
where $`a_1`$ is the P-wave partial amplitude. Thus
$$\mathrm{\Gamma }=\frac{P_0}{32\pi ^2m_\rho ^2}a_1^2(P_0).$$
(7.14)
Taking $`𝐏=Pe_z`$ and $`\lambda _\rho =0`$ gives
$$a_1(P)=\sqrt{\frac{4\pi }{3}}A(Pe_z,0),$$
(7.15)
therefore the width is given by
$$\mathrm{\Gamma }_{\rho 2\pi }^{(pwave)}=\frac{P_0}{24\pi m_\rho ^2}A^2(P_0e_z,0).$$
(7.16)
For large $`m`$ (in the nonrelativistic limit) the trace term tends to the value
$$\sqrt{2}(p^iP^i)ϵ^i(\lambda _\rho ),$$
(7.17)
and the amplitude (7.10) simplifies to
$$A(𝐏,\lambda _\rho )=\frac{8\mathrm{\Lambda }}{\sqrt{3}mN_\pi ^2(P)N_\rho (0)}\frac{d^3𝐩}{(2\pi )^3}(p^iP^i)ϵ^i(\lambda _\rho )[\psi _L^\pi ]^2\psi _L^\rho .$$
(7.18)
Here the normalization constants are given by (6.18), and in the orbital wave functions $`\psi _L`$ we need to expand the invariant masses and energies only up to terms quadratic in momenta.
Now we proceed to the $`{}_{}{}^{3}S_{1}^{}`$ model. For the $`\rho `$ meson the $`q\overline{q}g`$ component can be expanded in a basis of the $`a_0`$, $`a_1`$, $`a_2`$ wave functions, all having spin 1 and one unit of the orbital angular momentum between the quark and the antiquark (5.23), all coupled with a transverse gluon wave function to give the $`J^{PC}=1^{}`$ state. The wave functions for the $`\rho `$ are thus:
$$\mathrm{\Psi }_{q\overline{q}g(J)}^{\lambda _\rho }(\lambda _q,\lambda _{\overline{q}},\lambda _g)=\underset{\lambda _{q\overline{q}},\sigma =\pm 1}{}\mathrm{\Psi }_{q\overline{q}}^{J,\lambda _{q\overline{q}}}(𝐪,𝐥_{q\overline{q}}=𝐐,\lambda _q,\lambda _{\overline{q}})D_{\lambda _g\sigma }^{(1)}(\overline{𝐐})J,\lambda _{q\overline{q}};1,\sigma |1,\lambda _\rho .$$
(7.19)
where $`\mathrm{\Psi }_{q\overline{q}}^{J,\lambda _{q\overline{q}}}`$ are the $`a_0,a_1`$ and $`a_2`$ $`q\overline{q}`$ wave functions for $`J=0,1,2`$ respectively. The normalized wave functions for the $`\rho `$ are then given, similarly to those for the $`\pi _1`$ (5.33), by:
$`\mathrm{\Psi }_{q\overline{q}g(J=0)}^{\lambda _\rho }=\sqrt{{\displaystyle \frac{3}{8\pi }}}{\displaystyle \underset{\lambda _{q\overline{q}}}{}}\mathrm{\Psi }_{q\overline{q}}^{\lambda _{q\overline{q}}}(𝐪,𝐥_{q\overline{q}}=𝐐,\lambda _q,\lambda _{\overline{q}})[ϵ^{}(\lambda _{q\overline{q}})\overline{𝐪}][ϵ_c^{}(𝐐,\lambda _g)ϵ(\lambda _\rho )],`$
$`\mathrm{\Psi }_{q\overline{q}g(J=1)}^{\lambda _\rho }=\sqrt{{\displaystyle \frac{9}{32\pi }}}{\displaystyle \underset{\lambda _{q\overline{q}}}{}}\mathrm{\Psi }_{q\overline{q}}^{\lambda _{q\overline{q}}}(𝐪,𝐥_{q\overline{q}}=𝐐,\lambda _q,\lambda _{\overline{q}})[ϵ^{}(\lambda _{q\overline{q}})\times \overline{𝐪}][ϵ_c^{}(𝐐,\lambda _g)\times ϵ(\lambda _\rho )],`$
$`\mathrm{\Psi }_{q\overline{q}g(J=2)}^{\lambda _\rho }=\sqrt{{\displaystyle \frac{27}{160\pi }}}{\displaystyle \underset{\lambda _{q\overline{q}}}{}}\mathrm{\Psi }_{q\overline{q}}^{\lambda _{q\overline{q}}}(𝐪,𝐥_{q\overline{q}}=𝐐,\lambda _q,\lambda _{\overline{q}})`$
$`\times ϵ_c^{}(𝐐,\lambda _g)[ϵ^{}(\lambda _{q\overline{q}})\overline{𝐪}]ϵ(\lambda _\rho ),`$ (7.20)
where $`\mathrm{\Psi }_{q\overline{q}}^{\lambda _{q\overline{q}}}`$ is the spin-1 wave function (5.16). Writing this function more explicitly in terms of the quark momenta $`𝐩_q`$ and $`𝐩_{\overline{q}}`$ gives
$`\mathrm{\Psi }_{q\overline{q}g(J)}^{\lambda _\rho }(𝐩_q,𝐩_{\overline{q}},\lambda _q,\lambda _{\overline{q}},\lambda _g)={\displaystyle \frac{1}{\sqrt{2}m_{q\overline{q}}}}\overline{u}(𝐩_q,\lambda _q)\left[\gamma ^\mu {\displaystyle \frac{p_q^\mu p_{\overline{q}}^\mu }{m_{q\overline{q}}+2m}}\right]v(𝐩_{\overline{q}},\lambda _{\overline{q}})`$
$`\times \psi _{\mu (J)}(𝐩_q𝐩_{\overline{q}},\lambda _g,\lambda _\rho ),`$ (7.21)
where the gluon terms are respectively:
$`\psi _{\mu (J=0)}(𝐐,\lambda _g,\lambda _\rho )=\sqrt{{\displaystyle \frac{3}{8\pi }}}\left[g_\mu ^k{\displaystyle \frac{K_\mu K^k}{m_{q\overline{q}}(E_{q\overline{q}}+m_{q\overline{q}})}}\right](\delta ^{kl}\delta ^{mn})\overline{q}^lϵ_c^m(𝐐,\lambda _g)ϵ^n(\lambda _\rho ),`$
$`\psi _{\mu (J=1)}(𝐐,\lambda _g,\lambda _\rho )=\sqrt{{\displaystyle \frac{9}{32\pi }}}\left[g_\mu ^k{\displaystyle \frac{K_\mu K^k}{m_{q\overline{q}}(E_{q\overline{q}}+m_{q\overline{q}})}}\right](\delta ^{km}\delta ^{ln}\delta ^{kn}\delta ^{lm})\overline{q}^l`$
$`\times ϵ_c^m(𝐐,\lambda _g)ϵ^n(\lambda _\rho ),`$
$`\psi _{\mu (J=2)}(𝐐,\lambda _g,\lambda _\rho )=\sqrt{{\displaystyle \frac{27}{160\pi }}}\left[g_\mu ^k{\displaystyle \frac{K_\mu K^k}{m_{q\overline{q}}(E_{q\overline{q}}+m_{q\overline{q}})}}\right](\delta ^{km}\delta ^{ln}+\delta ^{kn}\delta ^{lm}{\displaystyle \frac{2}{3}}\delta ^{kl}\delta ^{mn})`$
$`\times \overline{q}^lϵ_c^m(𝐐,\lambda _g)ϵ^n(\lambda _\rho ),`$ (7.22)
and
$`m_{q\overline{q}}=m_{q\overline{q}}(𝐩_q,𝐩_{\overline{q}}),E_{q\overline{q}}=E(m_{q\overline{q}},𝐐),𝐊=𝐐,`$
$`K^0=E_{q\overline{q}}+m_{q\overline{q}},ϵ_c^0(𝐐,\lambda _g)=0.`$ (7.23)
As before, $`𝐪=𝐪(𝐩_q,𝐐)`$ denotes the quark momentum in the rest frame of the $`q\overline{q}`$ pair (6.5). The most general wave function will be given by a linear combination of the three components listed above, and the coefficients in this are provided by the $`q\overline{q}g`$ component mentioned at the beginning of this chapter.
The Hamiltonian matrix element leads again to two equal terms. Summation over flavor indices gives $`\pm 1/\sqrt{2}`$ as before, whereas for color one obtains
$$\frac{1}{12}\lambda _{bc}^a\lambda _{cb}^a=\frac{4}{3}.$$
(7.24)
Summation over spin gives
$$B_j^\mu \psi _{\mu j}^{(J)},$$
(7.25)
where
$$B^{\mu j}=\frac{Tr\left[(\mathit{}m)(\mathit{}m)\left(\gamma ^\mu +\frac{p^\mu l^\mu }{m_{q\overline{q}}(𝐩,𝐥)+2m}\right)(\mathit{}+m)(\mathit{}+m)\gamma ^j\right]}{2^{3/2}m_{q\overline{q}}(𝐩,𝐤)m_{q\overline{q}}(𝐫,𝐥)m_{q\overline{q}}(𝐩,𝐥)},$$
(7.26)
and
$$\psi _\mu ^{j(J)}(\lambda _\rho )=\underset{\lambda _g}{}\psi _{\mu (J)}(𝐐,\lambda _g,\lambda _\rho )ϵ_c^j(𝐐,\lambda _g).$$
(7.27)
The tensor $`B_j^\mu `$ corresponds to the first term contributing to the amplitude, and the notation is the same as in Fig. 7.2. The functions (7.27) are given by:
$`\psi _{\mu (J=0)}^j(𝐐,\lambda _g,\lambda _\rho )=\sqrt{{\displaystyle \frac{3}{8\pi }}}\left[g_\mu ^k{\displaystyle \frac{K_\mu K^k}{m_{q\overline{q}}(E_{q\overline{q}}+m_{q\overline{q}})}}\right]\overline{q}^l(\delta ^{jm}\overline{Q}^j\overline{Q}^m)ϵ^n(\lambda _\rho )(\delta ^{kl}\delta ^{mn}),`$
$`\psi _{\mu (J=1)}^j(𝐐,\lambda _g,\lambda _\rho )=\sqrt{{\displaystyle \frac{9}{32\pi }}}\left[g_\mu ^k{\displaystyle \frac{K_\mu K^k}{m_{q\overline{q}}(E_{q\overline{q}}+m_{q\overline{q}})}}\right]\overline{q}^l(\delta ^{jm}\overline{Q}^j\overline{Q}^m)ϵ^n(\lambda _\rho )`$
$`\times (\delta ^{km}\delta ^{ln}\delta ^{kn}\delta ^{lm}),`$
$`\psi _{\mu (J=2)}^j(𝐐,\lambda _g,\lambda _\rho )=\sqrt{{\displaystyle \frac{27}{160\pi }}}\left[g_\mu ^k{\displaystyle \frac{K_\mu K^k}{m_{q\overline{q}}(E_{q\overline{q}}+m_{q\overline{q}})}}\right]\overline{q}^l(\delta ^{jm}\overline{Q}^j\overline{Q}^m)ϵ^n(\lambda _\rho )`$
$`\times (\delta ^{km}\delta ^{ln}+\delta ^{kn}\delta ^{lm}{\displaystyle \frac{2}{3}}\delta ^{kl}\delta ^{mn}).`$ (7.28)
Consequently, the width can be determined from (7.16). In the nonrelativistic limit
$$B^{ij}\sqrt{2}m\delta ^{ij},$$
(7.29)
and the other components are of higher order in small quantities. Therefore:
$`B_j^\mu \psi _{\mu j}^{(0)}\sqrt{{\displaystyle \frac{3}{4\pi }}}m\overline{q}^iϵ^j(\lambda _\rho )(\delta ^{ij}\overline{Q}^i\overline{Q}^j),`$
$`B_j^\mu \psi _{\mu j}^{(1)}\sqrt{{\displaystyle \frac{9}{16\pi }}}m\overline{q}^iϵ^j(\lambda _\rho )(\delta ^{ij}+\overline{Q}^i\overline{Q}^j),`$
$`B_j^\mu \psi _{\mu j}^{(2)}\sqrt{{\displaystyle \frac{27}{80\pi }}}m\overline{q}^iϵ^j(\lambda _\rho ){\displaystyle \frac{1}{3}}(7\delta ^{ij}\overline{Q}^i\overline{Q}^j).`$ (7.30)
None of these functions vanishes. However, only two of them remain linearly independent.
### 7.2 Decay $`b_1\pi \omega `$
This process is a better test for this model because the ratio of the D-wave to the S-wave width rates is independent of the values of $`\mathrm{\Lambda }`$ and $`g`$. We begin with the $`b_1`$ as a $`q\overline{q}`$ bound state and the decay Hamiltonian (7.5). The amplitude of this mode is determined by the matrix element
$$\pi (𝐏),\omega (𝐏)|H|b_1.$$
(7.31)
This will lead to two equal terms, as before. Summation over color and flavor gives respectively $`1/\sqrt{3}`$ and $`\pm 1/\sqrt{2}`$, whereas the spin factor is given by
$$W^{\lambda _\omega }=\frac{Tr\left[(\mathit{}+m)(\mathit{}m)(\mathit{}m)(\mathit{}m)\left(\gamma ^\mu \frac{r^\mu l^\mu }{m_{q\overline{q}}(𝐫,𝐥)+2m}\right)\right]ϵ_\mu ^{}(𝐏,\lambda _\omega )}{2^{3/2}m_{q\overline{q}}(𝐩,𝐤)m_{q\overline{q}}(𝐫,𝐥)m_{q\overline{q}}(𝐩,𝐥)},$$
(7.32)
with the same notation as in the preceding section. In the nonrelativistic limit this expression tends to $`\sqrt{2}(p^iP^i)ϵ^i(\lambda _\omega )`$. The amplitude for this decay is
$`A(𝐏,\lambda _\omega ,\lambda _{b_1})={\displaystyle \frac{2\mathrm{\Lambda }}{\sqrt{6}N_\pi (P)N_\omega (P)N_{b_1}(0)}}{\displaystyle \frac{d^3𝐩}{(2\pi )^3}\frac{(E(m,𝐩)+E(m,𝐏𝐩))^2}{E(m,𝐩)E^2(m,𝐏𝐩)}}`$
$`\times W^{\lambda _\omega }(𝐩,𝐩𝐏,𝐏𝐩,𝐩)Y_{1\lambda _{b_1}}(𝐩/|𝐩|)\psi _L((E(m,𝐩)+E(m,𝐏𝐩))/\mu _{b_1})`$
$`\times \psi _L(m_{q\overline{q}}(𝐩,𝐏𝐩)/\mu _\pi )\psi _L(m_{q\overline{q}}(𝐩,𝐏𝐩)/\mu _\omega ),`$ (7.33)
which goes in the nonrelativistic limit to
$$A(𝐏,\lambda _\omega ,\lambda _{b_1})=\frac{4\mathrm{\Lambda }}{\sqrt{3}mN_\pi (P)N_\omega (P)N_{b_1}(0)}\frac{d^3𝐩}{(2\pi )^3}(p^iP^i)ϵ^i(\lambda _\rho )\psi _L^\pi \psi _L^\omega \psi _L^{b_1}.$$
(7.34)
This amplitude can be expanded into the partial waves
$$A(𝐏,\lambda _\omega ,\lambda _{b_1})=\underset{L,l}{}a_L(P)Y_{Ll}(𝐏/|𝐏|)L,l;1,\lambda _\omega |1,\lambda _{b_1},$$
(7.35)
where $`L=0`$ or $`2`$. The decay width is given by
$$d\mathrm{\Gamma }=\frac{1}{32\pi ^2}\frac{P_0|A(𝐏_0)|^2}{m_{b_1}^2}d\mathrm{\Omega },$$
(7.36)
with $`𝐏_0`$ satisfying
$$E(m_\rho ,𝐏_0)+E(m_\omega ,𝐏_0)=m_{b_1}.$$
(7.37)
This leads to
$$\mathrm{\Gamma }_L=\frac{P_0}{32\pi ^2m_{b_1}^2}a_L^2(P_0),\mathrm{\Gamma }=\underset{L}{}\mathrm{\Gamma }_L,$$
(7.38)
Taking $`𝐏=Pe_z`$ and $`\lambda _{b_1}=\lambda _{ex}=0`$ gives the first equation for the two partial amplitudes,
$$A(Pe_z,0,0)=A_{||}(P)=\sqrt{\frac{1}{4\pi }}a_0(P)+\sqrt{\frac{1}{2\pi }}a_2(P).$$
(7.39)
If $`𝐏=Pe_{}`$, where $`e_{}`$ is an arbitrary unit vector perpendicular to $`e_z`$, then the second equation for the partial amplitudes is
$$A(Pe_{},0,0)=A_{}(P)=\sqrt{\frac{1}{4\pi }}a_0(P)\sqrt{\frac{1}{8\pi }}a_2(P).$$
(7.40)
Finally, one obtains:
$`\mathrm{\Gamma }_{b_1\rho \omega }^{(swave)}={\displaystyle \frac{P_0}{72\pi m_{b_1}^2}}[A_{||}(P_0)+2A_{}(P_0)]^2,`$
$`\mathrm{\Gamma }_{b_1\rho \omega }^{(dwave)}={\displaystyle \frac{P_0}{36\pi m_{b_1}^2}}[A_{||}(P_0)A_{}(P_0)]^2.`$ (7.41)
Now we move to the $`b_1`$ treated as a gluonic bound state. The $`q\overline{q}g`$ wave function with $`J^{PC}=1^+`$, $`I=1`$ quantum numbers requires the $`q\overline{q}`$ to have the $`\pi `$ or $`\pi _2`$ quantum numbers. The corresponding, total wave functions are given by
$$\mathrm{\Psi }_{q\overline{q}g(J)}^{\lambda _{b_1}}(\lambda _q,\lambda _{\overline{q}},\lambda _g)=\underset{\lambda _{q\overline{q}},\sigma =\pm 1}{}\mathrm{\Psi }_{q\overline{q}}^{J,\lambda _{q\overline{q}}}(𝐪,𝐥_{q\overline{q}}=𝐐,\lambda _q,\lambda _{\overline{q}})D_{\lambda _g\sigma }^{(1)}(\overline{𝐐})J,\lambda _{q\overline{q}};1,\sigma |1,\lambda _{b_1},$$
(7.42)
with $`\mathrm{\Psi }_{q\overline{q}}^{J,\lambda _{q\overline{q}}}`$ being the $`\pi `$ ($`\pi _2`$) $`q\overline{q}`$ wave function for $`J=0`$ ($`J=2`$). The normalized spin wave function is thus given by
$$\mathrm{\Psi }_{q\overline{q}g(J)}^{\lambda _{b_1}}=\underset{\lambda }{}\mathrm{\Psi }_{q\overline{q}}^\lambda (𝐪,𝐐,\lambda _q,\lambda _{\overline{q}})\zeta _{(J)}(\overline{𝐐},\lambda ,\lambda _g,\lambda _\rho ),$$
(7.43)
where
$`\zeta _{(J=0)}=\sqrt{{\displaystyle \frac{3}{8\pi }}}[ϵ_c^{}(𝐐,\lambda _g)ϵ(\lambda _{b_1})]`$
$`\zeta _{(J=2)}=\sqrt{{\displaystyle \frac{27}{64\pi }}}\overline{𝐪}[ϵ_c^{}(𝐐,\lambda _g)ϵ(\lambda _{b_1})]\overline{𝐪},`$ (7.44)
with $`𝐐`$ being the gluon momentum and $`𝐪=𝐪(𝐩_q,𝐐)`$ being the relative momentum in the $`q\overline{q}`$ pair (6.5). The spin factor is given by (7.25), but with a different tensor $`B^{\mu j}`$:
$$B^{\mu j}=\frac{Tr\left[(\mathit{}m)(\mathit{}m)(\mathit{}m)\left(\gamma ^\mu \frac{r^\mu l^\mu }{m_{q\overline{q}}(𝐫,𝐥)+2m}\right)(\mathit{}+m)\gamma ^j\right]}{2^{3/2}m_{q\overline{q}}(𝐩,𝐤)m_{q\overline{q}}(𝐫,𝐥)m_{q\overline{q}}(𝐩,𝐥)}.$$
(7.45)
The notation used above is the same as in Fig. 7.2. In the nonrelativistic limit, however, this tensor tends to the same values as the $`B^{\mu j}`$ defined in (7.26). The widths can be calculated, as in the $`{}_{}{}^{3}P_{0}^{}`$ model, from (7.41).
### 7.3 Numerical results
In a simple constituent quark model one assumes that the mass difference between the mesons $`\pi `$ and $`\rho `$ arises only from spin. Therefore, we can write
$$m_M=\overline{m}_M+k(s_1s_2),$$
(7.46)
where $`M`$ denotes either meson and $`\overline{m}_M`$ is its “averaged” mass. Using the identity $`s(s+1)=s_1(s_1+1)+s_2(s_2+1)+2s_1s_2`$, we can solve for $`\overline{m}_M`$. We have $`s=0`$ for the $`\pi `$, and $`s=1`$ for the $`\rho `$. We have also $`s_1=s_2=1/2`$. Substituting $`m_\pi =`$140 MeV and $`m_\rho =`$770 MeV gives $`\overline{m}_M=`$612 MeV. Thus $`m_u=m_d=\overline{m}_M/2=`$306 MeV. A similar relation can be used for the $`K`$ and $`K^{}`$ (decays to strange mesons will be described in the next chapter), leading to $`\overline{m}_K=`$792 MeV and $`m_s=\overline{m}_Km_u=`$486 MeV.
The averaged mass of the $`\pi `$ and $`\rho `$ mesons should rather be used instead of their physical masses in the normalization constant $`(\text{6.8})`$. In similar fashion, the averaged mass of the $`K`$ and $`K^{}`$ should be used in $`(\text{8.45})`$. This procedure can also be followed for the $`b_1`$ and $`a_J`$ mesons, or for the $`h_1`$ and $`f_J`$ mesons. In this case, however, there is an additional term proportional to the spin-orbit interaction, $`SL`$. The averaged masses for the $`b_1`$ and $`f_1`$ are found to be close to their physical values. Therefore the physical values will be used in normalization for these mesons.
The weak decay constants $`(\text{6.14})`$ and $`(\text{8.46})`$ can be used to fit the parameters $`\mu _\pi `$ and $`\mu _K`$. Because our model is not exactly Lorentz covariant, the weak decay constants become functions of the meson momentum and we choose them to be equal to their experimental values at rest. Thus, setting $`f_\pi (0)=`$93 MeV and $`f_K(0)=`$113 MeV leads to $`\mu _\pi =`$221 MeV and $`\mu _K=`$275 MeV. The momentum dependence of $`f_\pi `$ in our model for $`m=306\text{ MeV}`$ and $`\mu _\pi =221\text{ MeV}`$ is presented in Fig. 7.3, which shows the difference between the rest-value and the infinity-value at the level of $`20\%`$.
The strong coupling constant at this scale is approximately $`g^2=10`$, and for the effective mass of the gluon we will take $`m_g=`$500 MeV, following what we said in Chapter 3. In Fig. 7.4 we present $`F^2(Q^2)`$ calculated with the same wave function parameters and compared with data . The agreement is good for small momentum transfer, whereas the discrepancy for larger $`Q^2`$ indicates a missing, high momentum component of the wave function. Actually, we could constrain the quark mass $`m`$ from $`f_\pi `$ and $`F_\pi `$ but it appears that both quantities are not too sensitive to $`m`$. Thus its value taken from the averaged mass of the $`\pi `$ and $`\rho `$ mesons works pretty well.
For the $`\rho `$ and $`b_1`$ decays we will choose the values of all parameters $`\mu `$ to be equal to $`\mu _\pi `$. The numerical predictions for the $`\rho `$ widths with $`\mathrm{\Lambda }=\mu _\pi `$ are presented in Table 7.1. As we discussed at the beginning of this chapter, these numbers correspond to each $`q\overline{q}`$ spin value in the hybrid component of a decaying meson. The experimental value of the width for $`\rho 2\pi `$ is $`149\text{ MeV}`$ . This number can be used to fit the free parameter $`\mathrm{\Lambda }`$, the coupling constant $`g`$, or the size parameter $`\mu _{ex}^{}`$ which need not to be of the same order as $`\mu _\pi `$. In order to do so, however, we need to know how each $`J`$ contributes to the total $`\rho `$ spin wave function.
The numerical predictions for the $`b_1`$ widths with $`\mathrm{\Lambda }=\mu _\pi `$ are presented in Table 7.2. The experimental value of the total width for the process $`b_1\pi \omega `$ is $`142\text{ MeV}`$, and for the ratio of the D-wave and S-wave width rates is $`0.08`$ . Our predictions give a value less than $`0.02`$ for this ratio in the $`{}_{}{}^{3}S_{1}^{}`$ model, and close to $`4`$ for the $`{}_{}{}^{3}P_{0}^{}`$ decay. Therefore the real mechanism should lie somewhere in between, although the $`{}_{}{}^{3}S_{1}^{}`$ mechanism gives more accurate result. However, in the $`{}_{}{}^{3}P_{0}^{}`$ model the D/S ratio is very sensitive to the free parameters $`\mu `$ and for $`\mu =`$400 MeV one obtains this ratio on the order of the experimental value . This value of $`\mu `$, however, is inconsistent with the weak decay constant and the elastic form factor for the pion.
The $`q\overline{q}`$g wave function component of the $`b_1`$ wave functions used here is that of Eq. (7.42), corresponding to a $`q\overline{q}`$ pair with the $`\pi `$ quantum numbers. For a $`q\overline{q}`$ with the $`\pi _2`$ quantum numbers, the numeric value for the width is much smaller than $`1\text{ MeV}`$ for the S-wave and approximately $`1\text{ MeV}`$ for the D-wave. The ratio $`D/S`$ is respectively $`230`$. In the nonrelativistic limit we obtain similar results.
We observe that treating the $`b_1`$ as the $`\pi _2+g`$ state increases dramatically the $`D/S`$ ratio. Therefore this may be an important component of the wave function. Relatively small values of the decay widths of $`b_1`$ with $`\pi _2`$ quantum numbers (L=2) compared to those of $`b_1`$ with pion quantum numbers resemble the situation for the process $`\pi _1\pi b_1`$, whose D-wave width was small compared to that in the S-wave.
## Chapter 8 Decays of $`\pi _1`$
In this chapter we will study the main subject of the presented work, i.e., a completely relativistic decay of the exotic meson $`\pi _1`$. In experiment we observe that most hadronic decays involve a minimal number of final state particles. This requires a small number of transitions at the quark level. Thus, an exotic meson is expected to decay into two normal mesons. In a constituent quark model the transverse gluon in the $`\pi _1`$ dissociates into a quark and an antiquark, and the two resulting quark-antiquark pairs rearrange themselves into two mesons, as shown in Fig. 8.1.
The possible decay modes of the $`\pi _1`$ for $`m_{\pi _1}=`$ 1600 MeV are listed in Table 8.1, where $`L`$ is the angular momentum between outgoing mesons. For brevity, we did not put there the decays into the antiparticles of the corresponding strange mesons. Hereinafter, such decays are understood to be included and have the same widths. In the following sections we will discuss the most important modes and calculate the corresponding widths.
### 8.1 Decay of $`\pi _1`$ into $`\pi \eta `$ and $`\pi b_1`$
The amplitudes for these modes are related to the matrix element
$$\pi (𝐏),M(𝐏)|H|\pi _1,$$
(8.1)
where $`M`$ is either $`\eta `$ or $`b_1`$. From rearranging the annihilation and creation operators one obtains two nonzero terms contributing to the total amplitude. We will show below that they are equal. Summing over color gives a factor $`4/3`$, as for decays of the $`q\overline{q}g`$ component of a normal meson. Summing over flavor leads to a trace of a product of Pauli matrices appearing in the isospin term (6.3),
$$2^{3/2}Tr(\sigma _i^{ex}\sigma _j^\pi )ϵ_i(I_3^{ex})ϵ_j^{}(I_3^\pi )$$
(8.2)
or
$$2^{3/2}Tr(\sigma _i^{ex}\sigma _j^\pi \sigma _k^{b_1})ϵ_i(I_3^{ex})ϵ_j^{}(I_3^\pi )ϵ_k^{}(I_3^{b_1}),$$
(8.3)
respectively (for all allowed isospin channels these factors are again equal to $`\pm 1/\sqrt{2}`$). The spin factor is given by (7.25) with $`S=J`$ and $`\lambda _\rho `$ replaced by $`\lambda _{ex}`$, where the tensor $`B^{\mu j}`$ was introduced in (7.26) and $`\psi _\mu ^{j(S)}`$ are given by (7.27). The notation is shown in Fig. 8.1. Using (5.36) we can write:
$`\psi _\mu ^{j(S=0)}=\sqrt{{\displaystyle \frac{3}{8\pi }}}g_\mu ^k(\delta ^{jk}\overline{Q}^j\overline{Q}^k)\overline{Q}^lϵ^l(\lambda _{ex}),`$
$`\psi _\mu ^{j(S=1)}=\sqrt{{\displaystyle \frac{3}{8\pi }}}\left[g_\mu ^k{\displaystyle \frac{K_\mu K^k}{m_{q\overline{q}}(E_{q\overline{q}}+m_{q\overline{q}})}}\right](\delta ^{jl}\overline{Q}^j\overline{Q}^l)\overline{Q}^kϵ^l(\lambda _{ex}),`$
$`\psi _\mu ^{j(S=2)}={\displaystyle \frac{3}{\sqrt{13}}}(\psi _\mu ^{j(S=1)}{\displaystyle \frac{2}{3}}\psi _\mu ^{j(S=0)}).`$ (8.4)
In the terms above we used the following notation:
$$m_{q\overline{q}}=m_{q\overline{q}}(𝐩,𝐏𝐫),𝐐=𝐏𝐩+𝐫,E_{q\overline{q}}=E(m_{q\overline{q}},𝐐),𝐊=𝐐,K^0=E_{q\overline{q}}+m_{q\overline{q}},$$
(8.5)
and assumed all quarks being on-shell particles.
Integration over all momenta gives for the first term $`(2\pi )^3\delta ^3(\mathrm{𝟎})A_1`$, where the amplitude $`A_1`$ is given by
$`A_{1(M)}^{(S)}(𝐏,\lambda _M,\lambda _{ex})={\displaystyle \frac{d^3𝐩}{(2\pi )^3}\frac{d^3𝐫}{(2\pi )^3}\frac{(E(m,𝐩)+E(m,𝐤))(E(m,𝐫)+E(m,𝐥))}{4E(m,𝐩)E(m,𝐤)E(m,𝐫)E(m,𝐥)E(m_g,𝐐)}}`$
$`\times (E(m,𝐩)+E(m,𝐥)+E(m_g,𝐐))\psi _L^{}(m_{q\overline{q}}(𝐩,𝐥)/\mu _{ex},m_{q\overline{q}g}(𝐩,𝐥,𝐐)/\mu _{ex}^{})`$
$`\times [N_\pi (P)N_M(P)N_{ex}]^1\psi _L(m_{q\overline{q}}(𝐩,𝐤)/\mu _\pi )\psi _L(m_{q\overline{q}}(𝐫,𝐥)/\mu _M)Y_{J_M\lambda _M}^{}(𝐪/|𝐪|)`$
$`\times g{\displaystyle \frac{4}{3}}{\displaystyle \frac{1}{\sqrt{2}}}B_j^\mu \psi _{\mu j}^{(S)}(\lambda _{ex}).`$ (8.6)
Here, $`M`$ denotes the second meson ($`\eta `$ or $`b_1`$), $`J`$ is its total spin, $`𝐪=𝐪(𝐫,𝐏)`$ is given in (6.5) and $`P=|𝐏|`$. If $`M=\eta `$, then the above expression must be proportional to $`ϵ^i(\lambda _{ex})P^i`$ (or $`Y_{1\lambda _{ex}}(𝐏/P)`$), being a vector function of the vector $`𝐏`$ (the outgoing mesons in the P-wave). However, if $`M=b_1`$ then the integral (8.6) is proportional to
$$ϵ^i(\lambda _{ex})ϵ^j(\lambda _M)[P^iP^j+f(P)\delta ^{ij}],$$
(8.7)
and can be represented as a superposition of spherical harmonics corresponding respectively to $`l=0`$ and $`l=2`$ (S-wave and D-wave).
The second term in the Hamiltonian matrix element $`A_2^{(S)}`$ is obtained from the first one by interchanging
$$𝐩𝐫,𝐏𝐏,I_3^{(1)}I_3^{(2)}$$
(8.8)
everywhere in (8.6), including $`𝐤,𝐥,𝐐`$, except in the third line. For $`\pi _1\pi \eta `$ we can write schematically:
$`𝐀_1(𝐏)={\displaystyle d^3𝐩d^3𝐫f_\eta (m_{q\overline{q}}(𝐩,𝐏𝐩))f_\pi (m_{q\overline{q}}(𝐫,𝐏𝐫))f_{ex}(m_{q\overline{q}}(𝐩,𝐏𝐫))}`$
$`\times (𝐏𝐩+𝐫),`$
$`𝐀_2(𝐏)={\displaystyle d^3𝐩d^3𝐫f_\eta (m_{q\overline{q}}(𝐩,𝐏𝐩))f_\pi (m_{q\overline{q}}(𝐫,𝐏𝐫))f_{ex}(m_{q\overline{q}}(𝐫,𝐏𝐩))}`$
$`\times (𝐏𝐫+𝐩).`$ (8.9)
We have not included here the energy and trace factors because they do not change under the above symmetry. Using $`m_{q\overline{q}}(𝐩,𝐫)=m_{q\overline{q}}(𝐫,𝐩)=m_{q\overline{q}}(𝐩,𝐫)`$ one can show that both terms are equal (for the same $`\lambda _M`$ and $`\lambda _{ex}`$),
$$𝐀_2(𝐏)=𝐀_1(𝐏)=𝐀_1(𝐏).$$
(8.10)
Thus, we have $`A=A_1+A_2=2A_1`$. In similar fashion one can prove this equality (up to a sign) for all decays of any $`q\overline{q}g`$ particle into two mesons.
If $`\mu _\eta =\mu _\pi `$, then each term ($`A_1`$ or $`A_2`$) is a product of a part that is symmetric under interchanging $`𝐩𝐫+𝐏`$ and a (vector) part that is antisymmetric. In this case the hybrid will not decay into $`\pi `$ and $`\eta `$. Neither can it decay into two pions because of a minus sign from interchanging the Pauli matrices in the isospin factor that makes both terms cancel. There is the same minus sign for the $`\pi _1\pi b_1`$, but now the amplitude is a scalar ($`L=0`$) or a tensor ($`L=2`$) function of $`𝐏`$ and again $`A_1(𝐏)=A_2(𝐏)`$. However, in this case $`\mu _{b_1}=\mu _\pi `$ does not imply $`A=0`$ because the orbital wave functions of these mesons are different. Thus we find that the $`1^+`$ isovector does not decay into identical pseudoscalars. This is a relativistic generalization of a symmetry found in other nonrelativistic decay models . One might think that a similar symmetry could occur for the decay $`\rho 2\pi `$. In this case, however, the amplitude does not vanish because its vector part comes from a spherical harmonic associated not with the gluon momentum, but with the relative momentum in a $`q\overline{q}`$ pair. This harmonic is asymmetric under interchanging $`𝐩𝐫+𝐏`$ and the amplitude for the $`\rho 2\pi `$ does not vanish.
The width is given by
$$d\mathrm{\Gamma }=\frac{1}{32\pi ^2}\frac{P_0|A(𝐏_0)|^2}{m_{ex}^2}d\mathrm{\Omega },$$
(8.11)
with $`𝐏_0`$ satisfying
$$E(m_\pi ,𝐏_0)+E(m_M,𝐏_0)=m_{ex}$$
(8.12)
and $`P_0=|𝐏_0|`$. The amplitudes must be, according to the Wigner-Eckart theorem, of the form:
$`A^{\pi \eta }(𝐏,\lambda _{ex})=a_1^{\pi \eta }(P)Y_{1\lambda _{ex}}(𝐏/P),`$
$`A^{\pi b_1}(𝐏,\lambda _{b_1},\lambda _{ex})={\displaystyle \underset{L,l}{}}a_L^{\pi b_1}(P)Y_{Ll}(𝐏/P)L,l;1,\lambda _{b_1}|1,\lambda _{ex},`$ (8.13)
which leads to Eq. (7.38) with $`m_{b_1}`$ replaced by $`m_{ex}`$. The amplitudes $`A(𝐏,\lambda _M,\lambda _{ex})`$ given in (8.6) are multiplied by $`2`$, since there are two equal terms. The expression for the $`\pi _1\pi \eta `$ width is similar to (7.16), whereas for the $`\pi _1\pi b_1`$ we can use the “parallel” and “perpendicular” amplitudes introduced for the decay $`b_1\pi \omega `$. The final results are:
$`\mathrm{\Gamma }_{\pi \eta }^{(pwave)}={\displaystyle \frac{P_0}{24\pi m_{ex}^2}}A_{\pi \eta }^2(P_0e_z,0),`$
$`\mathrm{\Gamma }_{\pi b_1}^{(swave)}={\displaystyle \frac{P_0}{72\pi m_{ex}^2}}[A_{||}^{\pi b_1}(P_0)+2A_{}^{\pi b_1}(P_0)]^2,`$
$`\mathrm{\Gamma }_{\pi b_1}^{(dwave)}={\displaystyle \frac{P_0}{36\pi m_{ex}^2}}[A_{||}^{\pi b_1}(P_0)A_{}^{\pi b_1}(P_0)]^2.`$ (8.14)
The $`\pi `$ and $`\eta `$ mesons have the same quantum numbers (except isospin) so $`\mu _\pi `$ and $`\mu _\eta `$ should be almost equal. This equality is not exact because the $`SU(3)_f`$ is only an approximate symmetry and there is a contribution of the $`s\overline{s}`$ in $`\eta `$. Therefore the amplitude for the $`\pi _1\pi \eta `$ should be close to zero, and of two channels $`\pi \eta `$, $`\pi b_1`$ the latter will be favored. However, the free parameters $`\mu `$ need not to be close to each other for decays of the $`\pi _1`$ into two mesons with different radial quantum number, which would make such channels significant.
### 8.2 Decay of $`\pi _1`$ into $`\pi \rho `$, $`\pi f_1`$, $`\pi f_2`$, $`\eta a_1`$ and $`\eta a_2`$
For these channels the procedure is analogous to that in the preceding section. The amplitude of this process is given by the matrix element $`M_1(𝐏),M_2(𝐏)|H|\pi _1`$ which is again a sum of two terms, and $`M_1`$ and $`M_2`$ denote the two outgoing mesons. The color and flavor factors are again $`4/3`$ and $`\pm 1/\sqrt{2}`$, but the spin factor is given now by
$$C_j^{\mu \nu }\psi _{\mu \nu j}^{(S)},$$
(8.15)
where (for the first term)
$`C^{\mu \nu j}=[2^{3/2}m_{q\overline{q}}(𝐩,𝐤)m_{q\overline{q}}(𝐫,𝐥)m_{q\overline{q}}(𝐩,𝐥)]^1Tr[(\mathit{}+m)(\gamma ^\mu {\displaystyle \frac{p^\mu l^\mu }{m_{q\overline{q}}(𝐩,𝐥)+2m}})`$
$`\times (\mathit{}m)(\gamma ^\nu {\displaystyle \frac{r^\nu l^\nu }{m_{q\overline{q}}(𝐫,𝐥)+2m}})(\mathit{}+m)\gamma ^j(\mathit{}m)\gamma ^5],`$ (8.16)
and
$$\psi _\mu ^{\nu j(S)}(\lambda ,\lambda _{ex})=\underset{\lambda _g}{}\psi _{\mu (S)}(𝐐,\lambda _g,\lambda _{ex})ϵ_c^j(𝐐,\lambda _g)ϵ^\nu (𝐏,\lambda )$$
(8.17)
or equivalently:
$`\psi _\mu ^{\nu j(S=0)}=\sqrt{{\displaystyle \frac{3}{8\pi }}}g_\mu ^k(\delta ^{jk}\overline{Q}^j\overline{Q}^k)\overline{Q}^lϵ^l(\lambda _{ex})ϵ^\nu (𝐏,\lambda ),`$
$`\psi _\mu ^{\nu j(S=1)}=\sqrt{{\displaystyle \frac{3}{8\pi }}}\left[g_\mu ^k{\displaystyle \frac{K_\mu K^k}{m_{q\overline{q}}(E_{q\overline{q}}+m_{q\overline{q}})}}\right](\delta ^{jl}\overline{Q}^j\overline{Q}^l)\overline{Q}^kϵ^l(\lambda _{ex})ϵ^\nu (𝐏,\lambda ),`$
$`\psi _\mu ^{\nu j(S=2)}={\displaystyle \frac{3}{\sqrt{13}}}(\psi _\mu ^{\nu j(S=1)}{\displaystyle \frac{2}{3}}\psi _\mu ^{\nu j(S=0)}).`$ (8.18)
The notation used above is the same as in (8.5) and Fig. 8.1.
The amplitude for the first term is given by
$`A_{1(M)}^{(S)}(𝐏,\lambda _M,\lambda _{ex})={\displaystyle \underset{\lambda ,l}{}}{\displaystyle \frac{d^3𝐩}{(2\pi )^3}\frac{d^3𝐫}{(2\pi )^3}\frac{(E(m,𝐩)+E(m,𝐤))(E(m,𝐫)+E(m,𝐥))}{4E(m,𝐩)E(m,𝐤)E(m,𝐫)E(m,𝐥)E(m_g,𝐐)}}`$
$`\times (E(m,𝐩)+E(m,𝐥)+E(m_g,𝐐))\psi _L^{}(m_{q\overline{q}}(𝐩,𝐥)/\mu _{ex},m_{q\overline{q}g}(𝐩,𝐥,𝐐)/\mu _{ex}^{})`$
$`\times [N_\pi N_MN_{ex}]^1\psi _L(m_{q\overline{q}}(𝐩,𝐤)/\mu _\pi )\psi _L(m_{q\overline{q}}(𝐫,𝐥)/\mu _M)Y_{L_{q\overline{q}(M)}l}^{}(𝐪/|𝐪|)`$
$`\times 1,\lambda ;L_{q\overline{q}(M)},l|J_M,\lambda _Mg{\displaystyle \frac{4}{3}}{\displaystyle \frac{1}{\sqrt{2}}}C_j^{\mu \nu }\psi _{\mu \nu j}^{(S)}(\lambda ,\lambda _{ex}),`$ (8.19)
where $`M`$ denotes the second meson ($`\rho `$ with $`L_{q\overline{q}}=0`$ or $`f_J`$ with $`L_{q\overline{q}}=1`$), $`J`$ is its total spin, $`𝐪=𝐪(𝐫,𝐏)`$ was defined in (6.5) and $`J=1,2`$. Again one can show (for the same $`\lambda `$’s)
$$A_2^{(S)}(𝐏)=A_1^{(S)}(𝐏).$$
(8.20)
If $`\mu _\rho =\mu _\pi `$, then the amplitude of the decay into $`\pi \rho `$ is not symmetric under interchanging $`𝐩𝐫`$ and does not vanish, unlike for the $`\pi \eta `$ case. Therefore this mode may be significant. The same holds for the $`\pi f_J`$ channels.
From the Wigner-Eckart theorem we have:
$`A^{\pi \rho }(𝐏,\lambda _\rho ,\lambda _{ex})={\displaystyle \underset{L,l}{}}a_L^{\pi \rho }(P)Y_{Ll}(𝐏/P)L,l;1,\lambda _\rho |1,\lambda _{ex},`$
$`A^{\pi f_J}(𝐏,\lambda _{f_J},\lambda _{ex})={\displaystyle \underset{L,l}{}}a_L^{\pi f_J}(P)Y_{Ll}(𝐏/P)L,l;J,\lambda _{f_J}|1,\lambda _{ex}.`$ (8.21)
In formulae (8.21) and (7.38) one must use the $`A(𝐏,\lambda ^{}s)`$ given in (8.19), multiplied by $`2`$. The mesons in the $`\pi \rho `$ channel go out in the P-wave, for the $`\pi f_1`$ it is either the S-wave or D-wave, whereas in the $`\pi f_2`$ they go out only in the D-wave. Taking $`𝐏=Pe_z`$ and $`\lambda _\rho =\lambda _{ex}=+1`$ leads to
$$a_1^{\pi \rho }(P)=\sqrt{\frac{8\pi }{3}}A^{\pi \rho }(Pe_z,1,1),$$
(8.22)
and therefore
$$\mathrm{\Gamma }_{\pi \rho }^{(pwave)}=\frac{P_0}{12\pi m_{ex}^2}A_{\pi \rho }^2(P_0e_z,1,1).$$
(8.23)
In the above $`𝐏_0`$ satisfies (8.12) with $`m_M=m_\rho `$. For decays into $`\pi f_J`$ one may follow the procedure with the parallel and perpendicular amplitudes described in the preceding section, or just use the general formula that is equivalent to (8.21),
$$a_L^{\pi f_J}(P)=\underset{L,l}{}𝑑\mathrm{\Omega }_𝐏A^{\pi f_J}(𝐏,\lambda _{f_J},\lambda _{ex})Y_{Ll}^{}(𝐏/P)L,l;J,\lambda _{f_J}|1,\lambda _{ex},$$
(8.24)
and substitute it into (7.38) with $`m_{b_1}`$ replaced by $`m_{ex}`$. Here, $`𝐏_0`$ satisfies (8.12) with $`m_M=m_{f_J}`$. All results for the decays $`\pi _1\pi f_J`$ are valid also for the $`\eta a_J`$ modes (with different normalization constants, parameters $`\mu `$ and $`P_0`$). In this case, $`\pi `$ must be replaced with $`\eta `$ and $`f`$ with $`a`$.
### 8.3 Decay $`\pi _1\rho \omega `$
The amplitude of this process is given by the matrix element $`\rho (𝐏),\omega (𝐏)|H|\pi _1`$, being again a sum of two terms. The color and flavor factors are the same as before, and the spin factor is equal to
$$D_j^{\mu \nu \rho }\psi _{\mu \nu \rho j}^{(S)},$$
(8.25)
where (for the first term)
$`D^{\mu \nu \rho j}=[2^{3/2}m_{q\overline{q}}(𝐩,𝐤)m_{q\overline{q}}(𝐫_{q_2},𝐥)m_{q\overline{q}}(𝐩,𝐥)]^1Tr[(\mathit{}m)(\gamma ^\nu {\displaystyle \frac{p^\nu k^\nu }{m_{q\overline{q}}(𝐩,𝐤)+2m}})`$
$`\times (\mathit{}+m)(\gamma ^\mu {\displaystyle \frac{p^\mu l^\mu }{m_{q\overline{q}}(𝐩,𝐥)+2m}})(\mathit{}m)(\gamma ^\rho {\displaystyle \frac{r^\rho l^\rho }{m_{q\overline{q}}(𝐫,𝐥)+2m}})(\mathit{}+m)\gamma ^j],`$ (8.26)
and
$$\psi _\mu ^{\nu \rho j(S)}=\underset{\lambda _g}{}\psi _{\mu (S)}(𝐐,\lambda _g,\lambda _{ex})ϵ_c^j(𝐐,\lambda _g)ϵ^\nu (𝐏,\lambda _\rho )ϵ^\rho (𝐏,\lambda _\omega )$$
(8.27)
or equivalently:
$`\psi _\mu ^{\nu \rho j(S=0)}=\sqrt{{\displaystyle \frac{3}{8\pi }}}g_\mu ^k(\delta ^{jk}\overline{Q}^j\overline{Q}^k)\overline{Q}^lϵ^l(\lambda _{ex})ϵ^\nu (𝐏,\lambda _\rho )ϵ^\rho (𝐏,\lambda _\omega ),`$
$`\psi _\mu ^{\nu \rho j(S=1)}=\sqrt{{\displaystyle \frac{3}{8\pi }}}\left[g_\mu ^k{\displaystyle \frac{K_\mu K^k}{m_{q\overline{q}}(E_{q\overline{q}}+m_{q\overline{q}})}}\right](\delta ^{jl}\overline{Q}^j\overline{Q}^l)\overline{Q}^kϵ^l(\lambda _{ex})`$
$`\times ϵ^\nu (𝐏,\lambda _\rho )ϵ^\rho (𝐏,\lambda _\omega ),`$
$`\psi _\mu ^{\nu \rho j(S=2)}={\displaystyle \frac{3}{\sqrt{13}}}(\psi _\mu ^{\nu \rho j(S=1)}{\displaystyle \frac{2}{3}}\psi _\mu ^{\nu \rho j(S=0)}).`$ (8.28)
The notation used above is the same as in the preceding sections.
Again one can demonstrate that the second term in the amplitude is equal to the first one. Therefore the total amplitude is
$`A^{(S)}(𝐏,\lambda _\rho ,\lambda _\omega ,\lambda _{ex})=2{\displaystyle \frac{d^3𝐩}{(2\pi )^3}\frac{d^3𝐫}{(2\pi )^3}\frac{(E(m,𝐩)+E(m,𝐤))(E(m,𝐫)+E(m,𝐥))}{4E(m,𝐩)E(m,𝐤)E(m,𝐫)E(m,𝐥)E(m_g,𝐐)}}`$
$`\times (E(m,𝐩)+E(m,𝐥)+E(m_g,𝐐))\psi _L^{}(m_{q\overline{q}}(𝐩,𝐥)/\mu _{ex},m_{q\overline{q}g}(𝐩,𝐥,𝐐)/\mu _{ex}^{})`$
$`\times [N_\rho N_\omega N_{ex}]^1\psi _L(m_{q\overline{q}}(𝐩,𝐤)/\mu _\rho )\psi _L(m_{q\overline{q}}(𝐫,𝐥)/\mu _\omega )g{\displaystyle \frac{4}{3}}{\displaystyle \frac{1}{\sqrt{2}}}D_j^{\mu \nu \rho }\psi _{\mu \nu \rho j}^{(S)}.`$ (8.29)
If $`\mu _\rho =\mu _\omega `$, then the symmetry of the orbital wave functions causes $`A=0`$ and the hybrid will not decay into $`\rho `$ and $`\omega `$. Because both parameters $`\mu `$ are expected to be on the same order, the $`\rho \omega `$ mode will not be favored. The width is given by (7.38) with $`m_{b_1}`$ replaced by $`m_{ex}`$, where
$$a_L^{\rho \omega }(P)=\underset{L,l,J^{},\lambda ^{}}{}𝑑\mathrm{\Omega }_𝐏A^{\rho \omega }(𝐏,\lambda _\rho ,\lambda _\omega ,\lambda _{ex})Y_{Ll}^{}(𝐏/P)1,\lambda _\rho ;1,\lambda _\omega |J^{},\lambda ^{}L,l;J^{},\lambda ^{}|1,\lambda _{ex},$$
(8.30)
and $`𝐏_0`$ is obtained from $`E(m_\rho ,𝐏_0)+E(m_\omega ,𝐏_0)=m_{ex}`$. In this process we have either $`L=1`$ or $`L=3`$.
### 8.4 Decay into strange mesons
The above results can be straightforwardly generalized to the case where $`m_q`$ and $`m_{\overline{q}}`$ are different, for example to decays into mesons with one strange quark ($`I=1/2`$). The spin wave function for a quark-antiquark pair in a $`J^P=0^{}`$ state is
$$\mathrm{\Psi }_{q\overline{q}}(𝐥_q,𝐥_{\overline{q}},\lambda _q,\lambda _{\overline{q}})=\frac{1}{\sqrt{2}\stackrel{~}{m}_{q\overline{q}}(𝐥_q,𝐥_{\overline{q}})}\overline{u}(m_q,𝐥_q,\lambda _q)\gamma ^5v(m_{\overline{q}},𝐥_{\overline{q}},\lambda _{\overline{q}}),$$
(8.31)
and the $`K`$ states (understood to be both $`K`$ and $`\overline{K}`$) are given by
$`|K(𝐏)={\displaystyle \underset{all\lambda ,c}{}}{\displaystyle \frac{d^3𝐩_q}{(2\pi )^32E(m_q,𝐩_q)}\frac{d^3𝐩_{\overline{q}}}{(2\pi )^32E(m_{\overline{q}},𝐩_{\overline{q}})}2(E(m_q,𝐩_q)+E(m_{\overline{q}},𝐩_{\overline{q}}))}`$
$`\times (2\pi )^3\delta ^3(𝐩_q+𝐩_{\overline{q}}𝐏){\displaystyle \frac{1}{\sqrt{3}}}\delta _{c_qc_{\overline{q}}}\mathrm{\Psi }_{q\overline{q}}(𝐩_q,𝐩_{\overline{q}},\lambda _q,\lambda _{\overline{q}}){\displaystyle \frac{1}{N(P)}}\psi _L(m_{q\overline{q}}(𝐩_q,𝐩_{\overline{q}})/\mu )`$
$`\times b_{𝐩_q\lambda _qf_qc_q}^{}d_{𝐩_{\overline{q}}\lambda _{\overline{q}}f_{\overline{q}}c_{\overline{q}}}^{}|0,`$ (8.32)
where the operators satisfy the anticommutation relations:
$`\{b_{𝐩\lambda fc},b_{𝐩^{}\lambda ^{}f^{}c^{}}^{}\}=(2\pi )^32E(m_q,𝐩)\delta ^3(𝐩𝐩^{})\delta _{\lambda \lambda ^{}}\delta _{ff^{}}\delta _{cc^{}},`$
$`\{d_{𝐩\lambda fc},d_{𝐩^{}\lambda ^{}f^{}c^{}}^{}\}=(2\pi )^32E(m_{\overline{q}},𝐩)\delta ^3(𝐩𝐩^{})\delta _{\lambda \lambda ^{}}\delta _{ff^{}}\delta _{cc^{}}.`$ (8.33)
The invariant mass is defined as
$$m_{q\overline{q}}=m_{q\overline{q}}(𝐥_q,𝐥_{\overline{q}})=\sqrt{(E(m_q,𝐥_q)+E(m_{\overline{q}},𝐥_{\overline{q}}))^2(𝐥_q+𝐥_{\overline{q}})^2}$$
(8.34)
and the “modified” invariant mass is
$$\stackrel{~}{m}_{q\overline{q}}=\sqrt{m_{q\overline{q}}^2(m_qm_{\overline{q}})^2}.$$
(8.35)
The flavor indices $`f_q`$ and $`f_{\overline{q}}`$ give four combinations: $`u\overline{s}`$, $`d\overline{s}`$, $`s\overline{u}`$ and $`s\overline{d}`$, corresponding to appropriate mesons.
If in (8.32) $`\mathrm{\Psi }_{q\overline{q}}`$ is given by
$$\mathrm{\Psi }_{q\overline{q}}^{\lambda _{q\overline{q}}}(𝐥_q,𝐥_{\overline{q}},\lambda _q,\lambda _{\overline{q}})=\frac{1}{\sqrt{2}\stackrel{~}{m}_{q\overline{q}}}\overline{u}(m_q,𝐥_q,\lambda _q)\left[\gamma ^\mu \frac{l_q^\mu l_{\overline{q}}^\mu }{m_{q\overline{q}}+m_q+m_{\overline{q}}}\right]v(m_{\overline{q}},𝐥_{\overline{q}},\lambda _{\overline{q}})ϵ_\mu (𝐥_{q\overline{q}},\lambda _{q\overline{q}}),$$
(8.36)
then one obtains the $`K^{}`$ states ($`J^P=1^{}`$). For a $`q\overline{q}`$ pair with spin $`S_{q\overline{q}}=0`$ and the orbital angular momentum $`L=1`$ we have
$$\mathrm{\Psi }_{q\overline{q}}^{\lambda _{q\overline{q}}}(𝐥_q,𝐥_{\overline{q}},\lambda _q,\lambda _{\overline{q}})=\frac{1}{\sqrt{2}\stackrel{~}{m}_{q\overline{q}}(𝐥_q,𝐥_{\overline{q}})}\overline{u}(m_q,𝐥_q,\lambda _q)\gamma ^5v(m_{\overline{q}},𝐥_{\overline{q}},\lambda _{\overline{q}})Y_{1\lambda _{q\overline{q}}}(\overline{𝐪}),$$
(8.37)
which corresponds to a $`K_1`$ meson ($`J^P=1^+`$). For $`S_{q\overline{q}}=1`$ we get
$`\mathrm{\Psi }_{q\overline{q}}^{\lambda _{q\overline{q}}}(𝐥_q,𝐥_{\overline{q}},\lambda _q,\lambda _{\overline{q}})={\displaystyle \underset{\lambda ,l}{}}{\displaystyle \frac{1}{\sqrt{2}\stackrel{~}{m}_{q\overline{q}}}}\overline{u}(m_q,𝐥_q,\lambda _q)\left[\gamma ^\mu {\displaystyle \frac{l_q^\mu l_{\overline{q}}^\mu }{m_{q\overline{q}}+m_q+m_{\overline{q}}}}\right]v(m_{\overline{q}},𝐥_{\overline{q}},\lambda _{\overline{q}})`$
$`\times ϵ_\mu (𝐥_{q\overline{q}},\lambda )Y_{1l}(\overline{𝐪})1,\lambda ;1,l|J,\lambda _{q\overline{q}},`$ (8.38)
where $`𝐪`$ was defined in (5.24). The last wave function characterizes mesons $`K_0^{}`$, $`K_1`$ and $`K_2^{}`$ ($`J^P=0,1,2^+`$). In a similar manner one can obtain states with higher orbital angular momenta $`L`$.
The decay of $`\pi _1`$ into two strange mesons proceeds through the dissociation of the gluon into a pair $`s\overline{s}`$. For the processes $`\pi _1K\overline{K}^{}`$ and $`\pi _1K\overline{K}_1`$ the Hamiltonian matrix element is $`H=(2\pi )^3\delta ^3(\mathrm{𝟎})A`$ (now there is only one term). The decay amplitude is given by
$`A_{(M)}^{(S)}(𝐏,\lambda _M,\lambda _{ex})={\displaystyle \underset{\lambda ,l}{}}{\displaystyle \frac{d^3𝐩}{(2\pi )^3}\frac{d^3𝐫}{(2\pi )^3}\frac{(E(m,𝐩)+E(m^{},𝐤))(E(m^{},𝐫)+E(m,𝐥))}{4E(m,𝐩)E(m^{},𝐤)E(m^{},𝐫)E(m,𝐥)E(m_g,𝐐)}}`$
$`\times (E(m,𝐩)+E(m,𝐥)+E(m_g,𝐐))\psi _L^{}(m_{q\overline{q}}(𝐩,𝐥)/\mu _{ex},m_{q\overline{q}g}(𝐩,𝐥,𝐐)/\mu _{ex}^{})`$
$`\times [N_KN_MN_{ex}]^1\psi _L(m_{q\overline{q}}(𝐩,𝐤)/\mu _\pi )\psi _L(m_{q\overline{q}}(𝐫,𝐥)/\mu _M)Y_{L_Ml}^{}(𝐪/|𝐪|)`$
$`\times 1,\lambda ;L_M,l|J_M,\lambda _Mg{\displaystyle \frac{4}{3}}{\displaystyle \frac{1}{\sqrt{2}}}(spin)(\lambda ,\lambda _{ex}).`$ (8.39)
Masses of quarks $`u`$,$`d`$ (assumed to be equal) and $`s`$ are denoted respectively by $`m`$ and $`m^{}`$, whereas $`M`$ stands for either $`\overline{K}^{}`$ or $`\overline{K}_1`$, and $`S=S_{q\overline{q}g}`$. The spin factor is given by
$$B_j^\mu \psi _{\mu j}^{(S)}(\lambda _{ex})\delta _{\lambda 0}$$
(8.40)
for kaons with $`S_{q\overline{q}}=1`$, and
$$C_j^{\mu \nu }\psi _{\mu \nu j}^{(S)}(\lambda ,\lambda _{ex})$$
(8.41)
if $`S_{q\overline{q}}=0`$, where
$$B^{\mu j}=\frac{Tr\left[(\mathit{}m^{})(\mathit{}m)\left(\gamma ^\mu +\frac{p^\mu l^\mu }{m_{q\overline{q}}(𝐩,𝐥)+2m}\right)(\mathit{}+m)(\mathit{}+m^{})\gamma ^j\right]}{2^{3/2}m_{q\overline{q}}(𝐩,𝐤)m_{q\overline{q}}(𝐫,𝐥)m_{q\overline{q}}(𝐩,𝐥)}$$
(8.42)
and
$`C^{\mu \nu j}=[2^{3/2}m_{q\overline{q}}(𝐩,𝐤)m_{q\overline{q}}(𝐫,𝐥)m_{q\overline{q}}(𝐩,𝐥)]^1Tr[(\mathit{}+m)(\gamma ^\mu {\displaystyle \frac{p^\mu l^\mu }{m_{q\overline{q}}(𝐩,𝐥)+2m}})`$
$`\times (\mathit{}m)(\gamma ^\nu {\displaystyle \frac{r^\nu l^\nu }{m_{q\overline{q}}(𝐫,𝐥)+m+m^{}}})(\mathit{}+m^{})\gamma ^j(\mathit{}m^{})\gamma ^5].`$ (8.43)
The notation used above is the same as that in (8.5) and Fig. 8.1, with the invariant mass (8.34) and some modifications due to difference between the masses of quarks:
$$p_{q_1}^0=E(m,𝐩),r^0=E(m^{},𝐫),k^0=E(m^{},𝐤),l^0=E(m,𝐥).$$
(8.44)
The normalization constants for strange mesons are
$`N_M^2(P)=(2E(m_M,𝐏))^1{\displaystyle \frac{d^3𝐤}{(2\pi )^3}\frac{(E(m,𝐤)+E(m^{},𝐏𝐤))^2}{E(m,𝐤)E(m^{},𝐏𝐤)}[Y_{L0}(\overline{𝐪}(𝐤,𝐏))]^2}`$
$`\times [\psi _L(m_{q\overline{q}}(𝐤,𝐏𝐤)/\mu _M)]^2,`$ (8.45)
where $`𝐪=𝐪(𝐤,𝐏)`$ is in (6.5). The kaon weak decay constant can be used to constrain the value of $`\mu _K`$:
$$f_K(P)=\frac{\sqrt{3}}{N_K(P)E(m_K,𝐏)}\frac{d^3𝐩}{(2\pi )^3}\frac{(p^0+q^0)(m^{}p^0+mq^0)}{p^0q^0m_{q\overline{q}}(𝐩,𝐏𝐩)}e^{\frac{m_{q\overline{q}}^2}{8\mu _K^2}}.$$
(8.46)
### 8.5 Nonrelativistic limit
For the decays $`\pi _1\pi \eta ,\pi b_1`$ the behaviour of the tensor (7.26) is given by Eq. (7.29). Therefore the spin factor $`B_j^\mu \psi _{\mu j}^{(S)}`$ is:
$`B_j^\mu \psi _{\mu j}^{(S=0)}\sqrt{{\displaystyle \frac{3}{\pi }}}m\overline{Q}^lϵ^l(\lambda _{ex}),`$
$`B_j^\mu \psi _{\mu j}^{(S=1)}\sqrt{{\displaystyle \frac{3}{4\pi }}}m(\delta ^{jl}\overline{Q}^j\overline{Q}^l)\overline{Q}^jϵ^l(\lambda _{ex})=0,`$
$`B_j^\mu \psi _{\mu j}^{(S=2)}{\displaystyle \frac{2}{\sqrt{13}}}B_j^\mu \psi _{\mu j}^{(S=0)}.`$ (8.47)
From the above it follows
$$\mathrm{\Gamma }_{(S=1)}0,\frac{\mathrm{\Gamma }_{(S=2)}}{\mathrm{\Gamma }_{(S=0)}}\frac{4}{13}.$$
(8.48)
The amplitudes $`A_{\pi \eta }(P)`$ and $`A_{\pi b_1}(P)`$ given by (8.13) are in this limit (for $`S=0`$)
$`A_{\pi M}(P)=2{\displaystyle \frac{4}{3}}{\displaystyle \frac{1}{\sqrt{2}}}m\sqrt{{\displaystyle \frac{3}{\pi }}}{\displaystyle \frac{2m+m_g}{m^2m_g}}{\displaystyle \frac{g}{N_\pi (P)N_M(P)N_{ex}}}{\displaystyle \frac{d^3𝐩}{(2\pi )^3}\frac{d^3𝐫}{(2\pi )^3}\overline{Q}_zY_{J_M0}^{}(𝐪/|𝐪|)}`$
$`\times \psi _L(m_{q\overline{q}}(𝐩,𝐤)/\mu _\pi )\psi _L(m_{q\overline{q}}(𝐫,𝐥)/\mu _M)\psi _L^{}(m_{q\overline{q}}(𝐩,𝐥)/\mu _{ex},m_{q\overline{q}g}(𝐩,𝐥,𝐐)/\mu _{ex}^{}),`$ (8.49)
where $`𝐪=𝐩_{q_2}+𝐏/2`$, $`\lambda _{ex}=0`$ and $`𝐤,𝐥,𝐐`$ were defined in (8.5). As before, $`M`$ denotes $`\eta `$ or $`b_1`$, with $`J=0`$ or $`1`$. In the nonrelativistic limit only the $`S=0`$ and $`S=2`$ hybrid components of the $`\pi _1`$ state contribute to its decays into $`\pi \eta `$ and $`\pi b_1`$. Therefore one may expect the $`S=1`$ component not to be too large for a fully relativistic case.
For $`\pi _1\pi \rho ,\pi f_{1,2}`$ we have
$$C^{ikj}\sqrt{2}imϵ^{0ikj}$$
(8.50)
and the other components are much smaller. This results in
$`C_j^{\mu \nu }\psi _{\mu \nu j}^{(S=0)}0,`$
$`C_j^{\mu \nu }\psi _{\mu \nu j}^{(S=1)}\sqrt{{\displaystyle \frac{3}{4\pi }}}im\overline{Q}^iϵ^j(\lambda _{ex})ϵ^k(\lambda )ϵ^{ikj},`$
$`C_j^{\mu \nu }\psi _{\mu \nu j}^{(S=2)}{\displaystyle \frac{3}{\sqrt{13}}}C_j^{\mu \nu }\psi _{\mu \nu j}^{(S=1)}.`$ (8.51)
Therefore we have
$$\mathrm{\Gamma }_{(S=0)}0,\frac{\mathrm{\Gamma }_{(S=2)}}{\mathrm{\Gamma }_{(S=1)}}\frac{9}{13}.$$
(8.52)
The amplitudes $`A_{\pi \rho }(P)`$ and $`A_{\pi f_{1,2}}(P)`$ given by (8.21) are in this limit (for $`S=1`$)
$`A_{\pi M}(P)=2{\displaystyle \frac{4}{3}}{\displaystyle \frac{1}{\sqrt{2}}}m\sqrt{{\displaystyle \frac{3}{4\pi }}}{\displaystyle \frac{2m+m_g}{m^2m_g}}{\displaystyle \frac{g}{N_\pi (P)N_M(P)N_{ex}}}{\displaystyle \frac{d^3𝐩}{(2\pi )^3}\frac{d^3𝐫}{(2\pi )^3}\overline{Q}_zY_{L_{q\overline{q}(M)}0}^{}(𝐪/|𝐪|)}`$
$`\times \psi _L(m_{q\overline{q}}(𝐩,𝐤)/\mu _\pi )\psi _L(m_{q\overline{q}}(𝐫,𝐥)/\mu _M)\psi _L^{}(m_{q\overline{q}}(𝐩,𝐥)/\mu _{ex},m_{q\overline{q}g}(𝐩,𝐥,𝐐)/\mu _{ex}^{})`$
$`\times (CG)_M,`$ (8.53)
with the notations from (8.5) and $`M`$ standing for $`\rho `$ ($`L_{q\overline{q}}=0`$ and $`(CG)=1`$) or $`f_{1,2}`$ ($`L_{q\overline{q}}=1`$ and $`(CG)=1/\sqrt{2}`$). In the nonrelativistic limit only the $`S=1`$ and $`S=2`$ $`q\overline{q}g`$ component of the $`\pi _1`$ state contribute to its decays into $`\pi \rho `$ and $`\pi f_{1,2}`$. Therefore one may expect the $`S=0`$ component not to be too large for a relativistic case. Similar conlusions can be made for the $`\eta a_{1,2}`$ modes.
Finally, for $`\pi _1\rho \omega `$ we have
$$D^{ijkl}\sqrt{2}m(\delta ^{ij}\delta ^{kl}\delta ^{ik}\delta ^{jl}+\delta ^{il}\delta ^{jk})$$
(8.54)
(the other components are much smaller), which gives:
$`D_j^{\mu \nu \rho }\psi _{\mu \nu \rho j}^{(S=0)}\sqrt{{\displaystyle \frac{3}{\pi }}}m\overline{Q}^iϵ^i(\lambda _{ex})ϵ^j(\lambda _\rho )ϵ^j(\lambda _\omega ),`$
$`D_j^{\mu \nu \rho }\psi _{\mu \nu \rho j}^{(S=1)}\sqrt{{\displaystyle \frac{3}{4\pi }}}m(\overline{Q}^i\delta ^{jk}\overline{Q}^j\delta ^{ik})ϵ^i(\lambda _\rho )ϵ^j(\lambda _\omega )ϵ^k(\lambda _{ex}).`$ (8.55)
The amplitude $`A_{\rho \omega }(P)`$ in this limit is described by formula (8.49) (for $`S=0`$) or (8.53) (for $`S=1`$), where $`\pi `$ is replaced by $`\rho `$ and $`M=\omega `$.
We can straightforwardly understand the difference in amplitudes coming from the spin wave function. If we assume $`m_\eta =m_\rho ,\mu _\eta =\mu _\rho `$ and $`m_{b_1}=m_{f_1}=m_{f_2},\mu _{b_1}=\mu _{f_1}=\mu _{f_2}`$ (the second condition for masses is satisfied to a good approximation), then comparing (8.49) with (8.53) and (8.14) with (8.23) gives
$`A_{\pi \rho }={\displaystyle \frac{1}{2}}A_{\pi \eta }\mathrm{\Gamma }_{\pi \rho }={\displaystyle \frac{1}{2}}\mathrm{\Gamma }_{\pi \eta }`$
$`A_{\pi f_1}={\displaystyle \frac{1}{2\sqrt{2}}}A_{\pi b_1}\mathrm{\Gamma }_{\pi f_1}={\displaystyle \frac{1}{8}}\mathrm{\Gamma }_{\pi b_1}.`$ (8.56)
Here, $`A_{\pi \eta }`$, $`A_{\pi b_1}`$ were taken for $`S=0`$ and $`A_{\pi \rho }`$, $`A_{\pi f_{1,2}}`$ for $`S=1`$. The relation between $`A_{\pi \eta }`$ and $`A_{\pi b_1}`$ (or between $`A_{\pi \rho }`$ and $`A_{\pi f_{1,2}}`$) is more complicated and depends on the orbital angular momentum wave functions $`\psi _L`$ and $`\psi _L^{}`$. If $`\mu _\rho =\mu _\pi `$, then in the nonrelativistic limit $`\pi _1`$ would not decay into $`\pi \rho `$ since the corresponding amplitude becomes proportional to that of the $`\pi \eta `$ mode. Therefore the width of the $`\pi \rho `$ mode is expected to be much smaller than that of $`\pi b_1`$, assuming the parameters $`\mu _{rho}`$ and $`\mu _\pi `$ are very close to one another.
Analogous calculations can be made for decays of $`\pi _1`$ into strange mesons. The amplitudes $`A_{K\overline{K}_1}`$ ($`S_{q\overline{q}}=0,1`$) behave similarly to $`A_{\pi b_1}`$ and $`A_{\pi f_1}`$, whereas $`A_{K\overline{K}^{}}`$ behaves like $`A_{\pi \rho }`$. Therefore the former will be dominant and the latter is expected to be much smaller. If $`\mu _K^{}=\mu _K`$, then in the nonrelativistic limit $`A_{K\overline{K}^{}}=0`$.
### 8.6 Numerical results
In Tables 8.3 and 8.3 we present the widths for various decay modes containing radially ground-state mesons. The numbers in parentheses correspond to calculations using the nonrelativistic formulae, and $`S`$ denotes the total spin of the $`q\overline{q}g`$ component in the $`\pi _1`$ wave function. As before, we are interested in relativistic effects in the $`\pi _1`$ decays. Thus it is sufficient to calculate the amplitudes for each value of $`S`$ separately. The procedure for obtaining the relative contributions for each $`S`$ was given at the beginning of Chapter 7. The values of parameters $`\mu `$ for all unflavored mesons and $`\pi _1`$ were taken equal to $`\mu _\pi `$, and for all strange mesons equal to $`\mu _K`$. This makes the width values for the modes $`\pi \eta `$, $`\pi \eta ^{}`$ and $`\rho \omega `$ identically equal to zero.
The mode $`\pi \rho `$ was expected to be rather small because in the nonrelativistic limit the corresponding amplitude was proportional to that of the mode $`\pi \eta `$, assuming $`\mu _\rho =\mu _\eta `$. We also assumed $`\mu _\rho =\mu _\pi `$. Combining both equalities led to this conclusion. We also found that the $`q\overline{q}g`$ component of $`\pi _1`$ state with $`S=0`$ did not contribute to the hybrid decay in this limit identically. For the relativistic case, however, we observe that the largest contribution in the $`\pi \rho `$ channel comes from the $`S=0`$ component. Our other predictions are confirmed by numerical results, i.e., the $`S=1`$ component for the $`\pi b_1`$ mode and the $`S=0`$ component for the $`\pi f_1`$ mode are negligible in the relativistic model as well.
The decay mode $`\pi b_1`$ is a dominant one. In the following, we will analyze how its widths depend on the free parameters, except the coupling constant $`g`$ which just multiplies the amplitude and $`\mu _{ex}`$ whose equality to $`\mu _\pi `$ seems to be justified. Because the D-wave value in this case is small as compared to the S-wave, we will only deal with the latter. Furthermore, in the following plots we will assume $`S=0`$ since $`S=1`$ gives negligible contribution (in the nonrelativistic limit it is zero) and thus the widths for $`S=2`$ are proportional to those of $`S=0`$ (8.48). The default values of the free parameters are those from Table 8.3.
In Fig. 8.2 we compare relativistic and nonrelativistic predictions for the width of the decay $`\pi _1\pi b_1`$ as a function of the mass of the light quark $`m`$. It also shows the semirelativistic values which include relativistic phase space and orbital wave functions, but with no Wigner rotations. For larger values of $`m`$ all three curves R, NR and SR converge, as it should be. The corresponding ratios NR/R and SR/R are shown in Fig. 8.4.
Let us proceed to analyze the contribution of the other free parameters. We show only relativistic and semirelativistic rates because the nonrelativistic ones correspond to different orbital wave functions and we are interested in corrections arising from Wigner rotations. The dependence of the width of $`\pi _1\pi b_1`$ on the mass of $`\pi _1`$ is presented in Fig. 8.4. We observe that Wigner rotations contribute more for larger values of $`m_{ex}`$, and at 2000 MeV the ratio SR/R is already about 50%. We also see that the maximal width occurs for $`m_{ex}`$ in the range $`15001600`$ MeV, i.e., for the values close to those experimentally reported. Thus, our predictions for the width are an upper limit with respect to $`m_{ex}`$.
In Fig. 8.6 we show the width dependence on $`\mu _{b_1}`$. If the value of $`\mu _{b_1}`$ is equal to $`\mu _\pi `$, then the width is approximately at its maximum. Probably it is a coincidence, but it tells us that our prediction is an upper limit for this width with respect to $`\mu _{b_1}`$. The corresponding ratio SR/R is given in Fig. 8.6. At masses around 1 GeV it already behaves like a constant.
There is one parameter whose choice was not justified, $`\mu _{ex^{}}`$. This quantity divides the invariant mass of the $`q\overline{q}g`$ system, and there is no reason why it should be equal to $`\mu _\pi `$, as we assumed. Its value may be obtained from decays of normal mesons such as $`\rho `$ or $`b_1`$. In Fig. 8.8 we show how the width of $`\pi _1\pi b_1`$ as a function of $`\mu _{ex^{}}`$. The ratio SR/R is given in Fig. 8.8 and we see that it is close to a constant, i.e., the choice of $`\mu _{ex^{}}`$ is crucial for the widths but not for relative relativistic corrections.
Finally, in Fig. 8.10 we show the width dependence on the effective mass of the gluon $`m_g`$, and in Fig. 8.10 the corresponding ratio SR/R. The latter displays clearly that the ratio is essentially independent of $`m_g`$. We observe that the relative relativistic corrections arising from Wigner rotations are insensitive to the free parameters referring to the $`\pi _1`$ at rest, i.e., $`m_{ex^{}}`$ and $`m_g`$.
Now we move to the decays into $`\pi `$ and various $`\eta `$ mesons. Since the amplitude vanishes for $`\mu _\eta =\mu _\pi `$ we want to observe what happens if $`\mu _\eta \mu _\pi `$. For brevity, we will consider only the relativistic case. In Fig. 8.11 we show how the width of the decay $`\pi _1\pi \eta `$ depends on the value of $`\mu _\eta `$, while the other parameters are the defaults. The biggest value approached is only a few MeV. That confirms our previous predictions and is in a good agreement with other models . The same dependence for the decay $`\pi _1\pi \eta ^{}(958)`$ is displayed in Fig. 8.12, and the maximum is around 10 MeV.
We see that the maximum of the width occurs when $`\mu _\eta `$ is approximately equal to 100 MeV. In Fig. 8.13 and Fig. 8.14 we show the width dependence on the mass of $`\pi _1`$ at this particular value $`\mu _\eta ^{max}`$, for $`\pi _1\pi \eta `$ and $`\pi _1\pi \eta ^{}`$ respectively. The reason for this is to see the largest values one can obtain for these decays. It turns out that these values have a maximum on the order of 10 MeV. However, the value of $`\mu _\eta `$ should not differ much from $`\mu _\pi `$ and thus the above widths will be on the order of 1 MeV.
In Tables 8.3 and 8.3 we presented the widths only for the modes including mesons with the lowest radial quantum number such as $`\pi `$ and $`K`$. It allowed us to choose the free parameters $`\mu `$ equal to either $`\mu _\pi `$ or $`\mu _K`$. This, however, cannot be justified for decays into mesons with higher values of the radial quantum number such as $`\eta (1295)`$ or $`\rho (1450)`$. The dependence of the decay width on the value of $`\mu _\eta `$ for $`\pi _1\pi \eta (1295)`$ with $`S=0`$ and $`m_{ex}=`$ 2 GeV is shown in Fig. 8.15. This mode may be a significant one (the width is at most about 10 MeV), which agrees with the results obtained in . Unfortunately, there is not enough experimental data to constrain the free parameters $`\mu `$ for $`\eta (1295)`$ and other radially excited mesons.
From these results it is clear that fully relativistic values are significantly different from nonrelativistic ones. There are two sources of this: the Wigner rotation which introduces relativistic coupling between spin and spatial degrees of freedom in the wave function, and different relations between energy, momentum and the invariant masses (in the phase space and most of all in the orbital wave functions). Both corrections actually appear to introduce corrections as large as 10% and thus should be included in phenomenological models.
We also observe that in our relativistic constituent model the $`S+P`$ selection rule, which was mentioned in Chapter 2, is obeyed. According to this rule the favored modes for the $`\pi _1`$ decay are $`\pi b_1`$, $`\pi f_{1,2}`$, $`\eta a_{1,2}`$, and both channels $`K\overline{K}_1`$. Our results support this prediction and agree with other nonrelativistic models .
## Chapter 9 Final state interactions
In the previous chapters we constructed meson states and studied kinematical relativistic effects at the quark and gluon level. In this chapter we will estimate the size of corrections to the $`\pi _1`$ decays originating from the meson exchange forces between mesons. Since the particle number is not conserved and momenta of particles are on the same order as their masses, this problem should be treated in a relativistic formalism. We will begin with the Lippmann-Schwinger equation applied to the $`\pi \rho `$ and $`\pi b_1`$ states. Then we will describe the computational procedure of solving the resulting integral equations. At the end of this chapter the numerical results will be given.
### 9.1 The Lippmann-Schwinger equation
Other models predict larger widths for the process $`\pi _1\pi \rho `$ than our model. The numerical values were given in Chapter 2. It is possible that this width is increased by final state interactions between the outgoing mesons. The $`b_1`$ created in the process $`\pi _1\pi b_1`$ can subsequently decay into $`\pi `$ and $`\omega `$, and then the $`\omega `$ can absorb the other $`\pi `$ and create the $`\rho `$, as shown in Fig. 9.1. Of course the reverse $`\pi \rho \pi b_1`$ process can occur as well. An $`\omega `$ exchange may occur more than once and we must sum up all possible amplitudes; the lowest order diagrams are presented in Fig. 9.2.
A bold horizontal solid line represents a hybrid and normal horizontal solid lines refer to mesons. A vertical dashed line corresponds to a single $`\omega `$ meson exchange. In order to describe a total contribution of the final state interactions to the decay widths of $`\pi _1`$, we need to solve the Lippmann-Schwinger equation
$$T=V+VGT,$$
(9.1)
which is equivalent to summing over all diagrams, including those with intermediate hybrid states.
Let the $`\pi _1`$ state be denoted by index $`\alpha `$, and the states $`\pi \rho `$ and $`\pi b_1`$ by Roman letters. For simplicity we will assume the $`\pi _1`$ spin wave function has only the $`S=0`$ $`q\overline{q}g`$ component; that gives the largest widths for both $`\pi b_1`$ and $`\pi \rho `$ modes. Thus our calculations should give the upper limit for FSI corrections. Let us introduce the matrix elements for the potential $`V`$:
$$V_{\alpha i}=\alpha |V|j,V_{ij}=i|V|j,$$
(9.2)
and similarly for $`T`$. The elements $`V_{\alpha i}`$ are just the amplitudes of the corresponding decays of $`\pi _1`$, whereas $`V_{ij}`$ are related to the final state interaction potential. In our state-space the Lippmann-Schwinger equation can be written as
$`T_{\alpha i}=V_{\alpha i}+V_{\alpha j}G_jT_{ji},`$
$`T_{ji}=V_{ji}+V_{jk}G_kT_{ki}+V_{j\alpha }G_\alpha T_{\alpha i},`$ (9.3)
where
$$G_i(E)=[EH_0(i)+iϵ]^1.$$
(9.4)
Both equations (9.3) are represented diagrammatically in Fig. 9.3.
A bold vertical solid line corresponds to the total amplitude of the interaction between two dimeson states, whereas a normal vertical solid line represents the sum over all intermediate states.
Hereinafter, if an index appears twice or more, then summation over this index is implicitly assumed. The Hamiltonian of the state $`|i`$ is $`H_0(i)`$, $`ϵ0^+`$, and $`E`$ is the energy which will equal the mass of the $`\pi _1`$. Calculating $`T_{ij}`$ from the second equation in (9.3) and its substitution to the first one leads to
$$T_{\alpha i}=V_{\alpha i}+V_{\alpha j}G_j[1VG]_{jk}^1(V_{ki}+V_{k\alpha }G_\alpha T_{\alpha i}).$$
(9.5)
Now we introduce the $`T`$ matrix acting only between states $`|\pi \rho `$ and $`|\pi b_1`$, denoted by $`t`$ and defined by
$$t=V+VGt,t_{ij}=[1VG]_{ik}^1V_{kj}.$$
(9.6)
Diagrammatically this represents the sum of all diagrams without hybrid intermediate states, as shown in Fig. 9.4.
Therefore
$`T_{\alpha i}=V_{\alpha i}+V_{\alpha j}G_jt_{ji}+V_{\alpha j}G_j[1VG]_{jk}^1V_{k\alpha }G_\alpha T_{\alpha i}=`$
$`=V_{\alpha i}+V_{\alpha j}G_jt_{ji}+V_{\alpha j}G_j[\delta _{jk}+(tG)_{jk}]V_{k\alpha }G_\alpha T_{\alpha i},`$ (9.7)
where we used $`[1VG]^1=1+tG`$ in the last term.
Using the identity $`[1GV]^1=1+Gt`$ on the first two terms on RHS leads to
$$T_{\alpha i}=(V[1GV]^1)_{\alpha i}+[V(1+Gt)GV]_{\alpha \alpha }G_\alpha T_{\alpha i}.$$
(9.8)
Solving for $`T_{\alpha i}`$ gives
$$T_{\alpha i}=[1(V(1+Gt)GV)_{\alpha \alpha }G_\alpha ]^1(V[1GV]^1)_{\alpha i}=G_\alpha ^1[G_\alpha ^1\mathrm{\Sigma }_\alpha ]^1(V[1GV]^1)_{\alpha i},$$
(9.9)
where $`\mathrm{\Sigma }_\alpha =(V(1+Gt)GV)_{\alpha \alpha }`$ is the self-energy of the $`\pi _1`$, shown in Fig. 9.5.
Both $`G_\alpha ^1`$ and $`\mathrm{\Sigma }_\alpha `$ are numbers, and thus
$$T_{\alpha i}=\frac{Em_{ex}}{Em_{ex}\mathrm{\Sigma }_\alpha }(V+VGt)_{\alpha i},$$
(9.10)
where we again used $`[1GV]^1=1+Gt`$. Renormalization of the theory is related to cutting off the self-energy term. Therefore we obtain finally
$$T_{\alpha i}=V_{\alpha i}+V_{\alpha j}G_jt_{ji},$$
(9.11)
with $`t_{ji}`$ defined by (9.6). This is equivalent to excluding an exotic intermediate state from the FSI corrections. Diagrammatic representation of Eq. (9.11) is given in Fig. 9.6.
Eq. (9.6), written in a full notation in the rest frame of the $`\pi _1`$, is
$$t(𝐩,𝐪,\lambda ,\lambda ^{})=V(𝐩,𝐪,\lambda ,\lambda ^{})+\underset{\lambda \mathrm{"}}{}\frac{d^3𝐤}{(2\pi )^34\omega _1(k)\omega _2(k)}V(𝐩,𝐤,\lambda ,\lambda \mathrm{"})G(k)t(𝐤,𝐪,\lambda \mathrm{"},\lambda ^{}),$$
(9.12)
where
$`\omega _i(k)=E(m_i,𝐤),H_0(k)=\omega _1(k)+\omega _2(k),`$
$`G(k)=[EH_0(k)+iϵ]^1,`$ (9.13)
and $`m_i(i=1,2)`$ are the meson masses in an intermediate two-meson state which is related to $`G`$. In the above $`𝐩`$ and $`𝐪`$ are the relativistic relative momenta between two mesons, and the dependence on the center-of-mass momentum has been already separated. Spins $`\lambda `$ refer to either $`b_1`$ or $`\rho `$.
We introduce the partial wave potentials
$`V_{LL^{}}(p,q)={\displaystyle \underset{M,M^{},\lambda ,\lambda ^{},j}{}}{\displaystyle 𝑑\mathrm{\Omega }_𝐩𝑑\mathrm{\Omega }_𝐪L,M;1,\lambda |J,jL^{},M^{};1,\lambda ^{}|J,j}`$
$`\times V(𝐩,𝐪,\lambda ,\lambda ^{})Y_{LM}(𝐩)Y_{L^{}M^{}}^{}(𝐪),`$ (9.14)
where $`d\mathrm{\Omega }_𝐤`$ is the element of the solid angle in the direction of the vector $`𝐤`$ and $`k=|𝐤|`$. Similarly we define $`t_{LL^{}}(p,q)`$. Substitution of $`V_{LL^{}}(p,q)`$ into (9.12) gives
$$t_{LL^{}}(p,q)=V_{LL^{}}(p,q)+\underset{L\mathrm{"}}{}\frac{k^2dk}{(2\pi )^34\omega _{L\mathrm{"},1}(k)\omega _{L\mathrm{"},2}(k)}V_{LL\mathrm{"}}(p,k)G_{L\mathrm{"}}(k)t_{L\mathrm{"}L^{}}(k,q),$$
(9.15)
where
$$\omega _{L\mathrm{"},i}(k)=E(m_{L\mathrm{"},i},𝐤),G_{L\mathrm{"}}(k)=[E\omega _{L\mathrm{"},1}(k)\omega _{L\mathrm{"},2}(k)+iϵ]^1.$$
(9.16)
In our state-space we can have $`L=0,2`$ (the relative angular momentum between $`\pi `$ and $`b_1`$) or $`L=1`$ (between $`\pi `$ and $`\rho `$). For a $`\pi _1`$ we also have $`J=1`$. Thus in Eq. (9.15) we must substitute:
$$m_{L,1}=m_\pi ,m_{0,2}=m_{2,2}=m_{b_1},m_{1,2}=m_\rho .$$
(9.17)
From the conservation of parity the only nonzero components of the final state interaction potential are $`V_{01}`$, $`V_{10}`$, $`V_{12}`$ and $`V_{21}`$. Moreover, from the CP conservation for strong interactions we have $`V_{01}=V_{10}^{}`$ and $`V_{12}=V_{21}^{}`$. The integral equation (9.15) cannot be solved in general analytically and one need to replace it with a set of matrix equations. The details will be given in the next section. When this is done and $`t_{LL^{}}`$ are found, we may go back to Eq. (9.11) which becomes
$$\stackrel{~}{a}_L(P)=a_L(P)+\underset{L^{}}{}\frac{k^2dk}{(2\pi )^3(2J+1)4\omega _1(k)\omega _2(k)}a_L^{}(k)G(k)_L^{}t_{L^{}L}(k,P),$$
(9.18)
where $`a_L(P)`$ are the partial amplitudes defined in (8.13) and (8.21). In order to obtain the corrected widths, these amplitudes must be replaced by the corrected amplitudes $`\stackrel{~}{a}_L(P)`$ in Eq. (7.38).
Finally, we proceed to the form of the final state interaction potential. The Lagrangian of the $`\rho \pi \omega `$ vertex is
$$L_{\rho \pi \omega }=g_{\rho \pi \omega }ϵ^{\mu \nu \lambda \sigma }_\mu \omega _\nu \pi ^i_\lambda \rho _\sigma ^i,$$
(9.19)
and for the $`b_1\pi \omega `$ vertex is
$$L_{b_1\pi \omega }=g_{b_1\pi \omega }_\mu b_1^{\mu i}\pi ^i\omega _\mu ,$$
(9.20)
where $`g_{\rho \pi \omega }`$ and $`g_{b_1\pi \omega }`$ are the hadrodynamical coupling constants and the index $`i`$ corresponds to isospin. The first one lies between 0.01 MeV<sup>-1</sup> and 0.02 MeV<sup>-1</sup> and we will take a value 0.014 MeV<sup>-1</sup> , whereas the second one can be obtained by calculating the width of the decay $`b_1\pi \omega `$ and comparing with its experimental value 142 MeV . Consequently, the amplitude of $`b_1\pi \omega `$ is equal to
$$A(𝐏,\lambda _{b_1},\lambda _\omega )=g_{b_1\pi \omega }ϵ^i(\lambda _{b_1})ϵ^i(\lambda _\omega ,𝐏),$$
(9.21)
where $`𝐏`$ is given by
$$E(m_\pi ,𝐏)+E(m_\omega ,𝐏)=m_{b_1}.$$
(9.22)
The width for the angular momentum $`L`$ between $`\pi `$ and $`\omega `$ is
$$\mathrm{\Gamma }_L=\frac{P|A|^2}{8\pi (2L+1)m_{b_1}^2},$$
(9.23)
where $`|A|^2`$ is summed over $`\lambda _\omega `$ and averaged with respect to $`\lambda _{b_1}`$. With the help of
$$\underset{\lambda }{}ϵ^\mu (\lambda ,𝐩)ϵ^\nu (\lambda ,𝐩)=g^{\mu \nu }+\frac{p^\mu p^\nu }{p_\rho p^\rho },$$
(9.24)
we have
$$|A|^2=g_{b_1\pi \omega }^2(1+\frac{P^2}{3m_\omega ^2}),$$
(9.25)
and from $`\mathrm{\Gamma }_{swave}=142`$ MeV$`\mathrm{\hspace{0.17em}0.92}=131`$ MeV (the factor 0.92 comes from the partial wave D/S ratio) we get $`g_{b_1\pi \omega }=3650`$ MeV. Thus $`g_{FSI}=g_{\rho \pi \omega }g_{b_1\pi \omega }`$ is around 50.
The final state interaction potential can be obtained from Lagrangians (9.19) and (9.20), written in the momentum space and dressed with the instantaneous $`\omega `$ propagator,
$$V(𝐩,𝐪,\lambda _{b_1},\lambda _\rho )=g_{FSI}ϵ_{\mu \nu \sigma \tau }p^\mu q^\nu \frac{1}{(𝐩𝐪)^2+m_\omega ^2}ϵ^\sigma (\lambda _{b_1},𝐩)ϵ^\tau (\lambda _\rho ,𝐪),$$
(9.26)
where $`𝐩`$ is the momentum of the $`\pi `$ in the $`|\pi b_1`$ state and $`𝐪`$ is the momentum of the $`\rho `$. For large values of $`p`$ and $`q`$ this potential grows with no limit which is a consequence of treating mesons as elementary particles. Therefore we must regulate this potential with an extra factor that tends to zero for large momenta. We choose an exponential function,
$$e^{|𝐩_\omega |/\mathrm{\Lambda }},$$
(9.27)
of the quantity which is invariant under translations and rotations. Here $`\mathrm{\Lambda }`$ is a scale parameter of order 1 GeV. This scale estimates a limit of treating mesons as elementary particles interacting via Lagrangians (9.19) and (9.20).
### 9.2 Computational procedure
The Lippmann-Schwinger integral equation (9.15) corresponds to outgoing wave boundary conditions. This means that the singularity of the term $`G(k)`$ is handled by giving the energy $`E`$ a small positive imaginary part $`iϵ`$. An integral of this form may be solved using the Cauchy principal-value prescription,
$$_0^{\mathrm{}}\frac{f(k)dk}{kk_0+iϵ}=\mathrm{}_0^{\mathrm{}}\frac{f(k)dk}{kk_0}i\pi f(k_0),$$
(9.28)
where the principal value is defined by
$$\mathrm{}_0^{\mathrm{}}f(k)𝑑k=\underset{ϵ0}{lim}\left[_0^{k_0ϵ}f(k)𝑑k+_{k_0+ϵ}^{\mathrm{}}f(k)𝑑k\right],$$
(9.29)
with $`k_0`$ being the zero of the real function $`f(k)`$. From the formula
$$\mathrm{}_{\mathrm{}}^+\mathrm{}\frac{dk}{kk_0}=0$$
(9.30)
one obtains
$$\mathrm{}_0^{\mathrm{}}\frac{dk}{k^2k_0^2}=0,$$
(9.31)
that can be generalized to the so-called Hilbert transform of a function $`f`$:
$$\mathrm{}_0^{\mathrm{}}\frac{f(k)dk}{k^2k_0^2}=_0^{\mathrm{}}\frac{[f(k)f(k_0)]dk}{k^2k_0^2}.$$
(9.32)
The integral in Eq. (9.15) has a slightly different singular denominator term $`[EH_0(k)]^1`$, and in this case
$`{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{f(k)dk}{EH_0(k)+iϵ}}=\mathrm{}{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{f(k)dk}{EH_0(k)}}i\pi f(k_0)\left({\displaystyle \frac{H_0(k)}{k}}\right)_{k=k_0}^1=`$
$`=\mathrm{}{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{f(k)dk}{k^2k_0^2}}{\displaystyle \frac{k^2k_0^2}{EH_0(k)}}i\pi f(k_0){\displaystyle \frac{\omega _1(k_0)\omega _2(k_0)}{k_0E}}=`$
$`={\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{dk}{k^2k_0^2}}\left[{\displaystyle \frac{f(k)(k^2k_0^2)}{EH_0(k)}}f(k_0)\underset{kk_0}{lim}{\displaystyle \frac{k^2k_0^2}{EH_0(k)}}\right]i\pi f(k_0){\displaystyle \frac{\omega _1(k_0)\omega _2(k_0)}{k_0E}}=`$
$`={\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{dk}{k^2k_0^2}}\left[{\displaystyle \frac{f(k)(k^2k_0^2)}{EH_0(k)}}{\displaystyle \frac{2f(k_0)\omega _1(k_0)\omega _2(k_0)}{E}}\right]i\pi f(k_0){\displaystyle \frac{\omega _1(k_0)\omega _2(k_0)}{k_0E}}.`$ (9.33)
Making a transition
$$f(k)\frac{k^2V(p,k)t(k,q)}{4(2\pi )^3\omega _1(k)\omega _2(k)},$$
(9.34)
and including the summation over partial waves, leads to a desired equation for $`t_{LL^{}}(p,q)`$ that no longer has a singularity:
$`t_{LL^{}}(p,q)=V_{LL^{}}(p,q)+{\displaystyle \underset{L\mathrm{"}}{}}{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{dk}{4(2\pi )^3(k^2k_{0,L\mathrm{"}}^2)}}[{\displaystyle \frac{k^2(k^2k_{0,L\mathrm{"}}^2)V_{LL\mathrm{"}}(p,k)t_{L\mathrm{"}L^{}}(k,q)}{[EH_{0,L\mathrm{"}}(k)]\omega _{L\mathrm{"},1}(k)\omega _{L\mathrm{"},2}(k)}}`$
$`+{\displaystyle \frac{2k_{0,L\mathrm{"}}^2}{E}}V_{LL\mathrm{"}}(p,k_{0,L\mathrm{"}})t_{L\mathrm{"}L^{}}(k_{0,L\mathrm{"}},q)]i\pi {\displaystyle }_{L\mathrm{"}}{\displaystyle \frac{k_{0,L\mathrm{"}}V_{LL\mathrm{"}}(p,k_{0,L\mathrm{"}})t_{L\mathrm{"}L^{}}(k_{0,L\mathrm{"}},q)}{4(2\pi )^3E}},`$ (9.35)
where
$$H_{0,L}(k)=\omega _{L,1}(k)+\omega _{L,2}(k)$$
(9.36)
and the quantities $`k_{0,L}`$ are defined by
$$H_{0,L}(k_{0,L})=E.$$
(9.37)
Because in the decay $`\pi _1\pi b_1`$ the D-wave amplitude is negligible as compared to that of the S-wave, we may reduce our angular momentum space to $`L=0,1`$. Therefore we need only the expression for $`V_{01}(p,q)`$. From the definition (9.14) and with the help of (5.31) we obtain
$$V_{01}(p,q)=\frac{i\sqrt{3}}{4\pi }\underset{\lambda ,\lambda ^{}}{}𝑑\mathrm{\Omega }_𝐩𝑑\mathrm{\Omega }_𝐪V(𝐩,𝐪,\lambda ,\lambda ^{})ϵ^{ijk}ϵ^i(\lambda )ϵ^j(\lambda ^{})q^k/q,$$
(9.38)
where $`V(𝐩,𝐪,\lambda ,\lambda ^{})`$ was given in (9.26). The S-matrix is obtained from the $`t`$ matrix via
$$S_{LL^{}}=\delta _{LL^{}}i\frac{\sqrt{k_{0,L}k_{0,L^{}}}}{16\pi ^2E}t_{LL^{}}(k_{0,L},k_{0,L^{}}),$$
(9.39)
and for a $`2\times 2`$ unitary matrix may be parametrized by scattering phase shifts $`\delta _0`$ and $`\delta _1`$,
$$S_{LL^{}}=\left(\begin{array}{cc}\eta e^{2i\delta _0}& i\sqrt{1\eta ^2}e^{i(\delta _0+\delta _1)}\\ i\sqrt{1\eta ^2}e^{i(\delta _0+\delta _1)}& \eta e^{2i\delta _1}\end{array}\right),$$
(9.40)
with $`0\eta 1`$.
Eq. (9.35) may be solved, as we already mentioned, by converting the integration over $`k`$ into the sum over $`N`$ integration points $`k_n,n=1,2,\mathrm{}N`$ (determined by Gaussian quadratures) with weights $`w_n`$ :
$`t_{LL^{}}(p,q)=V_{LL^{}}(p,q)+{\displaystyle \underset{L\mathrm{"},n}{}}{\displaystyle \frac{w_n}{4(2\pi )^3(k_n^2k_{0,L\mathrm{"}}^2)}}[{\displaystyle \frac{k_n^2(k_n^2k_{0,L\mathrm{"}}^2)V_{LL\mathrm{"}}(p,k_n)t_{L\mathrm{"}L^{}}(k_n,q)}{[EH_{0,L\mathrm{"}}(k_n)]\omega _{L\mathrm{"},1}(k_n)\omega _{L\mathrm{"},2}(k_n)}}`$
$`+{\displaystyle \frac{2k_{0,L\mathrm{"}}^2}{E}}V_{LL\mathrm{"}}(p,k_{0,L\mathrm{"}})t_{L\mathrm{"}L^{}}(k_{0,L\mathrm{"}},q)]i\pi {\displaystyle }_{L\mathrm{"}}{\displaystyle \frac{k_{0,L\mathrm{"}}V_{LL\mathrm{"}}(p,k_{0,L\mathrm{"}})t_{L\mathrm{"}L^{}}(k_{0,L\mathrm{"}},q)}{4(2\pi )^3E}}.`$ (9.41)
By taking $`p,q`$ equal to either $`k_n`$ or $`k_{0,L}`$ we can rewrite this equation as the $`(2N+2)\times (2N+2)`$ matrix equation
$$t_{mn}=V_{mn}+R_{mo}t_{on},m,n,o=\mathrm{1..2}N+2,$$
(9.42)
where
$`V_{mn}=0`$ $`m=1..N+1,n=1..N+1;`$ (9.43)
$`V_{mn}=V_{01}(l_m,l_n)`$ $`m=1..N+1,n=N+\mathrm{2..2}N+2;`$
$`V_{mn}=V_{01}(l_m,l_n)`$ $`m=N+\mathrm{2..2}N+2,n=1..N+1;`$
$`V_{mn}=0`$ $`m=N+\mathrm{2..2}N+2,n=N+\mathrm{2..2}N+2`$
and
$`l_n=k_n`$ $`n=1..N;`$ (9.44)
$`l_{N+1}=k_{0,0};`$
$`l_n=k_{nN1}`$ $`n=N+\mathrm{2..2}N+1;`$
$`l_{2N+2}=k_{0,1}.`$
The matrix $`R`$ is defined by
$$R_{mn}=\delta _{mn}V_{mn}D_n$$
(9.45)
with no summation over $`n`$, where
$`D_n={\displaystyle \frac{k_n^2w_n}{4(2\pi )^3\omega _{0,1}(l_n)\omega _{0,2}(l_n)[EH_{0,0}(l_n)]}}`$ $`n=1..N;`$ (9.46)
$`D_{N+1}={\displaystyle \frac{k_{0,0}}{4(2\pi )^3E}}\left[{\displaystyle \underset{j=n}{\overset{N}{}}}{\displaystyle \frac{2k_{0,0}w_n}{k_n^2k_{0,0}^2}}i\pi \right];`$
$`D_n={\displaystyle \frac{k_n^2w_n}{4(2\pi )^3\omega _{1,1}(l_n)\omega _{1,2}(l_n)[EH_{0,1}(l_n)]}}`$ $`n=N+\mathrm{2..2}N+1;`$
$`D_{2N+2}={\displaystyle \frac{k_{0,1}}{4(2\pi )^3E}}\left[{\displaystyle \underset{n=1}{\overset{N}{}}}{\displaystyle \frac{2k_{0,1}w_n}{k_n^2k_{0,1}^2}}i\pi \right].`$
Having inverted $`R_{mn}`$ and solved for $`t_{mn}`$, we subsitute it into (9.18) that in $`2N+2`$-dimensional space has the form
$`\stackrel{~}{a}_0(k_{0,0})=a_0(k_{0,0})+{\displaystyle \underset{n=1}{\overset{N+1}{}}}D_na_0(l_n)t_{n,N+1}+{\displaystyle \underset{n=N+2}{\overset{2N+2}{}}}D_na_1(l_n)t_{n,N+1},`$
$`\stackrel{~}{a}_1(k_{0,1})=a_1(k_{0,1})+{\displaystyle \underset{n=1}{\overset{N+1}{}}}D_na_0(l_n)t_{n,2N+2}+{\displaystyle \underset{n=N+2}{\overset{2N+2}{}}}D_na_1(l_n)t_{n,2N+2}.`$ (9.47)
The quantities $`k_{0,0}`$ and $`k_{0,1}`$ are according to (9.37) the relative momenta of $`\pi b_1`$ and $`\pi \rho `$ systems, respectively. Thus Eq. (9.47) gives the FSI-corrected amplitudes of $`\pi _1\pi b_1`$ and $`\pi _1\pi \rho `$ as functions of the energy $`E`$ that is equal to $`m_{ex}`$. The S-matrix in terms of $`t_{mn}`$ is
$$S_{ij}=\delta _{ij}i\frac{\sqrt{l_{i(N+1)}l_{j(N+1)}}}{16\pi ^2E}t_{i(N+1),j(N+1)},$$
(9.48)
where $`i,j=1,2`$ refer to channels $`\pi b_1`$ and $`\pi \rho `$.
### 9.3 Numerical results
The partial wave potential (9.38) obtained from the $`\pi b_1\pi \rho `$ final state interaction potential (9.26) is not separable, i.e., cannot be represented as a product of functions of only one variable,
$$V(p,q)f(p)g(q).$$
(9.49)
This corresponds to non-locality of this potential in the position space and results from a finite structure of mesons. If they were elementary and therefore $`V(p,q)`$ was a separable potential, then Eq. (9.35) could be solved analytically. In this section we will assume that this is the case and use it to test the accuracy of numeric results from the preceding section.
Consider the following symmetric and off-diagonal potential,
$$V_{LL^{}}(p,q)=\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right)g(p)g(q).$$
(9.50)
We seek a solution of (9.35) in the form
$$t_{LL^{}}(p,q)=\left(\begin{array}{cc}\lambda _{00}& \lambda _{01}\\ \lambda _{01}& \lambda _{11}\end{array}\right)g(p)g(q).$$
(9.51)
Substitution of the above matrices into (9.35) leads to
$$\left(\begin{array}{cc}\lambda _{00}& \lambda _{01}\\ \lambda _{01}& \lambda _{11}\end{array}\right)=\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right)+\left(\begin{array}{cc}I_1\lambda _{01}& I_1\lambda _{11}\\ I_0\lambda _{00}& I_0\lambda _{01}\end{array}\right),$$
(9.52)
where the quantities $`I_0`$ and $`I_1`$ are given by the integrals:
$`I_0={\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{k^2dkg^2(k)}{4(2\pi )^3\omega _{0,1}(k)\omega _{0,2}(k)[EH_{0,0}(k)+iϵ]}}`$
$`I_1={\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{k^2dkg^2(k)}{4(2\pi )^3\omega _{1,1}(k)\omega _{1,2}(k)[EH_{0,1}(k)+iϵ]}}.`$ (9.53)
For the $`t`$ matrix we obtain then
$$t_{LL^{}}(p,q)=\left(\begin{array}{cc}I_1& 1\\ 1& I_0\end{array}\right)\frac{g(p)g(q)}{1I_0I_1},$$
(9.54)
and the phase shifts can be computed from (9.39) and (9.40). The integrals $`I_0`$ and $`I_1`$ may be easily calculated using the principal-value prescription described in the preceding section.
In Fig. 9.7 we show the numerical values of the phase shift $`\delta _1`$ in degrees as a function of the number of grid points $`N`$, and compared to the corresponding analytical value. The toy potential was chosen such that
$$g(p)g(q)=V_0e^{(p^2+q^2)/\mu ^2},$$
(9.55)
with $`V_0=`$1000 (the FSI potential is dimensionless) and $`\mu =`$1 GeV, whereas the energy $`E=m_{ex}=`$1600 MeV. We see that the numeric results for this phase shift converge pretty fast to a value that is in a good agreement with the analytical one. Accurate and stable numbers are obtained already for $`N`$ being approximately 10.
Now we move to the original final state interaction and replace our toy potential with the real potential (9.38) obtained from Eq. (9.26). Fig. 9.8 and Fig. 9.9 show an increasing stability of phase shifts $`\delta _0`$ and $`\delta _1`$ (in degrees) for $`g_{FSI}=`$200 as the number of grid points $`N`$ grows. The default values in these plots are $`E=`$1600 MeV and $`\mathrm{\Lambda }=`$2 GeV. We see that, for an unseparable potential, a larger $`N`$ is needed to obtain stable results. For $`N=`$20 the discrepancy is still about $`1^{}`$. The unitarity condition is satisfied here with accuracy better than 0.1%.
Let us assume that the coupling constant $`g_{FSI}`$ is the variable and we introduce the potential strength $`I=g_{FSI}/g_{FSI}^{(0)}`$, where $`g_{FSI}^{(0)}=50`$. In Fig. 9.10 and Fig. 9.11 we present the phase shifts as functions of $`I`$. Their dependence on the energy $`E`$ for $`I=1`$ is shown in Fig. 9.12.
Within an experimental range of the hybrid mass $`m_{ex}=E`$ and the hadrodynamical coupling constant $`g`$ the phase shifts for both channels $`L=0`$ ($`\pi b_1`$) and $`L=1`$ ($`\pi \rho `$) are rather small. Therefore one expects that final state interaction would not change much the widths of $`\pi _1\pi b_1`$ and $`\pi _1\pi \rho `$.
In Tables 9.1 and 9.2 we compare the original widths (in MeV) with the FSI-corrected ones for various values of $`m_{ex}`$ and $`\mathrm{\Lambda }`$, for $`I=2`$. The strength was doubled in order to recompense damping caused by the regulating factor in the FSI potential. For $`\mathrm{\Lambda }`$ other than 2 GeV, the coupling constant $`g_{FSI}`$ was renormalized such that the value of $`V_{01}(p_0,q_0)`$ remained equal to that for $`\mathrm{\Lambda }=`$ 2 GeV. Here $`p_0`$ and $`q_0`$ are the physical values of the relative momenta between two mesons in the $`\pi b_1`$ and $`\pi \rho `$ channels, respectively. In Fig. 9.14 and Fig. 9.14 we show the width dependence on $`I`$ for $`m_{ex}=1.6`$ GeV and $`\mathrm{\Lambda }=2`$ GeV. The behaviour of $`\delta _1`$ indicates the existence of the $`\pi \rho `$ resonances for the free parameters given (a resonance occurs when the phase increases over $`n90^{}`$, where $`n=1,2,3\mathrm{}`$).
The numerical results confirm our predictions. Only large FSI potentials may change the width values for both modes by more than few MeV. However, the $`\pi \rho `$ channel is subjected to a rather large relative correction. It may be caused by a big value of the original width for the $`\pi b_1`$ mode. In any event, the width for the $`\pi _1\pi \rho `$ is always on the order of a few MeV which agrees with or and does not, for example, with .
There is one constraint for the possible values of the cut-off parameter $`\mathrm{\Lambda }`$ and the coupling constant $`g_{FSI}`$, given by the decay of the $`\pi _2(1670)`$. This meson almost does not decay into $`\pi b_1`$ but significantly does into $`\pi \rho `$. If the corrections from the final state interaction $`\pi b_1\pi \rho `$ were large, the $`\pi b_1`$ mode width could be increased by the $`\pi \rho `$ channel, resulting in a disagreement with experiment. Thus we arrive at the conclusion that FSI cannot introduce corrections that are too large. This agrees with our predictions for real $`I`$.
## Chapter 10 Conclusions
The quantitative description of the dynamics of gluons is complicated by the strong character of their interactions. Hadrons with excited gluonic degrees of freedom seem to be the only way to obtain information about the nature of low-energy gluons. Therefore they are of a great importance for our understanding of the quark-gluon interaction. In this work we focused on exotic mesons because they give insight into the dynamics of normal mesons.
The current candidates for the lightest exotic meson are $`\pi _1(1400)`$ and $`\pi _1(1600)`$. Most of the experimental resonances have large widths compared to normal mesons that decay strongly. Moreover, the observed decay channels should not be dominant according to the theory. Therefore, the reported signals do not completely agree with theoretical expectations and may originate from different phenomena such as rescattering. In order to resolve this problem it will be necessary to observe and measure exotic states with other quantum numbers. Searches for these states, as well as further exploration of the $`\pi _1`$ decays, are planned at Jefferson Lab (GlueX).
The dynamics of exotic mesons at the scale of $`12`$ GeV should be described by a relativistic theory. In this work we studied the size of relativistic effects in the decays of the $`\pi _1`$, and discussed a new picture of meson decays in the Coulomb gauge. Our considerations led to two important conclusions. First, numeric results showed significant relativistic corrections arising from the spin-orbit correlations introduced by Wigner rotation. The widths calculated using fully relativistic formulae are in general larger than the corresponding values calculated with no Wigner rotation (by a factor of order 10%), but smaller than completely nonrelativistic values. Some decays that are suppressed in the nonrelativistic limit (for example $`\pi _1\pi \rho `$ if the orbital wave functions of $`\pi `$ and $`\rho `$ are the same) acquire nonzero amplitudes in the relativistic case.
Second, the $`\pi _1`$ prefers to decay into two mesons, one of which has no orbital angular momentum and the other has $`L=1`$ (the $`S+P`$ selection rule). Thus this selection rule, found in other models seems to be quite general. Some decays ($`\pi \eta `$, $`\rho \omega `$, $`K\overline{K}^{}`$) are suppressed by symmetries in the orbital wave functions, and the assumption that the parameters $`\mu `$ for mesons with the same radial quantum numbers should be almost equal. We also noticed that the rates for the higher partial waves in decays of normal mesons where two waves are possible are larger in the $`{}_{}{}^{3}P_{0}^{}`$ then the $`{}_{}{}^{3}S_{1}^{}`$ model.
However, there is one problem which needs to be explored further. There are several components of the $`q\overline{q}g`$ normal and exotic meson wave functions, and in order to give accurate numerical predictions for the widths, one needs to know the relative amounts of these components. Since we were interested in relativistic effects and not in the absolute width values, we calculated the amplitudes for each component separately.
We also studied the interaction between $`\pi b_1`$ and $`\pi \rho `$ final states. For the physical range of the parameters used in our analysis we observed rather small absolute corrections arising from FSI. The decay width for the process $`\pi _1\pi b_1`$ decreased only by a factor of order 1 MeV, whereas for the $`\pi _1\pi \rho `$ increased by a similar amount. Therefore, the latter process seems to be suppressed anyway.
The experiments have reported exotic resonances that are light and rather broad. This work showed that relativistic calculations give widths smaller than nonrelativistic ones, thus the discrepancy between theory and experiment becomes larger. For the dominant $`\pi b_1`$ mode of the $`\pi _1(1600)`$ decay the maximum width is $`200`$ MeV as opposed to the experimental $`300`$ MeV. Furthermore, the width in the $`\pi \rho `$ mode is on the order of a few MeV, and the modes $`\pi \eta `$ and $`\pi \eta ^{}`$ are negligible.
The experimental data for the $`\pi _1`$ exotic meson are not well understood yet. Further experiments are needed to clarify the nature of the reported signals and explain the dynamics of exotic mesons. This work showed that, in order to compare these data with the theory, one should use models which are relativistic.
CURRICULUM VITAE
| |
| --- |
| |
| | Personal |
| |
| | Name : | Nikodem Janusz Poplawski |
| | Date of Birth: | March 1, 1975 |
| | Citizenship: | Polish |
| | Education |
| --- | --- |
| |
| | Sep 29, 2004 | PhD in Physics |
| | | Dissertation – *A relativistic description of hadronic decays of the* |
| | | *exotic meson $`\pi _1`$* |
| | | Supervisor – Prof. Adam Szczepaniak |
| Jan 31, 2004 | MS in Physics |
| --- | --- |
| 1999 – 2004 | Graduate studies – Department of Physics, Indiana University, |
| --- | --- |
| | Bloomington, IN |
| Jul 14, 1999 | MS in Astronomy |
| --- | --- |
| | Thesis – *A Michelson-Morley interferometer in the spacetime* |
| | *of a plane gravitational wave* |
| | Supervisor – Prof. Stanisław Baża$`\stackrel{´}{\mathrm{n}}`$ski |
| 1994 – 1999 | Undergraduate studies – Center for Interfaculty Individual |
| --- | --- |
| | Studies in Mathematical and Natural Sciences (MISMaP), |
| | University of Warsaw, Poland |
| 1993 – 1994 | Undergraduate studies – Department of Physics, Jagiellonian |
| --- | --- |
| | University, Cracow, Poland |
| | Work |
| --- | --- |
| |
| | 2004 – | Postdoctoral Fellow – Biocomplexity Institute, Indiana University |
| | 2001 – 2004 | Research Assistant – Nuclear Theory Center, Indiana University |
| | 1999 – 2001 | Associate Instructor – Department of Physics, Indiana University | |
warning/0507/math-ph0507029.html | ar5iv | text | # Bosons and fermions in external fields
## 1 Introduction
In this article we discuss quantum theories which describe systems of non-distinguishable particles interacting with external fields. Such models are of interest also in the non-relativistic case (in quantum statistical mechanics, nuclear physics, etc.), but the relativistic case has additional, interesting complications: in the latter case they are genuine quantum field theories, i.e. quantum theories with an infinite number of degrees of freedom, with non-trivial features like divergences and anomalies. Since interparticle interactions are ignored, such models can be regarded as a first approximation to more complicated theories, and they can be studied by mathematically precise methods.
Models of relativistic particles in external electromagnetic fields have received considerable attention in the physics literature, and interesting phenomena like the Klein paradox or particle-antiparticle pair creation in over-critical fields have been studied; see for an extensive review. We will not discuss these physics questions but only describe some proto-type examples and a general Hamiltonian framework which has been used in mathematically precise work on such models. The general framework for this latter work is the mathematical theory of Hilbert space operators (see e.g. ), but in our discussion we try to avoid presupposing knowledge of that theory. As shortly mentioned in the end, this work has had close relations to various topics of recent interest in mathematical physics, including anomalies, infinite dimensional geometry and group theory, conformal field theory, and noncommutative geometry.
We restrict our discussion to spin-$`0`$ bosons and spin-$`\frac{1}{2}`$ fermions, and we will not discuss models of particles in external gravitational fields but only refer the interested reader to . We also only mention in passing that external field problems have been also studied using functional integral approaches, and mathematically precise work on this can be found in the extensive literature on determinants of differential operators.
## 2 Examples
Consider the Schrödinger equation describing a non-relativistic particle of mass $`m`$ and charge $`e`$ moving in three dimensional space and interacting with an external vector- and scalar potential $`𝐀`$ and $`\varphi `$,
$$\mathrm{i}_t\psi =H\psi ,H=\frac{1}{2m}(\mathrm{i}+e𝐀)^2e\varphi $$
(1)
(we set $`\mathrm{}=c=1`$, $`_t=/t`$, and $`\psi `$, $`\varphi `$ and $`𝐀`$ can depend on the space and time variables $`𝐱^3`$ and $`t`$). This is a standard quantum mechanical model, with $`\psi `$ the one-particle wave function allowing for the usual probabilistic interpretation. One interesting generalization to the relativistic regime is the Klein-Gordon equation
$$\left[\left(\mathrm{i}_t+e\varphi \right)^2(\mathrm{i}+e𝐀)^2m^2\right]\psi =0$$
(2)
with a $``$-valued function $`\psi `$. There is another important relativistic generalization, the Dirac equation
$$\left[\left(\mathrm{i}_t+e\varphi \right)(\mathrm{i}+e𝐀)𝜶+m\beta \right]\psi =0$$
(3)
with $`𝜶=(\alpha _1,\alpha _2,\alpha _3)`$ and $`\beta `$ hermitian $`4\times 4`$ matrices satisfying the relations
$$\alpha _i\alpha _j+\alpha _j\alpha _i=\delta _{ij},\alpha _i\beta =\beta \alpha _i,\beta ^2=1$$
(4)
and a $`^4`$-valued function $`\psi `$ (we write $`1`$ also for the identity). These two relativistic equations differ by the transformation properties of $`\psi `$ under Lorentz transformations: in (2) it transforms like a scalar and thus describes spin-$`0`$ particles, and it transforms like a spinor describing spin-$`\frac{1}{2}`$ particles in (3). While these equations are natural relativistic generalizations of the Schrödinger equation, they no longer allow to consistently interpret $`\psi `$ as one-particle wave functions. The physical reason is that, in a relativistic theory, high energy processes can create particle-antiparticle pairs, and this makes the restriction to a fixed particle number inconsistent. This problem can be remedied by constructing a many-body model allowing for an arbitrary number of particles and anti-particles. The requirement that this many-body model should have a groundstate is an important ingredient in this construction.
It is obviously of interest to formulate and study many-body models of non-distinguishable already in the non-relativistic case. An important empirical fact is that such particles come in two kinds, bosons and fermions, distinguished by their exchange statistics (we ignore the interesting possibility of exotic statistics). For example, the fermion many-particle version of (1) for suitable $`\varphi `$ and $`𝐀`$ is a useful model for electrons in a metal. An elegant method to go from the one- to the many-particle description is the formalism of second quantization: one promotes $`\psi `$ to a quantum field operator with certain (anti-) commutator relations, and this is a convenient way to construct the appropriate many-particle Hilbert space, Hamiltonian, etc. In the non-relativistic case, this formalism can be regarded as an elegant reformulation of a pedestrian construction of a many-body quantum mechanical model, which is useful since it provides convenient computational tools. However, this formalism naturally generalizes to the relativistic case where the one-particle model no longer has an acceptable physical interpretation, and one finds that one can nevertheless can give a consistent physical interpretation to (2) and (3) provided that $`\psi `$ are interpreted as quantum field operators describing bosons and fermions, respectively. This particular exchange statistics of the relativistic particles is a special case of the spin-statistics theorem: integer spin particles are bosons and half-integer spin particles are fermions. While many structural features of this formalism are present already in the simpler non-relativistic models, the relativistic models add some non-trivial features typical for quantum field theories.
In the following we discuss a precise mathematical formulation of the quantum field theory models described above. We emphasis the functorial nature of this construction which makes manifest that it also applies to other situations, e.g., where the bosons and fermions are also coupled to a gravitational background, are considered in other spacetime dimensions than $`3+1`$, etc.
## 3 Second quantization: non-relativistic case
Consider a quantum system of non-distinguishable particles where the quantum mechanical description of one such particle is known. In general, this one-particle description is given by a Hilbert space $`h`$ and one-particle observables and transformations which are self-adjoint and unitary operators on $`h`$, respectively. The most important observable is the Hamiltonian $`H`$. We will describe a general construction of the corresponding many-body system.
Example. As a motivating example we take the Hilbert space $`h=L^2(^3)`$ of square-integrable functions $`f(𝐱)`$, $`𝐱^3`$, and the Hamiltonian $`H`$ in (1). A specific example for a unitary operator on $`h`$ is the gauge transformation $`(Uf)(𝐱)=\mathrm{exp}(\mathrm{i}\chi (𝐱))f(𝐱)`$ with $`\chi `$ a smooth, real-valued functions on $`^3`$.
In this example, the corresponding wave functions for $`N`$ identical such particles are the $`L^2`$-functions $`f_N(𝐱_1,\mathrm{},𝐱_N)`$, $`𝐱_j^3`$. It is obvious how to extend one-particle observables and transformations to such $`N`$-particle states: for example, the $`N`$-particle Hamiltonian corresponding to $`H`$ in (1) is
$$H_N=\underset{j=1}{\overset{N}{}}\frac{1}{2m}(\mathrm{i}_{𝐱_j}+e𝐀(t,𝐱_j))^2e\varphi (t,𝐱_j),$$
(5)
and the $`N`$-particle gauge transformation $`U_N`$ is defined through multiplication with $`_{j=1}^N\mathrm{exp}(\mathrm{i}\chi (𝐱_j))`$.
For systems of indistinguishable particles it is enough to restrict to wave functions which are even or odd under particle exchanges,
$$f_N(𝐱_1,\mathrm{},𝐱_j,\mathrm{},𝐱_k,\mathrm{},𝐱_N)=\pm f_N(𝐱_1,\mathrm{},𝐱_k,\mathrm{},𝐱_j,\mathrm{},𝐱_N)$$
(6)
for all $`1j<kN`$, with the upper and lower sign corresponding to bosons and fermions, respectively (this empirical fact is usually taken as postulate in non-relativistic many-body quantum physics). It is convenient to define the zero-particle Hilbert space as $``$ (complex numbers) and to introduce a Hilbert space containing states with all possible particle numbers: This so-called Fock space contains all states
$$\left(\begin{array}{c}f_0\\ f_1(𝐱_1)\\ f_2(𝐱_1,𝐱_2)\\ f_3(𝐱_1,𝐱_2,𝐱_3)\\ \mathrm{}\end{array}\right)$$
(7)
with $`f_0`$. The definition of $`H_N`$ and $`U_N`$ then naturally extends to this Fock space; see below.
General construction. The construction of Fock spaces and many-particle observables and transformations just outlined in a specific example is conceptually simple. An alternative, more efficient construction method is to use quantum fields which we denote as $`\psi (𝐱)`$ and $`\psi ^{}(𝐱)`$, $`𝐱^3`$. They can be fully characterized by the following (anti-) commutator relations,
$$[\psi (𝐱),\psi ^{}(𝐲)]_{}=\delta ^3(𝐱𝐲),[\psi (𝐱),\psi (𝐲)]_{}=0,$$
(8)
where $`[a,b]_{}abba`$, with the commutator and anti-commutators (upper and lower signs) corresponding to the boson and fermion case, respectively. It is convenient to ‘smear’ these fields with one-particle wave functions and define
$$\psi (f)=_^3\mathrm{d}^3x\overline{f(𝐱)}\psi (𝐱),\psi ^{}(f)=_^3\mathrm{d}^3x\psi ^{}(𝐱)f(𝐱)$$
(9)
for all $`fh`$. Then the relations characterizing the field operators can be written as
$$[\psi (f),\psi ^{}(g)]_{}=(f,g),[\psi (f),\psi (g)]_{}=0f,gh$$
(10)
where $`(f,g)=_^3\mathrm{d}^3x\overline{f(𝐱)}g(𝐱)`$ is the inner product in $`h`$. The Fock space $`_{}(h)`$ can then be defined by postulating that it contains a normalized vector $`\mathrm{\Omega }`$ called vacuum such that
$$\psi (f)\mathrm{\Omega }=0fh$$
(11)
and that all $`\psi ^{()}(f)`$ are operators on $`_{}(h)`$ such that $`\psi ^{}(f)=\psi (f)^{}`$ where $``$ is the Hilbert space adjoint. Indeed, from this we conclude that $`_{}(h)`$, as vector space, is generated by
$$f_1f_2\mathrm{}f_N\psi ^{}(f_1)\psi ^{}(f_2)\mathrm{}\psi ^{}(f_N)\mathrm{\Omega }$$
(12)
with $`f_jh`$ and $`N=0,1,2,\mathrm{}`$, and that the Hilbert space inner product of such vectors is
$$f_1f_2\mathrm{}f_N,g_1g_2\mathrm{}g_M=\delta _{N,M}\underset{PS_N}{}(\pm 1)^{|P|}\underset{j=1}{\overset{N}{}}(f_j,g_{Pj})$$
(13)
with $`S_N`$ the permutation group, with $`(+1)^{|P|}=1`$ always and $`(1)^{|P|}=+1`$ and $`1`$ for even and odd permutations, respectively. The many-body Hamiltonian $`q(H)`$ corresponding to the one-particle Hamiltonian $`H`$ now can be defined by the following relations,
$$q(H)\mathrm{\Omega }=0,[q(H),\psi ^{}(f)]=\psi ^{}(Hf)$$
(14)
for all $`fh`$ such that $`Hf`$ is defined. Indeed, this implies
$$q(H)f_1f_2\mathrm{}f_N=\underset{j=1}{\overset{N}{}}f_1f_2\mathrm{}(Hf_j)\mathrm{}f_N$$
(15)
which defines a self-adjoint operator on $`_{}(h)`$, and it is easy to check that this coincides with our down-to-earth definition of $`H_N`$ above. Similarly the many-body transformation $`Q(U)`$ corresponding to a one-particle transformation $`U`$ can be defined as
$$Q(U)\mathrm{\Omega }=\mathrm{\Omega },Q(U)\psi ^{}(f)=\psi ^{}(Uf)Q(U)$$
(16)
for all $`fh`$, which implies
$$Q(U)f_1f_2\mathrm{}f_N=(Uf_1)(Uf_2)\mathrm{}(Uf_N)$$
(17)
and thus coincides with our previous definition of $`U_N`$.
While we presented the construction above for a particular example, it is important to note that it actually does not make reference to what the one-particle formalism actually is. For example, if we had a model of particles on a space $``$ given by some ‘nice’ manifold of any dimension and with $`M`$ internal degrees of freedom, we would take $`h=L^2()^M`$ and replace (9) by
$$\psi (f)=_{}d\mu (𝐱)\underset{j=1}{\overset{M}{}}\overline{f_j(𝐱)}\psi _j(𝐱)$$
(18)
and its hermitian conjugate, with the measure $`\mu `$ on $``$ defining the inner product in $`h`$, $`(f,g)=d\mu (𝐱)_j\overline{f_j(𝐱)}g_j(𝐱)`$. With that, all formulas after (9) hold true as they stand. Given any one-particle Hilbert space $`h`$ with inner product $`(,)`$, observable $`H`$, and transformation $`U`$, the formulas above define the corresponding Fock spaces $`_{}(h)`$ and many-body observable $`q(H)`$ and transformation $`Q(U)`$. It is also interesting to note that this construction has various beautiful general (functorial) properties: the set of one-particle observables has a natural Lie algebra structure with the Lie bracket given by the commutator (strictly speaking: $`\mathrm{i}`$ times the commutator, but we drop the common factor $`\mathrm{i}`$ for simplicity). The definitions above imply
$$[q(A),q(B)]=q([A,B])$$
(19)
for one-particle observables $`A,B`$, i.e., the above-mentioned Lie algebra structure is preserved under this map $`q`$. In a similar manner, the set of one-particle transformations has a natural group structure preserved by the map $`Q`$,
$$Q(U)Q(V)=Q(UV),Q(U)^1=Q(U^1).$$
(20)
Moreover, if $`A`$ is self-adjoint, then $`\mathrm{exp}(\mathrm{i}A)`$ is unitary, and one can show that
$$Q(\mathrm{exp}(\mathrm{i}A))=\mathrm{exp}(\mathrm{i}q(A)).$$
(21)
For later use we note that, if $`\{f_n\}_n`$ is some complete, orthonormal basis in $`h`$, then operators $`A`$ on $`h`$ can be represented by infinite matrices $`(A_{mn})_{m,n}`$ with $`A_{mn}=(f_m,Af_n)`$, and
$$q(A)=\underset{m,n}{}A_{mn}^{}\psi _m^{}\psi _n^{}$$
(22)
where $`\psi _n^{()}=\psi ^{()}(f_n)`$ obey
$$[\psi _m^{},\psi _n^{}]_{}=\delta _{m,n},[\psi _m^{},\psi _n^{}]_{}=0$$
(23)
for all $`m,n`$. We also note that, in our definition of $`q(A)`$, we made a convenient choice of normalization, but there is no physical reason to not choose a different normalization and define
$$q^{}(A)=q(A)b(A)$$
(24)
where $`b`$ is some linear function mapping self-adjoint operators $`A`$ to real numbers. For example, one may wish to use another reference vector $`\stackrel{~}{\mathrm{\Omega }}`$ instead of $`\mathrm{\Omega }`$ in the Fock space, and then would choose $`b(A)=\stackrel{~}{\mathrm{\Omega }},q(A)\stackrel{~}{\mathrm{\Omega }}`$. Then the relation in (19) are changed to
$$[q^{}(A),q^{}(B)]=q^{}([A,B])+S_0(A,B)$$
(25)
where $`S_0(A,B)=b([A,B])`$. However, the $``$-number term $`S_0(A,B)`$ in the relations (25) is trivial since it can be removed by going back to $`q(A)`$.
Physical interpretation. The Fock space $`_{}(h)`$ is the direct sum of subspaces of states with different particle numbers $`N`$,
$$_{}(h)=\underset{N=0}{\overset{\mathrm{}}{}}h_{}^{(N)}$$
(26)
where the zero-particle subspace $`h_{}^{(0)}=`$ is generated by the vacuum $`\mathrm{\Omega }`$, and $`h_{}^{(N)}`$ is the $`N`$-particle subspace generated by the states $`f_1f_2\mathrm{}f_N`$, $`f_jh`$. We note that
$$𝒩q(1)$$
(27)
is the particle number operator, $`𝒩F_N=NF_N`$ for all $`f_Nh_{}^{(N)}`$. The field operators obviously change the particle number: $`\psi ^{}(f)`$ increases the particle number by one (maps $`h_{}^{(N)}`$ to $`h_{}^{(N+1)}`$), and $`\psi (f)`$ decreases it by one. Since every $`fh`$ can be interpreted as one-particle state, it is natural to interpret $`\psi ^{}(f)`$ and $`\psi (f)`$ as creation and annihilation operators, respectively: they create and annihilate one particle in the state $`fh`$. It is important to note that, in the fermion case, (10) implies $`\psi ^{}(f)^2=0`$, which is a mathematical formulation of the Pauli exclusion principle: it is not possible to have two fermions in the same one-particle state. In the boson case there is no such restriction. Thus, even though the formalisms used to describe boson- and fermion systems look very similar, they describe dramatically different physics.
Applications. In our example, the many-body Hamiltonian $`_0q(H)`$ can also be written in the following suggestive form,
$$_0=\mathrm{d}^3x\psi ^{}(𝐱)(H\psi )(𝐱),$$
(28)
and similar formulas hold true for other observables and other Hilbert spaces $`h=L^2()^n`$. It is rather easy to solve the model defined by such Hamiltonian: all necessary computations can be reduced to one-particle computations. For example, in the static case where $`𝐀`$ and $`\varphi `$ are time independent, a main quantity of interest in statistical physics is the free energy
$$\beta ^1\mathrm{log}\left(\mathrm{Tr}\left(\mathrm{exp}(\beta [_0\mu 𝒩])\right)\right)$$
(29)
where $`\beta >0`$ here is the inverse temperature, $`\mu `$ the chemical potential, and the trace over the Fock space $`_{}(h)`$. One can show that
$$=\pm \mathrm{tr}\left(\beta ^1\mathrm{log}(1\mathrm{exp}(\beta [H\mu ]))\right)$$
(30)
where the trace here is over the one-particle Hilbert space $`h`$. Thus, to compute $``$, one only needs to find the eigenvalues of $`H`$.
It is important to mention that the framework discussed here is not only for external field problems but can be equally well used to formulate and study more complicated models with interparticle interactions. For example, while the model with the Hamiltonian $`_0`$ above is often too simple to describe systems in nature, it is easy to write down more realistic models, e.g., the Hamiltonian
$$=_0+(e^2/2)\mathrm{d}^3x\mathrm{d}^3y\psi ^{}(𝐱)\psi ^{}(𝐲)|𝐱𝐲|^1\psi (𝐲)\psi (𝐱)$$
(31)
describes electrons in an external electromagnetic field interacting through Coulomb interactions. This illustrates an important point which we would like to stress: the task in quantum theory is two-fold, namely to formulate and to solve (exact of otherwise) models. Obviously, in the non-relativistic case, it is equally simple to formulate many-body models with and without inter-particle interactions, and the latter only are simpler because they are easier to solve: the two tasks of formulating and solving models can be clearly separated. As we will see, in the relativistic case, even the formulation of an external field problem is non-trivial, and one finds that one cannot formulate the model without at least partially solving it. This is a common feature of quantum field theories making them challenging and interesting.
## 4 Relativistic fermion and boson systems
We now generalize the formalism developed in the previous section to the relativistic case.
Field algebras and quasi-free representations. In the previous section we identified the field operators $`\psi ^{()}(f)`$ with particular Fock space operators. This is analog to identifying the operators $`p_j=\mathrm{i}_{x_j}`$ and $`q_j=x_j`$ on $`L^2(^M)`$ with the generators of the Heisenberg algebra, as usually done. (We recall: the Heisenberg algebra is the star algebra generated by $`P_j`$ and $`Q_j`$, $`j=1,2,\mathrm{},M<\mathrm{}`$, with the well-known relations,
$$[P_j,P_k]=\mathrm{i}\delta _{jk},[P_j,P_k]=[P_j,Q_k]=0,P_j^{}=P_j^{},Q_j^{}=Q_j^{}$$
(32)
for all $`j,k`$.) Identifying the Heisenberg algebra with a particular representation is legitimate since, as is well-known, all its irreducible representations are (essentially) the same (this statement is made precise by a celebrated theorem due to von Neumann).
However, in case of the algebra generated by the field operators $`\psi ^{()}(f)`$, there exist representations which are truly different from the ones discussed in the last section, and to construct relativistic external field problems such representations are needed. It is therefore important to distinguish the fields as generators of an algebra from the operators representing them. We thus define the (boson or fermion) field algebra $`𝒜_{}(h)`$ over a Hilbert space $`h`$ as the star algebra generated by $`\mathrm{\Psi }^{}(f)`$, $`fh`$, such that the map $`f\mathrm{\Psi }(f)`$ is linear and the relations
$$[\mathrm{\Psi }(f),\mathrm{\Psi }^{}(g)]_{}=(f,g),[\mathrm{\Psi }(f),\mathrm{\Psi }(g)]_{}=0,\mathrm{\Psi }^{}(f)^{}=\mathrm{\Psi }(f)$$
(33)
are fulfilled for all $`f,gg`$, with $``$ the star operation in $`𝒜_{}(h)`$.
The particular representation of this algebra discussed in the last section will be denoted by $`\pi _0`$, $`\pi _0(\mathrm{\Psi }^{()}(f)=\psi ^{()}(f)`$. Other representations $`\pi _P_{}`$ can be constructed from any projection operators $`P_{}`$ on $`h`$, i.e., any operator $`P_{}`$ on $`h`$ satisfying $`P_{}^{}=P_{}^2=P_{}^{}`$. Writing $`\widehat{\psi }^{()}(f)`$ short for $`\pi _P_{}(\mathrm{\Psi }^{()}(f))`$, this so-called quasi-free representation is defined by
$`\widehat{\psi }^{}(f)=\psi ^{}(P_+f)+\psi (\overline{P_{}f}),\widehat{\psi }(f)=\psi (P_+f)\psi ^{}(\overline{P_{}f})`$ (34)
where the bar means complex conjugation. It is important to note that, while the star operation is identical with the Hilbert space adjoint $``$ in the fermion case, we have
$$\widehat{\psi }(f)^{}=\psi (Ff)^{}\text{ with }F=P_+P_{}\text{ for bosons}$$
(35)
where $`F`$ is a grading operator, i.e., $`F^{}=F`$ and $`F^2=1`$. We stress that the ‘physical’ star operation always is $``$, i.e., physical observables $`A`$ obey $`A=A^{}`$.
The present framework suggests to regard quantization as the procedure which amounts to going from a one-particle Hilbert space $`h`$ to the corresponding field algebra $`𝒜_{}(h)`$. Indeed, the Heisenberg algebra is identical with the boson field algebra $`𝒜_{}(^M)`$ (since the latter is obviously identical with the algebra of $`M`$ harmonic oscillators), and thus conventional quantum mechanics can be regarded as boson quantization in the special case where the one-particle Hilbert space is finite dimensional. It is interesting to note that ‘fermion quantum mechanics’ $`𝒜_{}(^M)`$ is the natural framework for formulating and studying lattice fermion and spin systems which play an important role in condensed matter physics.
In the following we elaborate the naive interpretations of the relativistic equations in (2) and (3) as a quantum theory of one particle, and we discuss why they are unphysical. For simplicity we assume that the electromagnetic fields $`\varphi ,𝐀`$ are time independent. We then show that quasi-free representations as discussed above can provide physically acceptable many-particle theories. We first consider the Dirac case which is somewhat simpler.
### 4.1 Fermions
One-particle formalism: Recalling that $`\mathrm{i}_t`$ is the energy operator, we define the Dirac Hamiltonian $`D`$ by rewriting (3) in the following form,
$$\mathrm{i}_t\psi =D\psi ,D=(\mathrm{i}+e𝐀)𝜶+m\beta e\varphi .$$
(36)
This Dirac Hamiltonian is obviously is a self-adjoint operator on the one-particle Hilbert space $`h=L^2(^4)^4`$, but, different from the Schrödinger Hamiltonian in (1), it is not bounded from below: for any $`E_0>\mathrm{}`$ one can find a state $`f`$ such that the energy expectation value $`(f,Df)`$ is less than $`E_0`$. This can be easily seen for the simplest case where the external potential vanishes, $`𝐀=\varphi =0`$. Then the eigenvalues of $`D`$ can be computed by Fourier transformation, and one finds
$$E=\pm \sqrt{𝐩^2+m^2},𝐩^3.$$
(37)
Due to the negative energy eigenvalues we conclude that there is no ground state, and the Dirac Hamiltonian thus describes an unstable system which is physically meaningless.
To summarize: a (unphysical) one-particle description of relativistic fermions is given by a Hilbert space $`h`$ together with a self-adjoint Hamiltonian $`D`$ unbounded from below. Other observables and transformations are given by self-adjoint and unitary operators on $`h`$, respectively.
Many-body formalism: We now explain how to construct a physical many-body description from these data. To simplify notation we first assume that $`D`$ has a purely discrete spectrum (which can be achieved by using a compact space). We then can label the eigenfunctions $`f_n`$ by integers $`n`$ such that the corresponding eigenvalues $`E_n0`$ for $`n0`$ and $`E_n<0`$ for $`n<0`$. Using the naive representation of the fermion field algebra discussed in the last section we get (we use the notation introduced in (22))
$$q(D)=\underset{n0}{}|E_n|\psi _n^{}\psi _n^{}\underset{n<0}{}|E_n|\psi _n^{}\psi _n^{},$$
(38)
which is obviously not bounded from below and thus not physically meaningful. However, $`\psi _n^{}\psi _n^{}=1\psi _n^{}\psi _n^{}`$, which suggests that we can remedy this problem by interchanging the creation- and annihilation operators for $`n<0`$. This is possible: it is easy to see that
$$\widehat{\psi }_n^{}\psi _n^{}n0\text{ and }\widehat{\psi }_n^{}\psi _n^{}n<0$$
(39)
provides a representation of the algebra in (23). We thus define
$$\widehat{q}(D)\underset{n}{}E_n^{}:\widehat{\psi }_n^{}\widehat{\psi }_n^{}:$$
(40)
with the so-called normal ordering prescription
$$:\psi _m^{}\psi _n^{}:\psi _m^{}\psi _n^{}\mathrm{\Omega },\psi _m^{}\psi _n^{}\mathrm{\Omega },$$
(41)
where we made use of the freedom of normalization explained after (23) to eliminate unwanted additive constants. We get $`q(D)=_n|E_n^{}|\psi _n^{}\psi _n^{}`$, which is manifestly a non-negative self-adjoint operator with $`\mathrm{\Omega }`$ as groundstate. We thus found a physical many-body description for our model. We now can define for other one-particle observables,
$$\widehat{q}(A)\underset{n}{}A_{mn}^{}:\widehat{\psi }_m^{}\widehat{\psi }_n^{}:,$$
(42)
and by straightforward computations we obtain
$$[\widehat{q}(A),\widehat{q}(B)]=\widehat{q}([A,B])+S(A,B)$$
(43)
where $`S(A,B)=_{m<0}_{n0}(A_{mn}B_{nm}B_{mn}A_{nm})`$, i.e.,
$$S(A,B)=\mathrm{tr}\left(P_{}AP_+BP_{}P_{}BP_+AP_{}\right)$$
(44)
with $`P_{}=_{n<0}f_n(f_n,)`$ the projection onto the subspace spanned by the negative energy eigenvectors of $`D`$ and $`P_+=1P_{}`$. One can show that $`\widehat{q}(A)`$ no longer is defined for all operators but only if
$$P_{}AP_+\text{ and }P_+AP_{}\text{ are Hilbert-Schmidt operators}$$
(45)
(we recall that $`a`$ is a Hilbert-Schmidt operator if $`\mathrm{tr}(a^{}a)<\mathrm{}`$). The $``$-number term $`S(A,B)`$ in (43) is often called Schwinger term, and different from the similar term in (25) it now is non-trivial, i.e., it no longer is possible to remove it be a redefinition $`\widehat{q}^{}(A)=\widehat{q}(A)b(A)`$. This Schwinger term is an example of an anomaly, and it has various interesting implications.
In a similar manner, one can construct the many-body transformations $`\widehat{Q}(U)`$ of unitary operators $`U`$ on $`h`$ satisfying the very Hilbert-Schmidt condition in (45), and one obtains
$$\widehat{Q}(U)\widehat{Q}(V)=\chi (U,V)\widehat{Q}(UV)$$
(46)
with an interesting phase valued functions $`\chi `$.
More generally, for any one-particle Hilbert space $`h`$ and Dirac Hamiltonian $`D`$, the physical representation is given by the quasi-free representation $`\pi _P_{}`$ in (34) with $`P_{}`$ the projection onto the negative energy subspace of $`D`$. The results about $`\widehat{q}`$ and $`\widehat{Q}`$ mentioned hold true in any such representation.
Thus the one-particle Hamiltonian $`D`$ determines which representation one has to use, and one therefore cannot construct the ‘physical’ representation without specific information about $`D`$. However, not all these representations are truly different: If there is a unitary operator $`𝒰`$ on the Fock space $`_+(h)`$ such that
$$𝒰^{}\pi _{P_{}^{(1)}}(\psi ^{()}(f))𝒰=\pi _{P_{}^{(2)}}(\psi ^{()}(f))$$
(47)
for all $`fh`$, then the quasi-free representations associated with the different projections $`P_{}^{(1)}`$ and $`P_{}^{(2)}`$ are physically equivalent: one could equally well formulate the second model using the representation of the first. Two such quasi-free representations are called unitarily equivalent, and a fundamental theorem due to Shale and Stinespring states that two quasi-free representations $`\pi _{P_{}^{(1,2)}}`$ are unitarily equivalent if and only if $`P_{}^{(1)}P_{}^{(2)}`$ is a Hilbert-Schmidt operator (a similar result holds true in the boson case).
### 4.2 Bosons
One-particle formalism: Similarly as for the Dirac case, also the solutions of the Klein-Gordon equation in (2) do not define a physically acceptable one-particle quantum theory with a ground state: the energy eigenvalues in (37) for $`𝐀=\varphi =0`$ are a consequence the relativistic invariance and thus equally true for the Klein-Gordon case. However, in this case there is a further problem. To find the one-particle Hamiltonian one can rewrite the second order equation in (2) as a system of first order equations,
$$\mathrm{i}_t\mathrm{\Phi }=K\mathrm{\Phi },\mathrm{\Phi }=\left(\begin{array}{c}\psi \\ \pi ^{}\end{array}\right),K=\left(\begin{array}{cc}\hfill C& \hfill \mathrm{i}\\ \hfill \mathrm{i}B^2& \hfill C\end{array}\right)$$
(48)
with
$$B^2(\mathrm{i}+e𝐀)^2+m^2,Ce\varphi .$$
(49)
Thus one sees that the natural one-particle Hilbert space for the Klein-Gordon equation is $`h=L^2(^3)^2`$; here and in the following we identify $`h`$ with $`h_0h_0`$, $`h_0=L^2(^3)`$, and use a convenient $`2\times 2`$ matrix notation naturally associated with that splitting. However, the one-particle Hamiltonian is not self-adjoint but rather obeys
$$K^{}=JKJ,J\left(\begin{array}{cc}\hfill 0& \hfill \mathrm{i}\\ \hfill \mathrm{i}& \hfill 0\end{array}\right)$$
(50)
with $``$ the Hilbert space adjoint. It is important to note that $`J`$ is a grading operator. Thus, we can define a sequilinear form
$$(f,g)_J(f,Jg)f,gh,$$
(51)
with $`(,)`$ the standard inner product, and (50) is equivalent to $`K`$ being self-adjoint with respect to this sesquilinear form; in this case we say that $`K`$ is $`J`$-self-adjoint. Thus, in the Klein-Gordon case, this sesquilinear form takes the role of the Hilbert space inner product and, in particular, not $`(\mathrm{\Phi },\mathrm{\Phi })`$ but $`(\mathrm{\Phi },\mathrm{\Phi })_J`$ is preserved under time evolution. However, different from $`\mathrm{\Phi }^{}\mathrm{\Phi }`$, $`\mathrm{\Phi }^{}J\mathrm{\Phi }`$ is not positive definite, and it is therefore not possible to interpret it as probability density as in conventional quantum mechanics. For consistency one has to require that one-particle transformations $`U`$ are unitary with respect to $`(\mathrm{\Phi },\mathrm{\Phi })_J`$, i.e., $`U^1=JUJ`$. We call such operators $`J`$-unitary.
To summarize: a (unphysical) one-particle description of relativistic bosons is given by a Hilbert space of the form $`h=h_0h_0`$, the grading operator $`J`$ in (50), and a $`J`$-self-adjoint Hamiltonian $`K`$ of the form as in Eq. (48) where $`B0`$ and $`C`$ are self-adjoint operators on $`h_0`$. Other observables and transformations are given by $`J`$-self-adjoint and $`J`$-unitary operators on $`h`$, respectively.
Many-body formalism: We first consider the quasi-free representation $`\pi _{P_{}^{(0)}}`$ of the boson field algebra $`𝒜_{}(h)`$ so that the grading operator in (35) is equal to $`J`$, i.e., $`P_{}^{(0)}=(1J)/2`$. Writing $`\pi _{P_{}^{(0)}}(\mathrm{\Psi }^{()}(f))=\psi ^{()}(f)`$ one finds that
$$q(A)^{}=q(JAJ),Q(U)^{}=Q(JU^{}J),$$
(52)
and thus $`J`$-selfadjoint operators and $`J`$-unitary operators are mapped to proper observables and transformations. In particular, $`q(K)`$ is a self-adjoint operator, which resolves one problem of the one-particle theory. However, $`q(K)`$ is not bounded from below, and thus $`\pi _{P_{}^{(0)}}`$ is not yet the physical representation.
The physical representation can be constructed using the operators
$$T=\frac{1}{\sqrt{2}}\left(\begin{array}{cc}\hfill B^{1/2}& \hfill \mathrm{i}B^{1/2}\\ \hfill B^{1/2}& \hfill \mathrm{i}B^{1/2}\end{array}\right),F=\left(\begin{array}{cc}\hfill 1& \hfill 0\\ \hfill 0& \hfill 1\end{array}\right)$$
(53)
(for simplicity we restrict ourselves to the case $`C=0`$ and $`B>0`$; we use of the calculus of self-adjoint operators here) with the following remarkable properties,
$$T^1=JT^{}F,TKT^1=\left(\begin{array}{cc}\hfill B& \hfill 0\\ \hfill 0& \hfill B\end{array}\right)\widehat{K}.$$
(54)
One can check that
$$\widehat{\psi }^{}(f)\psi ^{}(Tf),\widehat{\psi }(f)\psi (T^1f),$$
(55)
is a quasi-free representation $`\pi _P_{}`$ of $`𝒜_{}(h)`$ with $`P_{}=(1F)/2`$. With that the construction of $`\widehat{q}`$ and $`\widehat{Q}`$ is very similar to the fermion case described above (the crucial simplification is that $`\widehat{K}`$ and $`F`$ now are diagonal). In particular, $`\widehat{q}(K)`$ is a non-negative operator with the ground state $`\mathrm{\Omega }`$, and $`\widehat{q}(A)`$ and $`\widehat{Q}(U)`$ is self-adjoint and unitary for every one-particle observable $`A`$ and transformation $`U`$, respectively. One also gets relations as in (43) and (46).
## 5 Further reading
The impossibility to construct relativistic quantum mechanical models played an important role in the early history of quantum field theory, as beautifully discussed in , Chapter 1.
The abstract formalism of quasi-free representations of fermion and boson field algebras was developed in many papers; see e.g. for explicit results on $`\widehat{Q}`$ and $`\chi `$. A nice textbook presentation with many references can be found in , Chapter 13 (this chapter is rather self-contained but mainly restricted to the fermion case).
Based on the Shale-Stinespring theorem there has been considerable amount of work to investigate whether the quasi-free representations associated with different external electromagnetic fields $`\psi _1,𝐀_1`$ and $`\psi _2,𝐀_2`$ are unitarily equivalent, if and which time dependent many-body Hamiltonians exist etc.; see , Chapter 13 and references therein.
The infinite dimensional Lie $`g_2`$ of Hilbert space operators satisfying the condition in (45) is an interesting infinite dimensional Lie algebra with a beautiful representation theory. This subject is closely related to conformal field theory; see e.g. for a textbook presentation and for a detailed mathematical account within the framework described by us.
It turns out that the mathematical framework discussed in Section 4 is sufficient for constructing fully interacting quantum field theories, in particular Yang-Mills gauge theories, in 1+1 but not in higher dimensions. The reason is that, in 3+1 dimensions, the one-particle observables $`A`$ of interest do not obey the Hilbert-Schmidt condition in (45) but only the weaker condition,
$$\mathrm{tr}(a^{}a)^n<\mathrm{},a=P_{}AP_\pm .$$
(56)
with $`n=2`$, and the natural analog of $`g_2`$ in 3+1 dimensions thus seems to be the Lie algebra $`g_{2n}`$ of operators satisfying this condition with $`n=2`$. Various results on the representation theory of such Lie algebras $`g_{2n>2}`$ have been developed; see where also various interesting relations to infinite dimensional geometry are discussed.
As mentioned, the Schwinger term $`S(A,B)`$ in (44) is an example of an anomaly. Mathematically it is a non-trivial 2-cocycles of the Lie algebra $`g_2`$, and analogs for the groups $`g_{2n>2}`$ have been found. These cocycles provide a natural generalization of anomalies (in the meaning of particle physics) to operator algebras. They not only shed some interesting light on the latter, but also provide a link to notions and results from non-commutative geometry; see e.g. . We believe that this link can provide a fruitful driving force and inspiration to find ways to deepen our understanding of quantum Yang-Mills theories in 3+1 dimensions .
## Keywords
Conformal field theory
anomalies
Noncommutative geometry
Dirac operators
Determinants of differential operators |
warning/0507/math0507524.html | ar5iv | text | # Weak Convergence of the Scaled Median of Independent Brownian Motions This work was supported in part by the VIGRE grants of both University of Washington and University of Wisconsin-Madison.
## 1 Introduction
Consider a dye diffusing in a homogeneous medium. When we view this phenomenon from a macroscopic perspective, what we see is a deterministic evolution of the density of the dye, governed by a partial differential equation. It is well understood that the solution of this equation can be represented probabilistically in terms of Brownian motion. The reason, of course, that Brownian motion enters into this situation is that, heuristically, we can imagine that each dye particle is performing such a random motion. In reality, however, a more accurate description of the particles is that they are following piece-wise linear trajectories and interacting through collisions.
In 1968, F. Spitzer provided a rigorous connection between a certain colliding particle model and the Brownian motion heuristics. In Spitzer’s model, we begin with countably many particles distributed along the real line according to a Poisson distribution. At time $`t=0`$, the particles begin moving with random velocities. These velocities are i.i.d., integrable, mean zero random variables. During their motion, the particles interact through elastic collisions. That is, whenever two particles meet, they exchange velocities (or, equivalently, they exchange trajectories). The particle which is closest to the origin at time $`t=0`$ is called the “tagged” particle and we denote its position at time $`t`$ by $`X(t)`$. Spitzer showed that the law on $`C[0,\mathrm{})`$ induced by the process $`tc^{1/2}X(ct)`$ converges weakly as $`c\mathrm{}`$ to the law of Brownian motion.
Spitzer’s work was preceded by that of T. E. Harris who showed that if the underlying motion of the particles is Brownian, instead of linear, then $`c^{1/4}X(ct)`$ converges to fractional Brownian motion with Hurst parameter $`H=1/4`$. These results were further generalized by Dürr, Goldstein, and Lebowitz in 1985. They showed, among other things, that if the individual particles perform fractional Brownian motion with Hurst parameter $`H`$, then $`c^{H/2}X(ct)`$ converges to fractional Brownian motion with Hurst parameter $`H/2`$.
One thing to note in these more general models is the definition of an “elastic collision.” When the particles perform Brownian motion, for example, the collisions are not isolated and it is not entirely clear how to exchange their trajectories at each point of intersection. In these situations, we generate the collision process by simply relabelling the particles at each time $`t`$ in a way that preserves their initial ordering. For instance, if there are only finitely particles, as there will be in our model, the location of the tagged particle is simply an order statistic of the locations of all of the particles. (In our model, it will be the median.)
In the work of Spitzer, Harris, and Dürr et al, the chief difficulty in proving convergence is establishing tightness. And in each of these models, the Poisson distribution of the initial particle configuration provides for tractable computations and is a central feature of the proofs. In this article, we will consider a model similar to Harris’s, but without the initial Poisson distribution. Namely, we consider a sequence $`\{B_j\}`$ of independent Brownian motions starting at the origin. We let $`M_n`$ denote the median of the first $`n`$ of these, and study the scaled process $`X_n=\sqrt{n}M_n`$. As with the other models, our chief difficulty will be to prove tightness. We will prove this, however, by making direct estimates on the path of the “tagged” particle, without relying on any special features of the initial particle distribution. In the end, we will discover a limit process which behaves locally like fractional Brownian motion with Hurst parameter $`H=1/4`$. This fact, formally stated in Theorem 2.1, lends support to the evident notion that Harris’s initial Poisson distribution is, to a certain degree, just a technical convenience, and does not play a significant role in determining the local behavior of the limit.
Before proceeding with the formal analysis of our model, let us preview some of the techniques in the proof. The first key ingredient in the proof will be given by Theorem 5.1, which establishes a formula for the conditional law of the median in terms of probabilities associated with a certain random walk. The second ingredient will be Lemma 6.4, which gives estimates for this random walk in terms of its parameters. And the third ingredient will be Lemma 7.1 (and its modification, Lemma 8.1) which estimates those parameters in terms of the motion of the individual particles.
Since it would be natural to conjecture that the results of Spitzer and Dürr et al would also hold in more general models, it is important to try to understand how these techniques might apply in a broader context. For example, we could try to generalize the results of Dürr et al by replacing the Brownian motions in our model with fractional Brownian motions. Or we could replace them with reflected processes if we wanted to consider particles in a “box,” reflecting off the walls of the box as well as each other. Such a model (in which the particles’ paths are piece-wise linear) was studied by P. F. Tupper in , although in that paper, a seemingly ad-hoc condition is imposed in order to prove tightness. (See the discussion after Theorem 2.3 in .) Other ways to generalize the model include giving our particles some nontrivial initial distribution, instead of starting them at the origin, or possibly considering a quantile (or even a family of quantiles) other than the median.
In any of these generalized models, the first and second ingredients outlined above would likely carry over with at most minor modifications. It is the third ingredient that would not transfer so easily. The estimates in Lemma 7.1 rely heavily on the fact that the individual particles are performing Brownian motion. Conceivably, analogous estimates could be worked out on a case-by-case basis for each different model under consideration. But the work of Harris, Spitzer, and Dürr et al suggests a deeper connection between the motion of the individual particles and the limit process. It is my belief that this connection would make itself known through these estimates.
But whether estimates can found in some general form or must be developed for each model individually, it is my hope that the techniques developed here can be used to extend the current family of results to a much broader range of colliding particle models.
## 2 The Model and Main Result
In our model, we will consider a sequence of independent, standard, one-dimensional Brownian motions, $`\{B_j(t)\}_{j=1}^{\mathrm{}}`$. Let $`M_n(t)`$ denote the median of the first $`n`$ Brownian motions. To be precise, define the median function $`_n:^n`$ as follows: if $`(x_1,\mathrm{},x_n)^n`$ and $`\tau `$ is a permutation of $`\{1,\mathrm{},n\}`$ such that $`x_{\tau (1)}x_{\tau (2)}\mathrm{}x_{\tau (n)}`$, then $`_n(x_1,\mathrm{},x_n)=x_{\tau (k)}`$, where $`k=(n+1)/2`$ and $`x`$ denotes the greatest integer less than or equal to $`x`$. We then define the (continuous) median process $`M_n(t)=_n(B_1(t),\mathrm{},B_n(t))`$.
In terms of colliding particles, what we have here is a sequence of particle systems. In the $`n`$-th system there are $`n`$ particles performing Brownian motion. If these particles interact through elastic collisions, then their trajectories are given by the order statistics of $`B_1(t),\mathrm{},B_n(t)`$. We will investigate the behavior of the center particle’s trajectory, $`M_n(t)`$.
In order to get a non-degenerate limit, we must consider the scaled median process $`X_n(t)=\sqrt{n}M_n(t)`$. The random variables $`X_n=\{X_n(t):0t<\mathrm{}\}`$ take values in the space $`C[0,\mathrm{})`$, which we endow with the topology of uniform convergence on compact sets. It will be shown that these processes converge weakly, by which we mean that they converge in law as $`C[0,\mathrm{})`$-valued random variables.
###### Theorem 2.1
There exists a continuous process $`X=\{X(t):0t<\mathrm{}\}`$ such that $`X_{2n+1}`$ converges weakly to $`X`$ as $`n\mathrm{}`$. Moreover, $`X`$ is a centered Gaussian process, which is locally Hölder continuous with exponent $`\gamma `$ for every $`\gamma (0,1/4)`$, and has covariance function
$$E[X(s)X(t)]=\sqrt{st}\mathrm{sin}^1\left(\frac{st}{\sqrt{st}}\right),$$
(2.1)
where $`\mathrm{sin}^1()`$ takes values in $`[\pi /2,\pi /2]`$.
It can be shown by (2.1) that, for $`ts`$ small, $`E|X(t)X(s)|^2\sqrt{ts}`$. In other words, the limit process has the same local fluctuations as fractional Brownian motion with Hurst parameter $`H=1/4`$.
The chief difficulty in proving Theorem 2.1 will be to establish the tightness of the processes $`X_{2n+1}`$. Before dealing with this issue, let us first establish the convergence of the finite-dimensional distributions and the existence of the limit process. To begin, we will need the following result, which is a special case of Theorems 7.1.1 and 7.1.2 in .
###### Theorem 2.2
Let $`\{\xi ^{(n)}\}_{n=1}^{\mathrm{}}`$ be an i.i.d. sequence of random vectors in $`^d`$ and define the component-wise median of $`\xi ^{(1)},\mathrm{},\xi ^{(n)}`$ to be the vector $`M^{(n)}`$ with components $`M_j^{(n)}=_n(\xi _j^{(1)},\xi _j^{(2)},\mathrm{},\xi _j^{(n)})`$. Let $`F_j(x)=P(\xi _j^{(1)}x)`$, $`G_{ij}(x,y)=P(\xi _i^{(1)}x,\xi _j^{(1)}y)`$, and $`\rho _{ij}=G_{ij}(0,0)1/4`$. If
(i) $`F_j(0)=1/2`$ and $`F_j^{}(0)>0`$ for all $`j`$, and
(ii) $`G_{ij}`$ is continuous at $`(0,0)`$ for all $`i`$ and $`j`$,
then $`\sqrt{n}M^{(n)}`$ converges in law to a jointly Gaussian random vector $`N`$ satisfying
$$EN_iN_j=\frac{\rho _{ij}}{F_i^{}(0)F_j^{}(0)}$$
and $`EN_i=0`$.
For our purposes, we will need the following.
###### Corollary 2.3
If $`\{\xi ^{(n)}\}_{n=1}^{\mathrm{}}`$ is an i.i.d. sequence of jointly Gaussian random vectors in $`^d`$ with mean zero and covariance matrix $`\sigma `$, then $`\sqrt{n}M^{(n)}`$ converges in law to a jointly Gaussian random vector $`Z`$ with mean zero and covariance matrix $`\tau `$, where
$$\tau _{ij}=EZ_iZ_j=\sqrt{\sigma _{ii}\sigma _{jj}}\mathrm{sin}^1\left(\frac{\sigma _{ij}}{\sqrt{\sigma _{ii}\sigma _{jj}}}\right)$$
and $`\mathrm{sin}^1()`$ takes values in $`[\pi /2,\pi /2]`$.
Proof: This follows easily from Theorem 2.2 and the well-known fact that if $`X`$ and $`Y`$ are jointly Gaussian with mean zero, then
$$P(X0,Y0)=\frac{1}{4}+\frac{1}{2\pi }\mathrm{sin}^1\left(\frac{EXY}{\sqrt{EX^2EY^2}}\right),$$
where $`\mathrm{sin}^1()`$ takes values in $`[\pi /2,\pi /2]`$. $`\mathrm{}`$
###### Theorem 2.4
There exists a centered Gaussian process $`X=\{X(t):0t<\mathrm{}\}`$ with covariance function (2.1) and which is locally Hölder continuous with exponent $`\gamma `$ for every $`\gamma (0,1/4)`$.
Proof: Let $`T`$ be the set of finite sequences $`𝐭=(t_1,\mathrm{},t_n)`$ of distinct, nonnegative numbers, where the length $`n`$ of these sequences ranges over the set of positive integers. For each $`𝐭`$ of length $`n`$, let $`𝐙_𝐭=(Z_1,\mathrm{},Z_n)`$ be a jointly Gaussian random vector with mean zero and covariance
$$EZ_iZ_j=\sqrt{t_it_j}\mathrm{sin}^1\left(\frac{t_it_j}{\sqrt{t_it_j}}\right).$$
(By Corollary 2.3, with $`\xi ^{(j)}=(B_j(t_1),\mathrm{},B_j(t_n))`$, such a $`𝐙_𝐭`$ exists.) Define the measure $`Q_𝐭`$ on $`^n`$ by $`Q_𝐭(A)=P(𝐙_𝐭A)`$. The family of finite-dimensional distributions, $`\{Q_𝐭\}_{𝐭T}`$, is clearly consistent, so there exists a real-valued process $`X=\{X(t):0t<\mathrm{}\}`$ that has the desired finite-dimensional distributions. It remains only to show that this process has a continuous modification, which is locally Hölder-continuous with exponent $`\gamma `$ for every $`\gamma (0,1/4)`$.
By the Kolmogorov-Čentsov Theorem (Theorem 2.2.8 in ), if, for each $`T>0`$,
$$E|X(t)X(s)|^\alpha C_T|ts|^{1+\beta }$$
for some positive constants $`\alpha `$, $`\beta `$, and $`C_T`$ (depending on $`T`$) and all $`0s<tT`$, then $`X`$ has a continuous modification which is locally Hölder-continuous with exponent $`\gamma `$ for every $`\gamma (0,\beta /\alpha )`$. Hence, it will suffice for us to show that for every $`\alpha >4`$ and every $`T>0`$,
$$E|X(t)X(s)|^\alpha C|ts|^{\alpha /4}$$
for some $`C>0`$ (depending only on $`T`$ and $`\alpha `$) and all $`0s<tT`$.
First, observe that $`X(t)X(s)`$ is normal with mean zero and variance
$`E|X(t)X(s)|^2`$ $`=EX(t)^2+EX(s)^22EX(t)X(s)`$
$`={\displaystyle \frac{\pi }{2}}t+{\displaystyle \frac{\pi }{2}}s2\sqrt{st}\mathrm{sin}^1\left(\sqrt{{\displaystyle \frac{s}{t}}}\right).`$
An application of L’Hôpital’s Rule shows that
$$\frac{\pi /2\mathrm{sin}^1x}{\sqrt{1x^2}}1$$
as $`x1`$. Hence, for some positive constant $`C^{}`$, we have $`\mathrm{sin}^1xC^{}\sqrt{1x^2}\pi /2`$ for all $`0x1`$. Now let $`x=s/t`$. Then
$`E|X(t)X(s)|^2`$ $`=t\left[{\displaystyle \frac{\pi }{2}}+{\displaystyle \frac{\pi }{2}}x2\sqrt{x}\mathrm{sin}^1(\sqrt{x})\right]`$
$`t\left[{\displaystyle \frac{\pi }{2}}+{\displaystyle \frac{\pi }{2}}x+2\sqrt{x}\left(C^{}\sqrt{1x}{\displaystyle \frac{\pi }{2}}\right)\right]`$
$`=t\left[{\displaystyle \frac{\pi }{2}}(1\sqrt{x})^2+2C^{}\sqrt{x}\sqrt{1x}\right].`$
Since $`1\sqrt{x}\sqrt{1x}`$ for $`0x1`$,
$`E|X(t)X(s)|^2`$ $`t\left({\displaystyle \frac{\pi }{2}}(1x)+2C^{}\sqrt{x}\sqrt{1x}\right)`$
$`t\left({\displaystyle \frac{\pi }{2}}\sqrt{1x}+2C^{}\sqrt{1x}\right)`$
$`=\sqrt{t}\left({\displaystyle \frac{\pi }{2}}+2C^{}\right)\sqrt{ts}`$
$`C^{\prime \prime }|ts|^{1/2},`$
where $`C^{\prime \prime }=\sqrt{T}(\pi /2+2C^{})`$.
Now, for every $`\alpha >0`$, there is a constant $`K_\alpha `$ such that if $`N`$ is normal with $`EN=0`$, then $`E|N|^\alpha =K_\alpha (EN^2)^{\alpha /2}`$. Thus, for any $`\alpha >4`$, $`E|X(t)X(s)|^\alpha C|ts|^{\alpha /4}`$, where $`C=K_\alpha (C^{\prime \prime })^{\alpha /2}`$. $`\mathrm{}`$
###### Theorem 2.5
Let $`X(t)`$ be as in Theorem 2.4 and let $`0t_1<\mathrm{}<t_d`$, $`d1`$, be arbitrary. Then $`(X_n(t_1),\mathrm{},X_n(t_d))`$ converges in law to $`(X(t_1),\mathrm{},X(t_d))`$ as $`n\mathrm{}`$.
Proof: This is an immediate consequence of Corollary 2.3. $`\mathrm{}`$
It now follows (see, for example, Theorem 2.4.15 in ) that Theorem 2.1 will be proved once we establish the following result.
###### Theorem 2.6
The sequence of processes $`\{X_{2n+1}\}_{n=1}^{\mathrm{}}`$ is tight.
## 3 Conditions for Tightness
A sufficient condition for tightness which will serve as the starting point for our analysis is the following.
###### Theorem 3.1
If $`\{Z_n\}`$ is a sequence of continuous stochastic processes such that
(i) $`sup_nP(|Z_n(t)Z_n(s)|\epsilon )C_T\epsilon ^\alpha |ts|^{1+\beta }`$ whenever $`0<\epsilon <1`$, $`T>0`$, and $`0s,tT`$, and
(ii) $`sup_nE|Z_n(0)|^\nu <\mathrm{}`$
for some positive constants $`\alpha `$, $`\beta `$, $`\nu `$, and $`C_T`$ (depending on $`T`$), then $`\{Z_n\}`$ is tight.
An alternative formulation of this theorem is one in which condition (i) is replaced by
$$\underset{n1}{sup}E|Z_n(t)Z_n(s)|^\alpha C_T|ts|^{1+\beta }.$$
(3.1)
For a proof of this alternative version, the reader is referred to Problem 2.4.11 in , which has a worked solution. An inspection of the proof shows that (3.1) is needed only to establish (via Chebyshev’s inequality) condition (i).
Since the median process inherits the scaling property of Brownian motion, we will find it convenient to reformulate Theorem 3.1. Specifically, for any real number $`c0`$ and any $`x^d`$, we have $`_n(cx)=c_n(x)`$. Hence, the processes $`X_n(c)`$ and $`\sqrt{c}X_n()`$ have the same law. For processes with this scaling property, we can modify Theorem 3.1 in the following way.
###### Theorem 3.2
Let $`\{Z_n\}`$ be a sequence of continuous stochastic processes. Suppose there exists $`r>0`$ such that for every $`c0`$ and every $`n`$, the processes $`Z_n(c)`$ and $`c^rZ_n()`$ have the same law. Suppose further that
(i) $`sup_nP(|Z_n(1+\delta )Z_n(1)|>\epsilon )C\epsilon ^\alpha \delta ^{1+\beta }`$ whenever $`0<\epsilon <1`$ and $`0<\delta <\delta _0`$
for some positive constants $`\delta _0`$, $`C`$, $`\alpha `$, and $`\beta `$. Define $`\gamma =\mathrm{min}(\alpha r,\beta r,1+\beta )`$. If $`\gamma >1`$ and
(ii) $`sup_nE|Z_n(1)|^{\gamma /r}<\mathrm{}`$,
then the sequence $`\{Z_n\}`$ is tight.
Theorem 3.2 follows directly from Theorem 3.1 (a complete proof can be found starting on p.36 of ). We will be applying it to the sequence $`Z_n=X_{2n+1}`$, in which case we have $`r=1/2`$. We will find it quite straightforward to verify condition (ii). To verify condition (i), we will utilize the following lemma, which will be the central focus of the remainder of our analysis.
###### Lemma 3.3
There exists a constant $`\delta _0>0`$ and a family of constants $`\{C_p\}_{p>2}`$ such that for each $`p>2`$,
$$\underset{n3}{sup}P\left(M_n(1+\delta )M_n(1)>\frac{\epsilon }{\sqrt{n}}\right)C_p(\epsilon ^1\delta ^{1/6})^p$$
(3.2)
whenever $`0<\epsilon <1`$ and $`0<\delta \delta _0`$.
It has already been remarked that the limit process $`X`$ behaves locally like a fractional Brownian motion with Hurst parameter $`H=1/4`$. It seems reasonable, then, to conjecture that the right-hand side of (3.2) could be replaced by $`C_p(\epsilon ^1\delta ^{1/4})^p`$. Although this sharper bound was not obtained, the choice of $`1/6`$ as the exponent in (3.2) appears to be arbitrary. Presumably, with minor modifications to the proofs presented here, the right-hand side of (3.2) could be replaced by $`C_p(\epsilon ^1\delta ^\nu )^p`$ for any fixed $`\nu <1/4`$.
Proof of Theorem 2.6, given Lemma 3.3. We apply Theorem 3.2 to $`Z_n=X_{2n+1}`$ with $`r=1/2`$. Choose any $`p>18`$, let $`\alpha =p`$, and let $`\beta =(p6)/6`$. Note that, in this case, $`\gamma =\beta /2>1`$.
To verify condition (i), let $`\delta _0`$ be as in Lemma 3.3. Since $`X_{2n+1}()`$ and $`X_{2n+1}()`$ have the same law,
$`\underset{n1}{sup}P(|X_{2n+1}(1+\delta )X_{2n+1}(1)|>\epsilon )`$ $`=2\underset{n1}{sup}P(X_{2n+1}(1+\delta )X_{2n+1}(1)>\epsilon )`$
$`2C_p(\epsilon ^1\delta ^{1/6})^p`$
$`=2C_p\epsilon ^\alpha \delta ^{1+\beta }`$
whenever $`0<\epsilon <1`$ and $`0<\delta <\delta _0`$.
To verify condition (ii), we will show that for any $`q>0`$,
$$\underset{n1}{sup}E|X_{2n+1}(1)|^q<\mathrm{}.$$
To see this, observe that for $`n`$ odd,
$`E|X_n(1)|^q`$ $`={\displaystyle _0^{\mathrm{}}}qy^{q1}P(|X_n(1)|>y)𝑑y`$
$`=2{\displaystyle _0^{\mathrm{}}}qy^{q1}P(X_n(1)<y)𝑑y.`$
It will therefore suffice to show that for any $`\kappa >2`$, there exists a finite constant $`K`$ such that
$$P(X_n(1)<y)Ky^\kappa $$
(3.3)
for all $`y>0`$ and all $`n`$.
To prove (3.3), we will consider two cases. First, assume $`y2\sqrt{n}`$. Note that by Theorem 1.3.2 in , $`M_n(1)`$ has density
$$f_n(x)=k\left(\genfrac{}{}{0pt}{}{n}{k}\right)\frac{1}{2\pi }\mathrm{\Phi }(x)^{k1}\mathrm{\Phi }(x)^{nk}e^{x^2/2}$$
(3.4)
where $`k=(n+1)/2`$ and $`\mathrm{\Phi }(x)=\frac{1}{2\pi }_{\mathrm{}}^xe^{u^2/2}𝑑u`$. Hence,
$`P(X_n(1)<y)`$ $`=P(M_n(1)<y/\sqrt{n})`$
$`={\displaystyle \frac{n!}{(nk)!(k1)!}}{\displaystyle _{\mathrm{}}^{y/\sqrt{n}}}\mathrm{\Phi }(x)^{k1}\mathrm{\Phi }(x)^{nk}\mathrm{\Phi }^{}(x)𝑑x`$
$`{\displaystyle \frac{n^k}{(k1)!}}{\displaystyle _{\mathrm{}}^{y/\sqrt{n}}}\mathrm{\Phi }(x)^{k1}\mathrm{\Phi }^{}(x)𝑑x`$
$`={\displaystyle \frac{n^k}{k!}}\mathrm{\Phi }(y/\sqrt{n})^k.`$
By Stirling’s formula, there exists a universal positive constant $`C`$ such that $`k!C^1k^ke^k`$. Also, writing $`_x^{\mathrm{}}e^{u^2/2}𝑑u=_x^{\mathrm{}}u^1ue^{u^2/2}𝑑u`$ and integrating by parts, it follows that
$$\sqrt{2\pi }\mathrm{\Phi }(x)x^1e^{x^2/2}$$
(3.5)
for all $`x>0`$. Thus,
$$P(X_n(1)<y)C\frac{n^k}{k^ke^k}\left(\frac{\sqrt{n}}{y}e^{y^2/2n}\right)^k.$$
Since $`y2\sqrt{n}`$ and $`n/k2`$, we have
$$P(X_n(1)<y)Ce^{k(1y^2/2n)}.$$
Since $`1y^2/2n<0`$ and $`kn/2`$, we have
$$P(X_n(1)<y)Ce^{n/2y^2/4}Ce^{y^2/8}.$$
Finally, given $`\kappa >2`$, there exists $`K`$ such that $`Ce^{y^2/8}Ky^\kappa `$ for all $`y>0`$, which verifies (3.3) in the case $`y2\sqrt{n}`$.
Now assume $`y<2\sqrt{n}`$. In this case,
$`P(X_n(1)<y)`$ $`=P(M_n(1)<y/\sqrt{n})`$
$`=P\left({\displaystyle \underset{j=1}{\overset{n}{}}}1_{\{B_j(1)<y/\sqrt{n}\}}\frac{n}{2}\right)`$
$`=P\left({\displaystyle \underset{j=1}{\overset{n}{}}}\xi _jn\left(\frac{1}{2}\mu \right)\right),`$
where $`\mu =\mathrm{\Phi }(y/\sqrt{n})`$ and $`\xi _j=1_{\{B_j(1)<y/\sqrt{n}\}}\mu `$. By Burkholder’s inequality (see, for example, Theorem 6.3.10 in ), there exists a constant $`K^{}`$, depending only on $`\kappa `$, such that
$$E\left|\underset{j=1}{\overset{n}{}}\xi _j\right|^\kappa K^{}E\left|\underset{j=1}{\overset{n}{}}|\xi _j|^2\right|^{\kappa /2}.$$
Hence, since $`\kappa >2`$, Jensen’s inequality and the fact that $`|\xi _j|1`$ a.s. imply
$$E\left|\underset{j=1}{\overset{n}{}}\xi _j\right|^\kappa K^{}n^{\kappa /2}E\underset{j=1}{\overset{n}{}}\frac{1}{n}|\xi _j|^kK^{}n^{\kappa /2}.$$
Chebyshev’s inequality now gives
$$P(X_n(1)<y)\frac{K^{}n^{\kappa /2}}{\left|n\left(\frac{1}{2}\mu \right)\right|^\kappa }=K^{}\left|\sqrt{n}\left(\frac{1}{2}\mu \right)\right|^\kappa .$$
Since
$$\sqrt{n}\left(\frac{1}{2}\mu \right)=\frac{\sqrt{n}}{\sqrt{2\pi }}_0^{y/\sqrt{n}}e^{u^2/2}𝑑u\frac{y}{\sqrt{2\pi }}e^{y^2/2n}\frac{y}{\sqrt{2\pi }}e^2,$$
we have that $`P(X_n(1)<y)Ky^\kappa `$, where $`K=K^{}(e^2/\sqrt{2\pi })^\kappa `$. This verifies (3.3) when $`y<2\sqrt{n}`$ and completes the proof. $`\mathrm{}`$
Our goal for the remainder of this article is to establish (3.2). Since each individual Brownian particle can be expected to move a distance of $`\sqrt{\delta }`$ between time $`t=1`$ and $`t=1+\delta `$, we will accomplish our goal by considering three different “jump regimes.” They are: the large jump regime in which $`\epsilon /\sqrt{n}`$ is much larger than $`\sqrt{\delta }`$, the small jump regime in which $`\epsilon /\sqrt{n}`$ is much smaller than $`\sqrt{\delta }`$, and the medium jump regime in which these two quantities are comparable. In the first two regimes, we will establish the sharp bound mentioned in the remark following Lemma 3.3. The bound in the medium jump regime will be established by modifying the techniques used in the small jump regime. This modification will result in the weaker bound given in (3.2).
## 4 The Large Jump Regime
The large jump regime is the easiest of the three to deal with. The probability that the median makes a large jump can be bounded above by the probability that at least one Brownian particle makes a large jump. Since the latter probability is exponentially small, the derivation of (3.2) is immediate.
###### Lemma 4.1
Fix $`p>0`$ and $`0<\mathrm{\Delta }<1/2`$. Suppose that $`\epsilon ,\delta (0,1)`$ and $`n`$ satisfy $`\epsilon /\sqrt{n}\delta ^{1/2\mathrm{\Delta }}`$. Then
$$P\left(M_n(1+\delta )M_n(1)>\frac{\epsilon }{\sqrt{n}}\right)C(\epsilon ^1\delta ^{1/4})^p,$$
where $`C`$ depends only on $`p`$ and $`\mathrm{\Delta }`$.
Proof: Suppose that $`B_j(1+\delta ,\omega )B_j(1,\omega )\epsilon /\sqrt{n}`$ for all $`j`$. Then, for each $`j`$ such that $`B_j(1,\omega )M_n(1,\omega )`$, we have $`B_j(1+\delta ,\omega )M_n(1,\omega )+\epsilon /\sqrt{n}`$. Note that there are at least $`k=(n+1)/2`$ such values of $`j`$. It follows that $`M_n(1+\delta ,\omega )M_n(1,\omega )+\epsilon /\sqrt{n}`$. Therefore,
$$\underset{j=1}{\overset{n}{}}\{B_j(1+\delta )B_j(1)\epsilon /\sqrt{n}\}\{M_n(1+\delta )M_n(1)\epsilon /\sqrt{n}\},$$
which gives
$`P\left(M_n(1+\delta )M_n(1)>{\displaystyle \frac{\epsilon }{\sqrt{n}}}\right)`$ $`P\left({\displaystyle \underset{j=1}{\overset{n}{}}}\{B_j(1+\delta )B_j(1)>\epsilon /\sqrt{n}\}\right)`$
$`n\mathrm{\Phi }(\epsilon /\sqrt{n\delta }).`$
For each $`r>0`$, there exists $`C_r`$ such that$`\mathrm{\Phi }(x)C_rx^r`$ for all $`x>0`$. Taking $`r=(p/4+1)/\mathrm{\Delta }`$ gives
$$P\left(M_n(1+\delta )M_n(1)>\frac{\epsilon }{\sqrt{n}}\right)nC_r\left(\frac{\epsilon }{\sqrt{n\delta }}\right)^rnC_r(\delta ^\mathrm{\Delta })^r=C_rn\delta ^{p/4+1}.$$
The proof is completed by observing that $`n\epsilon ^2\delta ^1\epsilon ^p\delta ^1`$. $`\mathrm{}`$
This establishes the necessary bound for the large jump regime. The other regimes, as we will see, are considerably more difficult to deal with.
## 5 Conditioning the Median
To establish (3.2) for the small and medium jump regimes, we will use conditioning. It may seem natural, at first, to condition on the locations of all the Brownian particles at time $`t=1`$. It turns out, however, that this is, in some sense, too much information. Rather, we shall condition only on the location of the median particle at time $`t=1`$.
Let us first give a heuristic description of this conditioning. Suppose that $`M_n(1)=x`$. This tells us three things. First, we have a single Brownian particle whose location is $`x`$. Second, we have roughly $`n/2`$ Brownian particles whose locations are less than $`x`$. Other than this condition on their locations, these particles are independent and identically distributed. We will refer to these particles as the “lower” particles. Third, we have roughly $`n/2`$ i.i.d. Brownian particles whose locations are greater than $`x`$. These will naturally be referred to as the “upper” particles.
Let us now fix $`y>0`$ and consider the event $`D=\{M_n(1+\delta )M_n(1)>y\}`$. This event will occur if and only if there are at least $`n/2`$ particles whose location at time $`t=1+\delta `$ is greater than $`x+y`$. Particles that satisfy this condition will be said to have “jumped.” Let $`U(j)`$ be the event that the $`j`$-th upper particle jumps, and let $`L(j)`$ be the event that the $`j`$-th lower particle does not jump. Then the total number of particles that jump is
$$1_{U(j)}+\left(\frac{n}{2}1_{L(j)}\right).$$
The event $`D`$ will occur if and only if this sum is at least $`n/2`$, which occurs if and only if $`Y_j0`$, where $`Y_j=1_{U(j)}1_{L(j)}`$ are i.i.d. $`\{1,0,1\}`$-valued random variables. Through conditioning, then, we are able to transform the event of interest into one involving an i.i.d. sum.
With these heuristics in place, let us establish the rigorous result. Define
$`p_1=p_1(x,y,\delta )`$ $`=P(B(1+\delta )<x+y|B(1)<x)`$ (5.1)
$`p_2=p_2(x,y,\delta )`$ $`=P(B(1+\delta )>x+y|B(1)>x)`$ (5.2)
$`=p_1(x,y,\delta )`$
and
$$q_j=1p_j.$$
(5.3)
In the language of our heuristics, $`p_1`$ is the probability that a lower particle does not jump and $`p_2`$ is the probability that an upper particle does jump.
Now, for each fixed triple $`(x,y,\delta )`$, let $`\{\xi _j^L\}_{j=1}^{\mathrm{}}`$ and $`\{\xi _j^U\}_{j=1}^{\mathrm{}}`$ be sequences of i.i.d $`\{0,1\}`$-valued random variables with $`P(\xi _j^L=1)=p_1`$ and $`P(\xi _j^U=1)=p_2`$. Define $`Y_j=\xi _j^U\xi _j^L`$. Observe that $`\{Y_j\}_{j=1}^{\mathrm{}}`$ is an i.i.d. sequence of $`\{1,0,1\}`$-valued random variables and, for future reference, define
$`\stackrel{~}{p}_1`$ $`=P(Y_j=1)=p_1q_2`$ (5.4)
$`\stackrel{~}{p}_2`$ $`=P(Y_j=1)=p_2q_1`$ (5.5)
$`\stackrel{~}{\epsilon }`$ $`=P(Y_j0)=\stackrel{~}{p}_1+\stackrel{~}{p}_2`$ (5.6)
$`\stackrel{~}{\mu }`$ $`=EY_j=\stackrel{~}{p}_1\stackrel{~}{p}_2.`$ (5.7)
Finally, let $`S_k=_{j=1}^kY_j`$ and $`\phi _k(x,y,\delta )=P(S_k0)`$.
Our heuristics suggest that
$$P(M_n(1+\delta )M_n(1)>y|M_n(1)=x)\phi _{n/2}(x,y,\delta ).$$
For a rigorous statement, the following inequality will serve our purposes.
###### Theorem 5.1
Let $`n3`$ and $`k=(n+1)/2`$. Then for all $`y>0`$ and all $`\delta >0`$,
$$P(M_n(1+\delta )M_n(1)>y)_{\mathrm{}}^{\mathrm{}}\phi _{k1}(x,y,\delta )f_n(x)𝑑x,$$
where $`f_n(x)`$ is the density of $`M_n(1)`$, given by (3.4).
Proof: First, let us observe that
$`\phi _k(x,y,\delta )=P(S_k0)`$ $`=P\left({\displaystyle \underset{j=1}{\overset{k}{}}}\xi _j^U{\displaystyle \underset{j=1}{\overset{k}{}}}\xi _j^L\right)`$
$`={\displaystyle \underset{\mathrm{}=0}{\overset{k}{}}}{\displaystyle \underset{m=\mathrm{}}{\overset{k}{}}}P\left({\displaystyle \underset{j=1}{\overset{k}{}}}\xi _j^L=\mathrm{},{\displaystyle \underset{j=1}{\overset{k}{}}}\xi _j^U=m\right)`$
$`={\displaystyle \underset{\mathrm{}=0}{\overset{k}{}}}{\displaystyle \underset{m=\mathrm{}}{\overset{k}{}}}\left({\displaystyle \genfrac{}{}{0pt}{}{k}{\mathrm{}}}\right)\left({\displaystyle \genfrac{}{}{0pt}{}{k}{m}}\right)p_1^{\mathrm{}}q_1^k\mathrm{}p_2^mq_2^{km}.`$
Let us also adopt the following notation: for $`h>0`$, let $`p_{1,h}=p_1(x+h,yh,\delta )`$ and
$$\phi _k^h(x,y,\delta )=\underset{\mathrm{}=0}{\overset{k}{}}\underset{m=\mathrm{}}{\overset{k}{}}\left(\genfrac{}{}{0pt}{}{k}{\mathrm{}}\right)\left(\genfrac{}{}{0pt}{}{k}{m}\right)p_{1,h}^{\mathrm{}}q_{1,h}^k\mathrm{}p_2^mq_2^{km},$$
where $`q_{1,h}=1p_{1,h}`$. Finally, let $`\mathrm{\Delta }M_n=M_n(1+\delta )M_n(1)`$.
Now, fix $`\delta >0`$ and $`y>0`$. Let $`K`$ and let $`h>0`$ with $`K/h`$. Then
$$P(\mathrm{\Delta }M_n>y,|M_n(1)|K)\underset{\begin{array}{c}xh\\ |x|K\end{array}}{}P\left(M_n(1+\delta )>x+y,M_n(1)[x,x+h)\right).$$
Let $`\mathrm{SS}_n=\{1,\mathrm{},n\}`$ and let $`S=S_n`$ denote the collection of all ordered pairs $`(I,j)`$ where $`I\mathrm{SS}_n`$ and $`j\mathrm{SS}_n`$ satisfy $`|I|=k1`$ and $`jI`$. For $`(I,j)S`$, $`x`$, and $`h>0`$, define $`I(j)^c=\mathrm{SS}_n(I\{j\})`$ and
$$\begin{array}{cc}\hfill A(I,j,x,h)& =\{B_j(1)[x,x+h)\}\hfill \\ & \{B_i(1)<B_j(1),iI\}\{B_i(1)>B_j(1),iI(j)^c\},\hfill \\ \hfill \stackrel{~}{A}(I,j,x,h)& =\{B_j(1)[x,x+h)\}\hfill \\ & \{B_i(1)<x+h,iI\}\{B_i(1)>x,iI(j)^c\}.\hfill \end{array}$$
Note that $`\{M_n(1)[x,x+h)\}=\{A(I,j,x,h):(I,j)S\}`$ up to a set of measure zero, and that this is a disjoint union. Therefore,
$`P(M_n(1+\delta )>x+y,M_n(1)[x,x+h))`$ $`={\displaystyle \underset{(I,j)S}{}}P(M_n(1+\delta )>x+y,A(I,j,x,h))`$
$`{\displaystyle \underset{(I,j)S}{}}P(M_n(1+\delta )>x+y,\stackrel{~}{A}(I,j,x,h)),`$
since $`A(I,j,x,h)\stackrel{~}{A}(I,j,x,h)`$.
Now fix $`(I,j)S`$ and $`x`$. Define
$`N_1`$ $`={\displaystyle \underset{iI}{}}\mathrm{\hspace{0.17em}\hspace{0.17em}1}_{\{B_i(1+\delta )<x+y\}}`$
$`N_2`$ $`={\displaystyle \underset{iI(j)^c}{}}1_{\{B_i(1+\delta )>x+y\}}`$
$`N`$ $`={\displaystyle \underset{i=1}{\overset{n}{}}}\mathrm{\hspace{0.17em}\hspace{0.17em}1}_{\{B_i(1+\delta )>x+y\}}`$
and note that $`\{M_n(1+\delta )>x+y\}=\{Nnk+1\}`$. Also note that, up to a set of measure zero,
$`N`$ $`=N_2+(k1)N_1+1_{\{B_j(1+\delta )>x+y\}}`$
$`N_2N_1+k.`$
Thus, if $`d(n)=n2k+1`$, then $`\{M_n(1+\delta )>x+y\}\{N_2N_1d(n)\}`$. This gives
$`P(M_n(1+\delta )>x+y,\stackrel{~}{A}(I,j,x,h))`$ $`P(N_2N_1d(n),\stackrel{~}{A}(I,j,x,h))`$
$`={\displaystyle \underset{\mathrm{}=0}{\overset{k1}{}}}{\displaystyle \underset{m=d(n)+\mathrm{}}{\overset{nk}{}}}P(N_1=\mathrm{},N_2=m,\stackrel{~}{A}(I,j,x,h)).`$
Hence, if we define
$`P_1(\mathrm{})`$ $`=P(\{N_1=\mathrm{}\}\{B_i(1)<x+h,iI\}),`$
$`P_2(m)`$ $`=P(\{N_2=m\}\{B_i(1)>x,iI(j)^c\}),`$
then we can write
$$P(M_n(1+\delta )>x+y,\stackrel{~}{A}(I,j,x,h))\underset{\mathrm{}=0}{\overset{k1}{}}\underset{m=d(n)+\mathrm{}}{\overset{nk}{}}P(B_j(1)[x,x+h))P_1(\mathrm{})P_2(m).$$
Since
$$P(\stackrel{~}{A}(I,j,x,h))=P(B_j(1)[x,x+h))\mathrm{\Phi }(x+h)^{k1}\mathrm{\Phi }(x)^{nk},$$
this gives
$$P(M_n(1+\delta )>x+y|\stackrel{~}{A}(I,j,x,h))\underset{\mathrm{}=0}{\overset{k1}{}}\underset{m=d(n)+\mathrm{}}{\overset{nk}{}}\frac{P_1(\mathrm{})}{\mathrm{\Phi }(x+h)^{k1}}\frac{P_2(m)}{\mathrm{\Phi }(x)^{nk}}$$
for each fixed $`I`$, $`j`$, and $`x`$.
To simplify this double sum, let
$$\begin{array}{cc}\hfill \psi (x,y,\delta )& =P(B(1+\delta )<x+y,B(1)<x)\hfill \\ & =_{\mathrm{}}^x\mathrm{\Phi }\left(\frac{x+yt}{\sqrt{\delta }}\right)\mathrm{\Phi }^{}(t)𝑑t.\hfill \end{array}$$
(5.8)
Then by symmetry and independence,
$`P_1(\mathrm{})`$ $`=\left({\displaystyle \genfrac{}{}{0pt}{}{k1}{\mathrm{}}}\right)[\psi (x+h,yh)]^{\mathrm{}}[\mathrm{\Phi }(x+h)\psi (x+h,yh)]^{k1\mathrm{}},`$
$`P_2(m)`$ $`=\left({\displaystyle \genfrac{}{}{0pt}{}{nk}{m}}\right)[\psi (x,y)]^m[\mathrm{\Phi }(x)\psi (x,y)]^{nkm}.`$
Also note that
$$\frac{\psi (x+h,yh)}{\mathrm{\Phi }(x+h)}=P(B(1+\delta )<x+y|B(1)<x+h)=p_{1,h}$$
and
$$\frac{\psi (x,y)}{\mathrm{\Phi }(x)}=P(B(1+\delta )>x+y|B(1)>x)=p_2,$$
which yields
$$P(M_n(1+\delta )>x+y|\stackrel{~}{A}(I,j,x,h))\underset{\mathrm{}=0}{\overset{k1}{}}\underset{m=d(n)+\mathrm{}}{\overset{nk}{}}\left(\genfrac{}{}{0pt}{}{k1}{\mathrm{}}\right)\left(\genfrac{}{}{0pt}{}{nk}{m}\right)p_{1,h}^{\mathrm{}}q_{1,h}^{k1\mathrm{}}p_2^mq_2^{nkm}$$
for each fixed $`I`$, $`j`$, and $`x`$.
Now suppose $`n`$ is odd. In this case, $`d(n)=0`$ and $`nk=k1`$, so
$$P(M_n(1+\delta )>x+y|\stackrel{~}{A}(I,j,x,h))\phi _{k1}^h(x,y,\delta ).$$
(5.9)
On the other hand, if $`n`$ is even, then $`d(n)=1`$ and $`nk=k`$, so
$`P(M_n(1+\delta )>x+y|\stackrel{~}{A}(I,j,x,h))`$ $`{\displaystyle \underset{\mathrm{}=0}{\overset{k1}{}}}{\displaystyle \underset{m=\mathrm{}+1}{\overset{k}{}}}\left({\displaystyle \genfrac{}{}{0pt}{}{k1}{\mathrm{}}}\right)\left({\displaystyle \genfrac{}{}{0pt}{}{k}{m}}\right)p_{1,h}^{\mathrm{}}q_{1,h}^{k1\mathrm{}}p_2^mq_2^{km}`$
$`={\displaystyle \underset{\mathrm{}=0}{\overset{k1}{}}}\left({\displaystyle \genfrac{}{}{0pt}{}{k1}{\mathrm{}}}\right)p_{1,h}^{\mathrm{}}q_{1,h}^{k1\mathrm{}}{\displaystyle \underset{m=\mathrm{}+1}{\overset{k}{}}}\left({\displaystyle \genfrac{}{}{0pt}{}{k}{m}}\right)p_2^mq_2^{km}.`$
But
$$\underset{m=\mathrm{}+1}{\overset{k}{}}\left(\genfrac{}{}{0pt}{}{k}{m}\right)p_2^mq_2^{km}=P\left(\underset{j=1}{\overset{k}{}}\xi _j^U>\mathrm{}\right)P\left(\underset{j=1}{\overset{k1}{}}\xi _j^U\mathrm{}\right)=\underset{m=\mathrm{}}{\overset{k1}{}}\left(\genfrac{}{}{0pt}{}{k1}{m}\right)p_2^mq_2^{k1m},$$
so (5.9) holds in this case as well.
Putting it all together, we have
$`P(\mathrm{\Delta }M_n>y,|M_n(1)|K)`$ $`{\displaystyle \underset{\begin{array}{c}xh\\ |x|K\end{array}}{}}{\displaystyle \underset{(I,j)S}{}}P(M_n(1+\delta )>x+y,\stackrel{~}{A}(I,j,x,h))`$
$`{\displaystyle \underset{\begin{array}{c}xh\\ |x|K\end{array}}{}}{\displaystyle \underset{(I,j)S}{}}\phi _{k1}^h(x,y,\delta )P(\stackrel{~}{A}(I,j,x,h))`$
$`={\displaystyle \underset{\begin{array}{c}xh\\ |x|K\end{array}}{}}{\displaystyle \underset{(I,j)S}{}}\phi _{k1}^h(x,y,\delta ){\displaystyle \frac{P(\stackrel{~}{A})}{P(A)}}P(A(I,j,x,h)).`$
Note that $`P(A(I,j,x,h))P(B_j(1)[x,x+h))\mathrm{\Phi }(x)^{k1}\mathrm{\Phi }(xh)^{nk}`$, so that
$$\frac{P(\stackrel{~}{A})}{P(A)}\left[\frac{\mathrm{\Phi }(x+h)}{\mathrm{\Phi }(x)}\right]^{k1}\left[\frac{\mathrm{\Phi }(x)}{\mathrm{\Phi }(xh)}\right]^{nk}.$$
If we denote the right-hand side of this inequality by $`g_h(x)`$, then by dominated convergence,
$`P(\mathrm{\Delta }M_n>y,|M_n(1)|K)`$ $`{\displaystyle \underset{\begin{array}{c}xh\\ |x|K\end{array}}{}}\phi _{k1}^h(x,y,\delta )g_h(x){\displaystyle \underset{(I,j)S}{}}P(A(I,j,x,h))`$
$`={\displaystyle \underset{\begin{array}{c}xh\\ |x|K\end{array}}{}}\phi _{k1}^h(x,y,\delta )g_h(x)P(M_n(1)[x,x+h))`$
$`{\displaystyle _K^K}\phi _{k1}(x,y,\delta )f_n(x)𝑑x.`$
Letting $`K\mathrm{}`$ finishes the proof. $`\mathrm{}`$
The estimate in Theorem 5.1 can be simplified even further and we will find it convenient to use the following.
###### Corollary 5.2
Let $`n3`$, $`k=(n+1)/2`$, $`y>0`$, and $`\delta >0`$. Then
$$P(M_n(1+\delta )M_n(1)>y)\phi _{k1}(x_0,y,\delta )+P(M_n(1)x_0)$$
(5.10)
for all $`x_0`$.
Proof: We will first show that $`x\phi _{k1}(x,y,\delta )`$ is decreasing, for which it will suffice to show that $`xp_1(x,y,\delta )`$ is increasing. To see this, recall that $`\phi _{k1}(x,y,\delta )=P(_{j=1}^{k1}Y_j0)`$. If $`xp_1(x,y,\delta )`$ is increasing, then $`xp_2(x,y,\delta )=p_1(x,y,\delta )`$ is decreasing. Hence, by (5.4) and (5.5), $`P(Y_j=1)=p_1(1p_2)`$ increases with $`x`$ and $`P(Y_j=1)=p_2(1p_1)`$ decreases with $`x`$, which shows that $`x\phi _{k1}(x,y,\delta )`$ is decreasing.
With $`\psi `$ as in (5.8), we have $`p_1=\psi /\mathrm{\Phi }(x)`$ and
$$_xp_1=\frac{\mathrm{\Phi }^{}(x)}{[\mathrm{\Phi }(x)]^2}\psi +\frac{1}{\mathrm{\Phi }(x)}\left[\mathrm{\Phi }\left(\frac{y}{\sqrt{\delta }}\right)\mathrm{\Phi }^{}(x)+\frac{1}{\sqrt{\delta }}_{\mathrm{}}^x\mathrm{\Phi }^{}\left(\frac{x+yt}{\sqrt{\delta }}\right)\mathrm{\Phi }^{}(t)𝑑t\right].$$
(5.11)
Integrating by parts gives
$$\psi (x,y,\delta )=\mathrm{\Phi }\left(\frac{y}{\sqrt{\delta }}\right)\mathrm{\Phi }(x)+\frac{1}{\sqrt{\delta }}_{\mathrm{}}^x\mathrm{\Phi }^{}\left(\frac{x+yt}{\sqrt{\delta }}\right)\mathrm{\Phi }(t)𝑑t.$$
Substituting this into (5.11) gives
$`_xp_1`$ $`={\displaystyle \frac{\mathrm{\Phi }^{}(x)}{[\mathrm{\Phi }(x)]^2\sqrt{\delta }}}{\displaystyle _{\mathrm{}}^x}\mathrm{\Phi }^{}\left({\displaystyle \frac{x+yt}{\sqrt{\delta }}}\right)\mathrm{\Phi }(t)𝑑t+{\displaystyle \frac{1}{\mathrm{\Phi }(x)\sqrt{\delta }}}{\displaystyle _{\mathrm{}}^x}\mathrm{\Phi }^{}\left({\displaystyle \frac{x+yt}{\sqrt{\delta }}}\right)\mathrm{\Phi }^{}(t)𝑑t`$
$`={\displaystyle \frac{1}{\mathrm{\Phi }(x)\sqrt{\delta }}}{\displaystyle _{\mathrm{}}^x}\mathrm{\Phi }^{}\left({\displaystyle \frac{x+yt}{\sqrt{\delta }}}\right)\left[{\displaystyle \frac{\mathrm{\Phi }^{}(t)}{\mathrm{\Phi }(t)}}{\displaystyle \frac{\mathrm{\Phi }^{}(x)}{\mathrm{\Phi }(x)}}\right]\mathrm{\Phi }(t)𝑑t.`$ (5.12)
Note that
$`{\displaystyle \frac{d}{dx}}\left[{\displaystyle \frac{\mathrm{\Phi }^{}(x)}{\mathrm{\Phi }(x)}}\right]`$ $`={\displaystyle \frac{\mathrm{\Phi }^{\prime \prime }(x)\mathrm{\Phi }(x)[\mathrm{\Phi }^{}(x)]^2}{[\mathrm{\Phi }(x)]^2}}`$
$`={\displaystyle \frac{1}{[\mathrm{\Phi }(x)]^2}}\left({\displaystyle \frac{1}{\sqrt{2\pi }}}xe^{x^2/2}\mathrm{\Phi }(x){\displaystyle \frac{1}{2\pi }}e^{x^2}\right)`$
$`={\displaystyle \frac{e^{x^2/2}}{\sqrt{2\pi }[\mathrm{\Phi }(x)]^2}}\left(x\mathrm{\Phi }(x)+{\displaystyle \frac{1}{\sqrt{2\pi }}}e^{x^2/2}\right).`$
Clearly, $`x\mathrm{\Phi }(x)+\frac{1}{\sqrt{2\pi }}e^{x^2/2}0`$ for $`x0`$. If $`x<0`$, then by (3.5),
$`x\mathrm{\Phi }(x)+{\displaystyle \frac{1}{\sqrt{2\pi }}}e^{x^2/2}`$ $`=x\mathrm{\Phi }(|x|)+{\displaystyle \frac{1}{\sqrt{2\pi }}}e^{x^2/2}`$
$`x{\displaystyle \frac{1}{\sqrt{2\pi }}}|x|^1e^{x^2/2}+{\displaystyle \frac{1}{\sqrt{2\pi }}}e^{x^2/2}`$
$`=0.`$
Thus, $`x\mathrm{\Phi }^{}(x)/\mathrm{\Phi }(x)`$ is decreasing, so by (5.12), $`_xp_10`$.
Hence, $`x\phi _{k1}(x,y,\delta )`$ is decreasing, and using Theorem 5.1,
$`P(M_n(1+\delta )M_n(1)>y)`$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}\phi _{k1}(x,y,\delta )f_n(x)𝑑x`$
$`{\displaystyle _{\mathrm{}}^{x_0}}\phi _{k1}(x,y,\delta )f_n(x)𝑑x+\phi _{k1}(x_0,y,\delta ){\displaystyle _{x_0}^{\mathrm{}}}f_n(x)𝑑x`$
$`{\displaystyle _{\mathrm{}}^{x_0}}f_n(x)𝑑x+\phi _{k1}(x_0,y,\delta ){\displaystyle _{\mathrm{}}^{\mathrm{}}}f_n(x)𝑑x`$
$`=P(M_n(1)x_0)+\phi _{k1}(x_0,y,\delta ),`$
where $`x_0`$ is arbitrary. $`\mathrm{}`$
Recall that our only remaining goal is to establish the inequality (3.2) for the small and medium jump regimes. In applying Corollary 5.2 to this task, we must set $`y=\epsilon /\sqrt{n}`$. Our choice for $`x_0`$, however, is less clear. On the one hand, we want $`x_0`$ to be large so that the first term on the right-hand of (5.10) is small. On the other hand, we need $`x_0`$ to be sufficiently far into the negative real line so that the second term is small. The value of $`x_0`$ that will strike a balance for us is given in the following lemma.
###### Lemma 5.3
Let $`\epsilon >0`$, $`\delta >0`$, and $`n`$. Define $`x_0=\epsilon /(\delta ^{1/4}\sqrt{n})`$. Then for all $`p>2`$,
$$P(M_n(1)x_0)C_p(\epsilon ^1\delta ^{1/4})^p,$$
where $`C_p`$ is a finite constant depending only $`p`$.
Proof: This follows immediately from (3.3). $`\mathrm{}`$
In light of this lemma and Corollary 5.2, we will establish inequality (3.2) once we verify that
$$\phi _{k1}(\frac{\epsilon }{\delta ^{1/4}\sqrt{n}},\frac{\epsilon }{\sqrt{n}},\delta )C_p(\epsilon ^1\delta ^{1/6})^p$$
(5.13)
for all values of $`\epsilon `$, $`\delta `$, and $`n`$ in the small and medium jump regimes.
## 6 Estimates for a Random Walk
In this section, we wish to find useful estimates for $`\phi _k(x,y,\delta )=P(S_k0)`$. The process $`\{S_n\}_{n=1}^{\mathrm{}}`$ is, of course, a biased random walk which, in the cases we are interested in, has a negative drift. Let us recall the definition of $`S_n`$. In this section, we will temporarily abandon the tilde notation for the sake of simplicity.
We take as given a sequence of $`\{1,0,1\}`$-valued random variables with $`p_1=P(Y_j=1)`$ and $`p_2=P(Y_j=1)`$. We define $`\epsilon =p_1+p_2`$ and $`\mu =p_1p_2`$, so that $`P(Y_j=0)=1\epsilon `$. We then define $`S_n=_{j=1}^nY_j`$.
As mentioned, we will be interested in the case where $`\mu >0`$, so that the walk has a negative drift. Besides this, however, we will also be interested in the case where $`\epsilon `$ is small. That is, besides the negative drift, our walk will have the property that, for most time steps, it does not move. Our first estimate is a straightforward application of Chebyshev’s inequality. It is a fairly simple result and serves as our starting point, but it will not be sufficient by itself. Note, in particular, that it does not make any noteworthy use of the fact that $`\epsilon `$ is small.
###### Lemma 6.1
If $`\epsilon >0`$ and $`\mu >0`$, then for all $`p>1`$, there exists $`C_p`$, depending only on $`p`$, such that
$$P(S_n0)C_p\frac{\epsilon }{n^p\mu ^{2p}}$$
(6.1)
for all $`n`$.
Proof: Since $`EY_j=\mu `$, Chebyshev’s inequality gives
$$P(S_n0)=P(S_n+n\mu n\mu )\frac{E|S_n+n\mu |^{2p}}{n^{2p}\mu ^{2p}}.$$
By Burkholder’s and Jensen’s inequalities,
$$E|S_n+n\mu |^{2p}=E\left|\underset{j=1}{\overset{n}{}}(Y_j+\mu )\right|^{2p}\stackrel{~}{C}_pE\left|\underset{j=1}{\overset{n}{}}|Y_j+\mu |^2\right|^p\stackrel{~}{C}_pn^pE|Y_1+\mu |^{2p}.$$
Also,
$`E|Y_1+\mu |^{2p}`$ $`=p_1(1\mu )^{2p}+(1\epsilon )\mu ^{2p}+p_2(1+\mu )^{2p}`$
$`2^{2p}(p_1+p_2)+\mu ^{2p}`$
$`(2^{2p}+1)\epsilon `$
since $`\mu \epsilon `$. Thus, (6.1) holds with $`C_p=\stackrel{~}{C}_p(2^{2p}+1)`$. $`\mathrm{}`$
As it stands, (6.1) will not suit our needs. We will find it necessary for the numerator on the right-hand side of (6.1) to contain $`\epsilon ^p`$ rather than $`\epsilon `$. To accomplish this, we must appeal to the fact that, for the most part, this random walk does not move. To this end, we begin with two lemmas.
###### Lemma 6.2
For $`n`$, $`k\{0,\mathrm{},n\}`$, $`p(0,1)`$, and $`x`$, let $`f(n,k,p)=\left(\genfrac{}{}{0pt}{}{n}{k}\right)p^kq^{nk}`$, where $`q=1p`$, and let $`g(n,x,p)=(2\pi npq)^{1/2}\mathrm{exp}\{(xnp)^2/2npq\}`$. Then
$$\underset{n}{sup}\left(\underset{k\{0,\mathrm{},n\}}{sup}\frac{f(n,k,p)}{g(n,k,p)}\right)<\mathrm{}$$
if and only if $`p=1/2`$. However, there exists a universal constant $`C`$, independent of $`p`$, such that $`f(n,k,p)/g(n,k,p)C`$ for all $`n`$ and all $`k\{0,\mathrm{},np\}`$, provided $`p1/2`$.
Proof: It will first be shown that there exists a universal constant $`C`$ such that
(i) if $`p1/2`$, then $`f(n,0,p)/g(n,0,p)C`$, and
(ii) if $`p1/2`$ and $`np1`$, then $`f(n,1,p)/g(n,1,p)C`$.
We will start by showing that if $`\alpha >0`$, then there exists a constant $`C_\alpha `$, depending only on $`\alpha `$, such that for all $`p1/2`$,
$$(np)^\alpha (qe^{p/2q})^nC_\alpha .$$
(6.2)
To prove this, first consider $`2/5p1/2`$. In this case, $`qe^{p/2q}\frac{3}{5}e^{1/2}<1`$. Thus,
$$(np)^\alpha (qe^{p/2q})^n\underset{n}{sup}\left[n^\alpha \left(\frac{3}{5}e^{1/2}\right)^n\right]<\mathrm{}.$$
Next, consider $`0<p<2/5`$. Since $`\frac{d}{dq}[\mathrm{log}(q^{5/6}e^{p/2q})]=(5q3)/6q^2>0`$ for $`q>3/5`$, it follows that in this case, $`q^{5/6}e^{p/2q}1`$. Hence,
$$(np)^\alpha (qe^{p/2q})^n(np)^\alpha q^{n/6}=(n^\alpha q^{n/6})p^\alpha .$$
Elementary calculus shows that $`xx^\alpha q^{x/6}`$ attains its maximum on $`[0,\mathrm{})`$ at $`x=6\alpha /\mathrm{log}q`$. Thus,
$$n^\alpha q^{n/6}p^\alpha \left(\frac{6\alpha }{e}\right)^\alpha \left(\frac{1q}{|\mathrm{log}q|}\right)^\alpha .$$
Since $`(q1)/\mathrm{log}q1`$ as $`q1`$, this proves (6.2). Thus, if $`p1/2`$, then
$$\frac{f(n,0,p)}{g(n,0,p)}=\sqrt{2\pi npq}q^ne^{np/2q}=\sqrt{2\pi q}(np)^{1/2}(qe^{p/2q})^n\sqrt{2\pi }C_{1/2},$$
and if $`p1/2`$ and $`np1`$, then
$`{\displaystyle \frac{f(n,1,p)}{g(n,1,p)}}`$ $`=\sqrt{2\pi npq}npq^{n1}\mathrm{exp}\left\{{\displaystyle \frac{np}{2q}}{\displaystyle \frac{1}{q}}+{\displaystyle \frac{1}{2npq}}\right\}`$
$`\sqrt{2\pi q}q^{n1}(np)^{3/2}e^{np/2q}`$
$`=\sqrt{{\displaystyle \frac{2\pi }{q}}}(np)^{3/2}(qe^{p/2q})^n`$
$`\sqrt{4\pi }C_{3/2},`$
which verifies (i) and (ii).
Now, for $`k\{1,\mathrm{},n1\}`$, Stirling’s formula implies that $`f(n,k,p)`$ is bounded above and below by universal, positive constant multiples of
$$\frac{n^{n+\frac{1}{2}}}{(nk)^{nk+\frac{1}{2}}k^{k+\frac{1}{2}}}p^kq^{nk}.$$
Let us define
$`F(k)=F(n,k,p)`$ $`=\mathrm{log}\left({\displaystyle \frac{n^{n+\frac{1}{2}}}{(nk)^{nk+\frac{1}{2}}k^{k+\frac{1}{2}}}}p^kq^{nk}\right)\mathrm{log}(\sqrt{2\pi }g(n,k,p))`$
$`=(n+1)\mathrm{log}n(nk+\frac{1}{2})\mathrm{log}(nk)(k+\frac{1}{2})\mathrm{log}k`$
$`+(k+\frac{1}{2})\mathrm{log}p+(nk+\frac{1}{2})\mathrm{log}q+(knp)^2/2npq,`$
so that there are universal, positive constants $`C_1`$ and $`C_2`$ such that
$$\mathrm{log}C_1+F(k)\mathrm{log}\left[\frac{f(n,k,p)}{g(n,k,p)}\right]\mathrm{log}C_2+F(k)$$
(6.3)
for all $`k\{1,\mathrm{},n1\}`$. Note that $`F(k)`$ is well-defined for all real $`k(0,n)`$.
We can directly compute that
$$F(n/2)=\frac{1}{2}\mathrm{log}(4pq)+\frac{n}{2}(G(p)+G(1p)),$$
where $`G(p)=\mathrm{log}2+\mathrm{log}p+1/(4p)1/2`$. Now, $`G^{}(p)=1/p1/(4p^2)`$, which gives
$$G^{}(p)G^{}(1p)=\left(\frac{qp}{pq}\right)\left(1\frac{1}{4pq}\right).$$
Since $`11/(4pq)<0`$ for all $`p1/2`$, the function $`pG(p)+G(1p)`$ is strictly decreasing on $`(0,1/2)`$ and strictly increasing on $`(1/2,0)`$. Since $`G(1/2)=0`$, we have that $`G(p)+G(1p)>0`$ for all $`p1/2`$. Thus, if $`p1/2`$, then $`F(n/2)\mathrm{}`$ as $`n\mathrm{}`$. It now follows from (6.3) that
$$\underset{n}{sup}\left(\underset{k\{0,\mathrm{},n\}}{sup}\frac{f(n,k,p)}{g(n,k,p)}\right)=\mathrm{}$$
whenever $`p1/2`$.
Now suppose $`p1/2`$ and let $`k[2,np]`$. We can compute that for all $`x(0,n)`$,
$`F^{}(x)`$ $`=\mathrm{log}(nx)+{\displaystyle \frac{1}{2(nx)}}\mathrm{log}x{\displaystyle \frac{1}{2x}}\mathrm{log}{\displaystyle \frac{p}{q}}+{\displaystyle \frac{x}{npq}}{\displaystyle \frac{1}{q}}`$
$`F^{\prime \prime }(x)`$ $`={\displaystyle \frac{1}{nx}}+{\displaystyle \frac{1}{2(nx)^2}}{\displaystyle \frac{1}{x}}+{\displaystyle \frac{1}{2x^2}}+{\displaystyle \frac{1}{npq}}`$
$`F^{\prime \prime \prime }(x)`$ $`={\displaystyle \frac{1}{(nx)^2}}+{\displaystyle \frac{1}{(nx)^3}}+{\displaystyle \frac{1}{x^2}}{\displaystyle \frac{1}{x^3}}`$
$`F^{(4)}(x)`$ $`={\displaystyle \frac{32(nx)}{(nx)^4}}+{\displaystyle \frac{32x}{x^4}}.`$
It is easily verified that $`F(np)=0`$ and $`F^{}(np)=(pq)/2npq`$, so that we may write
$`F(k)`$ $`={\displaystyle _k^{np}}F^{}(t)𝑑t`$
$`={\displaystyle _k^{np}}\left({\displaystyle \frac{pq}{2npq}}{\displaystyle _t^{np}}F^{\prime \prime }(s)𝑑s\right)𝑑t`$
$`{\displaystyle \frac{qp}{2q}}+{\displaystyle _k^{np}}{\displaystyle _k^s}F^{\prime \prime }(s)𝑑t𝑑s.`$
Since $`F^{(4)}0`$ on $`[2,n2]`$ and $`F^{\prime \prime \prime }(n/2)=0`$, it follows that $`F^{\prime \prime \prime }0`$ on $`[2,n/2]`$, which implies $`F^{\prime \prime }`$ is increasing on $`[2,n/2]`$. Since $`F^{\prime \prime }(np)=(p^2+q^2)/2n^2p^2q^2`$, we have
$$F(k)\frac{1}{2}+\frac{p^2+q^2}{2n^2p^2q^2}_k^{np}(sk)𝑑s\frac{1}{2}+\frac{p^2+q^2}{2n^2p^2q^2}n^2p^2\frac{3}{2}$$
for all $`p1/2`$.
It now follows from (6.3) and (i), (ii) that there is a universal constant $`C`$, independent of $`p`$, such that $`f(n,k,p)/g(n,k,p)C`$ for all $`n`$ and all $`k\{0,\mathrm{},np\}`$, provided $`p1/2`$. Also, if $`p=1/2`$, symmetry gives the same bound for $`k\{n/2+1,\mathrm{},n\}`$, and it follows that
$$\underset{n}{sup}\left(\underset{k\{0,\mathrm{},n\}}{sup}\frac{f(n,k,p)}{g(n,k,p)}\right)<\mathrm{},$$
which completes the proof. $`\mathrm{}`$
###### Lemma 6.3
Let $`0<\epsilon <1/2`$ and suppose that $`\{\xi _j\}_{j=1}^{\mathrm{}}`$ are i.i.d. $`\{0,1\}`$-valued random variables with $`P(\xi _1=1)=\epsilon `$. Let $`T_n=_{j=1}^n\xi _j`$. Then for each $`p>1`$, there exists a finite constant $`C_p`$, depending only on $`p`$, such that
$$E[T_n^p1_{\{T_n>0\}}]C_p\frac{1}{(\epsilon n)^p}$$
for all $`n`$.
Proof: Observe that
$`E[T_n^p1_{\{T_n>0\}}]`$ $`=E[T_n^p1_{\{1T_n\epsilon n/2\}}]+E[T_n^p1_{\{T_n>\epsilon n/2\}}]`$
$`P\left(T_n{\displaystyle \frac{\epsilon n}{2}}\right)+\left({\displaystyle \frac{\epsilon n}{2}}\right)^p.`$
Hence, it will suffice to show that
$$P\left(T_n\frac{\epsilon n}{2}\right)C_p\frac{1}{(\epsilon n)^p}.$$
To see this, let $`f`$ and $`g`$ be as in Lemma 6.2 with $`p=\epsilon `$, so that there exists a universal, finite constant $`C`$, independent of $`\epsilon `$, such that $`f(n,k,\epsilon )Cg(n,k,\epsilon )`$ for all $`n`$ and all $`k\{0,\mathrm{},\epsilon n\}`$. Let $`m=\epsilon n/2`$, so that
$$P\left(T_n\frac{\epsilon n}{2}\right)=P(T_nm)=\underset{k=0}{\overset{m}{}}P(T_n=k)C\underset{k=0}{\overset{m}{}}g(n,k,\epsilon ).$$
If $`\epsilon n4`$, then $`P(T_nm)14^p/(\epsilon n)^p`$, so that we may assume without loss of generality that $`\epsilon n>4`$. Note that $`xg(n,x,\epsilon )`$ is increasing on $`[0,\epsilon n]`$ and $`\epsilon n>4`$ implies $`m+1(\epsilon n/2)+1<3\epsilon n/4`$. Thus,
$`P(T_nm)`$ $`C{\displaystyle _0^{m+1}}g(n,x,\epsilon )𝑑x`$
$`C{\displaystyle _{\mathrm{}}^{3\epsilon n/4}}g(n,x,\epsilon )𝑑x`$
$`={\displaystyle \frac{C}{\sqrt{2\pi t}}}{\displaystyle _{\mathrm{}}^{3\epsilon n/4}}e^{(x\epsilon n)^2/2t}𝑑x,`$
where $`t=n\epsilon (1\epsilon )`$. By a change of variables,
$$P(T_nm)C\mathrm{\Phi }\left(\frac{\epsilon n}{4\sqrt{t}}\right)C\mathrm{\Phi }\left(\frac{\sqrt{\epsilon n}}{4}\right).$$
By (3.5),
$$P(T_nm)\frac{C}{\sqrt{2\pi }}\frac{4}{\sqrt{\epsilon n}}e^{\epsilon n/32}C\sqrt{\frac{2}{\pi }}e^{\epsilon n/32}.$$
Since there exists $`K_p<\mathrm{}`$ such that $`x^pe^{x/32}K_p`$ for all $`x[0,\mathrm{})`$, we have
$$P(T_nm)C\sqrt{\frac{2}{\pi }}K_p\frac{1}{(\epsilon n)^p},$$
which finishes the proof. $`\mathrm{}`$
With these lemmas in place, we may now make the needed improvement to Lemma 6.1.
###### Lemma 6.4
If $`0<\epsilon <1/2`$ and $`\mu >0`$, then for all $`p>1`$, there exists $`C_p`$, depending only on $`p`$, such that
$$P(S_n0)C_p\frac{\epsilon ^p}{n^p\mu ^{2p}}$$
(6.4)
for all $`n`$.
Proof: Let $`\{\stackrel{~}{Y}_j\}_{j=1}^{\mathrm{}}`$ be a sequence of i.i.d. $`\{1,1\}`$-valued random variables with $`P(\stackrel{~}{Y}_1=1)=p_1/\epsilon `$. Let $`\{\xi _j\}_{j=1}^{\mathrm{}}`$ be a sequence of i.i.d. $`\{0,1\}`$-valued random variables, independent of $`\{\stackrel{~}{Y}_j\}_{j=1}^{\mathrm{}}`$, with $`P(\xi _1=1)=\epsilon `$. Then $`\{\stackrel{~}{Y}_j\xi _j\}_{j=1}^{\mathrm{}}`$ is an i.i.d. sequence of random variables which has the same law as $`\{Y_j\}_{j=1}^{\mathrm{}}`$.
Let $`\stackrel{~}{S}_n=_{j=1}^n\stackrel{~}{Y}_j`$ and note that by Lemma 6.1,
$$P(\stackrel{~}{S}_n0)\stackrel{~}{C}_p\frac{1}{n^p(\mu /\epsilon )^{2p}}=\stackrel{~}{C}_p\frac{\epsilon ^{2p}}{n^p\mu ^{2p}}.$$
(6.5)
Define $`\xi ^{(n)}=(\xi _1,\mathrm{},\xi _n)`$, so that
$`P(S_n0)`$ $`=P\left({\displaystyle \underset{j=1}{\overset{n}{}}}\stackrel{~}{Y}_j\xi _j0\right)`$
$`={\displaystyle \underset{k=0}{\overset{n}{}}}{\displaystyle \underset{\begin{array}{c}\alpha \{0,1\}^n\\ |\alpha |=k\end{array}}{}}P\left({\displaystyle \underset{j=1}{\overset{n}{}}}\stackrel{~}{Y}_j\xi _j0,\xi ^{(n)}=\alpha \right)`$
$`={\displaystyle \underset{k=0}{\overset{n}{}}}{\displaystyle \underset{\begin{array}{c}\alpha \{0,1\}^n\\ |\alpha |=k\end{array}}{}}P\left({\displaystyle \underset{\{j:\alpha _j=1\}}{}}\stackrel{~}{Y}_j0,\xi ^{(n)}=\alpha \right),`$
where $`|\alpha |=\alpha _1+\mathrm{}+\alpha _n`$. If $`T_n=_{j=1}^n\xi _j`$, then by symmetry and independence,
$$P(S_n0)=\underset{k=0}{\overset{n}{}}\underset{\begin{array}{c}\alpha \{0,1\}^n\\ |\alpha |=k\end{array}}{}P\left(\underset{j=1}{\overset{k}{}}\stackrel{~}{Y}_j0\right)P(\xi ^{(n)}=\alpha )=\underset{k=0}{\overset{n}{}}P(\stackrel{~}{S}_k0)P(T_n=k).$$
Using (6.5) and Lemma 6.3,
$`P(S_n0)`$ $`P(T_n=0)+\stackrel{~}{C}_p{\displaystyle \frac{\epsilon ^{2p}}{\mu ^{2p}}}{\displaystyle \underset{k=1}{\overset{n}{}}}k^pP(T_n=k)`$
$`=(1\epsilon )^n+\stackrel{~}{C}_p{\displaystyle \frac{\epsilon ^{2p}}{\mu ^{2p}}}E[T_n^p1_{\{T_n>0\}}]`$
$`(1\epsilon )^n+\stackrel{~}{C}_p^{}{\displaystyle \frac{\epsilon ^{2p}}{\mu ^{2p}}}{\displaystyle \frac{1}{(\epsilon n)^p}},`$
Note that $`1\epsilon e^\epsilon `$, so that
$$(1\epsilon )^ne^{\epsilon n}\stackrel{~}{C}_p^{\prime \prime }\frac{1}{(\epsilon n)^p}=\stackrel{~}{C}_p^{\prime \prime }\frac{\epsilon ^p}{n^p\epsilon ^{2p}}\stackrel{~}{C}_p^{\prime \prime }\frac{\epsilon ^p}{n^p\mu ^{2p}},$$
which gives (6.4) with $`C_p=\stackrel{~}{C}_p^{\prime \prime }+\stackrel{~}{C}_p^{}`$. $`\mathrm{}`$
## 7 The Small Jump Regime
Let us now put the pieces together and establish (3.2) for the small jump regime. Recall from Section 5 that it will suffice to establish (5.13). Using the notation of (5.1)-(5.7), Lemma 6.4 will give us that, for $`p>1`$,
$$\phi _{k1}(x,y,\delta )C_p\frac{\stackrel{~}{\epsilon }^p}{(k1)^p\stackrel{~}{\mu }^{2p}},$$
(7.1)
provided $`\stackrel{~}{\epsilon }=\stackrel{~}{\epsilon }(x,y,\delta )<1/2`$ and $`\stackrel{~}{\mu }=\stackrel{~}{\mu }(x,y,\delta )>0`$. We will be applying this with $`x=\epsilon /(\delta ^{1/4}\sqrt{n})`$ and $`y=\epsilon /\sqrt{n}`$, but recall that in the small jump regime, we can write $`\epsilon /\sqrt{n}=\delta ^{1/2+\alpha }`$ for some $`\alpha >0`$. As such, the following lemma will help us check the provisions of (7.1).
###### Lemma 7.1
For each $`\mathrm{\Delta }>0`$, there exists $`\delta _0>0`$ such that
(i) $`\stackrel{~}{\mu }(\delta ^{1/4+\alpha },\delta ^{1/2+\alpha },\delta )\frac{1}{\sqrt{2\pi }}\delta ^{1/2+\alpha }`$, and
(ii) $`\stackrel{~}{\epsilon }(\delta ^{1/4+\alpha },\delta ^{1/2+\alpha },\delta )1000\delta ^{1/2}<\frac{1}{2}`$
for all $`\alpha \mathrm{\Delta }`$ and all $`0<\delta \delta _0`$.
Proof: For fixed $`\delta >0`$, let $`\psi (x,y)=\psi (x,y,\delta )`$ be given by (5.8). We wish to show that
$$\psi (x,y)=\frac{1}{2}\frac{1}{2\pi }\mathrm{tan}^1\sqrt{\delta }+\frac{x}{\sqrt{2\pi }}+\frac{y}{2\sqrt{2\pi }}+\frac{\sqrt{\delta }}{4\pi }(x+y)^2\frac{y^2}{4\pi \sqrt{\delta }}+\stackrel{~}{R}(x,y),$$
(7.2)
where
$$|\stackrel{~}{R}(x,y)|(|x|+|y|)^3+\frac{|x||y|^2}{\sqrt{\delta }}(|x|+|y|)+\frac{|y|^4}{\delta ^{3/2}}+\delta ^{3/2}(x+y)^2+\delta (|x|+|y|)$$
(7.3)
for all $`x,y`$.
We will first show that for $`i0`$ and $`j1`$,
$`_x^i\psi `$ $`={\displaystyle _{\mathrm{}}^x}\mathrm{\Phi }\left({\displaystyle \frac{x+yt}{\sqrt{\delta }}}\right)\mathrm{\Phi }^{(i+1)}(t)𝑑t,`$ (7.4)
$`_x^i_y^j\psi `$ $`=\left({\displaystyle \frac{1}{\sqrt{\delta }}}\right)^{j1}\mathrm{\Phi }^{(j1)}\left({\displaystyle \frac{y}{\sqrt{\delta }}}\right)\mathrm{\Phi }^{(i+1)}(x)+_x^{i+1}_y^{j1}\psi .`$ (7.5)
For $`i=0`$, (7.4) is just the definition of $`\psi `$. If (7.4) is true for some $`i0`$, then using integration by parts gives
$`_x^{i+1}\psi `$ $`=_x\left[{\displaystyle _{\mathrm{}}^x}\mathrm{\Phi }\left({\displaystyle \frac{x+yt}{\sqrt{\delta }}}\right)\mathrm{\Phi }^{(i+1)}(t)𝑑t\right]`$
$`=\mathrm{\Phi }\left({\displaystyle \frac{y}{\sqrt{\delta }}}\right)\mathrm{\Phi }^{(i+1)}(x)+{\displaystyle \frac{1}{\sqrt{\delta }}}{\displaystyle _{\mathrm{}}^x}\mathrm{\Phi }^{}\left({\displaystyle \frac{x+yt}{\sqrt{\delta }}}\right)\mathrm{\Phi }^{(i+1)}(t)𝑑t`$
$`={\displaystyle _{\mathrm{}}^x}\mathrm{\Phi }\left({\displaystyle \frac{x+yt}{\sqrt{\delta }}}\right)\mathrm{\Phi }^{(i+2)}(t)𝑑t,`$
so by induction, (7.4) holds for all $`i0`$. For (7.5), first consider $`j=1`$. Then
$`_x^i_y\psi `$ $`=_y\left[{\displaystyle _{\mathrm{}}^x}\mathrm{\Phi }\left({\displaystyle \frac{x+yt}{\sqrt{\delta }}}\right)\mathrm{\Phi }^{(i+1)}(t)𝑑t\right]`$
$`={\displaystyle _{\mathrm{}}^x}_y\left[\mathrm{\Phi }\left({\displaystyle \frac{x+yt}{\sqrt{\delta }}}\right)\right]\mathrm{\Phi }^{(i+1)}(t)dt`$
$`={\displaystyle _{\mathrm{}}^x}_x\left[\mathrm{\Phi }\left({\displaystyle \frac{x+yt}{\sqrt{\delta }}}\right)\right]\mathrm{\Phi }^{(i+1)}(t)dt`$
$`=_x\left[{\displaystyle _{\mathrm{}}^x}\mathrm{\Phi }\left({\displaystyle \frac{x+yt}{\sqrt{\delta }}}\right)\mathrm{\Phi }^{(i+1)}(t)𝑑t\right]\mathrm{\Phi }\left({\displaystyle \frac{y}{\sqrt{\delta }}}\right)\mathrm{\Phi }^{(i+1)}(x)`$
$`=\mathrm{\Phi }\left({\displaystyle \frac{y}{\sqrt{\delta }}}\right)\mathrm{\Phi }^{(i+1)}(x)+_x^{i+1}\psi ,`$
and (7.5) holds for all $`i0`$ when $`j=1`$. Now suppose (7.5) holds for some $`j1`$ and all $`i0`$. Then
$`_x^i_y^{j+1}\psi `$ $`=_y\left[\left({\displaystyle \frac{1}{\sqrt{\delta }}}\right)^{j1}\mathrm{\Phi }^{(j1)}\left({\displaystyle \frac{y}{\sqrt{\delta }}}\right)\mathrm{\Phi }^{(i+1)}(x)+_x^{i+1}_y^{j1}\psi \right]`$
$`=\left({\displaystyle \frac{1}{\sqrt{\delta }}}\right)^j\mathrm{\Phi }^{(j)}\left({\displaystyle \frac{y}{\sqrt{\delta }}}\right)\mathrm{\Phi }^{(i+1)}(x)+_x^{i+1}_y^j\psi .`$
By induction, (7.5) holds for all $`i0`$ and $`j1`$.
By Taylor’s Theorem we have that
$$\begin{array}{cc}\hfill \psi (x,y)& =\psi (0,0)+x\psi _x(0,0)+y\psi _y(0,0)\hfill \\ & +\frac{1}{2!}[x^2\psi _{xx}(0,0)+2xy\psi _{xy}(0,0)+y^2\psi _{yy}(0,0)]+R^{(1)}(x,y),\hfill \end{array}$$
(7.6)
where
$$R^{(1)}(x,y)=\frac{1}{3!}[x^3\psi _{xxx}(\overline{x},\overline{y})+3x^2y\psi _{xxy}(\overline{x},\overline{y})+3xy^2\psi _{xyy}(\overline{x},\overline{y})+y^3\psi _{yyy}(\overline{x},\overline{y})]$$
and $`(\overline{x},\overline{y})=(\theta x,\theta y)`$ for some $`\theta (0,1)`$. Using (7.4), (7.5), and direct integration, we can verify that (7.6) becomes
$$\begin{array}{cc}\hfill \psi (x,y)& =\frac{1}{2}\frac{1}{2\pi }\mathrm{tan}^1\sqrt{\delta }+\frac{x}{2\sqrt{2\pi }}\left(1+\frac{1}{\sqrt{1+\delta }}\right)+\frac{y}{2\sqrt{2\pi }\sqrt{1+\delta }}\hfill \\ & +\frac{(x+y)^2\sqrt{\delta }}{4\pi (1+\delta )}\frac{y^2}{4\pi \sqrt{\delta }}+R^{(1)}(x,y).\hfill \end{array}$$
Now,
$`{\displaystyle \frac{x}{2\sqrt{2\pi }}}\left(1+{\displaystyle \frac{1}{\sqrt{1+\delta }}}\right)`$ $`={\displaystyle \frac{x}{\sqrt{2\pi }}}+{\displaystyle \frac{x}{2\sqrt{2\pi }}}\left({\displaystyle \frac{1}{\sqrt{1+\delta }}}1\right)`$
$`{\displaystyle \frac{y}{2\sqrt{2\pi }\sqrt{1+\delta }}}`$ $`={\displaystyle \frac{y}{2\sqrt{2\pi }}}+{\displaystyle \frac{y}{2\sqrt{2\pi }}}\left({\displaystyle \frac{1}{\sqrt{1+\delta }}}1\right)`$
$`{\displaystyle \frac{(x+y)^2\sqrt{\delta }}{4\pi (1+\delta )}}`$ $`={\displaystyle \frac{\sqrt{\delta }}{4\pi }}(x+y)^2+{\displaystyle \frac{\sqrt{\delta }}{4\pi }}(x+y)^2\left({\displaystyle \frac{1}{1+\delta }}1\right).`$
Thus, if
$$R^{(2)}(x,y)=\frac{x+y}{2\sqrt{2\pi }}\left(\frac{1}{\sqrt{1+\delta }}1\right)\frac{\delta ^{3/2}(x+y)^2}{4\pi (1+\delta )},$$
then (7.2) holds with $`\stackrel{~}{R}=R^{(1)}+R^{(2)}`$.
Since $`|(1+\delta )^{1/2}1|<\delta `$, we have $`|R^{(2)}(x,y)|\delta (|x|+|y|)+\delta ^{3/2}(x+y)^2`$. To estimate $`R^{(1)}`$, we must estimate the third partial derivatives of $`\psi `$. Using (7.4), we have
$$|\psi _{xxx}(x,y)|=\left|_{\mathrm{}}^x\mathrm{\Phi }\left(\frac{x+yt}{\sqrt{\delta }}\right)\mathrm{\Phi }^{(4)}(t)𝑑t\right|_{\mathrm{}}^{\mathrm{}}|\mathrm{\Phi }^{(4)}(t)|𝑑t.$$
Since $`\mathrm{\Phi }^{(4)}(t)=(3tt^3)\mathrm{\Phi }^{}(t)`$, we have
$$|\psi _{xxx}(x,y)|2_0^{\mathrm{}}(3t+t^3)\mathrm{\Phi }^{}(t)𝑑t=\frac{10}{\sqrt{2\pi }}.$$
Similarly, by (7.5),
$$|\psi _{xxy}(x,y)|=\left|\mathrm{\Phi }\left(\frac{y}{\sqrt{\delta }}\right)\mathrm{\Phi }^{\prime \prime \prime }(x)+\psi _{xxx}(x,y)\right||\mathrm{\Phi }^{\prime \prime \prime }(x)|+\frac{10}{\sqrt{2\pi }}.$$
Since $`|\mathrm{\Phi }^{\prime \prime \prime }(x)|2(2\pi )^{1/2}`$ for all $`x`$, we have that
$$|\psi _{xxy}(x,y)|\frac{12}{\sqrt{2\pi }}.$$
Likewise, the formulas
$`\psi _{xyy}`$ $`={\displaystyle \frac{1}{\sqrt{\delta }}}\mathrm{\Phi }^{}\left({\displaystyle \frac{y}{\sqrt{\delta }}}\right)\mathrm{\Phi }^{\prime \prime }(x)+\psi _{xxy}`$
$`={\displaystyle \frac{x}{\sqrt{\delta }}}\mathrm{\Phi }^{}\left({\displaystyle \frac{y}{\sqrt{\delta }}}\right)\mathrm{\Phi }^{}(x)+\psi _{xxy}`$
and
$`\psi _{yyy}`$ $`={\displaystyle \frac{1}{\delta }}\mathrm{\Phi }^{\prime \prime }\left({\displaystyle \frac{y}{\sqrt{\delta }}}\right)\mathrm{\Phi }^{}(x)+\psi _{xyy}`$
$`={\displaystyle \frac{y}{\delta ^{3/2}}}\mathrm{\Phi }^{}\left({\displaystyle \frac{y}{\sqrt{\delta }}}\right)\mathrm{\Phi }^{}(x)+\psi _{xyy}`$
can be used to verify that
$`|\psi _{xyy}(x,y)|`$ $`(|x|\delta ^{1/2}+12\sqrt{2\pi })/(2\pi )`$
$`|\psi _{yyy}(x,y)|`$ $`(|y|\delta ^{3/2}+|x|\delta ^{1/2}+12\sqrt{2\pi })/(2\pi ).`$
Piecing this together, we have
$`|R^{(1)}(x,y)|`$ $`{\displaystyle \frac{1}{3!}}[{\displaystyle \frac{10|x|^3}{\sqrt{2\pi }}}+{\displaystyle \frac{36|x|^2|y|}{\sqrt{2\pi }}}+3|x||y|^2({\displaystyle \frac{|x|}{2\pi \sqrt{\delta }}}+{\displaystyle \frac{12}{\sqrt{2\pi }}})`$
$`+|y|^3({\displaystyle \frac{|y|}{2\pi \delta ^{3/2}}}+{\displaystyle \frac{|x|}{2\pi \sqrt{\delta }}}+{\displaystyle \frac{12}{\sqrt{2\pi }}})]`$
$`{\displaystyle \frac{1}{3!}}\left[{\displaystyle \frac{12}{\sqrt{2\pi }}}(|x|+|y|)^3+{\displaystyle \frac{3|x||y|^2}{2\pi \sqrt{\delta }}}(|x|+|y|)+{\displaystyle \frac{|y|^4}{2\pi \delta ^{3/2}}}\right]`$
$`(|x|+|y|)^3+{\displaystyle \frac{|x||y|^2}{\sqrt{\delta }}}(|x|+|y|)+{\displaystyle \frac{|y|^4}{\delta ^{3/2}}}.`$
Combined with the estimate for $`R^{(2)}`$, this verifies (7.3).
Now, observe that $`p_1(x,y,\delta )=\psi (x,y)/\mathrm{\Phi }(x)`$. Write $`\mathrm{\Phi }(x)=\frac{1}{2}+\frac{x}{\sqrt{2\pi }}+r_1(x)`$, where $`r_1(x)=\frac{1}{2}x^2\mathrm{\Phi }^{\prime \prime }(\overline{x})`$ and $`\overline{x}=\theta x`$ for some $`\theta (0,1)`$. Note that $`|r_1(x)|\frac{1}{2\sqrt{2\pi }}|x|^3`$. For $`x\sqrt{\pi /2}`$, write $`\mathrm{\Phi }(x)^1=(\frac{1}{2}+\frac{x}{\sqrt{2\pi }})^1+r_2(x)`$, where $`r_2(x)=r_1(x)\mathrm{\Phi }(x)^1(\frac{1}{2}+\frac{x}{\sqrt{2\pi }})^1`$. Similarly, we may write $`\mathrm{\Phi }(x)^1=2+r_3(x)`$, where
$`r_3(x)`$ $`=r_2(x)+{\displaystyle \frac{1}{\frac{1}{2}+\frac{x}{\sqrt{2\pi }}}}2`$
$`=r_2(x){\displaystyle \frac{4x}{\sqrt{2\pi }+2x}}.`$
Let us now assume $`|x|1`$. Then $`x\sqrt{\pi /2}`$ and the above applies. Note that
$$|r_2(x)|\frac{|r_1(x)|}{\mathrm{\Phi }(1)\left(\frac{1}{2}\frac{1}{\sqrt{2\pi }}\right)}$$
Since $`\mathrm{\Phi }(1)\frac{1}{2}\frac{1}{\sqrt{2\pi }}\frac{1}{10}`$, we have $`|r_2(x)|100|r_1(x)|\frac{50}{\sqrt{2\pi }}|x|^3`$. Also,
$$|r_3(x)||r_2(x)|+\left(\frac{4}{\sqrt{2\pi }2}\right)|x|\frac{50}{\sqrt{2\pi }}|x|^3+\frac{20}{\sqrt{2\pi }}|x|.$$
Since $`|x|1`$, this gives $`|r_3(x)|\frac{70}{\sqrt{2\pi }}|x|`$. Applying (7.2) yields
$`p_1(x,y,\delta )`$ $`=\psi (x,y)\mathrm{\Phi }(x)^1`$
$`=\left({\displaystyle \frac{1}{2}}+{\displaystyle \frac{x}{\sqrt{2\pi }}}\right)\left(\left({\displaystyle \frac{1}{2}}+{\displaystyle \frac{x}{\sqrt{2\pi }}}\right)^1+r_2(x)\right)`$
$`+\left({\displaystyle \frac{1}{2\pi }}\mathrm{tan}^1\sqrt{\delta }+{\displaystyle \frac{y}{2\sqrt{2\pi }}}+{\displaystyle \frac{\sqrt{\delta }}{4\pi }}(x+y)^2{\displaystyle \frac{y^2}{4\pi \sqrt{\delta }}}\right)(2+r_3(x))`$
$`+\stackrel{~}{R}(x,y)\mathrm{\Phi }(x)^1`$
$`=1{\displaystyle \frac{1}{\pi }}\mathrm{tan}^1\sqrt{\delta }+{\displaystyle \frac{y}{\sqrt{2\pi }}}+{\displaystyle \frac{\sqrt{\delta }}{2\pi }}(x+y)^2{\displaystyle \frac{y^2}{2\pi \sqrt{\delta }}}+R_\delta (x,y),`$ (7.7)
where
$`|R_\delta (x,y)|`$ $`|r_2(x)|+\left({\displaystyle \frac{\mathrm{tan}^1\sqrt{\delta }}{2\pi }}+{\displaystyle \frac{|y|}{2\sqrt{2\pi }}}+{\displaystyle \frac{\sqrt{\delta }}{4\pi }}(x+y)^2+{\displaystyle \frac{y^2}{4\pi \sqrt{\delta }}}\right)|r_3(x)|+{\displaystyle \frac{|\stackrel{~}{R}(x,y)|}{\mathrm{\Phi }(1)}}`$
$`{\displaystyle \frac{50}{\sqrt{2\pi }}}|x|^3+\left({\displaystyle \frac{\sqrt{\delta }}{2\pi }}+{\displaystyle \frac{|y|}{2\sqrt{2\pi }}}+{\displaystyle \frac{\sqrt{\delta }}{4\pi }}(x+y)^2+{\displaystyle \frac{y^2}{4\pi \sqrt{\delta }}}\right){\displaystyle \frac{70}{\sqrt{2\pi }}}|x|+10|\stackrel{~}{R}(x,y)|.`$
Hence,
$$\begin{array}{cc}\hfill |R_\delta (x,y)|& \frac{50}{\sqrt{2\pi }}|x|^3+\left(\frac{\sqrt{\delta }}{2\pi }+\frac{|y|}{2\sqrt{2\pi }}+\frac{\sqrt{\delta }}{4\pi }(x+y)^2+\frac{y^2}{4\pi \sqrt{\delta }}\right)\frac{70}{\sqrt{2\pi }}|x|\hfill \\ & +10\left[(|x|+|y|)^3+\frac{|x||y|^2}{\sqrt{\delta }}(|x|+|y|)+\frac{|y|^4}{\delta ^{3/2}}+\delta ^{3/2}(x+y)^2+\delta (|x|+|y|)\right]\hfill \end{array}$$
by (7.3).
Now suppose that $`\delta 1`$ and $`\alpha ,\beta `$. Let $`y=\delta ^{1/2+\alpha }`$, $`x=\delta ^{1/4+\beta }`$, and assume that $`yx1`$. Using the fact that $`|x|+|y|2|x|2`$, we have
$`|R_\delta (x,y)|`$ $`{\displaystyle \frac{50}{\sqrt{2\pi }}}|x|^3+{\displaystyle \frac{70}{(2\pi )^{3/2}}}|x|\left(\sqrt{\delta }+2|y|+2\sqrt{\delta }+{\displaystyle \frac{y^2}{\sqrt{\delta }}}\right)`$
$`+10\left(8|x|^3+2{\displaystyle \frac{|x||y|^2}{\sqrt{\delta }}}+{\displaystyle \frac{|y|^4}{\delta ^{3/2}}}+4\delta ^{3/2}x^2+2\delta |x|\right)`$
$`25|x|^3+5|x|\left(3\sqrt{\delta }+2|y|+{\displaystyle \frac{y^2}{\sqrt{\delta }}}\right)`$
$`+80|x|^3+20{\displaystyle \frac{|x||y|^2}{\sqrt{\delta }}}+10{\displaystyle \frac{|y|^4}{\delta ^{3/2}}}+40\delta ^{3/2}x^2+20\delta |x|`$
which reduces to
$$\begin{array}{cc}\hfill |R_\delta (x,y)|& 105\delta ^{3/4+3\beta }+15\delta ^{3/4+\beta }+10\delta ^{3/4+\alpha +\beta }+25\delta ^{3/4+2\alpha +\beta }\hfill \\ & +10\delta ^{1/2+4\alpha }+40\delta ^{2+2\beta }+20\delta ^{5/4+\beta }.\hfill \end{array}$$
To simplify further, suppose $`\alpha >0`$. Then
$`|R_\delta (x,y)|`$ $`105\delta ^{3/4+3\beta }+15\delta ^{3/4+\beta }+10\delta ^{3/4+\beta }+25\delta ^{3/4+\beta }`$
$`+10\delta ^{1/2+4\alpha }+40\delta ^{2+2\beta }+20\delta ^{3/4+\beta }`$
$`=105\delta ^{3/4+3\beta }+70\delta ^{3/4+\beta }+10\delta ^{1/2+4\alpha }+40\delta ^{2+2\beta }.`$
Now, if $`\beta 0`$, then $`2+2\beta >3/4+\beta `$, and $`|R_\delta (x,y)|115\delta ^{3/4+3\beta }+110\delta ^{3/4+\beta }+10\delta ^{1/2+4\alpha }`$. Otherwise, if $`\beta <0`$, then $`2+2\beta >3/4+3\beta `$, and $`|R_\delta (x,y)|145\delta ^{3/4+3\beta }+70\delta ^{3/4+\beta }+10\delta ^{1/2+4\alpha }`$. In either case,
$$|R_\delta (x,y)|150(\delta ^{3/4+3\beta }+\delta ^{3/4+\beta }+\delta ^{1/2+4\alpha })$$
whenever $`\alpha >0`$. On the other hand, suppose $`\alpha <0`$. Then
$`|R_\delta (x,y)|`$ $`105\delta ^{3/4+3\beta }+15\delta ^{3/4+2\alpha +\beta }+10\delta ^{3/4+2\alpha +\beta }+25\delta ^{3/4+2\alpha +\beta }`$
$`+10\delta ^{1/2+4\alpha }+40\delta ^{2+2\beta }+20\delta ^{3/4+2\alpha +\beta }`$
$`=105\delta ^{3/4+3\beta }+70\delta ^{3/4+2\alpha +\beta }+10\delta ^{1/2+4\alpha }+40\delta ^{2+2\beta }.`$
If $`\beta 0`$, then $`2+2\beta >3/4+\beta 3/4+2\alpha +\beta `$; if $`\beta <0`$, then $`2+2\beta >3/4+3\beta `$. We therefore have
$$|R_\delta (x,y)|150(\delta ^{3/4+3\beta }+\delta ^{3/4+2\alpha +\beta }+\delta ^{1/2+4\alpha })$$
whenever $`\alpha <0`$.
In summary, we have an expansion for $`p_1(x,y,\delta )`$ given by (7.7), together with a remainder estimate of the form
$$|R_\delta (x,y)|150(\delta ^{3/4+3\beta }+\delta ^{3/4+2(\alpha 0)+\beta }+\delta ^{1/2+4\alpha }),$$
(7.8)
valid for $`0<\delta 1`$ whenever $`y=\delta ^{1/2+\alpha }`$ and $`x=\delta ^{1/4+\beta }`$ satisfy $`yx1`$. Moreover, by symmetry, the same bound holds for $`|R_\delta (x,y)|`$.
Now fix $`\mathrm{\Delta }>0`$. Choose $`\delta _01`$ such that
$$900(\delta _0^{1/4}\delta _0^{3\mathrm{\Delta }})<(2\pi )^{1/2}.$$
(7.9)
Let $`\alpha \mathrm{\Delta }`$ and $`0<\delta \delta _0`$. Set $`\beta =\alpha `$, $`y=\delta ^{1/2+\alpha }`$, and $`x=\delta ^{1/4+\beta }`$. Note that by (5.1)-(5.7)
$$\stackrel{~}{\mu }(x,y,\delta )=p_1(x,y,\delta )p_1(x,y,\delta ),$$
so by (7.7)
$$\stackrel{~}{\mu }=\frac{2y}{\sqrt{2\pi }}+R_\delta (x,y)R_\delta (x,y).$$
Since $`\delta 1`$, we have $`yx1`$. Hence, by (7.8) and (7.9),
$`|R_\delta (x,y)R_\delta (x,y)|`$ $`300(2\delta ^{3/4+\alpha }+\delta ^{1/2+4\alpha })`$
$`=300(2\delta ^{1/4}+\delta ^{3\alpha })y`$
$`300(2\delta _0^{1/4}+\delta _0^{3\mathrm{\Delta }})y`$
$`900(\delta _0^{1/4}\delta _0^{3\mathrm{\Delta }})y<(2\pi )^{1/2}y.`$
Therefore, $`\stackrel{~}{\mu }(2\pi )^{1/2}y`$, which proves (i).
For (ii), observe that $`\stackrel{~}{\mu }>0`$ implies $`q_1<q_2`$. Hence $`\stackrel{~}{\epsilon }=p_1q_2+p_2q_12q_2`$. Moreover,
$$\begin{array}{cc}\hfill q_2& =1p_1(x,y,\delta )\hfill \\ & \frac{1}{\pi }\mathrm{tan}^1\sqrt{\delta }+\frac{|y|}{\sqrt{2\pi }}+\frac{\sqrt{\delta }}{2\pi }(x+y)^2+\frac{y^2}{2\pi \sqrt{\delta }}+|R_\delta (x,y)|\hfill \\ & \delta ^{1/2}+\delta ^{1/2+\alpha }+\delta ^{1+2\alpha }+\delta ^{1/2+2\alpha }+150(\delta ^{3/4+3\alpha }+\delta ^{3/4+\alpha }+\delta ^{1/2+4\alpha })\hfill \\ & 500\delta ^{1/2},\hfill \end{array}$$
(7.10)
so $`\stackrel{~}{\epsilon }1000\delta ^{1/2}`$. By making $`\delta _0`$ smaller if necessary we can ensure that $`1000\delta ^{1/2}<1/2`$. $`\mathrm{}`$
###### Lemma 7.2
Let $`p>2`$. Fix $`0<\mathrm{\Delta }<1/2`$ and let $`\delta _0`$ be as in Lemma 7.1. Suppose $`\epsilon >0`$, $`0<\delta \delta _0`$, and $`n3`$ satisfy $`\epsilon /\sqrt{n}\delta ^{1/2+\mathrm{\Delta }}`$. Then
$$P\left(M_n(1+\delta )M_n(1)>\frac{\epsilon }{\sqrt{n}}\right)C(\epsilon ^1\delta ^{1/4})^p,$$
where $`C`$ depends only on $`p`$ and $`\mathrm{\Delta }`$.
Proof: Let $`y=\epsilon /\sqrt{n}`$ and choose $`\alpha \mathrm{\Delta }`$ such that $`y=\delta ^{1/2+\alpha }`$. Set $`x_0=\delta ^{1/4+\alpha }`$. By Corollary 5.2, Lemma 5.3, Lemma 6.4, and Lemma 7.1,
$$P\left(M_n(1+\delta )M_n(1)>\frac{\epsilon }{\sqrt{n}}\right)C_{p/2}\frac{\stackrel{~}{\epsilon }^{p/2}}{(k1)^{p/2}\stackrel{~}{\mu }^p}+C_p(\epsilon ^1\delta ^{1/4})^p,$$
where $`\stackrel{~}{\epsilon }=\stackrel{~}{\epsilon }(x_0,y,\delta )1000\delta ^{1/2}<1/2`$ and
$$\stackrel{~}{\mu }=\stackrel{~}{\mu }(x_0,y,\delta )\frac{1}{\sqrt{2\pi }}\delta ^{1/2+\alpha }=\frac{1}{\sqrt{2\pi }}\frac{\epsilon }{\sqrt{n}}>0.$$
Hence,
$$\frac{\stackrel{~}{\epsilon }^{p/2}}{(k1)^{p/2}\stackrel{~}{\mu }^p}C\frac{\delta ^{p/4}}{n^{p/2}(\epsilon /\sqrt{n})^p}=C(\epsilon ^1\delta ^{1/4})^p,$$
which completes the proof. $`\mathrm{}`$
## 8 The Medium Jump Regime and Final Proof
Our analysis of the medium jump regime will require only minor modifications to the methods of Section 7.
###### Lemma 8.1
Fix $`0<\mathrm{\Delta }<1/16`$ and set $`\mathrm{\Delta }^{}=(116\mathrm{\Delta })/12>0`$. Then there exists $`\delta _0>0`$ such that
(i) $`\stackrel{~}{\mu }(\delta ^{1/4+\alpha },\delta ^{1/2+\alpha },\delta )\frac{1}{\sqrt{2\pi }}\delta ^{1/2+\mathrm{\Delta }}`$, and
(ii) $`\stackrel{~}{\epsilon }(\delta ^{1/4+\alpha },\delta ^{1/2+\alpha },\delta )1000\delta ^{1/24\mathrm{\Delta }^{}}<\frac{1}{2}`$
for all $`\mathrm{\Delta }^{}\alpha \mathrm{\Delta }`$ and all $`0<\delta \delta _0`$.
Proof: For fixed $`0<\mathrm{\Delta }<1/16`$, choose $`\delta _0>0`$ as in Lemma 7.1. By (5.1), $`p_1`$ is increasing in $`y`$. Hence, if $`x=\delta ^{1/4+\alpha }`$ and $`y=\delta ^{1/2+\mathrm{\Delta }}`$, then by (7.7) and (7.8),
$`\stackrel{~}{\mu }(x,\delta ^{1/2+\alpha },\delta )`$ $`=p_1(x,\delta ^{1/2+\alpha },\delta )p_1(x,\delta ^{1/2+\alpha },\delta )`$
$`p_1(x,y,\delta )p_1(x,y,\delta )`$
$`={\displaystyle \frac{2y}{\sqrt{2\pi }}}+R_\delta (x,y)R_\delta (x,y),`$
where
$`|R_\delta (x,y)R_\delta (x,y)|`$ $`300(\delta ^{3/4+3\alpha }+\delta ^{3/4+\alpha }+\delta ^{1/2+4\mathrm{\Delta }})`$
$`300(2\delta ^{3/43\mathrm{\Delta }^{}}+\delta ^{1/2+4\mathrm{\Delta }}).`$
However, note that $`3/43\mathrm{\Delta }^{}=1/2+4\mathrm{\Delta }`$. Hence, by (7.9),
$$|R_\delta (x,y)R_\delta (x,y)|900\delta ^{1/2+4\mathrm{\Delta }}=900\delta ^{3\mathrm{\Delta }}y<(2\pi )^{1/2}y.$$
Therefore, $`\stackrel{~}{\mu }(2\pi )^{1/2}y`$, which proves (i).
For (ii), observe that $`\stackrel{~}{\epsilon }2q_2`$ and, as in (7.10),
$`q_2`$ $`=1p_1(x,\delta ^{1/2+\alpha },\delta )`$
$`\delta ^{1/2}+\delta ^{1/2+\alpha }+\delta ^{1+2\alpha }+\delta ^{1/2+2\alpha }+150(\delta ^{3/4+3\alpha }+\delta ^{3/4+\alpha }+\delta ^{1/2+4\alpha })`$
$`4\delta ^{1/22\mathrm{\Delta }^{}}+150(2\delta ^{3/43\mathrm{\Delta }^{}}+\delta ^{1/24\mathrm{\Delta }^{}})`$
$`500\delta ^{1/24\mathrm{\Delta }^{}}.`$
Note that $`1/24\mathrm{\Delta }^{}>1/6`$, so that by making $`\delta _0`$ smaller if necessary, we can ensure that $`1000\delta ^{1/24\mathrm{\Delta }^{}}<1/2`$. $`\mathrm{}`$
###### Lemma 8.2
Fix $`p>2`$. Let $`\mathrm{\Delta }=1/18`$ and choose $`\delta _0>0`$ as in Lemma 8.1. Suppose $`\epsilon >0`$, $`0<\delta \delta _0`$, and $`n3`$ satisfy $`\delta ^{5/9}\epsilon /\sqrt{n}\delta ^{53/108}`$. Then
$$P\left(M_n(1+\delta )M_n(1)>\frac{\epsilon }{\sqrt{n}}\right)C(\epsilon ^1\delta ^{1/6})^p,$$
where $`C`$ depends only on $`p`$.
Proof: Let $`\mathrm{\Delta }=1/18`$ and $`\mathrm{\Delta }^{}=(116\mathrm{\Delta })/12=1/108`$ and observe that $`y=\epsilon /\sqrt{n}=\delta ^{1/2+\alpha }`$ for some $`\alpha [\mathrm{\Delta }^{},\mathrm{\Delta }]`$. Set $`x_0=\delta ^{1/4+\alpha }`$. By Corollary 5.2, Lemma 5.3, Lemma 6.4, and Lemma 8.1,
$$P\left(M_n(1+\delta )M_n(1)>\frac{\epsilon }{\sqrt{n}}\right)C_{p/2}\frac{\stackrel{~}{\epsilon }^{p/2}}{(k1)^{p/2}\stackrel{~}{\mu }^p}+C_p(\epsilon ^1\delta ^{1/4})^p,$$
where $`\stackrel{~}{\epsilon }=\stackrel{~}{\epsilon }(x_0,y,\delta )1000\delta ^{1/24\mathrm{\Delta }^{}}<1/2`$ and
$$\stackrel{~}{\mu }=\stackrel{~}{\mu }(x_0,y,\delta )\frac{1}{\sqrt{2\pi }}\delta ^{1/2+\mathrm{\Delta }}>0.$$
Note that $`n=\epsilon ^2y^2=\epsilon ^2\delta ^{12\alpha }`$. Hence,
$`{\displaystyle \frac{\stackrel{~}{\epsilon }^{p/2}}{(k1)^{p/2}\stackrel{~}{\mu }^p}}`$ $`C(\stackrel{~}{\epsilon }\epsilon ^2\delta ^{1+2\alpha }\stackrel{~}{\mu }^2)^{p/2}`$
$`C(\delta ^{1/24\mathrm{\Delta }^{}}\epsilon ^2\delta ^{12\mathrm{\Delta }^{}}\delta ^{12\mathrm{\Delta }})^{p/2}`$
$`=C(\epsilon ^2\delta ^{1/26\mathrm{\Delta }^{}2\mathrm{\Delta }})^{p/2}.`$
Since $`1/26\mathrm{\Delta }^{}2\mathrm{\Delta }=1/3`$, this completes the proof. $`\mathrm{}`$
With the completion of our lemmas, we have made short work of the only proof that remains.
Proof of Lemma 3.3: Take $`\mathrm{\Delta }=1/108`$ in Lemma 4.1 and, for each $`p>2`$, let $`C_{p,1}`$ be the constant that appears in that lemma. Then take $`\mathrm{\Delta }=1/18`$ in Lemma 8.1. Let $`\delta _0>0`$ be as in that lemma and note that the conclusions of Lemmas 7.2 and 8.2 hold for this choice of $`\delta _0`$. For each $`p>2`$, let $`C_{p,2}`$ be the larger of the constants appearing in those two lemmas and let $`C_p=C_{p,1}C_{p,2}`$.
Now let $`0<\epsilon <1`$, $`0<\delta \delta _0`$, and $`n3`$. Choose $`\alpha >1/2`$ such that $`\epsilon /\sqrt{n}=\delta ^{1/2+\alpha }`$. If $`\alpha 1/108`$, then by Lemma 4.1,
$$P\left(M_n(1+\delta )M_n(1)>\frac{\epsilon }{\sqrt{n}}\right)C_{p,1}(\epsilon ^1\delta ^{1/4})^pC_p(\epsilon ^1\delta ^{1/6})^p.$$
If $`\alpha 1/18`$, then by Lemma 7.2,
$$P\left(M_n(1+\delta )M_n(1)>\frac{\epsilon }{\sqrt{n}}\right)C_{p,2}(\epsilon ^1\delta ^{1/4})^pC_p(\epsilon ^1\delta ^{1/6})^p.$$
If $`1/108\alpha 1/18`$, then by Lemma 8.2,
$$P\left(M_n(1+\delta )M_n(1)>\frac{\epsilon }{\sqrt{n}}\right)C_{p,2}(\epsilon ^1\delta ^{1/6})^pC_p(\epsilon ^1\delta ^{1/6})^p,$$
and we are done. $`\mathrm{}`$
Acknowledgments. This material appeared in my doctoral dissertation and I would like to thank my advisors, Chris Burdzy and Zhen-Qing Chen, for their instruction and guidance. I also thank the referees for their many helpful suggestions. For helpful discussions and for sharing their insights, I would like to thank Jon Wellner, Davar Khoshnevisan, Wenbo Li, Bruce Erickson, Yaozhong Hu, and especially Tom Kurtz, to whom I am indebted for his assistance in preparing this article for publication. This work was done while supported by VIGRE, for which I thank the NSF, University of Washington, and University of Wisconsin-Madison. |
warning/0507/gr-qc0507022.html | ar5iv | text | # Some Statistical Mechanical Properties of Photon Black Holes
## I Introduction
In general, black holes are defined uniquely by their mass, angular momentum and charge (cf. chandra ). In this paper we shall deal exclusively with Schwarzschild black holes, where the angular momentum and charge are both zero. These black holes of mass $`M`$ are understood as systems defined uniquely by the condition $`RR_s`$ and have the following properties:
$$R_s=\frac{2GM}{c^2},$$
(1)
$$\frac{S}{k_B}=\frac{A_{BH}}{4A_P},$$
(2)
$$T_{BH}=\frac{hc}{8\pi ^2k_BR_s}.$$
(3)
Equation (1) defines the Schwarzschild radius in terms of the mass. Relation (2) establishes the entropy $`S`$ of the system as $`1/4`$ of the horizon area $`A_{BH}=4\pi R_s^2`$, in units of the Planck area $`A_P=\mathrm{}G/c^3`$. Equation (3) states that the black hole radiates as a black body of the given temperature $`T_{BH}`$. This emission process implies a loss of energy for the system, which results in an evaporation rate for the black hole given by: $`dM/dt(\mathrm{}c^4/G^2)M^2`$.
Notice that $`dS/dM=(c^2T)^1`$, fixing the internal energy of the black hole as $`U=Mc^2`$. Equation (1) has been about in speculative form since the 18th century, and was given a firm theoretical footing within the framework of general relativity during 1930s. Equations (2)-(3) are the result of the vigorous development in black hole thermodynamics of the 1960s and 1970s by various authors, notably Bekenstein and Hawking (cf. nov and references therein).
It has been suggested (e.g. 't Hooft , Suss or Bousso for a review) that for any physical system equation (2) should hold always, with $``$ replacing the equality, which in turn should hold only in the black hole regime. This is termed the holographic principle, from the fact that the information content of an object would be limited not by its 3D volume, but by its 2D bounding surface. The interesting connection implied between quantum mechanics through $`A_P`$ and gravity through the particle horizon, has raised the hope that the validity of the holographic principle would yield important clues regarding quantum theories of gravity. In this sense, even a heuristic study as to the possible origin of this principle should prove valuable.
We study the behaviour of a classical black body photon gas as it is compressed into a black hole, and propose a simple model for such a system using only photons confined to the Schwarzschild radius at their lowest possible momentum level. Two parameters determine the model, with a restriction only on the product of both of them. A formal statistical mechanical calculation is given through which these parameters are determined, leading to agreement with all black hole structural properties of equations (1-3). The study of a self–gravitating photon sphere (a geon) was first introduced by Wheeler (geons, ) in 1955. Much development in this area has grown since then related to different physical properties of boson stars (see e.g. (jetzer, ; schunck, )) and their stability.
Given recent proposals of a physical model for a black hole interior within the framework of loop quantum gravity Rov , we analyse our model in this context. Taking the view that the Bekenstein entropy has an statistical mechanical origin in terms of counting states on the surface defined by the Schwarzschild radius of a black hole, canonical quantum gravity has yielded scenarios in which this entropy can be derived from first principles. We find no inconsistencies with the quantum gravity approach, which in fact allows us to explicitly and independently re-evaluate the parameters introduced in the quantum simple model, obtaining the same results.
## II Classical Limit
Most of the material in this section can be found elsewhere, e.g. Sorkin , it is reproduced here for context. For a photon gas having a black body spectrum the following well known relations define the total electromagnetic energy $`E_{EM}`$ and entropy $`S`$ in terms of the volume and temperature:
$$E_{EM}=\frac{\pi ^2}{15}\frac{(k_BT)^4}{(\mathrm{}c)^3}\frac{4\pi R^3}{3},$$
(4)
$$\frac{S}{k_B}=\frac{4\pi ^2}{45}\frac{(k_BT)^3}{(\mathrm{}c)^3}\frac{4\pi R^3}{3},$$
(5)
for a spherical region of radius $`R`$.
If we think of an ideal adiabatic wall enclosing this spherical region, we immediately obtain the well known scaling of $`TR^1`$, and we can eliminate $`(k_BT)`$ from equation(4) in favour of $`S`$ and $`R`$ to obtain
$$E=C\frac{\mathrm{}c}{R}\left(\frac{S}{k_B}\right)^{4/3},$$
(6)
where $`C`$ is a numerical constant of order unity. If we think of the contraction as proceeding into the black hole regime, we would think of the radius as reaching $`R_s`$, which in this case would yield
$$R_s=\frac{2GE_{EM}}{c^4}.$$
(7)
In equation (7) we have used equation (1), replacing $`M`$ for $`E_{EM}/c^2`$. Substitution of $`E_{EM}`$ from equation (6) into equation (7) leads to
$$\left(\frac{S}{k_B}\right)^{4/3}=C^{}\frac{A_{BH}}{A_p},$$
(8)
where $`C^{}`$ is a numerical factor of order unity. Two interesting conclusions are immediately evident from this last equation. Firstly, it is obvious that for any black body photon gas having $`R_s>R_p`$, a Schwarzschild radius larger than the Plank length, the holographic principle will be valid throughout the contraction process, as the horizon area will always be larger than $`A_{BH}`$. Second, that the classical equations for the diluted photon gas, equations (4-5), cannot be valid into the black hole regime, since the required relationship for the entropy of such an object is equation (2) and the exponent of equation (8) is $`4/3`$ and not $`1`$. The inconsistency of equation (8) with equation (2) signals that physical processes which the system being modelled surely experiences, such as pair creation at high temperatures and quantum effects related to the dimensions of the system being comparable to the typical de Broglie wavelengths of the photons, are not been taken into account by the classical description of the photon gas through equations (4-5). So far, we have assumed that the photon gas was being compressed by some external agency, however, if it is to form a self gravitating object, an equilibrium configuration should exist, and possibly a collapse beyond this. This point can be estimated by evaluating the Jeans length $`R_J`$ of the problem for a sound speed $`v_s=c/\sqrt{3}`$,
$$R_J=\frac{c}{(3G\rho )^{1/2}}.$$
(9)
In this context, the mass density $`\rho `$ is equivalent to $`E_{EM}/V`$. Notice that equation (9) is derived directly from Einstein’s equations, where the pressure term is directly $`\mathrm{d}E/\mathrm{d}V`$, hence no assumption of particle interactions is being made (see the appendix for details on this). From equation (9) we see that since $`\rho `$ scales with $`T^4`$, $`R_J`$ will scale with $`T^2`$, which is interesting given that under adiabatic conditions the radius of the system will scale with $`T^1`$. This means that gravitational instability will occur, i.e. $`R>R_J`$ above a certain critical temperature, below a certain critical equilibrium radius $`R_c`$. The situation becomes increasingly unstable in going towards smaller radii and larger temperatures. In general, for a fluid of mass–energy $`M`$, radius $`R`$ and sound speed $`v_s`$,
$$R_J=v_s\left(\frac{4\pi R^3}{3GM}\right)^{1/2},$$
(10)
which expressing M in terms of the Schwarzschild radius through equation (1) reads,
$$R_J=\left(\frac{8\pi }{3}\right)^{1/2}\frac{Rv_s}{c}\left(\frac{R}{R_s}\right)^{1/2}.$$
If we take the critical condition $`R=R_J`$, the previous relation gives
$$\frac{R_J}{R_s}=\left(\frac{3}{8\pi }\right)\left(\frac{c}{v_s}\right)^2.$$
(11)
Equation(11) shows that, since for all non–relativistic systems $`v_sc`$, we should expect $`R_JR_s`$. Indeed, for most astrophysical applications the Jeans radius of a system is many orders of magnitude larger than the Schwarzschild radius. However, in going to a relativistic fluid, the condition $`v_sc`$ will apply, leading to $`R_JR_s`$. In other words, the self-gravitating regime will appear only close to the black hole regime. For the adiabatic photon gas we have studied, taking $`v_s=c/\sqrt{3}`$ and equations (4-5) and (9), in correspondence with equation (11) we obtain also $`R_JR_s`$. This last result shows that the self-gravitating regime for the photon gas we have studied does not appear until one is very close to the black hole regime, at scales where the analysis leading to equation (8) already showed that the structure equations for the gas (4) and (5) are no longer valid. In any case, the analysis following equation (9) together with equation (11), strongly suggests that any self-gravitating photon gas will be very close to catastrophic collapse and black hole formation.
The above results are in fact valid into the regime where the self-gravity of the radiation field is important, as shown by Sorkin , instability sets in for $`R<2R_s`$, but equilibrium maximum entropy configuration exist above this radius, which however also show the scaling of equation (8), c.f. their results following their equation (41).
## III Quantum limit
The gravitational collapse and transition between the classical regime of section II and a black hole will not be treated explicitly. Advances in that direction can be found e.g. in Sorkin Pavon and Ding .
Being subject to the extreme gravitational regime of $`RR_s`$, it is reasonable to expect that the photons will be highly limited in momentum space. At this point we introduce as a hypothesis that all photons will have a wavelength $`\lambda =\alpha R_s`$, with $`\alpha `$ a numerical constant. We can evaluate the internal energy of the system as:
$$E_{EM}=\frac{Nhc}{\alpha R_s},$$
(12)
where $`N`$ is the total number of photons. Establishing a correspondence between $`E_{EM}`$ and the internal energy of a black hole as $`E_{EM}=Mc^2`$, and using equation (1) to express $`M`$ in terms of $`R_s`$, the above expression yields:
$$N=\frac{\alpha }{16\pi ^2}\frac{A_{BH}}{A_p}.$$
(13)
If we think of a correspondence between the total entropy of the system and the total photon number given by $`S/k_B=\beta N`$, with $`\beta `$ a proportionality constant expected to be of order unity, we find by comparison to equation (2) that the configuration we propose will satisfy all required black hole properties if the condition $`\alpha \beta =4\pi ^2`$ is satisfied.
We can compute $`\beta `$ directly by calculating the entropy of the proposed system from first principles, through the thermodynamic potential $`\mathrm{\Omega }`$ given by:
$$\mathrm{\Omega }=k_BT\underset{k}{}\mathrm{ln}\left(1e^{[\mu ϵ_k]/k_BT}\right),$$
(14)
where the summation is over quantum states, $`\mu `$ is the chemical potential, and $`ϵ_k`$ is the energy of the $`kth`$ state. Since the total number of components of the system is given by:
$$N=\underset{k}{}\left(\frac{1}{e^{[ϵ_k\mu ]/k_BT}1}\right),$$
(15)
and given that in the system proposed all photons have the same energy, we can write
$$N=\frac{N}{e^{[ϵ\mu ]/k_BT}1}.$$
(16)
Note that each photon is assumed to be in a distinct detailed quantum level, and hence the analogy with a condensate is not complete. Now,
$$[ϵ\mu ]/k_BT=\mathrm{ln}(2),$$
(17)
which when substituted back into equation (14) gives:
$$\mathrm{\Omega }=k_BTN\mathrm{ln}(2).$$
(18)
This last result now yields the entropy for the system through $`S=\mathrm{\Omega }/(k_BT)|_{V,N}`$ as $`S=N\mathrm{ln}(2)`$, providing a justification for the assumption of $`S/k_B=\beta N`$ made above, in terms of simple statistical physics arguments. We hence obtain $`\beta =\mathrm{ln}(2)`$ and $`\alpha =4\pi ^2/\mathrm{ln}(2)`$. Note that the previous results are of general validity for bosons, the case of photons is obtained with $`\mu =0`$.
We also note that given the restriction of a fixed total internal energy, in the hypothesis that this is to be the sum of the energies of $`N`$ photons, the maximum entropy state will be the one with the most photons. In that case, all photons are at their lowest possible energy $`\lambda =\alpha R_s`$. Hence, the configuration proposed is also a maximum entropy state and is suggestive of a micro–physical origin for black hole entropy.
It has been argued (bekenstein82, ) that the shortest scale that can enter into any physical theory is the Planck length. Although so far we have been working under the assumption of macroscopic black holes, we can extrapolate to the very small scales as follows. In the context of the ideas presented here a natural lower limit for the Schwarzschild radius of order the Plank length appears by setting $`N=1`$ in equation (13), a single photon black hole. The energy $`e`$ associated to this single photon is $`e\mathrm{}c/(Gh/c^3)^{1/2}10^{28}\text{eV}`$.
Note that using equation (1) to substitute $`M`$ for $`R_s`$ in equation (13) gives the following quantisation for the mass $`M_N`$ of a black hole in units of the Planck mass:
$$\frac{M_N}{m_p}=\left(\frac{\pi }{\alpha }\right)^{1/2}N^{1/2},$$
(19)
where $`m_P=\left(\mathrm{}c/G\right)^{1/2}`$, a quantisation suggested already by Bekenstein’s entropy equation (2). For sufficiently small black hole masses, this equation suggests a discrete spectrum associated to the transitions $`N=10`$, $`N=21`$, etc. The change in mass $`\mathrm{\Delta }M_N`$ corresponding to a $`\mathrm{\Delta }N=1`$ transition is given by
$$\mathrm{\Delta }M_N=M_N\left\{\left[1+\frac{\pi }{\alpha }\left(\frac{m_P}{M_N}\right)^2\right]^{1/2}1\right\}.$$
(20)
In the limit of a macroscopic black hole, where $`M_Nm_p`$, the above equation implies that
$$\frac{\mathrm{\Delta }M_N}{m_P}=\frac{\pi }{2\alpha }\frac{m_P}{M_N}.$$
(21)
It is interesting to note that $`c^2\mathrm{\Delta }M_N`$ approximately corresponds to $`k_BT_{BH}`$ for a black hole mass $`M_N`$.
If we now identify the time scale $`\mathrm{\Delta }t`$ for the mass loss $`\mathrm{\Delta }M_N`$ to take place with the limit of the Heisenberg uncertainty principle, we can set up $`\mathrm{\Delta }t\mathrm{}/c^2\mathrm{\Delta }M_N`$. Under the above considerations, the mass evaporation rate for a black hole is
$$\frac{\mathrm{\Delta }M_N}{\mathrm{\Delta }t}\frac{\mathrm{}c^4}{G^2}\frac{1}{M_N^2},$$
(22)
which is within a numerical constant of the standard evaporation rate for a black hole (nov, ), seen here as the macroscopic limit of an intrinsically quantum process.
If the structure of the photon configuration we are describing is in any way related to quantum phenomena akin to Bose-Einstein condensation, we should expect the temperature to lie well below the critical temperature $`T_c`$ for Bose-Einstein condensation. This, for relativistic particles can be calculated in an analogous way to well known Bose-Einstein critical temperatures for non-relativistic particles (e.g. (landau, )), by integrating the expression
$$dN=\frac{gVp^2dp}{2\pi ^2\mathrm{}^3\left[e^{(ϵ\mu )/k_BT}\right]},$$
(23)
for $`\mu =0`$, $`g=2`$ (photons) and in this case $`ϵ=pc`$, giving
$$N=\frac{V(T_ck_B)^3}{\pi ^2\mathrm{}^3c^3}_0^{\mathrm{}}\frac{z^2dz}{e^z1},$$
where $`z=ϵ/(k_BT)`$. Evaluation of the above integral yields $`2.202`$ and so, using $`V=(4\pi /3)R_s^3`$ leads to
$$N=\left(\frac{8.808}{3\pi }\right)\frac{R_s^3(T_ck_B)^3}{(\mathrm{}c)^3},$$
(24)
If we now use the expression for $`N`$ in equation (13), and writing $`T_c`$ in units of $`T_{BH}`$ for a black hole of radius equal to the Planck length $`R_p`$, which correspond to a temperature $`T_{BHp}`$, we get
$$\left(\frac{T_c}{T_{BHp}}\right)^3=\left(\frac{6\pi ^3}{1.101}\right)\frac{\alpha R_p}{R_s},$$
(25)
which for $`\alpha =4\pi ^2/\mathrm{ln}(2)`$, as determined through the statistical mechanical calculation shown above yields,
$$\left(\frac{T_c}{T_{BHp}}\right)=21.3\left(\frac{R_p}{R_s}\right)^{1/3}.$$
(26)
As $`T_c`$ scales with $`R_s^{1/3}`$ and $`T_{BH}`$ scales with $`R_s^1`$, it is clear that for all black holes larger than a certain limit, the condition $`T_{BH}T_c`$ will be satisfied. A comparison of both temperatures is shown in figure 1, as a function of $`R_s`$, from which we see that $`T_c`$ is already over an order of magnitude larger than $`T_{BH}`$ at $`R_s=R_p`$. Any realistic black hole will be at a temperature much lower than the critical temperature for Bose-Einstein condensation for photons, showing the internal consistency of the model.
The physics described so far is clearly highly idealised, however in a core collapse process within a massive star, as for the central region $`RR_s`$, the typical speeds and $`v_s`$ of the constituent particles must necessarily tend to $`c`$, with de Broglie wavelengths not larger than the Schwarzschild radius.
At this point, quantum effects similar to Bose-Einstein condensation could take place, packing all (or most) photons into the lowest energy state. In this sense, typical wavelengths of the order of the Schwarzschild radius would be expected, as longer wavelengths would be prohibited by the containment imposed by gravity, and shorter wavelengths would imply an expansion in momentum space. In this sense, the identification we have maintained of the constituent particles as photons is shown to be largely arbitrary, any relativistic bosons will yield essentially identical conclusions.
## IV Loop Quantum Gravity Approach
We now explore the ideas of the previous sections within the framework of loop gravity. In particular, we will see that this allows us an independent re-evaluation the constants $`\alpha `$ and $`\beta `$ established in the last section, with the aid of some general results from loop quantum gravity.
The loop quantisation of $`3+1`$ general relativity is described in terms of a set of spin network states which span the Hilbert space on which the theory is based. These spin network states are labelled by closed abstract graphs with spins assigned to each link and intertwining operators assigned to each vertex.
A resent result that follows from the theory is that if a surface $`\mathrm{\Sigma }`$ is intersected by a link $`\mathrm{}_i`$ of a spin network carrying the label $`j_i`$ it acquires an area ashtek1 ; rov1
$$A_\mathrm{\Sigma }(j_i)=8\pi A_p\gamma \sqrt{j_i(j_i+1)},$$
(27)
where $`\gamma `$ is the Immirzi parameter.
Let us now consider for our purposes that the horizon $`\mathrm{\Sigma }`$ is intersected by a large number $`N_{\mathrm{}}`$ of links. Each intersection with $`\mathrm{\Sigma }`$ represents a puncture. In the limit of large $`N_{\mathrm{}}`$, one can say that each puncture is equipped with an internal space $`H_j`$ (the space of all flat $`U(1)`$ connections on the punctured sphere) of dimension alexan1
$$dimH_j=2j+1.$$
(28)
Each puncture of an edge with spin $`j`$ increases the dimension of the boundary Hilbert space by a factor of $`2j+1`$. Under these considerations, it follows that the entropy can be calculated by
$$S(j_p)=\mathrm{ln}\left(\underset{p}{}dimH_{j_p}\right).$$
(29)
Statistically, the most important contribution comes from those configurations in which the lowest possible spin $`j_{min}`$ dominates, so we can write the entropy (29) as
$$S(j_{min})=N_{\mathrm{}}\mathrm{ln}(2j_{min}+1),$$
(30)
where $`N_{\mathrm{}}`$ is given by
$$N_{\mathrm{}}=\frac{A_{BH}}{A_\mathrm{\Sigma }(j_{min})}.$$
(31)
Due to the fact that the assumed gauge group of loop quantum gravity is SU$`(2)`$, then it follows that $`j_{\text{min}}=1/2`$, and so the Immirzi parameter is given by rov1
$$\gamma =\frac{\mathrm{ln}2}{\pi \sqrt{3}},$$
(32)
and (31) becomes
$$N_{\mathrm{}}=\frac{1}{4\mathrm{ln}2}\frac{A_{BH}}{A_p}.$$
(33)
The number of links $`N_{\mathrm{}}`$ associated to the particles we are dealing with in this article must be proportional to the number of particles $`N`$. The simplest possible configuration is the one in which the proportionality factor is equal to unity and so,
$$N_{\mathrm{}}=N.$$
(34)
Using this relation and equations (13) and (33) we can evaluate $`\alpha `$ to obtain
$$\alpha =\frac{4\pi ^2}{\mathrm{ln}2}.$$
(35)
The parameter $`\beta `$ previously defined through the relation $`S=\beta N`$ can now be re-derived independently through equation (30) to yield:
$$\beta =\frac{N_{\mathrm{}}}{N}\mathrm{ln}(2j+1)_{j=1/2}=\mathrm{ln}2.$$
(36)
From (35) and (36) we can see that the product $`\alpha \beta =4\pi ^2`$ as required by the considerations on section III. It is interesting that the model proposed in the previous sections is seen not to be in conflict with a loop gravity approach, which indeed allows to re–evaluate the scaling parameters introduced earlier independently and in accordance with the expectations of the physics discussed above.
## V Conclusions
A photon gas contained within an adiabatic enclosure will satisfy the holographic principle, at least until before reaching the black hole regime, which approximately coincides with the self-gravitating condition and where the classical description is no longer valid.
A configuration where all photons have the same energy within $`R=R_S`$ can be constructed to satisfy all black hole conditions.
This configuration gives rise to a discrete evaporation spectrum for $`R_S`$ close to the Planck length, and in the macroscopic limit allows a re-derivation of the standard black hole evaporation rate.
Our results are consistent with the loop quantum gravity scheme, satisfying the constraints required to be in agreement with the Bekenstein-Hawking entropy for a black hole.
A comparison of both regimes studied is highly suggestive of a heuristic proof of the holographic principle, as any real system would require an increase of entropy (at fixed energy and volume) to be turned into the photon gas we have studied.
## VI Acknowledgements
X. Hernandez acknowledges the support of grant UNAM DGAPA (IN117803-3), CONACyT (42809/A-1) and CONACyT (42748). C. Lopez-Monsalvo thanks economic support from UNAM DGAPA (IN119203). S. Mendoza gratefully acknowledges financial support from CONACyT (41443) and UNAM DGAPA (IN119203). We thank the anonymous referee for his comments and suggestions which improved the final version of the paper.
*
## Appendix A Relativistic Jeans criterium for gravitational instability
In order to show that the Jeans gravitational instability limit is valid in the general relativistic regime, let us proceed as follows. The condition of hydrostatic equilibrium in general relativistic fluid mechanics is obtained by using the fact that the field is static. This means that one can describe the problem in a frame of reference in which the fluid is at rest, with all hydrodynamical quantities independent of time. This also implies that the mixed space and time components of the metric tensor are null. Under these assumptions, the equation of hydrostatic equilibrium is then given by (daufm, )
$$\frac{1}{w}\frac{p}{r}=\frac{1}{2}\frac{}{r}\mathrm{log}g_{00},$$
(37)
where $`g_{00}`$ is the time component of the metric tensor, $`w=e+p`$ is the enthalpy per unit volume, $`e`$ internal energy density and $`p`$ the pressure. Oppenheimer & Volkoff (oppenheimer, ) showed that equation (37) can be brought to the form (see e.g. MTW, )
$$\frac{\mathrm{d}p}{\mathrm{d}r}=\frac{\left(e+p\right)}{r\left(r\frac{2GM(r)}{c^2}\right)}\left\{\frac{GM(r)}{c^2}+\frac{4\pi }{c^2}Gr^3\rho \right\},$$
(38)
where the mass–energy $`M(r)`$ within a radius $`r`$ is given by
$$M(r)=4\pi _0^r\rho r^2dr,$$
(39)
and $`\rho (r):=e/c^2`$ is the mass–energy density of the fluid. Note that for the case of relativistic and non–relativistic dust particles, $`M(r)`$ represents the mass of particles within radius $`r`$. For the case of a photon gas, $`M(r)`$ is the mass corresponding to the internal energy.
We now assume that the plasma obeys a Bondi–Wheeler equation of state
$$p=\left(\kappa 1\right)e,$$
(40)
with constant index $`\kappa `$. This means that the sound velocity $`v_\text{s}`$ is given by the relation $`v_\text{s}^2=c^2\left(\kappa 1\right)`$ and so, the left hand side of equation (38) can be written as $`\left(v_\text{s}^2/c^2\right)\mathrm{d}e/\mathrm{d}r`$. Seen in this way, the left hand side of equation (38) no longer represents gradients of pressure which are in balance with self–gravitational forces related to the plasma. Indeed, the balance with the gravitational forces produced by the plasma is now related to the gradients of its proper internal energy density $`e`$ by
$$\left(\kappa 1\right)\frac{\mathrm{d}e}{\mathrm{d}r}=\frac{ke}{r\left(r\frac{2GM(r)}{c^2}\right)}\left\{\frac{GM(r)}{c^2}+\frac{4\pi }{c^2}Gr^3\rho \right\}.$$
(41)
Let us now take the absolute value on both sides of equation (41). To order of magnitude $`\mathrm{d}e/\mathrm{d}re/r`$ and $`M(r)\left(4/3\right)\pi r^3\rho `$. A gravitational instability occurs when the absolute value of the left hand side of equation (38) (or equivalently equation (40)) is less than the absolute value of its right hand side. Using all the above statements it follows that this instability occurs when the radial coordinate $`r`$ is such that
$$r\sqrt{\frac{3}{8\pi \left(3\kappa 1\right)}}\frac{v_\text{s}}{\sqrt{G\rho }}:=\mathrm{\Lambda }_\text{J}.$$
(42)
The quantity $`\mathrm{\Lambda }_\text{J}`$ on the right hand side of equation (42) is of the same order of magnitude as the standard Jeans length used in non–relativistic fluid dynamics. In other words, the criterium (42) means that the Jeans criteria for gravitational collapse is also valid in the relativistic regime as well.
For the particular case studied in this article, the constant $`\kappa =4/3`$ for a photon gas and the Jeans criteria can be applied to it, giving $`r\left(1/2\sqrt{2\pi }\right)v_\text{s}/\sqrt{G\rho }`$.
We see than the Jeans criterion can be generalised beyond an equilibrium between gas pressure and rest–mass self–gravity, to a very general equilibrium between energy–momentum flux and total self–gravity. |
warning/0507/hep-th0507169.html | ar5iv | text | # 1 Introduction
## 1 Introduction
In recent works the number of explicit examples of known five dimensional Einstein–Sasaki metrics was considerably enlarged by a new class of such metrics interpolating in a certain sense between the round $`S^5`$ sphere and the $`T^{1,1}`$ space . These are of cohomogeneity one, with principal orbits $`SU(2)\times U(1)\times U(1)`$ and in order to satisfy global regularity issues their parametric space is characterized by two coprime positive integers $`p`$ and $`q`$. Hence, these were called $`Y^{p,q}`$ spaces. This class has been further generalized by taking the BPS limits of Euclideanized Kerr–de Sitter black hole metrics with two independent angular momenta parameters . This construction leads to local Einstein–Sasaki metrics of cohomogeneity two, with $`U(1)\times U(1)\times U(1)`$ principal orbits. Similarly, these metrics, called $`L^{p,q,r}`$, are characterized by positive coprime integers $`p`$, $`q`$ and $`r`$ in order that they smoothly extend onto complete, non-singular compact manifolds. The $`Y^{p,q}`$ spaces come as special limits of the $`L^{p,q,r}`$ ones when the angular parameters coincide and a $`U(1)`$ symmetry factor gets enhanced into $`SU(2)`$. Similarly, the $`T^{1,1}`$ space results by a further symmetry enhancement.
One advantage of having explicit five-dimensional regular spaces is that they can be used as a base in the construction of six-dimensional Ricci-flat cones, which in turn are basic blocks for the ten-dimensional supergravity solutions representing the gravitational field of stacks of branes and the dual description of supersymmetric gauge theories within the gauge/gravity correspondence -. The usual cone one constructs suffers from a singularity in its tip and therefore part of the effort is to regularize it. A basic example is the six-dimensional cone based on the $`T^{1,1}`$ space in which the conical singularity were first smoothened out in the so called deformed and resolved conifolds , by introducing a parameter, and keeping finite at the tip of the cone either an $`S^2`$ or an $`S^3`$ factor. In addition, there is also the regularized conifold in which the original curvature singularity becomes a removable bolt singularity . Introducing D3-branes and taken into account their backreaction transforms the Ricci-flat solution of the cone times the Minkowski space into a warped solution of the full type-IIB supergravity -. We note that having a regular six-dimensional cone does not necessarily imply the regularity of the ten-dimensional solution (see, in particular, that emphasizes that). The purpose of this paper is to construct the six-dimensional supersymmetric cones based on the newly discovered $`Y^{p,q}`$ and $`L^{p,q,r}`$ spaces and use them for the construction of the ten-dimensional type-IIB supergravity solutions that include the brane backreaction.
This letter is organized as follows: In section 2 we present a brief review of the relevant aspects of the $`Y^{p,q}`$ and $`L^{p,q,r}`$ spaces. In section 3 we explicitly construct supersymmetric six-dimensional cone solutions based on these spaces. They depend on a constant moduli parameter as in the regularized conifold. In section 4 we construct supersymmetric supergravity solutions of a stack of D3- and D5-branes on these cones within type-IIB supergravity. They have the expected behaviour in the UV, but still suffer from a singularity in the IR.
## 2 Brief review of the $`Y^{p,q}`$ and $`L^{p,q,r}`$ spaces
In this section we provide a short review of some relevant to our construction aspects of the $`Y^{p,q}`$ and $`L^{p,q,r}`$ spaces and also comment on their relation. For details the reader should really consult the literature.
### 2.1 $`Y^{p,q}`$ geometry
The five dimensional $`Y^{p,q}`$ geometry in its canonical form is described by the following metric -
$$ds_5^2=ds_4^2+\left(\frac{1}{3}d\psi +\sigma \right)^2,$$
(2.1)
where the four dimensional metric is
$$ds_4^2=\frac{1cy}{6}\left(d\theta ^2+\mathrm{sin}^2\theta d\varphi ^2\right)+\frac{dy^2}{w(y)q(y)}+\frac{1}{36}w(y)q(y)(d\beta +c\mathrm{cos}\theta d\varphi )^2,$$
(2.2)
with
$$\sigma =\frac{1}{3}\mathrm{cos}\theta d\varphi +\frac{1}{3}y(d\beta +c\mathrm{cos}\theta d\varphi ).$$
(2.3)
and
$`w(y)={\displaystyle \frac{2(ay^2)}{1cy}},q(y)={\displaystyle \frac{a3y^2+2cy^3}{ay^2}}.`$ (2.4)
Therefore, it can be seen as a $`U(1)`$ bundle over a four-dimensional Einstein-Kähler metric with the Kähler two-form given by $`d\sigma =2J_4`$. It can be checked explicitly that the four-dimensional metric is Einstein with $`R_{\mu \nu }=6g_{\mu \nu }`$ and hence the five-dimensional metric is Einstein–Sasaki with $`R_{\mu \nu }=4g_{\mu \nu }`$. The coordinate $`y`$ ranges between the two smallest roots of the cubic equation $`a3y^2+2cy^3=0`$, so that the signature of the metric remains Euclidean. These are given in terms of the coprime integers $`p`$ and $`q`$ with explicit expressions that don’t concerns us here. In order also to obtain a compact manifold the coordinate $`\alpha `$ has a finite range. The remaining ones $`\theta `$, $`\varphi `$ and $`\psi `$ have periods $`\pi `$, $`2\pi `$ and $`2\pi `$, respectively.
### 2.2 $`L^{p,q,r}`$ geometry
The five-dimensional $`L^{p,q,r}`$ geometry is described by the following metric
$`ds_5^2=ds_4^2+\left(d\tau +\sigma \right)^2,`$ (2.5)
where the four-dimensional metric is
$`ds_4^2`$ $`=`$ $`{\displaystyle \frac{\rho ^2dx^2}{4\mathrm{\Delta }_x}}+{\displaystyle \frac{\rho ^2d\theta ^2}{\mathrm{\Delta }_\theta }}+{\displaystyle \frac{\mathrm{\Delta }_x}{\rho ^2}}\left({\displaystyle \frac{\mathrm{sin}^2\theta }{\alpha }}d\varphi +{\displaystyle \frac{\mathrm{cos}^2\theta }{\beta }}d\psi \right)^2`$ (2.6)
$`+{\displaystyle \frac{\mathrm{\Delta }_\theta \mathrm{sin}^2\theta \mathrm{cos}^2\theta }{\rho ^2}}\left[(1x/\alpha )d\varphi (1x/\beta )d\psi \right]^2`$
and
$`\sigma =(1x/\alpha )\mathrm{sin}^2\theta d\varphi +(1x/\beta )\mathrm{cos}^2\theta d\psi ,\rho ^2=\mathrm{\Delta }_\theta x,`$
$`\mathrm{\Delta }_x=x(\alpha x)(\beta x)\mu ,\mathrm{\Delta }_\theta =\alpha \mathrm{cos}^2\theta +\beta \mathrm{sin}^2\theta .`$ (2.7)
The five-dimensional metric has the standard form, as in the $`Y^{p,q}`$ case. The parameter $`\mu `$ is trivial and can be set to any non-zero constant by rescaling $`\alpha `$, $`\beta `$, and $`x`$, hence the metric depend on two parameters. The principal orbits $`U(1)\times U(1)\times U(1)`$ of the metric degenerate at $`\theta =0`$ and $`\theta =\frac{\pi }{2}`$ and at the roots of the cubic function $`\mathrm{\Delta }_x`$. In order to obtain metrics on non-singular manifolds the ranges of the coordinates should be $`0<\theta <\frac{\pi }{2}`$ and $`x_1<x<x_2`$, where $`x_1`$ and $`x_2`$ are the two smallest roots of the equation $`\mathrm{\Delta }_x=0`$. The ranges of the coordinates $`\varphi `$ and $`\psi `$ are determined using the notion of “surface gravity”, important in black hole solutions of Lorentzian signature. The analysis of the behavior at each collapsing orbit can be realized by examining the associated Killing vector $`\mathrm{}`$ whose length vanishes at the degeneration surface. By normalizing the Killing vector so that its “surface gravity” $`\kappa `$ is equal to unity, one obtains a translation generator $`/\chi `$, where $`\chi `$ is a local coordinate near the degeneration surface. The metric extends smoothly onto the surface if $`\chi `$ has period $`2\pi `$. The “surface gravity” is
$$\kappa ^2=\frac{g^{\mu \nu }(_\mu \mathrm{}^2)(_\nu \mathrm{}^2)}{4\mathrm{}^2},$$
(2.8)
in the limit the degeneration surface is reached. At the degeneration surfaces $`\theta =0`$ and $`\theta =\frac{\pi }{2}`$ the normalized killing vectors are $`/\varphi `$ and $`/\psi `$ respectively, so the periodicity of the coordinates $`\varphi `$ and $`\psi `$ is 0 to $`2\pi `$. At the degeneration surfaces $`x=x_1`$ and $`x=x_2`$, the associated normalized Killing vectors $`\mathrm{}_1`$ and $`\mathrm{}_2`$ are given by
$$\mathrm{}_i=c_i\frac{}{\tau }+a_i\frac{}{\varphi }+b_i\frac{}{\psi },$$
(2.9)
where the constants $`c_i`$, $`a_i`$ and $`b_i`$ are given by
$`a_i={\displaystyle \frac{\alpha c_i}{x_i\alpha }},b_i={\displaystyle \frac{\beta c_i}{x_i\beta }},c_i={\displaystyle \frac{(\alpha x_i)(\beta x_i)}{2(\alpha +\beta )x_i\alpha \beta 3x_i^2}}.`$ (2.10)
Similarly to the case of $`Y^{p,q}`$, all parameters are eventually given in terms of three coprime positive integers $`p`$, $`q`$ and $`r`$ so that the manifolds are complete and free of singularities.
#### 2.2.1 Connection with $`Y^{p,q}`$
If one sets $`p+q=2r`$, implying $`\alpha =\beta `$, the metric (2.5) reduce to (2.2) with $`Y^{p,q}=L^{pq,p+q,p}`$. Then the relation of variables and parameters is given by
$`x{\displaystyle \frac{\alpha }{3}}(1+2cy),\theta {\displaystyle \frac{\theta }{2}},\varphi \psi \varphi ,\varphi +\psi {\displaystyle \frac{\beta }{c}},3\tau +\varphi +\psi \psi `$ (2.11)
and
$$\mu =\frac{4}{27}(1ac^2)\alpha ^3.$$
(2.12)
After the coordinate transformation (2.11) the Killing vectors for the degeneration surfaces $`\theta =0`$ and $`\theta =\pi `$ are $`(/\varphi /\psi +c/\beta )`$ and $`(/\varphi /\psi +c/\beta )`$, respectively. At the degeneration surfaces $`y=y_1`$ and $`y=y_2`$, where $`y_1`$ and $`y_2`$ are the roots of the equation $`q(y)=0`$, the normalized killing vectors $`\mathrm{}_1`$ and $`\mathrm{}_2`$ are given by
$$\mathrm{}_i=\frac{}{\psi }\frac{1}{y_i}\frac{}{\beta },i=1,2$$
(2.13)
and correspond to the vectors in (2.9).
## 3 The six-dimensional cones
In this section we explicitly solve the supersymmetric Killing spinor equations and determine the six-dimensional cones. The latter are by construction Ricci-flat.
### 3.1 The cone over the $`Y^{p,q}`$ space
First we construct a six-dimensional supersymmetric cone over the $`Y_{p,q}`$ space as a base. The metric ansatz is
$`ds_6^2`$ $`=`$ $`dr^2+A(r)^2\left({\displaystyle \frac{1}{3}}d\psi +\sigma \right)^2+B(r)^2ds_4^2.`$ (3.1)
We will use the vielbein basis
$`e^1=B(r)\sqrt{{\displaystyle \frac{1cy}{6}}}d\theta ,e^2=B(r)\sqrt{{\displaystyle \frac{1cy}{6}}}\mathrm{sin}\theta d\varphi ,`$ (3.2)
$`e^3=B(r){\displaystyle \frac{dy}{\sqrt{w(y)q(y)}}},e^4=B(r){\displaystyle \frac{1}{6}}\sqrt{w(y)q(y)}(d\beta +c\mathrm{cos}\theta d\varphi ),`$
$`e^5=A(r){\displaystyle \frac{1}{3}}\left[d\psi \mathrm{cos}\theta d\varphi +y(d\beta +c\mathrm{cos}\theta d\varphi )\right],e^6=dr.`$
The non-vanishing components of the spin connection are
$`\omega ^{12}={\displaystyle \frac{1}{B}}\left[\mathrm{cot}\theta \left({\displaystyle \frac{6}{1cy}}\right)^{1/2}e^2+{\displaystyle \frac{A}{B}}e^5{\displaystyle \frac{c}{2(1cy)}}\sqrt{wq}e^4\right],`$
$`\omega ^{34}={\displaystyle \frac{1}{B}}\left[{\displaystyle \frac{}{y}}\sqrt{wq}e^4+{\displaystyle \frac{A}{B}}e^5\right],`$
$`\omega ^{14}={\displaystyle \frac{1}{B}}{\displaystyle \frac{c}{2(1cy)}}\sqrt{wq}e^2,\omega ^{15}={\displaystyle \frac{A}{B^2}}e^2,`$
$`\omega ^{13}={\displaystyle \frac{1}{B}}{\displaystyle \frac{c}{2(1cy)}}\sqrt{wq}e^1,\omega ^{25}={\displaystyle \frac{A}{B}}e^1,`$ (3.3)
$`\omega ^{24}={\displaystyle \frac{1}{B}}{\displaystyle \frac{c}{2(1cy)}}\sqrt{wq}e^1,\omega ^{45}={\displaystyle \frac{A}{B^2}}e^3,`$
$`\omega ^{23}={\displaystyle \frac{1}{B}}{\displaystyle \frac{c}{2(1cy)}}\sqrt{wq}e^2,\omega ^{35}={\displaystyle \frac{A}{B^2}}e^4,`$
$`\omega ^{i6}={\displaystyle \frac{B^{}}{B}}e^i,i=1\mathrm{}4,\omega ^{56}={\displaystyle \frac{A^{}}{A}}e^5,`$
where prime denotes differentiation with respect to $`r`$.<sup>1</sup><sup>1</sup>1One might try a more general ansatz than (3.1) by putting different functions of $`r`$ in front of every vielbein. However, it turns that the consistent with supersymmetry solution in the end simplifies the ansatz to that in (3.1). This is consistent with the observation of that, generically the $`Y^{p,q}`$ manifolds do not admit complex deformations. The Killing spinor equation are
$$_\mu ϵ+\frac{1}{4}\omega _\mu ^{ab}\mathrm{\Gamma }_{ab}ϵ=0.$$
(3.4)
In analyzing this set of equations we found necessary to impose the two projections
$$\mathrm{\Gamma }_{12}ϵ=\mathrm{\Gamma }_{34}ϵ=\mathrm{\Gamma }_{56}ϵ,$$
(3.5)
hence reducing supersymmetry to $`1/4`$ of the maximal. The Killing spinor turns out to be
$$ϵ=e^{\frac{1}{2}\psi \mathrm{\Gamma }_{12}}ϵ_0.$$
(3.6)
In addition we obtained the following system of differential equations that determine the functions $`A(r)`$ and $`B(r)`$
$`B^{}={\displaystyle \frac{A}{B}},A^{}=32{\displaystyle \frac{A^2}{B^2}}.`$ (3.7)
The general solution to the system is
$$B^2=R^2,A^2=R^2\left(1+\frac{C}{R^6}\right),$$
(3.8)
where $`C`$ is a constant. The relation of the two variables r and R is via the differential
$$dr=\left(1+\frac{C}{R^6}\right)^{1/2}dR.$$
(3.9)
Note that we have absorbed a second integration constant by a suitable redefinition of the variable $`R`$. After substituting the solution of the killing spinor equations to (3.1) the metric takes the simple form<sup>2</sup><sup>2</sup>2This solution belongs to the class of examples considered in by solving the second order field equations. We thank C. Pope for the information. A form of this solution was also obtained in but without any claim or proof on supersymmetry.
$$ds_6^2=\left(1+\frac{C}{R^6}\right)^1dR^2+R^2\left(1+\frac{C}{R^6}\right)\left(\frac{1}{3}d\psi +\sigma \right)^2+R^2ds_4^2.$$
(3.10)
We have checked that this metric has the same killing vectors with (2.2), with degeneration surfaces $`\theta =0`$, $`\theta =\pi `$, $`y=y_1`$ and $`y=y_2`$.
The asymptotic behavior for large values of $`R`$ takes the universal form
$$ds_6^2dR^2+R^2ds_5^2,\mathrm{as}R\mathrm{}$$
(3.11)
and it describes the usual cone whose base is given by the five dimensional metric (2.2). This solution is exact for all values of $`R`$ since it can be obtained by simply letting $`C=0`$. The constant $`C`$ changes the solution drastically towards the interior. When $`C0`$, the variable $`R0`$ and then the manifold has a curvature singularity at $`R=0`$. However, if $`C=a^6<0`$, where $`a`$ is a real positive constant, then the variable $`Ra`$. To examine the behaviour of the metric near $`R=a`$ we change into the new radial variable $`t=\sqrt{6a(Ra)}`$. We find
$`ds_6^2a^2ds_4^2+{\displaystyle \frac{1}{9}}dt^2+t^2\left({\displaystyle \frac{1}{3}}d\psi +\sigma \right)^2\mathrm{as}t0.`$ (3.12)
Therefore, near $`t=0`$ and for constant $`y`$, $`\theta `$, $`\beta `$ and $`\varphi `$, the metric behaves (up to $`1/9`$) as $`dt^2+t^2d\psi ^2`$ which shows that $`t=0`$ is a bolt singularity which is removable since the periodicity of the angle is $`0\psi <2\pi `$. The full solution interpolates between (3.12) for $`Ra`$ and (3.11) for $`R\mathrm{}`$. This is similar to that found in for the cones over the symmetric coset spaces $`SU(2)^n/U(1)^{n1}`$ that includes the regularization of the singular conifold on $`T^{1,1}`$ for $`n=2`$ . However, in our case we do not have a completely non-singular solution at the supergvavity level.<sup>3</sup><sup>3</sup>3We thank C. Pope for a correspondence on this. The Einstein–Kahler four-dimensional base is singular. At best it has orbifold singularities, when $`4p^23q^2=n^2`$, where $`nZ`$. The $`Y^{p,q}`$ metrics are then an orbifold $`U(1)`$ bundle over this Einstein–Kahler base orbifold . Nevertheless, string theory has probably more success with orbifold singularities than true curvature singularities since in some cases the singularity is resolved before the ”smoothening” . It is interesting to investigate this further.
### 3.2 The cone over the $`L^{p,q,r}`$ space
To construct the six-dimensional supersymmetric cone over $`L_{p,q,r}`$ we make the ansatz
$`ds_6^2`$ $`=`$ $`dr^2+A(r)^2\left(d\tau +\sigma \right)^2+B(r)^2ds_4^2`$ (3.13)
and use the vielbein basis
$`e^1=B(r){\displaystyle \frac{\rho }{\mathrm{\Delta }_\theta ^{1/2}}}d\theta ,e^2=B(r){\displaystyle \frac{\mathrm{\Delta }_\theta ^{1/2}\mathrm{sin}\theta \mathrm{cos}\theta }{\rho }}\left({\displaystyle \frac{\alpha x}{\alpha }}d\varphi {\displaystyle \frac{\beta x}{\beta }}d\psi \right),`$
$`e^3=B(r){\displaystyle \frac{\mathrm{\Delta }_x^{1/2}}{\rho }}\left({\displaystyle \frac{\mathrm{sin}^2\theta }{\alpha }}d\varphi +{\displaystyle \frac{\mathrm{cos}^2\theta }{\beta }}d\psi \right),e^4=B(r){\displaystyle \frac{\rho }{2\mathrm{\Delta }_x^{1/2}}}dx`$ (3.14)
$`e^5=A(r)\left(d\tau +\sigma \right),e^6=dr.`$
After some tedious algebra we found that the non-vanishing components of the spin connection are
$`\omega ^{12}={\displaystyle \frac{1}{B}}\left[\left(2\mathrm{cot}2\theta {\displaystyle \frac{\mathrm{\Delta }_\theta ^{1/2}}{\rho }}{\displaystyle \frac{\alpha \beta }{2}}\mathrm{sin}2\theta \left({\displaystyle \frac{1}{\rho \mathrm{\Delta }_\theta ^{1/2}}}{\displaystyle \frac{\mathrm{\Delta }_\theta ^{1/2}}{\rho ^3}}\right)\right)e^2+{\displaystyle \frac{\mathrm{\Delta }_x^{1/2}}{\rho ^3}}e^3+{\displaystyle \frac{A}{B}}e^5\right],`$
$`\omega ^{34}={\displaystyle \frac{1}{B}}\left[\left({\displaystyle \frac{3x^22(\alpha +\beta )x+\alpha \beta }{\rho \mathrm{\Delta }_x^{1/2}}}+{\displaystyle \frac{\mathrm{\Delta }_x^{1/2}}{\rho ^3}}\right)e^3{\displaystyle \frac{A}{B}}e^5+{\displaystyle \frac{\alpha \beta }{2}}\mathrm{sin}2\theta {\displaystyle \frac{\mathrm{\Delta }_\theta ^{1/2}}{\rho ^3}}e^2\right],`$
$`\omega ^{14}={\displaystyle \frac{1}{B}}\left[{\displaystyle \frac{\mathrm{\Delta }_x^{1/2}}{\rho ^3}}e^1{\displaystyle \frac{\alpha \beta }{2}}\mathrm{sin}2\theta {\displaystyle \frac{\mathrm{\Delta }_\theta ^{1/2}}{\rho ^3}}e^4\right],\omega ^{15}={\displaystyle \frac{A}{B^2}}e^2,`$
$`\omega ^{13}={\displaystyle \frac{1}{B}}\left[{\displaystyle \frac{\mathrm{\Delta }_x^{1/2}}{\rho ^3}}e^2{\displaystyle \frac{\alpha \beta }{2}}\mathrm{sin}2\theta {\displaystyle \frac{\mathrm{\Delta }_\theta ^{1/2}}{\rho ^3}}e^3\right],\omega ^{25}={\displaystyle \frac{A}{B^2}}e^1,`$ (3.15)
$`\omega ^{24}={\displaystyle \frac{1}{B}}\left[{\displaystyle \frac{\mathrm{\Delta }_x^{1/2}}{\rho ^3}}e^2{\displaystyle \frac{\alpha \beta }{2}}\mathrm{sin}2\theta {\displaystyle \frac{\mathrm{\Delta }_\theta ^{1/2}}{\rho ^3}}e^3\right],\omega ^{45}={\displaystyle \frac{A}{B^2}}e^3,`$
$`\omega ^{23}=+{\displaystyle \frac{1}{B}}\left[{\displaystyle \frac{\mathrm{\Delta }_x^{1/2}}{\rho ^3}}e^1{\displaystyle \frac{\alpha \beta }{2}}\mathrm{sin}2\theta {\displaystyle \frac{\mathrm{\Delta }_\theta ^{1/2}}{\rho ^3}}e^4\right],\omega ^{35}={\displaystyle \frac{A}{B^2}}e^4,`$
$`\omega ^{i6}={\displaystyle \frac{B^{}}{B}}e^i,i=1\mathrm{}4,\omega ^{56}={\displaystyle \frac{A^{}}{A}}e^5.`$
The set of projections obtained by analyzing the Killing spinor equations are the same as in the case of the cone over the $`Y^{p,q}`$ space in (3.5) and similarly the system of differential equations (3.7) determining the functions $`A(r)`$ and $`B(r)`$. The Killing spinor is
$$ϵ=e^{\frac{1}{2}(3\tau +\varphi +\psi )\mathrm{\Gamma }_{12}}ϵ_0.$$
(3.16)
Finally the solution takes the simple form
$$ds_6^2=\left(1+\frac{C}{R^6}\right)^1dR^2+R^2ds_4^2+R^2\left(1+\frac{C}{R^6}\right)(d\tau +\sigma )^2.$$
(3.17)
The metric above has the same killing vectors with (2.5), with degeneration surfaces $`\theta =0`$, $`\theta =\pi /2`$, $`x=x_1`$ and $`x=x_2`$. As in the case of the cone over $`Y^{p,q}`$, for $`C=a^6`$ the metric (3.17) is free of curvature singularities, but it has the singularities associated with the four-dimensional Einstein–Kahler space.
## 4 Warped type-IIB solutions
In order to construct ten-dimensional supersymmetric warped solutions we utilize the procedure developed in . We will use the cone over the $`L^{p,q,r}`$ space.<sup>4</sup><sup>4</sup>4For related work with the usual cone over $`L^{p,q,r}`$ see also and . This procedure was also recently used to construct a solution with the usual cone over $`Y^{p,q}`$ . The first step is to find a harmonic $`(2,1)`$ form $`\mathrm{\Omega }_{2,1}`$. For this reason we will use the local Kähler form $`J_4`$ on the Kähler-Einstein base
$`J_4`$ $`=`$ $`\stackrel{~}{e}^1\stackrel{~}{e}^2+\stackrel{~}{e}^3\stackrel{~}{e}^4`$
$`=`$ $`\mathrm{sin}\theta \mathrm{cos}\theta d\theta \left[(1x/\alpha )d\varphi (1x/\beta )d\psi \right]`$
$`{\displaystyle \frac{1}{2}}dx\left(1/\alpha \mathrm{sin}^2\theta d\varphi +1/\beta \mathrm{cos}^2\theta d\psi \right),`$
where $`\stackrel{~}{e}^i=e^i/A(r)`$. In turn, based on , it is possible to construct a $`\mathrm{\Omega }_{2,1}`$ form from a (1,1) form $`\omega `$ such that $`_4\omega =\omega `$, $`d\omega =0`$ and $`\omega J_4=0`$. Such a form is similar to the one proposed in and for the case of the usual cone on the $`Y^{p,q}`$ and $`L^{p,q,r}`$, respectively. We have explicitly that
$`\omega `$ $`=`$ $`{\displaystyle \frac{1}{\rho ^4}}(\stackrel{~}{e}^1\stackrel{~}{e}^2\stackrel{~}{e}^3\stackrel{~}{e}^4)`$
$`=`$ $`{\displaystyle \frac{1}{\rho ^4}}[\mathrm{sin}\theta \mathrm{cos}\theta d\theta ((1x/\alpha )d\varphi (1x/\beta )d\psi )`$
$`+{\displaystyle \frac{1}{2}}dx(1/\alpha \mathrm{sin}^2\theta d\varphi +1/\beta \mathrm{cos}^2\theta d\psi )],`$
where the overall factor in the first line has been fixed by demanding that $`d\omega =0`$.<sup>5</sup><sup>5</sup>5 A second possibility is $`\omega ={\displaystyle \frac{1}{\mathrm{sin}2\theta (\mathrm{\Delta }_\theta \mathrm{\Delta }_x)^{1/2}}}(\stackrel{~}{e}^1\stackrel{~}{e}^4\stackrel{~}{e}^2\stackrel{~}{e}^3)={\displaystyle \frac{1}{2\alpha \beta }}d\varphi d\psi +{\displaystyle \frac{\rho ^2}{2\mathrm{sin}2\theta \mathrm{\Delta }_\theta \mathrm{\Delta }_x}}d\theta dx.`$ (4.3) However, this form is singular at $`\theta =0,\pi `$ and $`x=x_1,x_2`$ and cannot be used to construct a complex 3-form with well defined associated charges. We thank C. Herzog for a correspondence on this. In order to check that the above form is indeed $`(1,1)`$ we introduce the set of complex coordinates (This should be equivalent to that presented in in a different coordinate system)
$`\eta _1={\displaystyle \frac{\mathrm{cot}\theta }{\mathrm{\Delta }_\theta }}d\theta +{\displaystyle \frac{\beta x}{2\mathrm{\Delta }_x}}dx+{\displaystyle \frac{i}{\alpha }}d\varphi ,`$
$`\eta _2={\displaystyle \frac{\mathrm{tan}\theta }{\mathrm{\Delta }_\theta }}d\theta +{\displaystyle \frac{\alpha x}{2\mathrm{\Delta }_x}}dx+{\displaystyle \frac{i}{\beta }}d\psi ,`$ (4.4)
$`\eta _3=\left(1{\displaystyle \frac{a^6}{R^6}}\right)^1{\displaystyle \frac{dR}{R}}+i\stackrel{~}{e}^5\eta _1(\alpha x)\mathrm{sin}^2\theta \eta _2(\beta x)\mathrm{cos}^2\theta .`$
It can be shown that the $`\eta _i`$’s indeed are closed and by construction $`(1,0)`$ forms. Using (4.4) we can solve for $`d\theta ,dx,d\varphi `$ and $`d\psi `$ in terms of $`\eta _{1,2}`$ and their complex conjugates. Then after substituting into (4) (and (4.3) for that matter) and some algebra we may show that both expressions indeed represent (1,1) forms.
Next we construct a (2,1) form as the wedge product of a (1,0) form and $`\omega `$
$$\mathrm{\Omega }_{2,1}=K\left[\left(1\frac{a^6}{R^6}\right)^1\frac{dR}{R}+i\stackrel{~}{e}^5\right]\omega ,$$
(4.5)
where K is a normalization constant. It is easily verified that the $`\mathrm{\Omega }_{2,1}`$ form is closed and imaginary self-dual in the six dimensional space, namely
$$d\mathrm{\Omega }_{2,1}=0,_6\mathrm{\Omega }_{2,1}=i\mathrm{\Omega }_{2,1}.$$
(4.6)
For the supergravity solution, we take the real RR $`F_3`$ and NSNS $`H_3`$ forms to be
$$iM\mathrm{\Omega }_{2,1}=F_3+\frac{i}{g_s}H_3$$
(4.7)
and therefore
$$F_3=MK\stackrel{~}{e}^5\omega ,H_3=g_sMK\left(1+\frac{C}{R^6}\right)^1\frac{dR}{R}\omega ,$$
(4.8)
where M is another normalization constant. The ansatz for the warped metric of the ten-dimensional type-IIB solution is
$$ds_{10}^2=H^{1/2}ds_4^2+H^{1/2}\left[\left(1\frac{a^6}{R^6}\right)^1dR^2+R^2ds_4^2+R^2\left(1\frac{a^6}{R^6}\right)(d\tau +\sigma )^2\right],$$
(4.9)
where the warp factor $`H`$ in generally depends on $`R`$, $`x`$ and $`\theta `$. There is no dilaton or axion field, while the self-dual five form is
$$g_sF_5=d(H^1)d^4x+_{10}[d(H^1)d^4x],$$
(4.10)
which after some algebra takes the form
$`g_sF_5`$ $`=`$ $`H^2\left({\displaystyle \frac{H}{R}}dR+{\displaystyle \frac{H}{x}}dx+{\displaystyle \frac{H}{\theta }}d\theta \right)d^4x`$
$`{\displaystyle \frac{H}{R}}R^5\left(1{\displaystyle \frac{a^6}{R^6}}\right){\displaystyle \frac{\mathrm{sin}2\theta }{4\alpha \beta }}\rho ^2d\tau d\theta dxd\varphi d\psi `$
$`{\displaystyle \frac{H}{x}}R^3{\displaystyle \frac{\mathrm{sin}2\theta }{\alpha \beta }}\mathrm{\Delta }_xd\tau dRd\theta d\varphi d\psi `$
$`+{\displaystyle \frac{H}{\theta }}R^3{\displaystyle \frac{\mathrm{sin}2\theta }{4\alpha \beta }}\mathrm{\Delta }_\theta d\tau dRdxd\varphi d\psi .`$
To determine the warped factor we substitute in the Bianchi identity
$$dF_5=H_3F_3$$
(4.12)
and obtain a second order partial differential equation whose precise form depends on which one of (4) or (4.3) we use to construct the 3-forms $`H_3`$ and $`F_3`$. If we use (4) in the Bianchi identity we obtain
$`{\displaystyle \frac{1}{R^3}}{\displaystyle \frac{}{R}}\left({\displaystyle \frac{H}{R}}R^5\left(1a^6/R^6\right)\right)+{\displaystyle \frac{4}{\rho ^2}}{\displaystyle \frac{}{x}}\left({\displaystyle \frac{H}{x}}\mathrm{\Delta }_x\right)`$
$`+{\displaystyle \frac{1/\rho ^2}{\mathrm{sin}2\theta }}{\displaystyle \frac{}{\theta }}\left({\displaystyle \frac{H}{\theta }}\mathrm{sin}2\theta \mathrm{\Delta }_\theta \right)=2{\displaystyle \frac{g_s^2M^2K^2}{\rho ^8R^4}}\left(1a^6/R^6\right)^1.`$ (4.13)
In the special case with $`\alpha =\beta `$ we can check that this equations indeed reduces to that in after we also make the consistent assumption that $`H`$ is $`\theta `$-independent. We were not able to find exact solutions of (4.13) in the generic case. In that respect note that it is not consistent to assume $`\theta `$-independence of the solutions. Perhaps the work of who study the Laplacian in the $`Y^{p,q}`$ spaces will be useful in that direction as well. Nevertheless, we may easily see that for large $`R`$ it exhibits the generic behaviour as $`H\mathrm{ln}R/R^4`$. Towards the infrared for $`Ra`$ it is seen that there is a singularity since $`H\mathrm{ln}^2(Ra)`$.
Perhaps the most important open issue concerns the construction of a supergravity solution utilizing the $`Y^{p,q}`$ and $`L^{p,q,r}`$ spaces and being dual to $`𝒩=1`$ gauge theories, in which the IR singularity is smoothened out. Let’s recall that some times a useful approach in constructing supersymmetric spaces representing cones with smoothened out singularities is via gauged supergravities. In particular, many such solutions having an $`SU(2)`$ isometry were found using the eight-dimensional supergravity of resulting from dimensionally reducing the eleven-dimensional supergravity of (see in particular the works -). The use of the lower dimensional gauged supergravity disentangles certain technical issues which are due to the complexity of the base manifolds (in our case see the expressions for the spin connections in section 3). We believe that at least for the case of solutions having the $`Y^{p,q}`$ manifold as an internal part the use of gauged supergravity could be proven quite useful.
Acknowledgments
We would like to thank C.N.Pope and D. Martelli for a correspondence and A. Paredes for a discussion. We acknowledge the financial support provided through the European Community’s program “Fundamental Forces and Symmetries of the Universe” with contract MRTN-CT-2004-005104, the INTAS contract 03-51-6346 “Strings, branes and higher-spin gauge fields”, as well as by the Greek Ministry of Education program $`\mathrm{\Pi }`$Y$`\mathrm{\Theta }`$A$`\mathrm{\Gamma }`$OPA$`\mathrm{\Sigma }`$ with contract 89194. In addition D.Z. acknowledges the financial support provided through the Research Committee of the University of Patras for a “K.Karatheodory” fellowship under contract number 3022. We also thank Ecole Polytechnique and CERN for hospitality and financial support during part of this work. |
warning/0507/cond-mat0507156.html | ar5iv | text | # Asymptotics of the dispersion interaction: analytic benchmarks for van der Waals energy functionals
(Original receipt date Feb 20 2005, revised MS July, August, Nov 05)
## Abstract
We show that the usual sum of $`R^6`$ contributions from elements separated by distance $`R`$ can give *qualitatively* wrong results for the electromagnetically non-retarded van der Waals interaction between non-overlapping bodies. This occurs for anisotropic nanostructures that have a zero electronic energy gap, such as metallic nanotubes or nanowires, and nano-layered systems including metals and graphene planes. In all these cases our analytic microscopic calculations give an interaction falling off with a power of separation different from the conventional value. We discuss implications for van der Waals energy functionals. The new nanotube interaction might be directly observable at sub-micron separations.
Dispersion interactions (part of the van der Waals, vdW energy)vdWBooks2 are especially significant in soft matter. The vdW physics that we expose here could be relevant in predicting the energetics of bundles of metallic nanowires or nanotubes, layered metallic systems, $`\pi `$conjugated systems including graphite, intercalated graphite, graphitic hydrogen storage systems and pi-stacked biomolecules, and other weakly bound (”soft”) layered and striated nanosystems. Standard local (LDA) and gradient (GGA) density functionals for the electronic energy do not obtain any distant dispersion interaction, but density functionals have been derived recently that obtain, in a natural fashion, both distant dispersion interactions and their saturation at small distances. These and other numerically practicable vdW energy schemes available to date vdWBooks2 ; LifshitzMolAttrFrcsSolsJETP56 ; UnivGraphiticPotlGirifalco ; HardMNosSoftMattRydbergEtal ; TractableNonlocVdWSlabs ; HardNosSoftMattRevisRydberg+ ; vdWFnalGenGeomDionPRL04 ; CHARMMPotls98 for the above systems (in the electromagnetically non-retarded regime) have a “universal” feature: the distant vdW interaction energy between sufficiently separated subsystems is given qualitatively by a sum of contributions of form $`R^6`$ between microscopic elements separated by distance $`R`$. This leads to “standard” power laws $`ED^p`$ for the interaction energy between various macroscopic bodies separated by distance $`D`$ (column 3 of Table 1). Although these ”universal” asymptotic results are indeed valid for most macroscopic systems, we show below that they fail for the anisotropic nanostructures mentioned above. Column 2 of Table 1 summarizes the asymptotic ($`D\mathrm{}`$) benchmarks that we propose below for the vdW energy of two parallel nanostructures of infinite extent.
To analyze these situations, we use the correlation energy $`E_c^{RPA}(D)`$ from the Random Phase Approximation (RPA)jfdvdw ; SurfEnPitarkeEguiluz ; RPAMolecsFurche ; RPAMolecsFuchsGonze02 ; DobsonWangPRL99 ; JeilJFDPabloRex04 , a basic microscopic theory that does not rely on assumptions of locality, additivity nor $`R^6`$ contributions. Going beyond the RPA does not change the asymptotic power laws predicted here, unless the exchange-correlation kernel $`f_{xc}`$ GrossKohnPRL85 ; EnOptFxc ; JeilJFDPabloRex04 has a slower spatial decay than the bare coulomb interaction, an unprecedented and unlikely scenario.
Where the separated subsystems exhibit lightly damped long-wavelength plasmons, we noteJFDIJQC04 that the principal contribution to $`E_c^{RPA}(D)`$ comes from the sum of coupled-plasmon zero-point energies: otherwise we use the full RPA. Some essential common features of these systems will be abstracted from these specific calculations. We obtain analytic results for the asymptotic $`(D\mathrm{})`$regime in all cases, but in section E we will also discuss systems near their equilibrium spacing.
*A: Distant attraction between metallic linear structures.* Consider two parallel, infinitely long conducting wires or tubes separated by a distance $`D`$ substantially exceeding their radius $`b`$, and with $`b<\lambda `$ where $`\lambda `$ is a bulk screening length. Both standard $`R^6`$analysisUnivGraphiticPotlGirifalco based on the vdW interaction between electrons localized in atoms or bonds, and recent functionals vdWInterParPolymersNanotubes+Hyldgaard+05 , give a vdW energy per unit length of the form $`ED^5`$. Instead we consider the zero-point energy of the delocalized coupled one-dimensional plasmon modes with wavenumber $`q`$parallel to the long axis CollectiveExcMetalNanotubesLongeBose93 ; JFDAWhiteUnpub .The radially-smeared intra-wire coulomb interaction is $`w_{11}(q)=w_{22}(q)=2e^2\mathrm{ln}(qb)`$ where 1 and 2 refer to the two wires, and we have assumed $`qb<<1`$, as appropriate when $`D>>b`$. In the same limit the bare density-density response for electronic motion parallel to the wire is $`\chi _{011}=\chi _{022}=N_0q^2/(m\omega ^2)`$ where $`\omega `$is the frequency, $`N_0`$ is the number of electrons per unit length and $`m`$ is the electron mass. RPA screening yields the interacting response of a single wire as $`\chi _{11}=\chi _{011}/(1w_{11}\chi _{011})`$. The inter-wire coulomb interaction in the present limit has a Bessel form, $`w_{12}=2e^2K_0(qD)`$. The RPA equation for coupled 1D plasmons on two identical wires is $`\chi _{11}^2w_{12}^2=1`$, giving two roots for each $`q`$: $`\omega _\pm (D)=c_{1D}\left|q\right|(\left|\mathrm{ln}(qb)\right|\pm K_0(qD))^{1/2}`$. Here $`c_{1D}=(2N_0e^2/m)^{1/2}`$ is a characteristic velocity. The vdW energy is the separation-dependent part of the sum of zero-point plasmon energies per unit length:
$$\frac{E^{vdW}}{L}=\frac{1}{2\pi }_{\mathrm{}}^{\mathrm{}}\frac{\mathrm{}}{2}\left(\omega _+(D)+\omega _{}(D)2\omega _+(\mathrm{})\right)𝑑q$$
For $`D>>b`$ we expanded to 2nd order in $`K_0(qD)/\left|ln(qb)\right|`$ which is small near the peak of the integrand. This gave JFDAWhiteUnpub approximately
$$E^{vdW}/L(16\pi )^1\mathrm{}c_{1D}D^2(\mathrm{ln}(2.39D/b))^{3/2}.$$
(1)
This approach is reasonable when the electron mean free path $`d_0`$ along the wire satisfies $`d_0>D`$. In fact bismuth nanowiresPosThermopowerBiNanowireGrozav04 and conducting nanotubes PhysPropsNanotubesSaitoDresselh can both have $`d_01`$ micron. Eq. (1) differs from the widely accepted result $`ED^5`$ by nearly three powers of $`D`$, and is necessarily dominant, at sufficiently large $`D,`$ over any such higher-power contributions (arising from the remaining bound sp<sup>2</sup> electrons (in nanotubes) and azimuthal $`\pi `$ plasmons). Our plasmon model does give $`D^5`$ if a pinning force is added to mimic an insulator JFDAWhiteUnpub .
*B: Distant attraction between thin conducting layers* Consider infinite parallel metallic plates separated by distance $`D`$ and of thickness $`b`$, with $`b<<D`$, $`b<\lambda `$ where $`\lambda `$ is the bulk screening length. As is already well-known FractvdWThinMetFilmsBostromSernelius ; JFDEtAlAustJChem02 , the zero point energy of long-wavelength coupled 2D plasmons leads to an attraction of form $`ED^{5/2}`$. The $`R^6`$ approach, correct for thin *insulators*, gives $`ED^4`$, different by 1.5 powers of $`D`$.
*C: Distant attraction between planar* $`\pi `$*-conjugated systems.* What does the physics of long-wavelength excitations imply for the energetics of layered planar $`\pi `$-conjugated systems, such as the controversial BenedictMeasGraphiteLayerAttr ; CohEnGraphiteThermDesorb+Hertel ; OrientationC60sTournusCharlier05 ; GraphiteBindingSemiEmpirHasegawa+04 and technologically important PhysPropsNanotubesSaitoDresselh ; HStorageCNanotubesGraphiticClustersSimonyan+ graphene-based systems? Firstly, an isolated graphene layer at T=0K is not a metal but a zero-gap insulatorPhysPropsNanotubesSaitoDresselh . Thus one cannot argue for a metallic $`D^{5/2}`$ energetics (as under (B) above) at large layer separation $`D`$ and $`T=0K`$, even though band overlap makes graphite weakly metallic at the equilibrium layer spacing. We briefly derive below, however, our new result that the attractive energy between two well-separated graphene planes at $`T=0K`$ is of form $`C_3D^3`$, closer to metallic $`D^{5/2}`$ behavior than to insulating $`D^4`$ behavior. All the new physics here comes from electrons close to the Fermi level: we can ignore the response of the tightly-bound covalent sp<sup>2</sup> electrons, whose finite energy gap ensures that they produce a conventional vdW attraction of 2D insulator type (energy $`D^4`$), negligible at large separations compared with the $`D^3`$ vdW attraction that we shall find between the $`\pi _z`$ electrons of interest here. The bonding and antibonding $`\pi `$ bands have a gapless bandstructurePhysPropsNanotubesSaitoDresselh . The energy near the K points where the bands touch is given by $`\epsilon ^{(1,2)}(\stackrel{}{p})=\mathrm{}v_0\left|\stackrel{}{p}\right|`$ where $`\stackrel{}{p}`$ is the 2D crystal momentum measured from a $`K`$ point, and $`v_0`$is a characteristic velocity (about $`5.7\times 10^5m/s`$ for graphene). From perturbation theory within a Wannier description NonUnivvdW ; JFDUnpubGeneral , the zero-temperature density-density response $`\chi _{KS}`$ of independent $`\pi _z`$electrons moving in the groundstate Kohn-Sham potential of a gapless $`\pi `$-layer is then of the form $`\chi _{KS}(\stackrel{}{q},\stackrel{}{0},\stackrel{}{0},z,z^{},\omega =iu)=S(q,z)S(q,z^{})^{}\overline{\chi }_0(\stackrel{}{q},iu),`$ with $`S𝑑z1`$as $`\stackrel{}{q}\stackrel{}{0}`$. We found NonUnivvdW ; JFDUnpubGeneral the effective 2D response $`\overline{\chi }_0`$ at small $`q`$ and imaginary frequency $`\omega =iu`$ to be
$$\overline{\chi }_0(\stackrel{}{q},iu)2\mathrm{}v_0q\left(1+u^2/(v_0q)^2\right)^{1/2},$$
(2)
consistent with previous real-$`\omega `$ resultsGrapheneRespMarginalFermiLiqu . Here we treat the response in each sheet as strictly two-dimensional, and ignore certain local-field effects, so that the only consequence of the periodic potential is to replace the 2D free-electron bare response $`n_0q^2/(mu^2)`$ by the zero-gap Bloch response (2). This is justified esewhere NonUnivvdW ; JFDUnpubGeneral . We consider electron density perturbations of form $`n_1\mathrm{exp}(i\stackrel{}{q}.\stackrel{}{r}+ut)`$ in layer #1, where $`\stackrel{}{q}`$and $`\stackrel{}{r}`$ are two-dimensional. Such charge disturbances interact via a Fourier transformed bare coulomb potential,
$$w_{11\lambda }(q)=2\pi \lambda e^2q^1,w_{12\lambda }(q)=2\pi \lambda e^2q^1\mathrm{exp}(qD),$$
(3)
for interactions within a layer and between two layers distant $`D`$, respectively. Then the RPA equation for the interacting density fluctuation in layer #1 as driven by an external potential $`v_1^{ext}\mathrm{exp}(i\stackrel{}{q}.\stackrel{}{r}+ut)`$ is of time-dependent mean-field form, $`n_1=\overline{\chi }_0(q,iu)\left(v_1^{ext}+w_{11}n_1\right)`$. This applies in the absence of layer #2 or equivalently for $`D\mathrm{}`$. Solving for $`n_1`$we find a single-layer density-density response
$$\chi _{11\lambda ,D\mathrm{}}n_1/v_1^{ext}=\overline{\chi }_0/(1w_{11\lambda }\overline{\chi }_0).$$
(4)
With two layers present, the density response obeys coupled RPA equations $`n_1=\chi _{11\lambda ,D\mathrm{}}(v_1^{ext}+w_{12\lambda }n_2)`$, $`n_2=\chi _{22\lambda ,D\mathrm{}}(v_2^{ext}+w_{21\lambda }n_1)`$. The solution is $`\stackrel{}{n}=\chi \stackrel{}{v}^{ext}`$ where $`\stackrel{}{n}=(n_1,n_2)^T`$ and similarly for $`\stackrel{}{v}^{ext},`$ while the components of the $`2\times 2`$ matrix $`\chi `$ are $`\chi _{11\lambda ,D}=\chi _{11\lambda ,D\mathrm{}}/(1w_{12\lambda }\chi _{11\lambda ,D\mathrm{}})`$ and $`\chi _{12\lambda ,D}=w_{12\lambda }\chi _{11\lambda ,D\mathrm{}}\chi _{11\lambda ,D}`$. For the case of two identical layers considered here, the other elements are $`\chi _{22\lambda ,D}=\chi _{11\lambda ,D}`$ and $`\chi _{21\lambda ,D}=\chi _{12\lambda ,D}`$.
In the present system the response (4) of a single layer, continued to the real frequency axis, yields no lightly damped plasmons (poles) for small $`q`$, so that a sum of plasmon zero-point energies cannot be used to evaluate the vdW interaction. Instead we consider the electromagnetically non-retarded groundstate electronic correlation energy, which for a general inhomogeneous electronic system is given exactly by the adiabatic connection fluctuation-dissipation theorem (see e.g. JFDIJQC04 ):
$$E_c=\frac{\mathrm{}}{2\pi }_0^1𝑑\lambda 𝑑\stackrel{}{r}𝑑\stackrel{}{r}^{}\frac{e^2}{\left|\stackrel{}{r}\stackrel{}{r}^{}\right|}_0^{\mathrm{}}\mathrm{\Delta }\chi _\lambda (\stackrel{}{r},\stackrel{}{r}^{},iu)𝑑u.$$
(5)
Here $`\mathrm{\Delta }\chi _\lambda =\chi _\lambda \chi _0`$, where $`\chi _\lambda `$ is the electron density-density response function at reduced coulomb interaction $`\lambda e^2/\left|\stackrel{}{r}\stackrel{}{r}^{}\right|`$. Applying (5) to the present layer geometry and Fourier transforming parallel to the layers we find that the separation-dependent part $`E^{vdW}/A`$of the energy per unit area is:
$`{\displaystyle \frac{E_c(D)E_c(\mathrm{})}{A}}={\displaystyle \frac{\mathrm{}}{\pi }}{\displaystyle _0^{\mathrm{}}}𝑑u{\displaystyle _0^1}{\displaystyle \frac{d\lambda }{\lambda }}{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{2\pi qdq}{(2\pi )^2}}`$ (6)
$`\times `$ $`\left(w_{11\lambda }\left(\chi _{11\lambda D}\chi _{11\lambda ,D\mathrm{}}\right)+w_{12\lambda }\chi _{12\lambda D}\right).`$
Within the RPA approximation, Eqs. (2) and (3), plus (4) and the equations following it, show that each term of form $`w\chi `$in (6) depends on $`u`$ solely through the dimensionless combination $`x=u/(v_0q)`$. The remaining dependence of $`w\chi `$ on $`q`$ is solely via $`y=qD.`$ Thus (6) has a scaling form
$`E^{vdW}/A`$ $`=`$ $`\mathrm{}{\displaystyle _0^1}{\displaystyle \frac{d\lambda }{\lambda }}{\displaystyle _0^{\mathrm{}}}q𝑑q{\displaystyle _0^{\mathrm{}}}𝑑uG(\lambda ,{\displaystyle \frac{u}{v_0q}},qD)`$ (7)
$`=`$ $`{\displaystyle \frac{\mathrm{}v_0}{D^3}}{\displaystyle _0^1}{\displaystyle \frac{d\lambda }{\lambda }}{\displaystyle _0^{\mathrm{}}}y^2𝑑y{\displaystyle _0^{\mathrm{}}}𝑑xG(\lambda ,x,y)`$
where $`G(\lambda ,x,y)`$is independent of $`D`$. We numerically evaluated the dimensionless 3D integral in (7) for graphene parametersNonUnivvdW ; JFDUnpubGeneral , giving the interaction energy per unit area in Gaussian esu units:
$$E^{vdW}/A=7.745_7\times 10^2\mathrm{}v_0D^3=2.003_6\times 10^2e^2D^3$$
(8)
This $`D^3`$form shows that the gapless $`\pi `$-conjugated planes behave in this respect more like metals ($`ED^{5/2}`$) than insulators ($`ED^4`$) , despite the lack of undamped 2D plasmon modes on a single $`\pi `$ sheet.
*D. Parallel metallic and* $`\pi `$*-conjugated planes*
Another interesting case is the interaction between a $`\pi `$-conjugated layer and a metallic 2D layer with fermi energy $`\epsilon _F`$ (e.g. an undoped and a doped graphene sheet). For $`D>>D_0=\mathrm{}^2v_0^2/(2\pi e^2\epsilon _F)`$ (=$`O(1nm)`$ for $`\epsilon _F=O(0.02eV)`$ as in a bulk graphite layer) the methods described above give an energy per unit area (c.f. (8))
$$E^{vdW}/ACe^2D^3\mathrm{ln}(D/D_0)(C\text{constant}).$$
(9)
As in the case of two non-metallic gapless $`\pi `$ layers, the result (9) disagrees with standard theories.
*E. Interaction energy near overlap* We now discuss possible difficulties with the non-asymptotic, near-equilibrium energetics of the present systems, especially graphenes. The commonest ab initio approach, the LDA, misses distant dispersion interactions entirely KohnMeirMakarovVdWPRL98 , and yet gives a good lattice spacing CharlierGonzeMichLDABSGraphite91 and good breathing phonon frequencies GrPhonons in graphite, (unlike GGAs DFTShortRangeBeyondRPAYanPrdwKrth ; HardMNosSoftMattRydbergEtal ). Recent experiments BenedictMeasGraphiteLayerAttr ; CohEnGraphiteThermDesorb+Hertel , however, lead one to suspectGraphiteBindingSemiEmpirHasegawa+04 that the LDA pays for its neglect of dispersion physics by severely underestimating the equilibrium binding energy of graphite. LDA also underbinds related fullerene systemsOrientationC60sTournusCharlier05 . This phenomenon has been investigated in layered jellium analogs, via fully nonlocal many-body correlation theory (JFDEVsDJellSlabs04 , Fig. 4 of JeilJFDPabloRex04 ). It was found that either layer-layer forces or binding energy have serious errors near equilibrium, when distant dispersion forces are underestimated. The underestimation is related in turn to the lack of distant correlated fluctuations, especially those oriented parallel to the layers. Thus these low-q fluctuations can have effects even near the equilibrium spacing. Addition of explicit $`R^6`$ vdW terms has been a common remedy for stretched graphitic systems UnivGraphiticPotlGirifalco ; GraphiteBindingSemiEmpirHasegawa+04 , and recently several seamless vdW schemes have been proposed DobsonWangPRL99 ; HardNosSoftMattRevisRydberg+ ; vdWFnalGenGeomDionPRL04 , based on approximations for response functions. Refs HardNosSoftMattRevisRydberg+ and vdWFnalGenGeomDionPRL04 are the most practical, and are qualitatively successful in graphitics. vdWFnalGenGeomDionPRL04 overestimates the binding energy of small systems but correspondingly obtains a large binding energy of two graphene layers (more than twice that from HardNosSoftMattRevisRydberg+ or from LDA, and consistent with experiment). Ideally a single theory should give reliable results for small and extended systems. Could it be that the key is a correct treatment of the fluctuations parallel to a long axis in the extended cases, the same fluctuations responsible for the unusual asymptotics exposed here that is absent in HardNosSoftMattRevisRydberg+ ; vdWFnalGenGeomDionPRL04 ? These fluctuations are of course dominant only at large separations but they might not be negligible near the equilibrium spacing, where all wavelengths can contribute. We speculate further that the same physics might apply in other large finite $`\pi `$-conjugated systems (e.g. planar melanin layers, carotenes, fullerenes OrientationC60sTournusCharlier05 ) where, as the system size increases, the electronic gap diminishes while longer-wavelength excitations become possible.
*Summary and Discussion.* Our new results (see (1),(8),(9) and Table 1) show that that usual sum of $`C_6R^6`$ terms incorrectly predicts the dependence of the dispersion energy on separation $`D`$ for a range of systems. Simple energy functionals presently available all have standard $`C_6R^6`$asymptotics. A finite sum of multipole, or triplet and higher terms will also not reproduce what we have discussed. The standard asymptotics fails when the component systems (i) are metallic (or have a zero electronic Bloch bandgap), and (ii) are spatially extented in at least one dimension, so that long-wavelength (low-$`q`$) charge fluctuations can occur, and (iii) are of nanoscopic dimensions in another spatial direction, so that the electron-electron screening is reduced compared with 3D bulk metallic systems, leaving a divergent screened polarizability at low frequency and wavenumber. (Thick metal slabs, for example, violate (iii): they have complete screening and exhibit a conventional power law, $`ED^2`$. See e.g. JFDEtAlAustJChem02 ). Where free low-$`q`$ plasmons exist, conditions (i) - (iii) imply that they will be gapless. The same conditions ensure that the usual spatially local approximation for the dielectric function LifshitzMolAttrFrcsSolsJETP56 is invalid. Our results provide unequivocal asymptotic benchmarks that are not satisfied by existing simplified van der Waals energy formulae, because they do not treat in enough detail the fluctuations along the extended space dimension. In Section E we have further motivated the possibility that the same fluctuation physics may be relevant in the systems considered here, even near their equilibrium spacing. Investigation of this question requires a seamless energy formalism that is fully nonlocal - e.g. RPA-like theories JFDEtAlAustJChem02 ; DobsonWangPRL99 ; JeilJFDPabloRex04 ; JFDIJQC04 . Such calculations are only now becoming possible for 3D systems RPAGreenFnEnergysolidsMiyake+PRB2002 ; BoronNitrideInRPA+MariniGarciaRubioPreprint05 , with no converged results available to date for the present zero-gap cases. Simplified vdW energy functionals are therefore certainly needed for routine modelling, and the above considerations suggest that existing functionals may need further refinement to take explicit account of large-scale geometry and/or nonlocal entities such as electronic bandgapJFDIJQC04 . We note finally that our work predicts novel differences in the forces between conducting and nonconducting nanotubes or wires, that might be directly measurable for low-index nanotubes at sub-micron separationsJFDAWhiteUnpub , and that could even affect self-assembly processses. These considerations might also affect the analysis of some seminal experimentsBenedictMeasGraphiteLayerAttr ; CohEnGraphiteThermDesorb+Hertel concerning graphitic cohesion, because these relied at some point on theory involving a sum of $`R^6`$ contributions. *Acknowledgments*. We thank I. D’Amico, P. Garcia-Gonzalez, J. Jung, L. Reining, E. Gray and P. Meredith for discussions. JFD acknowledges support from Australian Research Council grant DP0343926, UPV/EHU, Ecole Polytechnique and CNRS, and the hospitality of AR and Dr. L. Reining. AR was supported by the Network of Excellence NANOQUANTA (NMP4-CT-2004-500198), and a 2005 Bessel research award of the Humbolt Foundation. |
warning/0507/astro-ph0507653.html | ar5iv | text | # H I Observations of SA 68-6597: the faintest Blue Compact Dwarf Galaxy.
## 1. Introduction
Blue Compact Dwarfs (BCDs) are faint (M$`{}_{B}{}^{}`$-17 mag, e.g. Kong & Cheng, 2002), compact (diameters of the high surface brightness regions of less than 1 kpc, e.g. Thuan & Martin, 1981), and blue enough to suggest active star formation (e.g. Gordon & Gottesman, 1981; Thuan & Martin, 1981). They are typically low metallicity systems (Izotov et al., 1999). Two of the most extreme BCDs, in terms of luminosity, mass, and metallicity, are I Zw 18 and SBS 0335-052. I Zw 18 and SBS 0335-052 have luminosities of -12.8 mag and -14.3 mag, total masses of 10<sup>8.5-9.5</sup>$`M_{}`$, HI masses of 10<sup>7.8-9.3</sup>$`M_{}`$ (van Zee et al., 1998; Pustilnik et al., 2001), and oxygen abundances of the ionized gas of $`12+log(O/H)=`$ 7.17 & 7.34 (Izotov et al., 1999)–the lowest known in the universe. Because of these properties, I Zw 18 and SBS 0335-052 are believed to be undergoing early bursts of star formation (Izotov & Thuan, 2004; Lipovetsky et al., 1999). While much more luminous, BCD-like, HII galaxies at moderate redshift may evolve into galaxies like NGC 205 (Koo et al., 1994, 1995; Guzmán et al., 1996), these low luminosity, low mass BCDs may be the progenitors of dwarf spheroidal galaxies like Carina.
SA 68-6597 was discovered serendipitously during the first DEEP<sup>1</sup><sup>1</sup>1Deep Extragalactic Evolutionary Probe: see URL http://deep.ucolick.org run using the Keck LRIS instrument(Oke et al., 1995) with a 1200 l/mm grating (Koo et al. 2005, in preparation). This galaxy was selected because of its blue color and visually estimated compact non-stellar appearance and faint apparent magnitude. The LRIS data show that SA 68-6597 has a redshift of $`z`$=0.0186 implying that it is intrinsically extremely faint. Assuming a Hubble constant of 70 km s<sup>-1</sup> Mpc<sup>-1</sup>, SA 68-6597 is located at a distance of 80 Mpc and has a B magnitude of -12.4 mag as measured by the DEEP.team. Using the combination of the O III$`[\lambda 5007]`$/H$`\beta `$ and the N II$`[\lambda 6583]`$/H$`\alpha `$ line ratios as measured with HIRES, Koo et al. find that SA 68-6597 has an extremely low excitation temperature, $``$14,500 K, and metallicity $`12+log(O/H)`$7.4 ($``$0.05 Z). This places SA 68-6597 well away from the well-defined locus of H II galaxies (e.g. Lee et al., 2004) and Local Group dwarf irregulars (Mateo, 1998) in this parameter space. The most similar galaxy to SA 68-6597 in this space is the BCD SBS 0335-052. Follow-up observations (Koo et al. 2005, in preparation) with Keck HIRES (Vogt et al., 1994) also show that the emission lines have Gaussian velocity dispersions of $``$ 27 km/s. HST WFPC2 images reveal a very small galaxy, R<sub>25</sub> of 1.0$`\pm _{0.05}^{0.1}`$″= 400$`\pm _{20}^{40}`$ pc. Combined with the small optical linewidth, this implies that SA 68-6597 is a very low mass galaxy, $``$10<sup>7</sup>$`M_{}`$. Based on the flux of the H$`\alpha `$ line in the LRIS spectra, the star formation rate of this galaxy, 0.003 $`M_{}`$ yr<sup>-1</sup>, is similar to what is expected for a BCD given its inferred low total mass (Hopkins et al., 2002). These properties from Koo et al. (2005, in preparation) and summarized in Table 1, strongly suggest that SA 68-6597 is a blue compact dwarf that is fainter and smaller than other BCDs (e.g. Thuan & Martin, 1981; Salzer et al., 2002). It is a fainter, lower-mass, but slightly higher metallicity counterpart to the more famous BCDs I Zw 18 and SBS 0335-052. It is the faintest known BCD (Koo et al. 2005, in preparation). Its extreme nature makes this galaxy a particularly interesting probe of low mass galaxy formation (see Pustilnik et al., 2001, and references therein for a discussion).
Observations of the 21-cm line of neutral hydrogen (H I) can help us better understand the nature of SA 68-6597. The H I content of SA 68-6597 is an important measure of its potential for future star formation. The current burst of star formation is small in an absolute sense, 0.003 $`M_{}`$ yr<sup>-1</sup>, but if the H I mass is also low then it can still rapidly consume its H I and subsequently passively evolve, fade and possibly become a galaxy like the Carina dwarf spheroidal in the Local Group. If the H I mass is much higher, then it is more likely to continue forming stars for a long time and will retain its current appearance. Furthermore, while the optical emission lines have a width of 27 km s<sup>-1</sup>, this linewidth may not trace the entire gravitational potential of the galaxy. The H I gas tends to trace the gravitational potential to larger radii than the stars or ionized gas. In addition, ionized gas may be tracing galactic outflows; this is less likely for the neutral gas. For all of these reasons, H I provides the best measure of the total, dynamical mass of a galaxy. The dynamical mass is an important constraint on the evolution of a galaxy. If SA 68-6597 has a low total mass, then it may eject its neutral gas before it can consume it in star formation (e.g. Mac Low & Ferrara, 1999). If it is higher, then it may be too massive to evolve into a dwarf spheroidal galaxy. In this paper, we report on our Arecibo observations of H I in SA 68-6597. These observations help constrain the potential for future star formation in the galaxy and provide a more robust measure of the total mass of the galaxy constraining the current nature and future evolution of SA 68-6597, and its relation to other BCDs like I Zw 18.
## 2. Arecibo H I Observations & Reductions
We observed SA 68-6597 with the Arecibo<sup>2</sup><sup>2</sup>2The Arecibo Observatory is part of the National Astronomy and Ionosphere Center which is operated by Cornell University under a Cooperative Agreement with the National Science Foundation. 305 m telescope on 2004 August 1–4 and October 3–4. We observed only at night to minimize solar interference. We used the L-wide receiver for all our observations. This receiver has a system temperature of $``$27 K, and a gain of $``$10 K Jy<sup>-1</sup> as measured by the Arecibo staff. Both values are weakly dependent on the zenith angle of the observation. Data were processed through the interim correlator in both linear polarizations over total bandwidths of 25 MHz and 12.5 MHz, corresponding to a velocity range of $``$5000 km s<sup>-1</sup> and $``$2500 km s<sup>-1</sup>. Each bandwidth and polarization had 9-level sampling and 2048 channels, resulting in a velocity resolution per channel of 5.2 km s<sup>-1</sup> and 2.6 km s<sup>-1</sup>. Our observations utilized a high pass filter to block interference below 1370 MHz contaminating our band. The beam size of the L-wide receiver according to the Arecibo documentation is 3.1$`\mathrm{}\times `$ 3.5$`\mathrm{}`$. At the distance of SA 68-6597, 80 Mpc, this corresponds to a linear size of 72 kpc$`\times `$ 81 kpc. As the effective radius of the stellar emission is only 190 pc, this should be more than sufficient to encompass all of the H I associated with this galaxy. Yet this beam size is small enough that there are no known galaxies at a similar redshift that can contaminate our H I measurements; the closest galaxy is $``$400 kpc away and there are no known groups within 3 Mpc based on a NED<sup>3</sup><sup>3</sup>3The NASA/IPAC Extragalactic Database (NED) is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration. search. It is further unlikely that there are any H I-rich, optically invisible galaxies exist that could contaminate our measurements (Doyle et al., 2005).
We used the standard position switching algorithm for our observations spending 5 minutes on SA 68-6597 followed by 5 minutes offset to blank sky by 5 minutes in right ascension such that we tracked the same azimuth and zenith angle as the on-source scan. This was repeated for all six nights for an on-source integration time of 240 minutes. Our data were reduced using standard Arecibo IDL routines written by Phil Perrilat. Each bandwidth of each scan was calibrated separately and the polarizations were averaged together before a first or second order baseline was fit across the portion of the spectrum clean of any interference. All scans were then averaged together to produce the final spectrum. While some scans showed signatures of interference around 1400 MHz, since SA 68-6597 is at a frequency of 1394 MHz this should not affect our ability to detect the galaxy. The bandpass, however, was significantly better for the 12.5 MHz band as a result of the RFI at 1400 MHz, and so we proceeded only with this band. Because the two bands are split after the first amplification, their noise is not independent and, therefore, it would not have helped to combine these bands. The resulting noise was 0.22 mJy per 2.6 km s<sup>-1</sup> channel. For analysis, we binned this spectrum by 4 channels to a resolution of 10.8 km s<sup>-1</sup>, improving the noise to 0.10 mJy per channel. The observational details are listed in Table 1. Our flux measurements of a bright galaxy, UGC 199, are within 17% of previously published values (Schneider et al., 1990). This is a much smaller source of error than that of random noise and can be disregarded for this work.
## 3. Results
Figure 1 shows the binned H I spectrum of SA 68-6597 near the known optical velocity of the galaxy. The small inset in the lower left corner shows the entire spectrum (excluding the edges of the bandpass). The H I emission is very weak, but clearly detected. The line has a peak flux of 0.31 mJy located at a velocity of 5552 km s<sup>-1</sup>. This is only a 3$`\sigma `$ detection, but it is the brightest feature in the spectrum and is located within $``$40 km s<sup>-1</sup> of the optical velocity of SA 68-6597; which is within the 1$`\sigma `$ uncertainties of the optical redshift. To check its reality, we split the raw data into various subsets (e.g. by sets of days, polarization, etc.) and searched for the line in these data. The line was either visible in the subsets or its absence was consistent with the noise levels; thus we believe that it is a real emission line from SA 68-6597.
The vertical dotted lines in Figure 1 indicate the region over which we measured the H I properties of SA 68-6597. The integrated flux is 0.0095$`\pm `$0.0025 Jy km s<sup>-1</sup>, which translates to a $`M_{HI}`$ of $`(1.4\pm 0.4)\times `$10<sup>7</sup>$`M_{}`$–a 3.8$`\sigma `$ detection over five channels. We know of no galaxies within the 3.5$`\mathrm{}`$ beam of Arecibo that may be contaminating this measurement, so we believe that this H I is truly associated with SA 68-6597. Combining these H I data with the optical properties, we find a $`M_{HI}`$-to-$`L_B`$ ratio of 1.0$`\pm `$0.3$`M_{}`$/$`L_{}`$. The gas depletion timescale, $`\tau `$=$`M_{HI}`$/SFR, is 5$`\pm `$2 Gyr without accounting for helium, molecular gas, or recycling.
The H I linewidth at 50% of the peak flux, W<sub>50</sub> (FWHM), is measured to be 32$`\pm `$10 km s<sup>-1</sup>, centered at 5557$`\pm `$5 km s<sup>-1</sup>. Again, this is within the uncertainties of the optical recession velocity. Because of the low signal-to-noise ratio of the detection, this value is highly imprecise and potentially inaccurate. To address this issue, we have used a Monte Carlo simulation of a Gaussian line with a peak signal-to-noise ratio of 3$`\sigma `$ to relate the measured FWHM to the true FWHM. We find that the the true H I FWHM = 33$`\pm _{12}^{60}`$ km s<sup>-1</sup>. Converting this to a W<sub>20</sub> yields 51$`\pm _{19}^{93}`$ km s<sup>-1</sup>, assuming a Gaussian lineshape.
One of the main goals of our project is to determine the dynamical mass, $`M_{dyn}`$, of SA 68-6597 using the H I line. As discussed in Section 1, the H I line is generally believed to be a more reliable tracer of the gravitational potential than the H$`\beta `$ line. Because of the large uncertainties associated with our H I measurement, and the additional uncertainties from the unknown inclination of SA 68-6597, we are practically restricted to calculating a lower limit to $`M_{dyn}`$. We follow the same procedure to derive the dynamical mass as in Pisano et al. (2001) using the following standard formula assuming the H I is in circular rotation: $`M_{dyn}`$$`(<R)V_{rot}^2\times R/G`$.
In this case we take V<sub>rot</sub> to be half of the lower limit on W<sub>20</sub> uncorrected for inclination, for the radius we scale R<sub>25</sub> using a canonical scaling factor from Broeils & Rhee (1997) to get R<sub>HI</sub>=680$`\pm _{200}^{210}`$ pc. If we use these values to calculate a lower limit to $`M_{dyn}`$, we find it is greater than 3.0$`\times `$10<sup>7</sup>$`M_{}`$. This yields $`M_{HI}`$/ M$`{}_{dyn}{}^{}`$ 0.47, and $`M_{dyn}`$/$`L_B`$$``$2. See Pisano et al. (2001) for a discussion of the uncertainties involved in this calculation. All of these measured and derived properties of SA 68-6597 are summarized in Table 1.
## 4. Discussion
Our H I observations have revealed that SA 68-6597 is not only a low luminosity galaxy, but also has a low $`M_{HI}`$, and probably a low $`M_{dyn}`$ as well. By all three measures, SA 68-6597 reveals itself to be an extreme cousin of other BCDs. BCDs typically have $`M_{HI}`$$``$10<sup>8-9</sup>$`M_{}`$, with only a few as low as 10<sup>7</sup>$`M_{}`$ or as high as 10<sup>10</sup>$`M_{}`$. They have $`M_{dyn}`$$``$10<sup>8-10</sup>$`M_{}`$, and $`L_B`$$``$10<sup>8-10</sup>$`L_{}`$ (Chamaraux, 1977; Thuan & Martin, 1981; Hoffman et al., 1989; Staveley-Smith et al., 1992; Thuan et al., 1999; Salzer et al., 2002). SA 68-6597 represents the extreme low mass, low luminosity end of BCDs and is not a distinctly different class of galaxy as generally evidenced by its scale-free properties, such as the H I-mass-to-light ratio, $`M_{HI}`$/$`L_B`$, and the gas-mass fraction, $`M_{HI}`$/$`M_{dyn}`$.
SA 68-6597 has an $`M_{HI}`$/$`L_B`$ ratio that is consistent with that of the large samples of BCDs studied by Staveley-Smith et al. (1992); van Zee et al. (1998, 2001); Salzer et al. (2002); Hoffman et al. (2003); Thuan et al. (2004) who found ratios ranging from $``$0.33-1.46 $`M_{}`$/$`L_{}`$. SA 68-6597’s H I mass-to-light ratio is even within the range for luminous compact blue galaxies studied by Garland et al. (2004), but is about twice the median value. Only the study of Hoffman et al. (1989) found a significantly lower $`M_{HI}`$/$`L_B`$ value for 11 Virgo cluster BCDs of 0.04 $`M_{}`$/$`L_{}`$. The $`M_{HI}`$/$`M_{dyn}`$ values for BCDs are also generally consistent with SA 68-6597’s upper limit of 0.47. A variety of studies of BCDS find $`M_{HI}`$/$`M_{dyn}`$ ratios ranging from 0.01-0.78 (Hoffman et al., 1989; van Zee et al., 1998; Hoffman et al., 2003; Thuan et al., 2004). The ratio of dynamical mass-to-light, $`M_{dyn}`$/$`L_B`$, for other BCDs ranges from 0.18-2.62 $`M_{}`$/$`L_{}`$ (Hoffman et al., 1989; Staveley-Smith et al., 1992; Thuan et al., 2004), which is also consistent with the lower limit of 2 $`M_{}`$/$`L_{}`$ for SA 68-6597. It’s SFR and $`M_{HI}`$ are consistent with expectations for BCDs based on SA 68-6597’s H I linewidth (Hopkins et al., 2002). Even luminous compact blue galaxies have a median $`M_{dyn}`$/$`L_B`$ only slightly higher (5 $`M_{}`$/$`L_{}`$) than that of SA 68-6597(Garland et al., 2004). All of these ratios suggest that SA 68-6597 is an extremely faint, extremely low-mass version of a typical blue compact dwarf.
In terms of individual BCDs, SA 68-6597 is quite similar in its gaseous properties to I Zw 18 and Haro 4. I Zw 18 is still slightly more massive and more luminous with a $`M_{HI}`$ = 2.6$`\times `$10<sup>7</sup>$`M_{}`$, $`M_{dyn}`$=2.6$`\times `$10<sup>8</sup>$`M_{}`$, and $`L_B`$ = 3.5$`\times `$10<sup>7</sup>$`L_{}`$ (van Zee et al., 1998). Nevertheless, with $`M_{HI}`$/$`L_B`$ = 0.7 and $`M_{dyn}`$/$`L_B`$ = 5 I Zw 18 has mass-to-light ratios nearly identical to SA 68-6597. Its gas-mass fraction of 0.1 is also similar to SA 68-6597. Haro 4 is slightly less similar with $`M_{HI}`$ = 2$`\times `$10<sup>7</sup>$`M_{}`$, $`M_{HI}`$/$`L_B`$ = 0.17$`M_{}`$/$`L_{}`$, $`M_{dyn}`$=5$`\times `$10<sup>8</sup>$`M_{}`$, $`M_{dyn}`$/$`L_B`$ = 4.8$`M_{}`$/$`L_{}`$, and $`M_{HI}`$/$`M_{dyn}`$=0.03 (Bravo-Alfaro et al., 2004). While $`M_{HI}`$, $`M_{dyn}`$, and $`M_{dyn}`$/$`L_B`$ of Haro 4 are close to those of SA 68-6597, the $`M_{HI}`$/$`L_B`$ is lower, and the gas-mass fraction is lower than SA 68-6597, but still consistent with it.
The question is then raised: “what do these properties say about the evolutionary path of SA 68-6597?” Our derived gas depletion timescale, $`\tau `$, for SA 68-6597 is 5$`\pm `$2 Gyr. This value provides a rough measure of the time it will take for SA 68-6597 to consume all of its gas at its current rate of star formation (Kennicutt, 1983). This is similar to many of the measured values for a sample of 21 BCDs from Hopkins et al. (2002), but is much greater than a sample of 15 BCDs studied by Sage et al. (1992). It has a gas depletion timescale equal to the median value for the sample of field and cluster galaxies studied by Kennicutt (1983). SA 68-6597 has a larger $`\tau `$ than either I Zw 18 or SBS 0355-052 (Hopkins et al., 2002). It has a shorter $`\tau `$ than all Local Group dwarf irregular galaxies except IC 10 and NGC 6822 (Mateo, 1998). Our estimate does not account for the contribution of helium, recycling of gas, molecular gas, a decreasing SFR or less than 100% star formation efficiency–all of which would increase $`\tau `$–or the possible effects of outflow–which would decrease $`\tau `$. Mac Low & Ferrara (1999) suggest that galaxies with gas masses below 10<sup>6</sup>$`M_{}`$ can suffer complete blowout of their gas, while galaxies between 10<sup>6</sup> and 10<sup>7</sup>$`M_{}`$ may suffer partial blowout. Because the derived gas mass of SA 68-6597 is $``$10<sup>7</sup>$`M_{}`$, we expect that it could only have a partial outflow (Mac Low & Ferrara, 1999). The results of Mac Low & Ferrara (1999) are based on a dark matter halo approximately 10-100$`\times `$ larger than the gas mass. Overall, this means that, SA 68-6597 should evolve in a similar fashion to many less extreme BCDs and normal spiral and irregular galaxies; it will not rapidly consume its gas and passively evolve in the near future. The large $`M_{dyn}`$ of SA 68-6597 implies that if and when it consumes all its gas, it may be able to evolve into a moderate mass analog of the Local Group dwarf spheroidals (Mateo, 1998).
While the low signal-to-noise ratio of our observations make our velocity widths very uncertain, it is worth noting a potentially interesting property of SA 68-6597. The H$`\beta `$ velocity dispersion of SA 68-6597 is 27$`\pm _2^1`$ km s<sup>-1</sup>, while the H I dispersion is 14$`\pm _5^{25}`$ km s<sup>-1</sup>. If the H I linewidth is actually smaller than that of the ionized gas, then we may be seeing evidence of a galactic outflow in SA 68-6597. Such outflows could result in partial blowout of the ISM in the galaxy of anywhere between $``$1% to 100% (Mac Low & Ferrara, 1999) potentially reducing the gas depletion time. Outflows can also be particularly efficient at ejecting metals into the intergalactic medium (Mac Low & Ferrara, 1999; Ferrara & Tolstoy, 2000) permitting a galaxy like SA 68-6597 to form many generations of stars while maintaining its very low metallicity. Clearly more sensitive and detailed H I observations are needed to address this issue.
## 5. Conclusions
We have observed the recently discovered blue compact dwarf galaxy, SA 68-6597, with the Arecibo telescope to determine its gas content and better constrain its total mass using the H I line. SA 68-6597 is one of the faintest, lowest metallicity BCDs known.
SA 68-6597 has properties which indicate it is a typical blue compact dwarf galaxy in all ways, except for its extremely low luminosity and small H I and dynamical masses. In this way it represents the faint, low mass tail of the distribution of BCD properties. It is slightly fainter and less massive than the famous BCD I Zw 18 and only slightly more metal rich. SA 68-6597’s gas depletion timescale is similar to the value for other BCDs and normal field galaxies, yet is shorter than most dwarf irregulars in the Local Group. Nevertheless, SA 68-6597 can continue to form stars at its current, prolific rate for almost 5 Gyr and would, therefore, be unlikely to fade significantly in that time. When it does fade, its relatively large dynamical mass suggests it may be able to evolve into a massive dwarf spheroidal galaxy. Its future evolutionary path remains murky.
Because of the combination of the distance and low H I mass of SA 68-6597, our detection was only at the 3$`\sigma `$ level, meaning that the measured linewidth and derived dynamical mass are poorly constrained. Nevertheless, the most probable value of the H I linewidth is less than the measured H$`\beta `$ linewidth indicating that galactic outflows may be present in SA 68-6597. Because of the potential implications of such a situation on SA 68-6597’s evolution, more sensitive, spatially resolved H I observations of SA 68-6597 are essential to unravel its current nature and reveal its evolutionary path.
This research has made use of the NASA/IPAC Extragalactic Database (NED) which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration. The authors wish to thank the staff at Arecibo, especially Tapasi Ghosh, for their help with our observations. We also wish to thank the Arecibo observatory for granting us more observing time to improve the quality of our observations. We thank Eric Wilcots for assistance with the observing and helpful discussions. This research was performed while D.J.P. held a National Research Council Research Associateship Award at the Naval Research Laboratory. Basic research in astronomy at the Naval Research Laboratory is funded by the Office of Naval Research. D.J.P. also acknowledges generous support from the ATNF via a Bolton Fellowship and from NSF MPS Distinguished International Research Fellowship grant AST 0104439. D.C.K. acknowledges support from NSF AST-0071198 and HST GO-07339.01-96A. |
warning/0507/hep-ph0507036.html | ar5iv | text | # Higgs Localization in Split Fermion Models
## I Introduction
One of the indications for the incompleteness of the Standard Model (SM) is the hierarchy among its flavor parameters. An attractive solution for this flavor puzzle is presented in split fermion models AS , where the fermion zero modes are split over an extra dimension. The effective 4D Yukawa couplings are then suppressed by exponentially small overlaps between wavefunctions of different fermion zero modes. In some of these models MS ; ASmodels ; GP the fermions are localized at various positions in $`x_5`$, while in others KT , they are attached to 4D planes, with an exponential penetration into the bulk. A key ingredient in all these models is a real scalar field (the localizer) which acquires a bulk dependent VEV. More examples, discussions and experimental signatures of split fermion models can be found in PDG ; GGH ; GaussLocH ; LocH ; AGS .
The Higgs VEV is usually assumed either uniform MS ; GP ; GGH ; KT , or confined to a brane KT ; GaussLocH ; LocH . While in some works DS ; KT there are ideas about generating its profile, no comprehensive study was done. In order to motivate a certain profile, one has to study the coupling between the Higgs and the localizer, which is the purpose of this work. We work at the classical level, where we were able to realize the above two scenarios. A uniform Higgs is obtained if the above coupling is small, which requires fine tuning of the model parameters. Conversely, a scenario in which the Higgs is localized at a brane is obtained without fine tuning. We discuss an explicit solution of this sort in the large orbifold limit, and its phenomenological constraints. Finally, we show that there are other solutions which are not fined tuned and are phenomenologically viable.
## II Split Fermions
### II.1 Overview
Our interest lies in split fermion models with one extra dimension. While there is an extensive typology of such models, a feature common to many of them is the appearance of a real scalar ‘$`\mathrm{\Phi }`$’ called the localizer, which is a SM singlet. The remaining field content is similar to that of the SM, but with Dirac fermions (there are no chiral representations in five dimensions). By coupling to the fermions, the localizer VEV serves as a position dependent mass which vanishes at the 3-brane where the fermion is localized. In order to see that, we write the relevant part of the Lagrangian as
$$=\overline{\mathrm{\Psi }}_i\left(i\delta _{ij}\mathrm{\Gamma }^M_M\frac{\lambda _{ij}}{\sqrt{M_{}}}\mathrm{\Phi }M_{ij}\right)\mathrm{\Psi }_j,$$
(1)
where $`\mathrm{\Gamma }^M`$ are the five Dirac matrices, $`M_{}`$ is the UV cutoff so that the $`\lambda _{ij}`$ are dimensionless, and $`i,j`$ are flavor indices. As a first step we discuss an infinite extra dimension, where we denote the extra dimension coordinate as ‘$`z`$’. A further simplification is to take $`[\lambda ,M]=0`$. Then we can work in the mass basis where both $`\lambda `$ and $`M`$ are diagonal (this assumption is relaxed in twist ).
We need $`\mathrm{\Phi }(z)`$ to be a topological defect with co-dimension $`D4`$ in order to confine the other fields to a 4D universe. In an infinite extra dimension we assume the kink solution,
$$\mathrm{\Phi }(z)=\frac{\mu }{\sqrt{\lambda _\varphi /M_{}}}\mathrm{tanh}\left(\frac{\mu z}{\sqrt{2}}\right),$$
(2)
where $`\mu `$ is the 5D mass parameter of the localizer and $`\lambda _\varphi `$ is its dimensionless quartic coupling. The equations of motion (EOM) for the Kaluza-Klein (KK) wavefunctions of the fermions are obtained by solving the 5D EOM with separation of variables. In particular, the resulting wavefunctions for the zero modes are
$$f_{L,R}^{i(0)}(z)=N_{L,R}\mathrm{exp}\left(_0^z\left[\lambda _i\mathrm{\Phi }(z^{})+M_i\right]𝑑z^{}\right),$$
(3)
where $`M_i(\lambda _i)`$ are the eigenvalues of $`M(\lambda )`$. We can see that only one of each wavefunction pair is normalizable, depending on the sign of $`\lambda _i`$. In order to get rid of the mirror fermions in a realistic model, a common solution is to compactify the extra dimension on a $`S^1/Z_2`$ chiral orbifold GGH ; KT ; GP . The orbifold boundary conditions force right handed fermions to be odd, which is incompatible with Eq.(3), and therefore the right handed zero mode is projected out of the spectrum.
Having localized the SM chiral fields at various points in $`x_5`$, we now turn to show how small 4D Yukawa couplings arise naturally in this setup. For example, in the Arkani-Hamed–Schmaltz (AS) model, the relevant part of the Lagrangian for the leptons is
$`_{5\mathrm{D}}`$ $`=`$ $`\overline{L}_i(i/_5M_L{\displaystyle \frac{\lambda }{\sqrt{M_{}}}}\mathrm{\Phi })L_i+\overline{E}_i(i/_5M_E+{\displaystyle \frac{\lambda }{\sqrt{M_{}}}}\mathrm{\Phi })E_i`$ (4)
$``$ $`({\displaystyle \frac{Y_{ij}}{\sqrt{M_{}}}}\overline{L}_iHE_j+c.c.),`$
where $`Y_{ij}`$ is the 5D Yukawa couplings with the Higgs. Upon KK decomposition, we obtain the effective low-energy 4D Lagrangian,
$$_{4\mathrm{D}}=\left(\frac{Y_{ij}}{\sqrt{M_{}}}𝑑zf_i^{\mathrm{}}(z)\left[f^h(z)h(x)+v_H(z)\right]f_j^e(z)\right)\overline{\mathrm{}}_L^i(x)e_R^j(x)+c.c.,$$
(5)
where the zero modes are denoted in small Latin letters, and $`f^i(z)`$ are their wavefunctions. Note that in the KK decomposition of the Higgs field we distinguish between the VEV and the wavefunction of the lowest mode, which we call the ground KK mode.
As already mentioned, in realistic models the extra dimension is compactified on a $`S^1/Z_2`$ chiral orbifold, where the localizer and the right handed fermions are odd under the $`Z_2`$ reflection. The Higgs and the left handed fermions are even. Since 5D Dirac masses are forbidden by these boundary conditions, the fermions cannot be split up in the usual way, but a scenario with fermions in the bulk can still be constructed KT ; GP . Another approach is presented in KT , where the localizer VEV has a narrow domain wall, effectively a step function, and the Higgs VEV is confined to one of the fixed points. The fermion wavefunctions in this model are localized at each of the fixed points, with an exponential decay toward the other fixed point. The sign of the coupling to the localizer determines the localization point, and its magnitude fixes their width. More specifically, in this model the third generation quark doublet and the top singlet are localized at the Higgs brane whereas the other quarks are localized at the other fixed point. The 4D effective Yukawa is then proportional to the value of the wavefunction at the Higgs brane, thus giving a large Yukawa only for the top Yukawa, since both the top and $`Q_3`$ are maximal at the Higgs brane. In the next section we obtain such scenario as a specific classical solution.
In a milder, “regularized” version of the above, the Higgs VEV is rather localized, and not a delta function. A non-uniform Higgs VEV is possible as long as the mass of the $`W^\pm `$ ground mode is predicted correctly, namely,
$$g_4^2v_{\mathrm{EW}}^2=\frac{g_5^2}{M_{}}𝑑z\left|h(z)f_W(z)\right|^2,$$
(6)
where $`g_D`$ is the $`D`$ dimensional SU(2) gauge coupling, $`h(z)`$ is the 5D Higgs VEV, and $`f_W(z)`$ is the wavefunction of the $`W^\pm `$ ground mode. The relation between the 4D and 5D gauge couplings is given by
$$g_4=\frac{g_5}{\sqrt{M_{}}}_0^L𝑑zf_W(z)\left|f_\psi (z)\right|^2\frac{g_5}{\sqrt{M_{}L}},$$
(7)
provided that the $`W^\pm `$ wavefunction is nearly flat. Therefore the gauge couplings drop out and the phenomenological constraint Eq.(6) reads
$$v_{\mathrm{EW}}^2=_0^L𝑑zh^2(z).$$
(8)
This condition is necessary in order to comply with phenomenology.
We treat our model as an effective field theory, and therefore the presence of nonrenormalizable terms does not pose an essential problem. Throughout this work we consider operators which translate into renormalizable terms in the effective four dimensional theory, such as $`\varphi \overline{\psi }\psi `$ (dimension $`5\frac{1}{2}`$ in the 5D theory), or $`\varphi ^4`$ (dimension 6 in the 5D theory). Higher dimension operators are suppressed by appropriate powers of $`p/M_{}`$ and therefore can be neglected in the low energy limit, where the theory is effectively four dimensional.
Note that here we only give a classical treatment of split fermions, following the literature (see for example AS ; MS ; KT ). We do not discuss quantum corrections, and in particular anomalies in extra dimensions. Discussions concerning such issues can be found, for example, in anom .
### II.2 Tree Level FCNC
In SM extensions there can be various sources for tree-level Flavor Changing Neutral Currents (FCNCs). Such interactions can be mediated by the $`Z^0`$, by the Higgs or by new bosons. In split fermion models, every KK mode of the neutral gauge bosons or the Higgs can mediate FCNC. In particular, the KK modes of the gauge fields have $`x_5`$-dependent wavefunctions and therefore their couplings to the fermion zero modes are flavor dependent. The resulting FCNCs provide a bound on the size of the extra dimension KT . Here we are concerned with Higgs mediated FCNCs. In general, a scalar can mediate FCNC if its Yukawa term is not aligned with the fermion mass term. As an example we can think of the Higgs fields in multi Higgs models. In split fermion models the case is similar, although there is only one Higgs. This can be explained from a 5D or from a 4D point of view.
From the 5D point of view, the Higgs field is expanded about its VEV as
$$H(x,z)=v_H(z)+\stackrel{~}{H}(x,z).$$
(9)
The shifted field $`\stackrel{~}{H}`$ has vanishing VEV, and we decompose it into KK modes,
$$\stackrel{~}{H}(x,z)=\underset{n}{}h_n(x)f_n(z).$$
(10)
Retaining only the ground KK mode ($`n=0`$) we see that while the effective 4D Yukawa couplings are given by
$$y_{ij}=Y_{ij}𝑑z\frac{f^h(z)f^i(z)f^j(z)}{\sqrt{M_{}}},$$
(11)
the fermion mass term induced by the Higgs VEV is
$$m_{ij}=Y_{ij}𝑑z\frac{v(z)f^i(z)f^j(z)}{\sqrt{M_{}}},$$
(12)
leading to $`y_{ij}\propto ̸m_{ij}`$. This misalignment between the 4D Yukawa and mass matrices implies FCNC at the Lagrangian level.
In order to see the above from a 4D point of view, we distribute the non uniform Higgs VEV among the different KK modes of the Higgs. That is, we write
$$H(x,z)=\underset{n}{}\left[v_n+h_n(x)\right]f_n(z)$$
(13)
with
$$v_H(z)\underset{n}{}v_nf_n(z).$$
(14)
Such 4D theory contains many KK fields ($`h_n`$), each with its own VEV ($`v_n`$). The 4D mass term is given by
$$m_{ij}=Y_{ij}\underset{n}{}\frac{v_n}{\sqrt{M_{}}}𝑑zf_n^h(z)f^i(z)f^j(z),$$
(15)
while the Yukawa coupling to the $`n`$-th Higgs KK is
$$y_{ij}^n=\frac{Y_{ij}}{\sqrt{M_{}}}𝑑zf_n^h(z)f^i(z)f^j(z).$$
(16)
The resulting Lagrangian is that of a multi Higgs model, but without natural flavor conservation to prevent FCNC.
## III The Scalar Sector
Split fermions, then, provide a mechanism to localize the fermion zero modes. A side effect, however, is that a similar mechanism can (and therefore does) apply for the SM Higgs, as the latter couples to the localizer and inevitably gets localized. In order to find classical solutions for the Higgs VEV, we should study the scalar sector, which includes the SM Higgs and the localizer. More specifically, we are interested in the case of two scalars in one spatial dimension. In this work, for simplicity we replace the SM Higgs (four degrees of freedom before electroweak symmetry breaking) with a real field (one degree of freedom). We start with the potential
$$U(\varphi ,h)=\frac{1}{2}\mu ^2\varphi ^2+\frac{1}{4}\lambda \varphi ^4\frac{1}{2}\mu _h^2h^2+\frac{1}{4}\lambda _hh^4+\frac{1}{2}g\varphi ^2h^2.$$
(17)
The application to the five dimensional model involves adding the appropriate powers of the 5D cutoff scale $`M_{}`$. For example, $`\lambda ,\lambda _h,g`$ are couplings with mass dimension $`(1)`$. For now we work in natural units where $`M_{}=1`$.
Starting with the simplistic AS model, where the extra dimension is infinite, we know that the $`g0`$ limit leads to the uniform Higgs solution,
$$\varphi (z)=\frac{\mu }{\sqrt{\lambda }}\mathrm{tanh}\left(\frac{\mu z}{\sqrt{2}}\right);h(z)\frac{\mu _h}{\sqrt{\lambda _h}}.$$
(18)
We also keep in mind that we seek solutions in which the localizer $`\varphi (z)`$ is antisymmetric and the Higgs $`h(z)`$ is symmetric in order to match the orbifold boundary conditions upon compactification.
### III.1 A Uniform Higgs
Split fermion models require that there is a hierarchy between the localizer and the Higgs scales, in order that the fermion KK modes would not acquire $`𝒪(m_{\mathrm{EW}})`$ masses. The bound on the $`v_{\mathrm{EW}}/\mu `$ ratio depends on the details of the model, but characteristic values are roughly $`v_{\mathrm{EW}}/\mu (0.1\mathrm{TeV}/100\mathrm{TeV})`$ KT . Note that this is a direct bound on $`\mu `$, unlike the bound related to gauge KK modes, which constrains the size of the extra dimension. The $`v_{\mathrm{EW}}\mu `$ hierarchy suffers from fine tuning, since both scales get radiative corrections proportional to the UV cutoff. We also mention that the observation of universality in the weak interactions puts a bound on $`m_{EW}/\mu `$ which is of similar order.
In uniform Higgs scenarios, there is another fine tuning problem, related to the generating of the coupling $`g`$ in $`g\varphi ^2H^{}H`$. This operator is corrected even in the $`g_{\mathrm{tree}}0`$ limit. In one loop the only diagrams which contribute are those with fermions running in the loop (see Fig. 1). Since such diagrams depend strongly on the UV cutoff, the coupling $`g_{\mathrm{tree}}`$ must be fine tuned in order to cancel the radiative corrections. Note that models with uniform Higgs require the above fine tuning in addition to the $`\mu _h\mu `$ related fine tuning, which must be assumed in any split fermion model. In the rest of this work we do not consider quantum corrections, namely, we work exclusively at the classical level.
### III.2 A perturbative/adiabatic approximation
Going back to the classical problem, the equations of motion,
$$\varphi ^{\prime \prime }=\frac{U(\varphi ,h)}{\varphi };h^{\prime \prime }=\frac{U(\varphi ,h)}{h},$$
(19)
are coupled and cannot be integrated in a straightforward manner. However, a particularly simple scenario is obtained when $`gh(z)^2\mu ^2`$, that is, when the Higgs is affected by the localizer but not vice versa. This limit is realized by the condition
$$\frac{g}{\lambda _h}\mu _h^2\mu ^2.$$
(20)
Then we can approximate a solution as follows: The localizer ($`\varphi `$) is assumed to be the kink,
$$\varphi (z)=\frac{\mu }{\sqrt{\lambda }}\mathrm{tanh}\left(\frac{\mu z}{\sqrt{2}}\right),$$
(21)
and we wish to find the Higgs VEV. The Higgs potential is then given by
$$U(h)=\frac{1}{2}M^2(z)h^2+\frac{\lambda _h}{4}h^4,$$
(22)
where
$$M^2(z)\mu _h^2+\frac{g\mu ^2}{\lambda }\mathrm{tanh}^2\left(\frac{\mu z}{\sqrt{2}}\right)$$
(23)
stands for the squared bulk mass of the Higgs. The equation of motion for the static Higgs VEV is then
$$h^{\prime \prime }=\lambda _hh^3+M^2(z)h.$$
(24)
We expect the solution for $`h(z)`$ to get perturbative corrections of order $`𝒪\left(v_H/v_\varphi \right)`$. Unfortunately the zeroth order is already hard to solve. A further approximation is to neglect the left hand side of Eq.(24). In this “adiabatic” approximation, only in the regions where $`M^2(z)<0`$, the Higgs develops a VEV which is simply
$$h(z)=\sqrt{\frac{M^2(z)}{\lambda _h}}\left[1+𝒪\left(\frac{v_H}{v_\varphi }\right)+𝒪\left(\frac{h^{\prime \prime }(z)}{\mu _h^2}\right)\right].$$
(25)
Note that in this approximation $`h^{\prime \prime }(z)`$ diverges where $`M^2(z)=0`$. At these regions we expect large deviations. At other regions, $`h^{\prime \prime }(z)`$ is not very large.
With the above result, in which we neglected the curvature, we recognize four scenarios, depending on the parameters (See Fig. 2). Among these scenarios, we can identify one where the Higgs is localized at the domain wall. An exact solution of this form is obtained below, using a mechanical analogy coleman ; rajaraman . Before going on to the mechanical analogy for two fields, we recall the simpler case of one real scalar.
### III.3 One Scalar field
We recall that given a scalar potential $`U(\varphi )`$, the problem of finding static solutions depending only on one spatial coordinate, is equivalent to that of a non-relativistic particle coleman in the 1D potential $`V=U`$. The particle coordinate is $`\varphi `$, and the “time” is $`x_5`$. Considering the infinite dimension case (analogous to infinite time duration in the mechanical analogy), the field must approach two adjacent global minima at the boundaries (See Fig. 3).
If $`U`$ has at least two global minima, there exist non-trivial solutions interpolating between two adjacent global minima. For example, the potential
$$U(\varphi )=\frac{1}{4}\lambda \left(\varphi ^2\mu ^2/\lambda \right)^2,$$
(26)
has two global minima: $`\varphi _0=\pm \mu /\sqrt{\lambda }`$, and thus the possible solutions are the kink and the anti-kink:
$$\varphi (z)=\pm \frac{\mu }{\sqrt{\lambda }}\mathrm{tanh}\frac{\mu z}{\sqrt{2}}.$$
(27)
Proceeding to the more realistic case where the extra dimension is compact, it is obvious that the kink solution is incompatible with plain periodical boundary conditions. A common solution to this problem is to impose $`S^1/Z_2`$ orbifold boundary conditions, with the localizer odd under the $`Z_2`$. Note that the orbifold $`Z_2`$ symmetry of the Lagrangian is reflected in the fact that the solutions are either even or odd under the $`Z_2`$. In the large-$`L`$ limit the localizer VEV can be approximated by the kink-antikink (KAK) ansatz,
$$\varphi (z)=\varphi _k(z)\varphi _k(Lz);H(z)=\mathrm{const}.=v_H,$$
(28)
which appears in numerous models GP ; GGH . This is understood by considering the localizer field $`\varphi `$ with the potential (26). The orbifold implications on the mechanical analogy are that we look for a periodic motion with period $`2L`$ and with the analogue particle at the origin (though not at rest) in the start and the end points of the period (see Fig. 4a). An explicit expression for this KAK-like solution is given by
$$zz_0=\pm _{\varphi (z_0)}^{\varphi (z)}\frac{d\varphi }{\sqrt{2\left[U(\varphi )U(\varphi _{\mathrm{max}})\right]}}.$$
(29)
Unlike Eq. (27), this integral does not have a nice algebraic form for our potential GT . The relation between the orbifold size and the amplitude is given by
$$L=\sqrt{2}_0^{\varphi _{\mathrm{max}}}\frac{d\varphi }{\sqrt{U(\varphi )U(\varphi _{\mathrm{max}})}},$$
(30)
with a numerical evaluation depicted in Fig. 4b. For arbitrarily large-$`L`$ we can always find an appropriate motion whose amplitude is arbitrarily close to the maximum of the potential energy $`\varphi _{\mathrm{max}}=\mu /\sqrt{\lambda }`$. However, the period cannot be less than the small oscillation limit $`2\pi /\sqrt{U^{\prime \prime }(0)}`$. Thus there is a critical orbifold size $`L_c=\pi /\mu `$ which is the minimal one for a nontrivial solution. For $`L<L_c`$ the only solution is the trivial one, $`\varphi (z)0`$.
The intuitive picture is that there is a tension between the “natural” VEV $`\pm \mu /\sqrt{\lambda }`$ and the orbifold boundary conditions which set the odd field to zero at the fixed points. If we compare the total energy of the two configurations - the KAK-like vs. the identically-zero one, the total energy receives two kinds of contributions, the first comes from the potential difference while the second is the “shear” energy contributed from gradients in $`z`$. Unlike the shear contribution, the potential contribution is proportional to the bulk extent in which it resides, so we expect that when $`L`$ becomes small enough, the kinetic contribution takes over and the trivial solution becomes more economical. We also mention that the trivial configuration, $`\varphi (z)0`$, is always a classical solution, possibly not a minimum of the action. However, in the quantum theory such solution would tunnel to the true minimum which is the nontrivial solution if it exists.
### III.4 Two Scalar Fields
In approaching the two scalar case, we find that a notation similar to rajaraman can be useful. In this notation, we rewrite the potential as
$$U(\varphi ,h)=\frac{1}{4}\lambda \left(\varphi ^2u^2\right)^2+\frac{1}{2}k^2h^2+\frac{1}{4}\lambda _hh^4+\frac{1}{2}gh^2\left(\varphi ^2u^2\right),$$
(31)
where $`\lambda ,\lambda _h,g`$ have dimension $`1`$, $`u`$ has dimension $`3/2`$, and $`\lambda ,\lambda _h,u^2>0`$. Comparing to Eq.(17) we find the new parameters to be
$$u^2\frac{\mu ^2}{\lambda }\text{and}k^2\frac{g}{\lambda }\mu ^2\mu _h^2,$$
(32)
in the original notation.
#### III.4.1 Infinite extra dimension
The mechanical analogy for two scalar fields involves one particle in a two dimensional potential. A soliton is described by a zero-energy classical orbit starting and ending at global maxima. There are two types of such orbits: non-topological orbits start and end in the same global minimum while topological orbits connect two distinct global minima. We focus on the latter type since the former one is non stable, being in the same topological class as the trivial solution. In order to classify the solitons, we should study the configuration of the critical points in the potential. We distinguish between four configurations of the critical points (see Fig. 5). These four “types” of potentials are related to different regions in the parameter space $`(\mu _h,\mu ,\lambda _\varphi ,\lambda _h,g)`$.
* Type 0: For this type of potentials, $`\mu _h^2<0,g\mu ^2>\lambda \mu _h^2`$. This pattern has one maximum point at the origin and two global minima at $`(0,0),(\pm u,0)`$.
* Type-I: Another pattern occurs if
$$0<\mu _h^2<\frac{g\mu ^2}{\lambda }\text{and}\left(\frac{\mu _h}{\mu }\right)^2<\frac{\lambda _h}{g}.$$
(33)
* Type-II: An even richer pattern of critical points is achieved when the last inequality is reversed,
$$0<\mu _h^2<\frac{g\mu ^2}{\lambda }\text{and}\left(\frac{\mu _h}{\mu }\right)^2>\frac{\lambda _h}{g}.$$
(34)
* Type-III: If the first inequality in Type-I is inverted, we get a pattern of four global minima, one in each quadrant.
According to the mechanical analogy, type-0 does not yield a non trivial solution for both fields. Types I-II are the ones for which we obtain exact solutions. For type-III we conjecture qualitative characteristics of the solution without proof.
Unlike the single field case, here one must guess an orbit in the $`(\varphi ,h)`$ plane. If the guess is successful, an explicit solution as function of $`x_5`$ can be obtained. However, even if our guess is successful, the solution might be conditioned by constraints on the potential parameters. The full scheme is explained in the appendix. In Fig. 6, an exact solution is depicted, along with its corresponding mechanical orbit. This solution is given explicitly by
$$\varphi (z)=u\mathrm{tanh}[k(zz_0)];h(z)=\pm \sqrt{\frac{\lambda u^22k^2}{g}}\mathrm{sech}[k(zz_0)],$$
(35)
with
$$k=\sqrt{\frac{g}{\lambda }\mu ^2\mu _h^2}.$$
(36)
As can be seen in Fig. 6, in this solution the localizer acquires a kink-like profile and the Higgs VEV is bell-shaped, in accordance to the perturbative approximation. We note that the above solution has a constraint on the potential parameters, which is
$$\frac{\lambda _h}{\lambda }>\left(\frac{\mu _h}{\mu }\right)^4;\left(\frac{\mu }{\mu _h}\right)^2=\frac{2\lambda (g\lambda _h)}{g^22g\lambda _h+\lambda \lambda _h}.$$
(37)
That is, the above nice algebraic solution is valid only with this constraint. However, the mechanical analogy teaches us that similar solutions exist in the neighborhood of Eq.(37), although they might not have closed algebraic forms. With a suitable choice of parameters, the above solution can serve as a realization of the localized Higgs scenario. In fact, as can be seen from Eq.(36), the $`\mu _h/\mu `$ hierarchy makes sure that the above solution is tightly localized, provided that $`g/\lambda `$ is not too small. Another condition is the integral constraint Eq.(6). Putting the solution (35) into that constraint, we get
$$_0^L𝑑x_5h^2(x_5)=\frac{2}{k}\frac{\lambda u^22k^2}{g}=v_{\mathrm{EW}}^2,$$
(38)
or
$$v_{\mathrm{EW}}^2=\frac{2\mu ^2\left(12\frac{g}{\lambda }\right)+2\mu _h^2}{g\sqrt{\frac{g}{\lambda }\mu ^2\mu _h^2}}.$$
(39)
In all the solutions above, $`h(z)`$ maintains the same sign. This feature is expected to be violated in some of the solutions for type-III potentials which have four global minima at
$$(\varphi ,h)=(\pm \sqrt{\frac{g\mu _h^2\lambda _h\mu ^2}{g^2\lambda \lambda _h}},\pm \sqrt{\frac{g\mu ^2\lambda \mu _h^2}{g^2\lambda \lambda _h}}).$$
(40)
In this scenario the orbits which are relevant for our purposes are those connecting between two adjacent maxima with the same value of $`h`$. This is in order to match the orbifold boundary conditions. There are two kinds of conceivable orbits with such a feature, which are illustrated as “A” and “B” in Fig. (7a). In solution “A” the Higgs is always positive (or negative), realizing one of the “adiabatic” solutions depicted in Fig. 2, while in solution “B” it becomes negative in the vicinity of $`z=0`$. In both orbits the Localizer has a kink-like shape. At this point we could not obtain an explicit form of neither solutions, which may even not exist for the relevant region in the potential parameter space.
Regarding the conjectured solution “B”, two points are worth noting: First, this solution cannot appear in the adiabatic solution since it involves large gradients in $`h(z)`$, which is in contrast with the “adiabatic” assumption. Second, unlike the neutral $`h(z)`$ in our simplified discussion, the SM Higgs is charged under $`SU(2)\times U(1)`$, and thus the above picture of “sign alternating” should be replaced by one where the SU(2) phase rotates along the extra dimension.
#### III.4.2 Orbifold
While in an infinite extra dimension we can obtain an explicit solution, the orbifold case is much harder to solve. Nevertheless some interesting conjectures can be made. Again we seek oscillatory solutions with period $`2L`$. We start by discussing the large-$`L`$ case. In the case of infinite extra dimension, the analogue particle departs from $`(\mu /\lambda ,0)`$ and arrives at $`(+\mu /\lambda ,0)`$ in a zero energy path (see Fig. 8, dashed line). In a large but finite sized orbifold we expect closed (periodic) orbits with negative energy rather than zero energy. While we cannot prove nor verify the existence of such orbits, below we suggest the main features of such solutions if they exist. In an analogy to the single-field case, the particle starts at the fixed point with $`\varphi (0)=0;h(0)\sqrt{\alpha }u`$, with a non zero “velocity” ($`\dot{\varphi }0;\dot{h}=0`$). The particle barely misses the maximum at $`(+1,0)`$, then it proceeds to the opposite fixed point in a similar way. The motion in the “$`\varphi <0`$” plane is completely dictated by the orbifold $`Z_2`$ symmetry (that is, in a perfect reflection of the first half of the motion). The period of the above motion amounts to the complete circumference of the compact dimension (See Fig. 8). The orbifold symmetry requirement is invariance under reflection about the $`h`$-axis. In our case the symmetry of the potential further implies that the orbit is also invariant under reflection about the $`\varphi `$-axis. In the case of the $`SU(2)\times U(1)`$ Higgs, the orbifold condition implies that the orbit is invariant under reflection about the Higgs hyperplane.
We do not know if such an orbit (or a continuum of orbits) exists, but we do know that if some orbit intersects the axes with right angles $`\left(\dot{h}(0)=0,\dot{\varphi }(L/2)=0\right)`$ then it is periodic because of symmetry considerations. Here too, as in the case of one scalar field, there is a critical orbifold size, under which the only solution is
$$\varphi (z)=0;h(z)=\pm \sqrt{\frac{gu^2k^2}{\lambda _h}}=\pm \frac{\mu _h}{\sqrt{\lambda _h}},$$
(41)
since it always has a lower action than the trivial solution $`(\varphi ,h)=(0,0)`$. This solution is incompatible with split fermion models. For small but finite orbifold size, Eq. (41) cannot be perturbed with small oscillations, since this point is not a minimum of $`U(\varphi ,h)`$. Small oscillations can be found only about the origin, but since the problem is equivalent to an anisotropic harmonic oscillator, the only relevant solutions exist only when the ratio $`\mu _h/\mu `$ is rational, which requires fine tuning of the mass parameters as well as of the orbifold size. More specifically, elliptic and circular orbits exist only if $`\mu =\mu _h`$. Another possibility regarding type-III potentials is depicted in Fig. 9. This hypothetical solution is the orbifold version of the solutions “A” and “B”. In this case the Localizer is again KAK-like.
## IV A more generic scenario
As discussed before, the deviation of $`v_H(z)`$ from flatness is proportional to $`g`$, the coupling of $`\varphi ^2H^{}H`$. Such deviations lead to Tree-Level FCNC, and therefore we can translate the bounds from FCNC experimental data into a bound on $`g`$. In order to do this, first we estimate the deviation using the perturbativity condition Eq.(20). In this limit it turns out that $`f^h(z)v_H(z)`$ even if these functions are not uniform. This fact is demonstrated using the 5D point of view (see section II.2). We substitute the KK expansion,
$$H(x,z)=v_H(z)+\underset{n}{}h_n(x)f_n(z),$$
(42)
into the 5D equation of motion. Separation of variables for $`f_n(z)`$ yields
$$f_n_\mu ^\mu hhf^{\prime \prime }+(\mu _h^2+g\varphi ^2)hf+\lambda (v_H+hf)^3=0.$$
(43)
After linearization in $`hf`$ and separation, we obtain
$$f^{\prime \prime }+(\mu _h^2+g\varphi ^2+3\lambda v_H^2)f=0,$$
(44)
which, up to the scaling $`v_H(z)v_H(z)/\sqrt{3}`$ and upon substituting $`f=v_H`$, is similar to the equation for the VEV,
$$v_H^{\prime \prime }+(\mu _h^2+g\varphi ^2+\lambda v_H^2)v_H=0.$$
(45)
This approximation holds as long as the localizer is not affected by the Higgs VEV, namely when $`gv_H^2(z)\mu ^2`$, and thus we expect that
$$f(z)v_H(z)+𝒪\left(\frac{g\mu _h^2}{\lambda _h\mu ^2}\right).$$
(46)
This means that if the hierarchy $`\mu _h^2/\mu ^2`$ is resolved, the coupling $`g`$ does not have to be very small in order to suppress FCNC. In particular, we are interested in tree-level processes where the Higgs KK mediates flavor transitions. For example, such an effective operator contributing to $`K\overline{K}`$ mixing is e.g.:
$$\left(\frac{g\mu _h^2}{\lambda _h\mu ^2}\right)^2Y_{ij}^2\frac{s\overline{d}s\overline{d}}{m_{\mathrm{KK}}^2},$$
(47)
where the Dirac structure is suppressed. From experimental data of $`K\overline{K}`$ mixing and CP violation in Kaon decay, the suppression scale of a $`s\overline{d}s\overline{d}/\mathrm{\Lambda }^2`$ term is bounded by $`\mathrm{\Lambda }>10^4`$ TeV LP03 . Following the rationale of some common flavor models (see e.g.LNS ), we take $`Y_{sd}m_s/m_t(\mathrm{sin}\theta _c)^5`$. Furthermore, assuming $`\lambda _h1`$ one finds that there is no relevant bound on $`g`$. Similar arguments hold for the $`D^0`$ and $`B^0`$ systems. We conclude that configurations in which the Higgs VEV is neither uniform nor confined to a brane do not impose further constraints on the model parameters.
## V Conclusions
We discussed the implications of the Higgs coupling to the localizer in split fermion models. A scenario such as the Arkani-Hamed–Schmaltz model, where the Higgs VEV is uniform, requires this coupling to be small, implying fine tuning of the model parameters. We found an exact classical solution for the case of an infinite extra dimension, by applying a mechanical analogy rajaraman . This solution, which is not fine tuned, provides a realization of scenarios where the Higgs is confined to a brane KT . We also discussed qualitatively the more realistic case of a compact extra dimension. Furthermore, we showed that more generic configurations of the Higgs VEV are phenomenologically viable. The apparently dangerous Higgs mediated FCNCs are suppressed already for $`g𝒪(1)`$, since they are proportional to $`\mu _h^2/\mu ^2`$, which is already assumed small in any split fermion model KT .
Many assumptions were made in constructing the simplified model of AS . By now, most of these assumptions have been carefully studied and showed to be, indeed, only simplifying ones. The unrealistic infinite extra dimension is not needed when an orbifold is used GGH ; KT ; GP . The assumption that the coupling to the localizer and the bare mass term can be diagonalized simultaneously, was relaxed in twist . In this work we tested the assumption that the Higgs is flat. We found that this assumption too is not a crucial one, providing further reinforcement to the split fermion idea.
###### Acknowledgements.
I am grateful to Yuval Grossman, Oleg Khasanov and Gilad Perez for their extensive contribution to this work. I am also indebted to Andrey Katz, Israel Klich, Yossi Nir, Martin Schmaltz, Yael Shadmi and Tomer Volansky for their help throughout this work and for many fruitful discussions.
## Appendix A Exact Solutions
In this appendix we give some details of obtaining the exact solution (35) and other solutions. Here we follow a line similar to Rajaraman rajaraman .
### A.1 Orbits in the Mechanical Analogy
A first integration of the equations of motion (19) yields two coupled ordinary differential equations,
$$\frac{1}{2}\varphi ^2=\frac{U(\varphi ,h)}{\varphi }g\varphi +A;\frac{1}{2}h^2=\frac{U(\varphi ,h)}{h}gh+B,$$
(48)
compared with the one scalar case where there is one equation only. Here a primed field denotes its derivative with respect to $`z`$, and $`A,B`$ are integration constants.
A solution may be obtained as follows. First, we guess an equation for the mechanical orbit:
$$g(\varphi ,h)=0.$$
(49)
Differentiating both sides of (49) with respect to $`z`$ and squaring, yields
$$\left(\frac{g}{\varphi }\right)^2\varphi ^2=\left(\frac{g}{h}\right)^2h^2.$$
(50)
Inserting (48), we obtain
$$\left(\frac{g}{\varphi }\right)^2\left(_{\mathrm{orbit}}\frac{U(\varphi ,h)}{\varphi }g\varphi \right)=\left(\frac{g}{h}\right)^2\left(_{\mathrm{orbit}}\frac{U(\varphi ,h)}{h}gh\right),$$
(51)
where the integrals are evaluated along the orbit (49). Eq. (51) imposes relations among the parameters in (49) and those in the potential. Thus in general we must not expect that only the orbit parameters are constrained, while those of the potential remain intact, unless our guess of Eq. (49) is exceptionally successful.
The most obvious orbit connecting the two vacua is the straight line from $`(\varphi ,h)=(u,0)`$ to $`(+u,0)`$. A somewhat more complicated orbit could be the following: Consider the one parameter family of canonical ellipses which go through $`(0,\pm u)`$
$$g(\varphi ,h)=h^2+\alpha (\varphi ^2u^2)=0,$$
(52)
where the orbit parameterization is such that it starts ($`z\mathrm{}`$) at $`(\varphi ,h)=(u,0)`$ and ends ($`z+\mathrm{}`$) at $`(+u,0)`$. By differentiating we find that
$$\left(\frac{g}{\varphi }\right)^2=4\alpha ^2\varphi ^2=4\alpha \left(\alpha u^2h^2\right);\left(\frac{g}{h}\right)^2=4h^2,$$
(53)
and that
$$dh^2=\alpha d\varphi ^2.$$
(54)
The last relation is used for calculating the integrals
$`{\displaystyle \left(\frac{U}{\varphi }\right)𝑑\varphi }`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \left[\lambda (\varphi ^2u^2)+gh^2\right]𝑑\varphi ^2}={\displaystyle \frac{\lambda \alpha g}{2\alpha ^2}}{\displaystyle h^2𝑑h^2}`$ (55)
$`=`$ $`{\displaystyle \frac{\lambda \alpha g}{4\alpha ^2}}h^4`$
and
$`{\displaystyle \left(\frac{U}{h}\right)𝑑h}`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle (\lambda _hh^2+k^2gu^2+g\varphi ^2)𝑑h^2}={\displaystyle \frac{1}{2}}{\displaystyle \left(\frac{\alpha \lambda _hg}{\alpha }h^2+k^2\right)𝑑h^2}`$ (56)
$`=`$ $`{\displaystyle \frac{h^2}{4}}\left({\displaystyle \frac{\alpha \lambda _hg}{\alpha }}h^2+2k^2\right).`$
Substituting in (51) we get
$$(\lambda \alpha g)\left(\alpha u^2h^2\right)h^4=\left[2k^2\alpha +(\alpha \lambda _hg)h^2\right]h^4,$$
(57)
or
$$\alpha \left(\lambda u^2\alpha gu^22k^2\right)+\left[\lambda +g\alpha (g+\lambda _h)\right]h^2=0.$$
(58)
The above must vanish identically, giving
$$\alpha =\frac{\lambda u^22k^2}{gu^2};\lambda _h=\frac{g\left(gu^22k^2\right)}{\lambda u^22k^2}.$$
(59)
Since there is only one free parameter ($`\alpha `$), we have one equation too many. While the first condition gives us the desired orbit, the second relation rather constrains the potential. At this point we may wonder whether this illness could be remedied by replacing the orbit equation (52) with a two parameter family such as <sup>1</sup><sup>1</sup>1Note that both families of curves include also the parabolic orbit $`h+\alpha (\varphi ^2u^2)=0`$, which yields a solution with similar characteristics as the elliptic orbit discussed above.
$$h^2+\alpha \left(\varphi ^2u^2\right)+\beta \left(\varphi ^2u^2\right)^2=0\mathrm{or}\alpha h^2+\beta h+\left(\varphi ^2u^2\right)=0.$$
(60)
As we found out, the answer is negative. Substituting the above two-parameter orbit in (51) we do find, in addition to the two Rajaraman solutions, solutions with $`\alpha ,\beta 0`$. However, the constraints on the potential parameters are not relaxed. For example the orbit
$$\alpha =\beta =\frac{\lambda u^22k^2}{gu^2},$$
(61)
is a viable orbit only if the following two constraints are met:
$$gu^2=2\left(\lambda u^2+k^2\right);\lambda _h=\frac{8g}{3}.$$
(62)
### A.2 Explicit Solutions
With a legitimate one-particle orbit at hand we can finally decouple the equations of motion (19) by substituting the orbit. For the straight line ($`h=0`$) we have
$$\varphi ^{\prime \prime }=\lambda \varphi ^3\lambda u^2\varphi +g\varphi h^2=\lambda \varphi ^3\lambda u^2\varphi ,$$
(63)
which is solved by
$$\varphi (z)=u\mathrm{tanh}\left[\sqrt{\frac{\lambda }{2}}u(zz_0)\right];h(z)=0.$$
(64)
For the more interesting orbit (52) we have
$$\varphi ^{\prime \prime }=\lambda \varphi ^3\lambda u^2\varphi +g\varphi h^2=\frac{2k^2}{u^2}\varphi ^32k^2\varphi ,$$
(65)
whose solution is
$$\varphi (z)=u\mathrm{tanh}[k(zz_0)];h(z)=\pm \sqrt{\frac{\lambda u^22k^2}{g}}\mathrm{sech}[k(zz_0)].$$
(66)
There is no apparent reason that one of the above solutions has the globally minimal action. In absence of a uniqueness theorem for such nonlinear equations, we can only rule out a solution if we find another solution with smaller action. Considering the above solutions, we find that their actions are
$$S_{\mathrm{straight}}=\frac{2}{3}\sqrt{2\lambda };S_{\mathrm{ellipse}}=\frac{2}{3}\frac{k}{gu}\left(\lambda +2g\frac{2k^2}{u^2}\right).$$
(67)
Provided the conditions (37) are met, sometimes the straight line has a smaller action than the elliptic orbit and sometimes it is vice versa, depending on the potential parameters. Moreover, these two trajectories might be not minima but maxima or saddle points. |
warning/0507/hep-th0507022.html | ar5iv | text | # 1 Introduction.
## 1 Introduction.
We continue to investigate the possibility of describing the elementary fermions as knotted solitons.<sup>1</sup> These knots may be understood either as simply symbols (labels of particles) or as real physical structures such as knotted flux tubes. To relate the simplest particles to the simplest knots, we represent each of the 4 families of elementary fermions by a separate soliton labelled by one of the 4 possible trefoils, e.g. the family $`e,\mu ,\tau `$ is represented by a single trefoil, while the $`e,\mu `$, and $`\tau `$ particles are separately identified as different states of excitation of their common trefoil. In this paper we attempt to calculate interactions between these $`q`$-fermions mediated by the $`q`$-gauge vector, or alternatively to determine the $`q`$-currents.
## 2 The Origin of the Knots.
Our work is based on the possibility that $`SU_q(2)`$ is an effective phenomenological symmetry. If it is, the symmetry group of the standard electroweak theory may be regarded as a degenerate form of $`SU_q(2)`$. The linearized form of the theory based on the $`q`$-symmetry is indeed in approximate agreement with the standard theory in lowest order.<sup>2</sup>
To go beyond the linearization one may expand the quantum fields in irreducible representations $`(D_{mm^{}}^j(q|a,\overline{a},b,\overline{b}))`$ of $`SU_q(2)`$ where the arguments $`(a,\overline{a},b,\overline{b})`$ obey the algebra of $`SU_q(2)`$. Then the normal modes, besides describing states of momentum and spin, will also contain factors $`D_{mm^{}}^j(q|a,\overline{a},b,\overline{b})`$. These are polynomials in the non-commuting arguments $`(a,\overline{a},b,\overline{b})`$ with eigenstates $`|n`$. Since the different normal modes therefore have internal excited states, they may be described as solitons ($`q`$-solitons) rather than as point particles. A class of these normal modes may be related to knots and labelled by $`D_{\frac{w}{2}\frac{r+1}{2}}^{N/2}(q|a,\overline{a},b,\overline{b})`$ where $`(N,w,r)`$ mean the number of crossings, the writhe, and the rotation of the knot.<sup>1</sup> (To correctly represent a knot the three integers $`(N,w,r)`$ must satisfy certain knot constraints, e.g. $`w`$ and $`r`$ must be of opposite parity.)
## 3 Representation of the Elementary Particles.
We now propose that the elementary particles may be usefully labelled by the irreducible representations of $`SU_q(2)`$ in the form $`D_{\frac{w}{2}\frac{r+1}{2}}^{N/2}`$. Depending on whether $`N`$ is even or odd, we assume that $`D_{\frac{w}{2}\frac{r+1}{2}}^{N/2}`$ represents either a boson or a fermion respectively.
The lowest possible value of $`N`$ is 3 and the corresponding knot is a trefoil. It is then natural to associate the elementary fermions with the ground and lowest excited states of the trefoils. There are 4 trefoils described by
$$(w,r)=(3,2),(3,2),(3,2),(3,2)$$
(3.1)
There are also 4 families of elementary fermions, namely:
$$(e,\mu ,\tau ),(d,s,b),(u,c,t),(\nu _e,\nu _\mu ,\nu _\tau )$$
(3.2)
The four trefoils and associated $`D_{\frac{w}{2}\frac{r+1}{2}}^{N/2}`$ are shown in Fig. 1:
Figure 1.
For the last line of Fig. 1, see Eqs. (8.1)-(8.4).
Each of the 4 families of elementary fermions may be represented by one of the 4 possible trefoils. The 3 individual fermions belonging to a single family are then assumed to represent 3 different states of excitation of a single trefoil.
Members of the 4 families have the following values of $`(t,t_3,Q)`$, i.e. the isotopic spin, its 3-component and the charge, and we shall tentatively assume that $`w`$ and $`r`$ (labelling the writhe and rotation of their common trefoil) have the values shown in the same table:
$$\begin{array}{ccccccc}& \underset{¯}{t}& \underset{¯}{t_3}& \underset{¯}{Q}& \underset{¯}{w}& \underset{¯}{r}& \underset{¯}{D_{\frac{w}{2}\frac{r+1}{2}}^{N/2}}\\ (e,\mu ,\tau )& 1/2& 1/2& 1& 3& 2& D_{\frac{3}{2}\frac{3}{2}}^{3/2}\\ (\nu _e,\nu _\mu ,\nu _\tau )& 1/2& 1/2& 0& 3& 2& D_{\frac{3}{2}\frac{3}{2}}^{3/2}\\ (d,s,b)& 1/2& 1/2& 1/3& 3& 2& D_{\frac{3}{2}\frac{1}{2}}^{3/2}\\ (u,c,t)& 1/2& 1/2& 2/3& 3& 2& D_{\frac{3}{2}\frac{1}{2}}^{3/2}\end{array}$$
(3.3)
The assignment of $`w`$ and $`r`$ to the 4 families is discussed in paragraph 6 and in Ref. 1.
In the preceding table we have assumed the following relations between conventional (point particle) labels and knot (soliton) labels for the elementary fermions.
$$\begin{array}{cccc}\underset{¯}{e\mu \tau }& \underset{¯}{dsb}& \underset{¯}{uct}& \underset{¯}{\nu _e\nu _\mu \nu _\tau }\\ t=\frac{N}{6}& t=\frac{N}{6}& t=\frac{N}{6}& t=\frac{N}{6}\\ t_3=\frac{w}{6}& t_3=\frac{w}{6}& t_3=\frac{w}{6}& t_3=\frac{w}{6}\\ Q=\frac{1}{4}r\frac{1}{2}& Q=\frac{1}{4}r+\frac{1}{6}& Q=\frac{1}{4}r+\frac{1}{6}& Q=\frac{1}{4}r+\frac{1}{2}\end{array}$$
(3.4)
These relations between $`(t,t_3,Q)`$ and $`(N,w,r)`$ define a knot model. These linear relations satisfy (3.3). This trial knot model then establishes a unique match between the elementary fermion families and the trefoils.
## 4 Other Knots.
Since the trefoils characterized by $`(N=3,w=\pm 3,r=\pm 2)`$ are the simplest knots, they have been chosen to represent the simplest particles: the leptons and quarks. One may obtain higher knots by forming a connected sum of trefoils: These higher knots may be interpreted as bosonic or fermionic depending on whether $`N`$ is even or odd. In this way one may replicate the quark building up principle; then the mesons are two connected trefoils and the hadrons are three connected trefoils.
Any knot may be represented by $`D_{\frac{w}{2}\frac{r+1}{2}}^{N/2}(q|a,\overline{a},b,\overline{b})`$ which is a $`q`$-polynomial, just as any algebraic curve may be represented by a numerically valued polynomial. On the other hand, not every $`D_{mn}^j(q|a,\overline{a},b,\overline{b})`$ represents a knot; according to our ideas, however, these non-knots still represent states of excitation of the field, and their symbols, forming a complete orthogonal basis, would all be required in the underlying field theory.
## 5 Representation of $`W^+W^{}Z`$ and $`A`$.
Since the electroweak vector fields are responsible for pair production one might try to think of the knots associated with these vectors as fusions of the knots representing leptons or quarks. Since we are associating these elementary fermions with trefoils we shall represent the intermediate vectors as di-trefoils, as shown in the figures and tables.
Figure 2. Knot Representation of Gauge Vectors. The particle labelling is: (a) $`𝒟_{\mathrm{1\hspace{0.33em}\hspace{0.33em}0}}^1`$ ; (b) $`𝒟_{\mathrm{1\hspace{0.33em}\hspace{0.33em}0}}^1`$; (c) $`𝒟_{\mathrm{0\hspace{0.33em}\hspace{0.33em}0}}^1`$; (d)$`𝒟_{\mathrm{1\hspace{0.33em}\hspace{0.33em}1}}^1`$ .
| | $`t`$ | $`t_3`$ | $`Q`$ | $`t_0`$ | $`N`$ | $`w`$ | $`r`$ | $`𝒟_{t_3t_0}^t`$ |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| $`\begin{array}{c}W^+\\ W^{}\\ W_3\\ W_0\end{array}`$ | $`\begin{array}{c}\hfill 1\\ \hfill 1\\ \hfill 1\\ \hfill 1\end{array}`$ | $`\begin{array}{c}\hfill 1\\ \hfill 1\\ \hfill 0\\ \hfill 1\end{array}`$ | $`\begin{array}{c}\hfill 1\\ \hfill 1\\ \hfill 0\\ \hfill 0\end{array}`$ | $`\begin{array}{c}\hfill 0\\ \hfill 0\\ \hfill 0\\ \hfill 1\end{array}`$ | $`\begin{array}{c}\hfill 6\\ \hfill 6\\ \hfill 7\\ \hfill 7\end{array}`$ | $`\begin{array}{c}\hfill 6\\ \hfill +6\\ \hfill 1\\ \hfill 1\end{array}`$ | $`\begin{array}{c}\hfill 3\\ \hfill +3\\ \hfill 0\\ \hfill 0\end{array}`$ | $`\begin{array}{c}𝒟_{\mathrm{1\hspace{0.33em}\hspace{0.33em}0}}^1\hfill \\ 𝒟_{\mathrm{1\hspace{0.33em}\hspace{0.33em}0}}^1\hfill \\ 𝒟_{\mathrm{0\hspace{0.33em}\hspace{0.33em}0}}^1\hfill \\ 𝒟_{\mathrm{1\hspace{0.33em}\hspace{0.33em}1}}^1\hfill \end{array}`$ |
Table 1.
In this scheme negative charge corresponds to counter-clockwise rotation. Then $`r`$ is positive for both $`e^{}`$ and $`W^{}`$. Note that $`𝒟_{t_3t_0}^t`$ exhibits the particle rather than the knot labelling.
The linear relations between the quantum numbers $`(t,t_3,Q)`$ and the knot integers $`(N,w,r)`$ are shown in Table 2.
| $`W^+`$ and $`W^{}`$ | $`W_3`$ | $`W_0`$ |
| --- | --- | --- |
| $`\begin{array}{c}t=\frac{N}{6}\hfill \\ t_3=\frac{w}{6}\hfill \\ Q=\frac{1}{3}r\hfill \end{array}`$ | $`\begin{array}{c}t=\frac{N1}{6}\hfill \\ t_3=w1\hfill \\ Q=r\hfill \end{array}`$ | $`\begin{array}{c}t=\frac{N1}{6}\hfill \\ t_3=w\hfill \\ Q=r\hfill \end{array}`$ |
Table 2.
Since the charge is proportional to the rotation of the knot in this scheme, the component trefoils must have opposite rotations when the di-trefoil represents a neutral vector. If the two components do have opposite rotations, however, there must be an additional crossing in the knot diagram. The knots representing $`W_3`$ and $`W_0`$ then differ in the writhe of the crossing as shown. The degeneracy between $`W_3`$ and $`W_0`$ is thus resolved by the differing values of the writhe, and in the standard theory by coupling $`W_0`$ to $`U(1)`$ or by the introduction of the Weinberg angle. Here we follow the standard theory by defining
$$\begin{array}{ccc}\hfill Z& =& W_0\mathrm{sin}\theta +W_3\mathrm{cos}\theta \hfill \\ \hfill A& =& W_0\mathrm{cos}\theta +W_3\mathrm{sin}\theta \hfill \end{array}$$
(5.1)
We have assumed that the number of intersections is (even, odd) for (bosonic, fermionic) knots. Although the number of intersections for $`W_0`$ and $`W_3`$ separately is 7, this does not violate the (even, odd) rule for the physical fields since $`A`$ and $`Z`$ are linear combinations of $`W_0`$ and $`W_3`$. Hence the $`A`$ and $`Z`$ field quanta, being composite knots with 14 intersections, obey the (even, odd) rule. In Table 2 we have arranged the relation between the isotopic spin and knot labels so that all the di-trefoils lie in the same $`SU_q(2)`$ multiplet.
## 6 Masses of Fermions.<sup>1</sup>
We follow the standard theory in assuming that the masses of the fermions depend on the Higgs field $`(\phi )`$ at the minima in the Higgs potential. The mass operator in the Hamiltonian density is then taken to be
$$=(\overline{\psi }_L\phi \psi _R+\overline{\psi }_R\phi \psi _L)$$
(6.1)
Since $`\psi _R`$ is a singlet in the standard theory we assume that it is also a singlet in the $`SU_q(2)`$ theory. Then
$$=\overline{\psi }_L\phi +\phi \psi _L$$
(6.2)
Now replace the fields $`\psi _L`$ and $`\phi `$ by their normal modes that represent trefoils. We have been assuming that all fields including the Higgs field and therefore the Higgs potential lie in the $`q`$-algebra. Let the Higgs potential be chosen so that its minima lie at the trefoil points. The Higgs field at these points is then
$$\phi =\rho (w,r)D_{\frac{w}{2}\frac{r+2}{2}}^{3/2}(a,\overline{a}b,\overline{b})$$
(6.3)
where $`(w,r)`$ is a trefoil point. Then the mass operator (6.2) at $`(w,r)`$ becomes
$$(w,r)=\rho (w,r)[\overline{\psi }_LD_{\frac{w}{2}\frac{r+1}{2}}^{3/2}+D_{\frac{w}{2}\frac{r+1}{2}}^{3/2}\psi _L]$$
(6.4)
and the mass operator associated with any soliton $`(w^{},r^{})`$ and Higgs $`(w,r)`$ becomes
$$(w^{}r^{};wr)=\rho (w,r)\left[\overline{D}_{\frac{w^{}}{2}\frac{r^{}+1}{2}}^{3/2}D_{\frac{w}{2}\frac{r+1}{2}}^{3/2}+D_{\frac{w}{2}\frac{r+1}{2}}^{3/2}D_{\frac{w^{}}{2}\frac{r^{}+1}{2}}^{3/2}\right]$$
(6.5)
Here we have dropped the multiplier of $`D_{\frac{w}{2}\frac{r+1}{2}}^{3/2}`$ in the normal mode expansion that pertains to the momentum and spin of the fermions, since that factor would cancel out in the following discussion. The expectation value of $`(w^{},r^{};w,r)`$ vanishes unless $`w=w^{}`$ and $`r=r^{}`$. Then, since only the first term of (6.5) contributes to the expectation value, we have
$$n|(w,r)|n=\rho (w,r)n|\overline{D}_{\frac{w}{2}\frac{r+1}{2}}^{3/2}D_{\frac{w}{2}\frac{r+1}{2}}^{3/2}|n$$
(6.6)
To accommodate the 4 families one needs 4 minima in the Higgs potential. These minima may be labelled by the magnitudes of the Higgs field $`\phi `$ and by the associated Higgs trefoils. The mass scale of each family is determined by $`\phi `$ at the minimum for that family, and the trefoil for that family must agree with the trefoil for $`\phi `$. With this understanding, Eq. (6.6) implies
$$m_n(w,r)=\rho (w,r)n|\overline{D}_{\frac{w}{2}\frac{r+1}{2}}^{3/2}D_{\frac{w}{2}\frac{r+1}{2}}^{3/2}|n$$
(6.7)
where $`m_n(w,r)`$ is the mass of the $`(w,r)`$ soliton at the n<sup>th</sup> level. The different mass spectra corresponding to the different solitons are by Ref. (1) or by (6.7), and (8.1) and (8.3) as follows:
I $`m_n(3,2)=\rho (3,2)\mathrm{\Delta }({\displaystyle \frac{3}{2}},{\displaystyle \frac{3}{2}})(1q^{2n2}|\beta |^2)(1q^{2n4}|\beta |^2)(1q^{2n6}|\beta |^2)`$ (6.8a)
II $`m_n(3,2)=\rho (3,2)\mathrm{\Delta }({\displaystyle \frac{3}{2}},{\displaystyle \frac{1}{2}})[q^{4n}|\beta |^4q^{6n2}|\beta |^6]`$ (6.8b)
III $`m_n(3,2)=\rho (3,2)\mathrm{\Delta }({\displaystyle \frac{3}{2}},{\displaystyle \frac{1}{2}})[q^{2n}|\beta |^2(1q^{2n}|\beta |^2)(1q^{2n+2}|\beta |^2)]`$ (6.8c)
IV $`m_n(3,2)=\rho (3,2)\mathrm{\Delta }({\displaystyle \frac{3}{2}},{\displaystyle \frac{3}{2}})q^{6n}|\beta |^6`$ (6.8d)
Since all masses $`m_n(w,r)`$, within a single spectrum are proportional to $`\rho (w,r)\mathrm{\Delta }(w,r)`$, one may compute ratios of these masses without ambiguity. In calculating these ratios we assume that only the three lowest states of each soliton are occupied. Set
$$M=\frac{1||1}{0||0}\text{and}m=\frac{2||2}{1||1}$$
(6.9)
There is an equation for both $`M`$ and $`m`$ in each spectrum I-IV. These two equations may be rewritten for $`q`$ and $`|\beta |^2`$ as follows:
$`\text{ I}{\displaystyle \frac{m1}{mq^6}}`$ $`=`$ $`q^2{\displaystyle \frac{M1}{Mq^6}}|\beta |^2=q^6{\displaystyle \frac{M1}{Mq^6}}`$ (6.10a)
$`\text{ II}{\displaystyle \frac{mq^4}{mq^6}}`$ $`=`$ $`q^2{\displaystyle \frac{Mq^4}{Mq^6}}|\beta |^2={\displaystyle \frac{mq^4}{mq^6}}`$ (6.10b)
$`\text{III}{\displaystyle \frac{mq^2}{mq^6}}`$ $`=`$ $`q^2{\displaystyle \frac{Mq^2}{Mq^6}}|\beta |^2={\displaystyle \frac{Mq^2}{Mq^6}}`$ (6.10c)
$`\text{ IV}M`$ $`=`$ $`m=q^6`$ (6.10d)
The empirical input depends on the masses of the elementary fermions. These are well determined for the leptons $`(e,\mu ,\tau )`$, but for the quarks they are not even well defined. Since the quarks do not exist as free particles, the quoted masses depend on the theoretical procedure for defining them. There is then a range of “masses” given by the Particle Data Group.<sup>3</sup>
To solve the above equations for $`q`$ and $`|\beta |^2`$, we have chosen the following values for $`M`$ and $`m`$.
$$\begin{array}{ccc}& \underset{¯}{M}& \underset{¯}{m}\\ (1)e\mu \tau \hfill & 193& 16.7\\ (2)dsb\hfill & 37.5& 31.8\\ (3)uct\hfill & 750& 117\\ (4)\nu _e,\nu _\mu ,\nu _\tau \hfill & \mathrm{?}& \mathrm{?}\end{array}$$
(6.11)
These ratios are based on the masses of the quarks recorded here:
$$\begin{array}{ccccccc}u& d& c& s& t& b& \\ .002& .004& 1.5& .15& 176& 4.77& \text{GeV/c}^2\end{array}$$
(6.12)
One may try to match the four familiies (1)-(4) shown in (6.11) with the four spectra (I-IV) shown in (6.8). It is clear that none of the three families (1), (2), (3) match (IV). Therefore we assign (IV) to the neutrino family. Next write (6.10a)-(6.10c) as algebraic equations in $`q^2`$ and assign the equation of lowest degree to the lepton family (since the leptons do not have hypercharge or gluon charge.<sup>1</sup>) Then if we assign the I, II, and III spectra to $`(e,\mu ,\tau ),(d,s,b)`$ and $`(u,c,t)`$ respectively, we find that the roots of (6.10a)- (6.10c) where $`q`$ is closest to unity are
$$(e\mu \tau )q=1.46|\beta |=3.20$$
(6.13)
$$(dsb)q=1.76|\beta |=3.35$$
(6.14)
$$(uct)q=2.14|\beta |=1.07$$
(6.15)
It also turns out that any other match is also good, i.e., if $`q=q(M,m,w,r)`$, then it is found that $`q`$ depends mainly on $`M`$ and $`m`$, and is nearly independent of $`w`$ and $`r`$. We shall not, however, represent each of the three fermion families as a linear combination of the three trefoils, since they are topologically distinct, and consequently there is a topological obstruction to any dynamical transition between any two of them. Therefore in this simplified model we associate each family with a single trefoil, or equivalently with a single normal mode or irreducible representation, $`D_{\frac{w}{2},\frac{r+1}{2}}^{3/2}`$, where $`(w,r)`$ characterizes the trefoil. To match the families with the trefoils in a unique way, we tentatively postulate the knot model described by (3.4).
Higher knots designated by $`D_{\frac{w}{2}\frac{r+1}{2}}^{N/2}`$ with the same $`(w,r)`$ and $`N>3`$ are topologically equivalent and can therefore dynamically decay to trefoils. Moreover, if the dynamics requires that lower $`N`$, as well as lower $`n`$, implies lower energy, then only the trefoil solitons will be stable and recognizable as particles. (The topologically equivalent but dynamically unstable higher knots differ from the trefoils by a connected sum of curls.)
Eqs. (6.8a) through (6.8d) are of the form
$$m_n(w,r)=\rho (w,r)F(w,r;n,q,\beta )$$
(6.16)
By (6.13), (6.14), (6.15) one sees that $`F(w,r;n,q,\beta )`$ is negative in (6.8a) and (6.8b), but it is positive in (6.8c) and (6.8d). Therefore the first two minima $`(\rho (3,2),\rho (3,2))`$, must be negative while $`\rho (3,2)`$ and $`\rho (3,2)`$ must be positive to ensure that all masses $`m_n(w,r)`$ are positive.
The magnitude of $`\rho (w,r)`$ sets the energy scale and differs for each family. The choice of $`\rho (w,r)`$ and $`F(w,r;n,q,\beta )`$ for each family is determined by the knot model, i.e. by the postulated linear relation between $`(t_3,Q)`$ and $`(w,r)`$ in (3.3) or (3.4), as well as by the postulated relation between knots and the irreducible representations of $`SU_q(2)`$, namely $`D_{\frac{w}{2}\frac{r+1}{2}}^{N/2}`$.
The value of $`q`$ depends only weakly on $`(w,r)`$ in $`F(w,r;n,q,\beta )`$ but it does depend strongly on $`\beta `$ and $`m_n(w,r)`$. The parameter $`q`$ therefore behaves like a running coupling constant, where $`\beta `$ and $`m_n(w,r)`$ fix the energy scale.
We may interpret the numerical value of $`q`$ as a measure of the influence of the fields that play a role in the determination of the fermionic masses and that are excluded from the standard electroweak theory. Consistent with this view, $`q`$ is not far from unity; and the lepton family, having no gluon charge, has a $`q`$ value closer to unity than the quark families.
We have assumed that the three observed particles of each family occupy the 3 lowest states of the soliton representing that family. The model also permits higher excited states but if these lie at very high energies, they may have such short lifetimes that they would not be observable as particles. The tentative assignment that we have assumed in (3.3) leads to a fourth generation of (-1/3 quarks) at $`30m_b144`$ GeV and a fourth generation of (2/3 quarks) at $`100m_t17,600`$ GeV. The corresponding fourth generation lepton would appear at $`12m_\tau 21.3`$ GeV but is excluded by the known decays of the $`Z^0`$.<sup>7</sup> If the assignments of $`dsb`$ and $`uct`$ are interchanged so that $`dsb`$ corresponds to III and $`uct`$ to II then the fourth generation would appear at $`30.4m_b`$ and $`102m_t`$. If a fourth generation should be observed then a unique assignment of the $`(dsb)`$ and $`(uct)`$ families to trefoils could be put on an empirical basis. In any case further refinements of the model would depend on whether any or none of the fourth generation particles is observed. The predicted neutrino spectrum is a further test of the model. The neutrino data are very sparse, but are compatible with $`q1`$, leading by (6.8d) to a geometric hierarchy of nearly equal masses.<sup>4</sup>
## 7 Interactions Mediated by a Vector Field.
We are next interested in interactions that stem from the gauge invariant terms
$$\overline{\psi }/\psi $$
(7.1)
where $`/=\gamma ^\mu _\mu `$ is the gauge covariant derivative. This term gives rise to
$$\overline{\psi }W/\psi $$
(7.2)
where
$$W=$$
(7.3)
When $`\psi `$ and $`W`$ are expanded in absorption and emission operators, the normal modes will specify momentum, spin and species of soliton; in more detail it will specify the internal state of the soliton. To describe the interaction between fermions mediated by a vector particle one replaces the field operators by normal modes. Schematically
$$\overline{\psi }W/\psi \overline{D}_i^\alpha W/D_j^\beta $$
(7.4)
where
$$D_i^\alpha =D^\alpha |i$$
(7.5)
and $`|i`$ is an “internal” state, like a spin state. Here $`\alpha `$ runs over the 4 kinds of trefoils, i.e., $`\alpha `$ fixes $`(w,r)`$ while $`|i`$ labels the particle and the level of the trefoil spectrum. Hence
$$\overline{\psi }W/\psi i|\overline{D}_{\frac{w_1}{2}\frac{r_1+1}{2}}^{3/2}W/D_{\frac{w_2}{2}\frac{r_2+1}{2}}^{3/2}|j$$
(7.6)
where we have abstracted just the part of the matrix element that depends on the $`q`$-algebra.
In (7.6) $`W`$ is to be replaced by a normal mode or by a linear combination of normal modes. Lacking a firm a priori basis, this choice must be determined by empirical data. The problem here is similar to that faced in the earlier days of weak interaction theory where various linear combinations of the five fundamental forms were proposed before the decisive experiment requiring V-A was performed. Here we shall be guided on the one hand by the di-trefoil construction (Fig. 2) and on the other by the experimental requirement that each lepton be pair produced with only its “own” neutrino (lepton conservation) as well as by the additional restriction usually expressed as the universal Fermi interaction.
To satisfy these requirements we have made the following choices:
$$\begin{array}{ccccc}& \underset{¯}{W^{}}& \underset{¯}{W^+}& \underset{¯}{W^3}& \underset{¯}{W^0}\\ (a)\hfill & D_{30}^3& D_{30}^3& D_{00}^3& D_{11}^3\\ (b)\hfill & D_{+\frac{w}{2}\frac{r3}{2}}^{N/2}& D_{+\frac{w}{2}\frac{r+3}{2}}^{N/2}& D_{w1,r}^{(N1)/2}& D_{w,r+1}^{(N1)/2}\end{array}$$
(7.7)
Line (a) is chosen so that there is no change in level between initial and final states and therefore the usual fermion pairs are produced by $`W`$.
Line (b) is a relabelling of line (a) in terms of the knot signature $`(N,w,r)`$ according to Tables 1 and 2. A more general possibility is
$$C_{}(q,\beta )D_{30}^3W^{}+C_+(q,\beta )D_{30}^3W^++C_3(q,\beta )D_{00}^3W^3+C_0(q,\beta )D_{11}^3W^0$$
(7.8)
Since we are interested mainly in relative rates in this paper, we shall usually not be concerned with the possible $`(q,\beta )`$ dependence of the four coefficients.
The representatives of the charged vectors, namely $`D_{30}^3`$ and $`D_{30}^3`$, are conjugate monomials (up to $`q_1^3`$) while the corresponding representatives of the neutral vectors, namely $`D_{00}^3`$ and $`D_{11}^3`$, are polynomials in the $`(b,\overline{b})`$ subalgebra. These polynomials must satisfy the requirement that $`W^3`$ and $`W^0`$ have non-vanishing matrix elements between neutrino states. It follows that $`D_{00}^3`$ and $`D_{11}^3`$ must lie in the $`(b,\overline{b})`$ subalgebra since the neutrino states lie in this subalgebra. By (8.1) the conditions that $`D_{mm^{}}^j`$ lies in the $`(b,\overline{b})`$ algebra are
$$s+t=jm$$
(7.9)
and
$$s+t=j+m^{}$$
(7.10)
Hence
$$m+m^{}=0$$
(7.11)
By (8.1) the condition that $`D_{mm^{}}^j`$ be a function of only the product $`(b\overline{b})`$ requires in addition to (7.9) and (7.10)
$$s+t=j+m$$
(7.12)
By (7.10) and (7.12)
$$m=m^{}$$
(7.13)
Both $`D_{00}^3`$ and $`D_{11}^3`$ satisfy (7.11). Only $`D_{00}^3`$ satisfies (7.13) as well.
The neutral sector of the algebra is then determined by the neutral vectors and the neutrinos to be the $`(b,\overline{b})`$ subalgebra.
## 8 The “Internal” Modes.
To evaluate (7.6) one expresses the irreducible representations of $`SU_q(2)`$ as follows:<sup>1</sup>
$$\begin{array}{ccc}\hfill D_{mm^{}}^j(a,\overline{a},b,\overline{b})& =& \mathrm{\Delta }_{mm^{}}^j_{s,t}\begin{array}{c}n_+\\ s\end{array}_1\begin{array}{c}n_{}\\ t\end{array}_1q_1^{t(n_++1s)}(1)^t\delta (s+t,n_+^{})\hfill \\ & & \times a^sb^{n_+s}\overline{b}^t\overline{a}^{n_{}t}\hfill \end{array}$$
(8.1)
where
$$\begin{array}{ccc}\hfill n_\pm & =& j\pm m\hfill \\ \hfill n_\pm ^{}& =& j\pm m^{}\hfill \end{array}\begin{array}{c}n\\ s\end{array}_1=\frac{n_1!}{s_1!ns_1!}n_1=\frac{q_1^{2n}1}{q_1^21}$$
$$\mathrm{\Delta }_{mm^{}}^j=\left[\frac{n_+^{}_1!n_{}^{}_1!}{n_+_1!n_{}_1!}\right]^{1/2}q_1=q^1$$
The special cases (3.3) and (7.7a) when written out according to (8.1) are
Fermions
$$\begin{array}{ccccc}(w,r)& (3,2)& (3,2)& (3,2)& (3,2)\\ & D_{\frac{3}{2}\frac{3}{2}}^{3/2}& D_{\frac{3}{2}\frac{1}{2}}^{3/2}& D_{\frac{3}{2}\frac{1}{2}}^{3/2}& D_{\frac{3}{2}\frac{3}{2}}^{3/2}\\ & & & & \\ & a^3& \mathrm{\Delta }_{\frac{3}{2}\frac{1}{2}}^{3/2}\begin{array}{c}3\\ 1\end{array}_1ab^2& \mathrm{\Delta }_{\frac{3}{2}\frac{1}{2}}^{3/2}\begin{array}{c}3\\ 1\end{array}_1q_1\overline{b}\overline{a}^2& q_1^3\overline{b}^3\\ & & & & \\ & (e\mu \tau )& (dsb)& (uct)& (\nu _e\nu _\mu \nu _\tau )\end{array}$$
(8.2a)
where
$$\mathrm{\Delta }_{\frac{3}{2}\frac{1}{2}}^{3/2}\begin{array}{c}3\\ 1\end{array}_1=3_1^{1/2}$$
(8.2b)
To pass from particle to anti-particle we propose to take not only the usual charge conjugation operator, but in addition to take the $`q`$-conjugate as well. The $`q`$-antifermions are represented by the adjoint symbols, e.g., the $`(\overline{e}\overline{\mu }\overline{\tau })`$ family is represented by $`\overline{a}^3`$.
Vectors
$$\begin{array}{cccc}\underset{¯}{W^{}}& \underset{¯}{W^+}& \underset{¯}{W^3}& \underset{¯}{W^0}\\ D_{30}^3& D_{30}^3& D_{00}^3& D_{11}^3\\ \begin{array}{c}6\\ 3\end{array}_1^{1/2}a^3b^3& \begin{array}{c}6\\ 3\end{array}_1^{1/2}q_1^3\overline{b}^3\overline{a}^3& f_3(b\overline{b})& f_0(b,\overline{b})\end{array}$$
(8.3a)
where
$$\mathrm{\Delta }_{30}^3\begin{array}{c}6\\ 3\end{array}_1=\begin{array}{c}6\\ 3\end{array}_1^{1/2}$$
(8.3b)
and
$$f_0(b,\overline{b})=\left[\begin{array}{c}4\\ 2\end{array}_1q^2(1b\overline{b})(1q^2b\overline{b})2_14_1q_1^2(1b\overline{b})(b\overline{b})+q_1^{12}(b\overline{b})^2\right]\overline{b}^2$$
(8.4)
$$f_3(b\overline{b})=\underset{s=0}{\overset{2}{}}(1q^{2s}b\overline{b})q^23_1^2(b\overline{b})\underset{s=0}{\overset{1}{}}(1q^{2s}b\overline{b})+q_1^23_1^2(b\overline{b})^2(1b\overline{b})q_1^{12}(b\overline{b})^3$$
(8.5)
Note that flavor changing neutral currents are absolutely forbidden. Note also that
$$\overline{D}_{00}^3=D_{00}^3$$
$$\overline{D}_{11}^3(b,\overline{b})=D_{11}^3(\overline{b},b)$$
In reducing (7.6) the following relations are useful:<sup>1,5</sup>
$$\begin{array}{cc}ab=qba\hfill & a\overline{a}+b\overline{b}=1\hfill \\ a\overline{b}=q\overline{b}a\hfill & \overline{a}a+q_1^2\overline{b}b=1\hfill \\ b\overline{b}=\overline{b}b\hfill & \end{array}$$
(8.6)
$$\begin{array}{ccc}\hfill \overline{b}|n& =& q^n\beta ^{}|n\hfill \\ \hfill b\overline{b}|n& =& q^{2n}|\beta |^2|n\hfill \end{array}$$
(8.7)
$$\begin{array}{ccc}\hfill a|n& =& \lambda _n|n1\hfill \\ \hfill |\lambda _n|& =& (1q^{2(n1)}|\beta |^2)^{1/2}\hfill \end{array}$$
(8.8)
$$\begin{array}{ccc}\hfill \overline{a}|n& =& \mu _n|n+1\hfill \\ \hfill |\mu _n|& =& (1q^{2n}|\beta |^2)^{1/2}\hfill \end{array}$$
(8.9)
$$n|m=\delta (n,m)$$
(8.10)
$$\begin{array}{ccc}\hfill \overline{a}^na^n& =& \underset{s=0}{\overset{n1}{}}(1q_1^{2(s+1)}b\overline{b})=\underset{t=1}{\overset{n}{}}(1q_1^{2t}b\overline{b})\hfill \\ \hfill a^n\overline{a}^n& =& \underset{s=0}{\overset{n1}{}}(1q^{2s}b\overline{b})\hfill \end{array}$$
(8.11)
In the following we shall determine the dependence of the matrix elements on the $`q`$-algebra.
## 9 Lepton-Neutrino Couplings.
(a) Mediated by $`W^{}`$:
$$\overline{\mathrm{}}(j)+W^{}\overline{\nu }(i)\text{or}W^{}\mathrm{}(j)+\overline{\nu }(i)$$
(9.1)
The matrix element for the absorption of a $`\overline{\mathrm{}}(j)`$ and the emission of a $`\overline{\nu }(i)`$ is by (8.4) and (8.5)
$$\begin{array}{ccc}\hfill m(i,j)& =& i|\underset{\frac{3}{2}\frac{3}{2}}{\overset{3/2}{\stackrel{=}{D}}}D_{30}^3\overline{D}_{\frac{3}{2}\frac{3}{2}}^{3/2}|j\hfill \\ & =& \begin{array}{c}6\\ 3\end{array}_1^{1/2}q_1^3i|(\overline{b}^3)(a^3b^3)(\overline{a}^3)|j\hfill \end{array}$$
(9.2)
where the double bar signifies an antiparticle in the final state and where by (8.8) and (8.13)
$$\begin{array}{ccc}\hfill \overline{b}^3a^3b^3\overline{a}^3& =& q^9(\overline{b}b)^3a^3\overline{a}^3\hfill \\ & =& q^9(\overline{b}b)^3(1b\overline{b})(1q^2b\overline{b})(1q^4b\overline{b})\hfill \end{array}$$
(9.3)
Then
$$m(i,j)=\begin{array}{c}6\\ 3\end{array}_1^{1/2}q^{6+6n_i}|\beta |^6f(n_i)f(n_i+1)f(n_i+2)\delta (i,j)$$
(9.4)
where
$$f(n)=1q^{2n}|\beta |^2$$
(9.5)
The ratio of matrix elements at level $`(n+1)`$ to those at level $`(n)`$ is
$$\begin{array}{ccc}\hfill R_n=\frac{m(n+1)}{m(n)}& =& q^6\frac{f(n+1)f(n+2)f(n+3)}{f(n)f(n+1)f(n+2)}\hfill \\ & =& q^6\frac{1q^{2n+6}|\beta |^2}{1q^{2n}|\beta |^2}\hfill \end{array}$$
(9.6)
Then
$$\begin{array}{ccc}\hfill M& & R_0=q^6\frac{1q^6|\beta |^2}{1|\beta |^2}\hfill \\ \hfill m& & R_1=q^6\frac{1q^8|\beta |^2}{1q^2|\beta |^2}\hfill \end{array}$$
(9.7)
The Eqs. (9.7) may be rewritten as two equations for $`|\beta |^2`$, namely:
$$\begin{array}{ccc}\hfill |\beta |^2& =& \frac{Mq^6}{Mq^{12}}\hfill \\ \hfill |\beta |^2& =& q^2\frac{mq^6}{mq^{12}}\hfill \end{array}$$
(9.8)
By eliminating $`|\beta |^2`$ one finds
$$x^9+m2_xx^4M4_xx^3+Mm=0$$
(9.9)
where $`2_x`$ and $`4_x`$ are basic numbers $`\left(n_x=\frac{x^n1}{x1}\right)`$ and
$$x=q^2$$
(9.10)
If we assume that the universal Fermi interaction that holds for point particles in the standard theory also holds here, then
$$M=m=1$$
(9.11)
and (9.8) and (9.9) imply
$$q=1$$
(9.12)
This result differs sharply from the results of (6.13)-(6.15) and is a consequence of postulating lepton conservation as well as the universal Fermi interaction $`(M=m=1)`$ that holds for point particles. Depending on the extent that the U.F.I. may be violated between solitons, one would find solutions of (9.9) differing from but close to unity.
If $`q`$ is exactly unity, then $`|\beta |=\frac{1}{2}\sqrt{2}`$ by (9.8). If $`m`$ and $`M`$ differ from unity, there will be corresponding shifts in $`(q,\beta )`$ according to (9.8) and (9.9).
(b) Mediated by $`W^+`$:
Now
$$\mathrm{}(j)+W^+\nu (i)\text{or}W^+\nu (i)+\overline{\mathrm{}}(j)$$
(9.13)
The matrix element for this reaction is
$$\begin{array}{ccc}\hfill m(i,j)^{}& =& i|\overline{D}_{\frac{3}{2}\frac{3}{2}}^{3/2}D_{30}^3D_{\frac{3}{2}\frac{3}{2}}^{3/2}|j\hfill \\ & =& i|(q_1^3b^3)\left(\begin{array}{c}6\\ 3\end{array}_1^{1/2}q_1^3\overline{b}^3\overline{a}^3\right)a^3|j\hfill \\ & =& \begin{array}{c}6\\ 3\end{array}_1^{1/2}q^{6n_i6}|\beta |^6f(n_i1)f(n_i2)f(n_13)\delta (i,j)\hfill \end{array}$$
(9.14)
where $`f(n)`$ is given by (9.5). Then
$$R=\frac{m(i,j)^{}}{m(i,j)}=q^{12}\frac{f(n1)f(n2)f(n3)}{f(n)f(n+1)f(n+2)}$$
(9.15)
is the ratio of the matrix elements for the two charge conjugate reactions (9.1) and (9.13), up to the factor $`C_+(q,\beta )/C_{}(q,\beta )`$.
Since $`R`$ is empirically very close to unity, (9.15) suggests that $`q`$ is again very close to unity. Hence the charge conjugate symmetry as well as the universality of the Fermi interaction both imply that $`q`$ is near unity in the interaction of leptons and neutrinos. Therefore we conclude that the additional degrees of freedom associated with masses of the leptons and neutrinos are not excited in their pair production. These last remarks depend on the choice of $`C_+(q,\beta )`$ and $`C_{}(q,\beta )`$ that in turn are restricted by the relative masses of the vectors to be discussed later.
We may take the view that the internal $`SU_q(2)`$ algebra is an effective deformation of $`SU(2)`$ that depends on the background: in the case of the soliton spectra the deviations of $`q`$ from unity are relatively large but in the case of lepton-neutrino interactions, these deviations are suppressed, just as they would be if we were dealing with point particles rather than solitons, i.e. as if a weak charge were concentrated at the center of an approximately spherically symmetric soliton.
## 10 Charge Changing Quark Couplings.
We first consider
$$Q(j,\frac{1}{3})+W^+Q(i,\frac{2}{3})$$
(10.1)
where $`Q(j,\frac{1}{3})`$ is any quark of charge -1/3 and $`Q(i,\frac{2}{3})`$ is any quark of charge 2/3.
The matrix element for this process is by (8.4) and (8.5)
$$\begin{array}{ccc}\hfill m\left[\left(\frac{1}{3}j\right)\left(\frac{2}{3}i\right)\right]& =& i|\overline{D}_{\frac{3}{2}\frac{1}{2}}^{3/2}D_{30}^3D_{\frac{3}{2}\frac{1}{2}}^{3/2}|j\hfill \\ & =& 𝒞\mathrm{\Delta }_{30}^3\begin{array}{c}6\\ 3\end{array}_1q_1^4i|(a^2b)(\overline{b}^3\overline{a}^3)(ab^2)|j\hfill \end{array}$$
(10.2)
with
$$𝒞=\mathrm{\Delta }_{\frac{3}{2}\frac{1}{2}}^{3/2}\mathrm{\Delta }_{\frac{3}{2}\frac{1}{2}}^{3/2}\left(\begin{array}{c}3\\ 1\end{array}_1\right)^2$$
and where
$$\begin{array}{ccc}\hfill i|(a^2b)(\overline{b}^3\overline{a}^3)(ab^2)|j& =& q^8i|(b\overline{b}^3)(a^2\overline{a}^3a)b^2|j\hfill \\ & =& q^8i|(b\overline{b}^3)(a^2\overline{a}^2)(\overline{a}a)b^2|j\hfill \end{array}$$
(10.3)
Then
$$m\left[\left(\frac{1}{3}j\right)\left(\frac{2}{3}i\right)\right]=q^4𝒞\begin{array}{c}6\\ 3\end{array}_1^{1/2}q^{6n_i}|\beta |^6f(n_i)f(n_i+1)f(n_i1)\delta (i,j)$$
(10.4)
where $`f(n)`$ is defined by (9.5).
The ratio of matrix elements at level $`n+1`$ to those at level $`n`$ is by (10.4)
$$R_n=q^6\frac{1q^{2(n+2)}|\beta |^2}{1q^{2(n1)}|\beta |^2}$$
(10.5)
In particular
$$\begin{array}{ccc}\hfill R_0& =& q^6\frac{1q^4|\beta |^2}{1q^2|\beta |^2}=\frac{m(s+W^+c)}{m(d+W^+u)}\hfill \\ \hfill R_1& =& q^6\frac{1q^6|\beta |^2}{1|\beta |^2}=\frac{m(b+W^+t)}{m(s+W^+s)}\hfill \end{array}$$
(10.6)
Set
$$\begin{array}{ccc}\hfill M& =& R_0\hfill \\ \hfill m& =& R_1\hfill \end{array}$$
(10.7)
Then by (10.6)
$$\begin{array}{ccc}\hfill |\beta |^2& =& q^2\frac{Mq^6}{Mq^{12}}\hfill \\ \hfill |\beta |^2& =& \frac{mq^6}{mq^{12}}\hfill \end{array}$$
(10.8)
By the preceding equations for $`|\beta |^2`$
$$q^2\frac{Mq^6}{Mq^{12}}=\frac{mq^6}{mq^{12}}$$
(10.9)
or
$$q^{18}+mq^82_{q^2}Mq^64_{q^2}+Mm=0$$
(10.10)
Again if the matrix elements are equal for the following processes:
$$\begin{array}{ccc}\hfill d+W^+& & u\hfill \\ \hfill s+W^+& & c\hfill \\ \hfill b+W^+& & t\hfill \end{array}$$
(10.11)
we may set
$$M=m=1$$
(10.12a)
then (10.8) and (10.10) imply
$$q=1|\beta |=\frac{1}{2}\sqrt{2}=.707$$
(10.12b)
Since the diagonal elements of the Kobayashi-Maskawa matrix are not quite equal however (i.e. not strictly independent of $`n`$), Eq. (10.12a) is not exactly satisfied so that $`q`$ and $`|\beta |`$ must differ slightly from (10.12b).
Let us next consider processes mediated by $`W^{}`$:
$$Q(\frac{2}{3},j)+W^{}Q(\frac{1}{3},i)$$
(10.13)
For this reaction (7.6) becomes
$$𝒞\begin{array}{c}6\\ 3\end{array}_1^{1/2}q_1i|\overline{ab^2}a^3b^3\overline{b}\overline{a}^2|j=𝒞\begin{array}{c}6\\ 3\end{array}_1^{1/2}q_1i|\overline{b}^2(\overline{a}a^3)(b^3\overline{b})\overline{a}^2|j$$
(10.14)
where
$$\begin{array}{ccc}\hfill \overline{b^2}(\overline{a}a^3)(b^3\overline{b})\overline{a}^2& =& q^6(\overline{b}^2b^3)(\overline{a}a^3)\overline{b}\overline{a}^2\hfill \\ & =& q^8\overline{b}^3b^3\overline{a}a^3\overline{a}^2\hfill \\ & =& q^8(b\overline{b})^3(\overline{a}a)(a^2\overline{a}^2)\hfill \\ & =& q^8(b\overline{b})^3(1q_1^2\overline{b}b)(1b\overline{b})(1q^2b\overline{b})\hfill \end{array}$$
(10.15)
Then the matrix element (10.14) is by (10.15)
$$\begin{array}{ccc}\hfill m\left[\left(\frac{2}{3}j\right)\left(\frac{1}{3}i\right)\right]& =& 𝒞\begin{array}{c}6\\ 3\end{array}_1^{1/2}q^7i|(b\overline{b})^3(1q_1^2\overline{b}b)(1b\overline{b})(1q^2\overline{b}b)|j\hfill \\ & =& 𝒞q^7\begin{array}{c}6\\ 3\end{array}_1^{1/2}q^{6n_i}|\beta |^6(1q^{2(n_i1)}|\beta |^2)(1q^{2n_i}|\beta |^2)\hfill \\ & & \times (1q^{2(n_i+1)}|\beta |^2)\delta (i,j)\hfill \end{array}$$
(10.16)
or
$$m\left[\left(\frac{2}{3}i\right)\left(\frac{1}{3}i\right)\right]=𝒞q^7\begin{array}{c}6\\ 3\end{array}_1^{1/2}q^{6n_i}|\beta |^6f(n_i1)f(n_i)f(n_i+1)$$
(10.17)
where $`f(n)`$ is defined by (9.5).
This matrix element covers the following cases:
$$u+W^{}dc+W^{}st+W^{}b$$
(10.18)
In particular we have by (10.17) with $`n_i=0`$
$$m(ud)=𝒞q^7\begin{array}{c}6\\ 3\end{array}_1^{1/2}|\beta |^6f(1)f(0)f(1)$$
(10.19)
The ratio of matrix elements for the two reactions (10.4) and (10.17) is
$$\frac{m\left[Q\left(\frac{1}{3}i\right)+W^+Q\left(\frac{2}{3}i\right)\right]}{m\left[Q\left(\frac{2}{3}i\right)+W^{}Q\left(\frac{1}{3}i\right)\right]}=q_1^3$$
(10.20)
again up to the factor $`C_+(q,\beta )/C_{}(q,\beta )`$. Note that the $`q_1^3`$ appearing in (10.20) stems from the same factor in $`D_{30}^3`$ that appears in (8.5). If the symbol for $`W^+`$ is defined without this factor, then the symbols for $`W^+`$ and $`W^{}`$ are $`q`$-conjugate and $`q_1^3`$ does not appear in (10.20). This option is subsumed in the choice of the normalizing factors $`C_+(q,\beta )`$ and $`C_{}(q,\beta )`$.
## 11 The Kobayashi-Maskawa Matrix.
We want to compare the ratios calculated here with the Kobayashi-Maskawa matrix, namely:
$$\begin{array}{cccc}& d& s& b\\ & & & \\ u& 0.973& 0.23& 0\\ c& 0.24& 0.91& 0.06\\ t& 0& 0& 1\end{array}$$
(11.1)
without introducing the Cabibbo-GIM angles. (The matrix (11.1) is known more accurately but (11.1) is adequate for the present.)
The diagonal elements are all approximately unity. Then $`q1`$ in (10.10) if $`W^+`$ is represented by $`D_{30}^3`$ as in (7.7a). This choice of $`W^+`$, however, forbids
$$s+W^+u$$
(11.2)
$$d+W^+c$$
(11.3)
To include these forbidden processes as well we may replace $`D_{30}^3`$ and $`D_{30}^3`$ by
$$W^+D_{30}^3(1+a+\overline{a})$$
(11.4)
$$W^{}(1+a+\overline{a})D_{30}^3$$
(11.5)
The expressions appearing in (11.4) and (11.5) represent minimal modifications of $`W^+`$ and $`W^{}`$. We also have
$$\begin{array}{ccc}\hfill a& =& D_{\frac{1}{2}\frac{1}{2}}^{1/2}\hfill \\ \hfill \overline{a}& =& D_{\frac{1}{2}\frac{1}{2}}^{1/2}\hfill \end{array}$$
(11.6)
so that these modified forms may be written as
$$\begin{array}{ccc}\hfill W^+& & D_{30}^3(1+D_{\frac{1}{2}\frac{1}{2}}^{1/2}+D_{\frac{1}{2}\frac{1}{2}}^{1/2})\hfill \\ \hfill W^{}& & (1+D_{\frac{1}{2}\frac{1}{2}}^{1/2}+D_{\frac{1}{2}\frac{1}{2}}^{1/2})D_{30}^3\hfill \end{array}$$
(11.7)
$`W^+`$ and $`W^{}`$ may then be written as a linear combination of $`D_{mn}^j`$ terms with $`q`$-Clebsch-Gordan coefficients.
If the factor $`q_1^3`$ is dropped, $`\overline{D}_{30}^3=D_{30}^3`$ and $`\overline{W}^{}=W^+`$ in (11.4) and (11.5).
The assumptions (11.4) and (11.5) still forbid
$$\begin{array}{ccc}\hfill u+W^{}& & b\hfill \\ \hfill t+W^{}& & d\hfill \end{array}$$
(11.8)
as required by the approximate Kobayashi-Maskawa matrix.
We would expect the justification for modifying (7.8) by (11.4) and (11.5) for quarks to be found only in a refinement of the simple knot model described here. A similar modification of (7.8) for the lepton-neutrino system is forbidden by lepton conservation.
According to (11.4) the matrix element for the process: $`s+W^+u`$ is
$$\begin{array}{ccc}\hfill m(s+W^+u)& =& 0|\overline{D}_{\frac{3}{2}\frac{1}{2}}^{3/2}D_{30}^3(a+\overline{a})D_{\frac{3}{2}\frac{1}{2}}^{3/2}|1\hfill \\ & =& 𝒞0|(a^2b)q_1^4\begin{array}{c}6\\ 3\end{array}_1^{1/2}\overline{b}^3\overline{a}^3(a+\overline{a})(ab^2)|1\hfill \\ & =& q_1^4𝒞\begin{array}{c}6\\ 3\end{array}_1^{1/2}0|(a^2b)(\overline{b}^3\overline{a}^3)a(ab^2)|1\hfill \end{array}$$
(11.9)
where
$$\begin{array}{ccc}\hfill 0|(a^2b)(\overline{b}^3\overline{a}^3)a(ab^2)|1& =& 0|a^2(b\overline{b}^3)\overline{a}(\overline{a}^2a^2)b^2|1\hfill \\ & =& q^80|(b\overline{b}^3)a(a\overline{a})(\overline{a}^2a^2)b^2|1\hfill \\ & =& q^80|(b\overline{b}^3)a(1b\overline{b})(1q_1^2b\overline{b})(1q_1^4b\overline{b})b^2|1\hfill \\ & =& q^{10}0|(b\overline{b})^3(1q^2b\overline{b})(1b\overline{b})(1q_1^2b\overline{b})a|1\hfill \end{array}$$
(11.10)
Then by (8.10)
$$\begin{array}{ccc}\hfill m(s+W^+u)& =& q^6𝒞\begin{array}{c}6\\ 3\end{array}_1^{1/2}0|(b\overline{b})^3(1q^2b\overline{b})(1b\overline{b})(1q_1^2b\overline{b})(1|\beta |^2)^{1/2}|0\hfill \\ & =& q^6𝒞\begin{array}{c}6\\ 3\end{array}_1^{1/2}|\beta |^6(1q^2|\beta |^2)(1|\beta |^2)(1q_1^2|\beta |^2)(1|\beta |^2)^{1/2}\hfill \end{array}$$
or
$$m(s+W^+u)=q^6𝒞\begin{array}{c}6\\ 3\end{array}_1^{1/2}|\beta |^6f(1)f(0)f(1)(1|\beta |^2)^{1/2}$$
(11.11)
The corresponding matrix element for
$$d+W^+c$$
is
$$\begin{array}{ccc}\hfill m(d+W^+c)& =& 1|\overline{D}_{\frac{3}{2}\frac{1}{2}}^{3/2}D_{30}^3(a+\overline{a})D_{\frac{3}{2}\frac{1}{2}}^{3/2}|0\hfill \\ & =& 𝒞1|(q_1a^2b)(q_1^3\begin{array}{c}6\\ 3\end{array}_1^{1/2}\overline{b}^3\overline{a}^3)(a+\overline{a})(ab^2)|0\hfill \\ & =& q_1^4𝒞\begin{array}{c}6\\ 3\end{array}_1^{1/2}1|a^2b(\overline{b}^3\overline{a}^3)\overline{a}(a\overline{b})|0\hfill \end{array}$$
(11.12)
Here
$$\begin{array}{ccc}\hfill 1|(a^2b)(\overline{b}^3\overline{a}^3\overline{a})ab^2|0& =& q^81|(b\overline{b}^3)a^2\overline{a}^3\overline{a}ab^2|0\hfill \\ & =& q^81|(b\overline{b}^3)(a^2\overline{a}^2)\overline{a}(\overline{a}a)b^2|0\hfill \\ & =& q^81|(b\overline{b}^3)(1b\overline{b})(1q^2b\overline{b})\overline{a}(1q_1^2\overline{b}b)b^2|0\hfill \\ & =& q^81|(b\overline{b}^3)(1b\overline{b})(1q^2b\overline{b})(1q_1^4\overline{b}b)\overline{a}b^2|0\hfill \\ & =& q^61|(b\overline{b})^3(1b\overline{b})(1q^2b\overline{b})(1q_1^4\overline{b}b)\overline{a}|0\hfill \\ & =& q^61|q^6|\beta |^6(1q^2|\beta |^2)(1q^4|\beta |^2)(1q_1^2|\beta |^2)(1|\beta |^2)^{1/2}|1\hfill \end{array}$$
(11.13)
by (8.1) and (8.4). Then by (9.5)
$$m(d+W^+c)=q^8𝒞\begin{array}{c}6\\ 3\end{array}_1^{1/2}|\beta |^6f(1)f(2)f(1)(1|\beta |^2)^{1/2}$$
(11.14)
By (11.11) and the preceding equation
$$\begin{array}{ccc}\hfill \frac{m(s+W^+u)}{m(d+W^+c)}& =& \frac{q^6}{q^8}\frac{f(1)f(0)f(1)}{f(1)f(2)f(1)}\hfill \\ & =& q^2\frac{f(0)}{f(2)}=q^2\frac{1|\beta |^2}{1q^4|\beta |^2}\hfill \end{array}$$
(11.15)
Since the corresponding ratio in the Kobayashi-Maskawa matrix is very close to unity, Eq. (11.15) again implies
$$q1$$
(11.16)
One also finds by (10.4) and (11.14)
$$\frac{m(d+W^+u)}{m(d+W^+c)}=q_1^4\frac{f(0)}{f(2)}\frac{1}{(1|\beta |^2)^{1/2}}$$
(11.17)
By (11.15)
$$\begin{array}{ccc}\hfill \frac{m(d+W^+u)}{m(d+W^+c)}& =& q_1^2\frac{m(s+W^+u)}{m(d+W^+c)}\frac{1}{(1|\beta |^2)^{1/2}}\hfill \\ & & \frac{1}{(1|\beta |^2)^{1/2}}\hfill \end{array}$$
(11.18)
if we set $`q1`$ according to (11.16).
From the Kobayashi-Maskawa matrix we have
$$\frac{m(d+W^+u)}{m(d+W^+c)}=4.054$$
(11.19)
By (11.18) and (11.19)
$$|\beta |.968$$
(11.20)
Again $`q`$ and $`|\beta |`$ are approximately unity.
We next compare with the small Kobayashi-Maskawa entry $`(cb)`$
$$\begin{array}{ccc}\hfill m(b+W^+c)& =& 1|\overline{D}_{\frac{3}{2}\frac{1}{2}}^{3/2}D_{30}^3(a+\overline{a})D_{\frac{3}{2}\frac{1}{2}}^{3/2}|2\hfill \\ & =& 𝒞1|(q_1a^2b)(q_1^3\begin{array}{c}6\\ 3\end{array}_1^{1/2}\overline{b}^3\overline{a}^3)(a+\overline{a})(ab^2)|2\hfill \\ & =& q_1^4𝒞\begin{array}{c}6\\ 3\end{array}_1^{1/2}1|a^2b\overline{b}^3\overline{a}^3(a+\overline{a})ab^2|2\hfill \end{array}$$
(11.21)
where (11.21) may be reduced as follows:
$$\begin{array}{ccc}\hfill 1|a^2b\overline{b}^3\overline{a}^3(a+\overline{a})ab^2|2& =& q^81|b\overline{b}^3a^2\overline{a}^3a^2b^2|2\hfill \\ & =& q^{10}1|(b\overline{b})^3(a^2\overline{a}^2)(\overline{a}a)a|2\hfill \\ & =& q^{16}|\beta |^6(1q^2|\beta |^2)(1q^4|\beta |^2)(1|\beta |^2)(1q^2|\beta |^2)^{1/2}\hfill \end{array}$$
(11.22)
Then
$$m(b+W^+c)=q^{12}𝒞\begin{array}{c}6\\ 3\end{array}_1^{1/2}|\beta |^6(1q^4|\beta |^2)(1|\beta |^2)(1q^2|\beta |^2)^{3/2}$$
(11.23)
By (10.4) and the preceding equation
$$\frac{m(b+W^+c)}{m(d+W^+u)}=q^8\frac{1q^4|\beta |^2}{1q_1^2|\beta |^2}(1q^2|\beta |^2)^{1/2}$$
(11.24)
By comparing with the Kobayashi-Maskawa matrix one has
$$q^8\frac{1q^4|\beta |^2}{1q_1^2|\beta |^2}(1q^2|\beta |^2)^{1/2}=.0617$$
(11.25)
If one sets $`q=1`$ as in the previous case, one finds:
$$\begin{array}{ccc}\hfill (1|\beta |^2)^{1/2}& =& .0617\hfill \\ \hfill |\beta |& =& .998\hfill \end{array}$$
(11.26)
The approximate solution for this case is then $`(q,\beta )(1.00,.998)`$.
Finally the $`(ts)`$ element according to our model is
$$m(s+W^+t)=2|\overline{D}_{\frac{3}{2}\frac{1}{2}}^{3/2}D_{30}^3(a+\overline{a})D_{\frac{3}{2}\frac{1}{2}}^{3/2}|1$$
(11.27)
or
$$m(s+W^+t)=q_1^4𝒞\begin{array}{c}6\\ 3\end{array}_1^{1/2}2|a^2b\overline{b}^3\overline{a}^3\overline{a}(ab^2)|1$$
(11.28)
We find
$$\begin{array}{ccc}\hfill \frac{m(s+W^+t)}{m(b+W^+c)}& =& q^2\frac{(1q^6|\beta |^2)(1q^4|\beta |^2)(1|\beta |^2)(1q^2|\beta |^2)^{1/2}}{(1q^4|\beta |^2)(1|\beta |^2)(1q^2|\beta |^2)^{3/2}}\hfill \\ & =& q^2\frac{1q^6|\beta |^2}{1q^2|\beta |^2}\hfill \end{array}$$
(11.29)
If $`q=1`$, the matrix elements, $`(ts)`$ and $`(cb)`$, are equal. The vanishing $`(ts)`$ entry in the KM matrix may be compatible with (11.29) and the small $`(cb)`$ value already computed.
The $`(ub)`$ and $`(dt)`$ matrix elements vanish for (11.4) and (11.5) and also in the approximate KM matrix (11.1).
We have ignored the phase factors appearing in the empirical matrix elements as well as the phase factors stemming from $`\lambda _n`$ and $`\mu _n`$ (Eqs. (8.10) and (8.11)) and therefore appearing in the computed matrix elements as well. The results of this section are summarized in Table 3.
Results of Comparing with K.M. Matrix (without Cabibbo-GIM Angles).
$$\begin{array}{ccccc}\underset{¯}{\mathrm{Ratio}\mathrm{of}\mathrm{Matrix}\mathrm{Elements}}& \underset{¯}{\mathrm{Ratio}\mathrm{from}\mathrm{Model}}& \underset{¯}{\mathrm{K}.\mathrm{M}.\mathrm{Matrix}}& \underset{¯}{q}& \underset{¯}{\beta }\\ \frac{m(s+W^+u)}{m(d+W^+c)}& q^2\left(\frac{1|\beta |^2}{1q^4|\beta |^2}\right)& \frac{.23}{.24}=.958& 1& 1\\ & & & & \\ \frac{m(d+W^+u)}{m(d+W^+c)}& \frac{m(s+W^+u)}{m(d+W^+c)}\frac{q^2}{(1|\beta |^2)^{1/2}}& & & \\ & =.958\frac{q^2}{(1|\beta |^2)^{1/2}}& \frac{.973}{.24}=4.05& 1& .968\\ & & & & \\ \frac{m(b+W^+c)}{m(d+W^+u)}& q^8\frac{1q^4|\beta |^2}{1q^2|\beta |^2}(1q^2|\beta |^2)^{1/2}& \frac{.06}{.973}=.0617& 1& .998\\ & & & & \\ \frac{m(s+W^+t)}{m(b+W^+c)}& q^2\frac{1q^6|\beta |^2}{1q^2|\beta |^2}& 1& 1& 1\end{array}$$
Table 3.
Compare with $`(q=1,|\beta |=.707)`$ for lepton-neutrino production.
In comparing with the KM matrix we have found that $`(q,\beta )`$ remain stable and close to unity.
## 12 Neutral Couplings.
Lepton-Lepton Interactions.
(a) Mediated by $`W_3:`$
The matrix element for
$$\mathrm{}(n)+W_3\mathrm{}^{}(n)$$
is by (8.2), (8.3), and (8.4)
$$\begin{array}{ccc}\hfill n|\overline{D}_{\frac{3}{2}\frac{3}{2}}^{3/2}D_{00}^3D_{\frac{3}{2}\frac{3}{2}}^{3/2}|n& =& n|\overline{a}^3f_3(b\overline{b})a^3|n\hfill \\ & =& n|\overline{a}^3a^3|nf_3(q^{6+2n}|\beta |^2)\hfill \end{array}$$
(12.1)
(b) Mediated by $`W_0:`$
The corresponding matrix element for
$$\mathrm{}(n)+W_0\mathrm{}^{}(n)$$
is by (8.2), (8.3), and (8.5)
$$\begin{array}{ccc}\hfill n|\overline{D}_{\frac{3}{2}\frac{3}{2}}^{3/2}D_{11}^3D_{\frac{3}{2}\frac{3}{2}}^{3/2}|n& =& n|\overline{a}^3\widehat{f}_0(b\overline{b})\overline{b}^2a^3|n\hfill \\ & =& q^6n|\overline{a}^3a^3|n\widehat{f}_0(q^{6+2n}|\beta |^2)q^{2n}\overline{\beta }^2\hfill \end{array}$$
(12.2)
where by (8.6)
$$f_0(b,\overline{b})=\widehat{f}_0(b\overline{b})\overline{b}^2$$
Then the ratio of the $`W_3`$ to the $`W_0`$ matrix elements is
$$\frac{n|\overline{D}_{\frac{3}{2}\frac{3}{2}}^{3/2}D_{00}^3D_{\frac{3}{2}\frac{3}{2}}^{3/2}|n}{n|\overline{D}_{\frac{3}{2}\frac{3}{2}}^{3/2}D_{11}^3D_{\frac{3}{2}\frac{3}{2}}^{3/2}|n}=\frac{f_3(q^{6+2n}|\beta |^2)q_1^{2n6}}{\widehat{f}_0(q^{6+2n}|\beta |^2)\overline{\beta }^2}$$
(12.3)
Since we carry over the Weinberg-Salam relation between $`(W^3,W^0)`$ and $`(A,Z)`$, we have
$$n|A|n=n|\overline{a}^3a^3|n[\overline{\beta }^2q^6C_0\widehat{f}_0(q^{6+2n}|\beta |^2)\mathrm{cos}\theta +C_3f_3(q^{6+2n}|\beta |^2)\mathrm{sin}\theta ]$$
(12.4)
$$n|Z|n=n|\overline{a}^3a^3|n[\overline{\beta }^2q^6C_0\widehat{f}_0(q^{6+2n}|\beta |^2)\mathrm{sin}\theta +C_3f_3(q^{6+2n}|\beta |^2)\mathrm{cos}\theta ]$$
(12.5)
Neutrino-Neutrino Interactions.
(a) Mediated by $`W_3`$.
$$\begin{array}{ccc}\hfill n|\overline{D}_{\frac{3}{2}\frac{3}{2}}^{3/2}D_{00}^3D_{\frac{3}{2}\frac{3}{2}}^{3/2}|n& =& q_1^6n|b^3f_3(b\overline{b})\overline{b}^3|n\hfill \\ & =& q^{6n6}|\beta |^6f_3(q^{2n}|\beta |^2)\hfill \end{array}$$
(12.6)
(b) Mediated by $`W_0`$.
$$n|\overline{D}_{\frac{3}{2}\frac{3}{2}}^{3/2}D_{11}^3D_{\frac{3}{2}\frac{3}{2}}^{3/2}|n=q^{8n6}|\beta |^6\widehat{f}_0(q^{2n}|\beta |^2)\overline{\beta }^2$$
(12.7)
The ratio of these matrix elements is
$$\frac{n|\overline{D}_{\frac{3}{2}\frac{3}{2}}^{3/2}D_{00}^3D_{\frac{3}{2}\frac{3}{2}}^{3/2}|n}{n|\overline{D}_{\frac{3}{2}\frac{3}{2}}^{3/2}D_{11}^3D_{\frac{3}{2}\frac{3}{2}}^{3/2}|n}=q^{2n}\frac{f_3(q^{2n}|\beta |^2)}{\widehat{f}_0(q^{2n}|\beta |^2)}\frac{1}{\overline{\beta }^2}$$
(12.8)
Let
$`\widehat{f}_0^{}`$ $`=`$ $`C_0\widehat{f}_0`$
$`f_3^{}`$ $`=`$ $`C_3f_3`$ (12.9)
The matrix elements for $`A`$ and $`Z`$ are
$$n|A|n=q^{6n6}|\beta |^6[q^{2n}\widehat{f}_0^{}(q^{2n}|\beta |^2)\overline{\beta }^2\mathrm{cos}\theta +f_3^{}(q^{2n}|\beta |^2)\mathrm{sin}\theta ]$$
(12.10)
$$n|Z|n=q^{6n6}|\beta |^6[q^{2n}\widehat{f}_0^{}(q^{2n}|\beta |^2)\overline{\beta }^2\mathrm{sin}\theta +f_3^{}(q^{2n}|\beta |^2)\mathrm{cos}\theta ]$$
(12.11)
Since $`n|A|n`$ must vanish for neutrinos one demands
$$q^{2n}\widehat{f}_0^{}(q^{2n}|\beta |^2)\overline{\beta }^2\mathrm{cos}\theta +f_3^{}(q^{2n}|\beta |^2)\mathrm{sin}\theta =0$$
(12.12)
or
$$\mathrm{tan}\theta =q^{2n}\left(\frac{\widehat{f}_0^{}(q^{2n}|\beta |^2)}{f_3^{}(q^{2n}|\beta |^2)}\right)\overline{\beta }^2$$
(12.13)
For the neutrino family we set $`\beta =i|\beta |`$, since we take $`\mathrm{tan}\theta `$ positive.
The requirement (12.12) is equivalent to the requirement of standard theory that the photon interacts only with electric charge and not at all with hypercharge. Eq. (12.13) states that $`\mathrm{tan}\theta `$ is so chosen that $`W_3`$ and $`W_0`$ are mixed so that the photon has no role in the weak interactions.
Then
$$q^{2n}\frac{\widehat{f}_0^{}(q^{2n}|\beta |^2)|\beta |^2}{f_3^{}(q^{2n}|\beta |^2)}=\mathrm{tan}\theta $$
(12.14)
One requires that the Weinberg angle be independent of $`n`$. Then $`q=1`$ and
$$\begin{array}{ccc}\hfill \frac{C_0}{C_3}\frac{\widehat{f}_0(|\beta |^2)|\beta |^2}{f_3(|\beta |^2)}& =& \mathrm{tan}\theta \hfill \\ & =& g^{}/g\hfill \\ & =& .528\hfill \end{array}$$
(12.15)
## 13 Charge-Retention Interactions for Quarks.
We consider first the couplings of the $`(dsb)`$ family to $`W_3`$ and $`W_0`$.
(a) Mediated by $`W_3`$:
$$Q(n)+W_3Q(n)^{}$$
(13.1)
The matrix element for (13.1) is
$$\begin{array}{ccc}\hfill n|\overline{D}_{\frac{3}{2}\frac{1}{2}}^{3/2}D_{00}^3D_{\frac{3}{2}\frac{1}{2}}^{3/2}|n& =& 𝒞^{}n|(\overline{b}^2\overline{a})f_3(b\overline{b})(ab^2)|n\hfill \\ & =& 𝒞^{}n|\overline{b}^2(\overline{a}a)f_3(q_1^2b\overline{b})b^2|n\hfill \\ & =& 𝒞^{}n|(\overline{b}b)^2(\overline{a}a)|nn|f_3(q_1^2(b\overline{b})|n\hfill \end{array}$$
(13.2)
with
$$𝒞^{}=\left(\mathrm{\Delta }_{\frac{3}{2}\frac{1}{2}}^{3/2}\begin{array}{c}3\\ 1\end{array}_1\right)^2$$
(b) Mediated by $`W_0`$:
$$Q(n)+W_0Q(n)^{}$$
(13.3)
with the following matrix element:
$$\begin{array}{ccc}\hfill n|\overline{D}_{\frac{3}{2}\frac{1}{2}}^{3/2}D_{11}^3D_{\frac{3}{2}\frac{1}{2}}^{3/2}|n& =& 𝒞^{}n|\overline{b}^2\overline{a}\widehat{f}_0(b\overline{b})\overline{b}^2ab^2|n\hfill \\ & =& 𝒞^{}q_1^2n|(\overline{b}b)^2(\overline{a}a)|nn|\widehat{f}_0(q_1^2\overline{b}b)\overline{b}^2|n\hfill \end{array}$$
(13.4)
Then the ratio of the $`W_3`$ to the $`W_0`$ matrix elements is
$$\frac{n|\overline{D}_{\frac{3}{2}\frac{1}{2}}^{3/2}D_{00}^3D_{\frac{3}{2}\frac{1}{2}}^{3/2}|n}{n|\overline{D}_{\frac{3}{2}\frac{1}{2}}^{3/2}D_{11}^3D_{\frac{3}{2}\frac{1}{2}}^{3/2}|n}=q^2\frac{n|f_3(q_1^2b\overline{b})|n}{n|\widehat{f}_0(q_1^2b\overline{b})\overline{b}^2|n}$$
(13.5)
and we also have
$$n|A|n=n|(\overline{b}b)^2\overline{a}a|n[q_1^2n|\widehat{f}_0^{}(q_1^2\overline{b}b)\overline{b}^2|n\mathrm{cos}\theta +n|f_3^{}(q_1^2\overline{b}b)|n\mathrm{sin}\theta ]$$
(13.6)
$$n|Z|n=n|(\overline{b}b)^2\overline{a}a|n[q_1^2n|\widehat{f}_0^{}(q_1^2\overline{b}b)\overline{b}^2|n\mathrm{sin}\theta +n|f_3^{}(q_1^2\overline{b}b)|n\mathrm{cos}\theta ]$$
(13.7)
We next consider the corresponding couplings of the $`uct`$-family
(a) Mediated by $`W_3`$:
The matrix elements are
$$\begin{array}{ccc}\hfill n|\overline{D}_{\frac{3}{2}\frac{1}{2}}^{3/2}D_{00}^3D_{\frac{3}{2}\frac{1}{2}}^{3/2}|n& =& 𝒞^{}q_1^2n|a^2bf_3(b\overline{b})\overline{b}\overline{a}^2|n\hfill \\ & =& 𝒞^{}q_1^2q^4n|(b\overline{b})(a^2\overline{a}^2)|nn|f_3(q^4b\overline{b})|n\hfill \end{array}$$
(13.8)
(b) Mediated by $`W_0`$:
$$\begin{array}{ccc}\hfill n|\overline{D}_{\frac{3}{2}\frac{1}{2}}^{3/2}D_{11}^3D_{\frac{3}{2}\frac{1}{2}}^{3/2}|n& =& 𝒞^{}q_1^2n|(a^2b)(\widehat{f}_0(b\overline{b})\overline{b}^2)\overline{b}\overline{a}^2|n\hfill \\ & =& 𝒞^{}q_1^2q^8n|(b\overline{b})(a^2\overline{a}^2)|nn|\widehat{f}_0(q^4b\overline{b})\overline{b}^2|n\hfill \end{array}$$
(13.9)
Then the ratio of the $`W_3`$ to the $`W_0`$ matrix elements is
$$\frac{n|\overline{D}_{\frac{3}{2}\frac{1}{2}}^{3/2}D_{00}^3D_{\frac{3}{2}\frac{1}{2}}^{3/2}|n}{n|\overline{D}_{\frac{3}{2}\frac{1}{2}}^{3/2}D_{11}^3D_{\frac{3}{2}\frac{1}{2}}^{3/2}|n}=q^4\frac{n|f_3(q^4b\overline{b})|n}{n|\widehat{f}_0(q^4b\overline{b})\overline{b}^2|n}$$
(13.10)
and the $`A`$ and $`Z`$ matrix elements are
$$n|A|n=n|(b\overline{b})(a^2\overline{a}^2)|n[q^8n|\widehat{f}_0^{}(q^4b\overline{b})\overline{b}^2|n\mathrm{cos}\theta +q^4n|f_3^{}(q^4\overline{b}b)|n\mathrm{sin}\theta ]$$
(13.11)
$$n|Z|n=n|(b\overline{b})(a^2\overline{a}^2)|n[q^8n|\widehat{f}_0^{}(q^4b\overline{b})\overline{b}^2|n\mathrm{sin}\theta +q^4n|f_3^{}(q^4\overline{b}b)|n\mathrm{cos}\theta ]$$
(13.12)
## 14 Decays of the $`Z^0`$.
Decays of the $`Z^0`$ into Leptons.
The rates of these decays are described by
$$\mathrm{\Gamma }_n(Z^0\mathrm{}+\overline{\mathrm{}})\left|n|\overline{D}_{\frac{3}{2}\frac{3}{2}}^{3/2}(C_3D_{00}^3\mathrm{cos}\theta C_0D_{11}^3\mathrm{sin}\theta )D_{\frac{3}{2}\frac{3}{2}}^{3/2}|n\right|^2$$
or
$$=|a_n\mathrm{cos}\theta b_n\mathrm{sin}\theta |^2$$
where
$$a_n=C_3n|\overline{D}_{\frac{3}{2}\frac{3}{2}}^{3/2}D_{00}^3D_{\frac{3}{2}\frac{3}{2}}^{3/2}|n$$
(14.1)
$$b_n=C_0n|\overline{D}_{\frac{3}{2}\frac{3}{2}}^{3/2}D_{11}^3D_{\frac{3}{2}\frac{3}{2}}^{3/2}|n$$
(14.2)
Then
$$\begin{array}{ccc}\hfill \frac{\mathrm{\Gamma }_{n+1}}{\mathrm{\Gamma }_n}& =& \left|\frac{a_{n+1}\mathrm{cos}\theta b_{n+1}\mathrm{sin}\theta }{a_n\mathrm{cos}\theta b_n\mathrm{sin}\theta }\right|^2\hfill \\ & & \\ & =& \left|\frac{1\frac{b_{n+1}}{a_{n+1}}\mathrm{tan}\theta }{1\frac{b_n}{a_n}\mathrm{tan}\theta }\right|^2\left|\frac{a_{n+1}}{a_n}\right|^2\hfill \end{array}$$
(14.3)
One finds
$$a_n=n|\overline{a}^3a^3|nf_3^{}(q^{6+2n}|\beta |^2)$$
(14.4)
$$b_n=q^6n|\overline{a}^3a^3|n\widehat{f}_0^{}(q^{6+2n}|\beta |^2)\overline{\beta }^2$$
(14.5)
Then
$$\frac{b_n}{a_n}=q^6\frac{\widehat{f}_0^{}(q^{6+2n}|\beta |^2)\overline{\beta }^2}{f_3^{}(q^{6+2n}|\beta |^2)}$$
(14.6)
$$\frac{a_{n+1}}{a_n}=\frac{f_3^{}(q^{4+2n}|\beta |^2)}{f_3^{}(q^{6+2n}|\beta |^2)}\frac{n+1|\overline{a}^3a^3|n+1}{n|\overline{a}^3a^3|n}$$
(14.7)
If $`q=1`$, then by (14.7)
$$\left|\frac{a_{n+1}}{a_n}\right|^2=1$$
(14.8)
If $`q=1`$, then by (14.6)
$$\frac{b_{n+1}}{a_{n+1}}=\frac{b_n}{a_n}$$
(14.9)
and by (14.3)
$$\frac{\mathrm{\Gamma }_{n+1}}{\mathrm{\Gamma }_n}=1$$
(14.10)
The measured rates are<sup>7</sup>
$$\frac{\mathrm{\Gamma }(\mu ^+\mu ^{})}{\mathrm{\Gamma }(e^+e^{})}=1.0009$$
(14.11)
$$\frac{\mathrm{\Gamma }(\tau ^+\tau ^{})}{\mathrm{\Gamma }(e^+e^{})}=1.0019$$
(14.12)
The measured rates are thus compatible with $`q1`$.
Decay of $`Z^0`$ into Neutrinos.
We now have
$$\frac{\mathrm{\Gamma }(Z^0\nu _e+\overline{\nu }_e)}{\mathrm{\Gamma }(Z^0e+\overline{e})}=\left|\frac{n|\overline{D}_{\frac{3}{2}\frac{3}{2}}^{3/2}(C_3D_{00}^3\mathrm{cos}\theta C_0D_{11}^3\mathrm{sin}\theta )D_{\frac{3}{2}\frac{3}{2}}^{3/2}|n}{n|\overline{D}_{\frac{3}{2}\frac{3}{2}}^{3/2}(C_3D_{00}^3\mathrm{cos}\theta C_0D_{11}^3\mathrm{sin}\theta )D_{\frac{3}{2}\frac{3}{2}}^{3/2}|n}\right|^2$$
(14.13)
Let
$$R=\frac{\mathrm{\Gamma }(Z^0\nu _e+\overline{\nu }_e)}{\mathrm{\Gamma }(Z^0e+\overline{e})}$$
(14.14)
Then
$$R=\left|\frac{|\beta |^6}{(1|\beta |^2)^3}\right|^2\text{if}q=1$$
(14.15)
The measured value of $`R`$ is 1.98.<sup>7</sup> Then
$$\frac{|\beta |^2}{1|\beta |^2}=(1.98)^{1/6}$$
(14.16)
and
$$|\beta |=.727$$
(14.17)
Compare $`(q,\beta )`$ = (1,.727) with $`(q,\beta )`$ = (1,.707), the values found earlier in the discussion of lepton-neutrino production by charged $`W`$ (with the assumption of the universal Fermi interaction).
Pair Production of (2/3) Quarks.
In this case
$$\begin{array}{ccc}\hfill Z^0& & u+\overline{u}\hfill \\ & & c+\overline{c}\hfill \\ & & t+\overline{t}\hfill \end{array}$$
(14.18)
The relevant matrix element is
$$n|\overline{D}_{\frac{3}{2}\frac{1}{2}}^{3/2}(C_3D_{00}^3\mathrm{cos}\theta C_0D_{11}^3\mathrm{sin}\theta )D_{\frac{3}{2}\frac{1}{2}}^{3/2}|n$$
(14.19)
so that
$$\mathrm{\Gamma }_n|n|a^2\overline{a}^2|nq^{2n2}3_1|\beta |^2[(f_3^{}(q^{2n+4}|\beta |^2)\mathrm{cos}\theta f_0^{}(q^{2n+4}|\beta |^2)\mathrm{sin}\theta |^2$$
(14.20)
and
$$\frac{\mathrm{\Gamma }_{n+1}}{\mathrm{\Gamma }_n}=\left|\frac{n+1|a^2\overline{a}^2|n+1}{n|a^2\overline{a}^2|n}q^2\frac{[f_3^{}(q^{2n+6}|\beta |^2)\mathrm{cos}\theta f_0^{}(q^{2n+6}|\beta |^2)\mathrm{sin}\theta ]}{f_3^{}(q^{2n+4}|\beta |^2)\mathrm{cos}\theta f_0^{}(q^{2n+4}|\beta |^2)\mathrm{sin}\theta ]}\right|^2$$
(14.21)
If $`q1`$, there is little dependence of this ratio on $`n`$ and the three rates (14.22) are approximately equal.
Pair Production of (-1/3) Quarks.
Now
$$\begin{array}{ccc}\hfill Z^0& & d+\overline{d}\hfill \\ & & s+\overline{s}\hfill \\ & & b+\overline{b}\hfill \end{array}$$
(14.22)
with the matrix element
$$n|\overline{D}_{\frac{3}{2}\frac{1}{2}}^{3/2}[C_3D_{00}^3\mathrm{cos}\theta C_0D_{11}^3\mathrm{sin}\theta ]D_{\frac{3}{2}\frac{1}{2}}^{3/2}|n$$
(14.23)
so that
$$\mathrm{\Gamma }_n|q^{4n}|\beta |^43_1n|\overline{a}a|n[f_3^{}(q^{2n2}|\beta |^2)\mathrm{cos}\theta f_0^{}(q^{2n2}|\beta |^2)\mathrm{sin}\theta ]|^2$$
(14.24)
and
$$\frac{\mathrm{\Gamma }_{n+1}}{\mathrm{\Gamma }_n}=\left|\frac{n+1|\overline{a}a|n+1}{n|a\overline{a}|n}\frac{f_3^{}(q^{2n}|\beta |^2)\mathrm{cos}\theta f_0^{}(q^{2n}|\beta |^2)\mathrm{sin}\theta }{f_3^{}(q^{2n2}|\beta |^2)\mathrm{cos}\theta f_0^{}(q^{2n2}|\beta |^2)\mathrm{sin}\theta }\right|^2$$
(14.25)
If $`q1`$, the three rates (14.22) are again approximately equal.
Ratio of Pair Production of 2/3 and -1/3 Quarks by $`Z^0`$.
Let
$$R=\frac{\mathrm{\Gamma }\left(Z^0Q\left(\frac{1}{3}\right)+\overline{Q}\left(\frac{1}{3}\right)\right)}{\mathrm{\Gamma }\left(Z^0Q\left(\frac{2}{3}\right)+\overline{Q}\left(\frac{2}{3}\right)\right)}$$
(14.26)
By (13.7) and (13.12)
$$R=\left|q^{2n+2}|\beta |^2\frac{n|\overline{a}a|n}{n|a^2\overline{a}^2|n}\frac{[f_3^{}(q^{2n2}|\beta |^2)\mathrm{cos}\theta f_0^{}(q^{2n2}|\beta |^2)\mathrm{sin}\theta ]}{[f_3^{}(q^{2n+4}|\beta |^2)\mathrm{cos}\theta f_0^{}(q^{2n+4}|\beta |^2)\mathrm{sin}\theta ]}\right|^2$$
(14.27)
If $`q1`$,
$$R\left|\frac{|\beta |^2}{1|\beta |^2}\right|^2$$
(14.28)
By Ref. 6, the fractions of all decays of the $`Z^0`$ into $`(u\overline{u}+c\overline{c})/2`$ and $`(d\overline{d}+s\overline{s}+b\overline{b})/3`$ are 10.1% and 16.6% respectively. One may then estimate $`\beta `$ by setting
$$\left|\frac{|\beta |^2}{1|\beta |^2}\right|^2=\frac{16.6}{10.1}$$
(14.29)
Then
$$|\beta |=.750$$
(14.30)
One may compare $`(q,\beta )`$ for the following examples:
$$\begin{array}{cc}& \underset{¯}{q,\beta }\\ \mathrm{\Gamma }[W^+e^++\overline{\nu }_e]\hfill & (1,.707)\\ \mathrm{\Gamma }[Z^0e+\overline{e})/\mathrm{\Gamma }(Z^0\nu _e+\overline{\nu }_e)]\hfill & (1,.727)\\ \mathrm{\Gamma }\left[\left(Z^0Q\left(\frac{1}{3}\right)+\overline{Q}\left(\frac{1}{3}\right)\right)/\left(Z^0Q\left(\frac{2}{3}\right)+\overline{Q}\left(\frac{2}{3}\right)\right)\right]\hfill & (1,.750)\end{array}$$
(14.31)
In these tests the value of $`q`$ is simply assigned: $`q=1`$. Although a closer fit is possible if $`q`$ is allowed to vary, one already sees that $`\beta `$ is stable.
Relative Production of Lepton Pairs by $`A`$ and $`Z`$.
By (12.4) and (12.5)
$$\frac{n|A|n}{n|Z|n}=\frac{S+\mathrm{tan}\theta }{S\mathrm{tan}\theta +1}$$
(14.32)
where $`|n`$ is any lepton state and
$$S=q^6|\beta |^2\frac{\widehat{f}_0^{}(q^{6+2n}|\beta |^2)}{f_3^{}(q^{6+2n}|\beta |^2)}$$
(14.33)
since $`\overline{\beta }=\beta `$ for lepton states.
By (12.14)
$$\mathrm{tan}\theta =q^{2n}|\beta |^2\frac{\widehat{f}_0^{}(q^{2n}|\beta |^2)}{f_3^{}(q^{2n}|\beta |^2)}$$
(14.34)
and since $`\mathrm{tan}\theta `$ is independent of $`n`$, one has $`q=1`$ and by (14.33)
$$S=\mathrm{tan}\theta $$
(14.35)
Then by (14.32)
$$\frac{n|A|n}{n|Z|n}=\frac{2\mathrm{tan}\theta }{1\mathrm{tan}^2\theta }$$
(14.36)
in agreement with the Weinberg-Salam model for the ratio $`Q/Q^{}`$ of the electric to the hypercharge.
## 15 Covariant Derivative of Neutral States.
We are replacing the standard $`SU(2)_L\times U(1)`$ theory by a knot theory based on $`SU_q(2)_L`$ alone, i.e. we are assuming that the roles of charge and hypercharge in the standard theory can be carried by $`SU_q(2)`$ alone.
The transition from $`SU(2)\times U(1)`$ to $`SU_q(2)`$ may be partially described as follows:
$$g^{}W_0t_0+g\overline{W}\stackrel{}{t}\widehat{g}[C_0W_0D_{11}^3+C_3W_3D_{00}^3+C_{}W_{}D_{30}^3+C_+W_+D_{30}^3]$$
(15.1)
where the $`C_\tau `$ $`(\tau =0,3,,+)`$ are functions of $`q`$ and $`\beta `$. Here $`\widehat{g}`$ is the coupling constant of the $`SU_q(2)`$ theory while $`g`$ and $`g^{}`$ are the usual coupling constants of the $`SU(2)\times U(1)`$ model.
The neutral couplings in the knot theory are then described by
$$\widehat{g}[C_0W_0D_{11}^3+C_3W_3D_{00}^3]$$
(15.2)
By carrying over the relation between $`(W_0,W_3)`$ and $`(A,Z)`$ from the standard theory, we replace (15.2) by
$$\widehat{g}[𝒜A+𝒵Z]$$
(15.3)
where
$$𝒜=C_0\mathrm{cos}\theta D_{11}^3+C_3\mathrm{sin}\theta D_{00}^3$$
(15.4)
$$𝒵=C_0\mathrm{sin}\theta D_{11}^3+C_3\mathrm{cos}\theta D_{00}^3$$
(15.5)
Since $`𝒜`$ and $`𝒵`$ lie in the $`(b,\overline{b})`$ subalgebra, neutrino states are eigenstates of these operators. Since $`𝒜`$ is further restricted by the physical condition that photons do not interact with neutrinos, we have
$$\overline{\nu }𝒜\nu =\overline{\nu }𝒜^{}\nu =\overline{\nu }\nu 𝒜^{}=0$$
(15.6)
Then by (15.4)
$$C_0\mathrm{cos}\theta D_{11}^3+C_3\mathrm{sin}\theta D_{00}^3=0$$
(15.7)
and by (15.5)
$`𝒵\nu `$ $`=`$ $`[C_0\mathrm{sin}\theta D_{11}^3+C_3\mathrm{cos}\theta D_{00}^3]\nu `$ (15.8)
$`=`$ $`{\displaystyle \frac{C_3}{\mathrm{cos}\theta }}D_{00}^3\nu `$ (15.9)
The complete covariant derivative on a neutrino state, or any neutral state, is then by (15.1), (15.3), (15.6) and (15.9)
$$_\mu =_\mu +ig\left[C_{}W_{}D_{30}^3+C_+W_+D_{30}^3+\frac{C_3}{\mathrm{cos}\theta }D_{00}^3\right]$$
(15.10)
In addition
$$C_0=C_3\left(\frac{D_{00}^{}}{D_{11}^3}\right)\mathrm{tan}\theta $$
(15.11)
by (15.7).
## 16 Kinetic Energy of Neutral Higgs Scalar and Vector Masses.
Let us assign the neutral Higgs scalar to the lowest state of the trefoil previously identified with the neutrino family, namely (-3,2) carrying the representation $`D_{\frac{3}{2}\frac{3}{2}}^{3/2}`$. This trefoil lies entirely in the $`(b,\overline{b})`$ subalgebra as is also the case for the neutral vectors $`W_0`$ and $`W_3`$. We then have
$$\widehat{\phi }=\rho D_{\frac{3}{2}\frac{3}{2}}^{3/2}|0$$
(16.1)
where $`\widehat{\phi }`$ is the neutral Higgs scalar and the covariant derivative of this field is by (15.10)
$$_\mu \widehat{\phi }=\{_\mu \rho +i\widehat{g}\rho [C_{}W_\mu D_{30}^3+C_+W_{+\mu }D_{30}^3+\frac{C_3}{\mathrm{cos}\theta }Z_\mu D_{00}^3]\}D_{\frac{3}{2}\frac{3}{2}}^3|0$$
(16.2)
The kinetic energy of the lowest state of the neutral Higgs scalar is
$$\begin{array}{ccc}\hfill \overline{^\mu \widehat{\phi }}_\mu \widehat{\phi }& =& 0|\overline{D}_{\frac{3}{2}\frac{3}{2}}^{3/2}\{^\mu \rho _\mu \rho +\widehat{g}^2\rho ^2[|C_{}|^2W_{}^\mu W_\mu \overline{D}_{30}^3D_{30}^3+|C_+|^2W_+^\mu W_{+\mu }\overline{D}_{30}^3D_{30}^3\hfill \\ & & +\frac{|C_3|^2}{\mathrm{cos}^2\theta }Z^\mu Z_\mu \overline{D}_{00}^3D_{00}^3]\}D_{\frac{3}{2}\frac{3}{2}}^3|0\hfill \end{array}$$
(16.3)
where we have used orthogonality of the $`D_{mn}^3`$ that follows from
$$0|\mathrm{}\overline{a}^na^m\mathrm{}|0\delta ^{nm}$$
(16.4)
Then
$$\overline{^\mu \widehat{\phi }}_\mu \widehat{\phi }=I^\mu \rho _\mu \rho +\widehat{g}^2\rho ^2[II|C_{}|^2W_{}^\mu W_\mu +III|C_+|^2W_+^\mu W_{+\mu }+\frac{IV|C_3|^2}{\mathrm{cos}^2\theta }Z^\mu Z_\mu ]$$
(16.5)
where
$$\begin{array}{ccc}\hfill I& =& 0|\overline{D}_{\frac{3}{2}\frac{3}{2}}^{3/2}D_{\frac{3}{2}\frac{3}{2}}^{3/2}|0\hfill \\ \hfill II& =& 0|\overline{D}_{\frac{3}{2}\frac{3}{2}}^{3/2}(\overline{D}_{30}^3D_{30}^3)D_{\frac{3}{2}\frac{3}{2}}^{3/2}|0\hfill \\ \hfill III& =& 0|\overline{D}_{\frac{3}{2}\frac{3}{2}}^{3/2}(\overline{D}_{30}^3D_{30}^3)D_{\frac{3}{2}\frac{3}{2}}^{3/2}|0\hfill \\ \hfill IV& =& 0|\overline{D}_{\frac{3}{2}\frac{3}{2}}^{3/2}(\overline{D}_{00}^3D_{00}^3)D_{\frac{3}{2}\frac{3}{2}}^{3/2}|0\hfill \end{array}$$
(16.6)
To agree with the standard theory we now require
$$^\mu \widehat{\phi }_\mu \widehat{\phi }=^\mu \overline{\rho }_\mu \overline{\rho }+\widehat{g}^2\overline{\rho }^2[W_{}^\mu W_\mu +W_+^\mu W_{+\mu }+\frac{1}{\mathrm{cos}^2\theta }Z^\mu Z_\mu ]$$
(16.7)
where
$$\begin{array}{ccc}& & \overline{\rho }=I^{1/2}\rho \hfill \\ & & \frac{II}{I}|C_{}|^2=1\hfill \\ & & \frac{III}{I}|C_+|^2=1\hfill \\ & & \frac{IV}{I}|C_3|^2=1\hfill \end{array}$$
(16.8)
The $`I,II,III`$ and $`IV`$ are all explicit functions of $`(q,\beta )`$.
The four coefficients $`(C_{},C_+,C_3,C_0)`$ then follow from (16.6), (16.8) and (15.11):
$$\begin{array}{ccc}\hfill C_{}& =& \left[\begin{array}{c}6\\ 3\end{array}_1|\beta |^6\underset{1}{\overset{3}{}}(1q_1^{2t}|\beta |^2)\right]^{1/2}\hfill \\ \hfill C_+& =& \left[\begin{array}{c}6\\ 3\end{array}_1|\beta |^6\underset{0}{\overset{2}{}}(1q^{2s}|\beta |^2)\right]^{1/2}\hfill \\ \hfill C_3& =& f_3(|\beta |^2)^1\hfill \\ \hfill C_0& =& \frac{1}{f_0(|\beta |^2)}\mathrm{tan}\theta \hfill \end{array}$$
(16.9)
The mass relations between the neutral and charged vectors that follow from (16.7) are the same as for the standard theory. In order that the vertex functions be consistent with these relations, the vertex factors must be supplied with the $`(C_{},C_+,C_3,C_0)`$ given by (16.9).
## 17 Discussion.
We have been able to organize a class of data relevant to, but also not accessible from the standard theory. Although this model does not permit one to calculate absolute masses and reaction rates, it does provide a simple frame that describes fermionic spectra and reaction rates, and emerges quite naturally from the $`q`$-electroweak theory. The model may be fine-tuned and may be useful as a phenomenological model. To go further at a deeper level, one must be able to construct an effective field theory.
The relation of the knot model based on $`q`$-electroweak to standard electroweak resembles the relation of the Schrödinger to the Hamilton-Jacobi equation insofar as one adjoins in both cases a state space not present in the original description. Like the wave equation, the knot model may be applied in a variety of contexts. When the Schrödinger equation is applied to a single atom, or to molecules or other systems of arbitrary complexity, it has to be modified by changing the appropriate parameters. In hydrogenic systems, for example, the wave equation is applied with differing values of $`Z`$ and $`m`$. When the $`q`$-knot model is similarly applied to answer quite different questions, such as the masses of the fermions or reaction rates among them, it also has to be appropriately modified by choosing different values of $`q`$ and $`\beta `$. In every case the same algebra is used just as in every case the same wave equation is used.
We have computed only relative masses and relative rates. Given this restriction we find that $`q`$ and $`\beta `$ differ markedly from unity in the expressions for the mass ratios but are very close to unity in the corresponding expressions for relative rates.
If one regards $`SU_q(2)`$ as a fundamental symmetry, it may be possible to regard $`q`$ as a new constant with a single value that comes out differently in different contexts where external influences such as the gluon field have been ignored. Alternatively, $`q`$ may be regarded as a running coupling constant, where $`\beta `$ and $`m_n(w,r)`$ determine the energy scale.
The model in its present form predicts a fourth generation of fermions as well as a neutrino mass spectrum. In applications to fermionic mass spectra the parameters of the model ($`q`$ and $`\beta `$) have been fixed by two data $`(M,m)`$. If a fourth generation is found or not found in the neighborhood predicted by the model, then the model can be refined.
In the standard theory, and therefore here as well, there is no attempt to go beyond a provisional expression $`(\overline{\psi }\phi \psi )`$ for the fermionic masses, i.e. $`\overline{\psi }\phi \psi `$ could just as well be replaced by $`\overline{\psi }`$ $`F(\phi )\psi `$. There is no difficulty in cutting off the mass spectrum at three generations without changing the essential structure of the model.
The neutrino mass spectrum is also a strong constraint on the model; at present the data on this spectrum are compatible with $`q1`$. In applications to fermionic currents, both in the lepton-neutrino sector and in the Kobayashi-Maskawa sector, the data are compatible with $`q1`$.
The form of the vector coupling is restricted in the lepton sector by lepton conservation and by the Universal Fermi Interaction. In the quark sector it is restricted by the Kobayashi-Maskawa matrix. All of these restrictions can be satisfied approximately by the simple model described in this paper, but the model can be refined as more empirical input is utilized. Since gluon and gravitational couplings are not explicitly included, one may tentatively regard the deviation of $`q`$ from unity as a measure of their influence.
References.
1. R. J. Finkelstein, hep-th/0408218, to appear in Int. J. Mod. Phys. (A).
2. R. J. Finkelstein, hep-th/0110075, Lett. Math. Phys. 62, 199-210 (2002).
3. S. Eidelman et al., Phys. Lett. B592, 1 (2004).
4. V. Barger et al., Phys. Lett. B595, 55 (2004).
5. A. C. Cadavid and R. J. Finkelstein, J. Math. Phys. 36, 1912 (1995).
6. Phys. Rev. D Part I, Review of Particle Physics, 66 (2002). |
warning/0507/astro-ph0507216.html | ar5iv | text | # Structure and stellar content analysis of the open cluster M 11 with 2MASS photometry
## 1 Introduction
The open cluster M 11 (NGC 6705, Mel 213, Cr 391, OCl 76 - Alter et al. (1970)) is a concentrated, populous stellar system projected on the Scutum Cloud towards the central part of the Galactic disk \[$`\alpha (2000)=18\mathrm{h}51\mathrm{m}05\mathrm{s}`$, $`\delta (2000)=6^{}16\mathrm{}01\mathrm{}`$, $`l=27.30^{}`$, $`b=2.77^{}`$\]. Although dark clouds permeate the sky on the cluster direction, it is situated in a clear area characterized by a relatively low interestelar extintion nearby the Sagittarius arm. Were it a low surface brightness cluster it would be probably missed by surveys due to the rich field from the Galactic background stars. For several reasons, M 11 has captured attention over the years, not only for its intrinsic properties, but also for its contribution to the understanding of chemical and dynamical Galactic evolution.
Being closer to the Galactic center than the solar radius, M 11 suffers from relatively stronger tidal effects, as well as more frequent interactions with molecular clouds. The WEBDA database (Mermilliod 1996) provides a distance from the sun $`\mathrm{d}_{}=1877`$ pc, reddening $`\mathrm{E}(\mathrm{B}\mathrm{V})=0.426`$, apparent distance modulus $`(\mathrm{V}\mathrm{M}_\mathrm{V})=12.69`$, age t$`=200`$ Myr and metallicity $`[\mathrm{Fe}/\mathrm{H}]=0.13`$.
McNamara & Sanders (1977) studied proper motions in M 11 and obtained a velocity dispersion $`\sigma _\mathrm{v}=2.9\mathrm{km}\mathrm{s}^1`$, and an observed mass of $`3000`$ M. The inner cluster region may have isotropic orbits, while the orbits in the outer parts are probably eccentric with larger velocities in the radial direction.
Mathieu (1984) carried out a comprehensive analysis of M 11 based on proper motion and membership probability data (from McNamara et al. 1977) as well as photographic photometry reaching $`\mathrm{V}=20`$ and $`\mathrm{B}=21`$. By studying the cluster luminosity function out to a radius of 10’, evidence was found that inside 2’ the luminosity function is flatter than for the outer region, implying mass segregation. The total observed mass estimated inside the radius of 10’ and considering stellar masses down to $`0.7\mathrm{M}_{}`$ was $`4671\mathrm{M}_{}`$.
Santos et al. (1990) inferred the cluster overall mass function (MF) slope $`1<\chi <2.4`$ using a population synthesis method and the integrated spectrum aided by the HR diagram. The cluster visible light is dominated by the upper main sequence and turnoff stars (B6-A2).
Nilakshi et al. (2002) studied the spatial structure of a large sample of open clusters using photometric data from the DSS. For M 11 they derived a core radius of $`0.72\pm 0.10`$ pc, in agreement with the one obtained by Mathieu (1984).
Recently, Bonatto & Bica (2005), Bonatto et al. (2005), Bica et al. (2004), Bonatto & Bica (2003) and references therein undertook a systematic study of open cluster parameters, structure and other fundamental properties employing 2MASS photometry, making use of a spatial coverage as large as necessary for each case. For M 11, a deep UBVRI CCD study was carried out by Sung et al. (1999), including a spatial dependance of the MF. We intend to compare the performances of these optical CCD data and the 2MASS photometry. This is crucial for future cluster studies as 2MASS becomes widely used.
In the present work we explore M 11 with 2MASS photometry. In Sect. 2 the 2MASS photometry is presented. In Sect. 3 the cluster parameters are discussed. The cluster structure is analyzed in Sect. 4. Luminosity and mass functions are discussed in Sect 5. Concluding remarks are given in Sect. 6.
## 2 Database: 2MASS photometry
The 2MASS catalogue (Skrutskie et al. 1997) was employed in the present study because of the homogeneity and the possibility of large-area data extractions. The near-infrared photometry is also suitable for M 11, since its MS (and giant clump) stands out from the rich stellar field in CMDs as the one shown in Fig. 1. A circular data extraction with radius $`12\mathrm{}`$ centered in M 11 yielded 8432 stars surmounting by 941 the number of stars in the background field of same area (7491 stars), which is defined by an annulus with maximum radius of $`40\mathrm{}`$ (Fig. 1). From J=10 to J=14 the cluster MS seems to be little affected by field stars and should result in more precise determinations of the luminosity function (LF). A good account of the field is therefore necessary to obtain the LF of fainter stars. On this regard, we advance that a statistical approach was employed in which the number of cluster stars in a magnitude interval is obtained from the difference between the total number of stars within that interval at a given annulus and the same number at an external annulus supposedly containing only field stars. Before this procedure the data are submitted to a CMD filter, selecting only stars in the MS and giant cluster sequences (Fig. 1).
The distance from the cluster center used to extract a fiducial background field was chosen on the basis of a compromise between the premise that, on one side, the field is far enough in order to not contain cluster stars, and on the other side, the field is close enough to keep at small levels the irregularities produced by dust and stellar density gradient. Interesting to note that the field sequences correspond to disk stars as shown by means of CMDs simulations (Marigo et al. 2003): the vertical distribution around (J-H)=0.4 is formed by old disk turnoff stars (m$`0.9`$ M), the one at (J-H)=0.75 is associated to old disk low MS (m$`0.6`$ M) (but also may include reddened giants) and the stars redder than (J-H)=0.9 are probably disk giants.
### 2.1 Crowding effects
M 11 is a rich compact cluster, being classified as type “II2r” (detached, weak concentration; moderate range in brigthness; rich, more than 100 stars) by Trumpler (1930). Interestingly, Ruprecht (1966) has classified M 11 as a globular cluster of Trumpler type “I2r”, even though its first colour-magnitude diagram had already been obtained ten years before suggesting that the age of M 11 is intermediate between that of the Pleiades and that of Praesepe (Johnson et al. 1956). Not surprisingly, its concentration towards the center makes crowding an expected effect, enhanced by the instrumental limited spatial resolution. We have taken advantage of the complete analyses in the 2MASS database, which provide information on crowding for every source and band by means of a flag (“cc\_flg”). This flag identifies whenever a source/band had its photometry (flux) overestimated by at least 5% due to image artifacts, most of them associated to crowded fields.
We used this flag as an estimate of how significant is crowding over the cluster radius. The distribution of the ratio between the number of stars with photometry affected by crowding ($`N_c`$) and the total number of stars ($`N_u+N_c`$) was calculated as a function of radius. The counts were carried out within rings 2’ wide. The results are presented in Fig. 2. A nearly constant distribution of $`N_c/`$($`N_u+N_c`$) can be noticed except for the cluster inner regions (R$`<4`$’), in which crowding becomes important, as expected. What is the influence of crowding on the LF? Since most stars affected by crowding in the cluster inner regions follow the cluster sequences, i. e., the photometric precision is not severely degraded for those stars, the LF should preserve its shape if the magnitude bins are wider than the photometric uncertainty. Magnitude bins of 0.5 mag were used in the following since an uncertainty of 5% in flux corresponds to $`\sigma _\mathrm{J}=\frac{\sigma _{\mathrm{F}_\mathrm{J}}}{\mathrm{F}_\mathrm{J}}\frac{1}{0.4\mathrm{ln}10}`$0.05 mag, about one tenth of the magnitude bin. To reach 0.5 mag (bin width), the flux should be overestimated by 50%. Then, the crowding effects yield a negligible bias in the LF since large flux overestimates seem not to be the case according to Fig. 2, which shows that most of the stars affected by crowding are distributed over the same sequences as those unaffected. Indeed, significant flux overestimates caused by crowding would be immediately detected in the CMDs of Fig. 2 by an overall smearing of the sequences. Although completeness corrections were applied to M 11 by Mathieu (1984) and Sung et al. (1999), we have not applied such procedure since we focused most of our analysis outside the cluster core.
## 3 Cluster parameters
The interstellar reddening towards M 11 seems to be well established (E(B-V)=$`0.42\pm 0.03`$), with no evidence for a differential pattern across the cluster field (e. g. Sung et al. 1999). The apparent distance modulus has been measured with different methods resulting $`12.5<(\mathrm{V}\mathrm{M}_\mathrm{V})<12.92`$ (Sung et al. 1999; Brocato et al. 1993). The cluster metallicity is nearly above solar, \[Fe/H\]=$`0.136\pm 0.086`$, according to Twarog et al. (1997), which is approximately the value given in WEBDA (Sect. 1).
Cluster parameters were derived by fitting isochrones built using 2MASS filters (Bonatto et al. 2004) to the cluster central region (R$`<6`$’) CMDs J$`\times `$(J-H) and $`\mathrm{K}_\mathrm{S}\times (\mathrm{J}\mathrm{K}_\mathrm{S})`$, which best define the cluster sequences. The cluster central region corresponds to its visual diameter and it was chosen to maximize cluster members over field stars. Isochrones were adjusted to both CMDs using as constraints $`\mathrm{E}(\mathrm{B}\mathrm{V})=0.42\pm 0.03`$ and (m-M)$`{}_{}{}^{}=11.37\pm 0.23`$. Since each CMD was built from independent observations involving a mixture of different bands, they provide different data sets on which the isochrone matching should converge, giving more weight to the analysis. Fig. 3 shows the best matching solar metallicity isochrones where the data have been corrected for the extreme and average values that E(B-V) and (m-M) may assume due to errors. Selected stellar masses associated to the 224 Myr isochrone are also indicated in the top panels: the lower mass (1.2 M) corresponds approximately to the data instrumental limit, the intermediate mass (3.66 M) locates the turnoff, and the higher mass (3.82 M) marks the bluest point of the core He-burning phase (giant clump).
By fixing E(B-V) and (m-M) according to observational constraints, the free parameters were reduced to the isochrone age and metallicity. The well-defined main sequence (MS) and giant clump of M 11 allow an unambiguous derivation of its age with an uncertainty of nearly 10%.
A good overall match is obtained if the average reddening and average true distance modulus are used for the 250 Myr isochrone (Fig. 3, middle panels), with satisfactory results also being obtained for the 224 Myr isochrone and the 282 Myr one. In spite of this, a difference of $`0.05`$mag. is found for the giant clump mean locus in the M$`{}_{\mathrm{J}}{}^{}\times `$(J-H) CMD.
Taking into account the fact that the cluster has metallicity above solar (Twarog et al. 1997), the best match isochrone (250 Myr) is presented in Fig. 4 for two metallicities (Z=0.019 and Z=0.03 or \[Fe/H\]=0.20) together with the corrected data in both CMDs. Three mass values are shown connected to the corresponding isochrone. This comparison indicates that the colour of the cluster stars in the clump is not due to a cluster metallicity higher than solar, indeed the higher metallicity isochrone indicates a redder colour for the clump. Binaries may be affecting the cluster clump colour, since they would explain a brightening of clump stars if they were in binary systems and a blueing if they comprise a red giant and a blue turnoff star.
In conclusion, an age of t$`=250\pm 30`$Myr was obtained for M 11 and the solar metallicity isochrone was employed as representative of the cluster stellar population. Such a representation is relevant in the determination of the mass function since the isochrone is the source of the mass-luminosity relation used to transform the observed luminosity function (LF) into the mass function (MF). Indeed, as it is well known, the precise location of masses over the MS and the clump are influenced by age and metallicity as can be noticed by comparing Fig. 3 and Fig. 4. Thus, the mass range of observed stars and, in consequence, the cluster MF are partially determined by the isochrone chosen.
## 4 Cluster structure from the King-profile
The colour-magnitude filter (see Fig. 1) in the plane J$`\times `$(J-H) was applied in order to select the CMD regions containing the cluster evolutionary sequences. The magnitude cutoff at the lower MS end adopted for fitting a King-profile is based on the optimal separation of cluster stars and background field. Fig. 5 shows that at J=15.0 the density of cluster stars with respect to the background reaches a maximum value of 25.08 stars.arcmin<sup>-2</sup> at the central circle of 1 arcmin of radius. Therefore, the magnitude cutoff at J=15.0 was adopted.
A discussion on the effects of applying colour-magnitude filters and background selection are given in Bonatto et al. (2005). Because the background selection is critical for M 11, the filtered cumulative distribution of stars in the cluster outer regions was analyzed as a function of M<sub>J</sub>. In Fig. 6 the cumulative LF of the three outer rings 5 arcmin wide are compared. The counts were carried out within bins of $`\mathrm{\Delta }(\mathrm{M}_\mathrm{J})=0.1`$ mag and the range of M<sub>J</sub> presented in this Figure corresponds to the cluster MS. The LFs are normalized to the area of the outer ring ($`35<`$R(’)$`<40`$). A clear excess in the cumulative LF is noticed for the inner ring ($`25<`$R(’)$`<30`$), which is better visualized by the difference between its cumulative LF and that for the outer ring, and characterized as the “inner ring excess” in Fig. 6. The middle ring ($`30<`$R(’)$`<35`$) excess is also shown, which indeed does not reveal any significant difference between the middle and outer cumulative LFs, presumably for representing both a fiducial background field, little affected by cluster stars. In contrast, the inner ring clearly reveals the presence of cluster stars. Therefore cluster stars are present and dominant over the field for distances less than 30’ from its center.
An analysis of the cluster structure was performed on this selected sample. The radial distribution of stellar surface density (stars/arcmin<sup>2</sup>) was investigated by counting stars within rings of width 1 up to 40’ from the cluster center. As expected, the procedure carried out to maximize cluster stars over field stars helps to enhance the cluster structure keeping the background field at acceptable levels, as shown in the top panels of Fig. 7. In this Figure, a constant background was fitted to the outer region sampled, $`30<`$R(’)$`<40`$, its 1-$`\sigma `$ dispersion being shown in the top-right panel.
A reasonable limiting radius for M 11 is R$`{}_{lim}{}^{}=21\pm 1`$ ’, where the cluster star density begins to stand out from the background one. The fitted constant background was then subtracted from the overall surface density and a King-profile fitting was performed. Two-parameter (central stellar surface density, $`\sigma _0`$, and core radius, R<sub>c</sub>) and three-parameter ($`\sigma _0`$, R<sub>c</sub> and tidal radius, R<sub>t</sub>) King functions (King 1962, 1966) were employed. The fitted functions are presented in Fig 7 in log scales (bottom) together with the best fit parameters. The two-parametric King function should better represent the cluster inner regions and the three-parametric King function should provide a better estimate of the cluster overall structure. In both fittings the estimates of inner parameters ($`\sigma _0`$ and R<sub>c</sub>) agree within the uncertainties. The tidal radius was estimated with 50% precision because of the sensitivity of the three-parameter King model to the fluctuations in the density of cluster stars in its outskirts, almost at the background level. Such fluctuations (represented by poissonian errors in Fig. 7) are taken into account in the fitting by applying a weigthed least-squares method.
The adopted true distance modulus (m-M)$`{}_{}{}^{}=11.37\pm 0.23`$ translates into a distance from the Sun of d$`{}_{}{}^{}=1.89\pm 0.26`$ kpc. Therefore, the linear limiting radius of M 11 is R$`{}_{lim}{}^{}=11.5\pm 1.7`$ pc (Fig. 7). A galactocentric distance of d$`{}_{\mathrm{GC}}{}^{}=6.38\pm 0.21`$ kpc is obtained using d$`{}_{GC}{}^{}=8.0`$ kpc for the Sun Galactocentric distance (Reid 1993).
In Fig. 8 the same three-parametric King function shown in Fig. 7 is presented in absolute units, where 1 arcmin=0.55 pc. The cluster structural parameters concerning stars with J$`15.0`$ are $`\sigma _0=9.1\pm 3.0`$ stars.pc<sup>-2</sup>, R$`{}_{\mathrm{c}}{}^{}=1.23\pm 0.47`$ pc and R$`{}_{\mathrm{t}}{}^{}=29\pm 15`$ pc. The core radius is 1.7 times larger than that quoted by Nilakshi et al. (2002).
A deviation from the King profile can be seen between 6 and 9 pc (which is inside R<sub>lim</sub>, but well beyond R<sub>c</sub>) where the cluster star density is in excess with respect to the model (Fig. 8). Such an excess is expected if the cluster is in the process of loosing low mass stars by means of energy equipartition. If so, this excess of stellar surface density is also expected in the cluster outskirts, but detecting this effect beyond 9 pc is more difficult because of the uncertainties produced by the background field.
Alternative means of diagnosing mass segregation in clusters have been successfully employed by e.g. Raboud & Mermilliod (1998a), Raboud & Mermilliod (1998b), Mathieu (1984). Specifically, following these studies, we derive the cluster structure and cumulative distributions as characterized by stars in four mass ranges: $`1.17<`$m$`(\mathrm{M}_{})<1.68`$ (lower MS, $`2.00<`$M$`{}_{\mathrm{J}}{}^{}<3.27`$), $`1.68<`$m$`(\mathrm{M}_{})<3.07`$ (intermediate MS, $`0.00<`$M$`{}_{\mathrm{J}}{}^{}<2.00`$), $`3.07<`$m$`(\mathrm{M}_{})<3.50`$ (upper MS, $`2.00<`$M$`{}_{\mathrm{J}}{}^{}<0.00`$) and m$`(\mathrm{M}_{})>3.50`$ (giants, M$`{}_{\mathrm{J}}{}^{}<2.00`$). Two-parametric King models were fitted to the radial stellar density profile for each mass range separately. The results are presented in Fig. 9, where the best solution for the core radius is shown. The more extended populations (larger R<sub>c</sub>) correspond to those populations composed by less massive stars, as one would expect for a dynamically evolved cluster with conspicuous mass segregation. Although satisfactory fittings were obtained for the lower and intermediate MS, reasonable fittings were not achieved for the upper MS and giants, whose less numerous samples are subject to larger statistical errors. Therefore R<sub>c</sub> should be taken carefully as an indicator of mass segregation in the cluster.
Indeed, a better, direct account of mass segregation is revealed by the cumulative distribution of stars in the mass ranges as above (Fig. 10). The cumulative distribution (F(N)), field subtracted and determined up to 20 pc ($``$ 2$`\times `$R<sub>lim</sub>), shows that less stars in the lower MS sample are concentrated in the inner 6-7 pc in comparison with the other more massive samples. Closer to the cluster center, at R$`<3`$ pc, the number of stars more massive than 1.68 $`\mathrm{M}_{}`$ increases more sharply than the number of lower mass stars, indicating mass segregation. Comparing the lower MS with the intermediate MS, the former is less concentrated, in accord with the mass segregation effects. In the outer regions at R$`>10`$ pc, the number of giant stars appears to be still increasing, a feature already noticed by Mathieu (1984). Mathieu (1984) consider mass loss as a possible explanation for this unexpected effect: by losing mass the giant stars would reproduce the distribution of lower mass stars.
For being at the present time relatively close to the Galactic center, M 11 may have suffered the consequences of strong interactions with molecular clouds and/or tidal effects of the Galaxy determining its structure by affecting its tidal radius. In this respect, the internal dynamical processes occurring in M 11 are less effective in shaping the cluster overall structure than in the case of the older cluster NGC 188, which orbits the Galaxy in a path beyond the solar circle (Bonatto et al. 2005). An orbit calculation for M 11 would help to constrain its dynamical properties.
## 5 Radial luminosity and mass functions
LFs of the filtered CMD were determined for each 2MASS band separately and different annular regions corresponding to the cluster core, an intermediate annulus (R$`{}_{\mathrm{c}}{}^{}<`$R$`<5.5`$ pc), the halo (5.5 pc$`<R<`$R<sub>lim</sub>) and the overall cluster extension (0 pc$`<`$R$`<`$R<sub>lim</sub>). The LFs have been properly corrected for the background field by subtracting the counts measured in the outer field ($`30<`$R(’)$`<40`$) from the counts per magnitude bin in each region scaled to their area. The main-sequence LFs constructed independently for each band were converted in one MF for each cluster region by fitting mass-luminosity relations (in J, H and $`\mathrm{K}_\mathrm{S}`$) from the solar metallicity 250 Myr isochrone. The overall MF and the MFs of the selected spatial regions are presented in Fig. 11 together with power-law MF ($`\varphi (\mathrm{m})=\mathrm{A}.\mathrm{m}^{(1+\chi )}`$, where A is the MF normalization and $`\chi `$ is the MF slope) weighted fittings. The resulting MF slopes are shown in Table 1. The regions sampled are the core, the inner and outer halo and the overall cluster. The MF slope in the core is very flat, comparable to those of the cores of M 93, NGC 2477 and NGC 3680 (Bonatto & Bica 2005). M 11 presents a MF slope gradient from the core to the outer regions, which is characteristic of large-scale mass segregation. The overall MF slope value is similar to the standard Salpeter one, and comparable to most of the classical open clusters studied in Bonatto & Bica (2005).
In order to determine the total cluster mass from the overall MF, it was extrapolated to m$`<1.0`$ M down to the H-burning limit using the universal MF model by Kroupa (2001), which gives $`\chi =1.3\pm 0.3`$ for $`0.5<`$m(M)$`<1.0`$ and $`\chi =0.3\pm 0.5`$ for $`0.08<`$m(M)$`<0.5`$. The slope $`\chi =1.49\pm 0.09`$ was applied to the observed portion of the MF, which spans MS masses between $`1.09<`$m(M)$`<3.45`$. Following this procedure we get m(MS)=($`10.8\pm 3.8`$)$`\times 10^3`$ M.
The mass content in evolved stars was computed from the integrated LF above the turnoff at M$`{}_{\mathrm{J}}{}^{}=2.0`$ and the corresponding isochrone mass interval of the CMD observed stars. According to the 250 Myr isochrone, we observe evolved stars in the mass range $`3.50<`$m(M)$`<3.74`$ with a strong concentration around m=3.62 M in the giant clump region (see Fig. 4). We assume a typical mass for the evolved stars in M 11 of m$`=3.6\pm 0.1`$ M and use it to get the total mass in evolved stars, i. e., m(post-MS)=$`176\pm 26`$ M.
Therefore, the total mass obtained for M 11 inside a circular area of radius 11.5 pc is ($`11.0\pm 3.8`$)$`\times 10^3`$ M.
MFs in the selected regions of the cluster were used to compute the total mass of the subsystems. They are presented in Table 1.
The relation between the tidal radius and the cluster mass as given by King (1962) was also used to compute a total mass for M 11. Mathematically: m=3M$`{}_{\mathrm{G}}{}^{}(\frac{\mathrm{R}_\mathrm{t}}{\mathrm{R}_\mathrm{p}})_{}^{3}`$, where R<sub>p</sub> is the perigalacticon distance and M<sub>G</sub> is the Galactic mass inside R<sub>p</sub>. Assuming a nearly circular orbit for M 11, and consequently R<sub>p</sub>=d<sub>CG</sub>=$`6.38\pm 0.21`$ kpc, and M$`{}_{\mathrm{G}}{}^{}=8.0\times 10^{10}`$M (Carraro & Chiosi 1994), we get an expected mass of $`\mathrm{m}=(22\pm 35)\times 10^3\mathrm{M}_{}`$, the large error coming from the uncertainty in the tidal radius. Such a theoretical value is about 2 times larger than our estimate.
The relaxation time of a star system can be defined as t$`{}_{\mathrm{relax}}{}^{}=\frac{\mathrm{N}}{8\mathrm{l}\mathrm{n}\mathrm{N}}`$t<sub>cross</sub>, where t$`{}_{\mathrm{cross}}{}^{}=`$R/$`\sigma _\mathrm{v}`$ is the crossing time, N is the total number of stars and $`\sigma _\mathrm{v}`$ is the velocity dispersion (Binney & Tremaine 1987). t<sub>relax</sub> is the characteristic time scale in which the cluster reaches some level of kinetic energy equipartition with massive stars sinking to the core and low-mass stars being transferred to the halo. Using the velocity dispersion found for M 11 of $`\sigma _\mathrm{v}=2.9`$ km/s (McNamara & Sanders 1977) we obtain t$`{}_{\mathrm{relax}}{}^{}1300`$ Myr for the whole cluster and t$`{}_{\mathrm{relax}}{}^{}2`$ Myr for the cluster core. The MF slope flattening towards the center, as an evidence of mass segregation observed in M 11 is consistent with the cluster core being dynamically evolved in agreement with t<sub>relax</sub> in the core, which is smaller than the cluster age.
## 6 Concluding remarks
We employed 2MASS photometry to explore the structure and stellar content of the open cluster M 11, which is located internal to the Solar circle. The near-IR photometry basically confirmed previous photometric parameters derived from the optical. We studied this cluster with spatial resolution, owing to the wide-angle analysis allowed by 2MASS data. We obtained a core radius of 1.23 pc and a tidal radius of 29 pc. The latter value was obtained by fitting the three-parameter King profile to the radial distribution of stars, which is possible only for the more populous and highest-contrast open clusters.
The spatial distribution of mass functions showed a very flat one in the core ($`\chi =0.73`$) and a steep halo one ($`\chi =2.88`$), particularly in the outer region. Mass segregation is implied by the results. The overall mass function slope is similar to a standard Salpeter one. The spatial distribution of mass function slopes derived from 2MASS agrees with that derived from optical CCD data, which further confirms the reliability of 2MASS data for future analyses of this kind at comparable observational limits.
The cluster is massive, with a total (extrapolating the mass function to 0.08 M) mass of $`\mathrm{11\hspace{0.17em}000}\mathrm{M}_{}`$, which is somewhat larger than previous estimates.
The large cluster mass of M 11 is a slowing down factor of dynamical evolution because of a longer relaxation time. However, its position well within the Solar circle is expected to speed it up because of stronger tidal effects (e.g. de la Fuente Marcos & de la Fuente Marcos 2002; Bergond et al. 2001; Bonatto & Bica 2005).
###### Acknowledgements.
This publication makes use of data products from the Two Micron All Sky Survey, which is a joint project of the University of Massachusetts and the Infrared Processing and Analysis Center/California Institute of Technology, funded by the National Aeronautics and Space Administration and the National Science Foundation. We also thank the referee, Dr. J.-C. Mermilliod, for helping to improve the work and for the use of the WEBDA open cluster database. We acknowledge support from the Brazilian Institutions CNPq and FAPEMIG. |
warning/0507/math0507232.html | ar5iv | text | # Quantum cohomology of smooth complete intersections in weighted projective spaces and singular toric varieties
## 1. Results
See 2.1, 2.2 and 2.4 for necessary definitions.
### 1.1. Theorem
Let $`=(w_1,\mathrm{},w_k)`$ be a weighted projective space. Let $`X`$ be a smooth complete intersection of hypersurfaces $`X_1,\mathrm{},X_l`$ which do not intersect the singular locus of $``$. Assume that $`K_X>0`$ and $`\mathrm{Pic}X=H`$, where $`H`$ is the class dual to the hyperplane. Let $`i:X`$ be the natural embedding.
1) $`I`$-series for $`X`$ is the following.
$$I^X=e^{\alpha _Xq}\underset{d=0}{\overset{\mathrm{}}{}}q^di^{}\left(\frac{_{a=1}^l\text{(}X_a+1\text{)}_{𝒅\mathbf{}\mathrm{𝐝𝐞𝐠}𝑿_𝒂}}{_{a=1}^k\text{(}w_aH+1\text{)}_{𝒅\mathbf{}𝒘_𝒂}}\right).$$
Here $`\alpha _X=0`$ if the index of $`X`$ is $`2`$ or greater and $`\alpha _X=_{a=1}^l(\mathrm{deg}X_a)!/_{a=1}^kw_a!`$ if the index is $`1`$.
2) Let $`d_i`$ ($`1il`$) be the degrees (with respect to $`H`$) of the hypersurfaces $`X_i`$ and $`d_0=w_id_i`$ be the index of $`X`$.Let $`^X(q)=q^d((d_0d)!(e^{\alpha _X}I^X)_{d,H^0})`$. Consider the operator (which generalizes the Riemann–Roch operator, see \[Go01\])
$$L=\underset{i=1}{\overset{k}{}}\text{(}w_iD(w_i1)\text{)}_{𝒘_𝒊}q\underset{i=0}{\overset{l}{}}\text{(}d_iD+1\text{)}_{𝒅_𝒊}.$$
Then $`L[^X(q)]=0`$.
### 1.2. Theorem
Let $`Y`$ be a $``$-factorial toric variety and $`Y_1,\mathrm{},Y_k`$ be the divisors that correspond to the edges of the fan of $`Y`$. Consider a smooth complete intersection $`X`$ of hypersurfaces $`X_1,\mathrm{},X_l`$ that does not intersect the singular locus of $`Y`$. Assume that $`K_X>0`$ and $`\mathrm{Pic}X=`$. Let $`i:XY`$ be the natural embedding. Let $`\mathrm{}`$ be a nef generator of $`H_2(Y)`$. For $`\beta =d\mathrm{}`$ put $`q^\beta =q^d`$. Let $`\mathrm{\Lambda }H_2(X)`$ be the semigroup of algebraic curves as cycles on $`X`$.
1) $`I`$-series of $`X`$ is the following.
$$I^X=e^{\alpha _Xq}\underset{\beta \mathrm{\Lambda }}{}q^\beta i^{}\left(\frac{_{a=1}^l\text{(}X_a+1\text{)}_{𝜷\mathbf{}𝑿_𝒂}}{_{a=1}^k\text{(}Y_a+1\text{)}_{𝜷\mathbf{}𝒀_𝒂}}\right),$$
where $`\alpha _X=0`$ if the index of $`X`$ is greater than $`1`$, and $`\alpha _X=_{a=1}^l(\mathrm{}X_a)!/_{a=1}^k(\mathrm{}Y_k)!`$ if the index is $`1`$.
2) Let $`d_i`$ ($`1il`$) be the degrees of the hypersurfaces $`X_i`$ (with respect to $`\mathrm{}`$), $`w_i`$ ($`1ik`$) be the degrees of divisors that correspond to the edges of the fan of $`Y`$, and $`d_0`$ be the index of $`X`$.Let $`^X(q)=q^d((d_0d)!(e^{\alpha _X}I^X)_{d,H^0})`$. Consider the operator (which generalizes the Riemann–Roch operator, see \[Go01\])
$$L=\underset{i=1}{\overset{k}{}}\text{(}w_iD(w_i1)\text{)}_{𝒘_𝒊}q\underset{i=0}{\overset{l}{}}\text{(}d_iD+1\text{)}_{𝒅_𝒊}.$$
Then $`L[^X(q)]=0`$.
###### Remark 1.
By Lefschetz Theorem (see \[Do82\], Theorem 3.2.4, (i) and Remark 3.2.6) hypotheses of theorems 1.1 and 1.2 hold for complete intersections of dimension greater than $`2`$, which do not intersect the singular locus of ambient variety, if its Picard group is $``$ (this holds automatically by hypothesis of theorem 1.1).
## 2. Preliminaries
Throughout the paper we consider the invariants of *genus zero* (those that correspond to the rational curves).
An axiomatic treatment of *prime invariants* was given by M. Kontsevich and Yu. Manin in \[KM94\]. *Invariants with descendants* were introduced and constructed in \[BM96\] and \[Beh97\].
### 2.1. Moduli spaces of curves
Consider smooth variety $`X`$ such that $`K_X0`$.
#### 2.1.1. Definition
*The curve* is reduced scheme of pure dimension $`1`$. *The genus* of curve $`C`$ is the number $`h^1(𝒪_C)`$.
It is easy to see that the curve is of genus $`0`$ if and only if it is a tree of rational curves.
#### 2.1.2. Definition
The connected curve $`C`$ with $`n0`$ marked points $`p_1,\mathrm{},p_nC`$ is called *prestable* if it has at most ordinary double points as singularities and $`p_1,\mathrm{},p_n`$ are distinct smooth points (see \[Ma02\], III–2.1). The map $`f:CX`$ of connected curve of genus $`0`$ with $`n`$ marked points are called *stable* if $`C`$ is prestable and there are at least three marked or singular points on every contracted component of $`C`$ (\[Ma02\], V–1.3.2).
In the other words, a stable map of connected curve is the map that has only finite number of infinitesimal automorphisms.
#### 2.1.3. Definition
*The family of stable maps* (over the scheme $`S`$) of curves of genus $`0`$ with $`n`$ marked points is the collection $`(\pi :𝒞S,p_1,\mathrm{},p_n,f:𝒞X)`$, where $`\pi `$ is the following map. It is a smooth projective map with $`n`$ sections $`p_1,\mathrm{},p_n`$. Its geometric fibers $`(𝒞_s,p_1(s),\mathrm{},p_n(s))`$ are prestable curves of genus $`0`$ with $`n`$ marked points. Finally, the restriction $`f|_{𝒞_s}`$ on each fiber is a stable map.
Two families over $`S`$
$$(\pi :𝒞S,p_1,\mathrm{},p_n,f),(\pi ^{}:𝒞^{}S,p_1^{},\mathrm{},p_n^{},f^{})$$
are called *isomorphic* if there is an isomorphism $`\tau :𝒞𝒞^{}`$ such that $`\pi =\pi ^{}\tau `$, $`p_i^{}=\tau p_i`$, $`f=f^{}\tau `$.
#### 2.1.4.
Let $`\beta H_2^+(X)`$. Consider the following (contravariant) functor $`\overline{}_n(X,\beta )`$ from the category of (complex algebraic) schemes to the category of sets. Let $`\overline{}_n(X,\beta )(S)`$ be the set of isomorphism classes of families of stable maps of genus $`0`$ curves with $`n`$ marked points $`(\pi :𝒞S,p_1,\mathrm{},p_n,f)`$ such that $`f_{}([𝒞_s])=\beta `$, where $`[C_s]`$ is the fundamental class of $`C_s`$.
#### 2.1.5. Definition
*The moduli space of stable maps* of genus $`0`$ curves of class $`\beta H_2^+(X)`$ with $`n`$ marked points to $`X`$ is the Deligne–Mumford stack (see \[Ma02\], V–5.5) which is the coarse moduli space (see \[HM98\], Definition $`1.3`$) that represents $`\overline{}_n(X,\beta )`$. This space is denoted by $`\overline{M}_n(X,\beta )`$.
### 2.2. Gromov–Witten invariants and $`I`$-series
The definition of Gromov–Witten invariants is given in terms of intersection theory on smooth stacks $`\overline{M}_n(X,\beta )`$. It exists because locally such stack is a quotient of smooth variety by a finite group (see \[Vi89\]). However such stacks may have unexpected dimension. To use the products of cohomological cycles on them one should introduce *the virtual fundamental class* $`[\overline{M}_n(X,\beta )]^{\mathrm{virt}}`$ of virtual dimension $`\mathrm{vdim}\overline{M}_n(X,\beta )=dimX\mathrm{deg}_{K_X}\beta +n3`$ (see its construction in \[Ma02\], VI–1.1). In many cases (for instance, for homogenous spaces) virtual fundamental class coincides with the usual one.
Consider the map $`ev_i:\overline{M}_n(X,\beta )X`$, $`ev_i(C;p_1,\mathrm{},p_n,f)=f(p_i)`$. Let $`\pi _{n+1}:\overline{M}_{n+1}(X,\beta )\overline{M}_n(X,\beta )`$ be the forgetful map at the point $`p_{n+1}`$. It contracts unstable components after forgetting $`p_{n+1}`$. Let $`\sigma _i:\overline{M}_n(X,\beta )\overline{M}_{n+1}(X,\beta )`$ be the section that are correspond to the marked point $`p_i`$ constructed as follows. The image of the curve $`(C;p_1,\mathrm{},p_n,f)`$ under the map $`\sigma _i`$ is the curve $`(C^{};p_1,\mathrm{},p_{n+1},f^{})`$. Here $`C^{}=CC_0`$, $`C_0^1`$, and $`C_0`$ and $`C`$ intersect at the (non-marked on $`C^{}`$) point $`p_i`$. The points $`p_{n+1}`$ and new $`p_i`$ lie on $`C_0`$. The map $`f^{}`$ contracts $`C_0`$ and $`f^{}|_C=f`$.
Let $`L_i=\sigma _i^{}\omega _{\pi _{n+1}}`$, where $`\omega _{\pi _{n+1}}`$ is the relative dualizing sheaf $`\pi _{n+1}`$. The fiber $`L_i`$ over the point $`(C;p_1,\mathrm{},p_n,f)`$ is $`T_{p_i}^{}C`$.
#### 2.2.1.
Definition \[see \[Ma02\], VI–2.1\]. *A cotangent line class* is the class
$$\psi _i=c_1(L_i)H^2(\overline{M}_n(X,\beta )).$$
#### 2.2.2.
Definition \[see \[Ma02\], VI–2.1\]. Consider the cohomological classes $`\gamma _1,\mathrm{},\gamma _nH^{}(X)`$. Let $`a_1,\mathrm{},a_n`$ be the non-negative integers and $`\beta H_2(X)`$. *The Gromov–Witten invariant with descendants* that correspond to these classes is
$$\tau _{a_1}\gamma _1,\mathrm{},\tau _{a_n}\gamma _n_\beta =\psi _1^{a_1}ev_1^{}(\gamma _1)\mathrm{}\psi _n^{a_n}ev_n^{}(\gamma _n)[\overline{M}_n(X,\beta )]^{\mathrm{virt}}$$
if $`\mathrm{codim}\gamma _i+a_i=\mathrm{vdim}\overline{M}_n(X,\beta )`$ and $`0`$ otherwise. The invariants with $`a_i=0`$ (for each $`i`$) are called *prime*. They are equal to the expected numbers of rational curves of the class $`\beta `$ on $`X`$, which intersect the dual cycles to $`\gamma _1,\mathrm{},\gamma _n`$ (divided by the product of degrees of divisors from $`\gamma _1,\mathrm{},\gamma _n`$). We omit the symbols $`\tau _0`$.
#### 2.2.3.
Definition \[see \[Ga00\]\]. Let $`\mu _1,\mathrm{},\mu _N`$ be the basis of $`H^{}(X)`$, $`\stackrel{ˇ}{\mu }_1,\mathrm{},\stackrel{ˇ}{\mu }_N`$ be the dual basis, and $`\gamma _1,\mathrm{},\gamma _k`$ be the basis of $`H_2(X)`$. Each curve $`\beta `$ is of the form $`\beta =\beta _i\gamma _i`$. Consider the ring $`B=[[q]]`$, $`q=(q_1,\mathrm{},q_k)`$. Put $`q^\beta =q_i^{\beta _i}`$. Then *the $`I`$-series* for $`X`$ is the following.
$$I^X=\underset{\beta 0}{}I_\beta ^Xq^\beta BH^{}(X),I_\beta ^X=\underset{i,j0}{}\tau _i\mu _j_\beta \stackrel{ˇ}{\mu }_j$$
(we use the same symbol for $`\gamma H^{}(X)`$ and $`1\gamma BH^{}(X)`$).
*Fundamental term* of $`I`$-series is the series
$$I_{H^0}^X=\underset{\beta 0}{}\tau _{(K_X)\beta 2}\stackrel{ˇ}{\mathrm{𝟏}}_\beta q^\beta ,$$
where $`\stackrel{ˇ}{\mathrm{𝟏}}`$ is the class dual to the class of unity in the cohomology.
In the case $`k=1`$ and arbitrary series $`IBH^{}(X)`$ put $`I=_{q0}I_dq^d`$.
#### 2.2.4.
Lemma \[\[Ga99\], Lemma 5.5 or proof of Lemma 1 in \[LP01\]\]. Let $`YX`$ be a complete intersection and $`\phi :H^{}(X)H^{}(Y)`$ be the restriction homomorphism. Let $`\stackrel{~}{\gamma }_1\phi (H^{}(X))^{}`$ and $`\gamma _2,\mathrm{},\gamma _l\phi (H^{}(X))`$. Then for each $`\beta \phi (H_2(X))H_2(Y)`$ the Gromov–Witten invariant on $`Y`$ of the form
$$\tau _{d_1}\stackrel{~}{\gamma }_1,\tau _{d_2}\gamma _2,\mathrm{}\tau _{d_l}\gamma _l_\beta $$
vanishes.
#### 2.2.5. Definition
Consider a complete intersection $`YX`$. Let $`\phi :H^{}(X)H^{}(Y)`$ be the restriction homomorphism and $`R=\phi (H^{}(X))H^{}(Y)`$. Let $`\mu _1,\mathrm{},\mu _N`$ be the basis of $`R`$, $`\stackrel{ˇ}{\mu }_1,\mathrm{},\stackrel{ˇ}{\mu }_NR`$ be the dual basis, and $`\gamma _1,\mathrm{},\gamma _k`$ be the basis of $`H_2(Y)`$. Each curve $`\beta `$ is of the form $`\beta =\beta _i\gamma _i`$. Consider the ring $`B=[[q]]`$, $`q=(q_1,\mathrm{},q_k)`$. Put $`q^\beta =q_i^{\beta _i}`$. Then *restricted $`I`$-series* of $`Y`$ is the following.
$$\stackrel{~}{I}^Y=\underset{\beta 0}{}\stackrel{~}{I}_\beta ^Yq^\beta BRBH^{}(Y),\stackrel{~}{I}_\beta ^Y=\underset{i,j0}{}\tau _i\mu _j_\beta \stackrel{ˇ}{\mu }_j.$$
The invariants of complete intersection that correspond to cohomological classes restricted from the ambient variety are called restricted.
#### 2.2.6.
Remark. For a complete intersection $`YX`$ of dimension at least $`3`$ these series coincide, i. e. $`I^Y=\stackrel{~}{I}^Y`$. Indeed, $`H_2(Y)H_2(X)`$ and by lemma 2.2.4 and the divisor axiom 2.3.2 one-pointed invariants for primitive classes vanish.
#### 2.2.7. Remark
Two- and more- pointed invariants for primitive classes of middle dimension may be non-zero. For instance, for two primitive classes $`\alpha `$ and $`\beta `$, a hypersurface section $`H`$ and a line $`\mathrm{}`$ on cubic threefold
$$\alpha ,\beta ,H^2_{\mathrm{}}=\alpha \beta $$
(see \[Bea95\], Proposition 1).
### 2.3. Relations between multipointed invariants
#### 2.3.1.
In this paragraph we discuss relations between one-pointed Gromov–Witten invariants with descendants (i. e. coefficients of $`I`$-series) and multipointed prime ones (in particular, two-pointed ones, which are important for the following).
One-pointed invariants with descendants may be expressed in terms of multipointed ones by the following.
#### 2.3.2.
Divisor axiom for invariants with descendants \[\[KM98\], 1.5.2\]. Let $`Y`$ be smooth projective variety, $`\gamma _1,\mathrm{},\gamma _nH^{}(Y)`$ and $`\gamma _0H^2(Y)`$ be a divisor. Then
$$\begin{array}{c}\gamma _0,\tau _{d_1}\gamma _1,\mathrm{},\tau _{d_n}\gamma _n_\beta =(\gamma _0\beta )\tau _{d_1}\gamma _1,\mathrm{},\tau _{d_n}\gamma _n_\beta +\hfill \\ \hfill \underset{k,d_k1}{}\tau _{d_1}\gamma _1,\mathrm{},\tau _{d_k1}(\gamma _0\gamma _k),\mathrm{},\tau _{d_n}\gamma _n_\beta .\end{array}$$
The following formula (with divisor axiom) enables one to express recursively three-pointed invariants with descendants in terms of prime ones.
#### 2.3.3.
Theorem \[topological recursion relations, \[Ma02\], VI–6.2.1\]. Let $`Y`$ be a smooth projective variety, $`\{\mathrm{\Delta }^i\}`$ and $`\{\mathrm{\Delta }_i\}`$, $`i=1,\mathrm{},N`$, be dual bases of $`H^{}(Y)`$. Then
$$\tau _{d_1}\gamma _1,\tau _{d_2}\gamma _2,\tau _{d_3}\gamma _3_\beta =\underset{a,\beta _1+\beta _2=\beta }{}\tau _{d_11}\gamma _1,\mathrm{\Delta }^a_{\beta _1}\mathrm{\Delta }_a,\tau _{d_2}\gamma _2,\tau _{d_3}\gamma _3_{\beta _2}$$
(the sum is taken over all $`a,\beta _1,\beta _2`$ such that the expression in it is well-defined, i. e. $`a=1,\mathrm{},N`$, $`\beta _1,\beta _20`$).
#### 2.3.4.
Using this formulas, we can express one-pointed invariants in terms of prime two-pointed ones (see \[Pr04\], Proposition $`5.2`$). In fact, in many cases these expressions are invertible. More particular, the inverse expressions can be found recursively by the divisor axiom and the following theorem (see also \[BK00\], ⥮६ $`5.2`$ and \[Pr04\]).
#### 2.3.5.
Theorem \[Lee, Pandharipande, \[LP01\], Theorem 2, i\]. Let $`Y`$ be a smooth projective variety. Consider the self-dual ring $`RH^{}(Y)`$ generated by Picard group $`\mathrm{Pic}Y`$ such that for any $`\mu _1,\mathrm{},\mu _nR`$ and $`\nu R^{}`$
$$\tau ^{a_1}\mu _1,\mathrm{},\tau ^{a_n}\mu _n,\tau ^a\nu _\beta =0.$$
Let $`\gamma _1,\mathrm{},\gamma _nR`$, $`H\mathrm{Pic}Y`$, $`\mathrm{\Delta }^i`$ and $`\mathrm{\Delta }_i`$ be the dual bases of $`H^{}(Y)`$. Then one can algebraically express Gromov–Witten invariant
$$\tau ^{k_1}\gamma _1,\mathrm{},\tau ^{k_n}\gamma _n_\beta $$
in terms of one-pointed Gromov–Witten invariants with descendants using the following expressions
$$\begin{array}{c}\tau _{k_1}\gamma _1,\mathrm{},\tau _{k_n}H\gamma _n_\beta =\tau _{k_1}H\gamma _1,\mathrm{},\tau _{k_n}\gamma _n_\beta +\beta H\tau _{k_1+1}\gamma _1,\mathrm{},\tau _{k_n}\gamma _n_\beta \hfill \\ \hfill \underset{\beta _1+\beta _2=\beta }{}\beta _1H\underset{S^1S^n=S,a}{}\tau _{k_{s_1^1}}\gamma _{s_1^1},\mathrm{},\tau _{k_{s_a^1}}\gamma _{s_a^1},\mathrm{\Delta }^a_{\beta _1}\mathrm{\Delta }_a,\tau _{k_{s_1^n}}\gamma _{s_1^n},\mathrm{},\tau _{k_{s_b^n}}\gamma _{s_b^n}_{\beta _2}\end{array}$$
and
$$\begin{array}{c}\tau _{k_1}\gamma _1,\mathrm{},\tau _{k_n+1}\gamma _n_\beta =\tau _{k_1+1}\gamma _1,\mathrm{},\tau _{k_n}\gamma _n_\beta +\hfill \\ \hfill \underset{\beta _1+\beta _2=\beta ,\text{ }S^1S^n=S,a}{}\tau _{k_{s_1^1}}\gamma _{s_1^1},\mathrm{},\tau _{k_{s_a^1}}\gamma _{s_a^1},\mathrm{\Delta }^a_{\beta _1}\mathrm{\Delta }_a,\tau _{k_{s_1^n}}\gamma _{s_1^n},\mathrm{},\tau _{k_{s_b^n}}\gamma _{s_b^n}_{\beta _2},\end{array}$$
where $`H\mathrm{Pic}XR`$ and the latter summations are taken over partitions $`S^1=\{s_1^1,\mathrm{},s_a^1\}`$ and $`S^n=\{s_1^n,\mathrm{},s_b^n\}`$ of the set $`S=\{1,\mathrm{},n\}`$ such that $`1S^1`$ and $`nS^n`$.
#### 2.3.6.
The hypothesis of this theorem holds for subring of restricted cohomologies of complete intersection in toric variety (see \[Fu93\], 5.2) or for the subring of the cohomology ring of Fano threefold generated by Picard group. See expressions for Fano threefolds with Picard group $``$ in \[Pr04\].
*Consequently, theorems 1.1 and 1.2 enable us to find the restricted quantum cohomology ring of smooth complete intersections in weighted projective spaces and singular toric varieties.*
### 2.4. Toric varieties
The definition and the main properties of toric varieties see in \[Da78\] or in \[Fu93\]. Just remind that toric variety is a variety with action of torus $`T(^{})^n`$ such that one of its orbits is a Zariski open set. Toric variety is determined by its *fan*, i. e. some collection of cones with vertices in the points of lattice that is dual to the lattice of torus characters. Moreover, algebraic-geometric properties of toric variety can be formulated in terms of properties of this fan.
Remind some of them.
#### 2.4.1.
Every cone of the fan $`\sigma N=^n`$ of dimension $`r`$ corresponds to the orbit of the torus of dimension $`nr`$ (here $`n`$ is the dimension of toric variety). Thus, each edge (one-dimensional cone) correspond to the (equivariant) Weil divisor<sup>1</sup><sup>1</sup>1 That is, let $`\mathrm{\Sigma }N=^n`$ be a fan of the toric variety $`X_\mathrm{\Sigma }`$ and let $`\sigma \mathrm{\Sigma }`$ be any cone. Let $`M`$ be a lattice dual to $`N`$ with respect to some non-degenerate pairing $`,`$ and $`\sigma ^{}`$ be a dual cone for $`\sigma `$ (i. e. $`\sigma ^{}=\{lM|k\sigma l,k0\}`$). Let $`U_\sigma =\mathrm{Spec}[\sigma ^{}]`$ corresponds to $`\sigma `$. The variety $`X_\sigma `$ is obtained from the affine varieties $`U_\sigma `$, $`\sigma \mathrm{\Sigma }`$, by gluing together $`U_\sigma `$ and $`U_\tau `$ along $`U_{\sigma \tau }`$, $`\sigma ,\tau \mathrm{\Sigma }`$. Thus, if $`l\sigma \mathrm{\Sigma }`$ is an edge of the fan, then the divisor that is correspond to $`l`$ restricted on $`U_\sigma `$ as $`U_lU_\sigma `$. . The divisors which correspond to the edges of the fan generate divisor class group. A Weil divisor $`D=d_iM_i`$, where $`M_i`$ corresponds to edges, is Cartier if for each cone of the fan $`\sigma `$ there exist a vector $`n_\sigma `$ such that $`n_\sigma ,m_i=d_i`$ where $`m_i`$ is the primitive elements of the edges of this cone. If such vector is the same for all cones, then the divisor is principal. Hence if the toric variety is $`n`$-dimensional and the number of the edges is $`k`$, then the rank of the divisor class group is $`nk`$.
#### 2.4.2. Definition
The variety is called *$``$-factorial* if for each Weil divisor $`D`$ there exist some integer $`k`$ such that $`kD`$ is a Cartier divisor.
In particular, there exist an intersection theory for Weil divisors on the $``$-factorial variety.
#### 2.4.3.
Toric variety is $``$-factorial if and only if any cone of the fan, which corresponds to this variety, is simplicial. In this case Picard group is generated (over $``$) by divisors, which correspond to the edges of the fan.
#### 2.4.4.
Consider a weighted projective space $`=(w_0,\mathrm{},w_l)`$. The fan which correspond to it is generated by integer vectors $`m_0,\mathrm{},m_l^l`$ such that $`w_im_i=0`$. If $`w_0=1`$, then one can put $`m_0=(w_1,\mathrm{},w_l)`$, $`m_i=e_i`$, where $`e_i`$ is a basis of $`^l`$. The collection $`\{m_i\}`$ corresponds to the collection of standard divisors–strata $`\{D_i|w_iH|\}`$.
#### 2.4.5.
A toric variety is *smooth* if for any cone $`\sigma `$ in the fan that correspond to this variety the subgroup $`\sigma ^n`$ is generated by the subset of the basis of the lattice $`m_1^\sigma ,\mathrm{},m_k^\sigma `$. Adding the edge $`a=a_1m_1^\sigma +\mathrm{}+a_km_k^\sigma `$, $`a_i`$ to the cone (and connection it with “neighboring” faces) corresponds to weighted blow-up (along subvariety which correspond to $`\sigma `$) with weights $`1/r(\alpha _1,\mathrm{},\alpha _k)`$, where $`\alpha _i`$ and $`a_i=\alpha _i/r`$. Consecutively adding edges to the fan in this way we can get toric resolution of a toric variety.
#### 2.4.6.
So, singular locus of weighted projective space $`=(w_0,\mathrm{},w_l)`$ is the union of strata given by $`x_{i_1}=\mathrm{}=x_{i_j}=0`$, where $`x_{i_j}`$ is the coordinate of weight $`w_{i_j}`$ and $`\{i_1,\mathrm{},i_j\}`$ is the maximal set of indices such that greatest common factor of the others is greater than $`1`$.
### 2.5.
Givental’s Theorem \[\[Gi97\], Theorem $`0.1`$\]. Let $`X`$ be a smooth toric variety and $`Y`$ be a smooth complete intersection in it with positive anticanonical class. Let $`X_1,\mathrm{},X_k`$ be the divisors which correspond to the edges of a fan of $`X`$ and let $`Y`$ be given by divisors $`Y_1,\mathrm{},Y_r`$. Let $`\mathrm{\Lambda }H_2(X)`$ be the semigroup of algebraic curves as cycles on $`Y`$, and $`i:YX`$ be a natural embedding. Then
$$I^Y=e^{h(q)}\underset{\beta \mathrm{\Lambda }}{}q^\beta i^{}\left(\frac{_{a=1}^r\text{[}Y_a\text{]}_{𝜷\mathbf{}𝒀_𝒂\mathbf{+}\mathrm{𝟏}}}{_{a=1}^r\text{[}Y_a\text{]}_\mathrm{𝟏}}\frac{_{a=1}^k\text{[}X_a\text{]}_\mathrm{𝟏}}{_{a=1}^k\text{[}X_a\text{]}_{𝜷\mathbf{}𝑿_𝒂\mathbf{+}\mathrm{𝟏}}}\right),$$
where $`h(q)`$ is a polynomial supported by curves, whose intersection with anticanonical class is $`1`$ (i. e. $`h(q)=_\beta h_\beta q^\beta `$, where $`\beta (K_Y)=1`$).
### 2.6. Motivation
Consider a smooth Fano threefold $`X`$ with Picard group $``$. Put $`K=K_X`$.
#### 2.6.1.
Definition \[\[Go02\], 1.7, 1.10\]. *A counting matrix* is the matrix of Gromov-Witten invariants of $`X`$, namely the following matrix $`A\mathrm{Mat}(4\times 4)`$.
$$A=\left[\begin{array}{cccc}a_{00}& a_{01}& a_{02}& a_{03}\\ 1& a_{11}& a_{12}& a_{13}\\ 0& 1& a_{22}& a_{23}\\ 0& 0& 1& a_{33}\end{array}\right].$$
Numeration of rows and columns starts from $`0`$ and the elements are given by
$$a_{ij}=\frac{K^{3i},K^j,K_{ji+1}}{\mathrm{deg}X}=\frac{ji+1}{\mathrm{deg}X}K^{3i},K^j_{ji+1}$$
(the degree is taken with respect to the anticanonical class).
#### 2.6.2.
It is easy to see that the matrix $`A`$ is symmetric with respect to the secondary diagonal: $`a_{ij}=a_{3j,3i}`$. By definition, $`a_{ij}=0`$ if $`ji+1<0`$. If $`ji+1=0`$, then $`a_{ij}=1`$, because it is just a number of intersection points of $`K^{3i}`$, $`K^j`$, and $`K`$, which is $`\mathrm{deg}X`$; $`a_{00}=a_{33}=0`$. For the other coefficients $`a_{ij}`$’s are “expected” numbers of rational curves of degree $`ji+1`$ passing through $`K^{3i}`$ and $`K^j`$, multiplied by $`\frac{ji+1}{\mathrm{deg}X}`$. The only exception is the following: by divisor axiom
$$a_{01}=2(2ind(X)[\text{the number of conics passing through the general point}]).$$
#### 2.6.3.
Consider the following Fano threefolds (see \[Is77\], \[Is78\], \[Is79\], \[Is88\], \[IP99\], \[Mu92\]).
The variety $`V_1`$ of anticanonical degree 8 (a double covering of the cone over Veronese surface branched over a smooth cubic).
The variety $`V_2`$ of anticanonical degree 16 (a double covering of $`^3`$ branched over a smooth quartic).
The variety $`V_2^{}`$ of anticanonical degree 2 (a double covering of $`^3`$ branched over a smooth sextic).
#### 2.6.4.
Proposition. The varieties $`V_1`$, $`V_2`$, and $`V_2^{}`$ are of the following form.
Any variety of type $`V_1`$ is isomorphic to a smooth hypersurface of degree $`6`$ in $`(1,1,1,2,3)`$.
Any variety of type $`V_2`$ is isomorphic to a smooth hypersurface of degree $`4`$ in $`(1,1,1,1,2)`$.
Any variety of type $`V_2^{}`$ is isomorphic to a smooth hypersurface of degree $`6`$ in $`(1,1,1,1,3)`$.
#### 2.6.5. Proof
Double covering $`(w_0,\mathrm{},w_n)`$ branched over a divisor given by function $`f_k(x_0,\mathrm{},x_n)`$ of degree $`k`$ may be given by function $`x_{n+1}^2=f_k`$ in $`(w_0,\mathrm{},w_n,k/2)`$. The variables $`x_i`$ here have the weights $`w_i`$ and the variable $`x_{n+1}`$ has the weight $`k/2`$. ∎
#### 2.6.6. Theorem
The counting matrices for $`V_1`$, $`V_2`$, and $`V_2^{}`$ are as follows.
For $`V_1`$
$$M(V_1)=\left[\begin{array}{cccc}0& 240& 0& 57600\\ 1& 0& 1248& 0\\ 0& 1& 0& 240\\ 0& 0& 1& 0\end{array}\right].$$
For $`V_2`$
$$M(V_2)=\left[\begin{array}{cccc}0& 48& 0& 2304\\ 1& 0& 160& 0\\ 0& 1& 0& 48\\ 0& 0& 1& 0\end{array}\right].$$
For $`V_2^{}`$
$$M(V_2^{})=\left[\begin{array}{cccc}0& 137520& 119681240& 21690374400\\ 1& 624& 650016& 119681240\\ 0& 1& 624& 137520\\ 0& 0& 1& 0\end{array}\right].$$
#### 2.6.7.
Proof. By proposition 2.6.4 these varieties are hypersurfaces in weighted projective spaces. Find their one-pointed invariants using theorem 1.1. Apply theorem 2.3.5 for the subring of cohomology ring generated by the class dual to the class of hyperplane section and find prime two-pointed invariants. They are the coefficients of counting matrices. ∎
## 3. Proofs of the main theorems
### 3.1.
Proof of theorems 1.1 and 1.2.
The divisors in $``$ that correspond to the edges of its fan are $`w_1H,\mathrm{},w_kH`$. Thus, theorem 1.1 follows from theorem 1.2.
Prove theorem 1.2.
1) Let $`\mathrm{\Sigma }`$ be a fan of $`Y`$. Consider the sequence of simplicial fans $`\mathrm{\Sigma }=\mathrm{\Sigma }_0,\mathrm{\Sigma }_1,\mathrm{},\mathrm{\Sigma }_r`$ such that
* The fan $`\mathrm{\Sigma }_{i+1}`$ is a result of the following procedure applied to $`\mathrm{\Sigma }_i`$. Pick a cone whose edges are not part of a basis of the integer lattice containing this fan. Add an edge, which lies inside the cone. Add the faces which contain this edge and “neighboring” edges (i. e. replace all cones which contain the edge we add by all linear spans of this edge and faces of this cone that does not contain the edge);
* The fan $`\mathrm{\Sigma }_r`$ corresponds to a smooth variety.
Such sequence exists by \[Da78\], $`8.1`$$`8.3`$.
This way we get a toric resolution of singularities $`f:\stackrel{~}{Y}=X_{\mathrm{\Sigma }_r}Y`$ (where $`X_{\mathrm{\Sigma }_r}`$ is the toric variety that correspond to the fan $`\mathrm{\Sigma }_r`$, see remark 1). The exceptional set of this resolution is the union of divisors $`E_1,\mathrm{},E_r`$ which correspond to the edges we add. Let $`W_1,\mathrm{},W_k`$ be divisors that correspond to the edges of a fan of $`Y`$ and $`\stackrel{~}{W}_1,\mathrm{},\stackrel{~}{W}_k`$ be their strict transforms. Let $`\stackrel{~}{X}_1,\mathrm{},\stackrel{~}{X}_l`$ be strict transforms of $`X_1,\mathrm{},X_l`$, $`\stackrel{~}{X}=\stackrel{~}{X}_1\mathrm{}\stackrel{~}{X}_l`$, and $`j:\stackrel{~}{X}\stackrel{~}{Y}`$ be the natural embedding. Obviously, there exists an isomorphism $`g`$ such that the diagram
is commutative, because $`f`$ is an isomorphism between $`\stackrel{~}{Y}(_{i=1}^rE_i)`$ and $`Y\mathrm{Sing}Y`$.
Consider any toric variety $`V`$ associated to a complete simplicial fan. Then the canonical map of group of algebraic cycles modulo rational equivalence with rational coefficients to homology group with rational coefficients
$$A_{}(V)_{}H_{}(V,)$$
is an isomorphism (see \[Da78\], 10.9). So, we can extend an intersection theory to the homology group with rational coefficients.
The map $`f_{}:H_2(\stackrel{~}{Y},)H_2(Y,)`$ is surjective. Let $`\lambda XY`$ be an effective curve such that $`H_2(X)=\lambda `$ and $`\stackrel{~}{\lambda }=f^1(\lambda )`$. Put $`K=\mathrm{ker}f_{}`$. Then $`H_2(\stackrel{~}{Y},)=\stackrel{~}{\lambda }+K`$.
The cycle of any curve $`𝜷`$ which lies on $`\stackrel{~}{X}`$ equals $`\beta _0\stackrel{~}{\lambda }`$, where $`\beta _0`$ and $`\beta _00`$, because representatives of elements of $`K`$ lie on exceptional divisors, which do not intersect $`\stackrel{~}{X}`$. So, if $`\stackrel{~}{\mathrm{\Lambda }}`$ is a semigroup of algebraic curves on $`\stackrel{~}{X}`$, then $`\stackrel{~}{\mathrm{\Lambda }}=_0\stackrel{~}{\lambda }`$.
The multiplicity of intersection of divisor and a curve depends only on the neighborhood of this curve and $`f`$ is an isomorphism in a neighborhood of $`\stackrel{~}{X}`$, so $`\stackrel{~}{\lambda }\stackrel{~}{X}_i=\lambda X_i`$ and $`\stackrel{~}{\lambda }\stackrel{~}{W}_i=\lambda W_i`$. Thus, for any curve $`𝜷=\beta _0\stackrel{~}{\lambda }`$ on $`\stackrel{~}{X}`$ we have $`𝜷\stackrel{~}{W}_i=\beta _0(\lambda W_i)`$, $`𝜷\stackrel{~}{X}_i=\beta _0(\lambda X_i)`$, and $`𝜷E_i=0`$.
Divisors that correspond to the edges of a fan of $`\stackrel{~}{Y}`$ are $`\stackrel{~}{W}_1,\mathrm{},\stackrel{~}{W}_k,E_1,\mathrm{},E_r`$. But the factors in the expression for $`I`$-series for $`\stackrel{~}{X}`$ from Theorem 2.5 which include the divisors $`E_i`$ may be cancelled, so we may omit them. Consider rings $`H^{}(X)[[q]]`$ and $`H^{}(\stackrel{~}{X})[[\stackrel{~}{q}]]`$ and an isomorphism between them which sends $`q`$ to $`\stackrel{~}{q}`$ and acts on the coefficients as $`f^{}`$. Let $`j:\stackrel{~}{X}\stackrel{~}{Y}`$ be a natural embedding. Then
$$\begin{array}{c}I^{\stackrel{~}{X}}=e^{\stackrel{~}{h}(\stackrel{~}{q})}\underset{\beta =\beta _0\stackrel{~}{\lambda }\stackrel{~}{\mathrm{\Lambda }}}{}\stackrel{~}{q}^{\beta _0}j^{}\left(\frac{_{a=1}^l\text{[}\stackrel{~}{X}_a\text{]}_{𝜷\mathbf{}\stackrel{\mathbf{~}}{𝑿}_𝒂\mathbf{+}\mathrm{𝟏}}}{_{a=1}^l\text{[}\stackrel{~}{X}_a\text{]}_\mathrm{𝟏}}\frac{_{a=1}^k\text{[}\stackrel{~}{W}_a\text{]}_\mathrm{𝟏}}{_{a=1}^k\text{[}\stackrel{~}{W}_a\text{]}_{𝜷\mathbf{}\stackrel{\mathbf{~}}{𝑾}_𝒂\mathbf{+}\mathrm{𝟏}}}\right)=\hfill \\ \hfill j^{}f^{}\left(e^{h(q)}\underset{\beta _00}{}q^{\beta _0}\underset{a=1}{\overset{l}{}}\text{(}X_a+1\text{)}_{𝜷_\mathrm{𝟎}\mathbf{}\mathrm{𝐝𝐞𝐠}𝑿_𝒂}\frac{1}{_{a=1}^k\text{(}\stackrel{~}{W}_a+1\text{)}_{𝜷_\mathrm{𝟎}\mathbf{}𝒘_𝒂}}\right)=\\ \hfill g^{}\left(e^{h(q)}\underset{d=0}{\overset{\mathrm{}}{}}q^di^{}\left(\frac{_{a=1}^l\text{(}X_a+1\text{)}_{𝒅\mathbf{}\mathrm{𝐝𝐞𝐠}𝑿_𝒂}}{_{a=1}^k\text{(}\stackrel{~}{W}_a+1\text{)}_{𝒅\mathbf{}𝒘_𝒂}}\right)\right),\end{array}$$
where $`\stackrel{~}{h}(\stackrel{~}{q})`$ is a polynomial from theorem 2.5 and $`h(q)`$ is its pre-image. The second equality holds because we may choose divisors that intersect $`X`$ properly, and their strict transforms as representatives (over $``$) of divisor classes (because $`\mathrm{Pic}Y=`$ and every effective divisor on $`Y`$ is ample, and, therefore, it is equivalent to a rational multiple to a very ample one). The map $`f`$ after restriction on $`\stackrel{~}{X}`$ is isomorphism, which induces the isomorphism $`f^{}`$ of cohomology groups such that the image of any divisor under $`f^{}`$ is its strict transform. Finally, $`g^{}`$ is an isomorphism of cohomology rings $`X`$ and $`\stackrel{~}{X}`$, so $`I^{\stackrel{~}{X}}=g^{}(I^X)`$.
The only thing that has to be found is $`h(q)`$. If the index of $`X`$ is greater than $`1`$, then there are no curves which intersect the anticanonical class by $`1`$, so $`h(q)=0`$. If the index is 1, then any such curve is a line $`\mathrm{}`$ (with respect to the anticanonical class of $`X`$) and $`h(q)=\alpha _Xq`$. Besides, there are no lines on $`X`$ passing through the general point, i. e. $`H^{dimX}_{\mathrm{}}=0`$. This means that the coefficient $`e^{\alpha _Xq}I_{\mathrm{}}^X`$ at $`H^0`$ vanishes, so
$$\alpha _X=\frac{_{a=1}^l(\mathrm{}X_a)!}{_{a=1}^k(\mathrm{}Y_k)!}=\frac{_{a=1}^l(\mathrm{deg}X_a)!}{_{a=1}^kw_a!}.$$
2) It is easy to see that
$$^X(q)_{H^0}=\underset{n=0}{\overset{\mathrm{}}{}}\frac{_{i=0}^l(d_in)!}{_{j=1}^k(w_jn)!}q^n.$$
Consider the solution of the equation $`L[𝒥^X(q)]=0`$ as a series $`𝒥^X=_{n0}a_nq^n`$. Evidently, $`D[q^n]=nq^n`$. Hence
$$(\underset{i=1}{\overset{k}{}}\text{(}w_iD(w_i1)\text{)}_{𝒘_𝒊})[q^n]=(\underset{i=1}{\overset{k}{}}\text{(}w_in(w_i1)\text{)}_{𝒘_𝒊})q^n.$$
Analogously,
$$(q\underset{i=0}{\overset{l}{}}\text{(}d_iD+1\text{)}_{𝒅_𝒊})[q^{n1}]=(\underset{i=0}{\overset{l}{}}\text{(}d_i(n1)+1\text{)}_{𝒅_𝒊})q^n.$$
The differential equation on $`𝒥^X`$ translates into recursive equations on the coefficients $`a_n`$ using these equalities. Namely,
$$a_n=\frac{_{i=0}^l\text{(}d_i(n1)+1\text{)}_{𝒅_𝒊}}{_{i=1}^k\text{(}w_in(w_i1)\text{)}_{𝒘_𝒊}}a_{n1}.$$
Put $`a_0=1`$. By induction we get
$$a_n=\frac{_{i=0}^l(d_in)!}{_{j=1}^k(w_jn)!}q^n.$$
#### 3.1.1. Remark
It is easy to see that the Riemann–Roch operator is divisible by $`D^{dimYdimX}`$ from the left. Let $`L=D^{dimYdimX}\stackrel{~}{L}`$. Then $`\stackrel{~}{L}^X=0`$.
### 3.2.
Comments and examples.
#### 3.2.1.
Remarks.
In order to check that a complete intersection $`X=X_1\mathrm{}X_l`$ of general hypersurfaces of degrees $`d_1,\mathrm{},d_l`$ in a weighted projective space $`=(w_1,\mathrm{}w_k)`$ does not intersect the singular locus of $``$ one can use the following necessary (but not sufficient) condition.
$$\begin{array}{c}\text{The number of Cartier divisors among }X_i\text{ is greater than}\hfill \\ \hfill \text{the dimension of the singular locus of }\text{.}\end{array}$$
This means that
$$\begin{array}{c}(\text{a number of }d_i\text{’s that is divisible on each of }a_i\text{’s})>(\text{the maximal number of weights }w_i\hfill \\ \hfill \text{such that greatest common factor of the others is greater than 1})\end{array}$$
(see 2.4.6).
For an arbitrary toric variety the condition reformulates as follows. The number of movable divisors (i. e. those that we can move from any point) is greater than codimension of maximal cone whose edges are not subset of basis.
We assume that the Picard number of the ambient variety is $`1`$ just for simplicity. In cases of higher Picard number theorems 1.1 and 1.2 may be formulated and proved in a similar manner. We consider a multivariable $`q=(q_1,\mathrm{},q_m)`$ instead of the variable $`q`$ (here $`m=\mathrm{rk}\mathrm{Pic}X`$) a multidegree of a curve (with respect to the generators of Picard group) instead of a degree and so on. The only difference in this case is the expression for correction term $`e^{\alpha _Xq}`$.
If we lift the hypothesis $`\mathrm{Pic}X=`$ (or $`dimX3`$ in remark 1), then theorems 1.1 and 1.2 give *the restricted* $`I`$-series (see remark 2.2.6).
The main theorems can be generalized to Calabi–Yau varieties. Their proofs differ from proofs of theorems 1.1 and 1.2 only in formulas for correction term (see \[Gi97\]).
#### 3.2.2.
Examples.
Consider a general hypersurface $`X^k^n`$. Let $`k<n`$. Then, by theorem 1.1,
$$I^X=\underset{d=0}{\overset{\mathrm{}}{}}q^d\frac{\text{(}kH+1\text{)}_{𝒌𝒅\mathbf{+}\mathrm{𝟏}}}{(\text{(}H+1\text{)}_{𝒅\mathbf{+}\mathrm{𝟏}})^{n+1}}.$$
This coincides with the formula for $`I`$-series for a hypersurface in projective space from \[Gi96\].
If $`k=n`$, then, by theorem 1.1,
$$I^X=e^{k!q}\underset{d=0}{\overset{\mathrm{}}{}}q^d\frac{\text{(}kH+1\text{)}_{𝒌𝒅\mathbf{+}\mathrm{𝟏}}}{(\text{(}H+1\text{)}_{𝒅\mathbf{+}\mathrm{𝟏}})^{n+1}}.$$
This coincides with the formula 4.2.2 from \[Pr04\].
For projective space ($`w_i=1`$) theorem 1.1 is Givental’s Theorem for complete intersections in projective space.
## 4. Appendix: Golyshev’s conjecture
### 4.1.
Consider *a family* of counting matrices $`A^\lambda =A+\lambda \mathrm{𝟏}`$, where $`\mathrm{𝟏}`$ is the identity matrix (see 2.6.1 for the definition of counting matrix). Consider the one-dimensional torus $`𝔾_m=\mathrm{Spec}[z,z^1]`$ and the differential operator $`D=z\frac{}{z}`$. To construct the family of matrices $`M^\lambda `$, put its elements $`m_{kl}^\lambda `$ as follows
$$m_{kl}^\lambda =\{\begin{array}{cc}0,\hfill & \text{if }k>l+1,\hfill \\ 1,\hfill & \text{if }k=l+1,\hfill \\ a_{kl}^\lambda (\frac{}{z})^{lk+1},\hfill & \text{if }k<l+1.\hfill \end{array}$$
Consider the family of differential operators
$$\stackrel{~}{L}^\lambda =\mathrm{det}_{\mathrm{right}}(D\mathrm{𝟏}M^\lambda ),$$
where $`\mathrm{det}_{\mathrm{right}}`$ means “right determinant”, i. e. the determinant, which is calculated with respect to *the rightmost* column; all minors are calculated in the same way. Divide $`\stackrel{~}{L}^\lambda `$ on the left by $`D`$. We get the family of operators $`L^\lambda `$, so $`\stackrel{~}{L}^\lambda =DL^\lambda `$.
### 4.2.
Definition \[\[Go02\], 1.8\]. The equation of the family $`L^\lambda [\mathrm{\Phi }(z)]=0`$ is called *counting equation $`D3`$*.
### 4.3.
Golyshev’s conjecture \[V. Golyshev, \[Go02\]\]. The solutions of $`D3`$ equations for Fano threefolds with Picard group $``$ are modular. More precisely, let $`X`$ be such variety, $`i_X`$ be its index, and $`N=\frac{\mathrm{deg}X}{2i_X^2}`$. Then in the family of counting equations for $`X`$ there is one, $`L^{\lambda _X}[\mathrm{\Phi }(z)]=0`$, whose solution is an Eisenstein series of weight $`2`$ on $`X_0(N)`$.
More precise description of counting equations as Picard–Fuchs equations see in \[Go02\].
### 4.4.
Based on this conjecture, Golyshev gives a list of predictions of counting matrices of Fano threefolds. To check this conjecture one should find all such matrices.
There are $`17`$ families of smooth Fano threefolds with Picard group $``$ (V. Iskovskikh, \[Is77\], \[Is78\]). For 14 of them counting matrices were found by A. Beauville (\[Bea95\]), A. Kuznetsov (unpublished), V. Golyshev (unpublished) and the author (\[Pr04\]). The matrices for 2 other varieties are given by theorem 2.6.6.
All these matrices coincide with ones that were predicted by Golyshev. Thus, theorem 2.6.6 finishes the proof of Golyshev’s conjecture.
The author is grateful to I. Cheltsov, S. Galkin, A. Givental, V. Golyshev, M. Kazarian, S. Shadrin, K. Shramov, M. Tsfasman, and F. Zak for comments. |
warning/0507/hep-ph0507197.html | ar5iv | text | # Deconfinement transition dynamics and early thermalization in QGP
## I Introduction
Recent heavy ion collision experiments at RHIC have produced a wealth of data on hadron spectra and their anisotropies, in particular the magnitudes of radial and elliptical flows. This data reveals, perhaps unexpectedly, coherence in particle production and strong collective flow phenomena. It turns out that about 99% of the single hadron data ($`p_T<1.5`$ GeV) are very well described by the hydrodynamics of a near-perfect fluid, provided the initial condition of very rapid thermalization (in $`0.5fm/c`$) is introduced Heinz:2004pj \- Kolb:2003dz . This strongly indicates that the quark-gluon plasma (QGP) formed at RHIC energies is a strongly coupled fluid.
At asymptotically high temperatures above the deconfinement $`T_c`$, where the running coupling $`g(T)`$ is small, QCD is well-described as a gas of weakly coupled quasi-particles, and has been much studied by perturbative techniques. At the energy densities achieved in the high energy heavy ion collisions, however, perturbative treatment of the equilibration process appears not to be applicable. Various estimates of thermalization times based on perturbative scattering processes have been obtained, e.g. in the so called parton-cascade approach to the time evolution of hard partons, or the bottom-up scenario MG , BjV , BMSS incorporating saturation picture (see Kar , IV for review) initial conditions. They all result into thermalization times much longer than those needed by the hydrodynamical simulations. It has been argued that this is a generic feature of any dynamical evolution based only on perturbative scattering processes Kov .
Even within a weak coupling analysis, however, non-perturbative effects may contribute to the dynamics. It has been pointed out that, in a plasma with an anisotropic hard parton distribution, instabilities may develop in soft gauge field modes generated within the linear response approximation ALM \- H . It has been argued that these small-amplitude unstable modes are not stabilized by (non-Abelian) non-linearities; and thus can grow to amplitudes large enough to contribute $`O(1)`$ fraction to the total energy density, and drive isotropization of the hard modes faster than any hard collision equilibration time scale ALMY .
Various investigations of the evolution of these semiclassical instabilities have been carried out probing beyond the linear regime, and, in particular, within the hard-loop effective action Rebh1 \- Rebh2 . They generally indicate that such instabilities indeed persist when non-Abelian non-linearities are taken into account within the weak coupling regime. At some point, however, other non-perturbative dynamics at strong effective couplings defined at scales appropriate to the nonlinear interaction of such growing long range modes must enter. Still, consideration of such potential instabilities properly point to a basic underlying question which should be addressed from a more general point of view.
When systems are driven far from equilibrium by sudden changes in external conditions, the approach to a new equilibrium state involves, in general, complex nonequilibrium processes. Such situations arise, for example, in a ‘quench’ to the metastable region across a first-order transition boundary, or from a one-phase region to the multiphase coexistence region across a second-order transition. The study of the dynamics of such far-from-equilibrium processes is still at an early stage of development. Nonetheless, from many studies, mostly in condensed matter physics systems, two broad classes of responses have been roughly identified.
Immediately after a rapid quench, the state of the system, as characterized by the appropriate order parameter, is nearly identical to the state before. The fields must then find some way to adjust toward values appropriate to the conditions after the quench. One way the system may decay towards equilibrium is by the excitation of finite amplitude localized fluctuations that may grow or coalesce as in nucleation, or interact in other ways over time. This typically indicates that the system finds itself in some sort of metastable state. Another way is by the immediate development of a spatially modulated order parameter whose amplitude grows continuously from zero throughout the sample. Spinodal decomposition is a prime example of this type of response, and the term is often used loosely to generically denote such globally unstable behavior. It should be pointed out that the boundary between these two rough classes is not sharp in all cases.
At RHIC heavy-ion collisions in the central rapidity region achieve the sudden deposition of energy densities reaching $`30GeV/fm^3`$. The first most basic question that must be posed then is: which general type of dynamic response is characteristic of a rapid transition across the confinement-deconfinement boundary in QCD? In the case of heavy-ion collisions, the question must be further qualified by the inclusion of the effects of rapid expansion and temperature variation, as well as finite volume. This is the issue we explore in this paper.
To get some intuition, we first investigate this question briefly within an effective action approach in Section 2. Indeed, much of the current understanding of the early time evolution of systems out of equilibrium has been obtained by investigating classes of stochastic equations that are natural dynamical (time dependent) generalizations of the Landau-Ginzburg (LG) effective action models of the static (equilibrium) theory (gunton:1985 , Chaikin:1997 ). In the case of the confinement - deconfinement transition, the relevant order parameter is the Polyakov loop (Wilson line), and LG effective actions for it have been considered in Pisarski:2000eq \- HKW . Corresponding dynamical model generalizations can then be used to examine the question posed above. We consider the predictions of such a model briefly in Section 2 below.
Though the effective model approach often proves valuable, what is ultimately needed is an ab initio treatment in the full nonperturbative formulation of the exact theory, i.e. lattice gauge theory. Unfortunately, there is no established simulation formalism for directly extracting physical properties in non-equilibrium real-time evolution in quantum field theory. What one can do, however, is mimic Minkowski real-time dynamics by Glauber stochastic dynamics evolution. Thus starting with the system in thermalized equilibrium, one performs a temperature quench and follows the system, over successive simulation sweeps, on its path toward regaining equilibrium. Though this cannot be directly identified with the exact real-time evolution, it is known from many studies, mostly of condensed matter systems, to accurately reflect it. At the very least, the method provides a consistent picture of the basic features of the system’s real-time response. It has been extensively and successfully used for many systems exhibiting, in particular, first-order transitions. Indeed, in studies of binary alloys and binary fluids it has been found to reproduce the experimentally observed behavior in quantitative detail gunton:1983 . In the case of gauge theories, the method was first used in the pioneering studies in Miller:2000mr ; Miller:2000pd ; Miller:2001ym . More, recently, such studies of the dynamics of phase transitions in spin and gauge theory systems were undertaken and much extended in Bazavov:2004wc ; Berg:2003mn ; Berg:2003hc ; Berg:2004qb ; Velytsky:2002fn .
In this paper, building on these previous gauge theory studies, we consider $`SU(3)`$ gauge theory under conditions mimicking the situation encountered in heavy-ion collisions. This we do in Section 3 which constitutes the main part of the paper. Specifically, we examine the response after a sudden quench into the deconfinement phase under varying conditions of spatial expansion and temperature variation. In this first investigation, we consider only pure $`SU(3)`$ gauge theory, i.e. no quarks, as the deconfinement transition is driven by gluonic dynamics. In simulation measurements of structure functions and Polyakov loops, we follow the evolution of the system from the quench till its return to the confinement phase. The main outcome of our study is that two relevant, widely separated time scales emerge. First, a strong and robust signal of rapid growth of very long range modes is observed after the quench. Most importantly, the time scale of full development of these low momentum modes is essentially unaffected by the presence, over a wide range of parameters, of spatial expansion and temperature variation in the system. The development of these modes ‘isotropizes’ the system, which reaches full quasi-equilibrium shortly afterwards as signaled by the full decay of the structure function. It thus enters a stage of expansion and cooling, characterized by a second, much longer time scale, till its return to the confinement phase. The wide separation of these two scales accords well with what is seen in heavy-ion collision experiments. Having arrived at this robust qualitative picture, we also attempt to make a more quantitative comparison of what is seen in these simulations to hydrodynamical phenomenology (subsection 3.3). There are certainly uncertainties here, not least of which is the fact that we do not include fermions which can make a significant contribution especially at the late stage near the return to confinement. Still, one finds that the separation of scales is such that, for any reasonable choice of parameters, isotropization times well inside $`1fm/c`$ result naturally.
Finally, in Section 4 we briefly discuss our conclusions and directions for further work.
## II Effective action models
Effective action models allow one to build simple theories of initial time evolution of systems driven out of equilibrium by a quench. Such theories can be build for general classes of models. A simple linear theory (Cahn-Hilliard) was first proposed for models of type B with conserved order parameter Cahn:1958 . The generalization to models of type A with non-conserved order parameters is straightforward, see, for example, Berg:2004qb ; Velytsky:2004 . For a general overview see e.g. gunton:1985 ; Chaikin:1997 .
Though such effective models are not our primary focus in this paper, it is useful to briefly consider them as they provide useful insight into the possible behavior of the system that can be checked against the outcome of simulations in the actual gauge theory. For $`SU(3)`$ gauge theory the low energy degrees of freedom are represented by Polyakov loops, and standard potential models for them are known and well studied Pisarski:2000eq ; W ; HKW . Adopting the potential for the Polyakov loop $`l`$ (a complex quantity) in Pisarski:2000eq
$$𝒱(l)=\left(\frac{b_2}{2}|l|^2\frac{b_3}{6}(l^3+(l^{})^3)+\frac{1}{4}(|l|^2)^2\right)b_4T^4,$$
(1)
the coupled set of Langevin equations is
$$\frac{l}{t}=\mathrm{\Gamma }\frac{\delta S}{\delta l^{}}+\eta ,\text{and its}c.c.$$
(2)
Here $`S`$ is the standard Landau-Ginzburg action
$$S=d^3x\left(\frac{1}{2}|_il|^2+𝒱(l)\right),$$
(3)
$`\mathrm{\Gamma }`$ is the response coefficient, which defines the relaxation time scale of the system, and $`\eta `$ is a noise term. We will ignore the noise term, since it can be shown that it does not affect the resulting rate of growth of fluctuations F1 .
We are interested in the fluctuations of the Polyakov loop $`l`$ around some average value $`l_0`$:
$$l(\stackrel{}{r},t)=l_0+u(\stackrel{}{r},t),$$
(4)
Using (1) and (2), the system of equations for the fluctuations is
$$\frac{u}{t}=\mathrm{\Gamma }\left[\frac{1}{2}^2u+\left(c_1u+c_2u^{}+c_3\right)\right]\text{and its}c.c.,$$
(5)
where $`c_1=(b_2/2+|l_0|^2)b_4T^4`$, $`c_2=(1/2l_0^2b_3l_0^{})b_4T^4`$ and $`c_3=(b_2/2l_0b_3/2l_0^2+1/2l_0^2l_0^{})b_4T^4`$ are complex numbers. Note that $`c_1=c_1^{}`$.
We solve for the Fourier transforms $`u(\stackrel{}{k},t)`$ and $`v(\stackrel{}{k},t)`$ of the fluctuations $`u(\stackrel{}{r},t)`$ and $`u^{}(r,t)`$, respectively, which, from (5) satisfy:
$`{\displaystyle \frac{u(\stackrel{}{k},t)}{t}}+\mathrm{\Gamma }\left[({\displaystyle \frac{1}{2}}k^2+c_1)u(k)+c_2v(k)\right]`$ $`=`$ $`c(k)`$
$`{\displaystyle \frac{v(\stackrel{}{k},t)}{t}}+\mathrm{\Gamma }\left[({\displaystyle \frac{1}{2}}k^2+c_1^{})v(k)+c_2^{}u(k)\right]`$ $`=`$ $`c^{}(k),`$ (6)
where $`c(k)=c(k)=\mathrm{\Gamma }c_3_\stackrel{}{r}\mathrm{exp}(i\stackrel{}{k}\stackrel{}{r})`$. Note also that $`u^{}(\stackrel{}{k},t)=v(\stackrel{}{k},t)`$. The non-zero modes are governed by the homogeneous part of these equations. Its eigenvalues are
$$\omega _{1,2}(k)=\mathrm{\Gamma }(\frac{1}{2}k^2+c_1\pm |c_2|),$$
(7)
and the eigenvectors are
$$S_{1,2}=\left(\begin{array}{c}\pm c_2/|c_2|\\ 1\end{array}\right).$$
(8)
The solution is
$$\left(\begin{array}{c}u(\stackrel{}{k},t)\\ v(\stackrel{}{k},t)\end{array}\right)=C_1S_1e^{\omega _1(k)t}+C_2S_2e^{\omega _2(k)t}.$$
(9)
Here, an analysis of modes similar to that of the real case (spin systems, $`SU(2)`$) (Bazavov:2004wc \- Berg:2004qb ) can be applied. If $`c_1\pm |c_2|<0`$, one will observe exponential growth. Spinodal decomposition-like behavior then results if we have exponential growth in at least one exponent, i.e. when $`c_1<|c_2|.`$ The structure function (connected Polyakov 2-point function)
$$S(\stackrel{}{k},t)=<u(\stackrel{}{k},t)v(\stackrel{}{k},t)>$$
(10)
follows similar behavior (Cf. Bazavov:2004wc ).
Next, we want to study the effect of spatial expansion. We work in new coordinates of rapidity $`\eta =1/2\mathrm{ln}(t+z)/(tz)`$ and proper time $`\tau `$ :
$$t=\tau cosh(\eta ),z=\tau sinh(\eta ).$$
(11)
Here we assume the $`z`$-axes to be the axes of expansion (collision). The change in the corresponding part of the Minkowski metric is $`dt^2dz^2=d\tau ^2\tau ^2d\eta ^2`$, with the transverse coordinates left unchanged. Keeping the rapidity constant results in a constant rate of expansion in the system, the speed of expansion being proportional to the distance from the collision center: $`v=z/t`$, with $`t`$ the time after collision. In the new coordinates the metric is similar to the Robertson-Walker metric and is defined as $`ds^2=g_{\mu \nu }dx_\mu dx_\nu =d\tau ^2a^2(\tau )d(\tau _0\eta )^2dx_{}^2`$, i.e.
$$g_{\mu \nu }=\text{diag}(1,1,1,a^2(\tau ))$$
(12)
where $`a(\tau )=\tau /\tau _0`$ is the ‘scale factor’, and $`\tau _0`$ is the parameter controlling the rate of expansion.
The obvious naive generalization of the model above is to substitute
$$^2g_{ij}_i_j$$
in the action (3), and to consider dynamics in the proper frame.
Eq. (2) now gives
$$\frac{l}{\tau }=\mathrm{\Gamma }\left[_{}^2la^2(\tau )_{}^2l+\frac{V(l)}{l}\right],\text{and its}c.c.,$$
(13)
where $`_{}^2=_x^2+_y^2`$ and $`_{}^2=_\eta ^2`$. Going through the previous development amounts to the naive substitution $`\stackrel{}{k}^2\stackrel{}{k}_{}^2+a^2(\tau )k_{}^2`$ in the eigenvalues:
$$\omega _{1,2}(k)=\mathrm{\Gamma }\left[\frac{1}{2}(\stackrel{}{k}_{}^2+a^2(\tau )k_{}^2)+c_1\pm |c_2|\right].$$
(14)
This now gives a growth of fluctuations governed by a 3rd order polynomial in $`\tau `$ in the exponent:
$$\left(\begin{array}{c}u(\stackrel{}{k},t)\\ v(\stackrel{}{k},t)\end{array}\right)=C_1S_1\mathrm{exp}_0^\tau \omega _1(\stackrel{}{k},\tau ^{})𝑑\tau ^{}+C_2S_2\mathrm{exp}_0^\tau \omega _2(\stackrel{}{k},\tau ^{})𝑑\tau ^{}.$$
(15)
Thus we observe that if
$$\frac{1}{2}(\stackrel{}{k}_{}^2+a^2(\tau )k_{}^2)<c_1|c_2|,$$
there is an explosive growth, much faster than in the non-expanding case. Let us for simplicity consider only longitudinal modes. We see that the critical mode Chaikin:1997 is
$$k_{}^c=\frac{2}{a^2(\tau )}(c_1|c_2|).$$
(16)
This is rather remarkable dynamics where the critical mode (and all modes at momentum scales below it) is moving in time. During initial time there are more scales involved but as time progresses only very large scale regions participate. The prediction of growth under expansion by a higher than linear power exponent is tested against the simulation results in the following section.
## III Simulation study of deconfinement dynamics in $`SU(3)`$ LGT
In this section we present a numerical study of the effects of spatial expansion and varying temperature on the dynamical evolution following a sudden quench into the deconfined phase of the $`SU(3)`$ gauge theory on the lattice. We thus try to mimic conditions encountered in heavy-ion collisions. We use periodic boundary conditions, as we do not study the effects of finite size per se. A study of the role of the finiteness of the system would certainly be of interest, but is left for a future study. Also, in the customary heavy-ion collision picture the initial expansion is one-dimensional, becoming three-dimensional at later stages of the evolution Bjorken:1983 . Here, as we discuss further below, we simplify matters by considering a uniform expansion in all spatial directions. This is related to isotropic expansion in the proper frame at a fixed rapidity value. We first study the effect of such expansion; then we add the effect of varying temperature.
We use a field based heat bath update of the $`SU(3)`$ fields. To accelerate the dynamics there are 2 attempts to update the field per sweep. This is somewhat different from standard link-based updates; however, it is still in Glauber universality class. The time flow is proportional to the link-based heatbath with links visited in random order. Therefore, the peak values of the non expanding quench are slightly different from previous studies Bazavov:2004wc where a heat bath update is performed on links visited in systematic order.
We use a space-time anisotropic lattice in order to be able to independently control temperature and expansion. The anisotropic action is Karsch:1982ve
$$S=\beta _\xi /3\underset{x}{}\text{Re}\left[\xi ^1\underset{i>j}{}\text{Tr}U_{x,ij}+\xi \underset{i}{}\text{Tr}U_{x,0i}\right],$$
(17)
where $`i`$, $`j`$ runs over space-like directions, $`\xi =a/a_\tau `$ is the space-time anisotropy, $`\beta _\xi =6/g_\xi ^2`$, and $`g_\xi ^2=g_\sigma g_\tau `$ is the anisotropic coupling. Then, by varying two parameters it is possible to carry out expansion of the system while maintaining constant temperature, or also let the temperature drop thus allowing for cooling of the expanding plasma.
### III.1 Spatial expansion
Hubble-like uniform expansion of the metric amounts to varying the space-like lattice spacing $`a`$ as
$$a=a_0(1+\frac{\tau }{\tau _0}),$$
(18)
where $`\tau `$ is the proper time variable, and $`\tau _0`$ is the parameter which controls the rate of expansion.
We focus here solely on the role of the expansion. So, after the initial quench, we keep the temperature $`T=1/(N_\tau a_\tau )`$ constant throughout the expansion by fixing the time-like spacing $`a_\tau `$. This implies that the anisotropy $`\xi =a/a_\tau `$ traces the changes in $`a`$. Thus, assuming zero time anisotropy to be $`\xi (\tau =0)=1`$, one has
$$\xi (\tau )=1+\frac{\tau }{\tau _0}.$$
(19)
Next consider the dependence of the space-like lattice spacing on the coupling and anisotropy. The one-loop order renormalization group relationship is
$$a\mathrm{\Lambda }(\xi )=\mathrm{exp}\{1/(2b_0g_\xi ^2)\},$$
(20)
where $`b_0=113/(48\pi ^2)`$, and $`\mathrm{\Lambda }(\xi )`$ is dependent on $`\xi `$ Karsch:1982ve through
$$\mathrm{\Lambda }(\xi )/\mathrm{\Lambda }_E=\mathrm{exp}\{(c_\sigma (\xi )+c_\tau (\xi ))/4b_0,$$
(21)
where $`c_\sigma (\xi )`$ and $`c_\tau (\xi )`$ are known functions. Inclusion of the next order terms adds minor corrections for the range of the couplings and anisotropies used and is straightforward. It does, however, complicate numerical treatment, since it requires a numerical solving of the corresponding equation.
Following the procedure in Karsch:1982ve we compute $`\mathrm{\Lambda }(\xi )`$ for a range of anisotropy values as listed in table 1.
This allows us to estimate the necessary time evolution of $`\beta _\xi `$
$$\beta _\xi (\tau )=\beta (0)62b_0\mathrm{log}\left[(1+\frac{\tau }{\tau _0})\frac{\mathrm{\Lambda }(\xi )}{\mathrm{\Lambda }_E}\right].$$
(22)
It is important to indicate here that a non-perturbative further correction Boyd:1996bx need to be applied, when appropriate (see subsection 3.3 below).
Note that in the standard application of anisotropic lattices (such as in Karsch:1982ve ) the space-like lattice spacing is not varied; the anisotropy is varied by decreasing the time-like lattice spacing. This procedure keeps the coupling within the scaling window provided the initial coupling is close to the continuum limit. Here, on the contrary, we keep the time-like spacing constant (or, later, slowly increase it) as we vary the space-like coupling. This induces changes in the coupling that may drive its value out of the scaling regime ($`\beta 5`$). Therefore, our expansion has to be truncated whenever the value of $`\beta `$ falls below this cut-off value. This condition implies that, in order to follow the system evolution for longer time, one needs lattices with larger $`N_\tau `$, thus rendering the problem more computationally intensive.
We start simulations on smaller lattices where it is easier to gather satisfactory statistics for highly fluctuating quantities, such as the structure function. The quench is performed from $`\beta _\xi =5.5`$ to $`\beta _\xi =5.92`$ on $`16^3`$x$`4`$ lattice. The latter corresponds to a temperature after the quench $`T_{\mathrm{final}}=1.57T_c`$. The phase transition on this lattice at $`\xi =1`$ is at $`\beta =5.6902(2)`$. We use jack knife average over 10 bins, each of 50 configurations. The system is allowed to equilibrate for 200 lattice sweep, and then, after performing a quench, we allow the system to evolve for 800 sweeps, while measuring several lower modes of the structure function. We present here only averages of on-axis modes, such as permutations of $`(n,0,0)`$ \- this is the n-th mode in our notation. The first modes are presented in Fig. 1. We see that expansion significantly enhances the response. The faster the expansion rate, the higher are the peaks. On the other hand the shift in the location of the peaks is not as pronounced, an important point to which we return below.
Next, we look at higher modes of the structure function in the cases of no expansion, and expansion at $`\tau _0=1000`$ – see Fig. 2. We see that the second mode shows behavior similar to that observed for the first mode. The difference, however, is not that significant. The third mode in the expanding system outgrows the corresponding mode in the non-expanding case at early times, but then decreases faster. This is an indication of the shift of the critical mode with time, observed in the linear effective model (section 2). At later times the transition proceeds only through the lower modes. The error bars for some of the data points are also presented in Fig. 1 and 2.
To make a comparison to the linear effective theory of Section 2, we make fits to the exponent for the structure function data (Fig. 3). We use a $`32^3`$x$`4`$ lattice since we know from previous studies Berg:2004qb ; Bazavov:2004wc that the linear response behavior (pertinent to early times) manifests itself better on the larger lattices. Contrary to our expectations from the linear theory of section 2, however, we find there no substantial change in the behavior of the exponent between the expanding and non-expanding systems. A fit to
$$S(\tau )\mathrm{exp}(C\tau ^\alpha )$$
gives $`\alpha =1.22`$ for the expanding system, whereas it gives $`\alpha =1.18`$ for the non-expanding system. In Fig. 3 we also show fits to $`S(\tau )\mathrm{exp}(C\tau )`$, which in fact provides the best fit per parameter degree of freedom. Both fits are over the $`\tau `$ range from $`0`$ to $`250`$. In any case, there is no substantial deviation from exponential growth with linear $`\tau `$ dependence in the exponent. Furthermore, from the figures we observe divergence from exponential behavior at later times $`\tau >200`$. Also notable is the enhancement of the signal (by a factor $``$ 7) in the expanding system as compared to the non-expanding system. All this provides a manifestation of the limitations of the linear response effective models in Section 2. We note that there are 10 times as many points for the fitting as on the plot.
### III.2 Expansion accompanied by temperature fall-off
In this second part of the numerical study we try to mimic conditions similar to those in heavy ion collisions at RHIC. We want to follow the evolution of the expanding and cooling QGP. We let the temperature drop as
$$T=\frac{T_0}{(1+\tau /\tau _0^{})^\alpha }.$$
(23)
In the fire-tunnel (in the proper frame) the temperature is expected to decrease with the proper time $`\tau `$ as $`T\tau ^{1/3}`$ or slower Bjorken:1983 . We adopt this value of $`\alpha `$ in the following.
We first work on smaller lattices for elucidating the main physical features. Therefore, we do not set the freeze-out condition (see below) and simply take $`\tau _0^{}=\tau _0`$. With the $`\alpha =1/3`$ choice of temperature evolution, the anisotropy evolves as
$$\xi (\tau )=\left(1+\frac{\tau }{\tau _0}\right)^{2/3}$$
(24)
and $`\beta `$ evolves as in (22). We illustrate the effect of temperature drop on the structure function in Fig. 4.
There are two basic conclusions suggested by these plots.
The first is that the temperature evolution drives the system back towards the confined phase, but the expansion tends to prevent it as evidenced by the different peak heights. These are then two competing effects that tend to cancel, so that the return to structure function equilibration occurs at about the same time (here after about $`700`$ sweeps) as for the system in the absence of expansion and temperature fall-off. For the slower expansion rates ($`\tau _0>1000`$) and faster temperature fall-offs this cancellation effect is even more pronounced.
The second conclusion is suggested by the fact, also present and remarked upon in Fig. 1, that the location of the peaks is little affected by the presence of expansion and/or temperature evolution. This implies that the system’s response to the sudden quench is set by an internal dynamics scale that is faster than that of the expansion and accompanying temperature fall-off rates considered here. After the structure function is past its peak, the system is isotropized, and, after a relatively short time, any memory of the initial fast, spinodal-like, long range response to the violent quench across the deconfinemnt transition boundary disappears. The subsequent evolution is that of (quasi)equilibrium evolution of the system as it expands and cools towards its return to the confinement phase.
To elucidate this further we plot the time profile of the Polyakov loop average in Fig. 5 for the expanding and non-expanding systems as well as with and without drop in temperature. We now have to use a bigger, $`32^3\times 8`$ lattice in order to follow the evolution over a longer time interval.
First note that pure expansion drives the Polyakov loop towards larger values, i.e. more pronounced deconfined behavior. To get some insight into this behavior observe that, as it is evident from (17), spatial expansion enhances the timelike part of the action (2nd term in the square brackets on the r.h.s.) while suppressing the spacelike parts F3 . This tends to increase the expectation of timelike Polyakov lines Fsan .
On the other hand, decreasing temperature counteracts the expansion effect. This is clearly seen in Fig. 5. Eventually, under the combined effects of expansion and temperature fall-off, the Polyakov loop expectations drops to zero signaling the return of the system to the confinement phase. These qualitative features of Fig. 5 are rather generic, being stable under changes in the expansion and temperature fall-off rates (cf. Fig. 6 below).
The crucial feature characterizing the system’s overall evolution following the rapid quench into the deconfinement region is clearly revealed by examining Fig. 5 in conjunction with the plots of the structure function in Fig. 4. It is the fact that there are two scales involved in this evolution. The first is the scale set by the location of the peak of the structure function; the second is the scale set by the interval to return to the confinement phase. Furthermore, there is wide separation between these two scales. As noted above, the location of the peak ($`300`$ in Fig. 4) is very little affected by the conditions of expansion and temperature fall-off. It reflects strongly coupled dynamics at short time scales driving the exponential growth of very long range modes F4 leading to ‘isotropization’, by which we mean nothing more specific than that the full development of these modes appear to completely wipe out any remnants of the quenching event. Rapid equilibration follows within an interval of a few hundreds sweeps ($`300400`$ in fig. 4). The system then continues to evolve in quasi-equilibrium over a much longer period (typically of thousands of sweeps as in Fig. 5 and Fig. 6 below) expanding and cooling till it returns to the confinement phase. This separation of time scales is a very general feature over a wide range of parameters.
Having reached this qualitative physical picture, which accords well with that experimentally observed, it is interesting to explore whether it is possible to make an estimate of this separation in physical units, and establish some correspondence with heavy-ion collision phenomenology. This we do in the following subsection.
### III.3 Isotropization-thermalization and chemical freeze-out times
The expansion and temperature variation parameters ($`\tau _0^{}`$, $`\tau _0`$, $`T_{\mathrm{final}}`$) determine the precise time interval before returning to the confinement phase. Note that our requirement that the evolution remain inside the scaling window limits the time of observation. To deal with this and follow the evolution for substantial time we now work on a larger, $`32^3`$x$`8`$ lattice throughout, and attempt to set the parameters so as to reflect conditions in heavy ion collision at RHIC. There is strong evidence that the matter in the firetunnel reaches temperatures above $`2T_c`$ Kolb:2003dz . For our numerical simulation we quench to $`T_{\mathrm{final}}3T_c`$. (Note that $`T_c`$ here means critical temperature of pure gluodynamics). The system is equilibrated in the confinement phase at the same temperature as before $`0.8T_c`$. For $`N_\tau =8`$ and $`\xi =1`$ we recalculate the values of corresponding betas using the two-loop order formula connecting lattice spacing and the coupling, and reweigh it with non-perturbative correction factor above $`T_c`$ Boyd:1996bx . Such a correction is needed here since the bare gauge coupling is typically of order one for our range of betas. The value of $`\beta _c`$ is known from the lattice Polyakov loop susceptibility study Boyd:1996bx . We get
$`\beta _{\mathrm{initial}}`$ $`=5.90`$ $`0.76T_c`$
$`\beta _c`$ $`=6.0625`$ $`T_c`$ (25)
$`\beta _{\mathrm{final}}`$ $`=6.85`$ $`2.95T_c,`$
where ‘initial’ and ‘final’ refer to before and after the quench.
At high enough temperatures the system is close to an ideal gas of quasiparticles. Lattice data suggest that the equation of state does not actually deviate much from that of the ideal gas for all temperatures down to $`T_c`$, and is conveniently modeled as such in hydrodynamic descriptions. Hence, the energy and entropy densities scale as $`eT^4`$ and $`sT^3`$, respectively. Assuming adiabatic expansion the entropy per unit of rapidity is conserved. In real fire-tunnel evolution the initial expansion is one dimensional Bjorken:1983 , switching to three dimensional expansion at later times. This corresponds to $`T\tau ^{1/3}`$ at the early and mid time and $`T\tau ^1`$ at later time. The $`T\tau ^{1/3}`$ behavior is in fact seen in the hydrodynamic evolution almost down to $`T_c`$ Heinz:2004pj ; Kolb:2003gq ; Kolb:2003dz , and is the only one considered in the following. We are thus led to modeling of the spatial expansion and temperature drop on the lattice in terms of the two parameters $`\tau _0`$ and $`\tau _o^{}`$ as:
$`a_s`$ $`=`$ $`a_0(1+{\displaystyle \frac{\tau }{\tau _0}}),`$ (26)
$`a_\tau `$ $`=`$ $`a_0(1+{\displaystyle \frac{\tau }{\tau _0^{}}})^{1/3}`$ (27)
$`=`$ $`a_0(1+y{\displaystyle \frac{\tau }{\tau _0}})^{1/3},`$
where $`y=\tau _0/\tau _0^{}`$ is the ratio of the ‘speeds’. The evolution of the anisotropy then is
$$\xi (\tau )=\frac{1+\frac{\tau }{\tau _0}}{(1+y\frac{\tau }{\tau _0})^{1/3}}$$
(28)
As the chemical freeze-out temperature we choose $`T_{fo}=T_c`$ Kolb:2003dz ; Heinz:2004pj ; Kolb:2003gq . Before this freeze-out the plasma undergoes an $`x`$-fold expansion
$$x=\frac{a_s}{a_0}=1+\frac{\tau _{fo}}{\tau _0}$$
(29)
On the other hand
$$\frac{T_{fo}}{T_{final}}=\frac{1}{(1+y\frac{\tau _{fo}}{\tau _0})^{1/3}}$$
(30)
From these two we get
$$y=\frac{(T_{final}/T_{fo})^31}{x1}$$
(31)
Hydrodynamical model phenomenology yields all required parameters Kolb:2003gq . Thus we have $`x9`$ before the conversion to the confined phase is completed, and the freeze-out is observed. This corresponds to $`y3.25`$.
In Fig. 6 we plot the time profile of the Polyakov loop average in expanding and non-expanding systems with and without drop in temperature with these values of $`x`$, $`y`$, and $`\tau _0=1000`$. One again observes the same features discussed above in connection with Fig. 5.
In Fig. 7 we show the lowest mode structure function in the case of no expansion and temperature variation. The expanding-variable temperature case is very similar. We see that at $`\tau =400`$ sweeps the system has reached the peak. This is the point of isotropisation of the system when the long range fluctuations reach through the system. It is again to be contrasted to the much longer times needed to return to confinement (vanishing Polyakov loop expectation).
To explore this difference in detail, in Fig. 8 we present the Polyakov loop average evolution at different expansion rates. The figure on the top panel uses again $`x=9`$, so the chemical freeze-out point on this plot is at $`\tau /\tau _0=(x1)=8`$. We notice that in the range of $`\tau _0400600`$ the Polyakov loop expectation value gets close to zero just before this point. Actually, this should be considered as a lower bound on the range of $`\tau _0`$’s since this is a first order transition (or a rapid crossover in the presence of fermions). Thus, there is a latent heat period during which the system lingers, with Polyakov loop expectations close to zero, before it is fully converted to the confined phase. The values $`\tau _0=300`$ and less (higher ‘speeds’) do not show this behavior and therefore do not lead to a consistent picture. Taking then the range of $`\tau _0400600`$ gives an interval to freeze-out of $`32004800`$ sweeps, which in turn corresponds, from phenomenology, to $`9fm/c`$. This allows an upper bound estimate of the time around the peak of the structure function in physical units: $`0.751.1fm/c`$.
In the plot on the bottom panel in Fig. 8 we use the same quench to $`2.95T_c`$ but now take $`x=10`$, which gives $`y=2.89`$. This corresponds to a somewhat longer lifetime for the deconfined plasma. One sees that a value of $`\tau _0700800`$ gives a good fit to complete conversion to the confined phase by the time the freeze-out point is reached. This in turn gives $`0.560.63fm/c`$ as an upper bound estimate for the time around the peak of the structure function. There is of course an inherent uncertainty here as to what to take for the appropriate phenomenological value for the lifetime in physical units since we do not include fermions in out simulations; and fermions become important at the late stage of conversion back to confinement. In Fig. 8 we follow the curves as far as possible before running out of the scaling regime as explained in Section 3.1 above. Longer lifetimes result into even shorter isotropization times. To consider somewhat larger $`x`$ values, however, we need larger lattices than those employed in this study. But the main message extracted from the present decimations should be clear. The robust separation between the ‘fast’ dynamics scale of the exponential growth of the spinodal-like response to the sudden quench and that set by return to confinement is such that isotropization times well inside $`1fm/c`$ result naturally for any reasonable choice of parameters.
## IV Conclusions
In this paper we studied the response of the pure $`SU(3)`$ gauge theory to a rapid quench from its confined to its deconfined phase. In a series of simulations we followed the subsequent evolution of the system under varying conditions of temperature fall-off and/or spatial expansion. These conditions were chosen so as to reflect the type of variations presumed to hold in heavy ion collisions. Our main finding is that there are two distinct scales characterizing this evolution. There is one scale set by the development of very long range modes continuously from zero to their maximum amplitude over a short time interval. These modes, manifested in the response of the structure functions to the quench, drive isotropization and return to thermalization shortly afterwards. The scale set by this ‘fast’ dynamics is little affected by conditions of expansion and temperature variations. The second scale is set by the time interval to return to the confinement phase under quasi-equilibrium evolution, and is affected by the details of expansion and temperature fall-off. There is a robust wide separation between the two scales, which, translated to physical units under reasonable assumptions about the lifetime of the plasma, gives estimates of the ‘fast’ dynamics in the range of $`0.51fm/c`$.
There is a number of directions in which this study can be extended. The formalism used above can be extended to treat also anisotropy among the different space directions, thus allowing a more ‘realistic’ treatment of spatial expansion. Whether this makes much of difference, however, remains to be seen (cf. footnote Fsan ). Another interesting question is that concerning the effect of having a really finite physical system. In our simulations we, as usual, employed periodic boundary conditions. But one may explore different ones that would mimic a finite rather than an infinite system. The other obvious extension is the inclusion of fermions. The deconfinement transition is driven by gluonic dynamics. Fermions, however, are expected to contribute more substantially at the late stage of return to confinement and hadronization. Thus our study applies strictly to the gluonic plasma. Nonetheless, the qualitative picture of the separation of scales found above should not be affected in an essential way by the inclusions of fermions, though quantitative details related to the exact lifetime of the plasma, etc certainly will.
## Acknowledgment
We thank Academical Technology Services at UCLA for computer support. Our code is build on top of the SciDac qdp++ library, which also forms the foundation for chroma Edwards:2004sx . We also like to thank B. Berg for discussions. This work is partially supported by NSF-PHY-0309362. |
warning/0507/cond-mat0507650.html | ar5iv | text | # Quantum rings as electron spin beam splitters
## Abstract
Quantum interference and spin-orbit interaction in a one-dimensional mesoscopic semiconductor ring with one input and two output leads can act as a spin beam splitter. Different polarization can be achieved in the two output channels from an originally totally unpolarized incoming spin state, very much like in a Stern-Gerlach apparatus. We determine the relevant parameters such that the device has unit efficiency.
The Stern Gerlach experiment, where spatial and spin degrees of freedom become intertwined, has been playing a fundamental role in the conceptual foundations of Quantum Mechanics. Still, soon after the discovery of this effect, it was pointed out by Bohr and Mott Mott and Massey (1949) that, in contrast to atoms, electrons can not be spin-polarized in an inhomogeneous magnetic field. The recent spectacular development of spin electronics (spintronics) Žutić et al. (2004) in low dimensional semiconductor structures offers a new way of manipulating spin degrees of freedom. Quantum rings made of semiconducting material Viefers et al. (2004) exhibiting Rashba-type Rashba (1960) spin-orbit interaction (SOI) have been shown to be especially important due to their remarkable spin transformation properties Nitta et al. (1997, 1999); Molnár et al. (2004); Frustaglia and Richter (2004); Földi et al. (2005); Zhai and Xu (2005); Koga et al. (2004).
In the present paper we propose a device that can be considered to a large extent a spintronic analogue of the Stern-Gerlach apparatus: the incoming electrons are forced to split into two different spatial parts by the geometrical construction of the semiconductor device, see Fig. 1. Due to spin-sensitive quantum interference Souma and Nikolić (2005); Kato et al. (2005); Cserti et al. (2004) and spin-orbit interaction, electrons that are initially in a totally unpolarized spin state become polarized at the outputs with different spin directions. A similar polarizing effect has been predicted in a Y-shaped conductor as a consequence of scattering on impuritiesPareek (2004) (which is a different physical mechanism from the coherent spin transfer to be discussed here) or because of the presence of SOI in a localized area around the junction.Kiselev and Kim (2001) There are important proposals considering four terminal devices Ionicioiu and D’Amico (2003); Governale et al. (2002) as well, where the strength of the SOI is assumed to be different in the two arms of the interferometer. In our model SOI is uniform in the ring and absent in the leads. However, the latter requirement is not crucial, its purpose is to demonstrate clearly the role of the ring itself, while the effects caused by SOI in the leads can be included in a straightforward way. As our treatment is based on an exact, analytic solution of the spin dependent transport problem, it allows us to determine for which parameters the device is reflectionless, i.e, perfect polarization at the outputs takes place without losses.
We consider a ring Aronov and Lyanda-Geller (1993) of radius $`a`$ in the $`xy`$ plane and assume a tunable static electric field in the $`z`$ direction controlling the strength of the spin-orbit interaction characterized by the parameter $`\alpha `$.Nitta et al. (1997) The Hamiltonian Meijer et al. (2002); Molnár et al. (2004) in the presence of spin-orbit interaction for a charged particle of effective mass $`m^{}`$ is given by
$$H=\mathrm{}\mathrm{\Omega }\left[\left(i\frac{}{\phi }+\frac{\omega }{2\mathrm{\Omega }}(\sigma _x\mathrm{cos}\phi +\sigma _y\mathrm{sin}\phi )\right)^2\frac{\omega ^2}{4\mathrm{\Omega }^2}\right],$$
(1)
where $`\phi `$ is the azimuthal angle of a point on the ring, $`\mathrm{}\mathrm{\Omega }=\mathrm{}^2/2m^{}a^2`$ is the dimensionless kinetic energy of the charged particle and $`\omega `$ =$`\alpha /\mathrm{}a`$ is the frequency associated with the SOI. According to Ref. \[Földi et al., 2005\], in the $`|`$, $`|`$ eigenbasis of the $`z`$ component of the spin, the eigenstates of $`H`$ read:
$$\psi (\kappa ,\phi )=e^{i\kappa \phi }\left(\genfrac{}{}{0pt}{}{e^{i\phi /2}u(\kappa )}{e^{i\phi /2}v(\kappa )}\right).$$
(2)
The corresponding energy eigenvalues are
$$E=\mathrm{}\mathrm{\Omega }\left[\kappa ^2\mu \kappa w+1/4\right],\mu =\pm 1,$$
(3)
with $`w=\sqrt{1+(\omega ^2/\mathrm{\Omega }^2)}`$. The spinors in (2) are simultaneous eigenvectors of $`H`$, of the $`z`$ component of the total angular momentum: $`K=L_z+`$ $`S_z`$, and of the spin operator pointing in the direction determined by the angles $`\theta `$ and $`\phi `$:
$$S_{\theta \phi }=S_x\mathrm{sin}\theta \mathrm{cos}\phi +S_y\mathrm{sin}\theta \mathrm{sin}\phi +S_z\mathrm{cos}\theta ,$$
(4)
where $`\theta `$ is given by the constant $`\mathrm{tan}\theta =\omega /\mathrm{\Omega }:`$
$$K\psi (\kappa ,\phi )=\kappa _j^\mu \psi (\kappa _j^\mu ,\phi ),S_{\theta \phi }\psi (\kappa _j^\mu ,\phi )=\frac{\mu }{2}\psi (\kappa _j^\mu ,\phi ).$$
(5)
From geometrical point of view, the second eigenvalue equation above means that the direction of the spinors (2) are either parallel or antiparallel with the conserved (position dependent) direction defined by $`S_{\theta \phi }`$. Therefore, the expectation value of the vector $`\stackrel{}{S}`$ in these states rotates around the $`z`$ direction making always an angle $`\theta `$ with it, while $`\phi `$ is the actual azimuth along the ring.
In a closed ring $`\kappa \pm 1/2`$ should be integer, but the presence of the leads connected to the ring lifts this restriction: the energy is a continuous variable, and then the possible values of $`\kappa `$ are the solutions of Eq. (3):
$$\kappa _j^\mu =\mu (w/2+(1)^jq),j=1,2,\mu =\pm 1,$$
(6)
where $`q=\sqrt{(\omega /2\mathrm{\Omega })^2+E/\mathrm{}\mathrm{\Omega }}`$. The energy eigenvalues are fourfold degenerate, they can be classified Földi et al. (2005) by the quantum numbers $`\kappa `$ and $`\mu `$. The ratio of the components of the eigenvectors (2) is determined by $`v(\kappa _j^\mu )/u(\kappa _j^\mu )=(\mathrm{tan}\theta /2)_\mu =\mathrm{\Omega }/\omega \left(1\mu w\right).`$
The stationary states of the complete problem including the ring as well as the leads, can be obtained by fitting the solutions corresponding to the different domains. Using local coordinates as shown in Fig. 1, the incoming wave, $`\mathrm{\Psi }_3(x_3)`$, and the outgoing waves $`\mathrm{\Psi }_1(x_1),`$ $`\mathrm{\Psi }_2(x_2)`$ are built up as linear combinations of spinors with spatial dependence $`e^{ikx}`$ etc. corresponding to $`E=\mathrm{}^2k^2/2m^{}`$:
$`\mathrm{\Psi }_3\left(x_3\right)`$ $`=`$ $`\left(\begin{array}{c}f_{}\\ f_{}\end{array}\right)e^{ikx_3}+\left(\begin{array}{c}r_{}\\ r_{}\end{array}\right)e^{ikx_3},`$ (7a)
$`\mathrm{\Psi }_n\left(x_n\right)`$ $`=`$ $`\left(\begin{array}{c}t_{}^n\\ t_{}^n\end{array}\right)e^{ikx_n},`$ (7b)
where $`n=1,2.`$
The wave functions belonging to the same energy $`E`$ in all the three sections of the ring can be written as linear combinations of four eigenspinors:
$$\mathrm{\Psi }_i(\phi _i)=\underset{\begin{array}{c}j=1,2\\ \mu =\pm 1\end{array}}{}a_{ij\mu }\psi (\kappa _j^\mu ,\phi _i),$$
(8)
with $`i=I,II,III`$ identifying the sections. This superposition is no more an eigenvector of $`S_{\theta \phi },`$ as it contains states with both $`\mu =\pm 1,`$ and their coefficients are different in general. Additionally, spatial interference of terms describing clockwise and anticlockwise motion plays an essential role in determining the spin direction in the ring. Therefore the position dependencies of the corresponding spin expectation values $`\mathrm{\Psi }_i|\stackrel{}{S}|\mathrm{\Psi }_i`$ in the arms are more complicated than the simple precession of the eigenvectors given in Eq. (2). Nevertheless, this change of the spin direction along the ring can be calculated without difficulty, if one determines the values of the coefficients $`a_{ij\mu },`$ which can be done using the boundary conditions, to be discussed now.
Figure 1 indicates the wave functions to be fitted at different junctions: e.g. the incoming wave at $`x_3=0`$ should be fitted to $`\mathrm{\Psi }_I`$ at $`\phi _I=0`$ and to $`\mathrm{\Psi }_{III}`$ at $`\phi _{III}=2\pi `$. We require the continuity of the wave functions, as well as a vanishing spin current density at the junctions.Griffith (1953); Xia (1992); Molnár et al. (2004); Földi et al. (2005) The procedure is similar to the case of a single outgoing lead which was described in detail in Refs. \[Molnár et al., 2004,Földi et al., 2005\]. The results can be summarized by the aid of two transmission matrices which acting on the incoming spinor valued input wave functions provide the output:
$`T^{\left(n\right)}\left(\begin{array}{c}f_{}\\ f_{}\end{array}\right)`$ $`=`$ $`\left(\begin{array}{c}t_{}^n\\ t_{}^n\end{array}\right),`$ (9)
with $`n=1,2.`$ When the incoming electron is not perfectly spin-polarized, its state should be described by a $`2\times 2`$ density matrix $`\rho _{in},`$ we can write:
$$\rho ^n=T^{\left(n\right)}\rho _{in}\left(T^{\left(n\right)}\right)^{},$$
(10)
where $`\rho ^1`$ and $`\rho ^2`$ are the output density matrices in the respective leads. The matrix elements of $`T^{\left(1\right)}`$ and $`T^{\left(2\right)}`$ can be calculated analytically for arbitrary geometry, but we found that the spin polarizing properties of this device are most clearly seen for the case when the outgoing leads are in a symmetric position, i.e., $`\gamma _1=2\pi \gamma _2`$. Here we will limit ourselves to this symmetric geometry, yielding
$`T_{}^{\left(1\right)}`$ $`=`$ $`{\displaystyle \frac{8qak}{y}}e^{i\frac{\gamma _2}{2}}\left[\mathrm{cos}^2{\displaystyle \frac{\theta }{2}}\left(h_1+h_2\right)+\mathrm{sin}^2{\displaystyle \frac{\theta }{2}}\left(h_1^{}h_2^{}\right)\right],`$
$`T_{}^{\left(1\right)}`$ $`=`$ $`{\displaystyle \frac{8qak}{y}}e^{i\frac{\gamma _2}{2}}\mathrm{sin}{\displaystyle \frac{\theta }{2}}\mathrm{cos}{\displaystyle \frac{\theta }{2}}\left[\left(h_1+h_2\right)\left(h_1^{}h_2^{}\right)\right],`$
$`T_{}^{\left(1\right)}`$ $`=`$ $`{\displaystyle \frac{8qak}{y}}e^{i\frac{\gamma _2}{2}}\left[\mathrm{sin}^2{\displaystyle \frac{\theta }{2}}\left(h_1+h_2\right)+\mathrm{cos}^2{\displaystyle \frac{\theta }{2}}\left(h_1^{}h_2^{}\right)\right],`$
$`T_{}^{\left(1\right)}`$ $`=`$ $`e^{i\gamma _2}T_{}^{\left(1\right)},`$
where
$`h_1`$ $`=`$ $`ake^{i\frac{w}{2}\gamma _2}e^{iw\pi }\mathrm{sin}\left(q\left(2\pi \gamma _2\right)\right)\mathrm{sin}\left(2q\left(\pi \gamma _2\right)\right),`$
$`h_2`$ $`=`$ $`iqe^{i\frac{w}{2}\gamma _2}\left[e^{iw\pi }\mathrm{sin}\left(q\gamma _2\right)\mathrm{sin}\left(q\left(2\pi \gamma _2\right)\right)\right],`$
$`y`$ $`=`$ $`ia^3k^3[\mathrm{sin}\left(2q(3\pi 2\gamma _2)\right)2\mathrm{sin}\left(2q(\pi \gamma _2)\right)`$
$`\mathrm{sin}\left(2q\pi \right)]2qa^2k^2[\mathrm{cos}\left(2q(3\pi 2\gamma _2)\right)`$
$`+2\mathrm{cos}\left(2q(\pi \gamma _2)\right)]+6qa^2k^2\mathrm{cos}(2q\pi )`$
$`12iq^2ak\mathrm{sin}\left(2q\pi \right)+8q^3\left[\mathrm{cos}\left(w\pi \right)+\mathrm{cos}\left(2q\pi \right)\right].`$
Similarly, for the second output we obtain $`T_{}^{\left(2\right)}=T_{}^{\left(1\right)},`$ $`T_{}^{\left(2\right)}=T_{}^{\left(1\right)},`$ $`T_{}^{\left(2\right)}=T_{}^{\left(1\right)}`$ and $`T_{}^{\left(2\right)}=T_{}^{\left(1\right)}.`$ This symmetry is related to the chosen geometry $`\gamma _1=2\pi \gamma _2.`$ We note that the reflection matrix can also be calculated using the method described above, it turns out to be diagonal in the $`\{|`$, $`|\}`$ basis. We concentrate here on the transmission properties of the ring and consider reflection as a loss in the efficiency of spin transformation.
The most surprising physical consequence of our three terminal ring is its ability to deliver polarized output beams of electrons. Considering a completely unpolarized input, i.e., $`\rho _{in}`$ being proportional to the identity, the outputs will be generally partially polarized that could be detected by Faraday rotation experiments.Kato et al. (2005) However, we found that properly chosen parameters lead to output polarizations as high as $`100\%`$. The relevant output density operators in this case should be projectors (apart from the possible reflective losses):
$$\frac{1}{2}T^{\left(n\right)}\left(T^{\left(n\right)}\right)^{}=\eta _n|\varphi ^n\varphi ^n|.$$
(12)
The non-negative numbers $`\eta _1`$ and $`\eta _2`$ measure the efficiency of the polarizing device, i.e., $`\eta _1+\eta _2=1`$ means a reflectionless process. Direct calculation shows that, provided Eq. (12) is satisfied, the norms of the two outputs are equal, $`\eta _1=\eta _2\eta /2.`$ Eq. (12) is equivalent to requiring the determinants of $`T^{\left(n\right)}\left(T^{\left(n\right)}\right)^{}`$ to vanish. We found that these determinants are equal, and zero if $`h_1\pm h_2=0`$. Using Eqs. (Quantum rings as electron spin beam splitters), these conditions can be formulated as
$`\mathrm{cos}\left(w\pi \right)`$ $`=`$ $`{\displaystyle \frac{\mathrm{sin}\left(q\gamma _2\right)}{\mathrm{sin}\left(q\left(2\pi \gamma _2\right)\right)}},`$ (13a)
$`\mathrm{sin}\left(w\pi \right)`$ $`=`$ $`{\displaystyle \frac{ak}{q}}\mathrm{sin}\left(2q\left(\pi \gamma _2\right)\right),`$ (13b)
each of them lead to a $`k\omega `$ relation as depicted in Fig. 2 for a representative example corresponding to $`\gamma _2=3\pi /2`$. The crossing points of the gray (solution of Eq. (13a)) and black (solution of Eq. (13b)) curves in Fig. 2 are the parameters that can be used in an experimental realization of our proposal to achieve perfectly polarized outputs.
Similar figures can be drawn for arbitrary (symmetric) geometry. This implies that there are lines in three dimensional $`\{\gamma _2,\omega /\mathrm{\Omega },ka\}`$ space along which the ring polarizes a completely unpolarized input.
Now we can ask what the transmission probabilities are, *provided* perfect polarization occurs. Fig. 3 shows that along a line defined by $`h_1+h_2=0`$, $`\eta `$ is a quasiperiodic function of $`\gamma _2`$. A similar figure can be drawn for the condition $`h_1h_2=0.`$ As we can see, there are certain points (that is, parameter combinations), where the transmission probability is unity. This shows that it is possible to obtain $`100\%`$ spin polarized outputs from a perfectly unpolarized input, even without reflective losses.
Now we turn to the investigation of the outgoing spinors which arise as a consequence of the polarizing property of the ring. Clearly, these are the eigenstates $`|\varphi ^n`$ of the transmitted density matrices corresponding to the nonzero eigenvalues which are given by $`\eta _1=\eta _2=128q^2a^2k^2\left|h_1\right|^2/\left|y\right|^2.`$ Note that the quasiperiodic behavior of the transmission probability $`\eta =\eta _1+\eta _2`$ seen in Fig. 3 is related to the sine and cosine functions in $`h_1`$ and $`y`$. Focusing on the case of $`h_1+h_2=0`$, the eigenstates of the respective transmitted density matrices corresponding to the nonzero eigenvalues $`\eta _1`$ and $`\eta _2`$ read
$$|\varphi ^1_+=\left(\begin{array}{c}\mathrm{sin}\frac{\theta }{2}\\ e^{i\gamma _2}\mathrm{cos}\frac{\theta }{2}\end{array}\right),|\varphi ^2_+=\left(\begin{array}{c}e^{i\gamma _2}\mathrm{cos}\frac{\theta }{2}\\ \mathrm{sin}\frac{\theta }{2}\end{array}\right).$$
(14)
These results describe the connection between the strength of the spin-orbit coupling (encoded in $`\theta `$), the geometry of the device and its polarizing directions. We stress that this pair of spinors exhibits nontrivial spatial-spin correlation being a signature of quantum non-contextuality.Hasegawa et al. (2003) However, note that they are in general not orthogonal, their overlap is given by $`{}_{+}{}^{}\varphi ^2|\varphi ^1_{+}^{}=i\mathrm{sin}\theta \mathrm{sin}\gamma _2.`$ On the distinguishability of nonorthogonal states, see Ref. \[Bergou et al., 2003\]. Similarly, for $`h_1h_2=0`$, we have:
$$|\varphi ^1_{}=\left(\begin{array}{c}e^{i\gamma _2}\mathrm{cos}\frac{\theta }{2}\\ \mathrm{sin}\frac{\theta }{2}\end{array}\right),|\varphi ^2_{}=\left(\begin{array}{c}\mathrm{sin}\frac{\theta }{2}\\ e^{i\gamma _2}\mathrm{cos}\frac{\theta }{2}\end{array}\right).$$
(15)
Considering the transmission matrices themselves, it is clear that under the conditions given by Eqs. (13a -b), their determinants also vanish. That is, each $`T^{(n)}`$ has a zero eigenvalue, but – due to the nonhermiticity – its eigenspinors are not orthogonal. It can be verified that the eigenstates corresponding to the nonzero eigenvalue coincide with $`|\varphi ^n_+`$ and $`|\varphi ^n_{}`$, while the spinors annulled by the transmission matrices $`T^{(n)}|\varphi _0^n=0`$ have the following components:
$$|\varphi _0^1_+=\left(\begin{array}{c}\mathrm{cos}\frac{\theta }{2}\\ \mathrm{sin}\frac{\theta }{2}\end{array}\right),|\varphi _0^2_+=\left(\begin{array}{c}\mathrm{sin}\frac{\theta }{2}\\ \mathrm{cos}\frac{\theta }{2}\end{array}\right)$$
(16)
if $`h_1+h_2=0`$, and $`|\varphi _0^1_{}=|\varphi _0^2_+,`$ $`|\varphi _0^2_{}=|\varphi _0^1_+.`$
This shows that if the conditions given by Eqs. (13a -b) are satisfied, the device acts similar to a Stern-Gerlach apparatus in the sense that: 1) for unpolarized input, we have two different spin directions (14) in the outputs, 2) if we consider one of the eigenstates (14) as the input, its spin direction will not change in the appropriate output, and 3) there are spinors given by Eq. (16), for which the transmission probability into a given output lead is zero. However, the analogy is not perfect, the polarized spinors (14) are not orthogonal and the spinor which has zero probability to be transmitted through a given lead is not equal to the eigenstate corresponding to the nonzero eigenvalue of the other lead: $`|\varphi ^n|\varphi _0^n^{}`$ for $`nn^{}.`$ From this point of view, an optical polarizing beam splitter Loudon (2000); Asbóth et al. (2004) with nonorthogonal polarizing directions can be the closest analogue.
The present calculation was done for an idealized model system; in fact, our intention was showing that the discussed polarizing effect – in contrast to previous proposals – can be described in terms of pure Quantum Mechanics (i.e., spin precession and interference), thus it is of importance from a fundamental point of view, as well. On the other hand, there are results showing that the approximations of our model (transport is ballistic and one dimensional, i.e., the finite width of the ring-wire was not taken into account) can give valid description of actual physical systems under specific experimental situations. Currently, high mobility samples have become available such that at cryogenic temperatures transport is found to be ballistic over tens of microns. Similarly, phase coherence and spin coherence lengths Kikkawa and Awschalom (1998) have been found up to 100 $`\mu m`$. Our narrow ring implies the assumption of single mode propagation. Recently, it was found that the finite width of the rings has a small effect on the loss of coherence of the spin state; it has also been shown that in a multi-channel system the modulation of the transmitted spin states survive and under specific conditions the individual eigenchannel transmissions are very similar to the ones found in single channel rings.Frustaglia and Richter (2004); Souma and Nikolić (2004) A possible non-ideal coupling to the leads can be described through effective tunnel barriers. But in most of the current experimental systems the leads are connected in a rather adiabatic way which makes the coupling very close to ideal.
In conclusion, we showed that a quantum ring with one input and two output leads in the presence of Rashba-type SOI has remarkable similarities with a Stern-Gerlach apparatus. Parameter values, within the experimentally feasible range Nitta et al. (1997, 1999); Sato et al. (2001) were identified when the three terminal ring delivers perfectly polarized output beams of electrons without reflective losses. We found that appropriate spin polarized input states are transmitted without modification, but it is also possible to prepare inputs, for which the transmission into a given lead is forbidden. Thus our paper describes a realistic model in which spin sensitive quantum interference gives rise to fundamental polarization effects as well as to nontrivial spatial-spin correlations.
We note that similar rings can act as spintronic quantum gates Földi et al. (2005) or in the presence of an external magnetic field can be used also for spin filtering.Molnár et al. (2004) This points to the possibility to integrate spintronic beam splitters, gates and filters that can serve as elementary building blocks of a quantum network based on spin sensitive devices.Euges et al. (2003); Stepanenko et al. (2003); Yau et al. (2003); Frustaglia et al. (2001)
Acknowledgments: This work was supported by the Flemish-Hungarian Bilateral Programme, the Flemish Science Foundation (FWO-Vl), the Belgian Science Policy and the Hungarian Scientific Research Fund (OTKA) under Contracts Nos. T48888, D46043, M36803, M045596. |
warning/0507/hep-th0507088.html | ar5iv | text | # On the stability of the primordial closed string gas
## 1 Introduction
Treating a gas of free closed superstrings when all the spatial dimensions are closed and the system is kept in a thermal bath leads us to the well known conclusion that the Helmholtz free energy diverges as one approaches the Hagedorn temperature from the low temperature regime. This way, as the energy $`U`$ also diverges, one can conclude that, in a fixed temperature description, the Hagedorn temperature is a maximum one for the system. Let us recast in the next section what the details for the description in the macrocanonical ensemble with $`\mu =0`$ are. In particular, we will remind to the reader that energy fluctuations give us a measure about the very possibility of a fixed temperature description of the system. In section 3 we will compare our results with the thermodynamics that stems from the generalized ensemble description of an analogous extensive system. After all this, we will have then got some results to get into the treatment of the system at fixed energy in section 4. There, in a first subsection, we will remember and critically recast the well known computation by Brandenberger and Vafa . Next, we will present another computation that will reinforce the conclusion that the specific heat, as a function of energy, is divergent. A final subsection will be devoted to explain whether volume dependent corrections can change the picture. Finally, section 5 will present a few comments and serve as a reminder of the main results.
## 2 The macrocanonical description of closed strings at finite size
In a macrocanonical description at null chemical potential, the grand canonical partition function can be equated with the canonical partition function at a given number of strings $`N^{}`$. This number maximizes the canonical partition function $`Z(\beta ,R,N)`$ (i.e., $`\left[Z/N\right]_{N=N^{}}=0`$). When the fluctuations in the number of strings are small, this maximum coincides with the averaged number of strings, $`\overline{N}`$, as computed in the grand canonical ensemble. To be more concrete: $`\mathrm{\Theta }\left(\beta ,R,\mu =0\right)=_{N=0}^{\mathrm{}}Z(\beta ,R,N)Z(\beta ,R,N^{}(\beta ,R))`$ and $`Z`$ finally results a function of $`\beta `$ and $`R`$ only. This $`Z`$ is what we call the partition function for the system of free strings. Minus the logarithm of the partition function divided by $`\beta `$ is what we call the free energy of the system and it is only a function of $`R`$ and $`\beta `$ and not of the number of strings (see ). This will exactly correspond to the thermodynamical free energy whenever a thermodynamical limit can be defined. The computation for the gas of free superstrings gives
$$\begin{array}{c}\beta F\left(\beta \right)=\beta \frac{2^{\mathrm{\hspace{0.17em}6}}}{\pi \sqrt{\alpha ^{}}}_0^+\mathrm{}d\tau _2\tau _2^{3/2}\theta _2(0,\frac{\mathrm{i}\beta ^{\mathrm{\hspace{0.17em}2}}}{\pi ^{\mathrm{\hspace{0.17em}2}}\tau _2\alpha ^{}})_{1/2}^{1/2}d\tau _1\left|\theta _4(0,2\tau )\right|^{16}\hfill \\ \hfill \times \underset{\stackrel{}{m},\stackrel{}{n}}{}\mathrm{e}^{\pi \tau _2\left(\frac{R^{\mathrm{\hspace{0.17em}2}}}{\alpha ^{}}\stackrel{}{m}^{\mathrm{\hspace{0.17em}2}}+\frac{\alpha ^{}\stackrel{}{n}^{\mathrm{\hspace{0.17em}2}}}{R^{\mathrm{\hspace{0.17em}2}}}\right)+2\pi \mathrm{i}\tau _1\stackrel{}{m}\stackrel{}{n}}\end{array}$$
(1)
This is the free energy in the so called $`S`$-representation. It is the result one gets when computing the Helmholtz free energy in the light-cone gauge or by summing up over the field content of the string (analog model) (see ). In the conclusions we will comment more on this point and its relation to the $`F`$-representation, that coincides with the computation of the free energy as a vacuum energy for the Euclidean theory with Euclidean time of length $`\beta `$ including winding modes around it.
Now, let us introduce an ultraviolet dimensionless cutoff $`ϵ`$ in $`\tau _{\mathrm{\hspace{0.17em}2}}`$. This automatically produces the splitting of the free energy as $`F\left(\beta \right)=F^l\left(\beta \right)+F^h\left(\beta \right)`$ where $`F^l`$ is got<sup>1</sup><sup>1</sup>1$`F^l`$, as it is written in this proper time representation, shows a divergence when $`\tau _{\mathrm{\hspace{0.17em}2}}+\mathrm{}`$ for the massless excitations of the string when momentum and winding numbers vanish. This is an artifact of this representation that results from the second quantization of the vacuum state then producing a divergence as $`\mathrm{ln}\tau _{\mathrm{\hspace{0.17em}2}}`$ when $`\tau _{\mathrm{\hspace{0.17em}2}}`$ goes to infinity. This divergence should not be present with finite volume or, more precisely, when the momenta are not dense, and can be subtracted without affecting the results of our work. by integrating $`\tau _{\mathrm{\hspace{0.17em}2}}`$ from $`ϵ`$ to $`+\mathrm{}`$ and $`F^h`$ by integrating the same integrand over $`\tau _2`$ from $`0`$ to $`ϵ`$.
The integral over $`\tau _1`$ simply represents the left-right level matching condition, i.e., a Kronecker delta of the form $`\delta _{N\stackrel{~}{N}+\stackrel{}{m}\stackrel{}{n},\mathrm{\hspace{0.17em}0}}`$. Here, $`\stackrel{~}{N}`$ and $`N`$ stands for the right and left oscillator numbers.
Supposing $`ϵ1`$ has various implications. The Jacobi $`\theta _2`$ function, having the $`\beta `$ dependence, can then be approximated by the leading terms (two in fact) in the series expansion representing it. This is physically equivalent to taking the classical (Maxwell-Boltzmann) statistics approximation. One can also use the fact that $`\tau _2ϵ`$ to approximate with arbitrary precision the contribution from integrating over $`\tau _1`$, i.e., implementing the left-right level matching condition.
To compute first the integral over $`\tau _1`$, it is very useful to note that, when $`\tau _2`$ is small, the main contribution to the integrand will come from a neighborhood of $`\tau _1=0`$.
Furthermore, one can choose $`ϵ`$ small enough so as to have that $`ϵ\alpha ^{}/R^{\mathrm{\hspace{0.17em}2}}1`$ and, simultaneously, $`ϵR^{\mathrm{\hspace{0.17em}2}}/\alpha ^{}1`$. It is enough to choose $`ϵ`$ to fulfill the most stringent criterion. In fact, one criterion converts into the other by T-duality ($`R\alpha ^{}/R`$). If $`R\sqrt{\alpha ^{}}`$, one can choose $`ϵ\alpha ^{}/R^{\mathrm{\hspace{0.17em}2}}`$ in order to accomplish both criteria. When $`R\sqrt{\alpha ^{}}`$ we have that $`ϵR^{\mathrm{\hspace{0.17em}2}}/\alpha ^{}`$ suffices to satisfy both relations. We will see how these criteria, that define the physical system we are treating and the kind of thermodynamical limit we are taking, can be expressed in terms of energy density. This allows us to approximate the sum over $`\stackrel{}{m}`$ and $`\stackrel{}{n}`$ (i.e., over winding and momentum numbers) by a multiple integral over the whole $`^{\mathrm{\hspace{0.17em}18}}`$ on the real variables $`m_1,\mathrm{},m_9,n_1,\mathrm{},n_9`$. This is a valid approximation as given by the Euler-Maclaurin formula (see ). The multiple integral over windings and momenta shows us that, for the contribution coming from the ultraviolet degrees of freedom to $`F\left(\beta \right)`$ (i.e., the contribution encoded in $`F^h`$), no dependence on $`R`$, and then on the volume $`V=(2\pi R)^{\mathrm{\hspace{0.17em}9}}`$, will survive. This is so because the change of variables of unit Jacobian $`\stackrel{}{m}\stackrel{}{m}\sqrt{\alpha ^{}}/R`$, $`\stackrel{}{n}\stackrel{}{n}R/\sqrt{\alpha ^{}}`$ makes $`R`$ to disappear in the multiple integration. Everything happens as putting $`R=\sqrt{\alpha ^{}}`$ that simply shows the fact that a situation in which $`R`$ is in a neighborhood of $`\sqrt{\alpha ^{}}`$ also suffices to compute the sum by an integral. This will be linked to the cosmological situation in which all the spatial dimensions are of the order of the selfdual length and the system is, in some sense, small.
With all this together one can finally write
$$\begin{array}{cc}\hfill I(\tau _2)2^{\mathrm{\hspace{0.17em}8}}& _{1/2}^{+1/2}d\tau _1\left|\tau \right|^{\mathrm{\hspace{0.17em}8}}\left|\theta _2(0,1/(2\tau ))\right|^{16}_{^9\times ^9}d\stackrel{}{l}d\stackrel{}{k}\mathrm{e}^{\pi \tau _2(\stackrel{}{l}^2+\stackrel{}{k}^2)}\mathrm{e}^{2\pi \stackrel{}{l}\stackrel{}{k}\tau _1}\hfill \\ & =2^{17/2}\tau _2^{\mathrm{\hspace{0.17em}1}/2}\mathrm{e}^{\mathrm{\hspace{0.17em}2}\pi /\tau _2}\underset{i=0}{\overset{+\mathrm{}}{}}a_i\tau _2^i\hfill \end{array}$$
(2)
Where the modular properties of the transverse partition function have been used. We remark again, that the main contribution in the $`\tau _20`$ limit for the $`\tau _1`$ integral comes from the neighborhood of $`\tau _1=0`$. The coefficients $`a_i`$ are computable numbers (see the Appendix). In particular, $`a_0=1`$
$`F^h\left(\beta \right)`$ can then be approximated, for $`\mathrm{\hspace{0.17em}2}\pi \left(\beta ^{\mathrm{\hspace{0.17em}2}}\beta _H^{\mathrm{\hspace{0.17em}2}}\right)ϵ\beta _H^{\mathrm{\hspace{0.17em}2}}`$ , as<sup>2</sup><sup>2</sup>2The well known relation $`\mathrm{\Gamma }[a+1,x]=a\mathrm{\Gamma }[a,x]+x^a\mathrm{e}^x`$ is very useful to this purpose.
$$F^h\left(\beta \right)\mathrm{\Gamma }(0,\frac{2\pi \left(\beta ^{\mathrm{\hspace{0.17em}2}}\beta _H^{\mathrm{\hspace{0.17em}2}}\right)}{ϵ\beta _H^{\mathrm{\hspace{0.17em}2}}})\underset{n=0}{\overset{+\mathrm{}}{}}b_n\left(\frac{2\pi \left(\beta ^{\mathrm{\hspace{0.17em}2}}\beta _H^{\mathrm{\hspace{0.17em}2}}\right)}{\beta _H^{\mathrm{\hspace{0.17em}2}}}\right)^n$$
(3)
$`\beta _H=\pi \sqrt{8\alpha ^{}}`$ is the inverse Hagedorn temperature for closed superstrings type IIA and IIB (both have the same free energy because are indistinguishable at finite $`T`$) and $`\beta \beta _H>0`$. The $`b_n`$ are coefficients that can be directly connected to the $`a_i`$; for example, $`b_0=1/\beta _H`$, and, in general, the $`b_n1/\beta _H`$ are independent of $`ϵ`$ computable numbers. The important point now is that, as discussed in the Appendix,
$$\underset{n=0}{\overset{+\mathrm{}}{}}b_n\left(\frac{2\pi \left(\beta ^{\mathrm{\hspace{0.17em}2}}\beta _H^{\mathrm{\hspace{0.17em}2}}\right)}{\beta _H^{\mathrm{\hspace{0.17em}2}}}\right)^n=\underset{n=0}{\overset{+\mathrm{}}{}}(1)^n\frac{\left(\beta \beta _H\right)^n}{\beta _H^{n+1}}=\frac{1}{\beta }$$
(4)
As it is implied by the behavior of $`\mathrm{\Gamma }[0,z]`$ for big $`z`$, $`F^h`$ goes exponentially to zero as $`\beta \beta _H`$ grows. Because there is no dependence on the volume, it is not true that we can write $`F\left(\beta \right)=PV`$. This gives more importance and justifies the detailed way we have introduced $`F\left(\beta \right)`$ at the beginning of this section. It is now clear from the behavior of $`\mathrm{\Gamma }[0,z]`$ when $`z`$ goes to zero that the contribution of $`F^h`$ to the free energy when $`\beta \beta _h`$ is much bigger than that of $`F^l`$ because $`|F^h|`$ grows unbounded as $`\beta \beta _H^+`$ as long as $`F^l`$ gets the finite value $`F^l(\beta _H)`$.
The concrete behavior around $`\beta _H`$ can be made more explicit by using that $`\beta ^{\mathrm{\hspace{0.17em}2}}\beta _H^{\mathrm{\hspace{0.17em}2}}=2\beta _H(\beta \beta _H)+(\beta \beta _H)^{\mathrm{\hspace{0.17em}2}}2\beta _H(\beta \beta _H)`$. One finally gets
$$\begin{array}{cc}\hfill \beta F^h\left(\beta \right)& _0^+\mathrm{}dE\theta \left(E\frac{4\pi }{ϵ\beta _H}\right)\mathrm{e}^{\beta E}\frac{\mathrm{e}^{\beta _HE}}{E}\hfill \\ & =\mathrm{\Gamma }(0,\frac{4\pi \left(\beta \beta _H\right)}{ϵ\beta _H})\hfill \end{array}$$
(5)
This, by using inverse Laplace transformation, easily provides us the main ingredient to get the fixed energy description, $`\mathrm{\Omega }_1(E)`$, i.e., the single string density of states.
$$\mathrm{\Omega }_1^h\left(E\right)=\theta \left(E\mathrm{\Lambda }\right)\frac{\mathrm{e}^{\beta _HE}}{E}$$
(6)
The step function shows the utility of the dimensionless $`ϵ`$ parameter by imposing the condition $`E>\mathrm{\Lambda }=4\pi /(ϵ\beta _H)`$. This, when $`R\sqrt{\alpha ^{}}`$, finally enforces $`E4\pi R^{\mathrm{\hspace{0.17em}2}}/(\beta _H\alpha ^{})`$ as the condition for the validity of (6)<sup>3</sup><sup>3</sup>3It is important to remark that $`E`$ is here the energy of the states which are accessible by one string. This condition can be compared to the one in for the validity of converting the sum over winding and momenta into an integral in the direct calculation of $`\mathrm{\Gamma }_1`$ from its very definition ($`\mathrm{\Omega }_1=\mathrm{d}\mathrm{\Gamma }_1/\mathrm{d}E`$); the condition is $`ER/\alpha ^{}`$. It is clear that $`E4\pi R^{\mathrm{\hspace{0.17em}2}}/(\beta _H\alpha ^{})`$ implies $`ER/\alpha ^{}`$ if $`R\sqrt{\alpha ^{}}`$.. For the T-dual situation, one gets the dual condition.
It is now an immediate task to get $`U\left(\beta \right)`$ around $`\beta _H`$
$$U^h\left(\beta \right)=\frac{\left[\beta F^h\left(\beta \right)\right]}{\beta }=\frac{1}{\beta \beta _H}\mathrm{e}^{4\pi \left(\beta \beta _H\right)/\left(ϵ\beta _H\right)}\frac{1}{\beta \beta _H}$$
(7)
Where we have taken that $`0<\beta \beta _Hϵ\beta _H/(4\pi )=1/\mathrm{\Lambda }`$.
Fluctuations in the macrocanonical energy are very useful for our study. They can be computed to give, near $`\beta _H`$
$$\frac{\left[T^{\mathrm{\hspace{0.17em}2}}C_V^h(T,V)\right]^{1/2}}{U^h}=1+\mathrm{O}\left[\left(\beta \beta _H\right)^{\mathrm{\hspace{0.17em}1}}\right]$$
(8)
So energy relative fluctuations are finite and not negligible. This means that we may expect the fixed energy (”micro”) description and this macrocanonical picture not to be equivalent.
The entropy can easily be computed as a function of $`\beta `$ to give
$$S^h\left(\beta \right)=\beta ^{\mathrm{\hspace{0.17em}2}}\frac{F^h}{\beta }\frac{\beta _H}{\beta \beta _H}\mathrm{ln}\left[\mathrm{\Lambda }\left(\beta \beta _H\right)\right]$$
(9)
The fundamental relation in the entropic representation can now easily be obtained to be
$$S^h\beta _HU^h+\mathrm{ln}U^h$$
(10)
In the entropic representation it is manifest that the non extensive term $`\mathrm{ln}U^h`$ is the one responsible for the positivity of the specific heat. When energy is high, the logarithm of the averaged energy is very small as compared to energy itself. If one sees the big fluctuations as giving the error of the energy variable to produce $`S(\overline{E}=U)`$, one perhaps should neglect the logarithmic term because is smaller than the error.
The macrocanonical calculation gives us the number of strings as $`\beta F\left(\beta \right)`$ if Maxwell-Boltzmann statistics applies. Namely
$$\beta F^h\left(\beta \right)=\overline{N}^h\mathrm{ln}\left[\left(\beta \beta _H\right)\mathrm{\Lambda }\right]\mathrm{ln}U^h$$
(11)
One can write the entropy in terms of the number of strings as a function of energy to get
$$S^h\overline{N}(U^h)+\mathrm{e}^{\overline{N}(U^h)}$$
(12)
This expression for the entropy can be compared to the one for a regular system for which the entropy scales with the number of objects as in the black body, for instance. In our gas we have that the entropy near $`\beta _H`$ (that also gives high $`U`$) grows exponentially with the number of objects. This behavior in terms of the number of strings can also be compared to the open superstring case in which $`T_H`$ is a true maximum temperature . For it, the entropy grows as $`2\overline{N}(U^h)+K\overline{N}^{\mathrm{\hspace{0.17em}2}}(U^h)/V`$, with $`K`$ a constant, and is a degree one homogeneous function of energy and volume (extensivity is a property of the gas of open superstrings in the infinite volume limit).
It is then clear that the specific heat, as a function of the temperature, is positive for the high temperature phase, i.e., when $`T`$ is near $`T_H`$. The problem is that the order one energy fluctuations tell us that $`U^h`$ is a bad canonical average for the high energy of the non-isolated system.
## 3 The generalized ensemble and extensivity
The description at fixed temperature we have made is one very special. The reason is that our system does not depend on volume because this variable does not appear in the description of the system which is also at fixed temperature and null chemical potential.
A description through a generalized ensemble is one in which the system is characterized by intensive parameters. In a simple system they are pressure, temperature and chemical potential instead of volume, energy and the number of objects which are the corresponding extensive parameters. Since our string gas is one for which no volume dependence appears<sup>4</sup><sup>4</sup>4This is different from being a problem at zero pressure. In a system at zero pressure, the volume would be a function of temperature. In our case, there is no volume dependence at all and then the pressure vanishes., the ensemble we have named grand canonical is really a generalized ensemble.
This ensemble does not appear thoroughly treated in regular textbooks on Statistical Mechanics but can be found in . The main point to take into account when reading this textbook is that the treatment of the description using this ensemble depends on the fact that extensivity is assumed for the system. The first notable fact about this ensemble, if extensivity is assumed, is that fluctuations in volume, energy and the number of objects must be big. The reason is that, in this picture, the system is characterized by pressure, temperature and $`\mu `$ and then volume, total energy and the number of objects can get any value with equal probability. It is worth to remark that this must be so when extensivity holds.
Assuming extensivity in our problem would imply $`\beta F\left(\beta \right)=0`$ because Euler’s relation ($`UTS+PV\mu N=0`$) would hold and we have $`\mu =0`$ and no volume dependence. On the contrary, non extensivity allows a non vanishing free energy as computed previously, the relative energy fluctuations to be big (order one) and, simultaneously, the fluctuations in the number of strings to be small when energy is big enough. Indeed, the fluctuations in $`\overline{N}(T)`$ are small because, when Maxwell-Boltzmann statistics and the dilute gas approximation hold, they are given for any system by
$$\frac{\sqrt{\overline{\mathrm{\Delta }N^{\mathrm{\hspace{0.17em}2}}}}}{\overline{N}(\beta )}=\frac{\sqrt{\left(z_z\right)^{\mathrm{\hspace{0.17em}2}}\left(zq\right)}}{q}=\frac{1}{\sqrt{q}}=\frac{1}{\sqrt{\overline{N}\left(\beta \right)}}$$
(13)
where $`q=\beta F\left(\beta \right)`$ is the single object partition function, a function of $`T`$ and $`V`$ in general and only of $`T`$ in our problem. For the classical counting, $`q`$ gives the number of objects. The general application of this result is not in contradiction with the fact that, in the generalized ensemble when extensivity holds, the fluctuations in the number of objects are big. When the system is extensive all the subsystems in the generalized ensemble with different object numbers are equally probable and then, to get the partition function, the sum over the number of objects does not run up to infinity (see again).
On the other hand, the fact that for any extensive system at $`\mu =0`$ and with no volume dependence one has $`UTS=0`$ has immediate consequences. By putting $`T=U/S`$, $`USU/S=0`$ can be understood as a differential equation that can be solved to give $`S=\beta _0U`$ where $`\beta _0`$ is a constant of integration that gives the constant inverse temperature of the system. In our problem this $`\beta _0`$ is $`\beta _H`$. From the point of view of Legendre transformations, this is the most simple and extreme case for which the transformation cannot be defined because the system has an infinite specific heat. Furthermore, if we now associate a density of states to the obtained entropy we have $`\mathrm{\Omega }\left(E\right)=K\beta _0\mathrm{e}^{\beta _0E}`$, where $`K`$ is a positive dimensionless constant. Computing now back the partition function $`Z\left(\beta \right)`$ thorough Laplace transformation we obtain<sup>5</sup><sup>5</sup>5 We could have included a cutoff $`\varphi `$ to get $`Z\left(\beta \right)=K\beta _0\mathrm{e}^{\varphi \left(\beta \beta _H\right)}/\left(\beta \beta _0\right)`$. $`Z\left(\beta \right)=K\beta _0_0^+\mathrm{}dE\mathrm{e}^{\left(\beta \beta _0\right)E}=K\beta _0/\left(\beta \beta _0\right)`$. So we find the surprising result that the free energy is not zero, but actually $`\beta F\left(\beta \right)=\mathrm{ln}\left[\left(\beta \beta _0\right)/\left(K\beta _0\right)\right]`$. The free energy must be negative and then has physical meaning when $`\beta `$ is near and bigger than $`\beta _0`$ so, after all, there is a maximum temperature $`T_0`$ and there must also be an energy cutoff for the validity of $`\mathrm{\Omega }\left(E\right)`$. Finally, one exactly gets the singular dominant term in the incomplete Gamma function in (5) identifying $`T_0=T_H`$; i.e. the logarithmic contribution. This is a very simple example of non equivalence of ensembles so brightly explained in and that relies upon the interplay between Legendre and Laplace transformations. It is very important to notice that the single object partition function $`\beta F\left(\beta \right)`$ we get actually coincides with the one in (11).
Conversely, one can easily show that if the canonical energy is a function of $`T`$ such that, at $`T=T_0`$, $`\left(\beta \beta _0\right)U\left(\beta \right)`$ gives zero (or a constant) when $`\beta `$ approaches $`\beta _0`$ then, since $`S(\beta )=\beta U(\beta )`$ for a volume independent extensive system at $`\mu =0`$, one gets that $`S(\beta )=\beta _0U(\beta )`$ around $`\beta _0`$. But, in fact, we have already arrived at the fundamental thermodynamic relation $`S=\beta _0U`$ that implies that the temperature is constant and equal to $`T_0`$. This contradicts the hypothesis assuming that the internal energy is a function of a variable $`\beta `$. The final output is then that $`S=\beta _0U`$ only makes sense in microcanonical thermodynamics and, more than this, extensivity holds for the microcanonical thermodynamics we get.
It is certainly a notorious fact that, in this particular case, one can deduce the exponential growth of the density of states with energy from the hypotheses that extensivity holds, the system does not depend on volume and equilibrium is got at zero chemical potential. In fact, all the above reasoning simply tells us that $`\mathrm{\Omega }(E)`$, the density of states for the gas of strings, is a constant times $`\mathrm{e}^{\beta _HE}`$ when energy goes beyond a certain value, let’s call it $`\varphi `$. However, this contradicts the computation in in which a different non independent of the volume non extensive high energy entropy is found. This makes us suspect that there is something wrong in that computation.
On the other hand, the grand canonical description of the closed string gas near $`T_H`$ and under the condition that windings and momenta are equivalent is really a generalized ensemble description but without assuming extensivity. Non extensivity is usually related to a kind of smallness of the system in size or number of objects, the presence of long range interactions or a critical behavior. We will dwell a bit more on this point in the final section.
## 4 The fixed energy description of closed strings at finite size
In order to study the fixed energy description of the gas of closed strings, an expression for the multiple string density of states, $`\mathrm{\Omega }\left(E\right)`$, is needed. This has been done in the past in several ways. One could use, for example, a saddle point approximation to calculate the asymptotic expression for the density of states of the string gas, using the fact that $`\mathrm{\Omega }\left(E\right)=^1\left\{\mathrm{exp}\left(\beta F\left(\beta \right)\right)\right\}`$. However, once the single string density of states is provided, $`\mathrm{\Omega }\left(E\right)`$ can be obtained using the convolution theorem . This is the method used long time ago by R. Brandenberger and C. Vafa (see also ), who calculated<sup>6</sup><sup>6</sup>6Only for the Maxwell-Boltzmann statistics $`\mathrm{\Omega }_n`$ can be understood as the density of states for a gas with $`n`$ strings .
$`\mathrm{\Omega }_n\left(E\right)={\displaystyle \frac{1}{n!}}{\displaystyle _\mathrm{\Lambda }^E}{\displaystyle \underset{i=1}{\overset{n}{}}}\mathrm{d}E_i\mathrm{\Omega }_1(E_i)\delta \left({\displaystyle \underset{i}{}}E_iE\right).`$ (14)
$`\mathrm{\Omega }_1(E)`$ can be expressed as sum of two terms $`\mathrm{\Omega }_1^l(E)`$ and $`\mathrm{\Omega }_1^h(E)`$ where the superscripts $`l`$ and $`h`$ refer to the low and high energy expressions. This introduces a cutoff $`\mathrm{\Lambda }`$ separating both regimes that can exactly be written as our cutoff $`4\pi /(ϵ\beta _H)`$ on the energy of one string. In fact we can obtain $`\mathrm{\Omega }_N(E)`$ with a high degree of accuracy only from the convolutions of $`\mathrm{\Omega }_1^h(E)`$ as in the gas of open strings . The convolution between $`\mathrm{\Omega }_1^l(E)`$ and $`\mathrm{\Omega }_1^h(E)`$ to give a contribution to $`\mathrm{\Omega }_2^h(E)`$ is negligible, as can clearly be seen in Fig. 1 where the comparison between the exact calculation<sup>7</sup><sup>7</sup>7 This means that we have used an exact form for the coefficients giving the degeneration number at each mass level of the superstring and the sum over winding and momenta has been obtained by integration. of $`\omega _2(E,t)=\mathrm{\Omega }_1(Et)\mathrm{\Omega }_1(t)`$ and its approximation resulting from considering only the high energy part approximated by $`\mathrm{\Omega }_1^h`$ in (6) is shown at $`E=56`$, $`R=\sqrt{\alpha ^{^{}}}=1`$. $`\omega _2`$ is a measure of equipartition of energy between the two strings. In fact, it is Fig. 1 that lets us fix the cutoff $`\mathrm{\Lambda }=\frac{4\pi }{ϵ\beta _H}`$
### 4.1 The calculation of Brandenberger and Vafa revisited
Taking into account that<sup>8</sup><sup>8</sup>8For $`n=0`$ we have that $`\mathrm{\Omega }_0\left(E\right)`$ corresponds to the vacuum state, whose density of states is Dirac’s delta. $`\mathrm{\Omega }\left(E\right)=_{n=0}^{\mathrm{}}\mathrm{\Omega }_n\left(E\right)`$, and using (6), for the single string density of states when $`E>\mathrm{\Lambda }`$, Brandenberger and Vafa obtained
$$\mathrm{\Omega }^h\left(E\right)=\frac{\mathrm{e}^{\beta _HE}}{2\pi E}_{\mathrm{}}^{\mathrm{}}d\alpha \mathrm{e}^{\mathrm{i}\alpha }\mathrm{e}^{_{\frac{\mathrm{\Lambda }}{E}}^1\frac{\mathrm{d}x}{x}\mathrm{e}^{\mathrm{i}\alpha x}}=\frac{\mathrm{e}^{\beta _HE}}{\mathrm{\Lambda }}\left(a+b\frac{\mathrm{\Lambda }}{E}\right)$$
(15)
where:
$`a`$ $`=`$ $`{\displaystyle \frac{1}{2\pi }}{\displaystyle _{\mathrm{}}^{\mathrm{}}}d\alpha \mathrm{e}^{_0^\alpha dx\frac{\mathrm{cos}x1}{x}}\mathrm{cos}\left({\displaystyle _0^\alpha }dx{\displaystyle \frac{\mathrm{sin}x}{x}}\alpha \right)=0.56\pm 0.01`$
$`b`$ $`=`$ $`{\displaystyle \frac{1}{2\pi }}{\displaystyle _{\mathrm{}}^{\mathrm{}}}\alpha d\alpha \mathrm{e}^{_0^\alpha dx\frac{\mathrm{cos}x1}{x}}\mathrm{sin}\left({\displaystyle _0^\alpha }dx{\displaystyle \frac{\mathrm{sin}x}{x}}\alpha \right)=0.29\pm 0.01`$
The values of $`a`$ and $`b`$ were calculated numerically. Note that, although the exponent gives, when $`\alpha 1`$, a factor $`\mathrm{log}\alpha `$, the second integral is not convergent because of the oscillatory behavior of the integrand, and so it has to be regulated and they did it by using an exponentially decaying factor. The sign of $`b`$ produces a positive specific heat, and the Hagedorn temperature is then a maximum one in the microcanonical ensemble. We think, nevertheless, that this scenario needs a revision: there is nothing a priori unphysical in negative specific heats in the microcanonical ensemble, and anyhow, we are not sure that there could be a positive specific heat phase in the microcanonical ensemble for the gas of strings with windings and momenta.
In fact, it is easy to see how a contradiction appears, when $`b=0.29`$, in the following way: using (15) and the fact that $`Z\left(\beta \right)=\left\{\mathrm{\Omega }\left(E\right)\right\}`$ the following expression for the partition function can be obtained<sup>9</sup><sup>9</sup>9The multiple string density of states must have a cutoff that indicates the range of validity of the asymptotic approximation and that would be, in general, different from the single string cutoff. We are using $`\varphi `$ as the cutoff for the multiple string density of states.:
$$Z^h\left(\beta \right)=\frac{a\mathrm{e}^{\left(\beta \beta _H\right)\varphi }}{\mathrm{\Lambda }\left(\beta \beta _H\right)}+b\mathrm{\Gamma }[0,\varphi \left(\beta \beta _H\right)]$$
(16)
whereas from the single density of states it is possible to arrive also to an expression for the multiple string partition function via the equality $`\beta F\left(\beta \right)=\left\{\mathrm{\Omega }_1\left(E\right)\right\}`$
$$Z^h\left(\beta \right)=\mathrm{e}^{\beta F\left(\beta \right)}=\frac{\mathrm{e}^\gamma }{\mathrm{\Lambda }\left(\beta \beta _H\right)}+\mathrm{e}^\gamma +\mathrm{O}\left[\left(\beta \beta _H\right)^1\right]$$
(17)
Comparing (16) and (17), one immediately gets that<sup>10</sup><sup>10</sup>10This expression for $`a`$ was roughly deduced in , although no connection with the numerical value of was made there. $`a=\mathrm{exp}\left(\gamma \right)=0.5614`$, which perfectly agrees with the numerical value given in . But it is a very notorious fact that in (17) no term is found analogous to the logarithmically divergent one hidden in the incomplete gamma function of (16).
Furthermore, we have found an alternative calculation where the coefficient $`b`$ is actually zero and then, only one divergent term is present in both equations (16) and (17). As it is written in (6), the high energy dominant term in the single string density of states has a dimensionless factor which equals unity in the type II and the heterotic string (this factor could be volume dependent when considering open strings and branes). However, it is very useful to introduce in $`\mathrm{\Omega }_1\left(E\right)`$ a factor $`c`$ that could be thought just as a regulator (whereas we are going to give it a physical meaning at the end of this subsection). This change adds a factor $`c^n`$ to $`\mathrm{\Omega }_n`$ in (14). Only at the end the limit $`c`$ approaching to one will be taken. Furthermore, it is possible to work out the values of $`a`$ and $`b`$ analytically if a simple change is made in (14): the key ingredient is noting that the upper limit in the integrals can be taken to infinity since the Dirac delta function ensures that no value greater than $`E`$ will contribute to them. This will render the final integrals much easier to perform; then, we will have
$$_\mathrm{\Lambda }^EdE_i_\mathrm{\Lambda }^+\mathrm{}dE_i\mathrm{\Omega }^h\left(E\right)=\frac{\mathrm{e}^{\beta _HE}}{2\pi E}_{\mathrm{}}^+\mathrm{}d\alpha \mathrm{e}^{\mathrm{i}\alpha }\mathrm{e}^{c_{\frac{\mathrm{\Lambda }}{E}}^{\mathrm{}}\frac{\mathrm{d}x}{x}\mathrm{e}^{\mathrm{i}\alpha x}}$$
(18)
The calculation is now analogous to the one made in , taking the first terms in the series expansion in $`\mathrm{\Lambda }/E`$.
$`\mathrm{\Omega }^h\left(E\right)`$ $`=`$ $`{\displaystyle \frac{\mathrm{e}^{\beta _HE}}{2\pi E}}{\displaystyle _{\mathrm{}}^+\mathrm{}}d\alpha \mathrm{e}^{\mathrm{i}\alpha }\mathrm{e}^{c_{\frac{\alpha \mathrm{\Lambda }}{E}}^{\mathrm{}\mathrm{sg}\left(\alpha \right)}\frac{\mathrm{d}x}{x}\mathrm{cos}x}\mathrm{e}^{\mathrm{i}c_{\frac{\alpha \mathrm{\Lambda }}{E}}^{\mathrm{}\mathrm{sg}\left(\alpha \right)}\frac{\mathrm{d}x}{x}\mathrm{sin}x}`$ (19)
$`=`$ $`{\displaystyle \frac{\mathrm{e}^{\beta _HE}}{2\pi E}}{\displaystyle _{\mathrm{}}^+\mathrm{}}d\alpha \mathrm{e}^{c\mathrm{Ci}\left(\frac{\alpha \mathrm{\Lambda }}{E}\right)}\mathrm{cos}\left(\alpha +c\mathrm{si}\left({\displaystyle \frac{\alpha \mathrm{\Lambda }}{E}}\right)\right)`$
Where $`\mathrm{sg}\left(\alpha \right)`$ stands for the sign function. It is important to note that the $`\mathrm{Ci}`$ and the $`\mathrm{si}`$ functions only coincide with the standard cosine integral and sine integral functions for positive values of $`\alpha `$. $`\mathrm{Ci}\left(x\right)`$ is a real, even function of $`x`$ whereas $`\mathrm{si}\left(x\right)`$ is a real, odd function of its argument with a discontinuity at $`x=0`$ and $`\mathrm{si}\left(0\right)=0`$. This is easy to understand as a consequence of the integral upper limits including the $`\mathrm{sg}\left(\alpha \right)`$ term in the exponential factors in (19). The integrand is then even and one could perform a series expansion in terms of $`\alpha \mathrm{\Lambda }/E`$. Approximating $`\mathrm{Ci}\left(\frac{\alpha \mathrm{\Lambda }}{E}\right)=\gamma +\mathrm{log}\left(\frac{\alpha \mathrm{\Lambda }}{E}\right)+\mathrm{O}\left(\frac{\alpha \mathrm{\Lambda }}{E}\right)`$ and $`\mathrm{si}\left(\frac{\alpha \mathrm{\Lambda }}{E}\right)=\frac{\pi }{2}+\frac{\alpha \mathrm{\Lambda }}{E}+\mathrm{O}\left(\frac{\alpha \mathrm{\Lambda }}{E}\right)^2(\alpha >0)`$, we arrive at
$$\mathrm{\Omega }^h\left(E\right)=\mathrm{e}^\gamma \frac{\mathrm{e}^{\beta _HE}}{\pi E}\left(\frac{E}{\mathrm{\Lambda }}\right)^c_0^{\mathrm{}}\frac{\mathrm{d}\alpha }{\alpha ^c}\mathrm{cos}\left(\alpha c\frac{\pi }{2}+c\frac{\alpha \mathrm{\Lambda }}{E}\right)$$
(20)
Using that $`\mathrm{cos}\left(ab\right)=\mathrm{cos}a\mathrm{cos}b+\mathrm{sin}a\mathrm{sin}b`$ and taking the lowest order in $`\alpha \mathrm{\Lambda }/E`$ one finally gets
$$\mathrm{\Omega }^h\left(E\right)=\frac{E^{c1}}{\mathrm{\Lambda }^c}\mathrm{e}^{\beta _HE}\left[a(c)+b(c)\frac{\mathrm{\Lambda }}{E}+\mathrm{O}\left(\frac{\mathrm{\Lambda }}{E}\right)^2\right]$$
(21)
$`a\left(c\right)`$ $`=`$ $`{\displaystyle \frac{\mathrm{e}^{c\gamma }}{\pi }}{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{\mathrm{d}\alpha }{\alpha ^c}}\mathrm{cos}\left({\displaystyle \frac{c\pi }{2}}\alpha \right)={\displaystyle \frac{\mathrm{e}^{c\gamma }}{\mathrm{\Gamma }\left(c\right)}}\text{if }0<c<1`$
$`b\left(c\right)`$ $`=`$ $`{\displaystyle \frac{c\mathrm{e}^{c\gamma }}{\pi }}{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{\mathrm{d}\alpha }{\alpha ^{c1}}}\mathrm{sin}\left({\displaystyle \frac{c\pi }{2}}\alpha \right)={\displaystyle \frac{c\mathrm{e}^{c\gamma }}{\mathrm{\Gamma }\left(c1\right)}}\text{if }1<c<2`$
The integrals converge only for the indicated values of<sup>11</sup><sup>11</sup>11One could be tempted to put directly, in the expression for $`a\left(c\right)`$, that, when $`c=1`$, $`\mathrm{cos}\left(\frac{\pi }{2}\alpha \right)=\mathrm{sin}\alpha `$; but this would produce a wrong result since we would be forgetting that $`\mathrm{si}\left(0\right)=0`$. $`c`$. Clearly, in the $`c1`$ limit, the result for $`a`$ is fully compatible with the numerical value given in ; the problem is that it is very easy to see how in this limit $`b`$ goes to zero. As a matter of fact this method can be generalized and more terms of the form $`d_n\left(\mathrm{\Lambda }/E\right)^n(n)`$ can be computed, giving all of them zero for $`d_n`$ in the $`c1`$ limit.
Now, it is straightforward to see how the contradiction between the equations analogous to (16) and (17), that now depend on $`c`$, has disappeared. Using $`\mathrm{\Omega }_1\left(E\right)=c\mathrm{e}^{\beta _HE}/E`$ we have that
$$Z^h\left(\beta \right)=\mathrm{e}^{c\beta F^h\left(\beta \right)}\frac{\mathrm{e}^{c\gamma }}{\mathrm{\Lambda }^c\left(\beta \beta _H\right)^c}+\frac{c\mathrm{e}^{c\gamma }}{\mathrm{\Lambda }^{c1}\left(\beta \beta _H\right)^{c1}}+\mathrm{O}\left(\left(\beta \beta _H\right)^{2c}\right).$$
(22)
And making directly the Laplace transform of (21), the same expression for $`Z\left(\beta \right)=\left\{\mathrm{\Omega }\left(E\right)\right\}`$ can be found for $`c>1`$. When $`c=1`$ an annoying constant term appears preventing us to fix more than the only divergent term.
Once the value $`c=1`$ is taken, the expression for the density of states of the string gas in a finite size container is given by
$$\mathrm{\Omega }^h\left(E\right)=\mathrm{e}^\gamma \frac{\mathrm{e}^{\beta _HE}}{\mathrm{\Lambda }}E\mathrm{\Lambda }.$$
(23)
From this we can calculate both the entropy of the system and its temperature. The fundamental thermodynamic relationship giving the entropy as a function of the energy now looks like
$$S^h\left(E\right)=\beta _HE+\mathrm{log}\left(\frac{\mathrm{e}^\gamma }{\mathrm{\Lambda }}\right).$$
(24)
Comparing it with (10), the analogous expression in the fixed temperature, case we see how both ensembles seem to be inequivalent as pointed in (8). With this entropy we will also have that temperature is fixed to Hagedorn’s, and we would have to conclude that $`C_V\left(E\right)`$ would be infinite.
The constant $`c`$ has been introduced as a mere way of doing analytical continuation of ill-defined expressions, but we can give it a physical interpretation. Lets look at the expression defining $`\mathrm{\Omega }\left(E\right)`$ once $`c`$ is introduced
$$\mathrm{\Omega }\left(E\right)=\underset{n=0}{\overset{\mathrm{}}{}}c^n\mathrm{\Omega }_n\left(E\right)$$
(25)
We see that $`c`$ is really acting as the fugacity of the system, so that working with a generic value of $`c`$ and then performing the $`c1`$ limit is exactly the same as working with a generic non null chemical potential and then taking it to zero. This interpretation also lets us know that we have not been working in the microcanonical ensemble, but in the ”enthalpic” one for which energy and the chemical potential are given. This way $`\mathrm{\Omega }\left(E\right)`$ now depends on $`c`$ and becomes $`\mathrm{\Omega }(E,c)`$.
### 4.2 Fluctuations for the fixed energy description
It is now clear that the density of states of the string gas is given by
$$\mathrm{\Omega }^h\left(E\right)=\frac{\mathrm{e}^\gamma }{\mathrm{\Lambda }}\theta \left(E\varphi \right)\mathrm{e}^{\beta _HE}$$
(26)
The multi-string energy cutoff $`\varphi `$ cannot be completely determined without matching the high energy regime with the low energy phase because, imposing that $`\beta F^h\left(\beta \right)`$ must be obtained, $`\varphi `$ would appear as a factor of $`\left(\beta \beta _H\right)^{\mathrm{\hspace{0.17em}0}}`$ that is a regular term being $`F^l\left(\beta \right)`$ also regular at $`\beta _H`$. The matching can be done, but we are not going to dwell further on this point.
An immediate consequence of (24) is that the high energy microcanonical specific heat, $`C_V^h\left(E\right)`$, is divergent contrary to what, by using , has been assumed as true for more than fifteen years<sup>12</sup><sup>12</sup>12In the next subsection and the conclusions, we will treat and discuss the relevance that the radius corrections can have in relation to this thermodynamical statement..
Once $`\mathrm{\Omega }(E,c)`$ for the enthalpic ensemble is obtained, it is straightforward to compute the number of strings and its fluctuations as
$$\begin{array}{cc}\hfill \overline{N}\left(E\right)& =c_c\mathrm{log}\mathrm{\Omega }(E,c)|_{c=1}\mathrm{log}\left(\frac{E}{\mathrm{\Lambda }}\right)\hfill \\ \hfill \frac{\sqrt{\overline{\mathrm{\Delta }N^2}\left(E\right)}}{\overline{N}\left(E\right)}& =\frac{\sqrt{\left(c_c\right)^2\mathrm{log}\mathrm{\Omega }(E,c)|_{c=1}}}{\overline{N}\left(E\right)}\frac{1}{\sqrt{\overline{N}\left(E\right)}}\hfill \end{array}$$
with
$$\begin{array}{cc}\hfill \mathrm{\Omega }^h(E,c)=\theta \left(E\varphi \right)\frac{\mathrm{e}^{c\gamma }}{\mathrm{\Lambda }^c\mathrm{\Gamma }\left(c\right)}& E^{c1}\mathrm{e}^{\beta _HE}\hfill \\ & \left[1+\left(c1\right)\mathrm{O}\left(\frac{\mathrm{\Lambda }}{E}\right)+\mathrm{}\right]\hfill \end{array}$$
(27)
that has been obtained by Laplace inversion of $`\mathrm{e}^{cq\left(\beta \right)}`$.
### 4.3 Other refinements
In the preceding sections any dependence on $`R`$ has been lost as a result of being in a physical situation in which sums can be well approximated by integrals. Now, we would like to add the effects of introducing Euler-Maclaurin corrections in the integrals that represent sums over windings and momenta. This can also be done by means of a Poisson resummation. As a result, there appear more singular points in $`\beta F\left(\beta \right)`$ whose location depends on $`R`$ .
$$\beta F^h\left(\beta \right)=\underset{\stackrel{}{m},\stackrel{}{n},j}{}\mathrm{\Gamma }[0,\frac{2\pi \left(\beta ^{\mathrm{\hspace{0.17em}2}}\beta _{\stackrel{}{m},\stackrel{}{n},j}\left(R\right)^{\mathrm{\hspace{0.17em}2}}\right)}{ϵ\beta _H^{\mathrm{\hspace{0.17em}2}}}]$$
(28)
with
$$\beta _{\stackrel{}{m},\stackrel{}{n},j}^{\mathrm{\hspace{0.17em}2}}=\frac{\alpha ^{}\pi ^{\mathrm{\hspace{0.17em}2}}}{\left(j+1/2\right)^{\mathrm{\hspace{0.17em}2}}}\left[2\frac{R^{\mathrm{\hspace{0.17em}2}}}{\alpha ^{}}\stackrel{}{m}^{\mathrm{\hspace{0.17em}2}}\frac{\alpha ^{}}{R^{\mathrm{\hspace{0.17em}2}}}\stackrel{}{n}^{\mathrm{\hspace{0.17em}2}}\right]$$
(29)
Assuming $`R>\sqrt{\alpha ^{}}`$ the singularity nearest to $`\beta _H`$ is $`\beta _1=\beta _H\eta `$, $`\eta \alpha ^{}\beta _H/(4R^{\mathrm{\hspace{0.17em}2}})`$. Considering also the term depending on $`\beta _1`$ we get a corrected expression for the density of states
$$\mathrm{\Omega }^h(E,R)=\frac{\mathrm{e}^\gamma }{\mathrm{\Lambda }}\theta \left(E\varphi \right)\mathrm{e}^{\beta _HE}\left(1\frac{\mathrm{e}^{\eta E}\left(\eta E\right)^{17}}{\mathrm{\Gamma }\left(18\right)}\right)\left(\frac{\mathrm{e}^\gamma }{2\eta \mathrm{\Lambda }}\right)^{\mathrm{\hspace{0.17em}18}}$$
(30)
Where $`\eta E1`$ has been assumed. This expression depends on $`R`$ through $`\eta `$ and the $`R`$ corrections are actually non extensive. If taken into account, the effect of these corrections on $`\beta `$ would be
$$\beta \left(E\right)=\beta _H+\eta \frac{\mathrm{e}^{\eta E}}{\mathrm{\Gamma }\left(18\right)}\left(\eta E\right)^{17}$$
(31)
The second term on the right hand side of this expression has already been studied in the literature , and has been frequently claimed as being the cause of having a positive specific heat. It is easy to see that, after all, having enough energy for a given radius, the gas can always be in the regime in which there is no dependence on the volume and this is the regime of the Brandenberger-Vafa scenario, for which the specific heat diverges. The radius corrections are exponentially suppressed and, for our system, are not different from the finite volume corrections that are dropped when the thermodynamic limit is taken over the ideal gas of particles (see, for example, ).
## 5 Conclusions
As a first important result, it is crucial to remark that no published calculation has found the $`0.29\mathrm{e}^{\beta _HE}/E`$ term but the one presented by Brandenberger and Vafa in . This is the term that is needed to state, as these authors do, that the microcanonical specific heat is positive in the physical situation in which energy density is so high that the density of states does not depend on the volume. As far as we know, what any other calculation actually gets is that, for the same physical situation, the specific heat is divergent because a null $`b`$ coefficient is found (see eq. (15)). In reference , it is explicitly admitted that $`b=0.29`$ is not found, but the authors do not face the important question that the contradiction between their calculation and that of Brandenberger and Vafa rises. It seems that this is so because they consider they are using what they call a different ”formalism”. In our work, we clearly show that, using the same technique Brandenberger and Vafa used and a physically meaningful regularization<sup>13</sup><sup>13</sup>13Brandenberger and Vafa stay that they use an exponential regulator for a numerical computation., the $`b`$ coefficient vanishes (and also any other high energy correction). This is the first result of our work and, in our opinion, critically depends on understanding that the ”micro” ensemble is really the ”enthalpic” one, namely, a fixed energy and fixed chemical potential ensemble.
We have used the $`S`$-representation of the Helmholtz free energy to finally conclude that the behavior of the free energy around $`\beta _H`$ coincides with what is gotten from the $`F`$ representation. This might be expected but it is not a trivial fact<sup>14</sup><sup>14</sup>14In the previous versions of our work we thought on the contrary because we found only the leading order contribution in (2). In fact, a seemed very concerned referee, after reading the first version of our work, was fully convinced of the difference between the $`F`$ and $`S`$ representations for this calculation and used it to reject the work without any hesitation because he/she liked more the $`F`$-representation., because both representations do not coincide. That the $`S`$ and $`F`$ representations are not equal seems to be clearly commented for the first time in . The relationship between both representations is carefully studied for the family of the heterotic strings in (see also ). In those works it is clearly established that the $`F`$-representation does not provide an analytical function for complex $`\beta `$. In other words, it is false that the $`F`$-representation can be seen as providing the analytical continuation of the $`S`$-representation for heterotic strings. Then, it is not rigorous to say that the $`F`$-representation of the free energy can be used to get the density of states by inverse Laplace transformation of the corresponding partition function. What one can only do is to continue the $`S`$-representation. When one uses the $`F`$-representation to get the behavior around $`\beta _H`$ one is really using it in the interval $`(\beta _H,+\mathrm{})`$ where it coincides with the $`S`$-representation. Now, we are also providing a very concrete and explicit example of to what extent the $`S`$ and $`F`$ representations of the free energy are equal.
From other point of view, taking into account that the $`F`$-representation gives the free energy as computed for the compactification of the Euclidean time on a circle of length $`\beta `$ (including string windings along it), in there appears a proof for the noncritical $`c=1`$ string of the recently emphasized fact that, for strings, the free energy cannot be computed, at any temperature, by the compactification of time .
Another important point is to what extent the $`R`$ corrections presented in subsection (4.3) can be taken as showing that the system really has a positive specific heat. This seems to be the belief as expressed in and . In our opinion, the problem is that these nonextensive corrections are exponentially suppressed with the value of the energy and are then of no thermodynamical relevance for the system in the thermodynamical limit in which the energy density is very high and the sum over windings and momenta can be replaced by an integral. Those corrections are as the finite volume $`1/\sqrt{V}`$ corrections for the gas of free particles; they are irrelevant in the thermodynamical limit in which the sum over momenta can be replaced by an integral.
In the case we treat the gas in the fixed temperature (canonical) description, the big fluctuations might justify the exclusion of the nonextensive term that renders the canonical specific heat positive. In any case, those fluctuations would just make the canonical equilibrium description physically unusable.
From a cosmological point of view, one could think that there would not be any relevant cosmological implication from our results because, after all, the equation of state would still be the same, corresponding to pressureless dust matter ($`P=0`$). However, things are more intricate because a divergent microcanonical specific heat can be an indication of a phase transition. In our case, what we have done in the microcanonical (really enthalpic) treatment is a description of how the volume of phase space in the N-body problem changes when energy, which is a conserved quantity, increases . We have found that the system behaves very differently from the grand canonical ensemble in which $`T_H`$ would be a maximum temperature and the specific heat, as a function of temperature, would be positive.
What is clear is that a divergent microcanonical specific heat cannot be used ab initio as a criterion to drop as unphysical our gas of closed strings at finite size.
## Acknowledgments
The work of M. A. Cobas is partially supported by a Spanish MEC-FPI fellowship. M. Suárez is partially supported by a Spanish MEC-FPU fellowship. We all are partially supported by the Spanish MEC project BFM2003-00313.
## A. The UV limit in the S-representation partition function
This appendix is devoted to explain how (2) is obtained and to explicitly show that (4) holds.
First of all, the integral computing the sum over windings and momenta can be performed to give
$$_{^9\times ^9}d\stackrel{}{l}d\stackrel{}{k}\mathrm{e}^{\pi \tau _2(\stackrel{}{l}^{\mathrm{\hspace{0.17em}2}}+\stackrel{}{k}^{\mathrm{\hspace{0.17em}2}})}\mathrm{e}^{2\pi \mathrm{i}\stackrel{}{l}\stackrel{}{k}\tau _1}=\left|\tau \right|^9$$
(A.1)
Next, it has to be noticed that, when $`\tau 0`$, it holds that
$$\left|\theta _2(0,1/(2\tau ))\right|^{16}2^{16}\mathrm{e}^{2\pi \tau _2/\left|\tau \right|^{\mathrm{\hspace{0.17em}2}}}$$
because all the other terms are finite when $`\tau 0`$. One has then to perform the integral over $`\tau _1`$ as providing a function of $`\tau _2`$ given by $`\mathrm{e}^{\mathrm{\hspace{0.17em}2}\pi /\tau _2}`$ times a series expansion in powers of $`\tau _2`$ as it appears on the right hand side of (2). Namely, the left hand side of (2) (we called it $`I(\tau _2)`$) is now given by
$$\begin{array}{c}\hfill I(\tau _2)=\frac{1}{2^{\mathrm{\hspace{0.17em}8}}\tau _2}\mathrm{e}^{2\pi /\tau _2}_{1/2}^{+1/2}d\tau _1\left(1+\frac{\tau _1^{\mathrm{\hspace{0.17em}2}}}{\tau _2^{\mathrm{\hspace{0.17em}2}}}\right)^{1/2}\mathrm{e}^{2\pi \tau _1^{\mathrm{\hspace{0.17em}2}}/\tau _2^{\mathrm{\hspace{0.17em}3}}}\mathrm{e}^{\frac{2\pi }{\tau _2\left[1+\tau _1^{\mathrm{\hspace{0.17em}2}}/\tau _2^{\mathrm{\hspace{0.17em}2}}\right]}\frac{2\pi }{\tau _2}+\frac{2\pi \tau _1^{\mathrm{\hspace{0.17em}2}}}{\tau _2^{\mathrm{\hspace{0.17em}3}}}}\end{array}$$
(A.2)
It has been written in a way prepared to be rewritten in terms of a function $`\stackrel{~}{I}(\tau _2)`$ as $`I(\tau _2)=2^8\tau _2^{\mathrm{\hspace{0.17em}1}/2}\mathrm{e}^{\mathrm{\hspace{0.17em}2}\pi /\tau _2}\stackrel{~}{I}(\tau _2)`$ where
$$\stackrel{~}{I}(\tau _2)=_{\left(2\tau _2^{3/2}\right)^1}^{\left(2\tau _2^{3/2}\right)^1}dx\mathrm{e}^{2\pi x^{\mathrm{\hspace{0.17em}2}}}\mathrm{e}^{\frac{2\pi }{\tau _2}\left[\frac{1}{1+x^{\mathrm{\hspace{0.17em}2}}\tau _2}\left(1\tau _2x^{\mathrm{\hspace{0.17em}2}}\right)\right]}\left(1+\tau _2x^{\mathrm{\hspace{0.17em}2}}\right)^{1/2}$$
(A.3)
For which the change of variables $`\tau _1=x\tau _2^{3/2}`$ has been used. Now, for the product of the last two factors in the integrand, the following series expansion can be written
$$\begin{array}{cc}\hfill \mathrm{e}^{\frac{2\pi }{\tau _2}\left[\frac{1}{1+x^{\mathrm{\hspace{0.17em}2}}\tau _2}\left(1\tau _2x^{\mathrm{\hspace{0.17em}2}}\right)\right]}& \left(1+\tau _2x^{\mathrm{\hspace{0.17em}2}}\right)^{1/2}=\mathrm{e}^{2\pi x^{\mathrm{\hspace{0.17em}4}}\tau _2/(1+\tau _2x^{\mathrm{\hspace{0.17em}2}})}\left(1+\tau _2x^{\mathrm{\hspace{0.17em}2}}\right)^{1/2}\hfill \\ & =\underset{b=0}{\overset{+\mathrm{}}{}}\underset{a=0}{\overset{+\mathrm{}}{}}\frac{(1)^a(2\pi )^b\mathrm{\Gamma }\left(b+a+1/2\right)}{b!a!\mathrm{\Gamma }\left(b+1/2\right)}\tau _2^{b+a}x^{\mathrm{\hspace{0.17em}4}b+2a}\hfill \end{array}$$
(A.4)
Next we are able to perform the integral over $`x`$ of the term $`x^{4b+2a}`$ obtaining
$$\begin{array}{cc}\hfill _{\left(2\tau _2^{3/2}\right)^1}^{\left(2\tau _2^{3/2}\right)^1}dx\mathrm{e}^{2\pi x^{\mathrm{\hspace{0.17em}2}}}x^{4b+2a}& =\frac{\mathrm{\Gamma }\left(2b+a+1/2\right)\mathrm{\Gamma }(2b+a+1/2,2\pi /(4\tau _2^{\mathrm{\hspace{0.17em}3}}))}{(2\pi )^{2b+a+1/2}}\hfill \\ & (2\pi )^{2ba1/2}\mathrm{\Gamma }\left(2b+a+1/2\right)\hfill \end{array}$$
(A.5)
where the last approximation results from the exponential suppression of the contribution coming from the incomplete Gamma function when its argument gets big (what happens here when $`\tau _20`$).
Now one of the sums in the resulting double sum to get $`\stackrel{~}{I}`$ can be calculated (!!) to finally give
$$\stackrel{~}{I}(\tau _2)=\underset{b=0}{\overset{+\mathrm{}}{}}(2\pi )^{b1/2}\mathrm{\Gamma }\left(b+1/2\right)\tau _2^b$$
(A.6)
We are then able to get $`I(\tau _2)`$ and, in particular, the coefficients $`a_i`$ as
$$a_i=\frac{\mathrm{\Gamma }\left(i+1/2\right)}{(2\pi )^i\sqrt{\pi }}$$
(A.7)
The next step is to use $`I(\tau _2)`$ as written in (2) with the already known $`a_i`$ coefficients to compute $`F^h(\beta )`$ finding then the $`b_n`$ factors in (3). This is easily done to give
$$b_n=(1)^n\frac{a_n}{\beta _Hn!}$$
(A.8)
The final computation is that of the series generated by the $`b_n`$ coefficients as it appears on the left hand side of (4), namely
$$R(\beta )\underset{n=0}{\overset{+\mathrm{}}{}}b_n\left(\beta ^{\mathrm{\hspace{0.17em}2}}\beta _H^{\mathrm{\hspace{0.17em}2}}\right)^n\left(\frac{2\pi }{\beta _H^{\mathrm{\hspace{0.17em}2}}}\right)^n$$
(A.9)
Taking into account that $`\beta ^2\beta _H^{\mathrm{\hspace{0.17em}2}}=2\beta _H\left(\beta \beta _H\right)\left(1+\frac{\beta \beta _H}{2\beta _H}\right)`$, $`R(\beta )`$ can be written as a double sum
$$R(\beta )=\underset{n=0}{\overset{+\mathrm{}}{}}\left[\frac{1}{\sqrt{\pi }}\underset{q=0}{\overset{n}{}}\frac{(1)^q\mathrm{\hspace{0.17em}2}^{\mathrm{\hspace{0.17em}2}qn}\mathrm{\Gamma }\left(q+1/2\right)}{\left(nq\right)!\left(2qn\right)!}\right]\frac{\left(\beta \beta _H\right)^n}{\beta _H^{n+1}}$$
(A.10)
The term between square brackets can be computed to give exactly $`(1)^n`$. So, we have finally showed that
$$R(\beta )=1/\beta $$
(A.11)
by showing that it is given by the power series expansion of $`1/\beta `$ around $`\beta _H`$. |
warning/0507/math0507574.html | ar5iv | text | # Properties of Fixed Point Sets and a Characterization of the Ball in ℂ^𝑛
## 0. Introduction
Let $`D^n`$ be a bounded domain. Below we consider two families of self-maps of $`D`$. The first is the group $`Aut(D)`$ of holomorphic automorphisms of $`D`$; the second is the set $`H(D,D)`$ of all holomorphic maps from $`D`$ to $`D`$, i.e., the set of endomorphisms of $`D`$.
###### Definition 0.1.
A set $`KD`$ is called a determining subset of $`D`$ with respect to $`Aut(D)`$ (or $`H(D,D)`$ resp.) if, whenever $`g`$ is automorphism (resp. endomorphism) of $`D`$ such that $`g(k)=k`$ $`kK`$, then $`g`$ is the identity map of $`D`$.
The notion of a determining set was introduced earlier in a paper we wrote with our collaborators Steven G. Krantz and Kang-Tae Kim \[FK1\]. In that paper we attempted to find a higher dimensional analog of the following result of classical function theory (\[FF\],\[ES\],\[Mas\],\[PL\],\[Su\]): if $`f:MM`$ is a conformal self-mapping of a plane domain $`M`$ which fixes three distinct points then $`f(\zeta )=\zeta `$.
Determining sets have been further investigated in the following papers \[FK2\], \[KK\], \[Vi1\], \[Vi2\].
Let $`W_s(D)`$ denote the set of $`s`$-tuples $`(x_1,\mathrm{},x_s)`$, where $`x_jD`$, such that $`\{x_1,\mathrm{},x_s\}`$ is a determining set with respect to $`Aut(D)`$. Similarly, $`\widehat{W}_s(D)`$ denotes the set of $`s`$-tuples $`(x_1,\mathrm{},x_s)`$ such that $`\{x_1,\mathrm{},x_s\}`$ is a determining set with respect to $`H(D,D)`$. So $`\widehat{W}_s(D)W_s(D)D^s`$. We now introduce two numbers $`s_0(D)`$ and $`\widehat{s}_0(D)`$. In case $`Aut(D)=id`$, $`s_0(D)=0`$, otherwise $`s_0(D)`$ is the least integer $`s`$, such that $`W_s(D)\mathrm{}`$. The symbol $`\widehat{s}_0(D)`$ denotes the least integer $`s`$ such that $`\widehat{W}_s(D)\mathrm{}`$. Hence, $`s_0(D)\widehat{s}_0(D)`$.
In \[FK1\] we proved the inequality $`s_0(D)n+1`$ for many (but not all) bounded domains in $`^n`$. In \[Vi1\] J.-P. Vigué, using a different method proved this estimate for all bounded domains in $`^n`$ . Furthermore in \[Vi2\] Vigué proved the estimate $`\widehat{s}_0(D)n+1`$ for all bounded domains in $`^n`$.
Both estimates are the best possible, since for the unit ball $`B^n^n`$, $`s_0(B^n)=n+1`$. In section 1 we prove that the reverse is true: if $`s_0(D)=n+1`$ for a bounded domain $`D^n`$, then $`D`$ is biholomorphic to the unit ball $`B^n`$. Obviously, $`s_0(D)`$ depends on how large $`Aut(D)`$ is: for a smaller group, we expect a lower $`s_0(D)`$. This relationship is reflected below in Corollary 1.7.
If a positive integer $`ss_0(D)`$, then $`W_s(D)\mathrm{}`$, so there are $`s`$ points such that if an automorphism of $`D`$ fixes these points it will fix any point of $`D`$. Now the question arises whether the choice of these $`s`$ points is generic. To make it more precise we need to find out if $`W_s(D)`$ is open and everywhere dense in $`D^s`$. We consider this question in section 2. Refining and complementing the results of \[FK1, Vi1, Vi2\] we prove that $`W_s(D)`$ is open, and also dense if not empty. For the similar question related to $`\widehat{W}_s(D)`$ we provide examples to the contrary.
## 1. Estimates for $`s_0(D)`$ and a characterization of the ball in $`^n`$
### 1.1. Characterization of the ball by determining sets.
This section is devoted to the proof of the following theorem.
###### Theorem 1.1.
Let $`D`$ be a bounded domain in $`^n`$. Then $`s_0(D)=n+1`$ if and only if $`D`$ is biholomorphic to the unit ball $`B^n`$ in $`^n`$.
To verify the estimate $`s_0(B^n)=n+1`$ we need to prove that no $`n`$ points in $`B^n`$ form a determining set for $`Aut(B^n)`$. (This was done in \[FK2\], we repeat it here for completeness). Consider $`n`$ arbitrary points $`(p_0,p_1,\mathrm{},p_{n1})`$, where $`p_iB^n`$ for $`i=0,\mathrm{},n1`$. Consider $`gAut(B^n)`$ such that $`g(p_0)=0`$. Consider now $`n1`$ vectors $`g(p_i)`$, and the complex linear space $`\pi `$ spanned by these vectors. Since $`\mathrm{dim}(\pi )n1`$, there is a rotation $`fAut(B^n)`$ that is not the identity and keeps all the points of $`\pi `$ fixed. Now the automorphism $`h=g^1fg`$ $`Aut(B^n)`$ is not the identity and it fixes all $`n`$ points $`(p_0,p_1,\mathrm{},p_{n1})`$. We proved that $`W_n(B^n)=\mathrm{}`$, so $`s_0(B^n)=n+1`$.
The rest of this section will be devoted to the proof that $`s_0(D)=n+1`$ implies that $`D`$ is biholomorphic to the unit ball.
If $`H`$ is (isomorphic to) a subgroup of the unitary group $`U(n)`$, let $`k(H)`$ denote the least number $`k`$ of vectors $`u_1,\mathrm{},u_k`$ such that if $`hH`$ and if $`h(u_j)=u_j`$ for $`j=1,\mathrm{},k`$ then $`h=id`$. For $`zD`$ the isotropy group $`Aut_z(D)`$ is isomorphic to the group of its differentials at $`z`$, and these differentials are unitary with respect to the Bergman inner product on the tangent space $`T_z(D)`$. So $`Aut_z(D)`$ is isomorphic to a subgroup of $`U(n)`$.
###### Lemma 1.2.
For a bounded domain $`D`$ in $`^n`$, $`s_0(D)1+\mathrm{min}\{k(Aut_x(D)):xD\}`$.
###### Proof.
Choose $`zD`$ so that $`k(Aut_z(D))=\mathrm{min}\{k(Aut_x(D)):xD\}`$. Denote that number by $`k`$. Let $`u_1,\mathrm{},u_k`$ be vectors in $`T_zD`$ such that if $`hAut_z(D)`$ and if $`dh(z)(u_j)=u_j`$ for $`j=1,\mathrm{},k`$ then $`dh=id`$ (hence $`h=id`$). For each $`u_j`$, let $`z_j`$ be a point on the geodesic through $`z`$ in the direction $`u_j`$, so close to $`z`$ that the geodesic is the unique length minimizing geodesic from $`z`$ to $`z_j`$. Let $`f`$ be an automorphism of $`D`$ fixing $`z,z_1,\mathrm{},z_k`$. Then $`df(z)`$ fixes $`u_1,\mathrm{},u_k`$. It follows that $`df(z)=id`$ and $`f=id`$. Therefore, $`s_0(D)1+\mathrm{min}\{k(Aut_x(D)):xD\}`$. ∎
###### Lemma 1.3.
If $`H`$ is a subgroup of $`U(2)`$ and if $`H`$ is not transitive on $`S^3`$, then $`k(H)1`$.
###### Proof.
Let $`S=S^3`$ be the unit sphere in $`^2`$. It suffices to show that the set of fixed points in $`S`$ of nontrivial elements of $`H`$ (that is, each of these points is a fixed point of at least one nontrivial element of $`H`$) is not equal to $`S`$. For $`g,hU(2)`$ and $`xS`$, $`x`$ is a fixed point of $`h`$ iff $`g^1x`$ is a fixed point of $`g^1hg`$. So, without any loss of generality we can replace $`H`$ with a subgroup of $`U(2)`$ conjugate to $`H`$.
The Lie algebra $`Q`$ of $`U(2)`$ consists of skew Hermitian matrices, so it has as a basis the following elements:
$$a=\left(\begin{array}{cc}i& 0\\ 0& 0\end{array}\right),b=\left(\begin{array}{cc}0& 0\\ 0& i\end{array}\right),c=\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right),d=\left(\begin{array}{cc}0& i\\ i& 0\end{array}\right).$$
Their Lie bracket relations are
$$[a,b]=0,[a,c]=d,[a,d]=c,[b,c]=d,[b,d]=c,[c,d]=2a+2b.$$
If $`\mathrm{dim}H=4`$, then $`H=U(2)`$, which contradicts the hypothesis that $`H`$ is not transitive on $`S`$.
Suppose that $`\mathrm{dim}H=3`$. One can verify that the only 3-dimensional Lie subalgebra of $`Q`$ is spanned by $`\{c,d,ab\}`$. Hence, the identity component $`H_0`$ of $`H`$ is $`SU(2)`$, again contradicting the hypothesis that $`H`$ is not transitive on $`S`$.
Now suppose that $`\mathrm{dim}H=2`$. One can verify directly that $`U(2)`$ does not have a subgroup of dimension 2 and rank 1. Thus, $`H`$ has rank 2. Up to conjugation, the identity component $`H_0`$ of $`H`$ is $`T^2=\{diag(e^{i\alpha },e^{i\beta }):\alpha ,\beta \}`$. Each component of $`H`$ is $`gT^2`$. If $`hgT^2`$ has a nonzero fixed vector, $`h`$ must satisfy $`det(hid)=0`$. It follows that if $`\mathrm{dim}H2`$ then the set $`U`$ of the nontrivial elements of $`H`$ that have a fixed point on $`S`$ has dimension $`1`$, and for each $`gU`$, the set of fixed points of $`g`$ on $`S`$ has dimension 1. Thus the set $`P`$ of fixed points of nontrivial elements of $`H`$ has dimension $`2`$. It follows that $`PS`$. Therefore, $`k(H)1`$.
###### Lemma 1.4.
If $`H`$ is a subgroup of $`U(n)`$ with $`n2`$ and if $`H`$ is not transitive on $`S^{2n1}`$ then $`k(H)n1`$.
###### Proof.
The case where $`n=2`$ is the previous lemma. Suppose $`n>2`$ and $`HU(n)`$ is not transitive on $`S^{2n1}`$. Choose $`x,yS^{2n1}`$ so that no element of $`H`$ maps $`x`$ to $`y`$. Choose $`zS^{2n1}`$ orthogonal to both $`x`$ and $`y`$. Let $`S_1=\{vS^{2n1}:(v,z)=0\}`$, where $`(v,z)=v_j\overline{z}_j`$, and let $`H_1=\{gH:g(z)=z\}`$, and $`U_1=\{gU(n):g(z)=z\}`$. Now $`U_1U(n1)`$, and $`H_1`$ is a subgroup of $`U_1`$. By the induction hypothesis, since $`H_1`$ is not transitive on $`S_1`$, $`H_1`$ has a determining set of $`n2`$ vectors $`\{w_1,\mathrm{},w_{n2}\}`$. It follows that $`\{z,w_1,\mathrm{},w_{n2}\}`$ is a determining set for $`H`$. Therefore, $`k(H)n1`$. ∎
The proof of our main theorem follows from the following
###### Lemma 1.5.
If $`D`$ is a bounded domain in $`^n`$, and $`D\cong ̸B^n`$, then $`s_0(D)n`$.
###### Proof.
If $`n=1`$ the statement is obviously true. Assume $`n2`$. Let $`zD`$. Since $`D\cong ̸B^n`$, $`Aut_z(D)`$ is not transitive on the directions at $`z`$, by the main result of \[GK\]. By the Lemma 1.4, $`k(Aut_z(D))n1`$. It follows that $`s_0(D)1+k(Aut_z(D))n`$. ∎
Remark. For endomorphisms $`H(D,D)`$ we still have for the ball $`\widehat{s}_0(B^n)=n+1`$. However in the case of endomorphisms there are domains not biholomorphic to the ball but with the same maximum possible value of $`\widehat{s}_0`$. Here are two examples.
For $`n=1`$, $`\widehat{s}_0(D)=\widehat{s}_0(B^1)=2`$ for any bounded domain $`D`$.
We will show now that for $`n=2`$, for the unit polydisc $`\mathrm{\Delta }^2`$, $`\widehat{s}_0(\mathrm{\Delta }^2)=\widehat{s}_0(B^2)=3`$.
Indeed, consider any two distinct points $`p_1,p_2\mathrm{\Delta }^2`$. Since the $`Aut(\mathrm{\Delta }^2)`$ is transitive, we can find an automorphism $`g`$, such that $`g(p_1)=0`$. Let $`L`$ be the complex line through the origin and $`g(p_2)`$. In terms of the coordinates this line can always be described in one of the forms: $`z_2=\lambda z_1`$, or $`z_1=\lambda z_2`$ where $`|\lambda |1`$. One can check that the map $`P:(z_1,z_2)(z_1,\lambda z_1)`$ in the first case, or $`P:(z_1,z_2)(\lambda z_2,z_2)`$ in the second case will produce a holomorphic retraction of the polydisc, fixing $`g(p_1),g(p_2)`$. Now the map $`g^1Pg`$ is a holomorphic retraction of $`\mathrm{\Delta }^2`$fixing $`p_1,p_2`$. Therefore $`\widehat{s}_0(\mathrm{\Delta }^2)>2`$. Since this number is also $`3`$, we conclude $`\widehat{s}_0(\mathrm{\Delta }^2)=3`$.
### 1.2. An estimate for $`s_0(D)`$.
Let $`G`$ be a subgroup of $`Aut(D)`$. By $`s_0(D,G)`$ we denote the minimum number of distinct points in $`D`$ such that if $`gG`$, and $`g`$ fixes all these points, then $`g=id`$. So, $`s_0(D)=s_0(D,Aut(D)).`$
###### Theorem 1.6.
Let $`D`$ be a bounded domain in $`^n`$, let $`G`$ be a subgroup of $`Aut(D)`$, and let $`q=\mathrm{dim}G`$. If $`q1`$, then $`s_0(D,G)q`$. If $`q=0`$, then $`s_0(D,G)1`$.
###### Proof.
First we consider the case where $`q1`$. Let $`e`$ denote the identity element of $`G`$, and let $`Q=G\backslash \{e\}`$. For each $`gQ`$, the set $`\{xD:g(x)=x\}`$ is an analytic set of $`D`$ of dimension $`2n2`$. The set $`W_1:=\{(g,x)Q\times D:g(x)=x\}`$ is an analytic set of $`Q\times D`$ of dimension $`(2n2)+q2n1<dimD`$. Let $`W`$ denote the set of fixed points of nontrivial elements of $`G`$. Since $`W=\pi (W_1)`$, where $`\pi :Q\times DD`$ is the projection, and since $`dimW_1<dimD`$, we see that $`WD`$. Therefore, $`s_0(D,G)1`$.
Now we assume that $`q2`$. There must be an orbit $`Q`$ of $`G`$ of positive dimension. Let $`xQ`$, and let $`H:=G_x`$ be the subgroup of $`G`$ consisting of elements $`g`$ satisfying $`g(x)=x`$. Then $`\mathrm{dim}H<\mathrm{dim}G`$. By induction hypothesis, $`s_0(D,H)\mathrm{dim}G1`$. Therefore, $`s_0(D,G)1+s_0(D,H)\mathrm{dim}G`$. ∎
###### Corollary 1.7.
Let $`D`$ be a bounded domain in $`^n`$. If $`\mathrm{dim}(Aut(D))1`$, then $`s_0(D)\mathrm{dim}(Aut(D))`$. If $`\mathrm{dim}(Aut(D))=0`$, then $`s_0(D)1`$.
## 2. On topological properties of determining sets.
### 2.1. Determining sets $`W_s(D)`$ are open and dense
Our aim in this section is to prove the following theorem.
###### Theorem 2.1.
Let $`D`$ be a bounded domain in $`^n`$ and $`s1`$. Then $`W_s(D)D^s`$ is open; if in addition $`W_s(D)\mathrm{}`$, then $`W_s(D)`$ is dense in $`D^s`$.
The assertion that $`W_s(D)`$ is open and dense in $`D^s`$ was proved for some domains and $`sn+1`$ in \[FK1\]. Using analytic methods of \[Ca1\],\[Ca2\], J.P. Vigué (see \[Vi1\],\[Vi2\]) proved that $`W_s(D)`$ is open for all bounded domains and all $`s`$, and that it is dense for $`sn+1`$. By using the Bergman metric on a bounded domain we are able to use differential geometry methods and the Lie group properties (see \[GKM\], \[Kl\], \[MZ\], also \[BD\],\[FMP\], \[Ma\]) to prove the above general theorem.
First we introduce some notation. If $`G`$ is a subgroup of $`\mathrm{Aut}(D)`$, $`W_{s,G}(D)`$ denotes the set of $`s`$-tuples $`(x_1,\mathrm{},x_s)`$, where $`x_jD`$, such that each element $`gG`$ satisfying $`g(x_j)=x_j`$ for $`j=1,\mathrm{},s`$ has to be the identity.
We need the following lemma (Theorem 2.4 in \[Ma\]).
###### Lemma 2.2.
Let $`\mathrm{\Omega }`$ be a bounded domain in $`^n`$ containing the closure of the unit ball, and $`G`$ a compact Lie subgroup of $`Aut(\mathrm{\Omega })`$. Suppose that each $`G`$-orbit lies in a ball of radius $`1/2`$. Then $`G=\{id\}`$.
###### Lemma 2.3.
Suppose that $`D`$ is a bounded domain in $`^n`$ and $`G`$ is a subgroup of $`Aut(D)`$. Then $`W_{1,G}(D)`$ is open in $`D`$.
###### Proof.
We need to consider only the case $`W_{1,G}(D)\mathrm{}`$ . Suppose that $`xW_{1,G}(D)`$, and there is a sequence of points in $`D`$, $`x_kx`$ such that $`x_kW_{1,G}(D)`$ $`k`$. Let $`U`$ be a neighborhood of $`x`$, $`\overline{U}D`$, then there is a positive $`r`$ such that for large enough $`k`$ the ball with the center in $`x_k`$ and of radius $`r`$, $`b(x_k,r)`$ in the Bergman metric compactly lies in $`U`$. The assertion that $`x_kW_{1,G}(D)`$ means that the subgroup $`G_{x_k}`$ of $`G`$fixing $`x_k`$ is not the identity. This subgroup is a compact Lie subgroup of $`G`$, and also acts on $`b(x_k,r)`$ (since Bergman metric is an invariant metric). Applying (a properly adjusted form of) Lemma 2.2, one concludes that there exists an $`\epsilon >0`$, such that for large enough $`k`$ one has an automorphism $`g_kG_{x_k}`$ and a point $`y_kb(x_k,r)U`$ such that the Eucledian distance $`|g_k(y_k)y_k|>\epsilon `$. One can now find a subsequence $`\{k_j\}`$ such that (1) $`y_{k_j}y\overline{U}`$, and (2) $`g_{k_j}gG`$. We conclude now that $`g(x)=x`$ and that $`|g(y)y|\epsilon `$. This means that $`xW_{1,G}(D)`$ which is a contradiction. Therefore $`W_{1,G}(D)`$ is open in $`D`$. ∎
Let $`\rho (,)`$ denote the Bergman distance. Let $`b(z,r)`$ denote the Bergman ball with center $`z`$ and radius $`r`$. Let $`\overline{b}(z,r)`$ be the closure of $`b(z,r)`$ in $`D`$.
###### Lemma 2.4.
Suppose that $`D`$ is a bounded domain in $`^n`$ and $`G`$ is a subgroup of $`Aut(D)`$. If $`W_{1,G}(D)\mathrm{}`$ then $`W_{1,G}(D)`$ is dense in $`D`$.
###### Proof.
In this proof, let $`W=W_{1,G}(D)`$. Suppose that $`W`$ is not dense in $`D`$. Then the closure $`K`$ of $`W`$ in $`D`$ is not equal to $`D`$. Let $`p`$ be a boundary point of $`K`$ in $`D`$. Choose $`r>0`$ such that the closure of $`b(p,4r)`$ in $`D`$ is compact and such that each pair of points of $`b(p,4r)`$ is connected by a unique length-minimizing geodesic segment (in the Bergman metric). There exist points $`z,w`$ such that $`\rho (z,p)<r`$, $`\rho (w,p)<r`$, $`wW`$, and $`zK`$. Note that the orbit of $`w`$, $`G(w)W`$. Let $`Q=G(w)\overline{b}(p,4r)`$. Then $`Q`$ is compact and $`QW`$. Let $`u`$ be a point of $`Q`$ nearest to $`z`$. Then $`u`$ is also a point of $`G(w)`$ nearest to $`z`$, and $`R:=\rho (z,u)\rho (z,w)<2r`$. Choose a point $`y`$ on the unique length-minimizing geodesic segment from $`z`$ to $`u`$ such that $`yK`$ and $`yz`$. For each point $`x`$ of $`G(w)`$, we see that
$$\rho (z,y)+\rho (y,x)\rho (z,x)\rho (z,u),$$
and that the two equalities hold simultaneously only if $`x=u`$. Hence, $`\rho (z,y)+\rho (y,x)>\rho (z,u)=R`$ for each $`xG(w)`$, $`xu`$. It follows that $`\rho (y,x)>R\rho (z,y)=\rho (y,u)`$ for each $`xG(w)`$, $`xu`$. Therefore, $`u`$ is the unique point of $`G(w)`$ nearest to $`y`$. Since $`yK`$, there is a nontrivial $`gG`$ such that $`g(y)=y`$. Now $`\rho (y,u)=\rho (g(y),g(u))=\rho (y,g(u))`$ forces $`g(u)=u`$. Since $`uW`$, the map $`g`$ must be the identity, contradicting the fact that $`g`$ is not trivial. Therefore, $`W_{1,G}(D)`$ is dense in $`D`$. ∎
Proof of Theorem 2.1.
We need to prove this theorem only for $`W_s(D)\mathrm{}`$. For $`gAut(D)`$ let $`Q_s(g)`$ denote the mapping
$$Q_s(g):D^sD^s,Q_s(g)(z_1,\mathrm{},z_s)=(g(z_1),\mathrm{},g(z_s)).$$
Let $`G=\{Q_s(g):gAut(D)\}`$. Then $`GAut(D^s)`$, and $`W_{1,G}(D^s)=W_s(D)`$. By the previous lemmas, $`W_s(D)`$ is open and dense in $`D^s`$.
### 2.2. Determining sets $`\widehat{W}_s(D)`$ that are not open.
In \[Vi2\] it was proved that $`\widehat{W}_s(D)`$ is open in $`D^s`$ for any bounded taut domain in $`^n`$. Our aim in this section is to present an example in $`^2`$ of a bounded domain such that the determining set $`\widehat{W}_2(D)`$ is not open in $`D^2`$.
First we construct the set $`D^2`$.
Denote $`B_2=\{z^2||z|<2\},B_1=\{z^2||z|<1\},b_1=\{z=(z_1,z_2)^2||z_1|^2+|z_2+1|^2<(110^4)^2\},`$ $`b_2=\{z=(z_1,z_2)^2||z_10.02|^2+|z_2+1|^2<1\}`$. Let $`\mathrm{\Omega }=(B_2\backslash \overline{B_1})b_1(b_2B_2)`$.
Consider now pairs of points $`p_j,q_j\mathrm{\Omega }`$ , $`p_j=(1.5,2^j),q_j=(1.5,2^j)`$, and bydiscs $`U_j=\{z=(z_1,z_2)^2||z_1|<1.5,|z_22^j|<2^{2^j}\};j=1,2,\mathrm{}`$. And, finally domain $`D=\mathrm{\Omega }\underset{j=1}{\overset{\mathrm{}}{}}U_j`$.
Note the following properties:
1. $`D`$ is a connected domain, $`DB_2`$.
2. The entire complex disc $`\mathrm{\Delta }_j=\{z=(z_1,z_2)B_2|z_2=2^j\}D`$ for all $`j`$.
3. $`lim_j\mathrm{}p_j=p_0=(1.5,0),lim_j\mathrm{}q_j=q_0=(1.5,0);`$ and for the disc $`\mathrm{\Delta }_0=\{z=(z_1,z_2)B_2|z_2=0\}`$, $`\mathrm{\Delta }_0D=\{z^2|1<|z_1|<2,z_2=0)`$.
4. $`U_iU_j=\mathrm{}`$ for $`ij`$ for large enough $`i,j`$.
5. Let $`l_0=(0,10^4)\overline{b_1}`$ be the “tip” of this ball, and the point of $`\overline{b_1}`$ closest to the origin. Let $`d`$ denote the Kobayashi distance in $`B_2`$ from the origin to $`l_0`$, $`\overline{k}=\overline{k}(0,d)`$ the closed Kobayashi ball with the center at the origin and radius $`d`$. Then $`\overline{k}\overline{(b_1b_2)}=\{l_0\}`$.
We are going to show that for any $`j`$ the pair $`(p_j,q_j)\widehat{W}_2(D)`$, but their limit $`(p_0,q_0)\widehat{W}_2(D)`$, which will prove that $`\widehat{W}_2(D)`$ is not open in $`D^2`$.
Statement 1. For any $`j`$, there is a holomorphic retraction of $`B_2`$ (and therefore of $`DB_2`$) onto $`\mathrm{\Delta }_j`$, and since $`(p_j,q_j)\mathrm{\Delta }_j`$, the pair $`(p_j,q_j)\widehat{W}_2(D).`$
To prove this one needs to use first an automorphism $`g`$ of $`B_2`$ to move $`\mathrm{\Delta }_j`$ to $`\mathrm{\Delta }_0`$, use the natural projection $`P`$ of the ball $`B_2`$ onto $`\mathrm{\Delta }_0`$, and set the needed holomorphic retraction as $`g^1Pg`$.
Statement 2. Any holomorphic map $`f:DD`$ extends to a holomorphic map $`F:B_2B_2.`$
Statement 3. Any holomorphic map $`F:B_2B_2`$ that fixes our two points $`p_0,q_0`$ will fix all the points of $`\mathrm{\Delta }_0`$.
For proof see (\[Vi2\], ex. 1 in sec 4).
Now let $`f:DD`$ be a holomorphic map that fixes our two points $`p_0,q_0`$. Its extension $`F:B_2B_2`$ will be an identity on $`\mathrm{\Delta }_0`$.
Statement 4. $`F(l_0)=l_0`$.
Indeed, consider $`K`$the Kobayashi ball in $`B_2`$ with center at the origin that coincides with the standard unit ball in $`^2`$. $`KD`$ consists or nonintersecting connected pieces, only one of which, namely $`G=K(b_1b_2)`$ has a point $`(0.02,0)`$ on $`\mathrm{\Delta }_0`$ as a limit point. Since this point is fixed by $`F`$, and $`F`$ cannot increase the Kobayashi distance, we conclude that $`F(G)G`$.
Since it is also true that $`F(`$ $`\overline{k})`$ $`\overline{k}`$, the only possible image for $`l_0`$ under $`F`$ is the point itself (see property 5 above), so $`F(`$ $`l_0)=l_0`$.
We now conclude our observation by pointing out that $`F:B_2B_2`$ has three fixed points $`p_0,q_0,l_0`$. By (\[Vi2\], ex. 1 in sec 4) $`F`$ is the identity, so $`f:DD`$ is the identity, and therefore $`(p_0,q_0)\widehat{W}_2(D)`$.
### 2.3. Determining sets $`\widehat{W}_s(D)`$ that are not dense.
Our goal here is to present an example of a domain $`D`$ such that for any $`s`$ the determining set $`\widehat{W}_s(D)`$ is not dense in $`D^s`$.
J.-P. Vigué (see \[Vi2\], ex. 2 in sec 4) has provided such an example for $`D=\mathrm{\Delta }^2=\{z=(z_1,z_2)^2||z_1|<1,|z_2|<1\}`$ the polydisc in $`^2`$, and $`s=3`$. For completeness, using the same idea, we provide here an example for $`\mathrm{\Delta }^2`$ and any $`s3`$.
Fix $`s`$ points $`(A_j,0)`$, $`A_j=2^j,j=1,\mathrm{},s`$. The set $`T`$ of these points is a point in $`(\mathrm{\Delta }^2)^s`$. $`T\widehat{W}_s(\mathrm{\Delta }^2)`$, since $`(z_1,z_2)(z_1,0)`$ is a holomorphic retraction of $`\mathrm{\Delta }^2`$, fixing all these points. Let $`\stackrel{~}{T}=\{(a_j,b_j),j=1,\mathrm{},s\}(\mathrm{\Delta }^2)^s`$ be any $`\delta >0`$ perturbation of $`T`$. So, $`\underset{j=1}{\overset{s}{}}(|A_ja_j|^2+|b_j|^2)<\delta ^2`$. We will show that if $`\delta `$ is small enough $`\stackrel{~}{T}\widehat{W}_s(\mathrm{\Delta }^2)`$.
Consider the Lagrange interpolation polynomial $`\phi (w)=\underset{j=1}{\overset{s}{}}b_j\underset{ij}{}\frac{(wa_i)}{(a_ja_i)}`$. One can verify that if $`\delta >0`$ is small enough (say $`\delta <4^{s^2}`$) then $`|\phi (w)|<1`$ if $`|w|<1`$. Now the map $`(z_1,z_2)(z_1,\phi (z_1))`$is a holomorphic retraction of the unit polydisc that has $`\stackrel{~}{T}`$ in its set of fixed points. Therefore $`\stackrel{~}{T}\widehat{W}_s(\mathrm{\Delta }^2)`$.
As a remark we note that using this idea one can construct many such examples. Moreover the following theorem holds.
###### Theorem 2.5.
Consider the topological space $`\stackrel{ˇ}{D}_n`$ of all bounded domains in $`^n`$ with the topology induced by the Hausdorff distance between boundaries of the domains. Let $`\stackrel{˘}{G}_n\stackrel{ˇ}{D}_n`$ be such that if $`D\stackrel{˘}{G}_n`$ then $`\widehat{W}_s(D)`$ is not dense in $`D^s`$ for all $`s1`$. Then $`\stackrel{˘}{G}_n`$ is dense in the topological space $`\stackrel{ˇ}{D}_n`$. |
warning/0507/math0507289.html | ar5iv | text | # Kähler-Einstein metrics on orbifolds and Einstein metrics on spheres
## 1 Introduction
The aim of this paper is to explain how the methods of Arezzo, Ghigi, and Pirola can be applied to construct Kähler-Einstein metrics on compact complex orbifolds with positive first Chern class, and then use the approach of Boyer, Galicki, and Kollár to obtain new Einstein metrics on odd dimensional spheres.
The somewhat unusual aspect is that we work with orbifolds $`𝒳`$ that admit a map $`\pi :𝒳^n`$ which is the identity map set theoretically. Nonetheless, in the orbifold category $`\pi `$ is a nontrivial Galois cover, although with trivial Galois group.
The existence of Kähler-Einstein metrics on compact complex manifolds with positive first Chern class is still a difficult problem. For surfaces and toric manifolds a complete solution is known, due respectively to Tian and Wang-Zhu . Apart from these cases, there are two large classes of examples. The simplest are homogeneous spaces, for instance $`^n`$, quadrics, Grassmannians. In all these cases, the first Chern class is large, meaning for instance, that it is a large multiple of a generator of $`H_2(X,)`$. The opposite case, when the first Chern class is a small multiple of a generator of $`H_2(X,)`$ is also understood in many instances; see for a good overview.
A blending of these two approaches was developed in Arezzo, Ghigi, and Pirola to yield Kähler-Einstein metrics on certain manifolds $`X`$ which can be realized as Galois covers of another manifold $`Y`$ with a Kähler-Einstein metric. Since the method relies on finite group actions, it is most successfull when symmetries form a natural part of the complex structure, for instance for double covers of $`^n`$.
A construction of Einstein metrics on odd dimensional spheres was studied in Boyer, Galicki, and Kollár . The idea is that the quotient of an odd dimensional sphere by a circle action is frequently a complex orbifold, and a result of Kobayashi allows one to lift a Kähler-Einstein orbifold metric from the quotient to an Einstein metric on the sphere.
A frequently occurring case, studied by Orlik and Wagreich and Boyer, Galicki, and Kollár , appears when the quotient $`S^{2n+1}/S^1`$ is $`^n`$ as a manifold, and the orbifold structure is given by a $``$-divisor
$$\mathrm{\Delta }=\underset{i=0}{\overset{n+1}{}}\left(1\frac{1}{m_i}\right)D_i,$$
where
$$D_i=\{z_i=0\}\text{ for }i=0,\mathrm{},n,D_{n+1}=\{z_0+\mathrm{}+z_n=0\},$$
and the $`m_0,\mathrm{},m_{n+1}`$ are pairwise relatively prime ramification indices. (See Section 4 for precise definitions.) The orbifold first Chern class is
$$c_1(^n,\mathrm{\Delta })=(n+1)\underset{i=0}{\overset{n+1}{}}\left(1\frac{1}{m_i}\right)=\underset{i=0}{\overset{n+1}{}}\frac{1}{m_i}1,$$
where we have identified $`H^2(^n,)`$ with $``$. Thus $`c_1(^n,\mathrm{\Delta })`$ is positive iff
$$\underset{i=0}{\overset{n+1}{}}\frac{1}{m_i}1>0.$$
(1)
The existence result \[10, Theorem 34\] shows that $`(^n,\mathrm{\Delta })`$ has an orbifold Kähler-Einstein metric if in addition the following inequality is also satisfied
$$\underset{i=0}{\overset{n+1}{}}\frac{1}{m_i}1<\frac{n+1}{n}\underset{i}{\mathrm{min}}\{\frac{1}{m_i}\}.$$
(2)
This paper started with the observation that one can apply the method of to the identity map $`(^n,\mathrm{\Delta })^n`$ which is a Galois cover (with trivial Galois group). On the other hand, over the affine chart $`^n\{D_iD_j\}`$ the same map can be viewed as having cyclic Galois group of order $`_{ki,j}m_k`$. This approach improves the bound of by a factor of $`n`$, and we obtain
###### Theorem 1
Let $`D_0,\mathrm{},D_{n+1}^n`$ be hyperplanes in general position and $`m_0,\mathrm{},m_{n+1}`$ pairwise relatively prime natural numbers. Assume that
$$0<\underset{i=0}{\overset{n+1}{}}\frac{1}{m_i}1<(n+1)\underset{i}{\mathrm{min}}\{\frac{1}{m_i}\}.$$
(3)
Then there is an orbifold Kähler-Einstein metric on $`(^n,_{i=0}^{n+1}(1\frac{1}{m_i})D_i)`$.
Set $`M=_im_i`$ and $`w_i=M/m_i`$. As shown in the intersection of the unit sphere with the Brieskorn–Pham singularity
$$L(m_0,\mathrm{},m_{n+1}):=S^{2n+3}\left(\underset{i=0}{\overset{n+1}{}}z_i^{m_i}=0\right)^{n+2}$$
is homeomorphic to $`S^{2n+1}`$ and a Kähler-Einstein metric on the corresponding projective orbifold
$$(X,\mathrm{\Delta }_X):=(\left(\underset{i=0}{\overset{n+1}{}}z_i^{m_i}=0\right),\underset{i=0}{\overset{n+1}{}}(1\frac{1}{m_i})[z_i=0])(w_0,\mathrm{},w_{n+1})$$
lifts to a positive Ricci curvature Einstein metric on $`L(m_0,\mathrm{},m_{n+1})`$. The weighted projective space $`(w_0,\mathrm{},w_{n+1})`$ is not well formed and it is isomorphic to the ordinary projective space $`^{n+1}`$ by the map
$$(z_0,\mathrm{},z_{n+1})(x_0=z_0^{m_0},\mathrm{},x_{n+1}=z_{n+1}^{m_{n+1}}).$$
Under this isomorphism we get that
$$(X,\mathrm{\Delta }_X)(\left(\underset{i=0}{\overset{n+1}{}}x_i=0\right),\underset{i=0}{\overset{n+1}{}}(1\frac{1}{m_i})[x_i=0])^{n+1}.$$
By eliminating the variable $`x_{n+1}`$ we get that
$$(X,\mathrm{\Delta }_X)(^n,\mathrm{\Delta }).$$
The isometry class of the metric on the sphere determines the complex orbifold $`(^n,_{i=0}^{n+1}(1\frac{1}{m_i})D_i)`$, except possibly when $`(^n,_{i=0}^{n+1}(1\frac{1}{m_i})D_i)`$ has a holomorphic contact structure. The latter can happen only when $`n`$ is odd; see \[10, Lem.17\] for another necessary condition. (Note that $`n+2`$ hyperplanes in general position do not have moduli, so the numbers $`m_0,\mathrm{},m_{n+1}`$ alone determine the complex orbifold.)
Even with the improved bounds, the equations (3) are not easy to satisfy. Still, as in Example 43, we get 12 new Einstein metrics on $`S^5`$ corresponding to the ramification indices
$$m_0=2,m_1=3,m_2=5,m_3\{17,19,23,29,31,37,41,43,47,49,53,59\},$$
$`10^3`$ new Einstein metrics on $`S^7`$, $`10^6`$ new Einstein metrics on $`S^9`$
The above construction can be varied in many ways. For instance, one can take more than $`n+2`$ hyperplanes and quadrics. In all of these cases one gets an improvement by a factor roughly $`n`$ compared to the bounds in , but this gives many new cases only for $`n`$ large. (As shown by Orlik and Wagreich , taking higher degree hypersurfaces for the $`D_i`$ yields Einstein metrics on various rational homology spheres.)
As another application, we consider singular degree 2 Del Pezzo surfaces. These are all double covers of $`^2`$ ramified along a quartic curve. In the smooth case the existence of Kähler-Einstein metrics was proved by Tian . For singular surfaces we get the following.
###### Theorem 2
Let $`S`$ be a degree 2 Del Pezzo surface with only $`A_1`$ or $`A_2`$ singularities. Then $`S`$ has an orbifold Kähler-Einstein metric.
Anyone well versed in orbifolds, stacks and in the theory of Monge–Ampère equations should have no problem developing the theory of in the orbifold setting. Nonetheless, since the theory of orbifolds has too many “well known” but never proved theorems and not quite correct definitions and proofs, we felt that it makes sense to write down the arguments in some detail.
## 2 Analytic coverings
Let $`X`$ and $`Y`$ be reduced complex spaces. A map $`\pi :XY`$ is called *finite* if it is proper and has finite fibres. Since $`X`$ is locally compact a finite to one map is proper if and only if it is closed. Therefore a map is finite if and only if it is closed and has finite fibres. (By contrast note that $`\pi :\{1\}\{y^2=x^3+x^2\}^2`$ given by $`t(t^21,t^3t)`$ is a closed map of algebraic varieties with finite fibers but $`\pi `$ is not proper.)
The fundamental theorem on finite maps (see \[15, p. 179\]) states that when $`X`$ and $`Y`$ are irreducible any finite surjective map $`\pi :XY`$ is an *analytic covering*. This means that there is a thin subset $`TY`$ such that
* $`\pi ^1(T)`$ is thin in $`X`$, and
* the restriction $`\pi ^1(YT)YT`$ is locally biholomorphic (étale).
Put $`Y_0=YT`$ and $`X_0=\pi ^1(Y_0)`$. Then $`\pi :X_0Y_0`$ is a topological covering. We call it a *regular subcover* of $`\pi `$.
We assume that our spaces are irreducible so that “analytic covering” and “finite holomorphic surjection” can be regarded as synonyms.
Another important fact is that an analytic covering $`\pi :XY`$ with $`X`$ and $`Y`$ normal is an open map (see \[15, p. 135\]).
Let now $`\pi :XY`$ be an analytic covering among connected *normal* complex spaces. Put $`Y^{}=\{yY_{\mathrm{reg}}:\pi ^1(y)X_{\mathrm{reg}}\}`$ and $`X^{}=\pi ^1(Y^{})`$. Then $`X^{}`$ and $`Y^{}`$ are open sets with complements of codimension at least 2. Now $`\pi :X^{}Y^{}`$ is a finite surjective map between complex manifolds. Pick local coordinates $`z_1,\mathrm{},z_n`$ on a neighbourhood $`U`$ of a point in $`X^{}`$ and let $`w_1,\mathrm{},w_n`$ be coordinates around its image in $`Y^{}`$. Let $`w_i=\pi _i(z)`$ be the local expression of $`\pi `$. The divisors locally defined by the equation
$$\mathrm{det}\left(\frac{\pi _\mathrm{i}}{\mathrm{z}_\mathrm{j}}\right)=0$$
glue together yielding a well-defined divisor on $`X^{}`$. Since the complement of $`X^{}`$ has codimension at least 2, the Remmert-Stein extension theorem (see e.g. \[15, p. 181\]) ensures that the topological closure of this divisor is a divisor in $`X`$, called the *ramification divisor* of $`\pi `$, and denoted by $`R=R(\pi )`$. It satisfies the Hurwitz formula $`K_Y^{}=\pi ^{}K_X^{}+R`$. Write $`R=_jr_jR_j`$ with $`R_j`$ distinct prime divisors on $`X^{}`$. The reduced divisor $`R_{\mathrm{red}}=_jR_j`$ is called the *ramification locus*. By the implicit function theorem $`R_{\mathrm{red}}X^{}`$ is the set of points $`xX^{}`$ such that $`\pi `$ is not étale at $`x`$, that is the set of critical points of $`\pi `$. Since $`\pi `$ is finite, the image $`\pi (R_{\mathrm{red}})`$ is a divisor on $`Y`$, called the *branch divisor* of $`\pi `$.
Consider now the sets $`X^{\prime \prime }=X^{}\left((R_{\mathrm{red}})_{\mathrm{sing}}\pi ^1(B_{\mathrm{sing}})\right)`$ and $`Y^{}=\pi (X^{\prime \prime })`$. Both are open and have complements of codimension at least 2 in $`X`$ and $`Y`$ respectively. We use this notation often in the sequel. When we want to stress the dependence on $`\pi `$, we write $`X^{\prime \prime }(\pi )`$ and $`Y^{\prime \prime }(\pi )`$. If $`xX^{\prime \prime }`$ either $`xR_{\mathrm{red}}`$ or $`x`$ belongs to one and only one component $`R_j`$. In the first case we say that $`\pi `$ is *unramified* at $`x`$, in the latter case we say that the *ramification order of $`\pi `$ at $`x`$* is $`r_j+1`$. The ramification order of $`\pi `$ at $`x`$ will be denoted by $`\mathrm{ord}_\pi (x)`$. When $`\pi `$ is unramified at $`x`$, we put $`\mathrm{ord}_\pi (x)=1`$. If $`DX`$ is an irreducible divisor, then there is an open dense subset $`D^{\prime \prime }D`$ such that $`\mathrm{ord}_\pi (x)`$ does not depend on $`xD^{\prime \prime }`$. This common value is denoted by $`\mathrm{ord}_\pi (D)`$ and it is called the *ramification order of $`\pi `$ along $`D`$*.
We use some basic properties of analytic coverings and maps between them (see, for instance, \[6, Lemma 16.1\]).
###### Lemma 3
Let $`xX^{\prime \prime }`$. If $`\pi `$ is unramified at $`x`$, then $`\pi `$ is a local biholomorphism at $`x`$. If it has ramification order $`m>1`$, let $`R_j`$ be the component of $`R_{\mathrm{red}}`$ passing through $`x`$. Then there are local coordinates $`z_1,\mathrm{},z_n`$ on $`X^{\prime \prime }`$ and $`w_1,\mathrm{},w_n`$ on $`Y^{\prime \prime }`$ centred at $`x`$ and $`y=\pi (x)`$ respectively, such that locally $`R_j=\{z_1=0\}`$, $`B=\{w_1=0\}`$ and $`\pi (z_1,\mathrm{},z_n)=(z_1^m,z_2,\mathrm{},z_n)`$.
Since the complement of $`X^{\prime \prime }`$ has codimension 2, $`R_{\mathrm{red}}`$ is the closure of $`R_{\mathrm{red}}X^{\prime \prime }`$, that is the closure of the set of points where $`\pi `$ has ramification order $`>1`$.
The next lemma considers the problem of lifting in the simplest case. Denote by $`D(r)`$ the disc of radius $`r`$ centred at the origin, by $`D^{}(r)`$ the complement of $`\{0\}`$ in $`D(r)`$, and by $`P(r_1,\mathrm{},r_n)`$ the polydisc centred at the origin with polyradius $`(r_1,\mathrm{},r_n)`$.
###### Lemma 4
Let $`P_1=P(r_1,\mathrm{},r_n),P_2=P(\rho _1,\mathrm{},\rho _n)`$, $`Q_1=P(r_1^{m_1},r_2,\mathrm{},r_n)`$, $`Q_2=P(\rho _1^{m_2},\rho _2,\mathrm{},\rho _n)`$. Set $`P_1^{}=D^{}(r_1)\times P(r_2,\mathrm{},r_n)`$ and similarly for $`P_2^{},Q_1^{},Q_2^{}`$. Let $`\pi _i:P_iQ_i`$ be the maps $`\pi _1(z_1,..,z_n)=(z_1^{m_1},z_2,\mathrm{},z_n)`$, $`\pi _2(z_1,..,z_n)=(z_1^{m_2},z_2,\mathrm{},z_n).`$ Let $`f:Q_1Q_2`$ be a holomorphic map such that $`f(Q_1^{})Q_2^{}`$. If $`m_2|m_1`$ there are exactly $`m_2`$ liftings of $`f`$ (that is maps $`\stackrel{~}{f}:P_1P_2`$ such that $`\pi _2\stackrel{~}{f}=f\pi _1`$). Any local lifting of $`f`$ defined in a neighbourhood of some point $`xP_1`$ extends to one of these liftings defined on $`P_1`$.
###### Lemma 5
Let $`\pi _1:X_1Y`$ and $`\pi _2:X_2Y`$ be analytic coverings. For $`UX_1`$ set
$$𝔉(U)=\{\text{holomorphic maps }s:UX_2\text{ such that }\pi _1=\pi _2s\}.$$
Then $`𝔉`$ is a Hausdorff sheaf (of sets) over $`X_1`$. Assume that for any $`x_1X_1^{\prime \prime },x_2X_2^{\prime \prime }`$ with $`\pi _1(x_1)=\pi _2(x_2)`$
$$\mathrm{ord}_{\pi _2}(x_2)|\mathrm{ord}_{\pi _1}(x_1).$$
Then the restriction of $`𝔉`$ to $`X_1^{\prime \prime }\pi _1^1Y_2^{\prime \prime }(\pi _2)`$ is a finite topological covering. In particular, if $`X_1^{\prime \prime }`$ is simply connected, then for every $`x_1X_1^{\prime \prime }\pi _1^1Y_2^{\prime \prime }(\pi _2)`$ and $`x_2X_2^{\prime \prime }`$ such that $`\pi _1(x_1)=\pi _2(x_2)`$ there is an analytic map $`f:X_1^{\prime \prime }X_2`$ such that $`f(x_1)=x_2`$ and $`\pi _1=\pi _2f`$.
In fact, the above $`f`$ extends to $`X_1`$ by the following immediate consequence of the Riemann Extension Theorem (see e.g. \[15, p.144\])
###### Lemma 6
Let $`\pi _1:X_1Y`$ and $`\pi _2:X_2Y`$ be analytic coverings, $`X_1`$ normal and $`TX_1`$ a thin set. Let $`f^0:X_1TX_2`$ be an analytic map such that $`\pi _1=\pi _2f^0`$. Then $`f^0`$ extends to $`f:X_1X_2`$ such that $`\pi _1=\pi _2f`$.
## 3 The Galois group of coverings
Let $`f:XY`$ be an analytic covering of normal complex spaces. Put $`\mathrm{Gal}(\pi )=\{f\mathrm{Aut}(X):\pi f=\pi \}`$. $`\mathrm{Gal}(\pi )`$ is a finite subgroup of $`\mathrm{Aut}(X)`$. In fact fix $`xX^{\prime \prime }`$, $`y=\pi (x)`$, and let $`V`$ be a neighbourhood of $`y`$ in $`Y`$ such that $`\pi ^1(V)=_{i=1}^kU_i`$ with $`\pi :U_iV`$ a biholomorphism and $`xU_1`$. Then the stabiliser $`\mathrm{Gal}(\pi )_x`$ is a subgroup of finite index in $`\mathrm{Gal}(\pi )`$. Moreover any $`f\mathrm{Gal}(\pi )_x`$ maps $`U_1`$ to itself. Since $`\pi _{|U_1}`$ is injective, the restriction of $`f`$ to $`U_1`$ is the identity. By the connectedness of $`X`$, $`f=\mathrm{id}_X`$, so $`\mathrm{Gal}(\pi )_x=\{1\}`$ and $`\mathrm{Gal}(\pi )`$ is finite.
Since $`\pi `$ is $`\mathrm{Gal}(\pi )`$-invariant, the $`\mathrm{Gal}(\pi )`$-orbit of $`xX`$ is contained in $`\pi ^1\left(\pi (x)\right)`$. We say that an analytic covering $`\pi :XY`$ is *Galois* if the converse holds, that is two points of $`X`$ lie on the same fibre of $`\pi `$ only if they belong to the same $`\mathrm{Gal}(\pi )`$-orbit.
The branching divisor of a Galois cover can be described also in the following way. Given a prime divisor $`D`$ in $`X`$, set $`\mathrm{\Gamma }(D)=\{\gamma \mathrm{\Gamma }:D\mathrm{Fix}(\gamma )\}`$. For each prime divisor $`D`$ the image $`\pi (D)`$ is a prime divisor in $`Y`$. The prime divisors for which $`\mathrm{\Gamma }(D)0`$ are exactly the $`R_j`$. Set $`B_j=\pi (R_j)`$. In general different $`R_j`$’s can have the same image. Assume that $`\{B_i\}_{iI}`$ is the set of all images of the $`R_j`$’s (that is $`B_iB_k`$ if $`ik`$). Then
$$B(\pi )=\underset{iI}{}\left(1\frac{1}{|\mathrm{\Gamma }(R_i)|}\right)B_i.$$
(4)
###### Lemma 7
Let $`X`$ and $`Y`$ be normal complex spaces, $`\pi :XY`$ an analytic covering and $`Y_0Y`$ an open subset with thin complement. Put $`X_0=\pi ^1(Y_0)`$ and $`\pi _0=\pi _{|X_0}:X_0Y_0`$. Then the elements of $`\mathrm{Gal}(\pi _0)`$ extend to elements of $`\mathrm{Gal}(\pi )`$, and if $`\pi _0`$ is Galois, then $`\pi `$ is Galois too.
Proof. The first part follows from Lemma (6). For the second part, let $`x,x^{}X`$ be such that $`\pi (x)=\pi (x^{})=y`$. If $`yY_0`$ there is some $`g\mathrm{Gal}(\pi _0)`$ such that $`g.x=x^{}`$. Since we have just proved that $`\mathrm{Gal}(\pi _0)=\mathrm{Gal}(\pi )`$ the Galois condition is satisfied for these points. If instead $`yYY_0`$, choose neighbourhoods $`U_i`$ and $`V`$ as above. Assume $`x=x_1U_1`$ and $`x^{}=x_2U_2`$. Let $`\{z_n\}`$ be a sequence of points in $`X_0U_1`$ converging to $`x`$. Then $`y_n=\pi (z_n)`$ converge to $`y`$. Since $`\pi `$ is open, $`\pi (U_2)=V`$. Therefore there are points $`z_n^{}U_2X_0`$ such that $`\pi (z_n^{})=y_n`$. By the Galois condition on $`X_0`$, there are $`g_n\mathrm{Gal}(\pi )`$ such that $`z_n^{}=g_n.z_n`$. As $`\mathrm{Gal}(\pi )`$ is finite, we can extract a subsequence with $`g_ng`$. Since $`limz_n^{}=x_2`$ as $`\pi ^1(y)U_2=\{x_2\}`$, we get $`x_2=g.x_1`$. $`\mathrm{}`$
If $`\pi :XY`$ is a Galois covering, then $`\mathrm{Gal}(\pi )`$ acts freely on any regular subcover $`X_0`$. Therefore if $`x,x^{}X_0`$ and $`\pi (x)=\pi (x^{})`$, then there is a unique $`g\mathrm{Gal}(\pi )`$ such that $`g.x=x^{}`$. In particular the cardinality of $`\mathrm{Gal}(\pi )`$ equals that of the generic fibre. This condition is also sufficient: $`\pi `$ is Galois iff $`|\mathrm{Gal}(\pi )|`$ equals the cardinality of the general fibre iff $`\mathrm{Gal}(\pi )`$ is transitive on the general fibre.
For later reference we state the following simple lemma.
###### Lemma 8
Let $`X,Y`$ and $`Z`$ be irreducible complex spaces, and $`f:XZ`$, $`g:YZ`$, $`h:XY`$ analytic coverings such that $`gh=f`$. If $`f`$ is Galois, then $`h`$ is Galois too.
Proof. Thanks to Lemma 7 it is enough to consider the unramified case. Fix $`xX`$ and put $`y=h(x),z=f(x)=g(y)`$. We need to show that $`h_{}\pi _1(X,x)`$ is a normal subgroup of $`\pi _1(Y,y)`$. Since $`g_{}:\pi _1(Y,y)\pi _1(Z,z)`$ is injective it is enough to check that $`g_{}h_{}\pi _1(X,x)`$ is a normal subgroup of $`g_{}\pi _1(Y,y)`$. But $`f`$ being Galois $`f_{}\pi _1(X,x)=g_{}h_{}\pi _1(X,x)`$ is normal in $`\pi _1(Z,z)`$, hence a fortiori in $`g_{}\pi _1(Y,y)`$. $`\mathrm{}`$
For a general analytic covering $`\pi :XY`$ it is not possible to assign multiplicity to the branching divisor in any reasonable way. In fact, different points in the preimage of a point $`yB`$ have different branching orders. A typical example is $`X=\{z^33yz+2x=0\}^3`$ projecting on $`_{x,y}^2`$. Even shrinking the domain around the origin, one cannot separate the branches with different orders.
On the other hand, when the covering is Galois, for any $`yY^{\prime \prime }`$ all points in $`\pi ^1(y)`$ have the same branching order. Therefore we can assign multiplicities to the branch divisor according to the following rule. Let $`yY^{\prime \prime }B`$ and let $`x`$ be any point in $`\pi ^1(y)`$. Then we define the multiplicity of $`B`$ in $`y`$ to be $`11/\mathrm{ord}_\pi (x)`$. We still denote by $`B`$ the $``$-divisor given by the branching locus provided with these multiplicities. Note that with this convention $`R=\pi ^{}B`$, that is, the ramification divisor is the pull back of the branch divisor.
## 4 Orbifolds as pairs
As in , we look at orbifolds as a particular type of log pairs. $`(X,\mathrm{\Delta })`$ is a log pair if $`X`$ is a normal algebraic variety (or a normal complex space) and $`\mathrm{\Delta }=_id_iD_i`$ is an effective $``$-divisor where the $`D_i`$ are distinct, irreducible divisors and $`d_i`$. The number $`d_i`$ is called the multiplicity of $`\mathrm{\Delta }`$ along $`D_i`$, it is denoted by $`\mathrm{mult}_{D_i}\mathrm{\Delta }`$. We set $`\mathrm{mult}_D\mathrm{\Delta }=0`$ for every other irreducible divisor $`DD_ii`$.
Let $`X^{\prime \prime }(\mathrm{\Delta })`$ (or simply $`X^{\prime \prime }`$) be the complement of $`X_{\mathrm{sing}}\mathrm{\Delta }_{\mathrm{sing}}`$. For $`xX^{\prime \prime }`$ the multiplicity of $`\mathrm{\Delta }`$ at $`x`$ is a well defined rational number. For orbifolds, we need to consider only pairs $`(X,\mathrm{\Delta })`$ such that $`\mathrm{\Delta }`$ has the form
$$\mathrm{\Delta }=\underset{i}{}\left(1\frac{1}{m_i}\right)D_i,$$
where the $`D_i`$ are prime divisors and $`m_i`$. If $`(X,\mathrm{\Delta })`$ is such a pair then for any divisor $`DX`$ we put
$$\mathrm{ord}_\mathrm{\Delta }(D)=\frac{1}{1\mathrm{mult}_D\mathrm{\Delta }}.$$
The assumption on the multiplicities of $`\mathrm{\Delta }`$ amounts to saying that the order is always a nonnegative integer.
###### Definition 9
An *orbifold chart* on $`X`$ compatible with $`\mathrm{\Delta }`$ is a Galois covering $`\phi :U\phi (U)X`$ such that
1. $`U`$ is a domain in $`^n`$ and $`\phi (U)`$ is open in $`X`$;
2. the branch locus of $`\phi `$ is $`\mathrm{\Delta }_{\mathrm{red}}\phi (U)`$;
3. for any $`xU^{\prime \prime }(\phi )`$ such that $`\phi (x)D_i`$, $`\mathrm{ord}_\phi (x)=m_i`$.
Conditions (2) and (3) are equivalent to
$$B(\phi )=\mathrm{\Delta }\phi (U).$$
(5)
###### Definition 10
An orbifold is a log pair $`(X,\mathrm{\Delta })`$ such that $`X`$ is covered by orbifold charts compatible with $`\mathrm{\Delta }`$.
(For a slightly more general approach, see \[13, §14\].)
Let $`X`$ be a normal complex space and $`\pi :UX`$ a Galois cover where $`U`$ is a smooth. As discussed earlier, the branch divisor $`B(\pi )`$ of $`\pi `$ is defined and we get a log pair $`(X,B(\pi ))`$. If $`U`$ is simply connected, (which we can always assume by shrinking $`U`$ suitably) then by Lemma 5 the log pair $`(X,B(\pi ))`$ determines $`\pi :UX`$ up to biholomorhisms. Thus we recover the classical definition of orbifolds (as in for example).
###### Example 11
Let $`X`$ be a complex manifold and $`D=_{iI}D_i`$ a divisor with *local* normal crossing. By this we mean that for any point $`xX`$ there is a holomorphic coordinate system $`(V,z_1,\mathrm{},z_n)`$ such that $`DV=\{zV:z_1\mathrm{}z_k=0\}`$. If $`D_iV\mathrm{}`$ then $`D_iV`$ is the union of some of the hypersurfaces $`\{z_j=0\}`$. ($`D`$ is said to be a divisor with *global* normal crossing if, in addition, each $`D_i`$ is smooth.) For any $`iI`$, fix an integer $`m_i>1`$ and put $`\mathrm{\Delta }=_i(11/m_i)D_i`$. We claim that $`(X,\mathrm{\Delta })`$ is an orbifold. Indeed, fix a coordinate system as above and put $`m_j^{}=m_i`$ if $`\{z_j=0\}D_iV`$. Set
$$\phi :UV,\pi (x_1,\mathrm{},x_n)=(x_1^{m_1^{}},\mathrm{},x_k^{m_k^{}},x_{k+1},\mathrm{},x_n).$$
(6)
Then $`(U,\phi )`$ is orbifold chart on $`X`$ compatible with $`\mathrm{\Delta }`$ and so $`(X,\mathrm{\Delta })`$ is an orbifold.
In the same way, the usual definition of orbifold map is equivalent to the following one.
###### Definition 12
A finite holomorphic map $`f:XY`$ is an *orbifold map* $`f:(X,\mathrm{\Delta }_X)(Y,\mathrm{\Delta }_Y)`$ if
$$\mathrm{ord}_{\mathrm{\Delta }_Y}(f(D))|\mathrm{ord}_{\mathrm{\Delta }_X}(D)\mathrm{ord}_fD$$
(7)
for every divisor $`DX`$.
An *orbifold automorphism* is an orbifold map that is invertible with inverse an orbifold map. The group of automorphisms of $`(X,\mathrm{\Delta })`$ is denoted by $`\mathrm{Aut}(X,\mathrm{\Delta })`$.
###### Definition 13
An orbifold Galois covering $`f:(X,\mathrm{\Delta }_X)(Y,\mathrm{\Delta }_Y)`$ is an orbifold map such that $`f:XY`$ is a Galois analytic cover and $`\mathrm{Gal}(f)\mathrm{Aut}(X,\mathrm{\Delta }_X)`$.
By the *degree* of an orbifold Galois cover we mean its *degree* as an analytic cover.
###### Lemma 14
Let $`f:(X,\mathrm{\Delta }_X)(Y,\mathrm{\Delta }_Y)`$ be an orbifold map. Then given $`xX`$ and $`y=f(x)Y`$ there are orbifold charts $`(U,\phi )`$ and $`(V,\psi )`$ around $`x`$ and $`y`$ respectively such that $`f`$ has a lifting $`\stackrel{~}{f}:UV`$. If, in addition, $`f:XY`$ is a Galois covering then $`\stackrel{~}{f}:UV`$ is also a Galois covering.
Proof. Choose the chart $`(U,\phi )`$ such that $`U`$ is simply connected and $`f\left(\phi (U)\right)\psi (V)`$. If $`DU`$ is any divisor then
$$\mathrm{ord}_{f\phi }D=\mathrm{ord}_f\phi (D)\mathrm{ord}_\phi D=\mathrm{ord}_f\phi (D)\mathrm{ord}_{\mathrm{\Delta }_X}\phi (D).$$
By the definition of orbifold maps,
$$\mathrm{ord}_{\mathrm{\Delta }_Y}(f\phi )(D)|\mathrm{ord}_f\phi (D)\mathrm{ord}_{\mathrm{\Delta }_X}\phi (D),$$
hence we conclude that $`\mathrm{ord}_{\mathrm{\Delta }_Y}(f\phi )(D)`$ divides $`\mathrm{ord}_{f\phi }D`$. Thus the assumption of Lemma 5 is satisfied and so $`f\phi `$ lifts to $`\stackrel{~}{f}:UV`$. Assume next that $`f:XY`$ is a Galois covering and pick $`u_1,u_2U`$ such that $`\stackrel{~}{f}(u_1)=\stackrel{~}{f}(u_2)`$. Then $`f(\phi (u_1))=f(\phi (u_2))`$ hence there is a Galois automorphism $`\sigma `$ of $`f`$ such that $`\phi (u_1)=\sigma (\phi (u_2))`$. Applying Lemma 5 to $`\phi :UX`$ and $`\sigma \phi :UX`$ we conclude that $`\sigma `$ lifts to a biholomorphism $`\stackrel{~}{\sigma }`$ of $`U`$ such that $`\phi (u_1)=\phi (\stackrel{~}{\sigma }(u_2))`$. Since $`\phi :UX`$ is Galois, there is a biholomorphism $`\rho `$ of $`U`$ such that $`u_1=\rho (\stackrel{~}{\sigma }(u_2))`$. This shows that in the commutative diagram
$$\begin{array}{ccc}U& \stackrel{\stackrel{~}{f}}{}& V\\ \phi & & \psi & & \\ \phi (U)& \stackrel{f}{}& \psi (V).\end{array}$$
(8)
the composite $`f\phi `$ is Galois. But $`f\phi =\psi \stackrel{~}{f}`$ and by Lemma 8 $`\stackrel{~}{f}`$ is a Galois cover. $`\mathrm{}`$
###### Example 15
Let $`(X,\mathrm{\Delta })`$ be any orbifold, and let $`(X,0)`$ denote the orbifold structure on $`X`$ with trivial branching divisor. It is a nontrivial result that $`(X,0)`$ is an orbifold, that is, $`X`$ has quotient singularities (see ). (We use mainly the case when $`X`$ is smooth, and then the orbifold charts of $`(X,0)`$ are simply the manifold charts of $`X`$.)
The identity map $`\mathrm{id}_X:(X,\mathrm{\Delta })(X,0)`$ is trivially an orbifold Galois covering. In fact it is both an orbifold map and a Galois analytic cover, and $`\mathrm{Gal}(\mathrm{id}_X)=\{id_X\}\mathrm{Aut}(X,\mathrm{\Delta })`$.
If $`f:(X,\mathrm{\Delta })(Y,\mathrm{\Delta }_Y)`$ is an orbifold Galois covering the *orbifold ramification divisor* of $`f`$ is defined as
$$R^{orb}(\mathrm{\Delta }_X,\mathrm{\Delta }_Y,f)=R(f)+\mathrm{\Delta }_Xf^{}\mathrm{\Delta }_Y.$$
With this definition the logarithmic ramification formula
$$K_X+\mathrm{\Delta }_X=f^{}(K_Y+\mathrm{\Delta }_Y)+R^{orb}(\mathrm{\Delta }_X,\mathrm{\Delta }_Y,f)$$
is automatically satisfied. To understand the geometric meaning of $`R^{orb}`$ it is useful to look at the open set
$$X^{\prime \prime }(\mathrm{\Delta }_X,\mathrm{\Delta }_Y,f)=X_{\mathrm{reg}}f^1\left(Y_{\mathrm{reg}}(\mathrm{\Delta }_YB(f))_{\mathrm{sing}}\right)(\mathrm{\Delta }_XR(f))_{\mathrm{sing}}.$$
This means that $`xX^{\prime \prime }=X^{\prime \prime }(\mathrm{\Delta }_X,\mathrm{\Delta }_Y,f)`$ if (a) $`X`$ is smooth at $`x`$, (b) $`Y`$ is smooth at $`y=f(x)`$, (c) $`x`$ belongs to at most one component $`D`$ of $`\mathrm{\Delta }_X+R(f)`$ and in this case $`x`$ is a smooth point of $`D`$, (d) $`y`$ belongs to at most one component $`D^{}`$ of $`\mathrm{\Delta }_Y+B(f)`$ and in this case it is a smooth point of $`D^{}`$. As usual the complement of this set has codimension 2. Let $`D`$ be any smooth divisor passing through $`x`$ and $`D^{}`$ a smooth component passing through $`y`$. Assume first that $`f`$ is unbranched at $`x`$ and that locally $`\mathrm{\Delta }_X=(11/p)D`$ and $`\mathrm{\Delta }_Y=(11/q)D^{}`$. Then there is a local diagram like (8), with $`p=\mathrm{deg}\phi `$ and $`q=\mathrm{deg}\psi `$. Put $`k=\mathrm{deg}\stackrel{~}{f}`$. Since $`f`$ is unbranched we can assume that its restriction to $`\phi (U)`$ is a biholomorphism onto $`\psi (V)`$. Therefore $`p=qk`$. If $`p=1`$, then $`q=k=1`$, and as expected $`\mathrm{mult}_xR^{orb}=0`$. If $`p>1`$, then necessarily $`D^{}=f(D)`$ because of (7) and $`f^{}D^{}=D`$, since $`f`$ is étale. Therefore $`R^{orb}=(1/q1/p)D=(k1)/pD`$. If instead $`\mathrm{ord}_x(f)=m>1`$, then again $`D^{}=f(D)`$, $`R(f)=(m1)D`$, $`f^{}D^{}=mD`$, $`pm=qk`$ and $`R^{orb}=(m/q1/p)D=(k1)/pD`$ once more. Roughly the orbifold ramification divisor is the ramification of the lifting $`\stackrel{~}{f}`$ divided the degree of the local chart $`\phi `$.
Let $`(X,\mathrm{\Delta })`$ be an orbifold and $`\mathrm{\Gamma }\mathrm{Aut}(X,\mathrm{\Delta })`$ a finite subgroup. We want to define a quotient orbifold $`(Y,\mathrm{\Delta }^{})`$. By Cartan’s lemma $`Y=X/\mathrm{\Gamma }`$ is a normal analytic space and the canonical projection $`\pi :XY`$ is an analytic covering. The support of the branch divisor $`\mathrm{\Delta }^{}`$ is defined to be $`\pi (\mathrm{\Delta })B(\pi )`$, while the multiplicities are specified as follows. Let $`D`$ be an irreducible component of $`\pi (\mathrm{\Delta })B(\pi )`$. If $`D`$ is a component of $`\pi (\mathrm{\Delta })`$ and not of $`B(\pi )`$, then we assign to $`D`$ the multiplicity $`\mathrm{mult}_x(\mathrm{\Delta })`$, where $`x`$ is any point in $`X^{\prime \prime }(\mathrm{\Delta })`$ such that $`\pi (x)D`$ is a smooth point of $`\pi (\mathrm{\Delta })B(\pi )`$. If $`D`$ is a component of $`B(\pi )`$ and not of $`\pi (\mathrm{\Delta })`$ then we assign to $`D`$ the same multiplicity it has as a component of $`B(\pi )`$, that is $`11/\mathrm{ord}_\pi (x)`$ for any $`xX^{\prime \prime }(\pi )`$ such that $`\pi (x)D`$ is a smooth point of $`\pi (\mathrm{\Delta })B(\pi )`$. Finally, if $`D`$ is a common component of $`\pi (\mathrm{\Delta })`$ and $`B(\pi )`$ the we assign to it the multiplicity
$$1\frac{1\mathrm{mult}_x\mathrm{\Delta }}{\mathrm{ord}_\pi (x)}$$
for any $`xX^{\prime \prime }(\mathrm{\Delta })X^{\prime \prime }(\pi )`$ such that $`\pi (x)D`$ is a smooth point of $`\pi (\mathrm{\Delta })B(\pi )`$.
###### Proposition 16
Let $`(X,\mathrm{\Delta })`$ be an orbifold, and $`\mathrm{\Gamma }\mathrm{Aut}(X,\mathrm{\Delta })`$ a finite subgroup. Let $`Y=X/\mathrm{\Gamma }`$ be the quotient analytic space, and $`\mathrm{\Delta }^{}`$ the $``$-divisor defined above. Then $`(Y,\mathrm{\Delta }^{})`$ is an orbifold and the canonical projection
$$\pi :(X,\mathrm{\Delta }_X)(Y,\mathrm{\Delta }^{})$$
(9)
is an orbifold Galois covering.
Proof. We need to show that $`Y`$ is covered by orbifold charts compatible with $`\mathrm{\Delta }^{}`$. Fix $`yY`$, $`x\pi ^1(y)`$ and let $`\phi :U\phi (U)`$ be an orbifold chart with $`x\phi (U)`$. If the stabiliser $`\mathrm{\Gamma }_x`$ is trivial we can assume that $`\gamma \phi (U)\phi (U)=\mathrm{}`$ for any $`\gamma e`$. Then $`\pi :\phi (U)Y`$ is a biholomorphism onto its image. Put $`\psi =\pi \phi :UY`$. We claim that $`\psi `$ is an orbifold chart on $`Y`$ compatible with $`\mathrm{\Delta }^{}`$. In fact $`\psi `$ is Galois since $`\pi `$ is a biholomorphism on $`\phi (U)`$, and $`\pi ^{}B(\psi )=B(\phi )=\mathrm{\Delta }\phi (U)`$. On the other hand $`B(\pi )\psi (U)=\mathrm{}`$ since $`\pi :\phi (U)\psi (U)`$ is biholomorphic. Therefore on $`\psi (U)`$ the divisor $`\mathrm{\Delta }^{}`$ coincides with $`B(\psi )`$. This proves that $`\psi :UY`$ is an orbifold chart. If $`\mathrm{\Gamma }_y\{e\}`$ take a chart $`\phi :U\phi (U)X`$ such that $`\phi (U)`$ be a $`\mathrm{\Gamma }_x`$-invariant neighbourhood of $`x`$. Lemma 14 ensures that also in this case $`\psi =\pi \phi :U\psi (U)\phi (U)/\mathrm{\Gamma }_x`$ is a Galois covering. It is easy to verify that $`B(\psi )=\mathrm{\Delta }^{}`$ on $`\psi (U)`$. Finally that $`\pi `$ is an orbifold Galois covering is clear: a lifting of $`\pi :\phi (U)\psi (U)`$ is given by the identity map $`UU`$ which is trivially Galois. $`\mathrm{}`$
## 5 Basic estimates for orbifold Kähler-Einstein metrics
In this section we collect the orbifold versions of some fundamental results due to Aubin, Bando-Mabuchi and Tian, that are needed in the existence criteria in the next section. Most of the proofs are the same as in the case of a manifold and we just give appropriate references. For the basic definitions of differential geometry on orbifolds see , , and . Some information on Sobolev spaces and Laplace operators on orbifolds can be found e.g. in .
###### Remark 17
Note that if $`X`$ is a complex manifold and $`\mathrm{\Delta }`$ is a non trivial branching divisor, then smoothness in the orbifold sense is rather different from ordinary smoothness. For example, $`f(z)=|z|`$ is not smooth in the ordinary sense, but it belongs to $`C^{\mathrm{}}(,\mathrm{\Delta })`$, where $`\mathrm{\Delta }`$ is the divisor concentrated at the origin with multiplicity $`1/2`$. In fact the inclusions $`C^{\mathrm{}}(X)C^{\mathrm{}}(X,\mathrm{\Delta })`$ and $`^k(X)^k(X,\mathrm{\Delta })`$ are in general strict.
###### Definition 18
A *Fano orbifold* is a compact complex orbifold $`(X,\mathrm{\Delta })`$ such that $`(K_X+\mathrm{\Delta })`$ is ample.
By the Baily-Kodaira imbedding theorem this is equivalent to the fact that $`\mathrm{c}_1(X,\mathrm{\Delta })`$ contains an orbifold Kähler metric.
The following is the orbifold analogue of Bonnet-Myers Theorem. It follows, for example, from the Bishop volume comparison Theorem for orbifolds, see \[7, Prop. 20, Cor. 21\].
###### Theorem 19
Let $`X`$ be an $`m`$-dimensional orbifold and $`g`$ a Riemannian orbifold metric on $`X`$ with $`\mathrm{Ric}(g)\epsilon (m1)g`$ for some $`\epsilon >0`$. Then $`\mathrm{diam}(X,g)\pi /\sqrt{\epsilon }`$.
###### Theorem 20 (\[20, Theorem B\])
Let $`(X,g)`$ be a Riemannian orbifold of dimension $`m>2`$ with $`\mathrm{Ric}(g)(m1)\epsilon ^2g`$ for some $`\epsilon 0`$. Then there is a constant $`C>0`$ depending only on $`m`$ and $`\epsilon \mathrm{diam}(X,g)`$ such that
$$u_{L^2}C\frac{\mathrm{vol}(X,g)^{1/m}}{\mathrm{diam}(X,g)}u_{L^{2m/(m2)}}$$
(10)
for any $`uW^{1,2}(X)`$ with $`_Xu\mathrm{dvol}_g=0`$.
Combining the last two theorems one gets the following uniform Sobolev embedding.
###### Corollary 21
Let $`(X,\mathrm{\Delta })`$ be an $`n`$-dimensional Fano orbifold. For any $`\epsilon >0`$ there is a constant $`C=C(\epsilon )>0`$ such that for any metric $`\omega `$ in the class $`2\pi \mathrm{c}_1(X,\mathrm{\Delta })`$ with $`\mathrm{Ric}(\omega )\epsilon \omega `$ and any $`uW^{1,2}(X,\mathrm{\Delta })`$
$$u_{L^{2n/(n1)}}Cu_{W^{1,2}}^2.$$
(11)
If $`(X,\mathrm{\Delta })`$ is a Kähler orbifold, $`\omega ^{1,1}(X,\mathrm{\Delta })`$ is a closed smooth form and $`\phi C^{\mathrm{}}(X,\mathrm{\Delta })`$, put $`\omega _\phi =\omega +\mathrm{i}\overline{}\phi `$. We write $`\omega _\phi >0`$ to mean that it is a Kähler metric. If $`\omega `$ is such that
$$[\omega ]^n,[X]=_X\omega ^n>0$$
and $`\phi C^{\mathrm{}}(X,\mathrm{\Delta })`$ put
$`I_\omega (\phi )`$ $`={\displaystyle \frac{1}{[\omega ]^n,[X]}}{\displaystyle \phi (\omega ^n\omega _\phi ^n)}`$ (12)
$`J_\omega (\phi )`$ $`={\displaystyle _0^1}{\displaystyle \frac{I_\omega (s\phi )}{s}}𝑑s`$ (13)
$`F_\omega ^0(\phi )`$ $`=J_\omega (\phi ){\displaystyle \frac{1}{[\omega ]^n,[X]}}{\displaystyle \phi \omega ^n}.`$ (14)
###### Lemma 22
$$J_\omega (\phi )=\frac{1}{[\omega ]^n,[X]}\underset{k=0}{\overset{n1}{}}\frac{k+1}{n+1}_M\mathrm{i}\phi \overline{}\phi \omega ^k\omega _\phi ^{nk1}$$
(15)
$$I_\omega (\phi )J_\omega (\phi )=\frac{1}{[\omega ]^n,[X]}\underset{k=0}{\overset{n1}{}}\frac{nk}{n+1}_X\mathrm{i}\phi \overline{}\phi \omega ^k\omega _\phi ^{nk1}.$$
(16)
If $`\omega >0`$ and $`\omega _\phi >0`$, then $`I_\omega (\phi )`$, $`J_\omega (\phi )`$ and $`I_\omega (\phi )J_\omega (\phi )`$ are nonnegative and vanish only if $`\phi `$ is constant. Moreover $`J_\omega I_\omega (n+1)J_\omega `$.
For (15) see \[24, Lemma 2.2\] or \[1, Lemma 2.1\]. For (16) expand $`\omega ^n\omega _\phi ^n`$. The last statements follow diagonalising simultaneously $`\omega `$ and $`\mathrm{i}\overline{}\phi `$. $`\mathrm{}`$
###### Lemma 23
If $`\lambda `$ is a positive constant then
$$F_{\lambda \omega }^0(\lambda \phi )=\lambda F_\omega ^0(\phi ).$$
(17)
Let $`\omega _0`$ be a closed (1,1)-form such that $`[\omega _0]^n,[X]>0`$. Given $`\phi _{01}`$, $`\phi _{12}C^{\mathrm{}}(X,\mathrm{\Delta })`$ put $`\omega _1=\omega _0+\mathrm{i}\overline{}\phi _{01}`$, $`\phi _{02}=\phi _{01}+\phi _{12}`$. Then
$$F_{\omega _0}^0(\phi _{02})=F_{\omega _0}^0(\phi _{01})+F_{\omega _1}^0(\phi _{12}).$$
(18)
(Same proof as in \[27, pp. 60f\].)
###### Lemma 24 (\[27, p. 59\])
If $`\phi _t`$ is a differentiable family of smooth functions on $`(X,\mathrm{\Delta })`$ then
$`{\displaystyle \frac{\mathrm{d}}{\mathrm{dt}}}J_\omega (\phi _t)={\displaystyle \frac{1}{[\omega ]^n,[X]}}{\displaystyle _X}\dot{\phi }_t\left(\omega ^n\omega _t^n\right)`$ (19)
$`{\displaystyle \frac{\mathrm{d}}{\mathrm{dt}}}F_\omega ^0(\phi _t)={\displaystyle \frac{1}{[\omega ]^n,[X]}}{\displaystyle _X}\dot{\phi }_t\omega _t^n`$ (20)
Assume now that $`\omega `$ is a Kähler orbifold metric in the canonical class, that is $`\omega 2\pi \mathrm{c}_1(X,\mathrm{\Delta })`$. Let $`f=f(\omega )C^{\mathrm{}}(X,\mathrm{\Delta })`$ be the unique function such that
$$\mathrm{Ric}(\omega )\omega =\mathrm{i}\overline{}f(\omega )_Xe^{f(\omega )}=_X\omega ^n.$$
(21)
Put $`V=[\omega ]^n,[X]=n!\mathrm{vol}(X)`$ and define $`A_\omega ,F_\omega :C^{\mathrm{}}(X,\mathrm{\Delta })`$ by
$$A_\omega (\phi )=\mathrm{log}\left[\frac{1}{V}_Xe^{f(\omega )\phi }\omega ^n\right]F_\omega (\phi )=F_\omega ^0(\phi )A_\omega (\phi ).$$
(22)
Using the notation of Lemma 23 if $`\omega _0,\omega _1`$ and $`\omega _2`$ are Kähler metrics, then
$$F_{\omega _0}(\phi _{02})=F_{\omega _0}(\phi _{01})+F_{\omega _1}(\phi _{12}).$$
(23)
For $`G\mathrm{Aut}(X,\mathrm{\Delta })`$ a subgroup of isometries of $`(X,\mathrm{\Delta },\omega )`$ put
$$P_G(X,\mathrm{\Delta },\omega )=\{\phi C^{\mathrm{}}(X,\mathrm{\Delta }):\omega _\phi >0,\text{ and }\phi \text{ is }G\text{-invariant}\}.$$
(24)
If $`G=\{1\}`$ we simply write $`P(X,\mathrm{\Delta },\omega )`$.
In order to construct a Kähler-Einstein metric on $`(X,\mathrm{\Delta })`$ the continuity method is applied: fix a Kähler metric $`\omega `$ in the canonical class and consider the well-known equations
$$(\omega +\mathrm{i}\overline{}\phi _t)^n=e^{ft\phi _t}\omega ^n$$
$`()_t`$
for a smooth family of functions in $`C^{\mathrm{}}(X,\mathrm{\Delta })`$. Yau’s estimates hold for orbifold metrics, and in particular the Calabi conjecture is true, which implies that $`()_0`$ admits a unique solution. Denote by $`\mathrm{\Delta }`$ the negative definite $`\overline{}`$-Laplacian on functions (that is $`\mathrm{\Delta }=\overline{}^{}\overline{}`$) and by $`\lambda _j`$ its eigenvalues.
###### Lemma 25 (\[2, Theorem 4.20 p. 116\])
Let $`\omega `$ be a Kähler metric on the compact orbifold $`(X,\mathrm{\Delta })`$. If $`\mathrm{Ric}(\omega )\epsilon >0`$, then $`\lambda _11`$.
It follows that the times $`t`$ for which $`()_t`$ is solvable form an open subset $`S[0,1]`$ and that solutions $`\phi _t`$ are smooth in $`t`$, see \[27, pp. 63-66\]. Given a $`C^0`$-estimate for the solutions, Yau’s estimates ensure that $`S`$ is closed, thus yielding the solution up to $`t=1`$, which is a Kähler-Einstein metric.
###### Proposition 26
Let $`\phi _t`$ be a solution to $`()_t`$ for $`t[0,T_0)`$. Then $`I_\omega (\phi _t)J_\omega (\phi _t)`$ is nondecreasing and $`F_\omega ^0(\phi _t)0`$.
Proof. Differentiating $`()_t`$ with respect to $`t`$ one gets
$$(\mathrm{\Delta }_t+t)\dot{\phi }_t=\phi _t.$$
(25)
Therefore
$$\begin{array}{c}\frac{\mathrm{d}}{\mathrm{dt}}\left(I_\omega (\phi _t)J_\omega (\phi _t)\right)=\hfill \\ \hfill =\frac{1}{V}_X\phi _t(\phi _t+t\dot{\phi }_t)\omega _t^n=(1t^2)\frac{1}{V}_X\phi _t^2\omega _t^n+\frac{1}{V}_X|\overline{}\dot{\phi }_t|^2\omega _t.\end{array}$$
This gives the first result. For the second use (20) and (25):
$$\begin{array}{c}\frac{\mathrm{d}}{\mathrm{dt}}tF_\omega ^0(\phi _t)=F_\omega ^0(\phi _t)\frac{t}{V}_X\dot{\phi }_t\omega _t^n=\\ =F_\omega ^0(\phi _t)+\frac{1}{V}_X(\mathrm{\Delta }_t\dot{\phi }_t+\phi _t)\omega _t^n=J_\omega (\phi _t).\end{array}$$
Since $`J_\omega 0`$ the result follows. $`\mathrm{}`$
The following estimates depend on the uniform Sobolev embedding (Lemma 21) and their proof uses Moser iteration.
###### Theorem 27 (\[27, p. 67ff\])
If $`\phi _t`$ is a family of solutions to $`()_t`$ on the time interval $`[0,T_0)`$, then there is a constant $`C=C(T_0)>0`$ such that for any $`t<T_0`$
$$||\phi _t||_{\mathrm{}}C(1+J_\omega (\phi _t))$$
(26)
$$0\underset{X}{inf}\phi _tC\left(\frac{1}{V}_X(\phi _t)\omega _t^n+C\right)$$
(27)
$$F_\omega (\phi _t)A_\omega (\phi _t)C(1t)C.$$
(28)
###### Lemma 28 (\[5, §6\])
Let $`(X,\mathrm{\Delta })`$ be a Fano orbifold, $`\omega _{KE}`$ a Kähler-Einstein metric and $`\omega `$ a metric in the canonical class. Then there is $`g\mathrm{Aut}(X,\mathrm{\Delta })`$ such that $`\omega =g^{}\omega _{KE}+\mathrm{i}\overline{}\psi `$ with $`\psi `$ orthogonal to $`\mathrm{ker}(\mathrm{\Delta }_{g^{}\omega _{KE}}+1)`$ in $`L^2(X,\omega _{KE}^n)`$.
###### Proposition 29 (\[26, Prop. 5.3\])
Let $`(X,\mathrm{\Delta })`$ be a Fano orbifold and $`\omega _{KE}`$ a Kähler-Einstein metric in the canonical class. If $`\omega =\omega _{KE}+\mathrm{i}\overline{}\psi `$ is a Kähler metric, with $`\psi \mathrm{ker}(\mathrm{\Delta }_{KE}+1)`$ and $`_Xe^\psi \omega _{KE}^{}{}_{}{}^{n}=0`$, there is a solution $`\{\phi _t\}_{t[0,1]}`$ of $`()_t`$ with $`\phi _0=0`$ and $`\phi _1=\psi `$.
###### Theorem 30 (Ding-Tian)
If a Fano orbifold $`(X,\mathrm{\Delta })`$ admits a Kähler-Einstein metric $`\omega _{KE}`$, then $`F_\omega `$ is bounded from below on $`P(X,\mathrm{\Delta },\omega )`$ for any $`\omega `$ in the canonical class.
Proof. Thanks to (23) it is enough to bound $`F_{\omega _{KE}}`$. Given $`\phi P(X,\mathrm{\Delta },\omega _{KE})`$ put $`\omega =\omega _{KE}+\mathrm{i}\overline{}\phi `$ and let $`g`$ and $`\psi `$ be as in Lemma 28. Using again (23) it is enough to bound $`F_{g^{}\omega _{KE}}(\psi )`$. Take a path as in Lemma 29. Thanks to Proposition 26 $`F_{g^{}\omega _{KE}}(\psi )=F_\omega (\psi )=F_\omega (\phi _1)=F_\omega ^0(\phi _1)0`$. $`\mathrm{}`$
###### Remark 31
These estimates are enough to prove one half of Tian’s fundamental theorem, namely that properness of $`F_\omega `$ implies the existence of a Kähler-Einstein metric (see \[27, p. 63\]).
The following normalisation of potentials is useful:
$$Q_G(X,\mathrm{\Delta },\omega )=\{\phi P_G(X,\mathrm{\Delta },\omega ):A_\omega (\phi )=0\}.$$
(29)
For any $`\phi P_G(X,\mathrm{\Delta },\omega )`$, $`\phi +A_\omega (\phi )Q_G(X,\mathrm{\Delta },\omega )`$.
###### Proposition 32
Let $`(X,\mathrm{\Delta })`$ be a Fano orbifold, $`\omega 2\pi \mathrm{c}_1(X,\mathrm{\Delta })`$ a Kähler metric and $`G`$ a compact group of isometries of $`(X,\mathrm{\Delta },\omega )`$. If there are constants $`C_1,C_2>0`$ such that
$$F_\omega (\phi )C_1\underset{X}{sup}\phi C_2$$
(30)
for any $`\phi Q_G(X,\mathrm{\Delta },\omega )`$, then $`(X,\mathrm{\Delta })`$ admits a Kähler-Einstein metric.
Proof. Let $`\phi _t`$ be a solution of $`()_t`$ on $`[0,T_0)`$. Since $`\phi _t+A_\omega (\phi _t)`$$``$$`Q_G(X,\mathrm{\Delta },\omega )`$
$$\begin{array}{c}F_\omega (\phi _t)=F_\omega \left(\phi _t+A_\omega (\phi _t)\right)\\ C_1\underset{X}{sup}\left(\phi _t+A_\omega (\phi _t)\right)C_2=C_1\underset{X}{sup}\phi _t+C_1A_\omega (\phi _t)C_2\end{array}$$
(31)
Using (28)
$$C_1\underset{X}{sup}\phi _tF_\omega (\phi _t)C_1A_\omega (\phi _t)+C_2C_3+C_2+C_1C_3.$$
Hence $`sup_X\phi _t`$ is uniformly bounded. But $`F^0(\phi _t)0`$, so $`J_\omega (\phi _t)F_\omega ^0(\phi _t)+sup\phi _t`$ is bounded and (26) yields the required bound of the $`C^0`$ norm. $`\mathrm{}`$
###### Lemma 33 (\[1, Lemma 2.3\])
Let $`(X,\mathrm{\Delta })`$ be a Fano orbifold, and $`\omega 2\pi \mathrm{c}_1(M)`$ a Kähler metric. Then for any $`\beta >0`$ there are constants $`C_1,C_2>0`$ such that for any $`\phi Q(X,\mathrm{\Delta },\omega )`$
$$\mathrm{log}\left[\frac{1}{V}_Xe^{(1+\beta )\phi }\omega ^n\right]C_1\underset{X}{sup}\phi C_2.$$
(32)
###### Corollary 34
If there are constants $`C_1,C_2>0`$ and $`\beta >0`$ such that
$$F_\omega (\phi )C_1\mathrm{log}\left[\frac{1}{V}_Xe^{(1+\beta )\phi }\omega ^n\right]C_2$$
(33)
for any $`\phi Q_G(X,\mathrm{\Delta },\omega )`$, then $`(X,\mathrm{\Delta })`$ admits a Kähler-Einstein metric.
## 6 Existence theorems
A *current* on an orbifold $`(X,\mathrm{\Delta })`$ is a collection of $`\mathrm{Gal}(\phi )`$-invariant currents on any uniformiser $`(U,\phi )`$, satisfying the usual compatibility condition with respect to injections of uniformisers. In case $`X`$ is smooth, orbifold differential forms on $`(X,\mathrm{\Delta })`$ are *more* than ordinary differential forms on $`X`$. By duality orbifold currents on $`(X,\mathrm{\Delta })`$ are *less* than ordinary currents on $`X`$: they are the continuous functionals on $`^k(X)`$ that can be extended to the larger space $`^k(X,\mathrm{\Delta })`$. For *positive* $`(p,p)`$-currents there is no difference between the two notions, since every positive current has measure coefficients, and every orbifold differential form has continuous coefficients. If $`\gamma `$ is a continuous hermitian form on a compact orbifold $`(X,\mathrm{\Delta })`$, an orbifold *Kähler current* is a closed positive (orbifold) current $`T`$ of bidegree (1,1) such that for some positive constant $`c`$, $`Tc\gamma `$ in the sense of orbifold currents, that is $`Tc\gamma ,\eta 0`$ for any positive $`\eta ^{n1,n1}(X,\mathrm{\Delta })`$. The definition does not depend on the choice of $`\gamma `$, since $`X`$ is compact.
If $`(X,\mathrm{\Delta })`$ is a Fano orbifold, $`G\mathrm{Aut}(X,\mathrm{\Delta })`$ is a compact subgroup and $`\omega `$ is a $`G`$-invariant Kähler form in $`2\pi \mathrm{c}_1(X,\mathrm{\Delta })`$, put
$$P_G^0(X,\mathrm{\Delta },\omega )=\{\chi C^0(X):\omega +\mathrm{i}\overline{}\chi \text{ is a Kähler orbifold current}\}.$$
###### Proposition 35
(a) Any $`\chi P_G^0(X,\mathrm{\Delta },\omega )`$ is the $`C^0`$-limit of a sequence $`\phi _nP_G(X,\mathrm{\Delta },\omega )`$. (b) The functionals $`I_\omega `$, $`J_\omega `$, $`F_\omega ^0`$ and $`F_\omega `$ can be extended to $`P_G^0(X,\mathrm{\Delta },\omega )`$ and the extensions are continuous with respect to the $`C^0`$-topology.
(See Prop. 2.2 and 2.3 in .)
###### Lemma 36 (\[1, Lemma 2.6\] )
If $`\pi :(X,\mathrm{\Delta }_X)(Y,\mathrm{\Delta }_Y)`$ is an orbifold map between compact orbifolds, the direct image $`\pi _{}T`$ of a Kähler current $`T`$ on $`(X,\mathrm{\Delta })`$ is a Kähler current on $`(Y,\mathrm{\Delta }_Y)`$.
Proof. First of all observe that if $`f:(X,\mathrm{\Delta }_X)(Y,\mathrm{\Delta }_Y)`$ is an orbifold map of degree $`d`$ and $`\alpha ^{2n}(Y,\mathrm{\Delta }_Y)`$, then $`_Xf^{}\alpha =d_Y\alpha .`$ Next let $`\gamma _X`$ and $`\gamma _Y`$ be continuous hermitian forms on $`(X,\mathrm{\Delta }_X)`$ and $`(Y,\mathrm{\Delta }_Y)`$ respectively. Since $`\pi ^{}\gamma _Y`$ is continuous and $`\gamma _X`$ is positive definite, there is $`c_1>0`$ such that $`\gamma _Xc_1\pi ^{}\gamma _Y`$. If $`T`$ is a Kähler current on $`(X,\mathrm{\Delta })`$, by definition $`Tc_2\gamma _X`$ for some $`c_2>0`$, so $`Tc\pi ^{}\gamma _Y`$ with $`c=c_1c_2>0`$. We want to prove that for any positive form $`\eta ^{n1,n1}(Y,\mathrm{\Delta }_Y)`$, $`\pi _{}T,\eta c\mathrm{deg}\pi \gamma _Y,\eta `$. Choose orbifold charts $`(V,\psi )`$ on $`(Y,\mathrm{\Delta }_Y)`$ and $`(U_i,\phi _i)`$ on $`(X,\mathrm{\Delta })`$ such that $`\pi ^1\left(\psi (V)\right)=_i\phi _i(U_i)`$. Denote by $`\stackrel{~}{T}_i`$, $`\stackrel{~}{\eta }`$ and $`\stackrel{~}{\gamma }_Y`$ the local representations in the orbifold charts and by $`\stackrel{~}{\pi }_i:U_iV`$ the liftings of $`\pi `$. We can assume $`\mathrm{supp}(\eta )\psi (V)`$. Then
$$\begin{array}{c}\pi _{}T,\eta =T,\pi ^{}\eta =\underset{i}{}\frac{\stackrel{~}{T}_i,\stackrel{~}{\pi }_i^{}\stackrel{~}{\eta }}{|\mathrm{Gal}(\phi _i)|}\underset{i}{}\frac{c\stackrel{~}{\pi }_i^{}\stackrel{~}{\gamma }_Y^{},\stackrel{~}{\pi }_i^{}\stackrel{~}{\eta }}{|\mathrm{Gal}(\phi _i)|}=\\ =\underset{i}{}\frac{c}{|\mathrm{Gal}(\phi _i)|}_{U_i}\stackrel{~}{\pi }_i^{}\left(\stackrel{~}{\gamma }_Y\stackrel{~}{\eta }\right)=c\left(\underset{i}{}\frac{\mathrm{deg}\stackrel{~}{\pi }_i}{|\mathrm{Gal}(\phi _i)|}\right)_V\left(\stackrel{~}{\gamma }_Y\stackrel{~}{\eta }\right).\end{array}$$
Since
$$\underset{i}{}\frac{\mathrm{deg}\stackrel{~}{\pi }_i}{|\mathrm{Gal}(\phi _i)|}=\frac{\mathrm{deg}\pi }{|\mathrm{Gal}(\psi )|}$$
we finally get
$$\pi _{}T,\eta c_{\psi (V)}\left(\gamma _Y\eta \right)$$
and this proves the lemma. $`\mathrm{}`$
###### Lemma 37 (\[1, Lemma 2.7\])
Let $`\pi :(X,\mathrm{\Delta })(Y,\mathrm{\Delta }_Y)`$ be an orbifold map between $`n`$-dimensional Kähler orbifolds. Let $`\omega _Y`$ be a Kähler metric on $`(Y,\mathrm{\Delta }_Y)`$ and $`\chi P^0(Y,\mathrm{\Delta }_Y,\omega _Y)`$ a *continuous* potential such that $`\pi ^{}\chi C^{\mathrm{}}(X,\mathrm{\Delta })`$. Then
$$F_{\pi ^{}\omega _Y}^0(\pi ^{}\chi )=F_{\omega _Y}^0(\chi ).$$
(34)
###### Theorem 38
Let $`(X,\mathrm{\Delta }_X)`$ and $`(Y,\mathrm{\Delta }_Y)`$ be Fano orbifolds, $`\pi :(X,\mathrm{\Delta })(Y,\mathrm{\Delta }_Y)`$ an orbifold Galois covering of degree $`d`$ with $`G=\mathrm{Gal}(\pi )`$, $`\omega _Y`$ a Kähler-Einstein metric on $`(Y,\mathrm{\Delta }_Y)`$ and $`\omega 2\pi \mathrm{c}_1(X,\mathrm{\Delta })`$ a $`G`$-invariant Kähler metric. Assume that numerically $`R^{orb}(\pi )\beta (K_X+\mathrm{\Delta }_X)`$ for some $`\beta _+`$. Then there is a constant $`C`$ such that for any $`\phi P_G(X,\mathrm{\Delta },\omega )`$
$$F_\omega ^0(\phi )\frac{1}{1+\beta }\mathrm{log}\left[\frac{1}{V}_Xe^{(1+\beta )\phi }\pi ^{}\omega _Y^n\right]C.$$
(35)
The proof is identical to that of Theorem 2.2 in and depends on the previous lemmata. Notice that a $`G`$-invariant orbifold Kähler metric $`\omega `$ always exists since, according to Definition 13, $`G\mathrm{Aut}(X,\mathrm{\Delta })`$.
###### Theorem 39
Let $`(X,\mathrm{\Delta })`$, $`(X_1,\mathrm{\Delta }_1)`$, $`\mathrm{},`$ $`(X_k,\mathrm{\Delta }_k)`$ be $`n`$-dimensional Fano orbifolds. Assume that each $`(X_i,\mathrm{\Delta }_i)`$ admits a Kähler-Einstein metric and that $`\pi _i:(X,\mathrm{\Delta })(X_i,\mathrm{\Delta }_i)`$ are orbifold Galois coverings such that
1. the groups $`\mathrm{Gal}(\pi _i)`$ are all contained in some *compact* subgroup of $`\mathrm{Aut}(X,\mathrm{\Delta })`$;
2. $`R^{orb}(\pi _i)\beta _i(K_X+\mathrm{\Delta })`$ for some $`\beta _i_+`$.
Define $`\eta C^{\mathrm{}}(X,\mathrm{\Delta })`$ by
$$\frac{1}{k}\underset{i=1}{\overset{k}{}}\pi _i^{}\omega _i^n=\eta \omega ^n,$$
(36)
put $`c:=sup\{\lambda 0:\eta ^\lambda L^1(X,\omega ^n)\}`$ and $`\beta :=\mathrm{min}\beta _i`$. If
$$\frac{1}{c}<\beta ,$$
(37)
then $`(X,\mathrm{\Delta })`$ admits a Kähler-Einstein metric.
The proof is the same as that of Theorem 2.3 and Proposition 2.4 in .
###### Remark 40
If there is only one covering ($`k=1`$) and $`X`$ is smooth, then $`c`$ is simply the *complex singularity exponent* (that is the *log canonical threshold*) of the pair $`(X,R^{orb})`$ (see and ). On the other hand if there are enough coverings and the intersection of the ramification divisors $`R^{orb}(\pi _i)`$ is empty, then $`c=+\mathrm{}`$ and (37) is automatically satisfied.
## 7 Applications
Here we exhibit some concrete examples where Theorem 39 can be used to prove the existence of Kähler-Einstein metrics on orbifolds.
###### Theorem 41
Let $`X`$ be a Fano manifold, $`_{i=1}^ND_i`$ a divisor with local normal crossing and $`\omega `$ a Kähler-Einstein metric on $`X`$. Given integers $`m_i>1`$ put $`\mathrm{\Delta }=_i(11/m_i)D_i`$. If $`\mathrm{\Delta }\delta K_X`$ with $`\delta (0,1)`$ and
$$m_i1<\frac{\delta }{1\delta }$$
(38)
for any $`i=1,\mathrm{},N`$, then $`(X,\mathrm{\Delta })`$ is a Fano orbifold and has an orbifold Kähler-Einstein metric.
Proof. $`(X,\mathrm{\Delta })`$ is a Fano orbifold because $`K_X+\mathrm{\Delta }=(1\delta )K_X`$ and $`\delta <1`$. As observed in Example 15 the map $`\mathrm{id}:(X,\mathrm{\Delta })X`$ is an orbifold Galois cover and we want to apply Proposition 39 to it. The ramification divisor is just $`R^{orb}=\mathrm{\Delta }`$ so
$$R^{orb}(\mathrm{id})=\beta (K_X+\mathrm{\Delta })$$
with $`\beta =\delta /(1\delta ).`$ It remains to check that (38) implies (37). Let $`x`$ be any point in $`X`$. Choose a system of coordinates $`(V,z^1,\mathrm{},z^n)`$ on $`X`$ as in Example 11 and let $`(U,\phi )`$ be the corresponding orbifold chart for $`(X,\mathrm{\Delta })`$ as in (6). Then on $`\phi (U)=V`$
$$R^{orb}=\mathrm{\Delta }=\underset{j=1}{\overset{k}{}}\left(1\frac{1}{m_j^{}}\right)\{z_j=0\}$$
(39)
so that in the notation of (36), $`\eta (z)=\gamma (z)|f(z)|^2`$ on $`U`$, where $`f(z)=z_1^{m_11}\mathrm{}z_k^{m_k1}`$ and $`\gamma `$ is a smooth positive function. Set $`c_x=sup\{\lambda 0:_U|f|^{2\lambda }<+\mathrm{}\}.`$ Since
$$_U|f|^{2\lambda }=\mathrm{const}\underset{j=1}{\overset{k}{}}_D|z|^{2\lambda (m_j^{}1)}$$
(40)
where $`D`$ is the disk in $``$, we get that $`|f|^{2\lambda }L_{loc}^1`$ on $`U`$ iff $`\lambda <1/(m_j^{}1).`$ So $`c_x=\mathrm{min}\{1/(m_j^{}1):1jk\}`$,
$$c=\underset{xX}{sup}c_x=\underset{i}{\mathrm{min}}\frac{1}{m_i1}$$
(41)
and
$$\frac{1}{c}=\mathrm{max}(m_i1)<\frac{\delta }{1\delta }=\beta .$$
(42)
$`\mathrm{}`$
###### Example 42
Let some divisors $`D_i|𝒪_^n(d_i)|`$, and some integers $`m_i>1`$ be given for $`i=1,\mathrm{},N`$. Let $`m_1`$ be the greatest of the $`m_i`$’s. Put $`\mathrm{\Delta }=_i(11/m_i)D_i`$ and
$$\delta =\frac{_id_i\left(1\frac{1}{m_i}\right)}{n+1}.$$
(43)
Assume that
1. $`_iD_i`$ is local normal crossing;
2. $`\delta <1`$;
3. $`m_1(1\delta )<1`$.
Then the orbifold $`(X,\mathrm{\Delta })=(X,\mathrm{\Delta })`$ admits an orbifold Kähler-Einstein metric of positive scalar curvature.
###### Example 43 (Compare \[10, Note 36\])
Let $`D_i`$ be $`n+2`$ hyperplanes in general position in $`^n`$: $`D_i=\{z_i=0\}`$ for $`i=0,\mathrm{},n,`$ $`D_{n+1}=\{z_0+\mathrm{}+z_n=0\}`$. Set
$$\mathrm{\Delta }=\underset{i=0}{\overset{n+1}{}}(1\frac{1}{m_i})D_i.$$
Then $`(^n,\mathrm{\Delta })`$ has an orbifold Kähler-Einstein metric as soon as
$$1<\underset{i=0}{\overset{n+1}{}}\frac{1}{m_i}<1+(n+1)\underset{i}{\mathrm{min}}\frac{1}{m_i}$$
(44)
As in , many numerical examples come from Euclid’s or Sylvester’s sequence (cf. \[23, A000058\]). This is defined by the recursion relation
$$c_{k+1}=c_1\mathrm{}c_k+1=c_k^2c_k+1$$
beginning with $`c_1=2`$. The sequence grows doubly exponentially, and it starts as
$$2,3,7,43,1807,3263443,10650056950807,\mathrm{}$$
It is easy to see that
$$\underset{i=1}{\overset{n}{}}\frac{1}{c_i}=1\frac{1}{c_{n+1}1}=1\frac{1}{c_1\mathrm{}c_n}.$$
We get many new examples by taking
$$(m_0=c_1,m_1=c_2,\mathrm{},m_n=c_{n+1}2,m_{n+1}).$$
Then
$$\underset{i=0}{\overset{n}{}}\frac{1}{m_i}=1+\frac{1}{(c_{n+1}1)(c_{n+2}2)}.$$
Thus our conditions are satisfied as long as
$$c_{n+1}2<m_{n+1}<n(c_{n+1}1)(c_{n+2}2)$$
and $`m_{n+1}`$ is relatively prime to the other $`m_i`$.
Another case when Theorem 39 works is for degree 2 Del Pezzo surfaces $`S`$. Here we consider the case when $`S`$ is allowed to have cyclic quotient singularities. These are necessarily of the form $`^2/_n`$ where the group action is given by $`(u,v)(ϵu,ϵ^1v)`$ where $`ϵ`$ is a primitive $`n`$-th root of unity. The $`_n`$-invariant fuctions are generated by $`u^n,v^n,uv`$. This singularity is denoted by $`A_{n1}`$.
For any degree 2 Del Pezzo surface $`S`$ the anticanonical class is ample and it gives a degree 2 cover $`\pi :S^2`$. If $`H`$ denotes the hyperplane class on $`^2`$, then $`K_S=\pi ^{}H`$. The double cover $`\pi `$ ramifies along a quartic curve $`C`$, thus $`R=\frac{1}{2}\pi ^{}C=\pi ^{}2H`$, $`\beta =2`$ and to apply Theorem 39 we need to ensure that $`\eta ^\lambda `$ be integrable for $`\lambda \frac{1}{2}`$. The singularities of $`\pi `$ lie over the singularities of $`C`$, an $`A_{n1}`$–singularity of $`S`$ lies over an $`A_{n1}`$–singularity of $`C`$ (cf. \[6, p.87\]) and we can find local coordinates $`(x,y)`$ on $`^2`$ such that $`S`$ is locally isomorphic to some neighbourhood of the origin to the affine surface $`\{(x,y,t)^3:t^2=x^2+4y^n\}`$, the map $`\pi `$ being given simply by $`\pi (x,y,t)=(x,y)`$. An orbifold chart is given by $`\phi :U^2S`$ where $`\phi (u,v)=(u^nv^n,uv,u^n+v^n)`$. Thus $`\phi ^{}\pi ^{}(dxdy)=n(u^n+v^n)dudv`$ and $`\eta (u,v)=\text{const}|u^n+v^n|^2`$. It is easy to see by direct integration or by blowing up (see e.g. \[18, Prop. 6.39 p. 168\]) that for $`n2`$, $`|u^n+v^n|^{2\lambda }`$ is integrable if and only if $`\lambda <\frac{2}{n}`$. Thus Theorem 39 applies as long as $`\frac{1}{2}<c=\frac{2}{n}`$, that is for $`n<4`$. This proves Theorem 2.
One can also give a different proof of the following result of Mabuchi and Mukai \[19, Corollary C\].
###### Theorem 44
A diagonalizable singular Del Pezzo surface of degree 4 admits an orbifold Kähler-Einstein metric.
A quartic Del Pezzo surface $`S`$ is the intersection of two quadrics in $`^4`$, $`S=Q_1Q_2`$. It is said to be diagonalizable if both $`Q_1`$ and $`Q_2`$ can be put simultaneously in diagonal form. If $`S`$ is singular then in suitable coordinates it is given by equations
$$h_0:=x_0^2+x_1^2+x_2^2+x_3^2+x_4^2=0\text{and}h_1:=\lambda _2x_2^2+\lambda _3x_3^2+\lambda _4x_4^2=0$$
If two of the $`\lambda _i`$ coincide then $`S`$ is a quotient of $`^1\times ^1`$ and so has an orbifold Kähler-Einstein metric (see \[19, p.136\]). Thus assume that the $`\lambda _i`$ are distinct nonzero complex numbers. For $`i=2,3,4`$, the equation $`\lambda _ih_0h_1=0`$ does not involve $`x_i`$, and by dropping the $`x_i`$ variable we get smooth quadrics
$$Q_i=\{(\lambda _ih_0h_i=0)\}^3.$$
The map $`\pi _i:SQ_i`$ given by forgetting $`x_i`$ is a double cover ramified over the hyperplane section $`S\{x_i=0\}`$. Since the $`Q_i`$ are smooth two-dimensional quadrics, they are Kähler-Einstein. On the other hand, the divisors $`R^{orb}(\pi _i)`$ are disjoint, so $`\eta `$ is strictly positive on all $`S`$, $`c=\mathrm{}`$ and Theorem 39 yields that $`S`$ admits an orbifold Kähler-Einstein metric.
###### Acknowledgments
We thank C. Arezzo, Ch. Boyer, K. Galicki and G.P. Pirola for useful comments. J.K. was partially supported by the NSF under grant number DMS-0200883.
Università di Milano Bicocca
```
alessandro.ghigi@unimib.it
```
Princeton University, Princeton NJ 08544-1000
```
kollar@math.princeton.edu
``` |
warning/0507/quant-ph0507022.html | ar5iv | text | # From Dirac to Diffusion: Decoherence in Quantum Lattice Gases
## I Introduction
Lattice gases are arguably the simplest models for the simulation of classical physical systems. These models provide elementary microscopic dynamics whose hydrodynamic limits are, inter alia, the diffusion equation, Burgers’ equation and the Navier Stokes equations Doolen et al. (1990). They also provide a simple arena for the creation of new models of physical phenomena such as multicomponent flow and dynamical geometry Boghosian et al. (2000); Love (2002); Love et al. (2004); Hasslacher and Meyer (1998). Lattice gases possess deterministic, stochastic and quantum (unitary) formulations.
Simulation of quantum systems on quantum computers remains one of the very few applications for which exponential speedup over classical computation is provable. The quantum lattice gases defined by Meyer Meyer (2002) and by Boghosian and Taylor Boghosian and Taylor (1998a) may be simulated on a quantum computer in exponentially fewer steps than are required on a classical computer and with an exponential reduction in hardware. These models, together with other approaches by Lloyd Lloyd (1996); Abrams and Lloyd (1997) and by Ortiz and Gubernatis Ortiz et al. (2002, 2001), have made concrete Feynman’s observation that simulation of quantum systems is hard on classical computers, but easy on quantum computers Feynman (1984, 1986, 1982).
Meyer was led to the definition of the quantum lattice gas by consideration of quantum generalizations of cellular automata (CA). The simplest possible quantum CA model would be a map in which the updated state of a cell depended linearly on the state of its neighbours and the global evolution rule was unitary. Meyer showed that the only such maps are trivial, namely, the identity map and the left or right shift, possibly multiplied by a phase. This “No-Go” theorem proves that there are no nontrivial homogeneous scalar unitary cellular automata Meyer (1996a).
Quantum cellular automata models may evade this No-Go theorem Meyer (1996a) by having cell values which are not scalar, or by having local update rules which are not linear (although the global evolution in such models remains quantum mechanical, and therefore linear). Non-scalar models sacrifice some simplicity, while the determination of unitarity for nonlinear models is problematic Durr (1997, 2002); Meyer (1996b). A third possibility exists: We may partition our cellular automata, dividing our evolution into two substeps, acting on two distinct neighborhoods. This was the approach taken by Meyer Meyer (1996c) and also by Watrous Watrous (1995). If one of the substeps of the evolution is interpreted as propagation of the cell values to neighboring sites we may identify the partitioned cellular automata with a lattice gas model Meyer (1996c).
The dynamics of all lattice gases take place by propagation of particles to neighboring sites on the lattice, followed by a local collision operation. In a stochastic lattice gas which obeys semi-detailed balance, the collision step is a doubly-stochastic Markov matrix acting on the state of a single lattice site. For a lattice gas with only a single particle moving on the lattice, the stochastic model reduces to a classical random walk. In the collision step of a quantum lattice-gas model, the state at a site is acted on by a unitary scattering matrix. Just as in the classical case, the one-particle sector of the model possesses an interpretation as a discrete-time discrete-space quantum random walk.
Quantum lattice gases are explicitly formulated as discrete models for quantum physics, and so are subject to additional physical constraints. In particular, the unitary scattering matrix is constrained to be parity invariant. The quantum lattice gas was shown to yield the continuum propagator for the Dirac equation in one dimension Meyer (1996c), and also possesses the Schroedinger equation as a nonrelativistic continuum limit Boghosian and Taylor (1998b).
The quantum random walk has also been studied extensively from a point of view quite different from that of physical modeling. Quantum random walks may provide an alternative route for the development of new quantum algorithms. Such walks have the property that the variance of the walk grows linearly with time, in sharp contrast to classical walks where the variance grows as the square root of time. This echoes the computational advantage of Grover’s search algorithm, and indeed unstructured search can be reformulated as a quantum random walk problem. Discrete-time quantum walks have yielded exponential performance improvements in the hitting time on the hypercube, and continuous-time quantum walks have yielded an exponential improvement in the solution of a graph traversal problem. For an overview of the subject of quantum random walks we refer the reader to the review of Kempe Kempe (2003).
Two things distinguish quantum lattice gases from discrete quantum walks. Firstly, the absence of scalar homogeneous models means that discrete quantum walks must also introduce an extra degree of freedom. This degree of freedom is interpreted as a “coin” state which determines the motion of the walker at the next time step. In the context of the quantum lattice gas this degree of freedom is interpreted as the spin (or, in one dimension, helicity) degree of freedom of the particle. Secondly, the constraint of parity invariance imposed by Meyer in one dimension, and discrete rotation invariance imposed by Boghosian and Taylor in $`d`$-dimensions is infrequently applied to discrete time quantum random walks. The coin state is commonly updated by the Hadamard operation, which is not invariant under parity inversion. While parity inversion symmetry is a fundamental requirement for models of physics, from the point of view of the computational properties of quantum walks parity invariance is not an obvious requirement.
Unitary actions and measurements are the elementary operations allowed on closed quantum systems. However, no quantum system (excepting, possibly, the entire universe) is truly closed. Open quantum systems may be treated as subsystems of some larger closed quantum system. The unitary evolution of the entire closed system induces a (generally non-unitary) evolution on the open quantum subsystem. Such operations, which include measurements and unitary actions in their span, are referred to as quantum operations in the context of quantum information theory Nielsen and Chuang (2000). The theory of open quantum systems has been developed from the point of view of fundamental physics in the work of Feynman and Vernon, Caldeira and Leggett and Prokof’ev and Stamp Feynman and Vernon (1963); Caldeira and Leggett (1983); Prokof’ev and Stamp (2000); Weiss (1993). In this seminal work the environment degrees of freedom are included explicitly, and renormalization group arguments are adduced to model the environment as belonging to one of two universality classes: either the environment degrees of freedom are localized (spin bath) or delocalized (oscillator bath). “Random level” models of the environment have also been studied Wigner (1953); Hill and Wheeler (1953).
In keeping with the motivation of this issue we shall leave the fascinating computational properties of both discrete-time and continuous-time quantum random walks aside. We shall focus instead on the physical interpretation of the classical stochastic lattice gas as a microscopic model for diffusion and of the quantum lattice gas as a microscopic model for a single particle obeying the Dirac equation in the continuum limit. Our aim in the present paper will be the construction of a single model capable of capturing both types of behavior in different parameter regimes.
We shall consider decoherence arising from an interaction with an environment coupling only to the particles’ internal degrees of freedom and not to the particles’ position degrees of freedom. Such decoherence models are referred to as “coin” decoherence in the context of quantum random walks. In general, coin decoherence produces a transition from behavior characteristic of a quantum walk in which the standard deviation of the walk grows linearly in time to behavior characteristic of a classical walk in which the standard deviation varies as the square root of time Alagic and Russell (2005); Kendon and Tregenna (2002a, b); Kendon and Tregenna (2003); Alagic and Russell (2005); Brun et al. (2002a, b). Previous decoherence models for quantum walks considered a scattering step which is a unitary action with probability $`1p`$, and a unitary action followed by a projective measurement with probability $`p`$ Alagic and Russell (2005); Kendon and Tregenna (2002a, b); Kendon and Tregenna (2003). Models in which measurements on the coin yield less than total information have also been considered Kendon and Sanders (2004).
We begin by considering a decoherence model for quantum lattice gases (or equivalently, discrete-time quantum random walks) with an explicitly physical motivation. We require that the system-environment interaction be fixed and that the interaction preserve the parity invariance of the original lattice gas. We first describe the unitary lattice-gas dynamics and the extension of such dynamics to the density matrix formalism, necessary for the introduction of quantum operations into such dynamics. We then introduce the framework of quantum operations, and discuss two types of decoherence models for quantum lattice gases. We derive the constraints on the unitary operator coupling the system and environment arising from the requirement of parity invariance. Numerical results are presented for one parameterization of the decoherence model, which qualitatively verify the transition from quantum to classical (i.e., diffusive) behavior in the model. We close the paper with discussion and some directions for future work.
## II Decoherence
The state vector formulation of quantum mechanics is inadequate to describe situations in which we have imperfect knowledge of the quantum state. Such perfect knowledge is expressed by the system being in a pure state, that is, a vector of complex amplitudes whose moduli squared sum to one. Consider the imperfect preparation of a quantum state such that state $`|\psi `$ is prepared with probability $`P(|\psi )`$, where the (classical) probabilities $`P(|\psi )`$ sum to one. We cannot simply represent this by a real linear convex combination of states, as we already have a complex superposition over observable basis states. We must instead consider a linear real convex combination of the dyads $`|\psi \psi |`$,
$$\rho =P(\psi )|\psi \psi |.$$
(1)
Such a real linear convex combination is called a mixed state. The density matrix is the appropriate tool for simultaneously describing two probabilistic aspects of the theory – one arising from classical uncertainty about the state of the system, and one arising from the fundamental uncertainty arising from quantum superposition.
The time evolution of the density matrix may be obtained by linear extension of the evolution of the pure states. The time evolution of a pure state is given by $`|\psi ^{}=U|\psi `$ where $`U`$ is the unitary evolution operator of the system, and so the time evolution of the density matrix is given by conjugation: $`U`$: $`\rho ^{}=U\rho U^{}`$, where dagger indicates the Hermitian conjugate. The Hilbert space of a quantum system which can be divided into two subsystems $`A`$ and $`B`$ possesses a basis which can be tensor factored such that each basis vector $`|m`$ can be expressed as a tensor product $`|m_a|m_b`$, such that $`|m_aH_a`$ and $`|m_bH_b`$. The reduced density matrix of subsystem $`A`$ is obtained by taking the partial trace of the full density matrix over subsystem $`B`$.
We may now define a quantum operation on the density matrix of system $`A`$. We take the tensor product of the density matrix of the system $`A`$ with that of the environment $`B`$. A unitary operation $`U^{AB}`$ acts on the resulting density matrix by conjugation. The environmental subsystem $`B`$ is then traced over, resulting in a new density matrix for the system $`A`$. Such quantum operations are therefore maps from density matrices to density matrices. Such maps may be constructed without reference to an environment state by invoking the operator-sum representation. The theory of such maps may also be formulated axiomatically without reference to the constructive procedure adduced here Nielsen and Chuang (2000). In the present paper we utilize the unitary representation of quantum operations given above, while noting that a formulation of the noise model we shall construct in terms of the operator-sum representation is possible, and in fact may be a more convenient representation of the model.
## III Quantum Lattice-Gas Model
In the present paper we restrict attention to one dimensional lattice gases with two directions per site. The (classical) particle states are specified as follows: Each site on the lattice has two lattice vectors connecting it to its left and right neighbors. There may be at most one particle per site per vector <sup>2</sup><sup>2</sup>2This is sometimes called the “exclusion principle” for lattice gases, but it should be noted that it is unrelated to the Pauli exclusion principle.. The dynamics of all lattice gases take place in two substeps. First, the particles propagate along their vectors to neighboring sites, retaining their velocity as they do so. Second, the particles at each site undergo a collision changing the occupations of the vectors at each site. The propagation step clearly conserves any quantity that is obtained by summing a function of particle mass and velocity over all particles, since those quantities are not changed as particles propagate. The collision step is required to conserve a subset of these quantities that are of physical interest, such as mass, momentum, etc. In the following we shall consider models which conserve particle number only; if we regard the particles as each having unit mass, this may be thought of as conservation of mass.
For a one dimensional lattice with two vectors per site, the only deterministic rules which preserve particle number are trivial. The first nontrivial model occurs when one considers a stochastic collision in which a single particle at a site has probability $`1p`$ to reverse direction. The stochastic lattice gas with a single particle may be identified with a random walk. Generalizing to multiple particles we find the evolution of the single particle distribution function for a classical stochastic lattice gas obeys the diffusion equation. The stochastic models include the (trivial) deterministic models as the special case $`p=1`$.
The single timestep evolution operator $`U`$ of the quantum lattice gas without decoherence is the composition of advection and scattering steps:
$$\begin{array}{cc}\hfill \psi _{x,\alpha }|x,\alpha \underset{}{\mathrm{advect}}& \psi _{x,\alpha }|x+\alpha ,\alpha \hfill \\ \hfill \underset{}{\mathrm{scatter}}& \psi _{x,\alpha }S_{\alpha \alpha ^{}}|x+\alpha ,\alpha ^{},\hfill \end{array}$$
(2)
where the scattering matrix may be parameterized up to a global phase as:
$$S=\left(\begin{array}{cc}\mathrm{cos}\theta & i\mathrm{sin}\theta \\ i\mathrm{sin}\theta & \mathrm{cos}\theta \end{array}\right)$$
(3)
and the Hilbert space is $`2N`$ dimensional, where $`N`$ is the lattice size. This quantum lattice gas was shown to yield the continuum propagator for the Dirac equation for a particle with mass $`\mathrm{tan}\theta `$ in Meyer (1996c). The model may also be interpreted as a discrete-space discrete-time quantum random walk, subject to the constraint that the scattering rule be parity invariant.
We generalize the above model by first extending the dynamics to those of the density matrix for the lattice gas. This allows us to handle the mixed states which will arise when we introduce decoherence into the dynamics. We couple the particles’ internal degrees of freedom to a bath of arbitrary size and act on the internal degrees of freedom and the bath with a unitary matrix which is the product:
$$U_c\left[S𝕀_{bath}\right],$$
(4)
where $`𝕀`$ is the identity operator on the bath. We then trace over the environment degrees of freedom at each timestep.
This decoherence model corresponds to a generalization of the quantum lattice gas to the case where the collision operator is neither a unitary operator, nor a classical Markov matrix as in the stochastic classical lattice gas, but a quantum operation. The set of quantum operations contains both unitary actions and Markov actions as special cases. The unitary operator is clearly included as a special case when one considers $`U_c`$ which do not couple the system and the environment (i.e., which tensor factor into $`U_c=U_{c(sys)}U_{c(bath)}`$).
It is clear that the Markov operations of the classical stochastic lattice gas are included when one considers an initialization of the density matrix of the lattice gas in a completely classical state – that is, a density matrix at each site with only diagonal entries, $`p_l`$ and $`p_r`$. The action of the unitary part of the collision operator on this matrix is:
$$\left(\begin{array}{cc}\mathrm{cos}\theta & i\mathrm{sin}\theta \\ i\mathrm{sin}\theta & \mathrm{cos}\theta \end{array}\right)\left(\begin{array}{cc}p_r& 0\\ 0& p_l\end{array}\right)\left(\begin{array}{cc}\mathrm{cos}\theta & i\mathrm{sin}\theta \\ i\mathrm{sin}\theta & \mathrm{cos}\theta \end{array}\right)=\left(\begin{array}{cc}p_l\mathrm{cos}^2\theta +p_r\mathrm{sin}^2\theta & i\mathrm{sin}\theta \mathrm{cos}\theta (p_rp_l)\\ i\mathrm{sin}\theta \mathrm{cos}\theta (p_lp_r)& p_l\mathrm{sin}^2\theta +p_r\mathrm{cos}^2\theta \end{array}\right)$$
(5)
This reproduces the action of a classical Markov matrix on the vector of probabilities $`(p_l,p_r)`$, where the probability that the particle continues in its current state is $`\mathrm{cos}^2\theta `$ if the interaction with the environment induces the map:
$$\left(\begin{array}{cc}p_l\mathrm{cos}^2\theta +p_r\mathrm{sin}^2\theta & i\mathrm{sin}\theta \mathrm{cos}\theta (p_rp_l)\\ i\mathrm{sin}\theta \mathrm{cos}\theta (p_lp_r)& p_l\mathrm{sin}^2\theta +p_r\mathrm{cos}^2\theta \end{array}\right)\left(\begin{array}{cc}p_l\mathrm{cos}^2\theta +p_r\mathrm{sin}^2\theta & 0\\ 0& p_l\mathrm{sin}^2\theta +p_r\mathrm{cos}^2\theta \end{array}\right).$$
(6)
We can construct this mapping by coupling the system to a two-dimensional environment in the completely mixed state $`1/2𝕀`$ using the controlled-NOT gate where the system is the control qubit. More generally, the map given in Eq. (6) is the effect of a measurement on the system, and such actions are included in the set of quantum operations. The correspondence of unitary operations combined with measurements as equivalent to Markov operations is also discussed in the context of Type-II quantum computing in Love and Boghosian (2005).
In order to complete the demonstration that classical stochastic lattice gases are included as a subset of models described by a density matrix whose collision process is a quantum operation, we must show that a classical (i.e., diagonal) density matrix evolves to another classical density matrix under propagation. The classical density matrix is non-zero only in entries $`|x,\alpha x,\alpha |`$, which evolve under propagation to $`|x+\alpha ,\alpha x+\alpha ,\alpha |`$. Diagonal density matrices evolve to diagonal density matrices under propagation and a subset of collision operations evolving diagonal density matrices to diagonal density matrices are equivalent to the Markov matices implementing collisions of a classical stochastic lattice gas.
Hence a single-particle quantum lattice-gas simulation in which the entire density matrix is stored and in which the collision rule is a quantum operation includes as special cases quantum lattice-gas evolution and the stochastic evolution of a classical lattice gas. Deterministic classical lattice gases are included as they are a special case of stochastic lattice gases. Such a model therefore provides a bridge between the quantum lattice gas which possesses the Dirac equation as a continuum limit and classical stochastic lattice gases which possess the diffusion equation as a continuum limit.
## IV Parity-Preserving Noise
We now define the set of quantum operations giving our new scattering rule. Specification of a set of quantum operations defines a “quantum noise” model. Such a model has two ingredients: a model for the environment state and a model for the system-environment interaction. Before specifying the environment state we must specify how many dimensions the environment Hilbert space must have. Here we may invoke a theorem which states that for a $`d`$-dimensional system Hilbert space a $`d^2`$-dimensional environment is sufficient to produce every possible quantum operation on the system. This theorem states that for a quantum operation specified by a set of principal components in the operator-sum representation it is always possible to find a unitary operator coupling a $`d^2`$-dimensional environment to a $`d`$-dimensional system which reproduces this quantum operation Nielsen and Chuang (2000).
This suggests the following noise model: We initialize the four-dimensional environment in a fiducial state ($`|00`$ for example). We then sample from a distribution over the unitary group $`U(8)`$ and apply our sampled operator to the eight-dimensional system-environment pair, and then we trace over the environment. Such a model samples from the entire set of quantum operations. The preparation of the environment in a fiducial state does not imply a loss of generality, as the sampled operation can be considered to be first an operator acting only on the environment preparing a random environment state, followed by an operation coupling the system and environment. The distribution over the set of quantum operations is induced in a nontrivial way by the distribution over $`U(8)`$. Such a model would resemble a random level, or random matrix model, of the environment Wigner (1953); Hill and Wheeler (1953).
On the other hand, our aim is the construction of a noise model with physically inspired constraints on the system-environment interaction. For such physically motivated noise models we wish the system-environment coupling matrix to be a constant unitary operator, arising from a putative fixed system-environment interaction Hamiltonian (which we may or may not know). If we regard our quantum lattice gas as a discrete model for the Dirac equation, we note that the interaction of the helicity degree of freedom of a Dirac particle with an environment is indeed fixed by fundamental physics.
We must take care about the meaning of the theorem invoked above for noise models with a constant system-environment interaction. The theorem does not state that a fixed unitary operator coupling a $`d^2`$-dimensional environment to a $`d`$-dimensional system can reproduce every quantum operation on the system. In the sequel we construct our noise model for an environment of arbitrary dimension, although we revert to a four-dimensional environment for reasons of computational tractability for simulations.
Equation (3) gives an explicit parameterization of all parity-preserving two-dimensional unitary operators, up to a global phase. Such convenient parameterizations of quantum operations do not yet exist, and so we follow the explicit constructive procedure for such operations given above. We choose a model for the environment such that its state is a unimodular complex vector whose Cartesian components are independent random variables. The environment-system interaction is fixed, and we consider a wide class of such interactions, namely those which preserve the parity invariance of the original unitary quantum lattice gas. The unitary update $`S`$ obeys parity invariance $`ST=TS`$ where $`T`$ is the parity exchange operator. The parity preserving couplings $`U_s`$ have the property that they commute with the parity exchange operator acting on the system tensored with the identity operator acting on the environment.
$$[U_s,T𝕀]=0$$
(7)
Eq. (7) expresses a discrete symmetry of a discrete-time evolution operator. In physics we are more usually concerned with continuous symmetries and continuous time evolution operators. The usual statement of invariance of an interaction Hamiltonian under a particular symmetry transformation is that the infinitesimal generators of the symmetry transformation commute with the Hamiltonian. In the language of Lie groups this means that the Hamiltonian lies in the commutator subalgebra of the generators of the symmetry transformation in the Lie algebra of $`U(N)`$, where $`N`$ is the number of degrees of freedom of our system. We may also apply these ideas to a discrete symmetry of a discrete-time discrete-space model. Let $`t`$ be the Lie algebra element corresponding to $`T𝕀`$. Let $`u`$ be the Lie algebra element corresponding to $`U`$. A sufficient condition that Eq. (7) holds is:
$$[u,t]=0.$$
(8)
The Lie algebra of $`U(N)`$ is the set of anti-hermitian matrices. We use the basis arising from the root system of the Lie algebra of $`U(N)`$ Brocker and tomDieck (1985):
$$\begin{array}{cc}\hfill D_{kl}^p& =i\delta _{pk}\delta _{pl}1pN\hfill \\ \hfill S_{kl}^{qp}& =i(\delta _{kp}\delta _{ql}+\delta _{kq}\delta _{pl})1pNq<p\hfill \\ \hfill A_{kl}^{qp}& =(\delta _{kp}\delta _{ql}\delta _{kq}\delta _{pl})1pNq<p\hfill \end{array}$$
(9)
where $`\delta _{xy}`$ is the Kronecker delta, and there is no sum on repeated indices. We note that $`A^{qq}=0`$ and $`S^{qq}=2D^q`$. The convention for the antisymmetric matrices is chosen so that the labelling superscripts increase from left to right, and so that the negative entry is always in the upper triangular portion of the matrix.
The commutation relations of the Lie algebra follow from 9:
$$\begin{array}{cc}\hfill [D^p,D^q]_{kl}& =01pN\hfill \\ \hfill [D^r,S^{qp}]_{kl}& =\delta _{pr}A^{qr}+\delta _{rq}A^{pr}\hfill \\ \hfill [D^r,A^{qp}]_{kl}& =\delta _{rp}S^{rq}\delta _{rq}S^{rp}\hfill \\ \hfill [S^{rs},A^{qp}]_{kl}& =\delta _{sp}S^{rq}\delta _{sq}S^{rp}+\delta _{rp}S^{sq}\delta _{rq}S^{sp}\hfill \\ \hfill [S^{rs},S^{qp}]_{kl}& =\delta _{ps}A^{qr}+\delta _{qs}A^{pr}+\delta _{pr}A^{qs}+\delta _{qr}A^{ps}\hfill \\ \hfill [A^{rs},A^{qp}]_{kl}& =\delta _{ps}A^{qr}+\delta _{qs}A^{pr}+\delta _{pr}A^{qs}+\delta _{qr}A^{ps}\hfill \end{array}$$
(10)
The block diagonal form of $`T𝕀`$ makes it straightforward to diagonalize, and it is therefore straightforward to obtain the Lie algebra element $`t`$.
$$t=\mathrm{ln}T1=\frac{\pi }{2}\underset{r=1}{\overset{N}{}}D^r\frac{\pi }{2}\underset{s=1}{\overset{N/2}{}}S^{(2s1)2s}$$
(11)
A general element $`u`$ of the Lie algebra may be written:
$$u=\underset{r=1}{\overset{N}{}}\alpha _rD^r+\frac{1}{2}\underset{p=1}{\overset{N}{}}\underset{q=1}{\overset{N}{}}\left[\beta _{qp}S^{qp}+\gamma _{qp}A^{qp}\right]$$
(12)
where $`\beta _{qp}=\beta _{pq}`$ and $`\gamma _{qp}=\gamma _{pq}`$, and $`\beta _{qq}=\gamma _{qq}=0`$.
The commutator is then
$$\begin{array}{cc}\hfill [u,t]& =\frac{\pi }{2}[\underset{s=1}{\overset{N}{}}D^s\underset{s=1}{\overset{N/2}{}}S^{(2s1)2s},\underset{r=1}{\overset{N}{}}\alpha _rD^r+\frac{1}{2}\underset{p=1}{\overset{N}{}}\underset{q=1}{\overset{N}{}}\left[\beta _{qp}S^{qp}+\gamma _{qp}A^{qp}\right]]\hfill \end{array}$$
(13)
Applying the structure constants of the Lie algebra and using $`\beta _{qp}=\beta _{pq}`$ and $`\gamma _{qp}=\gamma _{pq}`$ gives:
$$\begin{array}{cc}\hfill [u,t]& =\frac{\pi }{2}\underset{s=1}{\overset{N/2}{}}\underset{p=1}{\overset{N}{}}\left[\beta _{p(2s)}A^{p(2s1)}+\beta _{p(2s1)}A^{p(2s)}+\gamma _{p(2s)}S^{(2s1)p}+\gamma _{p(2s1)}S^{(2s)p}\right]\hfill \end{array}$$
(14)
The constraint that this be zero imposes a set of constraints on the $`\beta `$ coefficients and a set of constraints on the $`\gamma `$ coefficients. Because the coefficients and the matrices $`A`$ are real, while the matrices $`S`$ are pure imaginary, we may rearrange terms involving the $`A`$’s and $`S`$’s separately to obtain these constraints. We write
$$[u,t]=C_\beta +C_\gamma .$$
(15)
Where $`C_\beta =0`$ and $`C_\gamma =0`$ are necessary conditions for $`[u,t]=0`$.
$$\begin{array}{cc}\hfill C_\beta & =\frac{\pi }{2}\underset{s=1}{\overset{N/2}{}}\underset{p=1}{\overset{N}{}}\left[\beta _{p(2s)}A^{p(2s1)}+\beta _{p(2s1)}A^{p(2s)}\right]\hfill \\ & =\frac{\pi }{2}\underset{seven}{\overset{N}{}}\underset{peven}{\overset{N}{}}\beta _{ps}A^{p(s1)}\frac{\pi }{2}\underset{seven}{\overset{N}{}}\underset{podd}{\overset{N1}{}}\beta _{ps}A^{p(s1)}\hfill \\ & \frac{\pi }{2}\underset{sodd}{\overset{N1}{}}\underset{peven}{\overset{N}{}}\beta _{ps}A^{p(s+1)}\frac{\pi }{2}\underset{sodd}{\overset{N1}{}}\underset{podd}{\overset{N1}{}}\beta _{ps}A^{p(s+1)}\hfill \end{array}$$
(16)
We wish to rearrange terms so that we have a unique set of $`A`$’s whose coefficients we can set to zero in order to obtain our constraints. In the first and fourth terms here the indices on the $`A`$’s have opposite parity, whereas in the second and third terms the indices have the same parity in each term, but the indices are both odd in the second term and both even in the third term. This means that the first and fourth terms may be combined by exchanging dummy indices and using the symmetry properties of the $`A`$’s, whereas the second and third terms must be dealt with separately. Denoting term $`x`$ in the right hand side of Eq. (16), $`C_\beta ^x`$, and taking the first and fourth terms:
$$\begin{array}{c}\hfill C_\beta ^1+C_\beta ^4=\frac{\pi }{2}\underset{sodd}{\overset{N1}{}}\underset{peven}{\overset{N}{}}\beta _{p(s+1)}A^{ps}\frac{\pi }{2}\underset{seven}{\overset{N}{}}\underset{podd}{\overset{N1}{}}\beta _{p(s1)}A^{ps}\end{array}$$
(17)
Exchanging $`p`$ and $`s`$ in the second term on the right hand side:
$$\begin{array}{c}\hfill C_\beta ^1+C_\beta ^4=\frac{\pi }{2}\underset{sodd}{\overset{N1}{}}\underset{peven}{\overset{N}{}}\left[\beta _{p(s+1)}A^{ps}+\beta _{s(p1)}A^{sp}\right]\end{array}$$
(18)
Using the antisymmetry of the $`A`$’s, we have
$$\begin{array}{c}\hfill C_\beta ^1+C_\beta ^4=\frac{\pi }{2}\underset{sodd}{\overset{N1}{}}\underset{peven}{\overset{N}{}}\left[\beta _{p(s+1)}\beta _{s(p1)}\right]A^{ps}\end{array}$$
(19)
Now consider the second and third terms in $`C_\beta `$. The second term is:
$$\begin{array}{cc}\hfill C_\beta ^2& =\frac{\pi }{2}\underset{sodd}{\overset{N1}{}}\underset{podd}{\overset{N1}{}}\beta _{p(s+1)}A^{ps}\hfill \\ & =\frac{\pi }{4}\underset{sodd}{\overset{N1}{}}\underset{podd}{\overset{N1}{}}\left[\beta _{p(s+1)}\beta _{s(p+1)}\right]A^{ps}\hfill \end{array}$$
(20)
The third term is:
$$\begin{array}{cc}\hfill C_\beta ^3& =\frac{\pi }{2}\underset{seven}{\overset{N}{}}\underset{peven}{\overset{N}{}}\beta _{p(s1)}A^{ps}\hfill \\ & =\frac{\pi }{4}\underset{seven}{\overset{N}{}}\underset{peven}{\overset{N}{}}\left[\beta _{p(s1)}\beta _{s(p1)}\right]A^{ps}\hfill \end{array}$$
(21)
These equations give the following set of conditions on the beta coefficients:
$$\begin{array}{c}\hfill \beta _{p(s+1)}\beta _{s(p1)}=0smod2=1pmod2=0\\ \hfill \beta _{p(s+1)}\beta _{s(p+1)}=0smod2=1pmod2=1\\ \hfill \beta _{p(s1)}\beta _{s(p1)}=0smod2=0pmod2=0\end{array}$$
(22)
These constraints are redundant. In fact a pair of constraints is sufficient to ensure $`C_\beta =0`$:
$$\begin{array}{cc}& \beta _{p(s+1)}\beta _{s(p1)}=0smod2=1s<p\hfill \\ & \beta _{p(s+1)}\beta _{s(p+1)}=0smod2=1pmod2=1s<p\hfill \end{array}$$
(23)
We now consider $`C_\gamma `$.
$$\begin{array}{cc}\hfill C_\gamma & =\frac{\pi }{2}\underset{s=1}{\overset{N/2}{}}\underset{p=1}{\overset{N}{}}\left[\gamma _{p(2s)}S^{(2s1)p}+\gamma _{p(2s1)}S^{(2s)p}\right]\hfill \\ & =\frac{\pi }{2}\underset{seven}{\overset{N}{}}\underset{peven}{\overset{N}{}}\gamma _{p(s1)}S^{sp}\frac{\pi }{2}\underset{seven}{\overset{N}{}}\underset{podd}{\overset{N1}{}}\gamma _{p(s1)}S^{sp}\hfill \\ & \frac{\pi }{2}\underset{sodd}{\overset{N1}{}}\underset{peven}{\overset{N}{}}\gamma _{p(s+1)}S^{sp}\frac{\pi }{2}\underset{sodd}{\overset{N1}{}}\underset{podd}{\overset{N1}{}}\gamma _{p(s+1)}S^{sp}\hfill \end{array}$$
(24)
Exchanging dummy indices in the third term, expanding the first and fourth terms, and utilizing the symmetry of the $`S`$’s and the antisymmetry of the $`\gamma `$’s gives:
$$\begin{array}{cc}\hfill C_\gamma & =\frac{\pi }{4}\underset{seven}{\overset{N}{}}\underset{peven}{\overset{N}{}}\left[\gamma _{p(s1)}+\gamma _{s(p1)}\right]S^{sp}\hfill \\ & \frac{\pi }{2}\underset{seven}{\overset{N}{}}\underset{sodd}{\overset{N1}{}}\left[\gamma _{p(s1)}\gamma _{(p+1)s}\right]S^{sp}\hfill \\ & \frac{\pi }{4}\underset{sodd}{\overset{N1}{}}\underset{podd}{\overset{N1}{}}\left[\gamma _{p(s+1)}+\gamma _{s(p+1)}\right]S^{sp}\hfill \end{array}$$
(25)
This yields the constraints on the $`\gamma `$’s:
$$\begin{array}{cc}& \gamma _{p(s1)}\gamma _{(p+1)s}=0smod2=0pmod2=1p<s1\hfill \\ & \gamma _{p(s1)}+\gamma _{s(p1)}=0smod2=0pmod2=0ps\hfill \end{array}$$
(26)
## V Simulations
Simulations were performed for a system whose internal degree of freedom is coupled to a four-dimensional bath. The coupling operator is therefore an element of $`U(8)`$. The general form of an element of the Lie algebra of $`U(8)`$ obeying the constraints derived above is
$$\begin{array}{c}\hfill i\left(\begin{array}{cccccccc}\alpha _1& \beta _{12}& \beta _{13}& \beta _{14}& \beta _{15}& \beta _{16}& \beta _{17}& \beta _{18}\\ \beta _{12}& \alpha _2& \beta _{14}& \beta _{13}& \beta _{16}& \beta _{15}& \beta _{18}& \beta _{17}\\ \beta _{13}& \beta _{14}& \alpha _3& \beta _{34}& \beta _{35}& \beta _{36}& \beta _{37}& \beta _{38}\\ \beta _{14}& \beta _{13}& \beta _{34}& \alpha _4& \beta _{36}& \beta _{35}& \beta _{38}& \beta _{37}\\ \beta _{15}& \beta _{16}& \beta _{35}& \beta _{36}& \alpha _5& \beta _{56}& \beta _{57}& \beta _{58}\\ \beta _{16}& \beta _{15}& \beta _{36}& \beta _{35}& \beta _{56}& \alpha _6& \beta _{67}& \beta _{57}\\ \beta _{17}& \beta _{18}& \beta _{37}& \beta _{38}& \beta _{57}& \beta _{67}& \alpha _7& \beta _{78}\\ \beta _{18}& \beta _{17}& \beta _{38}& \beta _{37}& \beta _{58}& \beta _{57}& \beta _{78}& \alpha _8\end{array}\right)+\left(\begin{array}{cccccccc}0& 0& \gamma _{13}& \gamma _{14}& \gamma _{15}& \gamma _{16}& \gamma _{17}& \gamma _{18}\\ 0& 0& \gamma _{14}& \gamma _{13}& \gamma _{16}& \gamma _{15}& \gamma _{18}& \gamma _{17}\\ \gamma _{13}& \gamma _{14}& 0& 0& \gamma _{35}& \gamma _{36}& \gamma _{37}& \gamma _{38}\\ \gamma _{14}& \gamma _{13}& 0& 0& \gamma _{36}& \gamma _{35}& \gamma _{38}& \gamma _{37}\\ \gamma _{15}& \gamma _{16}& \gamma _{35}& \gamma _{36}& 0& 0& \gamma _{57}& \gamma _{58}\\ \gamma _{16}& \gamma _{15}& \gamma _{36}& \gamma _{35}& 0& 0& \gamma _{58}& \gamma _{57}\\ \gamma _{17}& \gamma _{18}& \gamma _{37}& \gamma _{38}& \gamma _{57}& \gamma _{58}& 0& 0\\ \gamma _{18}& \gamma _{17}& \gamma _{38}& \gamma _{37}& \gamma _{58}& \gamma _{57}& 0& 0\end{array}\right)\end{array}$$
(27)
In all cases the system was initialized in a pure state corresponding to a gaussian spatial wavefunction centred at the origin with a standard deviation equal to one quarter of the lattice size, with equal amplitudes for both internal states of the particle. The system was a periodic lattice with $`64`$ sites, and the parameter $`\theta `$ in the unitary part of the collision operator was set equal to $`0.35`$.
Three types of simulations were performed. First, a simulation of the density matrix of the system and bath was performed in which the coupling matrix was the identity operator. In this case the quantum lattice gas reproduces the unitary evolution expected in the case that the system-bath coupling is zero. Second, simulations were performed in which the system-bath coupling operator is the Fourier transform over the cyclic group $`_8`$. Finally, simulations were performed in which all coefficients $`\alpha `$, $`\beta `$ and $`\gamma `$ in a matrix of the form (27) were set equal to one and the resulting matrix numerically exponentiated to obtain a parity-preserving coupling.
The simulation with no system-bath coupling shows typical unitary evolution on a cycle, with the wave-packet dispersing until the edges of the wave-packet reach the periodic boundaries of the system, at which time interference occurs between the original packet and the reentrant components. The reversibility and unitarity of the dynamics is apparent as the system never settles into a static equilibrium state. The simulations in which the system-bath coupling is given by the Fourier transform, which violates parity invariance, exhibit a driving of the system to the right. Additional simulations in which the coupling is given by the conjugate of the Fourier transform with the parity inversion operator show the same driving effect in the opposite direction, as expected. Simulations in which the system is coupled to the bath by our example parity-invariant unitary matrix show no driving effect. The system undergoes an irreversible evolution of the initial probability distribution to the uniform distribution on the cycle.
The behavior of the model shows its intermediate quantum-classical nature in two ways. First, the decay of the probability distribution shows some residual “wave-like” behavior in addition to the overall damping. Second, the final density matrix is not a completely mixed state, but retains non-zero off-diagonal components within a finite band. Both these effects indicate that the coupling matrix chosen implements a coupling strong enough to cause irreversible dynamics on an observable timescale but weak enough that the dynamics retains some interesting quantum characteristics.
## VI Conclusions
We have defined a new quantum lattice-gas model for a single particle (or equivalently, a discrete-space discrete-time quantum random walk) in which the scattering rule is given by a quantum operation, rather than a unitary, deterministic, or stochastic operation. We showed that the model so defined includes unitary, stochastic and deterministic models as special cases, as well as interesting intermediate behavior. Preliminary simulation results confirm this by exhibiting non-unitary diffusive decay of an initial gaussian pure state.
The noise model chosen here utilized a fixed unitary operator coupling a four-dimensional bath to the two-dimensional Hilbert space of the internal degree of freedom of a single quantum lattice-gas particle. This distinguishes our work from the noise model given by Kendon and Sanders Kendon and Sanders (2004) in which a single environment qubit is coupled by an interaction with a strength tunable from the case of no coupling to the case where the environment produced a projective measurement of the coin degree of freedom of the quantum random walk.
As noted above, a two-dimensional environment with a fixed interaction is insufficient to reproduce all quantum operations. However, the constraints of parity invariance on the coupling operator were obtained here for $`U(N)`$, and so this work could be extended to include an arbitrarily large environment. If the most general noise model is desired, the sampling procedure discussed but not implemented above, in which randomly sampled unitary operators couple a four-dimensional bath to the internal degree of freedom of the particle, provably includes all quantum operations. The parameterization of the noise model then involves a parameterization of the distribution of quantum operations induced by a given distribution on the unitary group.
The above discussion motivates several directions for future work. First, the noise model presented here is not conveniently parameterized. Ideally we would be able to smoothly vary the degree of coupling between the system and the environment from zero (where the time evolution would approximate the Dirac equation) to the case where the environment performs projective measurement of the particles’ internal degrees of freedom. This property is possessed by the noise model of Kendon and Sanders Kendon and Sanders (2004), and such a parameterization of the model presented here is certainly possible, although it may be tedious in practice. The work of Kendon and Sanders Kendon and Sanders (2004) discusses decoherence in quantum random walks from the point of view of complementarity. The work of the present paper was motivated instead by the principle of correspondence. The generalization here satisfies the requirement that the results agree with classical theory in the case that the particle is strongly coupled to an environment.
The two most natural generalizations of the work described here are to quantum lattice-gas models with multiple particles, and to models defined in multiple dimensions. The constraint of parity invariance becomes the constraint of invariance under discrete rotations in multiple dimensions, and one expects the constraints analogous to those derived here to be correspondingly more complex. Simulations of multiple particles, even in one dimension, have a classical computational cost which grows exponentially with the number of particles. However, one expects that few-particle simulations will be tractable.
The possibility of efficient quantum simulation of classical systems is an important open problem in the field of quantum computation for physical modeling. One of the central problems in this area is that most of the equations of classical physics of practical interest are irreversible macroscopic equations of motion. The work presented here shows that irreversibility may be simply included by the use of quantum operations instead of unitary matrices. The emergence of irreversible behavior in the degrees of freedom of the subsystem exhibited here is a manifestation of the “arrow of time” of non-equilibrium thermodynamics. The interesting practical question, which remains open, is whether there are systems for which the time complexity of such classical and quantum simulation is different. More glibly: Can time’s arrow may be made to move faster on a quantum computer?
## VII Acknowledgements
PJL and BMB were supported by DARPA QuIST program administered under AFOSR grant number F49620-01-1-0566, and by ARO contract number W911NF-04-1-0334. BMB was also supported by AFOSR award number FA9550-04-1-0176. Both authors would like to thank AFOSR for their hospitality at the Quantum Computation for Physical Modeling Workshop on Marthas Vineyard in 2004. The authors have great pleasure in thanking David Meyer, Gianluca Caterina, Howard Brandt, and Seth Lloyd for helpful discussions and questions. |
warning/0507/astro-ph0507372.html | ar5iv | text | # Interpreting Cosmological Vacuum Decay
## I Introduction
There is nowadays significant observational evidence that the expansion of the Universe is undergoing a late time acceleration perl ; wmap ; rnew ; allen ; revde . This, in other words, amounts to saying that in the context of Einstein’s general theory of relativity some sort of *dark energy*, constant or that varies only slowly with time and space, dominates the current composition of the cosmos (see, e.g., revde for some recent reviews on this topic). The origin and nature of such an *accelerating field* constitutes a completely open question and represents one of the major challenges not only to cosmology but also to our current understanding of fundamental physics.
Among many possible alternatives, the simplest and most theoretically appealing possibility for dark energy is the energy density stored on the true vacuum state of all existing fields in the Universe, i.e., $`\rho _\mathrm{\Lambda }=\mathrm{\Lambda }/8\pi G`$, where $`\mathrm{\Lambda }`$ is the cosmological constant. From the observational side, flat models with a relic cosmological term ($`\mathrm{\Lambda }`$CDM) seems to be in agreement with almost all cosmological observations, which makes them an excellent description of the observed universe. From the theoretical viewpoint, however, the well-known cosmological constant problem, i.e., the unsettled situation in the particle physics/cosmology interface, in which the cosmological upper bound ($`\rho _\mathrm{\Lambda }10^{47}\mathrm{GeV}^4`$) differs from theoretical expectations ($`\rho _\mathrm{\Lambda }10^{71}\mathrm{GeV}^4`$) by more than 100 orders of magnitude, originates an extreme fine-tuning problem revde1 or makes a complete cancellation (from an unknown physical mechanism) seem more plausible.
In this regard, a phenomenological attempt at alleviating such a problem is allowing $`\mathrm{\Lambda }`$ to vary<sup>3</sup><sup>3</sup>3Strictly speaking, in the context of classical general relativity any additional $`\mathrm{\Lambda }`$-type term that varies in space or time should be thought of as a new *time-varying field* and not as a cosmologial constant. Here, however, we adopt the usual nomenclature of time-varying or dynamical $`\mathrm{\Lambda }`$ models.. Cosmological scenarios with a time-varying or dynamical $`\mathrm{\Lambda }`$ were independently proposed almost twenty years ago in Refs. ozer (see also bron ). Afterward, a number of models with different decay laws for the variation of the cosmological term were investigated in Ref. lambdat0 and the confrontation of their predictions with observational data has also been analyzed by many authors lambdat1 . It is worth mentioning that the most usual critique to these $`\mathrm{\Lambda }`$(t)CDM scenarios is that in order to establish a model and study their observational and theoretical predictions, one needs first to specify a phenomenological time-dependence for $`\mathrm{\Lambda }`$. In this concern, an interesting step towards a more realistic decay law was given recently by Wang & Meng in Ref. wm . Instead of the traditional approach, they deduced a new decay law from a simple argument about the effect of the vacuum decay on the cold dark matter (CDM) expansion rate. Such a decay law is similar to the one originally obtained in Ref. shapiro1 from arguments based on renormalization group and seems to be very general, having many of the previous attempts as a particular case and being capable of reconciling $`\mathrm{\Lambda }`$(t)CDM models with an initially decelerated and late time accelerating universe, as indicated by current SNe Ia observations rnew .
The aim of the present paper is is twofold: first, to interpret thermodynamically the process of cosmological vacuum decay, as suggested in Ref. wm . From thermodynamic considerations, it is shown that such a process leads to two different effects, namely, a continuous creation of particles and an increasing in the mass of CDM particles given by $`m(t)=m_oa(t)^ϵ`$, where $`a(t)`$ is the cosmological scale factor and $`ϵ`$ is the parameter quantifying the decay vacuum rate; second, to analyze the dynamic modifications in the original Wang-Meng cosmic scenario by introducing explicitly the baryonic component. As we shall see, the presence of baryons alters considerably the accelerating redshift $`z_{}`$, that is, the redshift at which the Universe switches from deceleration to acceleration. In order to constrain the parametric space $`\mathrm{\Omega }_mϵ`$, we also perform a statistical analysis involving three sets of observables, namely, the latest Chandra measurements of the X-ray gas mass fraction in 26 galaxy clusters, as provided by Allen et al. allen , the so-called “gold” set of 157 SNe Ia, recently published by Riess et al. rnew , and the measurement of the CMB shift parameter, as given by WMAP, CBI, and ACBAR wmap . Finally, we extend the treatment of Ref. wm to a scenario in which the vacuum energy decays into photons. In this case, it is found that the temperature evolution law of radiation is modified to $`T=T_oa(t)^{ϵ/41}`$.
## II Vacuum Decay into CDM
Let us first consider the Einstein field equations
$$R^{\mu \nu }\frac{1}{2}Rg^{\mu \nu }=\chi \left[T^{\mu \nu }+\frac{\mathrm{\Lambda }}{\chi }g^{\mu \nu }\right],$$
(1)
where $`R^{\mu \nu }`$ and $`R`$ are, respectively, the Ricci tensor and the scalar curvature, $`T^{\mu \nu }`$ is the energy-momentum tensor of matter fields and CDM particles, and $`\chi =8\pi G`$ ($`c=1`$) is the Einstein’s constant. Note that according to the Bianchi identities, the above equations implies that $`\mathrm{\Lambda }`$ is necessarily a constant either if $`T^{\mu \nu }=0`$ or if $`T^{\mu \nu }`$ is separately conserved, i.e., $`u_\mu T^{\mu \nu };_\nu =0`$. In other words, this amounts to saying that (i) vacuum decay is possible only from a previous existence of some sort of non-vanishing matter and/or radiation, and (ii) the presence of a time-varying cosmological term results in a coupling between $`T^{\mu \nu }`$ and $`\mathrm{\Lambda }`$. For the moment, we will assume a coupling only between vacuum and CDM particles, so that
$$u_\mu 𝒯^{\mu \nu };_\nu =u_\mu (\frac{\mathrm{\Lambda }g^{\mu \nu }}{\chi });_\nu ,$$
(2)
or, equivalently,
$$\dot{\rho }_m+3\frac{\dot{a}}{a}\rho _m=\dot{\rho }_v,$$
(3)
where $`\rho _m`$ and $`\rho _v`$ are the energy densities of the CDM and vacuum, respectively, and $`𝒯^{\mu \nu }=\rho _mu^\mu u^\nu `$ denotes the energy-momentum tensor of the CDM matter.
As commented earlier, the traditional approach for $`\mathrm{\Lambda }`$(t)CDM models was first to specify a phenomenological decay law and then establish a cosmological scenario (see, e.g., lambdat0 ; lambdat1 ). Here, however, we follow the arguments presented in Ref. wm , in which a decay law is deduced from the effect it has on the CDM evolution. The qualitative argument is the following: since vacuum is decaying into CDM particles, CDM will dilute more slowly compared to its standard evolution, $`\rho _ma^3`$. Thus, if the deviation from the standard evolution is characterized by a positive constant $`ϵ`$, i.e.,
$$\rho _m=\rho _{mo}a^{3+ϵ},$$
(4)
Eq. (3) yields
$$\rho _v=\stackrel{~}{\rho }_{vo}+\frac{ϵ\rho _{m0}}{3ϵ}a^{3+ϵ},$$
(5)
where $`\rho _{mo}`$ is the current CDM energy density and $`\stackrel{~}{\rho }_{vo}`$ stands for what is named in Ref. wm “the ground state value of the vacuum”. As discussed there, such a decay law seems to be the most general one, having many of the previous phenomenological attempts as a particular case.
## III Thermodynamics of vacuum decay
Let us now investigate some thermodynamic features of the decaying vacuum scenario described in the last section. As discussed in Ref. Lima1 , the thermodynamic behavior of a decaying vacuum system is simplified if one assumes that the chemical potential of the vacuum component is zero, and also if the vacuum medium plays the role of a condensate carrying no entropy, as happens in the two fluid description employed in superfluid thermodynamics. In this case, the thermodynamic description require only the knowledge of the particle flux, $`N^\alpha =nu^\alpha `$, and the entropy flux, $`S^\alpha =n\sigma u^\alpha `$, where $`n=N/a^3`$ and $`\sigma =S/N`$ are, respectively, the concentration and the specific entropy (per particle) of the created component.
It is clear from last Section that in the Wang-Meng description the two component are changing energy, but it is not clear where the vacuum energy is going to or, in other words, where the CDM component is storing the energy received from the vacuum decay process. In principle, since the energy density of the cold dark matter is $`\rho =nm`$, there are two possibilities:
(i) the equation describing concentration, $`n`$, has a source term while the proper mass of CDM particles remains constant;
(ii) the mass $`m`$ of the CDM particles is itself a time-dependent quantity while the total number of CDM particles, $`N=na^3`$, remains constant.
The case (i) seems to be physically more realistic, and coincides exactly with the description presented in Ref. Lima1 . However, for the sake of completeness, in what follows we consider both cases.
### III.1 Case I: Vacuum decay into CDM particles
In this case, there is necessarily a source term in the current of CDM particles, that is, $`N^\alpha ;_\alpha =\psi `$. In terms of the concentration it can be written as
$$\dot{n}+3\frac{\dot{a}}{a}n=\psi =n\mathrm{\Gamma },$$
(6)
where $`\psi `$ is the particle source ($`\psi >0`$), or a sink ($`\psi <0`$), and we have written it in terms of a decay rate, $`\mathrm{\Gamma }`$. Since $`\rho =nm`$ we find from (4) that $`n=n_oa^{3+ϵ}`$. Inserting this result into the above equation it follows that
$$\mathrm{\Gamma }=ϵ\frac{\dot{a}}{a}.$$
(7)
The vacuum decay and the associated particle creation rate are the unique sources of irreversibility. Thermodynamically, the overall energy transfer from the vacuum to the fluid component may happens in several ways. In the most physically relevant case it has been termed adiabatic decaying vacuum Lima1 (see also pavon for more applications of adiabatic decay processes in cosmology). In this case, several equilibrium relations are preserved, and, perhaps, more important, the entropy of the created particles increases but the specific entropy (per particle) remains constant ($`\dot{\sigma }=0`$). This means that
$$\frac{\dot{S}}{S}=\frac{\dot{N}}{N}=\mathrm{\Gamma }.$$
(8)
On the other hand, from Eq. (7) we see that the total number of particles scales as a power law
$$N(t)=N_oa(t)^ϵ,$$
(9)
whereas the second law of thermodynamics, $`\dot{S}0`$, implies that $`ϵ0`$, as should be expected. To close the connection with the Wang-Meng scenario we need to show that the vacuum energy density follows naturally from the thermodynamic approach. Actually, for an adiabatic vacuum decay process one may write (see Eqs. (8) and (19) of Ref. Lima1 )
$$\dot{\rho }_v=\beta \psi ,$$
(10)
where the phenomenological parameter $`\beta `$ is defined by
$$\beta =\frac{\rho +p}{n}.$$
(11)
Finally, by considering that the CDM medium is pressureless, Eq. (10) can be rewritten as
$$\dot{\rho }_v=nmϵ\frac{\dot{a}}{a},$$
(12)
or still,
$$\dot{\rho }_v=\rho _{mo}ϵa^{4+ϵ}\dot{a},$$
(13)
whose integration reproduces expression (5) previously derived by Wang and Meng wm . Beyond the independent derivation of the decaying vacuum energy density, the interesting point here is that the sign of the “coupling constant”, $`ϵ`$, is constrained by the second law of thermodynamics.
### III.2 Case II: Variable Mass Particles
In this case, *there is no creation of CDM particles*, which means that the concentration satisfies the equation
$$\dot{n}+3\frac{\dot{a}}{a}n=0,$$
(14)
whose solution is $`n=n_oa^3`$ which implies that $`N(t)=constant`$. Naturally, if CDM particles are not being created, the unique possibility is an increasing in the proper mass of CDM particles. Actually, since $`\rho =nm`$, Eqs. (4) and (14) imply that the mass of the CDM particles scales as
$$m(t)=m_oa(t)^ϵ,$$
(15)
where $`m_o`$ is the present day mass of CDM particles (compare with expression (9)). Note that this approach for the vacuum decay process leads to a VAMP<sup>4</sup><sup>4</sup>4VAriable Mass Particles-type scenario, in which the interaction of CDM particles with the dark energy field imply directly in an increasing of the mass of CDM particles (see, e.g., vamp and references therein for more about VAMP models). To complete our thermodynamic approach for the vacuum decay, a similar treatment for the case in which the vacuum decays only into photons is briefly presented in Appendix A.
## IV Observational Aspects
In this Section we study some observational aspects of the cosmological scenario discussed above. The Friedmann equation for this modified $`\mathrm{\Lambda }`$(t)CDM cosmology reads
$$(\frac{H}{H_o})^2=\left[\mathrm{\Omega }_ba^3+\frac{3\mathrm{\Omega }_m}{3ϵ}a^{3+ϵ}+\stackrel{~}{\mathrm{\Omega }}_{vo}\right],$$
(16)
where $`\mathrm{\Omega }_b`$ and $`\mathrm{\Omega }_m`$ are, respectively, the baryon and CDM density parameters and $`\stackrel{~}{\mathrm{\Omega }}_{vo}`$ is the density parameter associated with “the ground state of vacuum”. Note that unlike Eq. (6) of Ref. wm , the above Friedmann equation has an additional term which accounts for the baryon contribution to the cosmic expansion. The presence of such a term – redshifting as $`(1+z)^3`$ – is justified here since the vacuum is assumed to decay only into CDM particles.
### IV.1 Transition epoch
Although subdominant at the present stage of cosmic evolution, the baryonic content may be important for reconciling $`\mathrm{\Lambda }`$(t)CDM models with some current cosmological observations. As an example, let us consider the transition redshift, $`z_{}`$, at which the Universe switches from deceleration to acceleration or, equivalently, the redshift at which the deceleration parameter vanishes. From Eq. (16), it is straightforward to show that the deceleration parameter, defined as $`q=a\ddot{a}/\dot{a}^2`$, now takes the following form
$$q(a)=\frac{3}{2}\frac{\mathrm{\Omega }_ba^3+\mathrm{\Omega }_ma^{3+ϵ}}{\mathrm{\Omega }_ba^3+\frac{3\mathrm{\Omega }_m}{3ϵ}a^{3+ϵ}+\stackrel{~}{\mathrm{\Omega }}_{vo}}1,$$
(17)
where we have set $`a_o=1`$.
Two important aspects concerning the above equation should be emphasized at this point. First, note that the presence in Eq. (17) of a non-null density parameter associated with the ground state of vacuum makes possible a transition deceleration/acceleration, as indicated by current SNe Ia observations rnew . As well discussed in Ref. wm , in most of the cases, $`\mathrm{\Lambda }`$(t)CDM models without such a term predict a universe which is either always accelerating or always decelerating from the onset of matter domination up to today. Second, note also that, due to the presence of the baryons, the transition epoch is delayed relative to previous cases (including the standard $`\mathrm{\Lambda }`$CDM model), which seems to be in better agreement with recent results indicating $`z_{}=0.46\pm 0.13`$ at 1$`\sigma `$ rnew .
To better visualize the effect of baryons on the transition epoch, we show in Fig. 1a the behavior of the deceleration parameter as a function of redshift \[Eq. (17)\] for selected values of the parameter $`ϵ`$. In agreement with WMAP estimates wmap we also assume $`\mathrm{\Omega }_m=0.27\pm 0.04`$ and $`\mathrm{\Omega }_b=0.044\pm 0.004`$. The best fit $`\mathrm{\Lambda }`$CDM case (the so-called ”concordance model”) is also showed for the sake of comparison. Note that at late times ($`z=0`$), since $`ϵ`$ is a positive quantity, the standard $`\mathrm{\Lambda }`$CDM scenario always accelerates faster than $`\mathrm{\Lambda }`$(t)CDM models, with the condition for current acceleration being $`\stackrel{~}{\mathrm{\Omega }}_{vo}>\frac{\mathrm{\Omega }_b}{2}+\frac{3\mathrm{\Omega }_m(1ϵ)}{62ϵ}`$. A closer look at the results shown in Fig. 1a is displayed in Fig. 1b. In Fig. 1c we show the transition redshift $`z_{}`$ as a function of the parameter $`ϵ`$, which is obtained from the expression
$$\mathrm{\Omega }_b(1+z_{})^3+\left[\frac{33ϵ}{3ϵ}\right]\mathrm{\Omega }_m(1+z_{})^{3ϵ}2\stackrel{~}{\mathrm{\Omega }}_{vo}=0.$$
(18)
Two different cases are shown. The scenario of Ref. wm (no baryons – dashed line) and the model presented here (solid line), in which the baryonic content accounts for $`4.4\%`$ of the critical density. As physically expected (due to the attractive gravity associated with the baryonic content), $`z_{}`$ is always smaller in the latter scenario than in the former. In particular, by considering the 2$`\sigma `$ interval $`0.2z_{}0.72`$ rnew (horizontal dashed lines) we find $`ϵ0.16`$, which is in fully agreement with the results of the statistical analysis performed in the next Section.
### IV.2 SNe Ia, Clusters and CMB Constraints
In order to delimit the parametric space $`\mathrm{\Omega }_mϵ`$ we perform in this Section a joint statistical analysis involving three complemetary sets of observations. We use to this end the latest Chandra measurements of the X-ray gas mass fraction in 26 galaxy clusters, as provided by Allen et al. allen along with the so-called “gold” set of 157 SNe Ia, recently published by Riess et al. rnew , and the estimate of the CMB shift parameter wmap , $`R\mathrm{\Omega }_m^{1/2}\mathrm{\Gamma }(z_{\mathrm{CMB}})=1.716\pm 0.062`$ from WMAP, CBI, and ACBAR wmap , where $`\mathrm{\Gamma }(z)`$ is the dimensionless comoving distance and $`z_{\mathrm{CMB}}=1089`$. In our analysis, we also include the most recent determinations of the baryon density parameter, as given by the WMAP team wmap , i.e., $`\mathrm{\Omega }_bh^2=0.0224\pm 0.0009`$ and the latest measurements of the Hubble parameter, $`h=0.72\pm 0.08`$, as provided by the HST key project hst (we refer the reader to refer for more details on the statistical analysis).
In Fig. 2 we show the results of our statistical analysis. Confidence regions ($`68.3\%`$, 95.4$`\%`$ and 99.7$`\%`$) in the plane $`\mathrm{\Omega }_mϵ`$ are shown for the particular combination of observational data described above. Note that, although the limits on the parameter $`ϵ`$ are very restrictive, the analysis clearly shows that the model presented here constitutes a small but significant deviation from the standard $`\mathrm{\Lambda }`$CDM dynamics. The best-fit parameters for this analysis are $`\mathrm{\Omega }_m=0.27`$ and $`ϵ=0.11`$, with the relative $`\chi _{min}^2/\nu 1.12`$ ($`\nu `$ is defined as degrees of freedom). Note that this value of $`\chi _{min}^2/\nu `$ is similar to the one found for the so-called “concordance model” by using SNe Ia data only, i.e., $`\chi _{min}^2/\nu 1.13`$ rnew . At 95.4$`\%`$ c.l. we also found $`\mathrm{\Omega }_m=0.26\pm 0.05`$ and $`ϵ=0.11\pm 0.12`$.
## V Conclusion
In this paper we have slightly modified and interpreted several features of the decaying vacuum scenario recently proposed by Wang and Meng wm . A baryonic component has been explicitly introduced, and we have seen that it has an important dynamic effect, namely, the transition epoch from a decelerating/acelerating regime is delayed relative to the one predicted by the original Wang-Meng scenario (including the standard $`\mathrm{\Lambda }`$CDM model). The importance of the baryonic contribution cannot be neglected because it reconciles the decaying vacuum scenario with the recent observations rnew (see figure 1, panel c). However, other details of the radiation and matter dominated phases are not modified. This is easily verified by computing the value of the redshift $`z`$ for which $`\rho _b=\rho _m`$. For the present values of the density parameters, $`\mathrm{\Omega }_{mo}0.3`$ and $`\mathrm{\Omega }_{bo}=0.04`$, one finds $`z10^{1/ϵ}`$. Therefore, for $`ϵ0.11`$ (the best-fit found in this paper), we obtain $`z10^{10}`$. In other words, after this redshift, the Universe is still radiation dominated but the baryons are already subdominant in comparison to the CDM component.
We have also discussed some thermodynamic aspects of such a scenario assuming that the baryonic component is identically conserved. In particular, if CDM particles are produced by the decaying vacuum, we shown that the sign of the coupling parameter, $`ϵ`$, is restricted by the second law of thermodynamics to assume only positive values. In this case, the total number of CDM particles is a time-dependent function given by $`N(t)=N_oa^ϵ`$. However, VAMP-type scenarios - VAriable mass particles - are also possible when the total number of particles remains constant. In this case, the mass scales as $`m(t)=m_oa^ϵ`$, that is, the energy of the vacuum decay process is totally transformed in mass of the the existing particles. Naturally, if photons are produced, the temperature law of radiation must also be affected. This case has been discussed with some detail in the Appendix A.
## Appendix A Vacuum decay into radiation
In this Appendix we briefly discuss how the Wang-Meng treatment can be extended to the case of radiation. Now, the energy conservation law reads
$$\dot{\rho }_r+4H\rho _r=\dot{\rho }_v,$$
(19)
where $`\rho _r`$ is the radiation energy density. By considering that radiation will dilute more slowly compared to its standard evolution, $`\rho _ma^4`$, and that such a deviation is characterized by a positive constant $`\alpha `$ we find
$$\rho _r=\rho _{ro}a(t)^{4+\alpha },$$
(20)
where $`\rho _{ro}`$ is the present day energy density of radiation. For an adiabatic vacuum decay the equilibrium relations are preserved Lima1 ; pavon , as happens with the Stefan law, $`\rho _r=aT^4`$. As a consequence, one may check that the product $`Ta^{1\alpha /4}`$ remains constant and, as such, this implies that the new temperature law scales with redshift as
$$T=T_o(1+z)^{1\alpha /4}.$$
(21)
By inserting (20) into (19) it follows that
$$\rho _v=\stackrel{~}{\rho }_{vo}+\frac{\alpha \rho _{ro}}{4\alpha }a^{4+\alpha },$$
(22)
which should be compared with Eq. (5) describing a decaying vacuum energy density into cold dark matter. Note that the ratio between the vacuum and radiation energy densities are:
$$\frac{\rho _v}{\rho _r}=\frac{\stackrel{~}{\rho }_{vo}}{\rho _{ro}}a^{4\alpha }+\frac{\alpha }{4\alpha }.$$
(23)
The first term is asymptotically vanishing at early times whereas the second one is smaller than unity. Therefore, a radiation dominated stage is always guaranteed in this kind of scenarios.
Acknowledgments: This work was supported by CNPq (Brazilian Research Agency). The authors thank R. Silva, R. C. Santos and J. F. Jesus for valuable discussions and Joan Solà and Hrvoje Stefancic for pointing a error in the first version of this paper. |
warning/0507/astro-ph0507214.html | ar5iv | text | # Integrated spectral analysis of 18 concentrated star clusters in the Small Magellanic Cloud
## 1 Introduction
The Small Magellanic Cloud (SMC) is a galaxy rich in star clusters of all ages and different types of field populations (Hodge 1988, 1989; Dolphin et al. 2001). An interesting feature in the chemical enrichment history of the SMC known up to now is that no very metal-poor old cluster has been observed in this galaxy (Da Costa 1991; Dutra et al. 2001). Piatti et al. (2001) studied 5 outlying intermediate-age clusters in the SMC and, combined to other data in the literature, studied the age-metallicity relationship, showing that epochs of sudden chemical enrichment take place in the age-metallicity plane. This favours a bursting star formation history for the SMC as opposed to a continuous one. Recently, Piatti et al. (2005) confirmed, with new observations, the occurrence of an important bursting star formation episode at $``$ 2.5 Gyr.
A star cluster spectral library at the SMC metallicity level can be useful for analyses of star clusters in dwarf galaxies observable by means of ground-based large telescopes as well as the Hubble Space Telescope (HST). In addition, such metal-poor library appears to be also useful for the study of a fraction of star clusters in massive galaxies, due to cannibalism. Indeed, in the Milky Way galaxy at least four globular clusters have been accreted from the Sagittarius dwarf galaxy (Da Costa & Armandroff 1995), and the open clusters AM-2 and Tombaugh 5 appear to be related to the Canis Major dwarf galaxy (Bellazzini et al. 2004).
In this sense, spectral libraries of stars (e.g., Silva & Cornell 1992), open clusters (Piatti et al. 2002a) or star clusters in general (Bica & Alloin 1986) are important datasets for spectral classifications and extraction of parameter information for target stars or star clusters (e.g., Piatti et al. 2002b) and galaxies (e.g., Bica 1988).
Samples of integrated spectra of SMC clusters were initially small, corresponding to the most prominent clusters such as NGC 121, NGC 419, NGC 330 and others (Bica & Alloin 1986; Santos et al. 1995). Ahumada et al. (2002) analysed integrated spectra in the range 3600-6800Å for 16 star clusters in the SMC, estimating ages and reddening values. That study has constituted a fundamental step forward towards a cluster spectral library at low metallicities.
A comprehensive catalogue of SMC clusters was produced by Bica & Schmitt (1995), and updated in Bica & Dutra (2000). In it, cross-identifications for different designations, coordinates, angular sizes and references to previous catalogues are provided. The angular distribution of SMC clusters has been discussed in Bica & Schmitt (1995): most clusters are projected on the SMC main body and a significant fraction are outlyers. The line-of-sight (LOS) depth of populous clusters in the SMC was analysed by Crowl et al. (2001), who found significant depth effects, with a triaxial shape of 1:2:4 for the declination, the right ascension, and the LOS depth of the SMC, respectively.
The present cluster sample complements previous ones, in an effort to provide a spectral library with several clusters per age bin. At the same time, we study the clusters themselves individually, determining their parameters and analysing the age distribution, in order to explore the SMC star formation history and its spatial extent. To estimate the clusters’ ages, we employ the new calibrations and diagnostic diagrams recently provided by Santos & Piatti (2004, hereafter SP) for visible integrated spectra, along with template spectra (e.g., Santos et al. 1995; Ahumada et al. 2002). We confirm the reliability of the procedure proposed by SP in determining clusters’ ages, since we included in the sample not only unstudied or poorly studied clusters but also some control clusters with well-known fundamental parameters.
In Sec. 2 we describe the different sets of observations and the reduction procedure performed. The analyses of the integrated spectra through the template matching and equivalent width methods are developed in Sec. 3, in which we also include some considerations for individual clusters. In Sec. 4 we discuss the present results in the light of the star formation history of the SMC. Finally, in Sec. 5 we summarize the main conclusions of this work.
## 2 Spectra acquisition and reduction
The objects studied here are part of a systematic spectroscopic survey of SMC star clusters which is being undertaken at Complejo Astronómico El Leoncito (CASLEO) in San Juan (Argentina) and Cerro Tololo Inter-American Observatory (CTIO, Chile). The first results of this survey dealt with 16 concentrated star clusters (Ahumada et al. 2002), approximately half of which constitute previously unstudied objects.
The observations analysed in this study were carried out with the CASLEO 2.15 m telescope during four nights in November 2001 and five nights in October 2002 and with the CTIO 1.5 m telescope during four nights in September 2003. In all the CASLEO runs we empoyed a CCD camera containing a Tektroniks chip of 1024 x 1024 pixels attached to a REOSC spectrograph (simple mode), the size of each pixel being 24 $`\mu `$m x 24 $`\mu `$m; one pixel corresponds to 0.94$`\mathrm{}`$ on the sky. The slit was set in the East-West direction and the observations were performed by scanning the slit across the objects in the North-South direction in order to get a proper sampling of cluster stars. The long slit corresponding to 4.7$`\mathrm{}`$ on the sky, allowed us to sample regions of the background sky. We used a grating of 300 grooves mm<sup>-1</sup>, producing an average dispersion in the observed region of $``$ 140 Å/mm (3.46 Å/pixel). The spectral coverage was $``$ 3600-6800 Å. The seeing during the CASLEO nights was typically 2.0$`\mathrm{}`$. The slit width was 4.2$`\mathrm{}`$, providing a resolution \[full width at half-maximum (FWHM)\] of $``$ 14 Å, as deduced from the comparison lamp lines. For the CTIO observations, we used a CCD Loral 1K chip of 1200 x 800 pixels (pixel diameter = 15 $`\mu `$m), controlled by the CTIO ARCON 3.9 data acquisition system at a gain of 2.05 e<sup>-</sup> ADU<sup>-1</sup> with a readout noise of 7.4 e<sup>-</sup> ADU<sup>-1</sup>. The same slit width of 4.2$`\mathrm{}`$ as in CASLEO was used at CTIO, thus providing a resolution of about 11 Å. The seeing during the CTIO observations was typically 1.0$`\mathrm{}`$. At least two exposures of 30 min of each object were taken in order to correct for cosmic rays. Standard stars from the list of Stone & Baldwin (1983) were also observed at both observatories for flux calibrations. Bias, darks, dome and twilight sky and tungsten lamp flats were taken and employed in the reductions.
The reduction of the spectra was carried out with the IRAF<sup>1</sup><sup>1</sup>1IRAF is distributed by the National Optical Astronomy Observatories, which is operated by the Association of Universities for Research in Astronomy, Inc., under contract with the National Science Foundation package at the Observatorio Astronómico de Córdoba (Argentina - CASLEO data) and at the Instituto de Astronomía y Física del Espacio (Argentina - CTIO data) following the standard procedures. Summing up, we subtracted the bias and used flat-field frames- previously combined - to correct the frames for high and low spatial frequency variations. We also checked the instrumental signature with the acquisition of dark frames. Then, we performed the background sky subtraction using pixel rows from the same frame, after having cleaned the background sky regions from cosmic rays. We controlled that no significant background sky residuals were present on the resulting spectra. The cluster spectra were extracted along the slit according to the cluster size and available flux. Five of these clusters (K 5, K 7, NGC 269, K 28 and NGC 411) have one very bright star located close to their main bodies. The spectra were then wavelength calibrated by fitting observed He-Ne-Cu (CASLEO) or He-Ar (CTIO) comparison lamp spectra with template spectra. The rms errors involved in these calibrations are in average 0.40 Å for both observatories. Finally, we applied to the cluster spectra extinction corrections and flux calibrations derived from the observed standard stars. We decided to use the sensitivity function derived from all the standard stars observed each night. This calibrated function turned out to be nearly the same as the nightly sensitivity functions, but more robustly defined and with a smaller rms error. In addition, cosmic rays on the cluster spectra were eliminated. Table 1 presents the cluster sample including the averaged signal-to-noise (S/N) ratios of the spectra.
## 3 Analysis of the cluster spectra
The cluster parameters were derived by means of two methods: the template matching method, in which the observed spectra are compared and matched to template spectra with well-determined properties (e.g. Piatti et al. 2002a, and references therein), and the equivalent width ($`EW`$) method, in which diagnostic diagrams involving the sum of $`EW`$s of selected spectral lines were employed together with their calibrations with age and metallicity (SP). In the first method, a high weight is assigned to the matching of the overall continuum, while in the second method the spectral lines are the observables that define cluster parameters. Both methods rely on the library of star cluster integrated spectra with well-determined properties, accomplished in various studies (e.g. Bica & Alloin 1986; Piatti et al. 2002a, and references therein) and made available through the CDS/Vizier catalogue database at http://vizier.u-strasbg.fr/cgi-bin/VizieR?-source=III/219 (Santos et al. 2002).
### 3.1 Template matching method
All 18 clusters in our sample are well represented by a wide variety of stellar populations, as may be noticed from their spectra overall appearance. The template spectra useful for the present sample are: Yb (5-10 Myr), Yd (40 Myr), Ye (45-75 Myr), Yg (200-350 Myr), Yh (0.5 Gyr), Ia (1 Gyr) and Ib (3-4 Gyr), which represent young and intermediate-age populations built from Galactic open clusters (Piatti et al. 2002a), and G3 ($`>10`$ Gyr, \[Fe/H\] = -1.0), G4 ($`>10`$ Gyr, \[Fe/H\] = -1.5) and G5 ($`>10`$ Gyr, \[Fe/H\] = -2.0), which represent old stellar populations built from Galactic globular clusters (Bica 1988).
The template matching method consists of achieving the best possible match between the analysed cluster spectrum and a template spectrum of known age and metallicity. In this process we selected, among the available template spectra, the ones which minimize the flux residuals, calculated as the difference (cluster - template)/cluster. Note that differences between the cluster and template spectra are expected to be found due to variations in the stellar composition of the cluster, such as the presence of a relatively bright star with particular spectral features or contamination of a field star close to the direction towards the cluster.
Since the continuum distribution is also affected by reddening, we first adopted a colour excess $`E(BV)`$ for each cluster, taking into account the Burstein & Heiles (1982, hereafter BH) and Schlegel et al. (1998, hereafter SFD) extinction maps, and then corrected the observed spectra accordingly before applying the template match method. We recall that one can deredden an integrated spectrum and simultaneously estimate the cluster age. However, in order to make the age estimate more robust, we preferred to match reddening corrected cluster spectra with template spectra. Thus, instead of having to handle two variables in the match (reddening and age), we limit it to find only the cluster age.
The maps of BH and SFD are frequently used to estimate the colour excesses of clusters located in the direction towards the Magellanic Clouds (see, e.g., Piatti et al. 2001; Dutra et al. 2001). SFD found that at high-latitude regions, their dust maps correlate well with maps of H I emission, but deviations are coherent in the sky and are especially conspicuous in regions of saturation of HI emission towards denser clouds and of formation of H<sub>2</sub> in molecular clouds. The SMC is quite transparent, the average foreground and internal reddenings being 0.01 and 0.04, respectively (Dutra et al. 2001). The typical reddening towards the SMC estimated from the median dust emission in annuli surrounding the galaxy is $`E`$($`BV`$)$`=0.037`$ (SFD). Therefore, we assume that relatively high SFD values are saturated and we then use the BH values. For clusters with non saturated SFD values, the difference between SFD and BH colour excesses resulted in, at the most, 0.02 mag; the SFD zero-point being made consistent with the BH maps by subtracting 0.02 mag in $`E(BV)_{\mathrm{SFD}}`$. The results are shown in Figs. 1 to 18.
### 3.2 Equivalent width method
Before measuring $`EW`$s in the observed spectra, they were set to the rest-frame according to the Doppler shift of H Balmer lines. Next, the spectra were normalized to F$`{}_{\lambda }{}^{}=1`$ at 5870 Å and smoothed to the typical resolution of the database ($``$ 10-15 Å).
Spectral fluxes at 3860, 4020, 4150, 4570, 4834, 4914 and 6630 Å were used as guidelines in order to define the continuum according to Bica & Alloin (1986). The $`EW`$s of H Balmer, K Ca II, G band (CH) and Mg I (5167 + 5173 + 5184 Å) were measured within the spectral windows defined by Bica & Alloin (1986) and using IRAF task splot. Boundaries for the K Ca II, G band (CH), Mg I, H$`\delta `$, H$`\gamma `$ and H$`\beta `$ spectral windows are, respectively, (3908-3952) Å, (4284-4318) Å, (5156-5196) Å, (4082-4124) Å, (4318-4364) Å, and (4846-4884) Å. Such a procedure has been applied consistently making the $`EW`$s from integrated spectra safely comparable with the well-known cluster database. Table 2 presents these measurements as well as the sum of $`EW`$s of the three metallic lines ($`S_m`$) and of the three Balmer lines H<sub>δ</sub>, H<sub>γ</sub> and H<sub>β</sub> ($`S_h`$). $`S_m`$ and $`S_h`$ are shown to be useful in the discrimination of old, intermediate-age and young systems (Rabin 1982; Dutra et al. 1999, SP). Typical errors of $``$ 10 % on individual $`EW`$ measurements were obtained by tracing slightly different continua. By using the sums of $`EW`$s $`S_h`$ and $`S_m`$ separately, the $`EW`$ relative errors are lowered ($`7\%`$ smaller range than the individual $`EW`$ errors), improving their sensitivity to cluster age and metallicity (SP).
The sums of $`EW`$s $`S_h`$ and $`S_m`$ presented in Table 2 were used to estimate cluster parameters according to their calibrations as a function of age and metallicity given by SP. Such calibrations are based on visible integrated spectra of Galactic and Magellanic Cloud clusters for which age and metallicity were well-determined and put within homogeneous scales. In summary, the calibrations, aided by diagnostic diagrams involving $`S_m`$ and $`S_h`$, allow one to obtain age for star clusters younger than $``$ 10 Gyr and metallicity for older ones. Yet, a degeneracy occurs for globular age-like clusters with $`[Fe/H]>1.4`$ and intermediate-age clusters ($`2.5`$ $`<`$ $`t`$ (Gyr) $`<`$ $`10`$), which cannot be discriminated using this method. In this case, it is necessary to constrain age by using an independent method (e.g., the template matching one) and then obtain metallicity with the SP’s calibration, if the cluster is old. It is worth mentioning that only 5 SMC clusters are included in the SP’s calibration, but since they follow the general trend of Galactic clusters in the diagnostic diagrams, we judged safe to apply that calibration to the present sample. The derived ages and metallicities for the cluster sample are summarized in Table 3. In columns 6 and 9, the methods used to obtain age and metallicity are indicated. Except for K 28, with a low S/N spectrum, all remaining clusters were age-ranked according to the $`EW`$ method based on $`S_h`$ and $`S_m`$ measurements. In the case of NGC 269, we decided to use only $`S_m`$, since the substraction of the spectrum of the symbiotic nova SMC 3 could affect the $`EW`$s of the cluster H Balmer lines (see details in Section 3.3.7). The template method was applied to the whole sample either independently from the $`EW`$ method (minus sign in column 6) or in conjunction with the $`EW`$ method (plus sign in column 6). Note that we only had to employ template and $`EW`$s methods in conjunction for clusters in the age-metallicity degeneracy range. We found a very good agreement between ages derived from both methods. The final cluster ages obtained from the weighted average of values taken from the literature (columns 3 and 4) and the measured present ones ($`t_\mathrm{m}`$) are listed in column 7. Their respective errors take into account the dispersion of the values averaged and/or the estimated uncertainties for $`t_\mathrm{m}`$. Column 2 lists the colour excesses adopted for the clusters.
The last two columns of Table 3 show the cluster metallicities adopted whenever possible and their corresponding sources, respectively. Some clusters have metal abundances directly averaged from published values. For K 3, we used eq. (8) of SP. Three clusters (L 5, K 5 and K 28) have metallicities derived from a technique involving morphological features in the cluster colour-magnitude diagram (CMD) (Piatti et al. 2002b, 2005), which we corrected for age degeneracy using the present ages. Finally, we fitted Padova isochrones (Girardi et al. 2002) to the K 6 CMD obtained by Matteucci et al. (2002) and yielded a cluster metallicity of \[Fe/H\] = -0.7, assuming for the cluster the reddening and age of Table 3 and the SMC apparent distance modulus $`(mM)`$ = 19.0 (Cioni et al 2000). The fit was performed on an extracted CMD containing stars distributed around 2$`\mathrm{}`$ from the cluster centre, with the aim of avoiding field star contamination.
### 3.3 Individual cluster analysis
We have revised the literature on the cluster parameters below. More weight has been assigned to ages determined from isochrone fitting to CMD data, but when such information was not available, ages based on integrated indices were also considered. No previous age information was found either for HW 8 (Fig. 6) or IC 1641 (Fig. 17).
#### 3.3.1 L 5
Piatti et al. (2005) have derived $`t=4.3`$ Gyr and $`[Fe/H]=1.2`$ for this cluster. Much like K 5, the age of L 5 has been estimated to be 0.8 Gyr according to both methods employed in the present work. A correction to the metallicity provided by Piatti et al. (2005) revised it to $`[Fe/H]=1.1`$ for its significantly younger age. Fig. 1 shows the best template combination for L 5, i. e., the average of Ia and Yh templates with a reddening of $`E(BV)`$ = 0.03. This is the cluster with the most discrepant age in the sample with respect to the published cluster ages. We did not find any reason for such difference, apart from a relative low S/N ratio in the observed spectrum.
#### 3.3.2 K 5
Bica et al. (1986) derived for K 5 the following parameters from integrated photometry of the H$`\beta `$ and G band absorption features: $`[Z/Z_{}]=1.1\pm 0.2`$ and $`t=3.2\pm 0.3`$ Gyr, while the recent study by Piatti et al. (2005) yields $`[Fe/H]=0.6`$ and $`t=2.0`$ Gyr. The template method estimate for K 5 age is $`t=0.8`$ Gyr, according to its spectral resemblance to an average of templates Ia and Yh, after applying a reddening correction of E(B-V) = 0.02 (Fig. 2). Its metallicity has been corrected to $`[Fe/H]=0.5`$, following an age revision on the Piatti et al. (2005) value.
#### 3.3.3 K 3
Rich et al. (1984) determined an age of 5-8 Gyr from BR photometry and isochrone fitting. K 3 was included in the integrated photometric study by Bica et al. (1986), who derived $`t`$ 10 Gyr and $`[Z/Z_{}]=1.5\pm 0.2`$. Mighell et al. (1998) obtained $`[Fe/H]=1.16\pm 0.09`$, $`t=6.0\pm 1.3`$ Gyr and $`E(BV)`$ = 0.0 from HST observations and morphological parameters defined in the CMD. More recently, Brocato et al. (2001) presented a HST CMD of K 3 making available its photometry, on which we have superimposed Padova isochrones (Girardi et al. 2002) to obtain essentially the same parameters as those derived by Mighell et al. (1998). In the present study, an intermediate age for K 3 is confirmed, being this the oldest cluster in the present sample. The template matching method gives for this cluster $`7`$ Gyr as a result of averaging the G3 and Ia templates (Fig. 3). Both age and metallicity obtained in the present analysis show good agreement with results from previous studies.
#### 3.3.4 K 6
From CCD $`BV`$ photometry selected for an inner region (r $`<`$ 35$`\mathrm{}`$) of K 6, Matteucci et al. (2002) derived an age of 1-1.3 Gyr for this cluster. The spectrum comparison leads to a match of K 6 spectrum with the template Ib (3-4 Gyr), combined with a reddening correction of $`E(BV)`$ = 0.03 (Fig. 4). However, a smaller age is suggested by the $`EW`$ method, being $`t`$ = 1.6 Gyr the final adopted value. By fitting Padova isochrones (Girardi et al. 2002) to the CMD data of Matteucci et al. (2002) and assuming the above mentioned age and the apparent distance modulus $`(mM)`$ = 19 (Cioni et al 2000), an estimate of the cluster metallicity was also obtained, i. e., \[Fe/H\] = -0.7.
#### 3.3.5 K 7
Mould et al. (1992) carried out CCD $`BR`$ photometry of K 7 obtaining $`t`$ = 3.5 $`\pm `$ 1 Gyr by isochrone fitting with $`E(BV)`$ = 0.04. The template spectrum Ib (3-4 Gyr) was initially tried as a match to the K 7 spectrum, but its redder colour cannot be accounted for by a large reddening correction exclusively. Mould et al. (1992) pointed out the presence of two carbon stars close to the cluster centre, which are the probable contributors to the red appearance of its integrated spectrum. In order to check whether this is the case, a combination of the Ib spectrum with a carbon star spectrum taken from Barnbaum et al. (1996) spectral library was tried. Specifically, the spectrum of the nearly solar metallicity carbon star BM Gem (Abia & Isern 2000) was used in the analysis. According to our observations, in the cluster spatial profile the presence of the bright star stands out over the bulk of the cluster light. We then extracted the integrated spectrum of the cluster plus the carbon star and of the carbon star spectrum alone. The flux ratio at 5870 Å between the carbon star spectrum and the integrated one turned out to be 0.35. As a matter of fact, there is a good match to K 7 spectrum if the template Ib is combined with the carbon star in a proportion of 65% and 35% of the total light at 5870Å, respectively, and the resulting spectrum is reddening corrected by $`E(BV)`$ = 0.02 (Fig. 5). Relatively large residual spectral differences still remain between the spectra, which may be attributed to the higher metallicity of the carbon star employed as template. No metallicity has been estimated for this cluster.
#### 3.3.6 NGC 269
This is an interesting case in which there is a bright emission line star contributing significantly to the cluster integrated spectrum. Such a situation, which we had found in previous cluster observations (e.g. Santos et al. 1995), has been successfully treated by subtracting the star spectrum from the total integrated one, leaving a spectrum which better represents the cluster average population. Although such a procedure introduced noise in the resulting spectrum, it allowed us to estimate the cluster age using the template matching method. The bright star in NGC 269 spectrum is SMC 3, a symbiotic nova composed by a M0 giant and a white dwarf orbiting each other in a period of $`4`$ years (Kahabka 2004). Its spectrum was published in the spectrophotometric atlas of Munari & Zwitter (2002). The OGLE database includes a CMD for this cluster (OGLE-CL-SMC0046), although an age estimate was not provided there (Pietrzynski et al. 1998; Pietrzynski & Udalski 1999). This CMD shows that SMC 3 is $``$ 2 mags brighter in $`V`$ than the next bright star in the cluster. In detail, the procedure adopted was to subtract the total integrated spectrum from a scaled SMC 3 spectrum by assuming that all the emission present in the integrated spectrum is due to SMC 3. In this manner, the difference between the spectra which minimizes the emission line residuals was obtained when the star contributes with 60% of the total flux at 5870 Å. Since the spectra were observed at different epochs and SMC 3 is variable, the small but clearly visible residuals reflect such irregularities. Another point that allows one to check the reliability of this procedure is the fact that absorption molecular bands present in SMC 3 spectrum almost disappear in the resulting spectrum. The subtracted spectrum was then submitted to the template matching method (Fig. 7), being similar to an average of the templates Yh and Ia (750 Myr). Such an age is in agreement with the clusters SWB type III-IV (Searle et al. 1980). González et al. (2004) have assigned an age of 500 Myr to NGC 269, based on the integrated colour parameterization (“s” parameter) by Elson & Fall (1988). However, it should be kept in mind that González et al. (2004) age ranking is intended to group clusters of similar integrated properties and their age groups encompass wide age ranges.
#### 3.3.7 L 39
The OGLE database includes a CMD for this cluster (OGLE-CL-SMC0054), with an isochrone based age estimate of 100 $`\pm `$ 23 Myr (Pietrzynski et al. 1998; Pietrzynski & Udalski 1999). A new age estimate based both on a different areal extraction of the same data and on the same isochrones revises it down to 80 $`\pm `$ 20 Myr (de Oliveira et al. 2000). According to González et al. (2004), L 39 is similar to 50 Myr old clusters. We have found that the cluster is 15 $`\pm `$ 10 Myr old, its steep continuum resembling those of L 51 and L 66 (Fig. 8).
#### 3.3.8 K 28
Piatti et al. (2001) obtained CCD Washington photometry for this cluster deriving \[Fe/H\] = -1.45 $`\pm `$ 0.13 and $`t`$ = 2.1 $`\pm `$ 0.5 Gyr, with a reddening within the range 0.06 $`<`$ $`E(BV)`$ $`<`$ 0.16. In the present analysis, we have not applied the $`EW`$ method to derive parameters for K 28 because its integrated spectrum has low S/N ratio, although it still seems to be adequate to the template matching method. Indeed, using the latter, we have got a good match for the template Ia (1 Gyr) combined with a reddening of $`E(BV)`$ = 0.06 (Fig. 9). By revising down the age obtained by Piatti et al. (2001), a corrected metallicity of \[Fe/H\] = -1.0 was derived.
#### 3.3.9 NGC 294
This cluster (OGLE-CL-SMC0090) has a CMD included in the OGLE database. Its estimated age based on the isochrone fitting method is 316 $`\pm `$ 73 Myr (Pietrzynski et al. 1998; Pietrzynski & Udalski 1999). The study published by de Oliveira et al. (2000) gives 300 $`\pm `$ 50 Myr. Since there is no SWB type assigned to this cluster, González et al. (2004) included NGC 294 in their 1 Gyr cluster group due to the similarity of its integrated colours to the colours of SWB IV clusters. This age seems to be in disagreement with the previously mentioned works and also with ours, which gives $`t`$ = 300 $`\pm `$ 100 Myr (Fig. 10). Visible cluster images do not show any bright star in the cluster core, and therefore the age discrepancy cannot be due to sampling effects.
#### 3.3.10 L 51
This cluster has spectral similarities with L 66, for which a CMD is available (see below). González et al. (2004) have assigned an age of 10 Myr to L 51 based on the “s” parameter (Elson & Fall 1988), in agreement with our estimate of $`t`$ = 15$`\pm `$ 10 Myr (Fig. 11).
#### 3.3.11 K 42
The OGLE database includes a CMD for this cluster (OGLE-CL-SMC0124), with an isochrone based age estimate of 39.8$`\pm `$ 9.2 Myr (Pietrzynski et al. 1998; Pietrzynski & Udalski 1999). The study by de Oliveira et al. (2000) obtained for it a younger age, 20 $`\pm `$ 10 Myr, which seems too low compared to our estimate, $`t`$ = 45 $`\pm `$ 15 Myr (Fig. 12).
#### 3.3.12 L 66
The OGLE database includes a CMD for this cluster (OGLE-CL-SMC0129), with an isochrone based age of 20.0 $`\pm `$ 4.6 Myr (Pietrzynski et al. 1998; Pietrzynski & Udalski 1999). This result is comparable to our age estimate of $`t`$ = 15 $`\pm `$ 10 Myr (Fig. 13).
#### 3.3.13 NGC 411
The spectral features and continuum slope of NGC 411 are comparable to the template spectrum Ia (1 Gyr), when a reddening correction of $`E(BV)`$ = 0.03 is applied to the observed spectrum (Fig. 14). Although this age estimate is lower than that obtained by Bica et al. (1986), i.e., 3.4 $`\pm `$ 0.3 Gyr (and $`[Z/Z_{}]=1.3\pm 0.2`$), the present result agrees with the study by Leonardi & Rose (2003) involving integrated spectroscopy at a higher resolution who obtained $`t`$ = 1.2 $`\pm `$ $`0.2`$ Gyr and \[Fe/H\] = -0.43 $`\pm `$ 0.14. In addition, two studies involving isochrone fittings to CMDs yield similar results: Da Costa & Mould (1986) obtain $`t`$ = 1.5 $`\pm `$ 0.5 Gyr and $`[Fe/H]`$ = -0.9 $`\pm `$ 0.3, adopting a reddening of $`E(BV)`$ = 0.04 and Alves & Sarajedini (1999) determined $`t`$ = 1.4 $`\pm `$ 0.2 Gyr and $`[Fe/H]`$ = -0.68 $`\pm `$ 0.07, based on HST data. By using CCD Washington photometry, a metallicity of $`[Fe/H]`$ = -0.84, which agrees with the previous, more recent estimates, was derived for this cluster by Piatti et al. (2002b).
#### 3.3.14 NGC 419
An age lower limit of 1 Gyr was obtained for this cluster (OGLE-CL-SMC0159) by isochrone fitting to OGLE data (Pietrzynski et al. 1998; Pietrzynski & Udalski 1999). This age limit is in accord with the other recent age estimates by Rich et al. (2000), who obtained 2.0 $`\pm `$ 0.2 Gyr by means of isochrone fitting to the cluster HST CMD, and by Durand et al. (1984), who estimated 1.2 $`\pm `$ 0.5 Gyr by using isochrone fitting to photographic CMDs. The template matching method yields $`t`$ = 0.75 Gyr (Fig. 15), but when this value is combined with an independent estimate from the $`EW`$ method, the age converges to 1.2 $`\pm `$ 0.4 Gyr, in close agreement with the result found by Durand et al. (1984). This age is consistent with \[Fe/H\] = -0.7 (Piatti et al. 2002b).
#### 3.3.15 NGC 422
As far as we know, the only age information on this cluster is based on its integrated colours. González et al. (2004) have considered NGC 422 as a member of the 50 Myr group (SWB type II), which is younger than our estimate of $`t`$ = 400 Myr (Fig. 16), based on the template method. The final adopted age for NGC 422 is $`t`$ = 300 $`\pm `$ 100 Myr, which results from the independent methods employed in the present work.
#### 3.3.16 NGC 458
Age determinations of NGC 458 based on isochrone fitting to the cluster CMD were made by Papenhausen & Schommer (1988), Stothers & Chin (1992) and more recently by Alcaino et al. (2003), who obtained 300 Myr, 100 Myr and 140 Myr, respectively. Independent age estimates using the template matching (Fig. 18) and the $`EW`$ methods give $`t`$ = 50 Myr. Taking into account the previous literature determinations, a final value of $`t`$ = 130 $`\pm `$ 60 Myr was adopted, which is consistent with \[Fe/H\] = -0.23 (Piatti et al. 2002b).
## 4 Discussion
Fig. 19 shows the positions of the studied clusters (crossed boxes) relative to the SMC optical centre (cross), assumed to be placed (J2000) at: 00<sup>h</sup> 52<sup>m</sup> 45<sup>s</sup>, -72$`\mathrm{°}`$ 49$`\mathrm{}`$ 43$`\mathrm{}`$ (Crowl et al. 2001). For the sake of completeness, we included 19 additional clusters (triangles) taken from Table 4 of Piatti et al. (2002b) and studied by Piatti et al. (2005), which have ages and metallicities put onto a homogeneous scale. The collection of these 37 objects constitutes at the present time the largest sample of SMC clusters used to address the issue of the galaxy chemical evolution. Thus, the results derived from this sample are valuable in the sense that they give us the opportunity to have some clues about the galaxy history, which obviously needs later confirmation from a larger database. Besides the SMC Bar, represented by a straight line in Fig. 19, we traced two ellipses centred at the SMC optical centre with their major axes aligned with the galaxy Bar. We adopted a $`b/a`$ ratio which equals to 1/2. The semi-major axes of the ellipses drawn in the figure have 2$`\mathrm{°}`$ and 4$`\mathrm{°}`$, respectively. Note that this elliptical geometry matches the space distribution of clusters more properly than a circular one.
When describing the cluster age and metallicity distributions, the interpretation of the results can depend on the spatial framework used. For example, one can adopt as a reference system the one corresponding to the right ascension and declination axes, or that centred on the galaxy with a coordinate axis parallel to the Bar. Thus, if there existed an abundance gradient from the centre and along the SMC Bar, its projection to the right ascension and declination axes would appear steeper. Similarly, it could be possible to affirm the existence of features which are actually the result of projection effects on these directions. By considering the distances of the clusters from the SMC centre instead of their projections onto the right ascension and declination axes, the genuine cluster age and metal abundance variations can be traced. Moreover, although it may be advantageous to plot ages and metallicities as a function of the distance from the galaxy centre, these plottings can result even more meaningful when the spatial variable reflects the flattening of the system. In the case of the SMC, this can be accomplished by using ellipses instead of circles around the SMC centre.
In order to examine how the cluster ages vary in terms of the distances from the SMC centre, we computed for each cluster the value of the semi-major axis ($`a`$) that an ellipse would have if it were centred at the SMC centre, had a $`b/a`$ ratio of 1/2, and one point of its trajectory coincided with the cluster position. Fig. 20 shows the result obtained, in which we used the same symbols as in Fig. 19. The figure reveals that there are very few clusters younger than 4 Gyr in the outer disk, defined as the portion of the SMC disk with $`a`$ $``$ 3.5$`\mathrm{°}`$. Conversely, it would appear that there are very few clusters older than 4 Gyr in the inner disk. Furthermore, in the inner disk, the older the clusters, the larger their corresponding semi-major axes, which astonishingly suggests the possibility that the clusters were formed outside in, like in a relatively rapid collapse. As far as we are aware, this is the first time such an evidence is presented.
Harris & Zaritsky (2004) recently determined the global star formation and chemical enrichment history of the SMC within the inner 4$`\mathrm{°}`$x4.5$`\mathrm{°}`$ area of the main body, based on UBVI photometry of $``$ 6 million stars from their Magellanic Clouds Photometric Survey (Zaritsky et al. 1997). Among other results, they found that there was a rise in the mean star formation rate during the most recent 3 Gyr punctuated by bursts at 2.5 Gyr, 400 Myr, and 60 Myr. The two older events coincide with past perigalactic passages of the SMC around the Milky Way (see, e.g., Lin et al. 1995). In addition, Harris & Zaritsky (2004) derived a chemical enrichment history in agreement with the age-metallicity relation of the SMC clusters and field variable stars. This chemical enrichment history is consistent with the model of Pagel & Tautvaisiene (1999), lending further support to the presence of a long quiescent period (3 $`<`$ age(Gyr) $`<`$ 8.4) in the SMC early history. Piatti et al. (2005) confirmed that $``$ 2.5 Gyr ago the SMC reached the peak of a burst of cluster formation, which corresponds to a very close encounter with the LMC according to recent dynamic models of Bekki et al. (2004). It would seem reasonable, therefore, to accept that the burst which took place 2 Gyr ago formed both clusters and stars simultaneously. Particularly, the 2.5 Gyr star burst appears to have an annular structure and an inward propagation spanning $``$ 1 Gyr (Harris & Zaritsky 2004).
Piatti et al. (2005) studied 10 clusters mainly located in the southern half of the SMC with ages and metallicities in the ranges 1.5 - 4 Gyr and -1.3 $`<`$ \[Fe/H\] $`<`$ -0.6, respectively. They also favoured a bursting cluster formation history as opposed to a continuous one for the SMC. The age-position relation shown in Fig. 20 for clusters younger than 4 Gyr adds, if it is confirmed, a new piece of evidence to the bursting conception of cluster formation. In the case of the cluster formation episode peaking at $``$ 2.5 Gyr (Piatti et al. 2005), the burst could have triggered the formation process which continued producing clusters from the outermost regions to the innermost ones in the inner SMC disk. On this basis, the inner disk could have been formed during this period.
The distribution of the cluster metal abundances as a function of the distances from the SMC centre is depicted in Fig. 21, where we used the same symbols as in Fig. 19. Note that in the outer disk, there are no clusters with iron-to-hydrogen ratios larger than \[Fe/H\] = -1.2, with only one exception. On the other hand, the inner disk is shared by both metal-poor and metal-rich clusters, the averaged metallicity being clearly larger than that for the outer disk. We thus confirm the existence of a metal abundance gradient for the SMC disk, in the sense that the farther a cluster from the galaxy centre, the poorer its metal content. However, all the clusters with \[Fe/H\] $`>`$ -1.2 in the inner disk were formed during the last 4 Gyr, whereas the metal-poor ones are as old as those in the outer disk (see Fig. 20). Consequently, the abundance gradient seems to reflect the combination between an older and more metal-poor population of clusters spread throughout the SMC and a younger and metal-richer one mainly formed in the inner disk. Note that some few clusters were also formed in the inner disk with \[Fe/H\] $``$ -1.2 (Fig. 21). We also recall that the present cluster sample follows the age-metallicity relation discussed in a previous work (Piatti et al. 2005, see their Figure 6).
## 5 Concluding remarks
As part of a systematic spectroscopic survey of star clusters in the SMC, we present and analise in the current paper flux-calibrated integrated spectra of 18 concentrated star clusters which, with a few exceptions, lie within the inner parts of the SMC. The sample of SMC clusters studied by means of integrated spectroscopy has now been considerably increased. Therefore, the present cluster spectral library at the SMC metallicity level can be useful for future analyses of star clusters in dwarf galaxies as well as for the study of a fraction of star clusters in massive galaxies.
E(B-V) colour excesses were derived for the present cluster sample by interpolation between the extinction maps published by Burstein & Heiles (1982) and by Schlegel et al. (1998). Using template spectra with well determined cluster properties and equivalent widths (EWs) of the Balmer and several metallic lines, we determined ages and, in some cases, metallicities as well. For the SMC clusters HW 8 and IC 1641, the ages have been determined for the first time, while for the rest of the studied sample the ages derived from the template matching and EW methods exhibit very good agreement. Metal abundances have been derived for five clusters (L 5, K 5, K 3, K 6 and K 28), while we have adopted averaged metallicities from published values for other 4 clusters (K 7, NGC 411, NGC 419 and NGC 458). By combining the present cluster sample with 19 additional SMC clusters with ages and metallicities in a homogeneous scale, we analise the age and metallicity distributions in different regions of the SMC to probe the galaxy chemical enrichment and its spatial distribution. Very few clusters younger than 4 Gyr are found in the outer disk and, conversely, very few clusters older than 4 Gyr lie in the inner disk. Furthermore, the present age-position relation for the SMC clusters in the inner disk suggests not only the possibility that the clusters were formed outside in, like in a relatively rapid collapse, but also that the inner disk itself could have been formed during a bursting formation mechanism, with an important cluster formation event centred at $``$ 2.5 Gyr. According to the recent results obtained by Harris & Zaritsky (2004), this cluster burst, which occurred $``$ 2.5 Gyr ago, is clearly related to an episode of enhanced star formation having taken place about the same time ago. Evidence is also presented on the existence of a radial metal abundance gradient for the SMC disk, which reflects the combination between an older and more metal-poor population of clusters distributed throughout the SMC and a younger and metal-richer one mainly formed in the inner disk.
###### Acknowledgements.
We are grateful for the use of the CCD and data acquisition system at CASLEO, supported under US National Science Foundation (NSF) grant AST-90-15827. This work is based on observations made at CTIO, which is operated by AURA, Inc., under cooperative agreement with the NSF. We thank the staff members and technicians at CASLEO and CTIO for their kind hospitality and assistance during the observing runs. We gratefully acknowledge financial support from the Argentinian institutions CONICET, Agencia Nacional de Promoción Científica y Tecnológica (ANPCyT) and Agencia Córdoba Ciencia. We thank Dr. Munari for sending us the spectrum of SMC 3. This work was also partially supported by the Brazilian institution FAPEMIG and CNPq. |
warning/0507/hep-ph0507250.html | ar5iv | text | # 1 Introduction
## 1 Introduction
Extra dimension scenarios are now essential part of the studies of physics beyond the Standard Model (SM). They provide an alternate view of the hierarchy between the electroweak and the Planck scale. Some of these extra dimension models invoke the brane world scenarios to hide the extra spacial dimensions from current observation. Two such models that are phenomenologically widely studied are the Arkani-Hamed, Dimopoulos and Dvali (ADD) and the Randall-Sundrum (RS) models.
In the ADD case the compactified extra dimensions could be large and the large volume of the compactified extra spacial dimension would account for the dilution of gravity in 4-dimensions and hence the hierarchy. In this model, new physics can appear at a mass scale of the order of a TeV. A viable mechanism to hide the extra spacial dimension, is to introduce a 3-brane with negligible tension and localise the SM particles on it. Only the graviton is allowed to propagate the full $`4+d`$ dimensional space time. As a consequence of these assumptions, it follows from Gauss Law that the effective Planck scale $`M_P`$ in 4-dimension is related to the $`4+d`$ dimensional fundamental scale $`M_S`$ through the volume of the compactified extra dimensions . The extra dimensions are compactified on a torus of common circumference $`R`$. The number of extra spacial dimension possible is $`d>2`$ from current experimental limits on deviation from inverse square law . The space time is factorisable and the 4-dimensional spectrum consists of the SM confined to 4-dimensions and a tower of Kaluza-Klien (KK) modes of the graviton propagating the full $`4+d`$ dimensional space time.
The interaction of the KK modes $`h_{\mu \nu }^{(\stackrel{}{n})}`$ with the SM fields localised on the 3-brane is given by
$`_{int}{\displaystyle \frac{1}{M_P}}{\displaystyle \underset{\stackrel{}{n}=0}{\overset{\mathrm{}}{}}}T^{\mu \nu }(x)h_{\mu \nu }^{(\stackrel{}{n})}(x),`$ (1.1)
where $`T^{\mu \nu }`$ is the energy-momentum tensor of the SM fields on the 3-brane. The zero mode corresponds to the usual 4-dimensional massless graviton. The KK modes are all $`M_P`$ suppressed but the high multiplicity could lead to observable effects at present and future colliders. The Feynman rules are given in .
In the RS model there is only one extra spacial dimension and the extra dimension is compactified to a circle of circumference $`2L`$ and further orbifolded by identifying points related by $`yy`$. Two branes are placed at orbifold fixed points, $`y=0`$ with positive tension called the Planck brane and a second brane at $`y=L`$ with negative tension called the TeV brane. For a special choice of parameters, it turns out that the 5-dimensional Einstein equations have a warped solution for $`0<y<L`$ with metric $`g_{\mu \nu }(x^\rho ,y)=\mathrm{exp}(2ky)\eta _{\mu \nu }`$, $`g_{\mu y}=0`$ and $`g_{yy}=1`$. This space is not factorisable and has a constant negative curvature— $`AdS_5`$ space-time. $`k`$ is the curvature of the $`AdS_5`$ space-time and $`\eta _{\mu \nu }`$ is the usual 4-dimensional flat Minkowski metric. In this model the mass scales vary with $`y`$ according to the exponential warp factor. If gravity originates on the brane at $`y=0`$, TeV scales can be generated on the brane at $`y=L`$ for $`kL10`$. The apparent hierarchy is generated by the exponential warp factor and no additional large hierarchies appear. The size of the extra dimension is of the order of $`M_P^1`$. Further it has been showed that the value of $`kL`$ can be stabilised without fine tuning by minimising the potential for the modulus field which describes the relative motion of the two branes. In the RS model graviton and the modulus field can propagate the full 5-dimensional space time while the SM is confined to the TeV brane. The 4-dimensional spectrum contains the KK modes, the zero mode is $`M_P`$ suppressed while the excited modes are massive and are only TeV suppressed. The mass gap of the KK modes is determined by the difference of the successive zeros of the Bessel function $`J_1(x)`$ and the scale $`m_0=ke^{\pi kL}`$. As in the ADD case the phenomenology of the RS model concerns the effect of massive KK modes of the graviton, though the spectrum of the KK mode is quite different.
In the RS model the massive KK modes $`h_{\mu \nu }^{(n)}(x)`$ interacts with the SM fields
$`_{int}{\displaystyle \frac{1}{M_P}}T^{\mu \nu }(x)h_{\mu \nu }^{(0)}(x){\displaystyle \frac{e^{\pi kL}}{M_P}}{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}T^{\mu \nu }(x)h_{\mu \nu }^{(n)}(x),`$ (1.2)
where $`T^{\mu \nu }`$ is the energy-momentum tensor of the SM fields on the 3-brane at $`y=L`$. The masses of $`h_{\mu \nu }^{(n)}(x)`$ are given by $`M_n=x_nke^{\pi kL}`$, where $`x_n`$ are the zeros of the Bessel function $`J_1(x)`$. In the RS model there are two parameters which are $`c_0=k/M_P`$, the effective coupling An alternate definition is $`\overline{c}_0=k/\overline{M}_P`$, where $`\overline{M}_P=M_P/\sqrt{8\pi }`$, hence $`\overline{c}_0=c_0\sqrt{8\pi }`$. and $`M_1`$ the mass of the first KK mode. Expect for an overall warp factor the Feynman rule of RS is the same as those of the ADD model.
Next to leading order (NLO) QCD corrections have been recently calculated in the ADD case for $`e^+e^{}`$ hadrons and various distributions of invariant lepton pair production at both LHC and Tevatron . This was further extended to the RS case . In this paper, for the ADD and RS models, we consider the un-integrated distribution with respect to $`\mathrm{cos}\theta ^{}`$ to NLO in QCD, where $`\theta ^{}`$ is the scattering angle of the lepton with an initial hadron in the c.o.m frame of the lepton pair. This is particularly important in the dilepton production case to achieve maximum sensitivity to the model parameters, as $`\mathrm{cos}\theta ^{}`$ integrated cross section is independent of the interference between SM and gravity . To leading order (LO), this double differential $`d\sigma /dQ/d\mathrm{cos}\theta ^{}`$ was analysed in . At hadron colliders the NLO-QCD corrections are important especially in models of extra dimension as gluon-gluon subprocess contributes at the same LO as quark-antiquark subprocess. DØ Collaboration recently reported searches for large extra dimensions in the dimuon channel for the double differential cross section , this updates the earlier Run-I results . The first direct search of the RS KK modes using the dileptons have been reported by DØ Collaboration .
Rest of the paper is organised as follows: In section 2 we evaluate the NLO coefficient functions to the subprocess that contribute to the double differential cross section $`d\sigma /dQ/d\mathrm{cos}\theta ^{}`$. Finally in section 3, we discuss the impact of the NLO results.
## 2 Drell-Yan $`\mathrm{𝐜𝐨𝐬}𝜽^{\mathbf{}}`$ distribution
We consider the Drell-Yan process and study the double differential cross section with respect to the invariant mass of the final lepton pair and $`\mathrm{cos}\theta ^{}`$ the cosine of the angle between the final state lepton momenta and the initial state hadron in the $`c.o.m`$ frame of the lepton pair <sup>§</sup><sup>§</sup>§An alternate definition of the angle has been considered in to study the lepton helicity distribution in polarised Drell-Yan process.. The relevant kinematical formulation is detailed in . In the QCD improved parton model, the hadronic cross section can be expressed in terms of partonic cross sections convoluted with appropriate parton distribution functions. The coefficient functions to NLO in QCD are evaluated for both ADD and RS models. The difference between the two models depend on the spectrum of the KK modes and hence summation of the KK modes that contribute to the dilepton production leads to different results .
The hadronic part involves the computation of various processes that contribute to $`Q`$ or $`X_F`$ or rapidity distributions that are presented in the reference . The angular distributions which are ”odd” in $`\mathrm{cos}\theta ^{}`$ come mainly from the interferences terms. The non-vanishing odd contribution in the standard model sector comes from the interference of photon mediated processes with $`Z`$-boson mediated processes. We also find that non-vanishing odd contributions come from the interference of standard model diagrams with the graviton exchange diagrams. These inference diagrams are absent in the computation of $`Q,X_F`$ and rapidity distributions where only even functions of $`\mathrm{cos}\theta ^{}`$ contribute. We have regularised all the divergences using dimensional regularisation. The mass singularities are removed by the mass factorisation, for details refer to .
We first present the angular distribution which is ”even” in $`\mathrm{cos}\theta ^{}`$.
$`2S{\displaystyle \frac{d\sigma _e^{P_1P_2}}{dQ^2d\mathrm{cos}\theta ^{}}}`$ $`=`$ $`{\displaystyle \underset{q}{}}_{SM,q}{\displaystyle _0^1}𝑑x_1{\displaystyle _0^1}𝑑x_2{\displaystyle _0^1}𝑑z\delta (\tau zx_1x_2)`$
$`\times [H_{q\overline{q}}(x_1,x_2,\mu _F^2)(\mathrm{\Delta }_{q\overline{q}}^{(0),\gamma /Z}(z,Q^2,\mu _F^2)+a_s\mathrm{\Delta }_{q\overline{q}}^{(1),\gamma /Z}(z,Q^2,\mu _F^2))`$
$`+H_{qg}(x_1,x_2,\mu _F^2)a_s\mathrm{\Delta }_{qg}^{(1),\gamma /Z}(z,\mu _F^2)`$
$`+H_{gq}(x_1,x_2,\mu _F^2)a_s\mathrm{\Delta }_{gq}^{(1),\gamma /Z}(z,\mu _F^2)]`$
$`+{\displaystyle \underset{q}{}}_{SMGR,q}{\displaystyle _0^1}𝑑x_1{\displaystyle _0^1}𝑑x_2{\displaystyle _0^1}𝑑z\delta (\tau zx_1x_2)`$
$`\times [H_{q\overline{q}}(x_1,x_2,\mu _F^2)(\mathrm{\Delta }_{q\overline{q}}^{(0),G\gamma /Z}(z,Q^2,\mu _F^2)+a_s\mathrm{\Delta }_{q\overline{q}}^{(1),G\gamma /Z}(z,Q^2,\mu _F^2))`$
$`+H_{qg}(x_1,x_2,\mu _F^2)a_s\mathrm{\Delta }_{qg}^{(1),G\gamma /Z}(z,\mu _F^2)`$
$`+H_{gq}(x_1,x_2,\mu _F^2)a_s\mathrm{\Delta }_{gq}^{(1),G\gamma /Z}(z,\mu _F^2)]`$
$`+{\displaystyle \underset{q}{}}_{GR}{\displaystyle _0^1}𝑑x_1{\displaystyle _0^1}𝑑x_2{\displaystyle _0^1}𝑑z\delta (\tau zx_1x_2)`$
$`\times [H_{q\overline{q}}(x_1,x_2,\mu _F^2)(\mathrm{\Delta }_{q\overline{q}}^{(0),G}(z,Q^2,\mu _F^2)+a_s\mathrm{\Delta }_{q\overline{q}}^{(1),G}(z,Q^2,\mu _F^2))`$
$`+H_{qg}(x_1,x_2,\mu _F^2)a_s\mathrm{\Delta }_{qg}^{(1),G}(z,Q^2,\mu _F^2)`$
$`+H_{gq}(x_1,x_2,\mu _F^2)a_s\mathrm{\Delta }_{gq}^{(1),G}(z,Q^2,\mu _F^2)`$
$`+H_{gg}(x_1,x_2,\mu _F^2)(\mathrm{\Delta }_{gg}^{(0),G}(z,Q^2,\mu _F^2)+a_s\mathrm{\Delta }_{gg}^{(1),G}(z,Q^2,\mu _F^2))],`$
where $`H_{ab}(x_1,x_2,\mu _F^2)`$ are the renormalised partonic distributions and $`\mathrm{\Delta }_{ab}(z,Q^2,\mu _F^2)`$ are the coefficient function corresponding to various subprocess cross section to NLO in QCD. The factors $`_{SM,q},_{GR}`$ correspond to pure SM and gravity (GR) part respectively and are given in , the factor that corresponds to the interference of SM and gravity is
$`_{SMGR,q}`$ $`=`$ $`{\displaystyle \frac{\alpha \kappa ^2Q^2}{4\pi }}|𝒟(Q^2)|\left[{\displaystyle \frac{Q^2(Q^2M_Z^2)}{\left((Q^2M_Z^2)^2+M_Z^2\mathrm{\Gamma }_Z^2\right)c_w^2s_w^2}}g_q^Ag_e^A\right],`$ (2.2)
where $`\alpha `$ is the fine structure constant, $`\kappa =\sqrt{16\pi }/M_P`$. The summation of the KK modes leads to $`𝒟(Q^2)`$ for ADD and RS case has been given in .
The leading order results read
$`\mathrm{\Delta }_{q\overline{q}}^{(0),\gamma /Z}`$ $`=`$ $`{\displaystyle \frac{2\pi }{N}}\delta (1z)\left[{\displaystyle \frac{3}{8}}\left(1+\mathrm{cos}^2\theta ^{}\right)\right],`$
$`\mathrm{\Delta }_{q\overline{q}}^{(0),G\gamma /Z}`$ $`=`$ $`{\displaystyle \frac{\pi }{8N}}\delta (1z)\left[1+3\mathrm{cos}^2\theta ^{}\right],`$
$`\mathrm{\Delta }_{q\overline{q}}^{(0),G}`$ $`=`$ $`{\displaystyle \frac{\pi }{8N}}\delta (1z)\left[{\displaystyle \frac{5}{8}}(13\mathrm{cos}^2\theta ^{}+4\mathrm{cos}^4\theta ^{})\right],`$
$`\mathrm{\Delta }_{gg}^{(0),G}`$ $`=`$ $`{\displaystyle \frac{\pi }{2(N^21)}}\delta (1z)\left[{\displaystyle \frac{5}{8}}(1\mathrm{cos}^4\theta ^{})\right],`$ (2.3)
and the next to leading order results read
$`\mathrm{\Delta }_{q\overline{q}}^{(1)\gamma /Z}`$ $`=`$ $`\left({\displaystyle \frac{2\pi }{N}}\right)4C_F\{[(4+2\zeta (2))\delta (1z)+2{\displaystyle \frac{1}{(1z)_+}}\mathrm{ln}\left({\displaystyle \frac{Q^2}{\mu _F^2}}\right)`$
$`+4\left({\displaystyle \frac{\mathrm{ln}(1z)}{1z}}\right)_++{\displaystyle \frac{3}{2}}\delta (1z)\mathrm{ln}\left({\displaystyle \frac{Q^2}{\mu _F^2}}\right)(1+z)\mathrm{ln}\left({\displaystyle \frac{Q^2(1z)^2}{\mu _F^2z}}\right)`$
$`2{\displaystyle \frac{\mathrm{ln}(z)}{1z}}]\left({\displaystyle \frac{3}{8}}(1+\mathrm{cos}^2\theta ^{})\right)+[1z]{\displaystyle \frac{3}{8}}(13\mathrm{cos}^2\theta ^{})\},`$
$`\mathrm{\Delta }_{q(\overline{q})g}^{(1)\gamma /Z}`$ $`=`$ $`\left({\displaystyle \frac{2\pi }{N}}\right)T_F\{[4z\mathrm{log}(z)+2(12z+2z^2)\mathrm{ln}\left({\displaystyle \frac{Q^2(1z)^2}{\mu _F^2z}}\right)7z^2]`$
$`\times \left({\displaystyle \frac{3}{8}}(1+\mathrm{cos}^2\theta ^{})\right)+\left[{\displaystyle \frac{15}{8}}+{\displaystyle \frac{3}{4}}z+3z\mathrm{log}(z)\right]`$
$`+[{\displaystyle \frac{33}{8}}+{\displaystyle \frac{27}{4}}z3z\mathrm{log}(z)]\mathrm{cos}^2\theta ^{}\},`$
$`\mathrm{\Delta }_{gq(\overline{q})}^{(1)\gamma /Z}`$ $`=`$ $`\left({\displaystyle \frac{2\pi }{N}}\right)T_F\{[4z\mathrm{log}(z)+2(12z+2z^2)\mathrm{ln}\left({\displaystyle \frac{Q^2(1z)^2}{\mu _F^2z}}\right)]`$
$`\times \left({\displaystyle \frac{3}{8}}(1+\mathrm{cos}^2\theta ^{})\right)+\left[{\displaystyle \frac{3}{8}}{\displaystyle \frac{3}{4}}z+{\displaystyle \frac{3}{8}}z^2{\displaystyle \frac{3}{2}}z\mathrm{log}(z)\right]`$
$`+[{\displaystyle \frac{3}{8}}+{\displaystyle \frac{45}{4}}z{\displaystyle \frac{93}{8}}z^2+{\displaystyle \frac{21}{2}}z\mathrm{log}(z)]\mathrm{cos}^2\theta ^{}\},`$
$`\mathrm{\Delta }_{q\overline{q}}^{(1)G\gamma /Z}`$ $`=`$ $`\left({\displaystyle \frac{\pi }{8N}}\right)C_F[(1212z+{\displaystyle \frac{8}{1z}})\mathrm{log}(z)+(8+8z)\mathrm{log}(1z)`$
$`16\left({\displaystyle \frac{\mathrm{log}(1z)}{1z}}\right)_++\left(4+4z{\displaystyle \frac{8}{(1z)_+}}6\delta (1z)\right)\mathrm{log}\left({\displaystyle \frac{Q^2}{\mu _F^2}}\right)`$
$`8\zeta (2)\delta (1z)12+12z+18\delta (1z)\left]\right(13\mathrm{cos}^2\theta ^{}),`$
$`\mathrm{\Delta }_{q(\overline{q})g}^{(1)G\gamma /Z}`$ $`=`$ $`\left({\displaystyle \frac{\pi }{8N}}\right)T_F[(6+4z^2)\mathrm{log}(z)+(4+8z8z^2)\mathrm{log}(1z)`$
$`+(2+4z4z^2)\mathrm{log}\left({\displaystyle \frac{Q^2}{\mu _F^2}}\right)52z+7z^2\left]\right(13\mathrm{cos}^2\theta ^{}),`$
$`\mathrm{\Delta }_{gq(\overline{q})}^{(1),G\gamma /Z}`$ $`=`$ $`\left({\displaystyle \frac{\pi }{8N}}\right)T_F[(228z+4z^2)\mathrm{log}(z)+(4+8z8z^2)\mathrm{log}(1z)`$
$`+(2+4z4z^2)\mathrm{log}\left({\displaystyle \frac{Q^2}{\mu _F^2}}\right)922z+31z^2\left]\right(13\mathrm{cos}^2\theta ^{}),`$
$`\mathrm{\Delta }_{q\overline{q}}^{(1)G}`$ $`=`$ $`\left({\displaystyle \frac{\pi }{8N}}\right)C_F\{[({\displaystyle \frac{5}{2}}+{\displaystyle \frac{5}{2}}z{\displaystyle \frac{5}{1z}})\mathrm{log}(z)+(55z)\mathrm{log}(1z)`$
$`+10\left({\displaystyle \frac{\mathrm{log}(1z)}{1z}}\right)_++\left({\displaystyle \frac{5}{2}}{\displaystyle \frac{5}{2}}z+{\displaystyle \frac{5}{(1z)_+}}+{\displaystyle \frac{15}{4}}\delta (1z)\right)`$
$`\times \mathrm{log}\left({\displaystyle \frac{Q^2}{\mu _F^2}}\right)+5\zeta (2)\delta (1z){\displaystyle \frac{15}{2}}+{\displaystyle \frac{10}{3z}}+{\displaystyle \frac{15}{2}}z{\displaystyle \frac{10}{3}}z^2`$
$`{\displaystyle \frac{25}{2}}\delta (1z)]+[({\displaystyle \frac{45}{2}}+{\displaystyle \frac{45}{2}}z+{\displaystyle \frac{15}{1z}})\mathrm{log}(z)`$
$`+(15+15z)\mathrm{log}(1z)30\left({\displaystyle \frac{\mathrm{log}(1z)}{1z}}\right)_++({\displaystyle \frac{15}{2}}+{\displaystyle \frac{15}{2}}z`$
$`{\displaystyle \frac{15}{(1z)_+}}{\displaystyle \frac{45}{4}}\delta (1z)\left)\mathrm{log}\right({\displaystyle \frac{Q^2}{\mu _F^2}})15\zeta (2)\delta (1z)+{\displaystyle \frac{225}{2}}`$
$`{\displaystyle \frac{225}{2}}z+{\displaystyle \frac{75}{2}}\delta (1z)]\mathrm{cos}^2\theta ^{}+[(4040z{\displaystyle \frac{20}{1z}})\mathrm{log}(z)`$
$`+(2020z)\mathrm{log}(1z)+40\left({\displaystyle \frac{\mathrm{log}(1z)}{1z}}\right)_++(1010z`$
$`+{\displaystyle \frac{20}{(1z)_+}}+15\delta (1z)\left)\mathrm{log}\right({\displaystyle \frac{Q^2}{\mu _F^2}})+20\zeta (2)\delta (1z)150`$
$`{\displaystyle \frac{10}{3z}}+150z+{\displaystyle \frac{10}{3}}z^250\delta (1z)]\mathrm{cos}^4\theta ^{}\},`$
$`\mathrm{\Delta }_{q(\overline{q})g}^{(1)G}`$ $`=`$ $`{\displaystyle \frac{\pi }{8N}}T_F\{[({\displaystyle \frac{35}{4}}{\displaystyle \frac{10}{z}}10z{\displaystyle \frac{5}{2}}z^2)\mathrm{log}(z)+({\displaystyle \frac{35}{2}}+{\displaystyle \frac{20}{z}}+5z+5z^2)`$
$`\times \mathrm{log}(1z)+\left({\displaystyle \frac{35}{4}}+{\displaystyle \frac{10}{z}}+{\displaystyle \frac{5}{2}}z+{\displaystyle \frac{5}{2}}z^2\right)\mathrm{log}\left({\displaystyle \frac{Q^2}{\mu _F^2}}\right)`$
$`{\displaystyle \frac{15}{8}}{\displaystyle \frac{15}{2z}}+{\displaystyle \frac{75}{4}}z{\displaystyle \frac{35}{8}}z^2]+[({\displaystyle \frac{135}{4}}+45z+{\displaystyle \frac{15}{2}}z^2)\mathrm{log}(z)`$
$`+\left({\displaystyle \frac{15}{2}}+15z15z^2\right)\mathrm{log}(1z)+\left({\displaystyle \frac{15}{4}}+{\displaystyle \frac{15}{2}}z{\displaystyle \frac{15}{2}}z^2\right)\mathrm{log}\left({\displaystyle \frac{Q^2}{\mu _F^2}}\right)`$
$`+{\displaystyle \frac{645}{8}}{\displaystyle \frac{375}{4}}z+{\displaystyle \frac{105}{8}}z^2]\mathrm{cos}^2\theta ^{}+[(65+{\displaystyle \frac{10}{z}}35z10z^2)\mathrm{log}(z)`$
$`+\left(30{\displaystyle \frac{20}{z}}30z+20z^2\right)\mathrm{log}(1z)+\left(15{\displaystyle \frac{10}{z}}15z+10z^2\right)`$
$`\times \mathrm{log}\left({\displaystyle \frac{Q^2}{\mu _F^2}}\right){\displaystyle \frac{205}{2}}+{\displaystyle \frac{15}{2z}}+{\displaystyle \frac{215}{2}}z{\displaystyle \frac{35}{2}}z^2]\mathrm{cos}^4\theta ^{}\},`$
$`\mathrm{\Delta }_{gq(\overline{q})}^{(1)G}`$ $`=`$ $`{\displaystyle \frac{\pi }{8N}}T_F\{[({\displaystyle \frac{35}{4}}{\displaystyle \frac{10}{z}}+{\displaystyle \frac{25}{2}}z{\displaystyle \frac{5}{2}}z^2)\mathrm{log}(z)+({\displaystyle \frac{35}{2}}+{\displaystyle \frac{20}{z}}+5z+5z^2)`$
$`\times \mathrm{log}(1z)+\left({\displaystyle \frac{35}{4}}+{\displaystyle \frac{10}{z}}+{\displaystyle \frac{5}{2}}z+{\displaystyle \frac{5}{2}}z^2\right)\mathrm{log}\left({\displaystyle \frac{Q^2}{\mu _F^2}}\right)`$
$`+{\displaystyle \frac{285}{8}}{\displaystyle \frac{35}{2z}}{\displaystyle \frac{15}{4}}z{\displaystyle \frac{75}{8}}z^2]+[({\displaystyle \frac{465}{4}}{\displaystyle \frac{345}{2}}z+{\displaystyle \frac{15}{2}}z^2)\mathrm{log}(z)`$
$`+\left({\displaystyle \frac{15}{2}}+15z15z^2\right)\mathrm{log}(1z)+\left({\displaystyle \frac{15}{4}}+{\displaystyle \frac{15}{2}}z{\displaystyle \frac{15}{2}}z^2\right)\mathrm{log}\left({\displaystyle \frac{Q^2}{\mu _F^2}}\right)`$
$`{\displaystyle \frac{1695}{8}}{\displaystyle \frac{20}{z}}+{\displaystyle \frac{615}{4}}z+{\displaystyle \frac{625}{8}}z^2]\mathrm{cos}^2\theta ^{}+[(185+{\displaystyle \frac{10}{z}}+215z10z^2)`$
$`\times \mathrm{log}(z)+(30{\displaystyle \frac{20}{z}}30z+20z^2)\mathrm{log}(1z)+(15{\displaystyle \frac{10}{z}}`$
$`15z+10z^2)\mathrm{log}\left({\displaystyle \frac{Q^2}{\mu _F^2}}\right)+{\displaystyle \frac{395}{2}}+{\displaystyle \frac{545}{6z}}{\displaystyle \frac{385}{2}}z{\displaystyle \frac{605}{6}}z^2]\mathrm{cos}^4\theta ^{}\},`$
$`\mathrm{\Delta }_{gg}^{(1)G}`$ $`=`$ $`\left({\displaystyle \frac{\pi }{2(N^21)}}\right)\{C_A[(10{\displaystyle \frac{5}{z}}5z+5z^2{\displaystyle \frac{5}{1z}})\mathrm{log}(z)`$ (2.4)
$`+\left(20+{\displaystyle \frac{10}{z}}+10z10z^2\right)\mathrm{log}(1z)+10\left({\displaystyle \frac{\mathrm{log}(1z)}{1z}}\right)_+`$
$`+\left(10+{\displaystyle \frac{5}{z}}+5z5z^2+{\displaystyle \frac{5}{(1z)_+}}+{\displaystyle \frac{55}{12}}\delta (1z)\right)\mathrm{log}\left({\displaystyle \frac{Q^2}{\mu _F^2}}\right)`$
$`+5\zeta _2\delta (1z)+{\displaystyle \frac{25}{4}}{\displaystyle \frac{85}{12z}}{\displaystyle \frac{25}{4}}z+{\displaystyle \frac{85}{12}}z^2{\displaystyle \frac{1015}{72}}\delta (1z)]`$
$`+C_A\left[\left(3030z\right)\mathrm{log}(z)60{\displaystyle \frac{5}{z}}+60z+5z^2\right]\mathrm{cos}^2\theta ^{}`$
$`+C_A[(40+{\displaystyle \frac{5}{z}}+55z5z^2+{\displaystyle \frac{5}{1z}})\mathrm{log}(z)+(20{\displaystyle \frac{10}{z}}`$
$`10z+10z^2)\mathrm{log}(1z)10({\displaystyle \frac{\mathrm{log}(1z)}{1z}})_++(10{\displaystyle \frac{5}{z}}5z`$
$`+5z^2{\displaystyle \frac{5}{(1z)_+}}{\displaystyle \frac{55}{12}}\delta (1z)\left)\mathrm{log}\right({\displaystyle \frac{Q^2}{\mu _F^2}})5\zeta (2)\delta (1z)`$
$`+{\displaystyle \frac{255}{4}}+{\displaystyle \frac{305}{12z}}{\displaystyle \frac{255}{4}}z{\displaystyle \frac{305}{12}}z^2+{\displaystyle \frac{1015}{72}}\delta (1z)]\mathrm{cos}^4\theta ^{}`$
$`+T_Fn_f\left[{\displaystyle \frac{5}{3}}\delta (1z)\mathrm{log}\left({\displaystyle \frac{Q^2}{\mu _F^2}}\right)+{\displaystyle \frac{175}{36}}\delta (1z)\right]`$
$`+T_Fn_f[{\displaystyle \frac{5}{3}}\delta (1z)\mathrm{log}\left({\displaystyle \frac{Q^2}{\mu _F^2}}\right){\displaystyle \frac{175}{36}}\delta (1z)]\mathrm{cos}^4\theta ^{}\},`$
where $`C_F=(N^21)/2N`$, $`C_A=N`$ and $`T_F=1/2`$ are the $`SU(N)`$ colour factors and $`n_f`$ is the number of flavours. The ”plus” functions appearing in the above results are the distributions which satisfy the following equation
$`{\displaystyle _0^1}𝑑zf_+(z)g(z)`$ $`=`$ $`{\displaystyle _0^1}𝑑zf(z)\left(g(z)g(1)\right),`$
where
$`f_+(z)`$ $`=`$ $`\left({\displaystyle \frac{\mathrm{ln}^i(1z)}{1z}}\right)_+,i=0,1`$
and $`g(z)`$ is any well behaved function in the region $`0z1`$.
In Eqs. (2.3) and (2.4) the term $`(13\mathrm{cos}^2\theta ^{})`$ corresponds to the interference between the SM and GR and within the SM interference between $`\gamma `$ and $`Z`$ diagrams. Though this combination is even in $`\mathrm{cos}\theta ^{}`$ it vanishes in the angular integrated cross section and also does not contribute to the forward-backward asymmetry $`A_{FB}`$. Hence the un-integrated cross section is very useful to study this contribution to the interference effect in the Drell-Yan process.
We present below the angular distribution which is ”odd” in $`\mathrm{cos}\theta ^{}`$:
$`2S{\displaystyle \frac{d\sigma _o^{P_1P_2}}{dQ^2d\mathrm{cos}\theta ^{}}}`$ $`=`$ $`{\displaystyle \underset{q}{}}\delta _{SM,q}{\displaystyle _0^1}𝑑x_1{\displaystyle _0^1}𝑑x_2{\displaystyle _0^1}𝑑z\delta (\tau zx_1x_2)`$ (2.5)
$`\times [\delta H_{q\overline{q}}(x_1,x_2,\mu _F^2)(\delta \mathrm{\Delta }_{q\overline{q}}^{(0),\gamma Z}(z,Q^2,\mu _F^2)+a_s\delta \mathrm{\Delta }_{q\overline{q}}^{(1),\gamma Z}(z,Q^2,\mu _F^2))`$
$`+\delta H_{qg}(x_1,x_2,\mu _F^2)\left(a_s\delta \mathrm{\Delta }_{qg}^{(1),\gamma Z}(z,\mu _F^2)\right)`$
$`+\delta H_{gq}(x_1,x_2,\mu _F^2)\left(a_s\delta \mathrm{\Delta }_{gq}^{(1),\gamma Z}(z,\mu _F^2)\right)]`$
$`+{\displaystyle \underset{q}{}}\delta _{SMGR,q}{\displaystyle _0^1}𝑑x_1{\displaystyle _0^1}𝑑x_2{\displaystyle _0^1}𝑑z\delta (\tau zx_1x_2)`$
$`\times [\delta H_{q\overline{q}}(x_1,x_2,\mu _F^2)(\delta \mathrm{\Delta }_{q\overline{q}}^{(0),G\gamma /Z}(z,Q^2,\mu _F^2)+a_s\delta \mathrm{\Delta }_{q\overline{q}}^{(1),G\gamma /Z}(z,Q^2,\mu _F^2))`$
$`+\delta H_{qg}(x_1,x_2,\mu _F^2)\left(a_s\delta \mathrm{\Delta }_{qg}^{(1),G\gamma /Z}(z,\mu _F^2)\right)`$
$`+\delta H_{gq}(x_1,x_2,\mu _F^2))\left(a_s\delta \mathrm{\Delta }_{gq}^{(1),G\gamma /Z}(z,\mu _F^2)\right)].`$
The constants $`\delta _{SM,q},\delta _{SMGR,q}`$ are given by
$`\delta _{SM,q}`$ $`=`$ $`2\alpha ^2[{\displaystyle \frac{(Q^2M_Z^2)}{\left((Q^2M_Z^2)^2+M_Z^2\mathrm{\Gamma }_Z^2\right)c_w^2s_w^2}}Q_qQ_eg_q^Ag_e^A`$ (2.6)
$`+{\displaystyle \frac{2Q^2}{\left((Q^2M_Z^2)^2+M_Z^2\mathrm{\Gamma }_Z^2\right)c_w^4s_w^4}}g_q^Vg_e^Vg_q^Ag_e^A],`$
$`\delta _{SMGR,q}`$ $`=`$ $`{\displaystyle \frac{\alpha \kappa ^2Q^2}{4\pi }}|𝒟(Q^2)|\left[Q_qQ_e+{\displaystyle \frac{Q^2(Q^2M_Z^2)}{\left((Q^2M_Z^2)^2+M_Z^2\mathrm{\Gamma }_Z^2\right)c_w^2s_w^2}}g_q^Vg_e^V\right].`$ (2.7)
The renormalised incoming partonic fluxes are defined by
$`\delta H_{q\overline{q}}(x_1,x_2,\mu _F^2)`$ $`=`$ $`f_q^{P_1}(x_1,\mu _F^2)f_{\overline{q}}^{P_2}(x_2,\mu _F^2)f_{\overline{q}}^{P_1}(x_1,\mu _F^2)f_q^{P_2}(x_2,\mu _F^2),`$
$`\delta H_{gq}(x_1,x_2,\mu _F^2)`$ $`=`$ $`f_g^{P_1}(x_1,\mu _F^2)\left(f_q^{P_2}(x_2,\mu _F^2)f_{\overline{q}}^{P_2}(x_2,\mu _F^2)\right),`$
$`\delta H_{qg}(x_1,x_2,\mu _F^2)`$ $`=`$ $`\delta H_{gq}(x_2,x_1,\mu _F^2).`$ (2.8)
The LO coefficient functions corresponding to Eq. (2.5) are
$`\delta \mathrm{\Delta }_{q\overline{q}}^{(0),\gamma Z}`$ $`=`$ $`{\displaystyle \frac{2\pi }{N}}\delta (1z)\left[\mathrm{cos}\theta ^{}\right],`$
$`\delta \mathrm{\Delta }_{q\overline{q}}^{(0),G\gamma /Z}`$ $`=`$ $`{\displaystyle \frac{\pi }{8N}}\delta (1z)\left[2\mathrm{cos}^3\theta ^{}\right].`$ (2.9)
The NLO contributions are given by
$`\delta \mathrm{\Delta }_{q\overline{q}}^{(1)\gamma Z}`$ $`=`$ $`{\displaystyle \frac{2\pi }{N}}C_F[(8+8z{\displaystyle \frac{8}{1z}})\mathrm{log}(z)+(88z)\mathrm{log}(1z)`$
$`+16\left({\displaystyle \frac{\mathrm{log}(1z)}{1z}}\right)_++(44z+{\displaystyle \frac{8}{(1z)_+}}+6\delta (1z))\mathrm{log}\left({\displaystyle \frac{Q^2}{\mu _F^2}}\right)`$
$`+8\zeta (2)\delta (1z)+44z16\delta (1z)]\mathrm{cos}\theta ^{},`$
$`\delta \mathrm{\Delta }_{qg}^{(1)\gamma Z}`$ $`=`$ $`{\displaystyle \frac{2\pi }{N}}T_F[(24z^2)\mathrm{log}(z)+(48z+8z^2)\mathrm{log}(1z)`$
$`+(24z+4z^2)\mathrm{log}\left({\displaystyle \frac{Q^2}{\mu _F^2}}\right)+1+6z7z^2]\mathrm{cos}\theta ^{},`$
$`\delta \mathrm{\Delta }_{gq}^{(1)\gamma Z}`$ $`=`$ $`{\displaystyle \frac{2\pi }{N}}T_F[(24z+12z^2)\mathrm{log}(z)+(4+8z8z^2)\mathrm{log}(1z)`$
$`+(2+4z4z^2)\mathrm{log}\left({\displaystyle \frac{Q^2}{\mu _F^2}}\right)+12z+z^2]\mathrm{cos}\theta ^{},`$
$`\delta \mathrm{\Delta }_{q\overline{q}}^{(1)G\gamma /Z}`$ $`=`$ $`{\displaystyle \frac{\pi }{8N}}C_F\{[{\displaystyle \frac{16}{1z}}\mathrm{log}(z)+(1616z)\mathrm{log}(1z)`$
$`+32\left({\displaystyle \frac{\mathrm{log}(1z)}{1z}}\right)_++(88z+{\displaystyle \frac{16}{(1z)_+}}+12\delta (1z))\mathrm{log}\left({\displaystyle \frac{Q^2}{\mu _F^2}}\right)`$
$`+16\zeta (2)\delta (1z)48+48z36\delta (1z)]\mathrm{cos}^3\theta ^{}`$
$`+[2424z]\mathrm{cos}\theta ^{}\},`$
$`\delta \mathrm{\Delta }_{qg}^{(1)G\gamma /Z}`$ $`=`$ $`{\displaystyle \frac{\pi }{8N}}T_F\{[(1224z8z^2)\mathrm{log}(z)+(816z+16z^2)\mathrm{log}(1z)`$
$`+(48z+8z^2)\mathrm{log}\left({\displaystyle \frac{Q^2}{\mu _F^2}}\right)38+52z14z^2]\mathrm{cos}^3\theta ^{}`$
$`+[24z\mathrm{log}(z)+2424z]\mathrm{cos}\theta ^{}\},`$
$`\delta \mathrm{\Delta }_{gq}^{(1)G\gamma /Z}`$ $`=`$ $`{\displaystyle \frac{\pi }{8N}}T_F\{[(36+72z+24z^2)\mathrm{log}(z)+(8+16z16z^2)\mathrm{log}(1z)`$ (2.10)
$`+(4+8z8z^2)\mathrm{log}\left({\displaystyle \frac{Q^2}{\mu _F^2}}\right)+98100z+2z^2]\mathrm{cos}^3\theta ^{}`$
$`+[48z\mathrm{log}(z)48+48z]\mathrm{cos}\theta ^{}\}.`$
These coefficient functions Eq. (2) and (2), which are odd in $`\mathrm{cos}\theta ^{}`$, are due to the interference of $`\gamma `$ and $`Z`$ in SM and between $`SM`$ and $`GR`$ in the full theory. Note that the $`q\overline{q}`$ in the $`qg`$ subprocess leads to a negative sign which has been taken care of in the renormalised parton fluxes Eq. (2). $`A_{FB}`$ picks up this odd parts which contributes to the interference terms. To NLO the $`A_{FB}`$ coefficient functions have been evaluated in In the last three equations of Eq. (6.11), the RHS should read $`\mathrm{\Delta }_{ab}^{(1)\gamma /Z}`$. and the effects analysed for the ADD case. In the next section, the impact of the NLO-QCD correction derived in this section is discussed.
## 3 Discussions
In this section, the effect of the NLO QCD corrections on the angular distribution of lepton pair are presented. We present these distributions for the LHC ($`\sqrt{S}=14\mathrm{TeV}`$) and Run II of Tevatron ($`\sqrt{S}=1.96\mathrm{TeV}`$) for typical values of ADD and RS model parameters. The effort here is mainly to emphasise the impact of QCD correction on the bounds rather than to extract bounds on $`M_S`$.
For ADD model, we choose the parameters $`M_S=2`$ TeV and $`d=3`$. For RS we choose $`c_0=0.01`$, $`M_1=1.5`$ TeV(for LHC) and $`M_1=700`$ GeV(for Tevatron). The SM parameters which enter our analysis are $`\alpha =1/137.03604`$, $`M_Z=91.1876`$ GeV, $`\mathrm{\Gamma }_Z=2.4952`$ GeV and $`\mathrm{sin}^2\theta _W=0.227`$. For the parton density sets, we adopt in leading order, the MRST 2001 LO ($`\mathrm{\Lambda }=0.1670`$ GeV) and in next-to-leading order, the MRST 2001 NLO ($`\mathrm{\Lambda }=0.2390`$ GeV). The renormalisation scale $`\mu _R`$ and factorisation scale $`\mu _F`$ are taken to be equal to $`Q`$ unless mentioned otherwise.
For the coefficient functions which are even in $`\mathrm{cos}\theta ^{}`$, the parton density combinations are even under the interchange of $`x_1`$ and $`x_2`$, while for the odd terms in $`\mathrm{cos}\theta ^{}`$, the parton density combinations are odd under this exchange. Hence, the quark-antiquark initiated contributions from these odd terms to LHC cross sections are zero, but small contribution comes from quark-gluon initiated processes. This is not the case for Tevatron.
In the SM part, at LO level, the quark-antiquark initiated processes behave as $`1+\mathrm{cos}^2\theta ^{}`$ for pure $`\gamma `$ and $`Z`$ intermediate states, and as $`\mathrm{cos}\theta ^{}`$ for $`\gamma Z`$ interference terms. In the Gravity part, at LO, gluon-gluon initiated process is of the form $`1\mathrm{cos}^4\theta ^{}`$, quark-antiquark process is of the form $`13\mathrm{cos}^2\theta ^{}+4\mathrm{cos}^4\theta ^{}`$. The interference between SM and GR always behaves as $`\mathrm{cos}\theta ^{}`$, $`\mathrm{cos}^3\theta ^{}`$ and $`13\mathrm{cos}^2\theta ^{}`$. At NLO, quark-gluon initiated processes contribute to both SM, GR and the interference terms.
We first discuss the phenomenology at LHC using ADD model. In Fig. 1a, we plot the angular distribution at $`Q=700`$ GeV with NLO corrected cross sections. We find that the gravity contribution is large compared to that of SM. Since the gluon flux is large at LHC, the gluon-gluon initiated subprocess dominates over the rest. Also, the interference between SM and Gravity is negligible over the entire range of $`\mathrm{cos}\theta ^{}`$ at this $`Q=700`$ GeV. In the Fig. 1b, we have plotted the K-factor at LHC. In general K-factor is defined as
$`K=\left[{\displaystyle \frac{d\sigma _{LO}^I(Q,\mathrm{cos}\theta ^{})}{dQd\mathrm{cos}\theta ^{}}}\right]^1\left[{\displaystyle \frac{d\sigma _{NLO}^I(Q,\mathrm{cos}\theta ^{})}{dQd\mathrm{cos}\theta ^{}}}\right],`$ (3.1)
where $`I=SM,TOT`$, SM means Standard Model, TOT means sum of SM and Gravity contributions. Since the gluon-gluon initiated process dominates over the rest the K-factor is around $`1.4`$ to $`1.5`$ in the entire range of $`\mathrm{cos}\theta ^{}`$. In the Fig. 1c we have plotted the R-ratio in order to check whether the NLO results improve the scale uncertainty. Here we have chosen $`\mu _R=\mu _F=\mu `$ and we follow the same throughout our analysis. The R-ratio is defined as
$`R_{LO}^I`$ $`=`$ $`\left[{\displaystyle \frac{d\sigma _{LO}^I(\mu =\mu _0))}{dQd\mathrm{cos}\theta ^{}}}\right]^1\left[{\displaystyle \frac{d\sigma _{LO}^I(\mu ))}{dQd\mathrm{cos}\theta ^{}}}\right]_{Q=700GeV},`$ (3.2)
$`R_{NLO}`$ $`=`$ $`\left[{\displaystyle \frac{d\sigma _{NLO}^I(\mu =\mu _0))}{dQd\mathrm{cos}\theta ^{}}}\right]^1\left[{\displaystyle \frac{d\sigma _{NLO}^I(\mu ))}{dQd\mathrm{cos}\theta ^{}}}\right]_{Q=700GeV}.`$ (3.3)
We have chosen $`\theta ^{}=45^0`$ for the plot. As we can see the NLO results improve the scale uncertainty.
Let us now repeat the similar study for the RS model at LHC energies. In the Fig. 2a, we have plotted the angular distribution at first resonance $`M_1=1.5`$ TeV and $`c_0=0.01`$. We find that the gravity contribution is well above the standard model one. In particular, the gluon-gluon initiated contribution is the dominant one. Since the SM and interference contributions are of the same order and are extremely small due to large $`Q`$ which is $`1.5`$ TeV, the total contribution is purely due to the gluon-gluon initiated process. Unlike the ADD case (see Fig. 1a), the total contribution mainly comes from the gravity mediated process at the first resonance. In Fig. 2b, we have plotted the K-factor defined in Eq. (3.1) at the first resonance. Because of this, the dominant contribution comes from the gravity mediated processes. Since $`Q=1.5`$ TeV, both quark-antiquark as well as the gluon initiated processes contribute at the same level because their partonic fluxes are comparable at this energy. The shape of the K-factor in Fig. 1b looks different from Fig. 2b because at $`Q=0.7`$ TeV, only gluon initiated process dominates. In Fig. 2c, we have plotted the R-ratio defined in Eq. (3.2, 3.3) for RS resonance $`M_1=1.5`$ TeV, $`c_0=0.01`$ at $`\theta ^{}=45^0`$. From the plot it is clear that the NLO corrections improve the scale uncertainty.
Next we discuss the phenomenology at Tevatron. We start with ADD for the parameters $`M_S=2`$ TeV, $`d=3`$ and $`Q=400`$ GeV. In Fig. 3a we have plotted angular distribution with NLO results. We find that the SM dominates over the rest. In Tevatron, both the even and odd in $`\mathrm{cos}\theta ^{}`$ contribute significantly unlike in the LHC. This leads to the asymmetry in the angular distribution as shown in Fig. 3a at Tevatron. At $`Q=400`$ GeV the dominant contribution at Tevatron is from SM. In Fig. 3b, we have plotted the K-factor at $`Q=400`$ GeV. Since it is the quark-antiquark initiated process that dominates, the K-factor is similar to the SM value which is around 1.3. Fig. 3c shows the sensitivity of the results to the scale variation. As expected NLO improves the result.
We now study the phenomenology at Tevatron in the case of RS model. We choose the RS model parameters $`M_1=700`$ GeV and $`c_0=0.01`$ which are not excluded by the recent searches by DØ. In Fig. 4a, we have plotted the angular distribution at the first resonance using NLO corrected cross sections. Being in the resonance region, gravity mediated process dominates over the SM. At $`Q=700`$ GeV, the quark initiated processes contribute significantly, as the $`q\overline{q}`$ flux dominates. In Fig. 4b, we have plotted the K-factor at $`Q=700`$ GeV. Since the quark initiated process dominates, the K-factor is close to the SM value. Fig. 4c shows the sensitivity of the results with respect to scale. One can easily notice that NLO gives reliable predictions.
In summary, we have computed the cross sections $`d\sigma /dQ/d\mathrm{cos}\theta ^{}`$ up to next to leading order in QCD. Along with the standard model results, we have presented the contributions from all the subprocesses that are due to the graviton in the context of TeV-scale gravity models. Our main conclusion is that the NLO QCD corrections are very significant at the LHC because of the large incident gluon flux. At the Tevatron where the gluon flux is small, the NLO effects are moderate for ADD and RS in the resonance region. But, significantly, at both the colliders the inclusion of the NLO QCD corrections help stabilise the cross-section with respect to scale variations. The extraction of bounds from both the colliders will, therefore, require the inclusion of these NLO QCD corrections.
Acknowledgments:
The work of PM is part of a project (IFCPAR Project No. 2904-2) on ‘Brane-World Phenomenology’ supported by the Indo-French Centre for the Promotion of Advanced Research, New Delhi, India. We thank S. Raychaudhuri for providing the code that evaluates the RS KK mode sum in the propagator.
Figure Caption
Figure 1. (a) The double differential cross section $`d\sigma /dQ/d\mathrm{cos}\theta ^{}`$ is plotted as a function of $`\mathrm{cos}\theta ^{}`$ for $`Q=700`$ GeV at LHC. The typical ADD parameters chosen are $`M_S=2`$ TeV, $`d=3`$. (b) The corresponding K-factor for $`\mathrm{cos}\theta ^{}`$ distribution SM and SM plus gravity (TOT). (c) Scale variation at LO and NLO as defined in Eq. (3.2), (3.3) for $`Q=700`$ GeV and $`\theta ^{}=45^0`$.
Figure 2. (a) The double differential cross section $`d\sigma /dQ/d\mathrm{cos}\theta ^{}`$ is plotted as a function of $`\mathrm{cos}\theta ^{}`$ for $`Q=1.5`$ TeV at LHC. The RS model parameters are $`M_1=1.5`$ TeV and $`c_0=0.01`$. (b) The K-factor for the distribution in (a) is plotted for the $`\mathrm{cos}\theta ^{}`$ range \[-1,1\]. (c) The scale variation of the ratio R is plotted as a function of $`\mu /\mu _0`$ at the first RS KK resonance region and $`\theta ^{}=45^0`$.
Figure 3. (a) For Tevatron energies, $`d\sigma /dQ/d\mathrm{cos}\theta ^{}`$ is plotted as a function of $`\mathrm{cos}\theta ^{}`$ at $`Q=400`$ GeV for typical value of ADD parameters $`M_S=2`$ TeV and $`d=3`$. (b) The K-factor for $`\mathrm{cos}\theta ^{}`$ distribution for the same ADD parameters in (a) is plotted. (c) The variation of the R-ratio with respect to the scale $`\mu /\mu _0`$ for the ADD parameters in (a) at $`\theta ^{}=45^0`$.
Figure 4. (a) The double differential cross section $`d\sigma /dQ/d\mathrm{cos}\theta ^{}`$ is plotted as a function of $`\mathrm{cos}\theta ^{}`$ for $`Q=700`$ GeV at the Tevatron. RS parameters $`M_1=700`$ GeV and $`c_0=0.01`$. (b) The K-factor for the distribution in (a) is plotted for the $`\mathrm{cos}\theta ^{}`$ range. (c) The scale variation of the ratio R is plotted as a function of $`\mu /\mu _0`$ for $`\theta ^{}=45^0`$ and $`Q=700`$ GeV. |
warning/0507/math0507588.html | ar5iv | text | # 1. Introduction
## 1. Introduction
B.L. van der Waerden proved that for any positive integers $`k`$ and $`r`$, there is a positive integer $`w(k,r)`$ such that any $`r`$-coloring of $`\{1,2,\mathrm{},w(k,r)\}`$ must admit a monochromatic $`k`$-term arithmetic progression. In , a generalization of van der Waerden’s theorem for 3-term arithmetic progressions was investigated. Namely, for integers $`1ab`$, define an $`(a,b)`$-triple to be any 3-term sequence of the form $`(x,ax+d,bx+2d)`$, where $`x,d`$ are positive integers. Taking $`a=b=1`$ gives a 3-term arithmetic progression, and by van der Waerden’s theorem the associated van der Waerden number $`w(3,r)`$ is finite for all $`r`$.
Throughout this note, we assume that $`a`$ and $`b`$ are integers and that $`1ab`$. For $`r1`$, denote by $`n=n(a,b;r)`$ the least positive integer, if it exists, such that every $`r`$-coloring of $`[1,n]`$ admits a monochromatic $`(a,b)`$-triple. If no such $`n`$ exists, we write $`n(a,b;r)=\mathrm{}`$. We say that $`(a,b)`$ is regular if $`n(a,b;r)<\mathrm{}`$ for each $`r`$. By van der Waerden’s theorem $`(1,1)`$ is regular. If $`(a,b)`$ is not regular, the degree of regularity of $`(a,b)`$, denoted dor$`(a,b)`$, is the largest integer $`r`$ such that $`(a,b)`$ is $`r`$-regular.
In , it is shown that for a wide class of pairs $`(a,b)(1,1)`$, $`(a,b)`$ is not regular, i.e., dor$`(a,b)<\mathrm{}`$, and its authors conjectured that, in fact, $`(1,1)`$ is the only regular pair. In Section 2 we confirm this conjecture.
Also in , it was shown that
dor($`a,b`$) = 1 if and only if $`b=2a`$, (1)
and upper bounds on dor($`a,b`$) are given for those pairs which are shown not to be regular. Further, those authors speculated that dor($`a,b)\{1,2,\mathrm{}\}`$ for all pairs $`(a,b)`$. In Section 3 we show this conjecture to be false. We also obtain upper bounds on dor($`a,b`$) for all $`(a,b)(1,1)`$, which improve upon the results of , and provide an alternate proof that (1,1) is the only regular triple.
## 2. The Only Regular Triples are Arithmetic Progressions
In this section we give a short proof which shows that (1,1)-triples are the only regular $`(a,b)`$-triples. The proof makes use of Rado’s regularity theorem (see ) which states, in particular, that the linear equation $`a_1x_1+a_2x_2+\mathrm{}+a_kx_k=0`$ has a monochromatic solution in $``$ under any finite coloring of $``$ if and only if some nonempty subset of the nonzero coefficients sums to zero. It also uses the following lemma.
###### Lemma 1
For all $`1ab`$, and all $`i1`$,
$$n(a,b;r)n(a+i,b+2i;r),$$
and hence dor$`(a,b)dor(a+i,b+2i)`$.
Proof. Let $`a,b,i`$ be given. To prove the lemma, it suffices to show that every $`(a+i,b+2i)`$-triple is also an $`(a,b)`$-triple. Let $`X=(x,y,z)`$ be an $`(a+i,b+2i)`$-triple. So $`y=(a+i)x+d`$ and $`z=(b+2i)x+2d`$ for some $`d>0`$. But then $`X`$ is also an $`(a,b)`$ triple, since $`y=ax+(ix+d)`$ and $`z=bx+2(ix+d)`$. $`\mathrm{}`$
###### Theorem 1
Let $`1ab`$. If $`(a,b)(1,1)`$, then $`(a,b)`$ is not regular.
Proof. Since the triple $`\{x,ax+d,bx+2d\}`$ satisfies the equation $`(2ab)x2y+z=0`$, by Rado’s regularity theorem an $`(a,b)`$-triple is regular only if $`b2a\{2,1,1\}`$. Hence, this leaves three cases to consider: (i) $`b=2a+1`$, (ii) $`b=2a1`$, and (iii) $`b=2a2`$. In it was shown that dor$`(1,3)3`$, dor$`(2,3)=2`$, and dor$`(2,2)5`$. By Lemma 1, these three facts cover Cases (i), (ii), and (iii), respectively. $`\mathrm{}`$
Remark 1 In Section 3 we will show that dor$`(2,2)4`$. We see from this fact, the proof of Theorem 1, and (1), that $`2`$ dor$`(a,2a2)4`$ for all $`a2`$; that dor($`a,2a1)=2`$ for all $`a2`$; and that $`2`$ dor$`(a,2a+1)3`$ for all $`a1`$.
## 3. More on the Degree of Regularity
Using the Fortran program AB.f, available from the third author’s website<sup>1</sup><sup>1</sup>1http://math.colgate.edu/$``$aaron/programs.html, we have found that $`n(2,2;3)=88`$. This implies
dor$`(2,2)3`$, (2)
which is a counterexample to the suggestion made in that dor$`(a,b)\{1,2,\mathrm{}\}`$ for all $`(a,b)`$. The program uses a well-known backtracking algorithm (see , Algorithm 2, page 31) which checks that all $`3`$-colorings of $`[1,88]`$ contain a monochromatic $`(2,2)`$-triple.
Although (2) shows the existence of a pair besides (1,1) whose degree of regularity is greater than two, we wonder if dor$`(a,b)=2`$ for “almost all” $`(a,b)`$. In particular, we pose the following questions.
> Question 1 Is it true that dor$`(a,b)2`$ whenever $`b2a2`$ and $`a2`$?
>
> Question 2 For $`b2a`$, are there only a finite number of pairs $`(a,b)`$ such that
> dor$`(a,b)2`$?
While we do not yet have the answers to these questions, we have been able to improve the upper bounds for dor$`(a,b)`$, as established in , for many $`(a,b)`$-triples. These new bounds are supplied by the next two theorems. The proofs of both theorems use the following coloring.
Notation Let $`c3`$ be an integer and let $`p=2\frac{2}{c}`$. Denote by $`\gamma _c`$ the $`c`$-coloring of $``$ defined by coloring, for each $`k0`$, the interval $`[p^k,p^{k+1})`$ with color $`k(modc)`$.
Theorem 2 Let $`a,i,c`$ such that $`a2`$ and $`c5`$. Define $`p=2\frac{2}{c}`$ and let $`0ip^c(p^{c1}2)`$. If $`a\frac{p^c}{c1}`$, then dor$`(a,a+i)c1`$.
Proof. We use the $`c`$-coloring $`\gamma _c`$. Assume, for a contradiction, that $`\{x,ax+d,(a+i)x+2d\}`$ is a monochromatic $`(a,a+i)`$-triple under $`\gamma _c`$. Let $`x[p^k,p^{k+1})`$. Since $`p<2`$ and $`a2`$, we have that $`ax+d[p^{k+cj},p^{k+cj+1})`$ for some $`j`$. This gives us that $`d>p^{k+cj}ap^{k+1}`$, which, in turn, gives us $`(a+i)x+2d>2p^{k+cj}ap^{k+1}+ip^k`$. We now show that this lower bound is more that $`p^{k+cj+1}`$: By choice of $`a`$ we have $`ap^{c1}(2p)`$ so that $`2\frac{a}{p^{cj}}p`$ for all $`j`$. This gives us $`2p^{k+cj}ap^{k+1}>p^{k+cj+1}`$ which is sufficient for all $`i0`$.
Next, we will show that $`(a+i)x+2d<p^{k+c(j+1)}`$. Since $`d<ax+d<p^{k+cj+1}`$ and $`ix<ip^{k+1}`$ it suffices to show that $`2p^{k+cj+1}+ip^{k+1}<p^{k+cj+c}`$. We have $`ip^c(p^{c1}2)`$, which implies that $`2+\frac{i}{p^cj}<p^{c1}`$ for all $`j`$, which, in turn, implies the desired bound.
Hence, we have $`p^{k+cj+1}<(a+i)x+2d<p^{k+c(j+1)}`$. By the definition of $`\gamma _c`$, we see that if $`x`$ and $`ax+d`$ are the same color, then $`(a+i)x+2d`$ must be a different color under $`\gamma _c`$, a contradiction. $`\mathrm{}`$
Example By Theorem 2 and (2), dor$`(2,2)\{3,4\}`$.
Theorem 3 Let $`b,c`$ such that $`b2`$ and $`c5`$. Let $`p=2\frac{2}{c}`$. If $`b<\frac{2+p^c}{p}`$, then dor$`(1,b)c1`$.
Proof. The proof is quite similar to that of Theorem 2. Assume, for a contradiction, that $`\{x,x+d,bx+2d\}`$ is monochromatic under $`\gamma _c`$. Let $`x[p^k,p^{k+1})`$ so that $`bx+2d[p^{k+cj},p^{k+cj+1})`$ (since $`b2>c`$) for some $`j`$. This gives $`d\frac{1}{2}p^{k+cj}\frac{b}{2}p^{k+1}`$ so that $`x+d>p^k+\frac{1}{2}p^{k+cj}\frac{b}{2}p^{k+1}`$. The condition on $`b`$ implies that this last bound is larger than $`p^{k+1}`$.
We next show that $`x+d<p^{k+cj}`$. We have $`d<\frac{1}{2}p^{k+cj+1}`$ so that $`x+d<p^{k+1}+\frac{1}{2}p^{k+cj+1}`$. Since $`2<p^{c1}(2p)`$ for all $`c5`$, we have $`p^{k+1}+\frac{1}{2}p^{k+cj+1}<p^{k+cj}`$ for all $`j`$. Hence, $`p^{k+1}<x+d<p^{k+cj}`$ so that $`x+d`$ is not the same color, under $`\gamma _c`$, as $`x`$ and $`bx+2d`$, a contradiction. $`\mathrm{}`$
Corollary 1 For $`a1`$ and $`1j5`$, dor($`a,2a+j)4`$.
Proof. This follows from Theorem 3 and Lemma 1. $`\mathrm{}`$
Remark 2 Theorems 2 and 3, along with the following result from , provide an alternate proof of Theorem 1 without the use of Rado’s regularity theorem.
###### Lemma 2
Assume $`b(2^{3/2}1)a2^{3/2}+2`$. Then dor$`(a,b)2\mathrm{log}_2c,`$ where $`c=b/a`$.
Below we give a table showing the known bounds on the degrees of regularity for some small values of $`a`$ and $`b`$. The entries in the table that improve the previously known bounds are marked with \*; all others are from . The improved bounds for dor(1,5), dor(1,6), dor(1,7), dor(1,8), and dor(1,9) follow from Theorem 3; the upper bound on dor(2,10) follows from Theorem 2; and the upper bounds on dor(3,4) and dor(3,7) follow from Lemma 1.
$$\begin{array}{cccccc}(a,b)\hfill & \mathrm{dor}(a,b)\hfill & (a,b)\hfill & \mathrm{dor}(a,b)\hfill & (a,b)\hfill & \mathrm{dor}(a,b)\hfill \\ & & & & & \\ (1,1)\hfill & \mathrm{}\hfill & (2,2)\hfill & 3^{}4^{}\hfill & (3,3)\hfill & 25\hfill \\ (1,2)\hfill & 1\hfill & (2,3)\hfill & 2\hfill & (3,4)\hfill & 23^{}\hfill \\ (1,3)\hfill & 23\hfill & (2,4)\hfill & 1\hfill & (3,5)\hfill & 2\hfill \\ (1,4)\hfill & 24\hfill & (2,5)\hfill & 23\hfill & (3,6)\hfill & 1\hfill \\ (1,5)\hfill & 24^{}\hfill & (2,6)\hfill & 23\hfill & (3,7)\hfill & 23^{}\hfill \\ (1,6)\hfill & 24^{}\hfill & (2,7)\hfill & 24\hfill & (3,8)\hfill & 23\hfill \\ (1,7)\hfill & 24^{}\hfill & (2,8)\hfill & 24\hfill & (3,9)\hfill & 23\hfill \\ (1,8)\hfill & 25^{}\hfill & (2,9)\hfill & 24\hfill & (3,10)\hfill & 24\hfill \\ (1,9)\hfill & 25^{}\hfill & (2,10)\hfill & 24^{}\hfill & (3,11)\hfill & 24\hfill \end{array}$$
Acknowledgement The result that $`(1,1)`$ is the only regular pair has been independently shown by Fox and Radoicic . They show that, in fact, dor$`(a,b)23`$ for all $`(a,b)(1,1)`$.
## References
T. Brown and B. Landman, Monochromatic Arithmetic Progressions with Large Differences, Bull. Australian Math. Soc. 60 (1999), 21-35.
J. Fox and R. Radoicic, preprint
B. Landman and A. Robertson, On Generalized van der Waerden Triples, Disc. Math. 256 (2002), 279-290.
B. Landman and A. Robertson, Ramsey Theory on the Integers, STML 24, Am. Math. Soc., 2004.
B. L. van der Waerden, Bewis einer Baudetschen Vermutung, Nieuw. Arch. Wisk. 15 (1927), 212-216. |
warning/0507/astro-ph0507386.html | ar5iv | text | # Force-Free Magnetohydrodynamic Waves: Non-Linear Interactions and Effects of Strong Gravity
## I Introduction
Magnetohydrodynamics can be studied in the ultra-relativistic limit, where the energy density in the conducting matter which enforces $`𝐄𝐁=0`$ is much less than in the electromagnetic field itself. The Euler equation reduces to the force-free equation $`J^\mu F_{\mu \nu }=0`$, which provides a useful starting point for investigating some aspects of the dynamics of accreting and outflowing matter around compact stars, and classical gamma-ray bursts (e.g. Blandford and Znajek (1977), Uchida (1997), Thompson and Blaes (1998), Lyutikov and Blandford (2003)). In this regime a uniform magnetofluid supports two modes: incompressible torsional waves analogous to Alfvén waves in non-relativistic magnetohydrodynamics (MHD), and a compressible wave analogous to the fast mode. Both of these modes have exact non-linear solutions in a uniform magnetic field in Minkowski space, which means in turn that there is no spontaneous decay of one type of mode into the other.
This fundamental property of a relativistic magnetofluid changes in the presence of a strong gravitational field. We focus in this paper on the effect of spacetime curvature on the interactions between torsional and compressive MHD modes. There is a non-linear interaction if the background spacetime is static, which appears at linear order if it is slowly rotating. We analyze the effect first in the weak-field regime, and consider both spherically and cylindrically symmetric gravitational fields. As a particular example where strong-field effects are important, we also consider the case of a uniform magnetofluid immersed in the spacetime of a black string.
Our main goal here is to explore the fundamentals of the interactions between MHD waves and gravitational fields. Although the scattering of gravitational waves, scalar waves, and vaccum electromagnetic waves have long been studied in black hole spacetimes Chandrasekhar (1992), analogous work on force-free magnetohydrodynamics has focused on the short-wavelength limit Uchida (1997). There are two well-established astrophysical contexts in which the dynamics may be well described, in a first approximation, by a relativistic force-free fluid: the magnetosphere of a bursting magnetar (e.g. Thompson and Duncan (1995)); and the polar regions of an accreting black hole Blandford and Znajek (1977). We restrict ourselves, in this paper, to the simplest case where the background magnetofluid does not maintain a large-scale current (as it does surrounding a Kerr black hole).
In a dynamic situation, a fluid formulation of force-free magnetohydrodynamics provides a physically transparent set of variables. Two such formalisms have been developed, by Achterberg Achterberg (1983) and Thompson & Blaes Thompson and Blaes (1998). These are, in fact, dual versions of each other, and are connected by interchanging the dynamical role of the Lagrangian fluid coordinates and the Eulerian coordinates of the background spacetime. In reviewing these formalisms, we generalize the second to an arbitrary curvilinear coordinate system.
Another of our goals is to extend the discussion of non-linear mode interactions in force-free MHD, begun in Thompson and Blaes (1998). We examine in more detail how obliquely propagating Alfvén modes and fast modes will perturb each other, including the effects of spacetime curvature. The effect of field line wandering on mode collisions has received considerable attention in recent years, in the case of non-relativistic and incompressible Alfvén turbulence. We examine its effect on the collisions between compressive (fast) modes in a relativistic magnetofluid in flat Minkowski space. In contrast with the case of colliding Alfvén waves, the effect of a net field-line displacement cannot be expressed in kinematic terms through a zero-frequency component of one of the colliding waves.
The sections are organized as follows. In Section II, we given an overview the electric and magnetic lagrangian fluid formulations of force-free MHD. As a toy example, we construct the solution for the uniform magnetic field in a spherically symmetric spacetime (with non-vanishing Ricci tensor). In Section III we review the normal modes of a uniform, force-free magnetofluid. Section IV is devoted to the scattering of a torsional wave in a curved spacetime: we first consider the propagation of such a wave in a general, cylindrically symmetric spacetime; and then consider the scattering of such a wave in a shallow, spherical gravitational potential. Section V generalizes the results of the previous section to the case of two colliding, axisymmetric torsional modes in a gravitational field. The final Section VI considers the more general case of collisions between non-axisymmetric MHD modes in Minkowski space, focusing on the three-mode couplings of two colliding Alfvén and fast waves.
## II Lagrangian Formulations of Ultrarelativistic MHD
The equations describing the behavior of ideal fluids coupled to electromagnetic fields in curved spacetimes are inherently very complicated. A significant simplification is obtained by assuming that the spacetime metric is fixed; i.e., that the mass-energy of the fluid can be neglected in comparison with that of the source of gravity. In this paper we make a second simplification: the role of the charged matter fields is restricted to a source of electric currents which cancel off the electric field in the fluid rest frame. The mass-energy of these matter fields is, in other words, negligible in comparison with that of the electromagnetic field.
We begin by summarizing the two Lagrangian fluid formulations of relativistic, force-free MHD.
### II.1 MHD equations in the force-free limit
The covariant Maxwell Equations are
$$4\pi J^\mu =_\nu F^{\mu \nu }=\frac{1}{\sqrt{g}}_\nu \left(\sqrt{g}g^{\mu \alpha }g^{\nu \beta }F_{\alpha \beta }\right)$$
(1)
and
$$_\alpha F_{\mu \nu }+_\nu F_{\alpha \mu }+_\mu F_{\nu \alpha }=_\alpha F_{\mu \nu }+_\nu F_{\alpha \mu }+_\mu F_{\nu \alpha }=0.$$
(2)
When the only force acting on the charged matter in a local inertial frame is the Lorentz force, the dynamics of the electromagnetic field is described by the force-free equation
$$J^\mu F_{\mu \nu }=\frac{1}{4\pi }_\alpha F^{\mu \alpha }F_{\mu \nu }=0$$
(3)
in combination with
$$F_{\mu \nu }\stackrel{~}{F}^{\mu \nu }=0.$$
(4)
Here
$$\stackrel{~}{F}^{\mu \nu }=\frac{1}{2\sqrt{g}}\epsilon ^{\mu \nu \alpha \beta }F_{\alpha \beta }$$
(5)
is the dual of the electromagnetic field tensor, expressed in terms of the antisymmetric symbol with unit norm $`|\epsilon ^{\mu \nu \alpha \beta }|=1`$.
### II.2 Electric lagrangian variables
Achterberg Achterberg (1983) developed a variational principle for relativistic magnetohydrodynamics in which the Lagrangian coordinates $`x_0^\mu `$ of the fluid are the field variables, moving in a fixed Eulerian space $`x^\mu `$. In the ideal MHD limit, where the electric field vanishes in the rest frame of the fluid, the Lie derivative of the Faraday two-form vanishes, $`\mathrm{\pounds }_uF=0`$, along the direction $`u`$ of fluid motion. One thereby obtains a simple relation between the background (reference) state and the dynamic state of the magnetofluid. Solving for the $`x_0^\mu `$ allows one to obtain all relevant information about the system.
The Faraday two-form is related to its background counterpart by
$$F_{\mu \nu }=\frac{x_{0}^{}{}_{}{}^{\alpha }}{x^\mu }\frac{x_{0}^{}{}_{}{}^{\beta }}{x^\nu }F_{\alpha \beta }^{(0)}.$$
(6)
In the force-free limit, the action is
$$S=d^4x\frac{1}{4}\sqrt{g}g^{\mu \rho }g^{\nu \sigma }F_{\mu \nu }F_{\rho \sigma }.$$
(7)
By extremizing this action, the equation of motion
$$J^\mu F_{\mu \nu }=0$$
(8)
can be obtained. Written in terms of the coordinate fields this is
$$\frac{x_0^\alpha }{x^\mu }F_{\alpha \nu }^{(0)}\frac{}{x^\rho }\left(\sqrt{g}g^{\rho \sigma }g^{\mu \tau }\frac{x_0^\gamma }{x^\sigma }\frac{x_0^\delta }{x^\tau }F_{\gamma \delta }^{(0)}\right)=0.$$
(9)
Here, we have eliminated a factor of $`x_0^\beta /x^\nu `$ from $`F_{\mu \nu }`$. The Bianchi identities are automatically satisfied as long as they are satisfied by the background field. Once the equations for these coordinate fields have been solved, the electromagnetic field is obtained from Eq. (6).
### II.3 Magnetic lagrangian variables
Two new MHD formalisms in the extreme relativistic limit were recently developed by Thompson and Blaes Thompson and Blaes (1998). The first of these is related by a duality transformation to the action presented by Achterberg, and is of interest here. In this formalism, the dual tensor $`\stackrel{~}{F}`$ (rather than $`F`$) plays a central role and the dynamical fields are the perturbed coordinates $`x^\mu `$ (rather than the $`x_0^\mu `$). In this section, we generalize the treatment of Thompson and Blaes (1998) to include gravity.
Once the fields $`x^\mu `$ are solved, the electromagnetic fields can be found from the transformation
$$\stackrel{~}{F}^{\mu \nu }=\frac{\sqrt{g(x_0)}}{J_4\sqrt{g(x)}}\frac{x^\mu }{x_0^\alpha }\frac{x^\nu }{x_0^\beta }\stackrel{~}{F}_0^{\alpha \beta }.$$
(10)
Here
$$J_4=det\left(\frac{x^\mu }{x_0^\alpha }\right)$$
(11)
is the Jacobian of the transformation $`x_0^\alpha x^\mu `$ from Lagrangian to Eulerian variables, and the factor of $`\sqrt{g_0/g}`$ is necessary in a curvilinear coordinate system. We show in Appendix A that the force-free equation $`J^\mu F_{\mu \nu }=0`$ may be derived from the action
$$S^{}=d^4x_0L^{}=d^4x\frac{1}{4}\sqrt{g}g_{\mu \rho }g_{\nu \sigma }\stackrel{~}{F}^{\mu \nu }\stackrel{~}{F}^{\rho \sigma }=d^4x_0\frac{J_4}{4}\sqrt{g}g_{\mu \rho }g_{\nu \sigma }\stackrel{~}{F}^{\mu \nu }\stackrel{~}{F}^{\rho \sigma }.$$
(12)
Along the way, we obtain the equation of motion
$$\stackrel{~}{F}^{\delta \epsilon }\frac{}{x^\epsilon }\left[g_{\beta \delta }g_{\gamma \mu }\stackrel{~}{F}^{\beta \gamma }\right]\frac{1}{2}\stackrel{~}{F}^{\delta \epsilon }\frac{}{x^\mu }\left[g_{\beta \delta }g_{\gamma \epsilon }\stackrel{~}{F}^{\beta \gamma }\right]=0.$$
(13)
In cases where the background magnetic field asymptotes to a constant at large distances, we can use the simple background
$$\stackrel{~}{F}_0^{\mu \nu }=B_0(\delta _0^\mu \delta _3^\nu \delta _3^\mu \delta _0^\nu );\sqrt{g_0}=1,$$
(14)
and define proper time $`\tau t_0`$ and distance $`\sigma z_0`$. The equation of motion (13) then takes the form
$$\frac{x^\epsilon }{x_0^\alpha }\left(\frac{x^\delta }{\tau }\frac{}{\sigma }\frac{x^\delta }{\sigma }\frac{}{\tau }\right)\mathrm{\Sigma }_{\delta \epsilon }\left(\frac{x^\delta }{\tau }\frac{x^\epsilon }{\sigma }\right)\frac{}{x_0^\alpha }\mathrm{\Sigma }_{\delta \epsilon }=0,$$
(15)
where
$$\mathrm{\Sigma }_{\mu \nu }\frac{g_{\mu \alpha }g_{\nu \beta }}{\sqrt{g}J_4}\left(\frac{x^\alpha }{\tau }\frac{x^\beta }{\sigma }\frac{x^\beta }{\tau }\frac{x^\alpha }{\sigma }\right)=\mathrm{\Sigma }_{\nu \mu }.$$
(16)
Although Eq. (13) has four components, it is straightforward to check that the longitudinal components $`\mu =0,3`$ vanish. Physically, this is equivalent to the statement that MHD waves have two independent transverse polarizations. Notice also that it is possible to simplify Eq. (13) further by choosing a time slicing in which
$$\sqrt{g}J_4=1.$$
(17)
The cost of this gauge choice is that one can no longer identify the background time coordinate $`t`$ with the proper time $`\tau `$ of the magnetofluid.
### II.4 Uniform magnetic field: Spherically symmetric spacetime
There is a simple prescription for constructing a uniform magnetic field ($`J^\varphi =0`$) in a vacuum spacetime with an axial killing vector $`\xi ^\mu =g_{}^{\mu }{}_{\varphi }{}^{}`$. One identifies Wald (1974)
$$A_\mu =\frac{1}{2}B_0\xi _\mu =\frac{1}{2}B_0g_{\varphi \mu }.$$
(18)
Thus $`F_{\mu \nu }`$ takes exactly the same form outside a black hole of vanishing charge and spin, in the standard Schwarzschild coordinates ($`g_{\varphi \varphi }=r^2\mathrm{sin}^2\theta `$), as it does in Minkowski space:
$$F_{r\varphi }=B_0r\mathrm{sin}^2\theta ;F_{\theta \varphi }=B_0r^2\mathrm{sin}\theta \mathrm{cos}\theta .$$
(19)
More generally this method is of limited utility, since it depends on the vanishing of the Ricci tensor of the background spacetime. To give a further illustration of the lagrangian fluid method, we show how it may be used to construct a uniform magnetic field in a general spherical spacetime. The metric may be written as
$$ds^2=g_{tt}dt^2+g_{rr}dr^2+r^2(d\theta ^2+\mathrm{sin}^2\theta d\varphi ^2).$$
(20)
We imagine that the currents supporting the magnetic field are flowing far outside the self-gravitating mass, which itself is not electrically conducting. Then
$$F_{\mu \nu }=(_\mu R_0_\nu \varphi _0_\mu \varphi _0_\nu R_0)F_{R\varphi }^0=\frac{1}{2}(_\mu R_0^2_\nu \varphi _\mu \varphi _0_\nu R_0^2)B_0.$$
(21)
Here $`R_0=r_0\mathrm{sin}\theta _0`$ is the cylindrical radius and $`F_{R\varphi }^0=B_0R_0`$ is the field strength that would be supported by the same currents in flat space. The vanishing of
$$4\pi \sqrt{g}J^\varphi =_r\left(\sqrt{g}g^{\varphi \varphi }g^{rr}F_{\varphi r}\right)+_\theta \left(\sqrt{g}g^{\varphi \varphi }g^{\theta \theta }F_{\varphi \theta }\right)$$
(22)
combined with the separation of variables
$$r_0^2\mathrm{sin}^2\theta _0=(r)𝒯(\theta )$$
(23)
gives the coupled ordinary differential equations
$$\frac{r^2}{\sqrt{g_{tt}g_{rr}}}\frac{d}{dr}\left(\sqrt{\frac{g_{tt}}{g_{rr}}}\frac{d}{dr}\right)=\frac{\mathrm{sin}\theta }{𝒯}\frac{d}{d\theta }\left(\frac{1}{\mathrm{sin}\theta }\frac{d𝒯}{d\theta }\right).$$
(24)
Because the equation for $`𝒯`$ is the same as in flat space, we may write
$$\mathrm{sin}\theta _0=\mathrm{sin}\theta $$
(25)
and
$$F_{r\varphi }=\frac{1}{2}B_0\frac{d}{dr}\mathrm{sin}^2\theta ;F_{\theta \varphi }=B_0\mathrm{sin}\theta \mathrm{cos}\theta ,$$
(26)
where $``$ is a solution to
$$r^2\frac{d}{dr}\left(\sqrt{\frac{g_{tt}}{g_{rr}}}\frac{d}{dr}\right)2\sqrt{g_{tt}g_{rr}}=0.$$
(27)
If the Ricci tensor of the background spacetime vanishes at large radius, then the appropriate solution to (27) must match onto $`(r)=r^2`$ at large $`r`$ (where $`g_{tt}=g_{rr}^1=12GM/r`$).
### II.5 Uniform electromagnetic field: Cylindrically symmetric spacetime
We will be concerned with the propagation of MHD waves in cylindrically symmetric geometries of the following form
$$ds^2=g_{tt}dt^2+g_{rr}dr^2+g_{\varphi \varphi }d\varphi ^2+g_{zz}dz^2+2g_{t\varphi }dtd\varphi .$$
(28)
When $`g_{t\varphi }0`$, the frame dragging of the background $`z`$-magnetic field generates a radial electric field. Thus, the background electromagnetic field is a stationary, axisymmetric, and $`z`$-independent solution to the equation $`J^\mu =0`$ with non-vanishing components $`F_{r\varphi }^{(0)}`$ and $`F_{rt}^{(0)}`$,
$$_r\left(\sqrt{g}g^{tt}g^{rr}F_{tr}^{(0)}+\sqrt{g}g^{t\varphi }g^{rr}F_{\varphi r}^{(0)}\right)=0;$$
(29)
$$_r\left(\sqrt{g}g^{\varphi \varphi }g^{rr}F_{\varphi r}^{(0)}+\sqrt{g}g^{\varphi t}g^{rr}F_{tr}^{(0)}\right)=0.$$
(30)
Since the electric field vanishes in the static limit, one has
$$F_{rt}^{(0)}=\frac{g_0^{t\varphi }}{g_0^{tt}}F_{r\varphi }^{(0)}$$
(31)
and
$$F_{r\varphi }^{(0)}=\frac{B_0}{\sqrt{g_0}g_0^{rr}[g_0^{\varphi \varphi }(g_0^{t\varphi })^2/g_0^{tt}]}=\frac{g_{0rr}g_{0\varphi \varphi }}{\sqrt{g_0}}B_0.$$
(32)
\[Here $`g_0g(x_0)`$.\]
We will investigate perturbations to the fields, subject to the constraint $`F^{\mu \nu }\stackrel{~}{F}_{\mu \nu }=0`$, in spacetimes with varying degrees of curvature. It is useful first to consider the case of a nearly flat and non-rotating cylindrical spacetime. In general, static spacetimes which are symmetric about the z-axis may be written in following form Synge (1960)
$$ds^2=e^{2\lambda }dt^2+e^{2(\nu \lambda )}(dr^2+dz^2)+r^2e^{2\lambda }d\varphi ^2.$$
(33)
The function $`\lambda `$ satisfies Laplace’s equation in cylindrical coordinates, and $`\nu `$ is second order in $`\lambda `$. In the weak field limit this metric reduces to
$$ds^2=(1+2\lambda )dt^2+(12\lambda )(dr^2+dz^2)+r^2(12\lambda )d\varphi ^2,$$
(34)
and
$$\lambda =Gd^3x^{}\frac{\rho (x^{})}{|xx^{}|}$$
(35)
becomes the usual Newtonian potential. The background fields (32) and (31) now reduce to
$$F_{r\varphi }^{(0)}=(12\lambda )B_0r;F_{rt}^{(0)}=0.$$
(36)
A cylindrically symmetric spacetime with some attributes of $`3+1`$-dimensional black holes can be constructed from the product of $`2+1`$-dimensional black hole (sitting in the $`r`$-$`\varphi `$ plane) and an infinite line segment along the $`z`$-axis. The $`2+1`$ black hole solution was originally found by Banados, Teitelboim and Zanelli Banados et al. (1992); its properties are reviewed by Carlip Carlip (1995). The key element in its construction is the introduction of a negative cosmological constant $`\mathrm{\Lambda }=\frac{1}{\mathrm{}^2}`$, so that the spacetime asymptotes to anti-de Sitter space at large radius. The BTZ metric is
$$ds^2=\left(M+\frac{r^2}{\mathrm{}^2}+\frac{J^2}{4r^2}\right)dt^2+\left(M+\frac{r^2}{\mathrm{}^2}+\frac{J^2}{4r^2}\right)^1dr^2+r^2\left(d\varphi \frac{J}{2r^2}dt\right)^2+dz^2,$$
(37)
or equivalently
$$ds^2=(ZR)dt^2+(ZR)^1dr^2+r^2\left(d\varphi \frac{J}{2r^2}dt\right)^2+dz^2,$$
(38)
where
$$ZM+\frac{r^2}{\mathrm{}^2}$$
(39)
and
$$R1\frac{g_{t\varphi }^2}{g_{tt}g_{\varphi \varphi }}=1+\frac{J^2}{4Zr^2}.$$
(40)
(The variable $`M`$, being a mass per unit length, is dimensionless in units where $`G=c=1`$.) The background Maxwell fields in this spacetime are
$$F_{r\varphi }^{(0)}=\frac{B_0r}{ZR};$$
(41)
$$F_{rt}^{(0)}=\frac{J}{2r^2}F_{r\varphi }^{(0)}=\left(\frac{J}{2r^2}\right)\frac{B_0r}{ZR}.$$
(42)
These expressions reduce to
$$F_{r\varphi }^{(0)}=\frac{B_0r}{r^2/\mathrm{}^2M};F_{rt}^{(0)}=0$$
(43)
in the static limit ($`J=0`$). Minkowski space is recovered by taking $`M1`$ and $`\mathrm{}^2\mathrm{}`$.
## III Force-Free MHD Waves in Curved Spacetimes
### III.1 MHD modes: Force-free limit
We first review the normal modes of a uniform magnetofluid in flat Minkowski space, before considering their interactions and the effects of spacetime curvature. The background magnetic field is $`𝐁=B_0\widehat{z}`$, or equivalently $`F_{r\varphi }=B_0r`$ in cylindrical coordinates. The fluid supports two distinct modes in the force-free limit: the fast mode, which involves a compressive disturbance and has an isotropic dispersion relation $`\omega =|𝐤|`$; and the Alfvén mode, which is incompressible and has a group velocity directed along the background magnetic field, $`d\omega /dk_z=\pm 1`$. There is no slow mode.
The simplest way of seeing that the fluid supports only two normal modes is to note that the perturbation to the Faraday tensor
$$\delta F_{\mu \nu }=F_{\mu \nu }F_{\mu \nu }^{(0)},$$
(44)
can be expressed in terms of two variables $`\delta r`$ and $`\delta \varphi `$. To first order in the perturbation, we have
$$\delta F_{r\varphi }^{(1)}=\frac{}{r}\left(F_{r\varphi }^{(0)}\delta r\right)+\frac{\delta \varphi }{\varphi }F_{r\varphi }^{(0)};\delta F_{t\varphi }^{(1)}=\frac{\delta r}{t}F_{r\varphi }^{(0)};\delta F_{z\varphi }^{(1)}=\frac{\delta r}{z}F_{r\varphi }^{(0)},$$
(45)
and
$$\delta F_{tr}^{(1)}=\frac{\delta \varphi }{t}F_{r\varphi }^{(0)};\delta F_{zr}^{(1)}=\frac{\delta \varphi }{z}F_{r\varphi }^{(0)};\delta F_{tz}^{(1)}=0.$$
(46)
As expected, the component of the electric field parallel to $`𝐁_0`$ vanishes to linear order. The absence of the slow mode can be related to the invariance of $`F_{\mu \nu }^{(0)}`$ under reparameterizations of the $`z`$-coordinate. The full expressions for $`\delta F_{\mu \nu }`$, written in terms of the Lagrangian fluid variables, are collected in Appendix A.
More generally, the Alfvén mode and fast mode involve perturbations to both fluid coordinates. These perturbations are related by the condition of incompressibility
$$\frac{1}{r}\frac{(r\delta r)}{r}+\frac{\delta \varphi }{\varphi }=0(\mathrm{Alfven})$$
(47)
in the case of the Alfvén mode; and the condition of vanishing torsion
$$\frac{\delta r}{\varphi }\frac{(r^2\delta \varphi )}{r}=0(\mathrm{Fast})$$
(48)
in the case of the fast mode. The fluctuating current associated with the field perturbations (45), (46) is
$$4\pi \sqrt{g}J^{t(1)}=_r\left(\sqrt{g}g^{tt}g^{rr}\delta F_{tr}^{(1)}\right)+_\varphi \left(\sqrt{g}g^{tt}g^{\varphi \varphi }\delta F_{t\varphi }^{(1)}\right);$$
(49)
$$4\pi \sqrt{g}J^{z(1)}=_r\left(\sqrt{g}g^{zz}g^{rr}\delta F_{zr}^{(1)}\right)+_\varphi \left(\sqrt{g}g^{zz}g^{\varphi \varphi }\delta F_{z\varphi }^{(1)}\right);$$
(50)
$$4\pi \sqrt{g}J^{r(1)}=_t\left(\sqrt{g}g^{rr}g^{tt}\delta F_{rt}^{(1)}\right)+_z\left(\sqrt{g}g^{rr}g^{zz}\delta F_{rz}^{(1)}\right)+_\varphi \left(\sqrt{g}g^{rr}g^{\varphi \varphi }\delta F_{r\varphi }^{(1)}\right);$$
(51)
and
$$4\pi \sqrt{g}J^{\varphi (1)}=_r\left(\sqrt{g}g^{\varphi \varphi }g^{rr}\delta F_{\varphi r}^{(1)}\right)+_t\left(\sqrt{g}g^{\varphi \varphi }g^{tt}\delta F_{\varphi t}^{(1)}\right)+_z\left(\sqrt{g}g^{\varphi \varphi }g^{zz}\delta F_{\varphi z}^{(1)}\right).$$
(52)
Expressed in terms of the coordinate perturbations, and specializing to Minkowski space, this becomes
$$4\pi rJ^{t(1)}=B_0_t\left[_r\left(r^2\delta \varphi \right)_\varphi \delta r\right];$$
(53)
$$4\pi rJ^{z(1)}=B_0_z\left[_r\left(r^2\delta \varphi \right)_\varphi \delta r\right];$$
(54)
$$4\pi rJ^{r(1)}=B_0r^2\left\{_t^2\delta \varphi +_z^2\delta \varphi +\frac{1}{r^2}_\varphi \left[\frac{1}{r}_r\left(r\delta r\right)+_\varphi \delta \varphi \right]\right\};$$
(55)
and
$$4\pi rJ^{\varphi (1)}=B_0\left\{_t^2\delta r+_z^2\delta r+_r\left[\frac{1}{r}_r\left(r\delta r\right)+_\varphi \delta \varphi \right]\right\}.$$
(56)
The dynamical equations for the two normal modes are recovered by imposing the force-free condition. To linear order one has
$$J^{r(1)}F_{r\varphi }^{(0)}=0=J^{\varphi (1)}F_{\varphi r}^{(0)},$$
(57)
hence $`J^{r(1)}=J^{\varphi (1)}=0`$. In the case of the Alfvén mode, applying the constraint (47) gives the one-dimensional wave equation
$`_t^2\delta r+_z^2\delta r`$ $`=`$ $`0;`$
$`_t^2\delta \varphi +_z^2\delta \varphi `$ $`=`$ $`0,`$ (58)
with a vanishing longitudinal field perturbation
$$\delta F_{r\varphi }^{(1)}=0,$$
(59)
and (in general) a non-vanishing longitudinal current
$$J^{t(1)},J^{z(1)}0.$$
(60)
A force-free magnetofluid supports torsional waves of both helicities. It should be noted that, in this limit, there is no basic distinction between the two helicities, as there is a non-relativistic cold, magnetized plasma.
In the case of the fast mode, one instead finds that the current vanishes entirely and the coordinate perturbations obey three-dimensional wave equations,
$`_t^2\delta \varphi +_z^2\delta \varphi +{\displaystyle \frac{1}{r^3}}_r\left[r_r\left(r^2\delta \varphi \right)\right]+{\displaystyle \frac{1}{r^2}}_\varphi ^2\delta \varphi `$ $`=`$ $`0;`$
$`_t^2\delta r+_z^2\delta r+_r\left[{\displaystyle \frac{1}{r}}_r\left(r\delta r\right)\right]+{\displaystyle \frac{1}{r^2}}_\varphi ^2\delta r`$ $`=`$ $`{\displaystyle \frac{2}{r}}_\varphi \delta \varphi .`$ (61)
By applying the operator $`r^1_rr`$ to the first equation, one sees that an axisymmetric field perturbation
$$\frac{\delta B^z}{B_0}=\frac{\delta F_{\varphi r}^{(1)}}{F_{\varphi r}^{(0)}}=\frac{1}{r}\frac{}{r}\left(r\delta r\right)$$
(62)
satisfies the (axisymmetric) cylindrical wave equation.
Thus, the fast mode is equivalent to a vacuum electromagnetic wave superposed on the uniform background magnetic field, but with its polarization restricted to $`\delta 𝐄𝐁_0=0`$. This way of constructing the fast mode also makes it immediately apparent that infinite plane waves (restricted to this single polarization state) are another general class of solutions.
Inspection of Eqs. (55) and (56) shows that another projection,
$$\frac{1}{r^2}\frac{\delta r}{\varphi }\frac{\delta \varphi }{r}=0,$$
(63)
allows the equations of motion for $`\delta r`$ and $`\delta \varphi `$ to be entirely separated,
$`_t^2\delta \varphi +_z^2\delta \varphi +{\displaystyle \frac{1}{r^3}}_r\left(r^3_r\delta \varphi \right)+{\displaystyle \frac{1}{r^2}}_\varphi ^2\delta \varphi `$ $`=`$ $`0;`$
$`_t^2\delta r+_z^2\delta r+_r\left[{\displaystyle \frac{1}{r}}_r\left(r\delta r\right)\right]+{\displaystyle \frac{1}{r^2}}_\varphi ^2\delta r=0.`$ (64)
At the same time, the longitudinal currents do not vanish,
$$J^{t(1)}=\frac{B_0}{2\pi }_t\delta \varphi ;J^{z(1)}=\frac{B_0}{2\pi }_z\delta \varphi .$$
(65)
### III.2 Decomposition into normal modes in a curved spacetime
We have seen that, even in Minkowski space, there is an ambiguity in the definition of the fast mode. Axially symmetric perturbations decompose directly into torsional modes, which are supported only by a perturbation $`\delta \varphi `$, and compressive modes, which are supported only by a perturbation $`\delta r`$. In the more general case of non-axisymmetric perturbations, one can still define unambiguously an incompressible mode (47). But one is faced with a choice either of defining the fast mode to be current free through Eq. (48), in which case the equations of motion (III.1) for $`\delta r`$ and $`\delta \varphi `$ do not entirely separate; or instead of using the alternative projection (63) which separates these equations but leaves a non-vanishing longitudinal current (65).
This ambiguity remains when one considers the case of a magnetofluid in a gravitational field. We focus here on a general static ($`g_{t\varphi }=0`$) and cylindrically symmetric spacetime. In this case, the longitudinal currents (49) and (50) cannot both vanish, as is seen by writing them in terms of the coordinate perturbations:
$$\frac{4\pi }{B_0}\sqrt{g}J^{t(1)}=\frac{}{t}\left[\frac{}{r}\left(\frac{g_{\varphi \varphi }}{g_{tt}}\delta \varphi \right)+\frac{g_{rr}}{g_{tt}}\frac{\delta r}{\varphi }\right];$$
(66)
and
$$\frac{4\pi }{B_0}\sqrt{g}J^{z(1)}=\frac{}{z}\left[\frac{}{r}\left(\frac{g_{\varphi \varphi }}{g_{zz}}\delta \varphi \right)+\frac{g_{rr}}{g_{zz}}\frac{\delta r}{\varphi }\right].$$
(67)
In these expressions, $`g_{tt}`$ and $`g_{zz}`$ have different dependences on radius $`r`$, and we have made use of the diagonality of the metric. We conclude that, to linear order in the field perturbation, non-axisymmetric excitations of the fast mode are current-carrying.
It is, nonetheless, possible to choose a generalization of the projection (63) which does allow the equations of motion for $`\delta r`$ and $`\delta \varphi `$ to decouple at linear order. From Eqs. (66) and (67), the obvious choice is
$$\frac{g_{rr}}{g_{\varphi \varphi }}\frac{\delta r}{\varphi }\frac{\delta \varphi }{r}=0.(\mathrm{Fast})$$
(68)
Then Eq. (52) becomes
$$\left(\frac{1}{g_{tt}}\frac{^2\delta r}{t^2}+\frac{1}{g_{zz}}\frac{^2\delta r}{z^2}\right)F_{r\varphi }^{(0)}+\frac{1}{g_{\varphi \varphi }}\frac{^2\delta r}{\varphi ^2}F_{r\varphi }^{(0)}+\frac{g_{\varphi \varphi }}{\sqrt{g}}\frac{}{r}\left[\frac{\sqrt{g}}{g_{\varphi \varphi }g_{rr}}\frac{}{r}\left(\delta rF_{r\varphi }^{(0)}\right)\right]=0,$$
(69)
or, upon substituting expression (32),
$$\frac{1}{g_{tt}}\frac{^2\delta r}{t^2}+\frac{1}{g_{zz}}\frac{^2\delta r}{z^2}+\frac{1}{g_{\varphi \varphi }}\frac{^2\delta r}{\varphi ^2}+\frac{1}{g_{rr}}\frac{}{r}\left[\frac{\sqrt{g}}{g_{\varphi \varphi }g_{rr}}\frac{}{r}\left(\delta r\frac{g_{rr}g_{\varphi \varphi }}{\sqrt{g}}\right)\right]=0.$$
(70)
The longitudinal currents then are
$$\frac{4\pi }{B_0}\sqrt{g}J^{t(1)}=\frac{}{r}\left(\frac{g_{\varphi \varphi }}{g_{tt}}\right)\frac{\delta \varphi }{t};\frac{4\pi }{B_0}\sqrt{g}J^{z(1)}=\frac{}{r}\left(\frac{g_{\varphi \varphi }}{g_{zz}}\right)\frac{\delta \varphi }{z}.$$
(71)
The perturbation $`\delta \varphi `$ can be recovered from $`\delta r`$ using the constraint (68). Similarly, the incompressible Alfvén mode is defined by
$$\frac{\sqrt{g}}{g_{rr}g_{\varphi \varphi }}\frac{}{r}\left(\frac{g_{rr}g_{\varphi \varphi }}{\sqrt{g}}\delta r\right)+\frac{\delta \varphi }{\varphi }=0,(\mathrm{Alfven})$$
(72)
to linear order.
For example, the wave equation for a purely compressive disturbance in a BTZ black string spacetime is
$$\frac{1}{Z^2}\frac{^2\delta r}{t^2}+\frac{1}{Z}\frac{^2\delta r}{z^2}+\frac{1}{r^2Z}\frac{^2\delta r}{\varphi ^2}+\frac{}{r}\left[\frac{Z}{r}\frac{}{r}\left(\frac{r\delta r}{Z}\right)\right]=0.$$
(73)
(As before, $`Z(r)r^2/\mathrm{}^2M`$.) When $`_z\delta r=_\varphi \delta r=0`$, this equation can also be re-expressed directly in terms of the field perturbation
$$\frac{\delta B^z}{B^z}=\frac{Z}{r}\frac{}{r}\left(\frac{r\delta r}{Z}\right)$$
(74)
as
$$\frac{^2}{t^2}\left(\frac{\delta B^z}{B^z}\right)+\frac{Z}{r}\frac{}{r}\left[Zr\frac{}{r}\left(\frac{\delta B^z}{B^z}\right)\right]=0.$$
(75)
More generally, the field perturbations have some dependence on $`z`$, and it is more useful to work with the coordinate perturbation.
## IV Scattering of Torsional Waves by Spacetime Curvature
We now turn to an important, but subtle, difference between the compressive (fast) and torsional (Alfvén) modes in a curved spacetime. We focus on a uniform magnetofluid in which the background current vanishes. Both modes are exact non-linear solutions to the force-free equation in Minkowski space. This property is retained by a purely radial oscillation ($`\delta \varphi =0`$), in any static cylindrically symmetric (or spherically symmetric) spacetime. The resulting fast mode is equivalent to an electromagnetic wave polarized in the $`\varphi `$-direction, with $`\delta 𝐄𝐁_0=0`$. Hence it can have arbitrary amplitude.
An axisymmetric torsional wave is not, by contrast, a non-linear solution to $`J^\mu F_{\mu \nu }=0`$, because the wave speed varies with radius. We now show that, in a static spacetime, such an Alfvén wave will excite a compressive motion transverse to the background magnetic field. The amplitude of this compressive motion is second order in the amplitude $`\epsilon `$ of the Alfvén wave itself. This effect is somewhat analogous to the non-linear coupling between a fast mode and an Alfvén wave that is propagating through a non-relativistic magnetofluid with a gradient in density and in Alfvén speed (e.g. Nakariakov et al. (1997), Tsiklauri et al. (2002)). In a spherical spacetime, we show that a torsional wave originating at infinity is scattered into a spherical outgoing compressive wave.
To this end, we expand the field and currents perturbations in powers of $`\epsilon 1`$:
$$F_{\mu \nu }=F_{\mu \nu }^{(0)}+\delta F_{\mu \nu }^{(1)}+\delta F_{\mu \nu }^{(2)}+O(\epsilon ^3);$$
(76)
$$J^\mu =\delta J^{\mu (1)}+\delta J^{\mu (2)}+O(\epsilon ^3).$$
(77)
Here, $`\delta F_{\mu \nu }^{(N)}=O(\epsilon ^N)`$, etc.
When considering the effects of spacetime curvature, we also restrict ourselves to axisymmetric perturbations. In a static spacetime, this means that $`\delta r`$ and $`\delta \varphi `$ decouple to linear order. Such a linear coupling is re-established in a rotating spacetime.
### IV.1 Torsional waves in a cylindrically symmetric spacetime
#### IV.1.1 Static spacetime ($`g_{t\varphi }=0`$)
Torsional deformations of the magnetic field are supported by electric currents which, to leading order in $`\epsilon `$, propagate along the background magnetic field. We first derive the analog, in a curved cylindrical spacetime, of the linear wave equation (III.1). With this in hand, we find solutions to $`J^\mu F_{\mu \nu }=0`$ to second order in $`\epsilon `$.
The force-free equation implies $`J^{r(1)}=0`$ to linear order (Eq. ). Combining expression (51) with the with the Bianchi identity $`_t\delta F_{rz}^{(1)}=_z\delta F_{rt}^{(1)}`$ gives
$$g^{tt}_t^2\left(\delta F_{rz}^{(1)}\right)+g^{zz}_z^2\left(\delta F_{rz}^{(1)}\right)=0.$$
(78)
The electric field $`\delta F_{rt}^{(1)}`$ satisfies the same wave equation. The solution is
$$\delta F_{rz}^\pm =\delta F_{rz}^\pm (zv_zt);\delta F_{rt}^\pm =\delta F_{rt}^\pm (zv_zt).$$
(79)
where
$$v_z\sqrt{\frac{g_{tt}}{g_{zz}}}.$$
(80)
The two fluctuating field components are related by
$$\delta F_{tr}^\pm =v_z\delta F_{zr}^\pm .$$
(81)
The net result is that the Alfvén wave propagates along the $`z`$-axis, to leading order in $`\epsilon `$, with a phase speed equal to that of a null geodesic that lies tangent to the background magnetic field.
An Alfvén wave front propagating along the background magnetic field will be bent as the result of the dependence of $`g_{tt}`$ (and $`v_z`$) on radius. This, in turn, forces a radial oscillation of the magnetofluid that is second order in $`\epsilon `$. Indeed, inspection of the force-free equation
$$J^{t(1)}\delta F_{tr}^{(1)}+J^{z(1)}\delta F_{zr}^{(1)}+J^{\varphi (2)}F_{\varphi r}^{(0)}=0,$$
(82)
shows that the first two terms generally do not cancel. Using Eq. (81) to eliminate $`\delta F_{tr}`$ in terms of $`\delta F_{zr}`$, one obtains
$$g^{rr}g^{zz}(\delta F_{zr}^{(1)})^2_r(\mathrm{ln}v_z)+4\pi J^{\varphi (2)}F_{\varphi r}^{(0)}=0.$$
(83)
The radial fluctuation of the magnetofluid supports the additional current perturbation,
$$4\pi \sqrt{g}J^{\varphi (2)}=B_0\left\{\frac{g_{rr}}{g_{tt}}\frac{^2\delta r}{t^2}+\frac{g_{rr}}{g_{zz}}\frac{^2\delta r}{z^2}+\frac{}{r}\left[\frac{\sqrt{g}}{g_{\varphi \varphi }g_{rr}}\frac{}{r}\left(\delta r\frac{g_{rr}g_{\varphi \varphi }}{\sqrt{g}}\right)\right]\right\}.$$
(84)
Let us first consider the special case of a nearly flat cylindrical spacetime. In such a background, $`v_z=1+2\lambda `$ and Eq. (83) reduces to
$$\frac{1}{2\pi }(\delta F_{zr}^{(1)})^2\frac{d\lambda }{dr}+J^{\varphi (2)}F_{\varphi r}^{(0)}=0,$$
(85)
to lowest order in $`\lambda `$. Re-expressing this equation in terms of the perturbations $`\delta r`$ and $`\delta \varphi `$ gives
$$\frac{^2\delta r}{t^2}+\frac{}{r}\left[\frac{1}{r}\frac{}{r}(r\delta r)\right]+\frac{^2\delta r}{z^2}=2r^2\frac{d\lambda }{dr}\left(\frac{\delta \varphi }{z}\right)^2.$$
(86)
The analogous expressions for the BTZ black string spacetime are not much more complicated. In this case, $`v_z=Z^{1/2}=(r^2/\mathrm{}^2M)^{1/2}`$, which vanishes at the horizon $`r_H=M^{1/2}\mathrm{}`$. The force-free equation reduces to
$$\frac{1}{2}(\delta F_{zr}^{(1)})^2_rZ+4\pi J^{\varphi (2)}F_{\varphi r}^{(0)}=0.$$
(87)
We then have
$$\frac{1}{Z^2}\frac{^2\delta r}{t^2}+\frac{1}{Z}\frac{^2\delta r}{z^2}+\frac{}{r}\left[\frac{Z}{r}\frac{}{r}\left(\frac{r\delta r}{Z}\right)\right]=\frac{1}{Z}\frac{r^3}{l^2}\left(\frac{\delta \varphi }{z}\right)^2.$$
(88)
#### IV.1.2 Non-static spacetime $`(g_{t\varphi }0)`$
When the spacetime rotates $`(g_{t\varphi }0`$), one finds that the Alfvén wave ansatz (81) fails to satisfy the force-free equation to linear order in $`\epsilon `$. Physically, this is because the torsional and radial modes of the magnetofluid are coupled through conservation of angular momentum. One must therefore introduce simultaneous perturbations in $`\delta r`$, $`\delta \varphi `$, and $`\delta t`$. In terms of these variables, the field perturbations become (to first order)
$$\delta F_{r\varphi }^{(1)}=\frac{}{r}(F_{r\varphi }^{(0)}\delta r);\delta F_{t\varphi }^{(1)}=\frac{(\delta r)}{t}F_{r\varphi }^{(0)};\delta F_{z\varphi }^{(1)}=\frac{(\delta r)}{z}F_{r\varphi }^{(0)};$$
(89)
$$\delta F_{rt}^{(1)}=\frac{(\delta \stackrel{~}{\varphi })}{t}F_{r\varphi }^{(0)}\frac{}{r}\left(\frac{g^{t\varphi }}{g^{tt}}F_{r\varphi }^{(0)}\delta r\right);\delta F_{rz}^{(1)}=\frac{(\delta \stackrel{~}{\varphi })}{z}F_{r\varphi }^{(0)};\delta F_{tz}^{(1)}=\frac{\delta r}{z}\left(\frac{g^{t\varphi }}{g^{tt}}\right)F_{\varphi r}^{(0)}.$$
(90)
Here
$$\delta \stackrel{~}{\varphi }\delta \varphi \left(\frac{g^{t\varphi }}{g^{tt}}\right)\delta t=\delta \varphi +\left(\frac{g_{t\varphi }}{g_{\varphi \varphi }}\right)\delta t.$$
(91)
In this situation, the $`t`$-$`\varphi `$ components of the metric are not diagonal: $`g^{tt}=g_{\varphi \varphi }/det(t,\varphi )`$, $`g^{\varphi \varphi }=g_{tt}/det(t,\varphi )`$, and $`g^{t\varphi }=g_{t\varphi }/det(t,\varphi )`$, where $`det(t,\varphi )g_{tt}g_{\varphi \varphi }g_{t\varphi }^2`$.
The force-free equations are, again to linear order,
$$J^{r(1)}F_{r\varphi }^{(0)}=0=J^{r(1)}F_{rt}^{(0)},$$
(92)
or equivalently
$$J^{r(1)}=0;$$
(93)
and
$$J^{t(1)}F_{tr}^{(0)}+J^{\varphi (1)}F_{\varphi r}^{(0)}=0.$$
(94)
Finally, the currents are given by
$`4\pi \sqrt{g}J^{t(1)}`$ $`=`$ $`_r(\sqrt{g}g^{tt}g^{rr}\delta F_{tr}^{(1)}+\sqrt{g}g^{t\varphi }g^{rr}\delta F_{\varphi r}^{(1)})`$ (95)
$`+`$ $`_z(\sqrt{g}g^{tt}g^{zz}\delta F_{tz}^{(1)})+_z(\sqrt{g}g^{t\varphi }g^{zz}\delta F_{\varphi z}^{(1)});`$
$$4\pi \sqrt{g}J^{r(1)}=_t(\sqrt{g}g^{rr}g^{t\varphi }\delta F_{r\varphi }^{(1)}+\sqrt{g}g^{rr}g^{tt}\delta F_{rt}^{(1)})+_z(\sqrt{g}g^{rr}g^{zz}\delta F_{rz}^{(1)});$$
(96)
and
$`4\pi \sqrt{g}J^{\varphi (1)}`$ $`=`$ $`_r(\sqrt{g}g^{\varphi \varphi }g^{rr}\delta F_{\varphi r}^{(1)}+\sqrt{g}g^{\varphi t}g^{rr}\delta F_{tr}^{(1)})_t(\sqrt{g}(g^{t\varphi })^2\delta F_{\varphi t}^{(1)})`$
$`+`$ $`_t(\sqrt{g}g^{\varphi \varphi }g^{tt}\delta F_{\varphi t}^{(1)})+_z(\sqrt{g}g^{\varphi \varphi }g^{zz}\delta F_{\varphi z}^{(1)})+_z(\sqrt{g}g^{\varphi t}g^{zz}\delta F_{tz}^{(1)}).`$
Substituting for the field perturbations, Eq. (93) becomes
$$g^{tt}\frac{^2(\delta \stackrel{~}{\varphi })}{t^2}+g^{zz}\frac{^2(\delta \stackrel{~}{\varphi })}{z^2}=g^{tt}\frac{}{r}\left(\frac{g_{t\varphi }}{g_{\varphi \varphi }}\right)\frac{\delta r}{t}.$$
(98)
One quickly obtains an integral of motion from this equation when the gradient in the $`z`$-direction vanishes:
$$\frac{\stackrel{~}{\varphi }}{t}+\delta r\frac{}{r}\left(\frac{g_{t\varphi }}{g_{\varphi \varphi }}\right)=0.$$
(99)
The torsional perturbation $`\delta \varphi `$ is slaved to the compressive perturbation $`\delta r`$ through conservation of angular momentum. The manipulation of the force-free equation (94) is a bit more complicated; some details are given in Appendix B. The final result is
$`\left(g^{tt}{\displaystyle \frac{^2\delta r}{t^2}}+g^{zz}{\displaystyle \frac{^2\delta r}{z^2}}\right)F_{r\varphi }^{(0)}`$ $`+`$ $`{\displaystyle \frac{g_{\varphi \varphi }}{\sqrt{g}}}{\displaystyle \frac{}{r}}\left[{\displaystyle \frac{\sqrt{g}}{g_{\varphi \varphi }g_{rr}}}{\displaystyle \frac{}{r}}(\delta rF_{r\varphi }^{(0)})\right]`$ (100)
$`=`$ $`{\displaystyle \frac{g^{tt}g_{\varphi \varphi }}{g_{rr}}}\left[{\displaystyle \frac{\stackrel{~}{\varphi }}{t}}+\delta r{\displaystyle \frac{}{r}}\left({\displaystyle \frac{g_{t\varphi }}{g_{\varphi \varphi }}}\right)\right]{\displaystyle \frac{}{r}}\left({\displaystyle \frac{g_{t\varphi }}{g_{\varphi \varphi }}}\right)F_{r\varphi }^{(0)}.`$
The wave operator on the left-hand side of Eqs. (98) and (100) is now mixed with an additional term involving the other coordinate perturbation. This source term can be treated as a perturbation when the spacetime is slowly rotating ($`g_{t\varphi }^2|g_{tt}g_{\varphi \varphi }|`$). It is also worth noting a simplification which takes place when the perturbations are uniform in $`z`$: then the right-hand side of Eq. (100) vanishes and, effectively, the fast mode equation involves only a single coordinate perturbation. That is, the time-evolution of the angular perturbation $`\delta \varphi `$ is fixed by conservation of angular momentum, and does not react back on the equation of motion for $`\delta r`$. More generally this is not the case, and the dynamics involves two independent fields.
### IV.2 Scattering of a cylindrical Alfvén wave in a spherical gravitational field
A uniform magnetofluid is perturbed by a spherical gravitational field (Section II.4). We now consider the perturbation to a torsional Alfvén wave propagating along this background magnetic field. We focus on a weak, spherical gravitational field, and allow for the possibility that the gravitating mass is extended in radius. Then the metric departs from the Schwarzschild metric; we write it as
$$g_{tt}=1+\mathrm{\Delta }_{tt};g_{rr}=1+\mathrm{\Delta }_{rr}.$$
(101)
(In this section, as in Section II.4, $`r`$ is the spherical radius.) In the weak field limit ($`|\mathrm{\Delta }_{tt}|`$, $`|\mathrm{\Delta }_{rr}|1`$) the radial coordinate of the magnetofluid is perturbed by an amount $`\delta r/r=O(\mathrm{\Delta })`$.
The static perturbation $`\delta r`$ is a solution to the linearized version of Eq. (27),
$$\frac{d^2(r\delta r)}{dr^2}2\frac{(r\delta r)}{r^2}=\frac{1}{2}(\mathrm{\Delta }_{tt}+\mathrm{\Delta }_{rr})\frac{1}{2}\frac{d}{dr}\left[r\left(\mathrm{\Delta }_{tt}\mathrm{\Delta }_{rr}\right)\right].$$
(102)
Notice that the right-hand side of this equation vanishes in the Schwarzschild metric, $`\mathrm{\Delta }_{rr}=\mathrm{\Delta }_{tt}=2GM/r`$. When the enclosed gravitating mass $`M`$ is itself a function of radius, Eq. (102) becomes
$$\frac{d^2(r\delta r)}{dr^2}2\frac{(r\delta r)}{r^2}=2G\frac{dM}{dr}.$$
(103)
We expand the background magnetic field $`F^{(0)}`$, the Alfvén wave $`\delta F^{(1)}`$, and its current $`\delta J^{(1)}`$ in powers of $`\mathrm{\Delta }`$,
$$F^{(0)}=F_{0\mathrm{\Delta }}^{(0)}+F_{1\mathrm{\Delta }}^{(0)}$$
(104)
and
$$\delta F^{(1)}=\delta F_{0\mathrm{\Delta }}^{(1)}+\delta F_{1\mathrm{\Delta }}^{(1)};\delta J^{(1)}=\delta J_{0\mathrm{\Delta }}^{(1)}+\delta J_{1\mathrm{\Delta }}^{(1)},$$
(105)
so that $`F_{1\mathrm{\Delta }}^{(0)}`$, $`\delta F_{1\mathrm{\Delta }}^{(1)}`$ and $`\delta J_{1\mathrm{\Delta }}^{(1)}`$ are first order in $`\delta r`$. For example, expanding $`=r_0^2r^2+2r\delta r`$ in Eq. (26) for the background field, we have
$$\left[F_{0\mathrm{\Delta }}^{(0)}\right]_{r\varphi }=B_0r\mathrm{sin}^2\theta ;\left[F_{0\mathrm{\Delta }}^{(0)}\right]_{\theta \varphi }=B_0r^2\mathrm{sin}\theta \mathrm{cos}\theta ,$$
(106)
and
$$\left[F_{1\mathrm{\Delta }}^{(0)}\right]_{r\varphi }=B_0\frac{d(r\delta r)}{dr}\mathrm{sin}^2\theta ;\left[F_{1\mathrm{\Delta }}^{(0)}\right]_{\theta \varphi }=2B_0r\delta r\mathrm{sin}\theta \mathrm{cos}\theta .$$
(107)
In the absence of gravity, the torsional wave propagates along the magnetic field. The associated charge and current densities
$`\delta \rho `$ $`=`$ $`\delta J_0(zt)\mathrm{exp}\left[{\displaystyle \frac{1}{2}}\left({\displaystyle \frac{r\mathrm{sin}\theta }{R_0}}\right)^2\right];`$
$`\delta 𝐉`$ $`=`$ $`\delta J_0(zt)\mathrm{exp}\left[{\displaystyle \frac{1}{2}}\left({\displaystyle \frac{r\mathrm{sin}\theta }{R_0}}\right)^2\right]\widehat{z}`$ (108)
are localized within a cylindrical radius $`r\mathrm{sin}\theta R_0`$.
We will focus on the case where the wave is strongly sheared, i.e., where its frequency is small compared with $`R_0^1`$:
$$R_0\left|\frac{\delta J_0^{}(zt)}{\delta J_0(zt)}\right|1.$$
(109)
To lowest order in the gravitational potential, the components of the fluctuating current and field are
$$\left[\delta J_{0\mathrm{\Delta }}^{(1)}\right]^t=\delta J;\left[\delta J_{0\mathrm{\Delta }}^{(1)}\right]^r=\mathrm{cos}\theta \delta J;\left[\delta J_{0\mathrm{\Delta }}^{(1)}\right]^\theta =\frac{1}{r}\mathrm{sin}\theta \delta J;$$
(110)
and
$`\left[\delta F_{0\mathrm{\Delta }}^{(1)}\right]_{\theta r}`$ $`=`$ $`{\displaystyle \frac{4\pi \delta J_0R_0^2}{\mathrm{sin}\theta }}\left\{1\mathrm{exp}\left[{\displaystyle \frac{1}{2}}\left({\displaystyle \frac{r\mathrm{sin}\theta }{R_0}}\right)^2\right]\right\};`$
$`\left[\delta F_{0\mathrm{\Delta }}^{(1)}\right]_{tr}`$ $`=`$ $`{\displaystyle \frac{\mathrm{sin}\theta }{r}}\left[\delta F_{0\mathrm{\Delta }}^{(1)}\right]_{\theta r};\left[\delta F_{0\mathrm{\Delta }}^{(1)}\right]_{t\theta }=\mathrm{cos}\theta \left[\delta F_{0\mathrm{\Delta }}^{(1)}\right]_{\theta r}.`$ (111)
The static gravitational perturbation to the background magnetic field is known implicitly through the solution to Eq. (102). We next solve for the perturbations $`\delta F_{1\mathrm{\Delta }}^{(1)}`$ and $`\delta J_{1\mathrm{\Delta }}^{(1)}`$. The current perturbation is determined from the linear component of the force-free equation. Making use of $`\left[\delta J_{0\mathrm{\Delta }}^{(1)}\right]^\mu \left[F_{0\mathrm{\Delta }}^{(0)}\right]_{\mu \nu }=0`$ and neglecting the terms that are second-order in $`\mathrm{\Delta }`$, we have
$$\left[\delta J_{0\mathrm{\Delta }}^{(1)}\right]^\mu \left[F_{1\mathrm{\Delta }}^{(0)}\right]_{\mu \nu }+\left[\delta J_{1\mathrm{\Delta }}^{(1)}\right]^\mu \left[F_{0\mathrm{\Delta }}^{(0)}\right]_{\mu \nu }=0.$$
(112)
The force-free equation (112) may be re-expressed by substituting Eqs. (106), (107) and (110),
$$\left[\delta J_{1\mathrm{\Delta }}^{(1)}\right]^r\mathrm{sin}\theta +\left[\delta J_{1\mathrm{\Delta }}^{(1)}\right]^\theta r\mathrm{cos}\theta =\delta J(r\mathrm{sin}\theta ,z,t)\left[r\frac{d(\delta r/r)}{dr}\right]\mathrm{sin}\theta \mathrm{cos}\theta .$$
(113)
In the low frequency regime, the equation of current conservation simplifies to
$$\frac{1}{r^2}\frac{}{r}\left\{r^2\left[\delta J_{1\mathrm{\Delta }}^{(1)}\right]^r\right\}+\frac{1}{\mathrm{sin}\theta }\frac{}{\theta }\left\{\mathrm{sin}\theta \left[\delta J_{1\mathrm{\Delta }}^{(1)}\right]^\theta \right\}=0.$$
(114)
For the particular choice (IV.2) of incident current density, the solution to the combined equations (113) and (114) is
$$\left[\delta J_{1\mathrm{\Delta }}^{(1)}\right]^r=\left[2\frac{\delta r}{r}\mathrm{cos}\theta \frac{r\delta r}{R_0^2}\mathrm{cos}\theta \mathrm{sin}^2\theta \right]\delta J$$
(115)
and
$$\left[\delta J_{1\mathrm{\Delta }}^{(1)}\right]^\theta =\left[\frac{1}{r^2}\frac{d(r\delta r)}{dr}\mathrm{sin}\theta +\frac{\delta r}{R_0^2}\mathrm{sin}^3\theta \right]\delta J.$$
(116)
Notice that the current $`\delta J_{1\mathrm{\Delta }}^{(1)}`$ receives a contribution from the metric perturbation as well as from $`\delta F_{1\mathrm{\Delta }}^{(1)}`$, e.g.
$$4\pi \left[\delta J_{1\mathrm{\Delta }}^{(1)}\right]^r=\frac{1}{r^2\mathrm{sin}\theta }\frac{}{\theta }\left\{\mathrm{sin}\theta (g^{rr}1)\left[\delta F_{0\mathrm{\Delta }}^{(1)}\right]_{r\theta }\right\}+\frac{1}{r^2\mathrm{sin}\theta }\frac{}{\theta }\left\{\mathrm{sin}\theta \left[\delta F_{1\mathrm{\Delta }}^{(1)}\right]_{r\theta }\right\}.$$
(117)
Here, the metric perturbation is $`g^{rr}1=2GM/r`$ to lowest order. Thus the current components $`[\delta J_{1\mathrm{\Delta }}^{(1)}]^{r,\theta }`$ are sourced by the fluctuating field
$$\left[\delta F_{1\mathrm{\Delta }}^{(1)}\right]_{\theta r}=\left[4\pi (r\delta r)\mathrm{sin}\theta \right]\delta J(r\mathrm{sin}\theta ,z,t)+\frac{2GM}{r}\left[\delta F_{0\mathrm{\Delta }}^{(1)}\right]_{\theta r}.$$
(118)
The gravitational perturbation of the charge density, and the electric field which it sources, can be obtained by combining the Bianchi identity
$$_\theta \left[\delta F_{1\mathrm{\Delta }}^{(1)}\right]_{tr}_r\left[\delta F_{1\mathrm{\Delta }}^{(1)}\right]_{t\theta }=_t\left[\delta F_{1\mathrm{\Delta }}^{(1)}\right]_{\theta r}0,$$
(119)
with the MHD condition
$`\left\{\left[\delta F_{0\mathrm{\Delta }}^{(1)}\right]_{t\theta }+\left[\delta F_{1\mathrm{\Delta }}^{(1)}\right]_{t\theta }\right\}\left\{\left[F_{0\mathrm{\Delta }}^{(0)}\right]_{\varphi r}+\left[F_{1\mathrm{\Delta }}^{(0)}\right]_{\varphi r}\right\}`$ (120)
$``$ $`\left\{\left[\delta F_{0\mathrm{\Delta }}^{(1)}\right]_{tr}+\left[\delta F_{1\mathrm{\Delta }}^{(1)}\right]_{tr}\right\}\left\{\left[F_{0\mathrm{\Delta }}^{(0)}\right]_{\varphi \theta }+\left[F_{1\mathrm{\Delta }}^{(0)}\right]_{\varphi \theta }\right\}=0.`$
At leading order in the gravitational potential, this last equation becomes
$$\left[\delta F_{1\mathrm{\Delta }}^{(1)}\right]_{t\theta }=r\mathrm{cot}\theta \left[\delta F_{1\mathrm{\Delta }}^{(1)}\right]_{tr}\mathrm{cos}\theta \frac{d(r^3\delta r)}{d\mathrm{ln}r}\left[\delta F_{0\mathrm{\Delta }}^{(1)}\right]_{\theta r}.$$
(121)
Substituting this expression into Eq. (119), one can solve for the perturbation to the electric field,
$$\left[\delta F_{1\mathrm{\Delta }}^{(1)}\right]_{tr}=4\pi \delta J_0\left[R_0^2\frac{d(\delta r/r)}{dr}\left(1e^{R^2/2R_0^2}\right)+\mathrm{sin}^2\theta \delta re^{R^2/2R_0^2}\right],$$
(122)
and
$$\left[\delta F_{1\mathrm{\Delta }}^{(1)}\right]_{t\theta }=[4\pi (r\delta r)\mathrm{sin}\theta \mathrm{cos}\theta ]\delta J_0e^{R^2/2R_0^2}.$$
(123)
Here $`R=r\mathrm{sin}\theta `$. The charge density perturbation is then
$`4\pi \left[\delta J_{1\mathrm{\Delta }}^{(1)}\right]^t={\displaystyle \frac{1}{r^2}}{\displaystyle \frac{}{r}}\left\{r^2\left[\delta F_{1\mathrm{\Delta }}^{(1)}\right]_{tr}\right\}`$ $``$ $`{\displaystyle \frac{1}{r^2\mathrm{sin}\theta }}{\displaystyle \frac{}{\theta }}\left\{\mathrm{sin}\theta \left[\delta F_{1\mathrm{\Delta }}^{(1)}\right]_{t\theta }\right\}`$ (124)
$`+`$ $`{\displaystyle \frac{1}{r^2\mathrm{sin}\theta }}{\displaystyle \frac{}{\theta }}\left\{\mathrm{sin}\theta \left(g^{tt}+1\right)\left[\delta F_{0\mathrm{\Delta }}^{(1)}\right]_{t\theta }\right\}.`$
Here $`g^{tt}+1=2GM/r`$ to lowest order.
Our principal goal here is to work out the second-order current $`[\delta J_{1\mathrm{\Delta }}^{(2)}]^\varphi `$. This may be obtained from
$$\left[\delta J_{0\mathrm{\Delta }}^{(1)}\right]^\mu \left[\delta F_{1\mathrm{\Delta }}^{(1)}\right]_{\mu \nu }+\left[\delta J_{1\mathrm{\Delta }}^{(1)}\right]^\mu \left[\delta F_{0\mathrm{\Delta }}^{(1)}\right]_{\mu \nu }=\left[\delta J_{1\mathrm{\Delta }}^{(2)}\right]^\varphi \left[F_{0\mathrm{\Delta }}^{(0)}\right]_{\varphi \nu },$$
(125)
by substituting Eqs. (110), (111), (116), (118), (122) and (124). The resulting expression is fairly complicated, but it simplifies dramatically when the gravitating mass is concentrated at a radius much smaller than the transverse scale $`R_0`$ of the incident Alfvén wave. In this regime, one has $`\delta r/rG(dM/dr)GM/r`$, and the dominant field and current perturbations are
$`\left[\delta F_{1\mathrm{\Delta }}^{(1)}\right]_{\theta r}`$ $``$ $`{\displaystyle \frac{2GM}{r}}\left[\delta F_{0\mathrm{\Delta }}^{(1)}\right]_{\theta r}`$ (126)
$`=`$ $`\left({\displaystyle \frac{2GM}{r}}\right){\displaystyle \frac{4\pi R_0^2}{\mathrm{sin}\theta }}\left[1e^{R^2/2R_0^2}\right]\delta J_0(r\mathrm{cos}\theta t),`$
and
$$\left[\delta J_{1\mathrm{\Delta }}^{(1)}\right]^t\frac{2GM}{r}\left[\mathrm{cos}^2\theta e^{R^2/2R_0^2}\frac{R_0^2}{r^2}\left(1e^{R^2/2R_0^2}\right)\right]\delta J_0(r\mathrm{cos}\theta t).$$
(127)
Combining the terms on the left-hand side of Eq. (125), and substituting the background field (19), we find
$$\left[\delta J_{1\mathrm{\Delta }}^{(2)}\right]^\varphi =\frac{8\pi (\delta J_0)^2}{B_0\mathrm{sin}^2\theta }\left(\frac{GM}{r}\right)\left(\frac{R_0}{r}\right)^4\left(1e^{R^2/2R_0^2}\right)^2\left[1+\frac{R^2/R_0^2}{e^{R^2/2R_0^2}1}\right].$$
(128)
This expression simplifies further at small cylindrical radius ($`RR_0`$),
$$\left[\delta J_{1\mathrm{\Delta }}^{(2)}\right]^\varphi 6\pi \frac{|\delta J_0|^2}{B_0}\left(\frac{GM}{r}\right)\mathrm{sin}^2\theta e^{2i\omega (zt)},$$
(129)
when the incident Alfvén wave is a pure harmonic. The second-order current is the source for an outgoing compressive wave
$$\frac{^2(r\delta r_I)}{t^2}\frac{1}{r^2\mathrm{sin}\theta }\frac{}{\theta }\left(\mathrm{sin}\theta \frac{(r\delta r_I)}{\theta }\right)\frac{^2(r\delta r_I)}{r^2}=\frac{4\pi r^2}{B_0}\left[\delta J_{1\mathrm{\Delta }}^{(2)}\right]^\varphi ,$$
(130)
of amplitude $`\delta r_I`$. As in the case of a torsional wave propagating in a cylindrical spacetime, the compressive mode is second harmonic and second-order in the wave amplitude.
The Greens function solution to Eq. (130) is dominated by the fluid at radius $`rR_0`$. As a result, the shape of the asymptotic fast wave is sensitive to the distribution of current within the incoming Alfvén mode. We will not examine the pattern of the outgoing compressive wave in full detail here.
## V Collisions of Axisymmetric Waves in a Cylindrical Spacetime
In this section, we study the collision of two torsional Alfvén waves in a uniform magnetofluid in a static cylindrical spacetime ($`g_{t\varphi }=0`$). The effects of spacetime curvature are subtle enough that we specialize to axisymmetric modes, supported only by a perturbation $`\delta \varphi `$. The result easily generalizes to the case of flat space, and in the next section we examine more general types of wave interactions in a Minkowski background.
The two Alfvén waves, propagating oppositely along the magnetic field, are labeled $`+`$ and $``$, and the component of the current generated by their interaction is labelled $`I`$. The colliding waves are each assumed to be supported within a cylindrical shell, between radii $`R_{\mathrm{min}}`$ and $`R_{\mathrm{max}}`$. We start from the force-free equation
$$(J_+^t+J_{}^t)(\delta F_{tr}^++\delta F_{tr}^{})+(J_+^z+J_{}^z)(\delta F_{zr}^++\delta F_{zr}^{})+J_I^\varphi F_{\varphi r}^{(0)}=0.$$
(131)
The background magnetic field is defined by $`J^{\varphi (0)}=0`$ \[Eq. (32)\]. The fluctuating fields are
$`\delta F_{zr}^{}{}_{}{}^{\pm }`$ $`=`$ $`{\displaystyle \frac{\varphi _\pm }{z}}F_{\varphi r}^{(0)};`$
$`\delta F_{tr}^{}{}_{}{}^{\pm }`$ $`=`$ $`{\displaystyle \frac{\varphi _\pm }{t}}F_{\varphi r}^{(0)}=v_z\delta F_{zr}^{}{}_{}{}^{\pm },`$ (132)
where $`v_z`$ is given by Eq. (80). The associated currents are
$$J_\pm ^t=\frac{1}{4\pi \sqrt{g}}\frac{}{r}\left(\sqrt{g}g^{tt}g^{rr}\delta F_{tr}^\pm \right)=\pm \frac{1}{4\pi \sqrt{g}}\frac{}{r}\left(\sqrt{g}\frac{g^{zz}g^{rr}}{v_z}\delta F_{zr}^\pm \right)$$
(133)
and
$$J_\pm ^z=\frac{1}{4\pi \sqrt{g}}\frac{}{r}\left(\sqrt{g}g^{zz}g^{rr}\delta F_{zr}^\pm \right).$$
(134)
Combining these expressions, the force-free equation becomes
$`{\displaystyle \frac{4\pi }{B_0}}\sqrt{g}J_I^\varphi `$ $`=`$ $`{\displaystyle \frac{g_{rr}}{g_{tt}}}{\displaystyle \frac{^2\delta r}{t^2}}+{\displaystyle \frac{g_{rr}}{g_{zz}}}{\displaystyle \frac{^2\delta r}{z^2}}+{\displaystyle \frac{}{r}}\left[{\displaystyle \frac{\sqrt{g}}{g_{\varphi \varphi }g_{rr}}}{\displaystyle \frac{}{r}}\left(\delta r{\displaystyle \frac{g_{rr}g_{\varphi \varphi }}{\sqrt{g}}}\right)\right]`$ (135)
$`=`$ $`2\left({\displaystyle \frac{g_{zz}}{g_{\varphi \varphi }}}\right){\displaystyle \frac{}{r}}\left[\left({\displaystyle \frac{g_{\varphi \varphi }}{g_{zz}}}\right)^2{\displaystyle \frac{\varphi _+}{z}}{\displaystyle \frac{\varphi _{}}{z}}\right]+`$
$`\left({\displaystyle \frac{g_{\varphi \varphi }}{g_{zz}}}\right)\left[2{\displaystyle \frac{\varphi _+}{z}}{\displaystyle \frac{\varphi _{}}{z}}\left({\displaystyle \frac{\varphi _+}{z}}\right)^2\left({\displaystyle \frac{\varphi _{}}{z}}\right)^2\right]_r(\mathrm{ln}v_z).`$
In contrast with the case of a single (unidirectional) Alfvén wave, a source term appears even when the wave speed has a vanishing gradient ($`_rv_z=0`$). For that reason we must, in general, retain the curvature corrections in the wave operator for $`\delta r`$.
### V.1 Collision in a nearly flat, cylindrically symmetric spacetime
We now specialize to a nearly flat cylindrical spacetime with Newtonian potential $`\lambda `$ and metric (34), and focus on the source terms in Eq. (135) that are proportional to $`(_z\varphi _+)(_z\varphi _{})`$. The torsional wave speed is $`v_z=1+2\lambda `$. Comparing with flat space, the new effects we are interested in are already present in the case of a very long wavelength twist in the magnetic field,
$$\left|\frac{\varphi _\pm }{z}\right|R_{\mathrm{max}}1,$$
(136)
so we neglect the derivatives in $`t`$ and $`z`$. The perturbation to $`B^z`$ is
$$\frac{\delta B^z}{B_0}=\frac{\delta F_{\varphi r}^I}{F_{\varphi r}^{(0)}}=\frac{1}{r}\frac{(r\delta r)}{r}2\frac{d\lambda }{dr}\delta r,$$
(137)
and the equation of motion (135) simplifies to
$$\frac{}{r}\left(\frac{\delta B^z}{B_0}\right)=\frac{2}{r^2}\frac{}{r}\left(r^4\frac{\varphi _+}{z}\frac{\varphi _{}}{z}\right)+4r^2\frac{d\lambda }{dr}\frac{\varphi _+}{z}\frac{\varphi _{}}{z}.$$
(138)
Our first goal is to calculate the amplitude $`\delta R_{\mathrm{max}}(t)\delta r(R_{\mathrm{max}},t)`$ of the hydrodynamic fluctuation at the outer boundary of the annulus supporting the Alfvén waves. The amplitude of the fast mode at large radius can then easily be obtained, to leading order in $`\lambda `$, from the equation of motion (III.1) in Minkowski space. Taking the two waveforms to be <sup>1</sup><sup>1</sup>1Generally the wavenumbers $`k_+`$ and $`k_{}`$ are not equal; but the translational invariance of the background magnetofluid in the $`z`$-direction allows them to be made equal through an appropriate Lorentz boost.
$$\delta \varphi _\pm (t,r,z)=\delta \varphi _{0\pm }\mathrm{sin}\left[k(zt)\right]f_\pm (r),$$
(139)
one sees that the interaction term
$$\frac{\varphi _+}{z}\frac{\varphi _{}}{z}=\frac{1}{2}k^2\delta \varphi _{0+}\delta \varphi _0\left[\mathrm{cos}(2kt)+\mathrm{cos}(2kz)\right]f_+(r)f_{}(r)$$
(140)
is the sum of a uniform oscillation of the cylinder, and a static deformation that excites no wave motion outside the cylinder. Thus the calculation of the fast mode amplitude reduces to the problem of calculating the amplitude $`\delta R_{\mathrm{max}}`$ of the oscillation at the outer boundary of the cylinder.
We start by integrating Eq. (137) outward from $`\delta R=0`$ at $`r=0`$ to obtain
$$R_{\mathrm{min}}\delta R_{\mathrm{min}}=\left(\frac{1}{2}R_{\mathrm{min}}^2+_0^{R_{\mathrm{min}}}\frac{d\lambda }{dr}r^2𝑑r\right)\frac{\delta B^z}{B_0}|_{R_{\mathrm{min}}},$$
(141)
to first order in $`\lambda `$. The expansion of the magnetofluid is nearly incompressible in the near zone outside the cylinder ($`kr1`$), so the relevant inner and outer boundary conditions are
$$\frac{\delta B^z}{B_0}(kR_{\mathrm{max}})^2\frac{\delta R_{\mathrm{max}}}{R_{\mathrm{max}}}0,$$
(142)
and
$$\frac{\varphi _\pm }{z}|_{R_{\mathrm{max}}}=\frac{\varphi _\pm }{z}|_{R_{\mathrm{min}}}=0.$$
(143)
The field perturbation is nearly uniform inside radius $`R_{\mathrm{min}}`$; we will only need the leading term
$$\frac{\delta B^z}{B_0}|_{R_{\mathrm{min}}}=0+2_{R_{\mathrm{min}}}^{R_{\mathrm{max}}}\frac{1}{r^2}\frac{}{r}\left(r^4\frac{\varphi _+}{z}\frac{\varphi _{}}{z}\right)𝑑r=4\mathrm{\Phi }__+(R_{\mathrm{max}}),$$
(144)
where
$$\mathrm{\Phi }__+(r)_{R_{\mathrm{min}}}^rr\frac{\varphi _+}{z}\frac{\varphi _{}}{z}𝑑r.$$
(145)
The radial perturbation within the annulus $`R_{\mathrm{min}}<r<R_{\mathrm{max}}`$ can be found iteratively from
$`r\delta r`$ $``$ $`R_{\mathrm{min}}\delta R_{\mathrm{min}}={\displaystyle _{R_{\mathrm{min}}}^r}r^{}\left[{\displaystyle \frac{\delta B^z}{B_0}}+2{\displaystyle \frac{d\lambda }{dr^{}}}\delta r(r^{})\right]𝑑r^{}`$
$`=`$ $`{\displaystyle \frac{1}{2}}R_{\mathrm{min}}^2{\displaystyle \frac{\delta B^z}{B_0}}|_{R_{\mathrm{min}}}+{\displaystyle \frac{1}{2}}r^2{\displaystyle \frac{\delta B^z}{B_0}}{\displaystyle _{R_{\mathrm{min}}}^r}{\displaystyle \frac{1}{2}}r^2{\displaystyle \frac{}{r^{}}}\left({\displaystyle \frac{\delta B^z}{B_0}}\right)𝑑r^{}+2{\displaystyle _{R_{\mathrm{min}}}^r}r^{}{\displaystyle \frac{d\lambda }{dr^{}}}\delta r(r^{})𝑑r^{},`$
after substituting Eq. (138) in the third term on the right-hand side. Evaluating this expression at $`r=R_{\mathrm{max}}`$ we find
$$R_{\mathrm{max}}\delta R_{\mathrm{max}}=2_{R_{\mathrm{min}}}^{R_{\mathrm{max}}}r^2\frac{d\lambda }{dr}\left(2\mathrm{\Phi }__+(r)+r\frac{d\mathrm{\Phi }__+}{dr}\right)𝑑r+4\mathrm{\Phi }__+(R_{\mathrm{max}})_0^{R_{\mathrm{max}}}r^2\frac{d\lambda }{dr}𝑑r.$$
(147)
The first remark to make about this expression is that $`\delta R_{\mathrm{max}}`$ vanishes in flat space ($`d\lambda /dr=`$ constant). The same turns out to be true in the special case of a cylindrical line mass ($`d\lambda /dr=K/r`$), but not for a more general cylindrical mass distribution that is extended in radius.
We now obtain the fast mode amplitude in the wave zone $`kr1`$. The cylindrical wave equation for $`\delta B^z`$ has the outgoing wave solution
$$\frac{\delta B^z}{B_0}=Ae^{ikt}H_{0}^{}{}_{}{}^{(1)}(kr)A\sqrt{\frac{2}{\pi kr}}e^{ik(rt)i\pi /4}(kr1)$$
(148)
in Minkowski space. The Hankel function scales as $`H_0^{(1)}(kr)(2i/\pi )\mathrm{ln}(kr)`$ at small radius. To obtain the normalization factor, we note that in the near zone ($`kr1`$) where the magnetofluid is nearly incompressible, the rate of transport of magnetic flux is nearly constant in radius,
$$\frac{\mathrm{\Phi }}{t}=2\pi r\frac{(\delta r)}{t}B^z\mathrm{constant}(kr1).$$
(149)
This implies
$$\frac{\delta r}{t}=\left[\frac{(\delta r)}{t}\right]_{R_{\mathrm{max}}}\left(\frac{r}{R_{\mathrm{max}}}\right)^1.$$
(150)
Substituting this expression into $`\stackrel{}{E}=_t(\delta r)B_0(\widehat{r}\times \widehat{z})`$, and thence into
$$\frac{\stackrel{}{E}}{t}=\stackrel{}{}\times \stackrel{}{B},$$
(151)
and integrating over radius, gives
$$\frac{\delta B^z}{B_0}\frac{\delta B^z(R_{\mathrm{max}})}{B_0}=\left[\frac{^2(\delta r)}{t^2}\right]_{R_{\mathrm{max}}}\mathrm{ln}\left(\frac{r}{R_{\mathrm{max}}}\right)R_{\mathrm{max}},$$
(152)
and
$$Ae^{ikt}=\frac{i\pi }{2}\left(\omega ^2R_{\mathrm{max}}\right)\delta r(R_{\mathrm{max}},t).$$
(153)
The general solution is
$$\frac{\delta B^z(r,t)}{B_0}=R_{\mathrm{max}}_{\mathrm{}}^{tr}\frac{^2(\delta r)}{t^2}\frac{dt^{}}{\sqrt{(tt^{})^2r^2}}.$$
(154)
### V.2 Collision in a static black string spacetime
As another illustrative example, we consider the collision between two torsional waves in the spacetime of a static black string \[Eq. (37) with $`J=0`$\]. The background magnetic field $`F_{r\varphi }^{(0)}=B_0r/(r^2/\mathrm{}^2M)`$ is aligned with the axis of the string. We assume, as before, that the colliding waves are supported only within a cylindrical annulus $`R_{\mathrm{min}}<r<R_{\mathrm{max}}`$, and take $`R_{\mathrm{min}}`$, $`R_{\mathrm{max}}\mathrm{}`$.
A radial disturbance of frequency $`\omega \mathrm{}^1`$ accumulates a phase
$$\varphi \omega \left(\frac{g_{rr}}{g_{tt}}\right)^{1/2}𝑑r=\omega \frac{dr}{r^2/\mathrm{}^2M}.$$
(155)
This integral converges at large radius but diverges near the horizon $`r_H=M^{1/2}\mathrm{}`$. When $`M(\omega \mathrm{})^2`$, there is a well defined wave zone which is localized at a small radius, $`r\stackrel{<}{}\omega \mathrm{}^2`$. The effect of a small finite angular momentum $`J`$ is to impose a reflecting barrier for the magnetosonic wave at a radius $`r(J^2/4\omega )^{1/3}`$, and therefore to form a resonant cavity. We will not consider the case of finite angular momentum here.
The medium outside radius $`R_{\mathrm{max}}`$ will, at the same time, respond to a low-frequency disturbance with a uniform compression or rarefaction, $`\delta F_{\varphi r}F_{\varphi r}^{(0)}`$. Thus $`\delta rr`$ at a large radius ($`r\mathrm{}`$). The amplitude of the radial perturbation $`\delta R_{\mathrm{max}}`$ can be obtained by noting that the magnetic flux
$$\mathrm{\Phi }=2\pi 𝑑rF_{r\varphi }=𝑑r\frac{2\pi rB_0}{r^2/\mathrm{}^2M}$$
(156)
diverges at large radius. At the same time, the background field energy
$$\frac{dE_B}{dz}=2\pi 𝑑r\sqrt{g_{rr}g_{\varphi \varphi }}\left[g^{tt}g^{rr}g^{\varphi \varphi }\frac{F_{r\varphi }^2}{8\pi }\right]=\frac{B_0^2}{4}\frac{rdr}{(r^2/\mathrm{}^2M)^{5/2}}$$
(157)
is finite. As a result, an oscillation $`\delta R_{\mathrm{max}}`$ of the outer boundary of the annulus is accompanied by a small field perturbation, $`\delta B_z(R_{\mathrm{max}})0`$.
The nature of the wave solutions at small radius is most easily demonstrated by taking the limit $`M0`$, $`Z(r)r^2/\mathrm{}^2`$. Then the wave equation (75), which applies to the case of a uniform radial oscillation $`\delta r(R_{\mathrm{max}},z,t)e^{i\omega t}`$, simplifies considerably. It transforms under the change of variable $`r_{}=\mathrm{}^2/r`$ to
$$\frac{^2}{t^2}\left(\frac{\delta B^z}{B^z}\right)+r_{}\frac{}{r_{}}\left[\frac{1}{r_{}}\frac{}{r_{}}\left(\frac{\delta B^z}{B^z}\right)\right]=0.$$
(158)
A further change of variable $`\delta B^z/B^z=h(r_{})(\omega r_{})e^{i\omega t}`$ yields the Bessel equation
$$r_{}^2\frac{^2h}{r_{}^2}+r_{}\frac{h}{r_{}}+(\omega ^2r_{}^21)h=0,$$
(159)
which has the wave solution
$$\frac{\delta B^z}{B^z}=(\omega r_{})H_1^{(2)}(\omega r_{})Ae^{i\omega t}A(\omega r_{})\sqrt{\frac{2}{\pi (\omega r_{})}}e^{i\omega (r_{}+t)3\pi /4}$$
(160)
propagating to small $`r`$ (large negative $`r_{}`$). It will be noted that the amplitude of the wave diverges in the wave zone, because the radial phase velocity
$$\frac{\omega }{k}=\sqrt{\frac{g_{tt}}{g_{rr}}}\left(\frac{\mathrm{}}{r_{}}\right)^2$$
(161)
asymptotes to zero.
To relate the coefficient $`A`$ to the amplitude of the oscillation at radius $`R_{\mathrm{min}}`$, we note that the rate of transport of magnetic flux
$$2\pi F_{r\varphi }\frac{(\delta r)}{t}=\frac{2\pi B_0\mathrm{}^2}{r}\frac{(\delta r)}{t}$$
(162)
is constant in the near zone ($`|\omega r_{}|1`$). Hence $`_r(\delta r)\delta r/r`$. From Eq. (73) we have,
$$\frac{^2\delta r_{}}{t^2}=\frac{}{r_{}}\left(\frac{\delta B^z}{B^z}\right).$$
(163)
Combining this with the low frequency expansion $`(\omega r_{})H_1^{(2)}(\omega r_{})2i/\pi (i/\pi )(\omega r_{})^2\mathrm{ln}(\omega r_{}/2)`$ gives
$$Ae^{i\omega t}\frac{i\pi /2}{\mathrm{ln}(\omega R_{\mathrm{min}}^{}/2)}\frac{\delta R_{\mathrm{min}}^{}(t)}{R_{\mathrm{min}}^{}}.$$
(164)
The last step in obtaining the amplitude of the fast mode that emerges from the wave collision, is to express $`\delta R_{\mathrm{min}}`$ in terms of the amplitudes of the colliding Alfvén modes. The terms involving $`(_z\varphi _+)(_z\varphi _{})`$ in Eq. (135) can be combined to give
$$\frac{}{r}\left(\frac{\delta B^z}{B^z}\right)=\frac{}{r}\left[r\frac{}{r}\left(\frac{\delta r}{r}\right)\right]=\frac{2}{r}\frac{}{r}\left(r^3\frac{\varphi _+}{z}\frac{\varphi _{}}{z}\right).$$
(165)
Integrating with respect to radius, this becomes
$`{\displaystyle \frac{\delta R_{\mathrm{min}}}{R_{\mathrm{min}}}}{\displaystyle \frac{\delta R_{\mathrm{max}}}{R_{\mathrm{max}}}}`$ $`=`$ $`{\displaystyle _{R_{\mathrm{min}}}^{R_{\mathrm{max}}}}𝑑r\left\{{\displaystyle \frac{}{r}}\left[\mathrm{ln}\left({\displaystyle \frac{r}{R_{\mathrm{min}}}}\right){\displaystyle \frac{\delta B^z}{B^z}}\right]\mathrm{ln}\left({\displaystyle \frac{r}{R_{\mathrm{min}}}}\right){\displaystyle \frac{}{r}}\left({\displaystyle \frac{\delta B^z}{B^z}}\right)\right\}`$ (166)
$`=`$ $`2{\displaystyle _{R_{\mathrm{min}}}^{R_{\mathrm{max}}}}𝑑rr\left[1\mathrm{ln}\left({\displaystyle \frac{r}{R_{\mathrm{min}}}}\right)\right]{\displaystyle \frac{\varphi _+}{z}}{\displaystyle \frac{\varphi _{}}{z}},`$
since $`_z\varphi _\pm (R_{\mathrm{min}})=_z\varphi _\pm (R_{\mathrm{max}})=0`$ and $`\delta B_z(R_{\mathrm{max}})=0`$. The field perturbation at the inner boundary is obtained by substituting $`R_{\mathrm{min}}R_{\mathrm{max}}`$ in the logarithms in the first line of Eq. (166). After subtracting the two equations, one obtains
$$\frac{\delta B^z}{B^z}|_{R_{\mathrm{min}}}=2_{R_{\mathrm{min}}}^{R_{\mathrm{max}}}r\frac{\varphi _+}{z}\frac{\varphi _{}}{z}𝑑r.$$
$`(166b)`$
## VI Collisions of Non-Axisymmetric Waves in Minkowski Space
We will now broaden our discussion of mode collisions to non-axisymmetric modes, and consider both fast and Alfvén modes. Three-mode interactions in the force-free limit were previously considered in Thompson and Blaes (1998), using the magnetic Lagrangian formalism developed there. We return to this subject and clarify the properties of i) the three-mode coupling between two Alfvén waves; ii) the interaction between two fast waves; and iii) the interaction between a fast wave and an Alfvén wave.
Basic constraints on the mode interactions arise from conservation of energy and momentum. Because the relativistic magnetofluid is inherently compressible, there is guaranteed to be a three-mode coupling between two Alfvén waves and a fast wave Thompson and Blaes (1998). The Alfvén waves ($`A1`$, $`A2`$) satisfy the dispersion relation $`\omega _A=\pm k_A^z`$; and the fast mode $`\omega _F=|𝐤_F|`$. Conservation of energy
$$\omega _{A1}+\omega _{A2}=\omega _F$$
(167)
and longitudinal momentum
$`k_{A1}^z+k_{A2}^z`$ $`=`$ $`\omega _{A1}\omega _{A2}`$ (168)
$`=`$ $`k_F^z=\omega _F\mathrm{cos}\theta _F`$
($`\mathrm{cos}\theta _F=𝐤_F𝐁_0/k_FB_0`$) can be combined to give
$$\mathrm{cos}\theta _F=\frac{\omega _{A1}\omega _{A2}}{\omega _{A1}+\omega _{A2}}.$$
(169)
Thus in a frame in which the two colliding Alfvén modes have equal frequencies, the fast mode is emitted perpendicular to the background field and with a frequency $`\omega _F=2\omega _A`$ – just as we found in Section V.1.
There is also a three-mode interaction between the two colliding Alfvén waves and a third Alfvén wave. From a kinematic viewpoint, this interaction is non-vanishing only if at least one of the colliding modes has a zero-frequency component (Montgomery and Matthaeus (1995), Ng and Bhattacharjee (1996)), as is easily checked by setting $`\mathrm{cos}\theta _F=1`$ in Eqs. (167) and (168). An alternative description is that the magnetic field lines experience a net displacement (or braiding) across at least one of the colliding wavepackets Goldreich and Sridhar (1997), Thompson and Blaes (1998). (A further requirement for a non-vanishing three-mode interaction, as we detail shortly, is that the colliding modes are non-axisymmetric.)
Next consider the interaction between two fast waves. It is here that a significant difference arises between these two viewpoints. If the background magnetic field suffers a net displacement across the fast wavefront, then a zero-frequency component is present in the fourier decomposition of the field. From a kinematic viewpoint, this zero-frequency component would facilitate a three-wave interaction between the two colliding fast modes and a third fast mode. Two colliding fast waves of finite frequency can generate a third fast wave only if the waves are colinear: only in that case is it possible to satisfy conservation of energy
$$\omega _{F1}+\omega _{F2}=\omega _{F3}$$
(170)
and momentum
$$𝐤_{F1}+𝐤_{F2}=𝐤_{F3}.$$
(171)
(Two photons propagating obliquely in vacuum cannot merge to form a single photon, in part because these kinematic conditions cannot be satisfied.) From the same kinematic viewpoint a three-wave interaction between two colliding fast waves and an Alfvén mode is possible only in the degenerate case where both colliding waves propagate along the background magnetic field.<sup>2</sup><sup>2</sup>2We thank Maxim Lyutikov for an illuminating discussion of this point. This restriction would appear to be somewhat relaxed if one of the waves has a zero-frequency component: then the direction of propagation of that wave can be arbitrary.
As our second task in this section, we calculate the form of this three-mode interaction for two colliding planar fast waves. We find that, in fact, no fast wave is emitted during the collision, but an Alfvén wave is. The Alfvén wave is emitted for any direction of propagation of the two colliding waves. Moreover, when there is a field line displacement in one fast wave, and the other is a pure sinusoid, then the outgoing Alfvén wave is also a pure sinusoid but its frequency is not obtainable from a three-wave resonance condition. It is only when the two colliding modes are directed along the background magnetic field that the resonance condition can be satisfied with a zero-frequency component.
The fast mode also undergoes a four-mode interaction analogous to photon scattering, with the conservation relations
$$\omega _{F1}+\omega _{F2}=\omega _{F3}+\omega _{F4}$$
(172)
and
$$𝐤_{F1}+𝐤_{F2}=𝐤_{F3}+𝐤_{F4}.$$
(173)
However, the fast mode also has a non-linear interaction with an Alfvén mode. We show that if the Alfvén wave spectrum is that of a critically balanced cascade (Goldreich and Sridhar (1995)), then this second interaction dominates the self-interaction of the fast mode (as was claimed without detailed justification in Thompson and Blaes (1998)).
### VI.1 Collision between two Alfvén waves
A collision between two axisymmetric torsional modes generates a compressive motion transverse to the axis of the background magnetic field. The amplitude of this compressive motion is, however, strongly suppressed when the colliding wavepackets are highly elongated ($`k_zR_{\mathrm{max}}1`$). In the case of two monochromatic torsional waves with frequencies $`\omega =k_z`$, this radial disturbance vanishes at the outer boundary of the zone supporting the colliding Alfvén waves. More generally, $`\delta R_{\mathrm{max}}`$ is suppressed by a factor $`(k_zR_{\mathrm{max}})^2`$, compared with the result of simple dimensional analysis,
$$\frac{\delta R_{\mathrm{max}}}{R_{\mathrm{max}}}(k_zR_{\mathrm{max}})^2\times R_{\mathrm{max}}^2\frac{\varphi _+}{z}\frac{\varphi _{}}{z}.$$
(174)
In other words, the three-mode interaction between two Alfvén waves and the fast mode is suppressed for elongated wavepackets (such as are created during a weak turbulent cascade Goldreich and Sridhar (1995)).
There is, nonetheless, a much stronger three-mode interaction involving a third Alfvén wave. This interaction is not apparent if one restricts the two colliding waves to be axially symmetric, and so we will broaden the analysis here to include Alfvén modes supported by coordinate perturbations $`\delta \varphi `$ and $`\delta r`$. The relevant component of the force-free equation is
$$J^{t(1)}F_{t\varphi }^{(1)}+J^{z(1)}F_{z\varphi }^{(1)}+J^{r(2)}F_{r\varphi }^{(0)}=0.$$
(175)
In Minkowski space, $`\delta F_{t\varphi }^{(1)}=(B_0r)_t\delta r`$ and $`\delta F_{z\varphi }^{(1)}=(B_0r)_z\delta r`$. The second-order current vanishes for the two separate modes $`+`$ and $``$, but where they overlap the first two terms in Eq. (175) do not cancel:
$$J^{t(1)}F_{t\varphi }^{(1)}+J^{z(1)}F_{z\varphi }^{(1)}=\frac{B_0^2}{2\pi }\left[_\varphi \left(_z\delta r_+_z\delta r_{}\right)_z\delta r_+_r\left(r^2_z\delta \varphi _{}\right)_z\delta r_{}_r\left(r^2_z\delta \varphi _+\right)\right].$$
(176)
When at least one of the colliding Alfvén wavepackets is not axisymmetric (with both perturbations $`\delta \varphi `$ and $`\delta r`$ excited), one also finds an explicit second order contribution to the fields,
$$\delta F_{rt}^{(2)}=\left[\frac{1}{r}_r\left(r\delta r_+\right)_t\delta \varphi _{}+\frac{1}{r}_r\left(r\delta r_{}\right)_t\delta \varphi _+_t\delta r_+_r\delta \varphi _{}_t\delta r_{}_r\delta \varphi _+\right]B_0r;$$
(177)
and similarly for $`\delta F_{r\varphi }^{(2)}`$ and $`\delta F_{rz}^{(2)}`$. The second-order current therefore has an explicit contribution from the colliding modes. As before, an additional interaction component $`\delta r_I`$, $`\delta \varphi _I`$ to the coordinate fields is required to solve the force-free equation:
$`4\pi J^{r(2)}`$ $`=`$ $`_t\delta F_{rt}^{(2)}+_z\delta F_{rz}^{(2)}+{\displaystyle \frac{1}{r^2}}_\varphi \delta F_{r\varphi }^{(2)}`$ (178)
$`=`$ $`2B_0r\left[_z\delta r_+_r_z\delta \varphi _{}+_z\delta r_{}_r_z\delta \varphi _++_\varphi \left(_z\delta \varphi _+_z\delta \varphi _{}\right)\right]+{\displaystyle \frac{1}{r^2}}_\varphi \delta F_{r\varphi }^{(2)}`$
$`+`$ $`B_0r\left\{_t^2\delta \varphi _I+_z^2\delta \varphi _I+_\varphi \left[{\displaystyle \frac{1}{r^3}}_r\left(r\delta r_I\right)+{\displaystyle \frac{1}{r^2}}_\varphi \delta \varphi _I\right]\right\}.`$
Thus Eqs. (175) and (176) combine to give the equation of motion
$`_t^2\delta \varphi _I+_z^2\delta \varphi _I`$ $`+`$ $`{\displaystyle \frac{1}{r^2}}_\varphi \left[{\displaystyle \frac{1}{r}}_r\left(r\delta r_I\right)+_\varphi \delta \varphi _I+{\displaystyle \frac{1}{B_0r}}\delta F_{r\varphi }^{(2)}\right]`$
$`=`$ $`{\displaystyle \frac{2}{r}}\left[2_z\delta r_+_r\left(r_z\delta \varphi _{}\right)+2_z\delta r_{}_r\left(r_z\delta \varphi _+\right)+r_\varphi \left(_z\delta \varphi _+_z\delta \varphi _{}\right){\displaystyle \frac{1}{r}}_\varphi \left(_z\delta r_+_z\delta r_{}\right)\right].`$
In this equation, the transverse components of the laplacian on the left-hand side suppress the interaction unless they vanish – that is, unless the the new mode is an Alfvén wave (Eq. 47). Even without making this restriction, the dominant three-mode interaction can be obtained by transforming to the variable
$$\mathrm{\Gamma }_\pm 𝑑\varphi \delta \varphi _\pm (\varphi ,r);\mathrm{\Gamma }_I𝑑\varphi \delta \varphi _I(\varphi ,r).$$
(180)
Here, the integral is performed at constant cylindrical radius. Applying the operator $`𝑑\varphi `$ to Eq. (VI.1) kills off the last term on each side, and we are left with
$$_t^2\mathrm{\Gamma }_I+_z^2\mathrm{\Gamma }_I=\frac{4}{r}𝑑\varphi \left[_z\delta r_+_r\left(r_z\delta \varphi _{}\right)+_z\delta r_{}_r\left(r_z\delta \varphi _+\right)\right].$$
(181)
Transforming to light-cone variables $`z_\pm =z\pm t`$, $`_\pm =\frac{1}{2}(_z\pm _t)`$, the three-mode correction to each Alfvén mode can be calculated from
$`_{}\mathrm{\Gamma }_+|_{\mathrm{out}}_{}\mathrm{\Gamma }_+|_{\mathrm{in}}`$ $`=`$ $`{\displaystyle 𝑑z_+_+_{}\mathrm{\Gamma }_I}`$ (182)
$`=`$ $`{\displaystyle \frac{1}{r}}{\displaystyle 𝑑\varphi \left[_z\delta r_+_r\left(r_z\mathrm{\Delta }\varphi _{}\right)+_z\mathrm{\Delta }r_{}_r\left(r_z\delta \varphi _+\right)\right]}.`$
The correction to $`\mathrm{\Gamma }_{}`$ is obtained by interchanging $`+`$ and $``$ in this equation.
As advertised, the resonant three-mode interaction depends on the presence of a net twist (radial displacement) of the magnetofluid across the wavepacket,
$$\mathrm{\Delta }\varphi _\pm =𝑑z_{}_z\delta \varphi _\pm (z_{});\mathrm{\Delta }r_\pm =𝑑z_{}_z\delta r_\pm (z_{}).$$
(183)
Because the colliding modes are Alfvén modes, $`\mathrm{\Delta }r_\pm `$ is related to $`\mathrm{\Delta }\varphi _\pm `$ through the condition of incompressibility,
$$\frac{1}{r}_r\left(r\mathrm{\Delta }r_\pm \right)+_\varphi \mathrm{\Delta }\varphi _\pm =0.$$
(184)
The three-mode interaction also depends essentially on the non-axisymmetry of the colliding wavepackets (i.e., on the simultaneous presence of coordinate perturbations $`\delta r`$ and $`\delta \varphi `$). Finally we note that applying the operator $`𝑑\varphi `$ to the constraint equation (48) for the fast mode gives
$$\frac{}{r}\left(r^2𝑑\varphi \delta \varphi \right)=0,$$
(185)
hence $`r^2\mathrm{\Gamma }=`$ constant. As long as $`\mathrm{\Gamma }`$ does not have a singularity at $`r=0`$, we see that $`\mathrm{\Gamma }=0`$ for the fast mode.
### VI.2 Collision between two fast waves
In order to tackle the collision between two fast waves, we consider the special case of two planar waves. (There are inherent complications associated with the choice of cylindrical geometry, due to the the non-locality of waveforms propagating in two spatial dimensions – in contrast with one or three dimensions, the Greens function is not a delta function.)
Choosing a background magnetic field $`𝐁_0=B_0\widehat{z}`$, the perturbed Faraday tensor is
$$F_{\mu \nu }=\left(_\mu x^0_\nu y^0_\mu y^0_\nu x^0\right)B_0.$$
(186)
Each wave, by itself, is equivalent to a plane electromagnetic wave superposed on a uniform background magnetic field, and the associated coordinate perturbation satisfies the usual wave equation,
$$x^0=x+\delta x_{1,2}^0(x^\mu );_\nu ^\nu \left[\delta x_{1,2}^0(x^\mu )\right]=0.$$
(187)
We can boost along the background magnetic field into a frame in which one of the colliding fast waves propagates in a direction perpendicular to $`𝐁_0`$, say
$$\delta 𝐱_1^0=\delta x_1^0(xt)\widehat{x}.$$
(188)
A specific example of such a waveform is the harmonic perturbation $`𝐤_1=k_1\widehat{x}`$, $`\delta x_1^0=\delta X_1e^{ik_1(xt)}`$. Because neither fast mode involves a propagating current in the force-free limit, the interaction vanishes if the electric vectors of the two modes are parallel and their direct superposition involves no violation of the constraint $`\delta 𝐄𝐁_0=0`$. Thus the waveform of the second mode is chosen to have a non-vanishing derivative in the $`y`$-direction. (In general all three components of $`𝐤_2`$ are non-vanishing.) The corresponding coordinate fluctuation is
$$\delta 𝐱_2^0=\delta y_2^0(𝐧_2𝐱t)\left[\widehat{y}+\frac{n_{2x}}{n_{2y}}\widehat{x}\right],$$
(189)
where $`𝐧_2`$ is an arbitrary unit vector. The relative normalization of the fluctuations in $`x^0`$ and $`y^0`$ is fixed by the vanishing of the current, $`_y\delta x_2^0_x\delta y_2^0=0`$. We allow for the possibility that there is a non-vanishing displacement across (at least) one of the colliding waveforms, e.g.
$$\mathrm{\Delta }x_1^0=𝑑\mathrm{}(\delta x_1^0)^{}(\mathrm{})0;\mathrm{\Delta }y_2^0=𝑑\mathrm{}(\delta y_2^0)^{}(\mathrm{})0.$$
(190)
The current obtained by superposing $`\delta 𝐱^0=\delta 𝐱_1^0+\delta 𝐱_2^0`$ is
$$4\pi J_{FF}^\mu =_\nu \left(^\mu x^0^\nu y^0^\mu y^0^\nu x^0\right)B_0=\left(_\nu ^\mu \delta x_1^0^\nu \delta y_2^0_\nu ^\mu \delta y_2^0^\nu \delta x_1^0\right)B_0.$$
(191)
Because the linear currents $`J^{\mu (1)}`$ vanish, the force-free condition is $`J^{x(2)}=J^{y(2)}=0`$ to second order in the wave amplitude. We must then introduce an interaction field $`x_{}^{\mu }{}_{I}{}^{0}`$ and associated current
$$4\pi J_I^\mu =\left[_y^\mu x_I^0_x^\mu y_I^0+\delta _x^\mu \left(_\nu ^\nu y_I^0\right)\delta _y^\mu \left(_\nu ^\nu x_I^0\right)\right]B_0.$$
(192)
The conditions
$$J_I^x+J_{FF}^x=0=J_I^y+J_{FF}^y$$
(193)
correspond to the equations of motion
$`_\nu ^\nu y_I^0`$ $`+`$ $`_x(_y\delta x_I^0_x\delta y_I^0)=`$
$`(1n_{2x})\left[(\delta x_1^0)^{\prime \prime }(xt)(\delta y_2^0)^{}(𝐧_2𝐱t)n_{2x}(\delta x_1^0)^{}(xt)(\delta y_2^0)^{\prime \prime }(𝐧_2𝐱t)\right]`$
and
$$_\nu ^\nu x_I^0_y(_y\delta x_I^0_x\delta y_I^0)=n_{2y}(1n_{2x})(\delta x_1^0)^{}(xt)(\delta y_2^0)^{\prime \prime }(𝐧_2𝐱t).$$
(196)
It is useful to project these equations onto the outgoing fast ($`_xx_I^0+_yy_I^0`$) and Alfvén ($`_xy_I^0_xy_I^0`$) modes,
$$_\nu ^\nu \left(_xx_I^0+_yy_I^0\right)=2(1n_{2x})n_{2y}(\delta x_1^0)^{\prime \prime }(xt)(\delta y_2^0)^{\prime \prime }(𝐧_2𝐱t)$$
(197)
and
$`(_t^2+_z^2)\left(_xy_I^0_yx_I^0\right)=(1n_{2x})`$ $`[`$ $`(\delta x_1^0)^{\prime \prime \prime }(xt)(\delta y_2^0)^{}(𝐧_2𝐱t)`$
$`(n_{2x}^2+n_{2y}^2)(\delta x_1^0)^{}(xt)(\delta y_2^0)^{\prime \prime \prime }(𝐧_2𝐱t)].`$
To find the outgoing perturbation to the fast wave 1, we start with the Greens function solution to Eq. (197),
$$\left(_xx_I^0+_yy_I^0\right)(𝐱,t)=2n_{2y}(1n_{2x})d^3x^{}\frac{1}{|𝐱𝐱^{}|}(\delta x_1^0)^{\prime \prime }(x^{}t+|𝐱𝐱^{}|)(\delta y_2^0)^{\prime \prime }(𝐧_2𝐱^{}t+|𝐱𝐱^{}|),$$
(199)
(using the retarded time $`t|𝐱𝐱^{}|`$) and take the limit $`x,t\mathrm{}`$ at $`y=z=0`$. Each wavepacket is localized within a distance $`L`$ along its direction of propagation. Transforming $`x^{}x^{}x`$ and then setting $`\{x^{},y^{},z^{}\}=r\{\mathrm{sin}\theta \mathrm{cos}\varphi ,\mathrm{sin}\theta \mathrm{sin}\varphi ,\mathrm{cos}\theta \}`$, one has
$`(_xx_I^0+_yy_I^0)(x,0,0,t)=2n_{2y}(1n_{2x}){\displaystyle }rdr\mathrm{sin}\theta d\theta d\varphi \{(\delta x_1^0)^{\prime \prime }[r(1+\mathrm{sin}\theta \mathrm{cos}\varphi )+xt]`$ (200)
$`(\delta y_2^0)^{\prime \prime }[r(1+𝐧_2𝛀)+n_{2x}xt]\}.`$ (201)
The integrand is non-vanishing only in a small region $`\mathrm{\Delta }rL`$ near $`r=(tn_{2x}x)/(1+𝐧_2𝛀)`$ and $`\mathrm{\Delta }\theta \mathrm{\Delta }\varphi L/t`$ near $`\theta =\pi /2`$, $`\varphi =\pi `$. When integrating over $`r`$, the argument of $`(\delta x_1^0)^{\prime \prime }`$ varies only by a small amount $`L^2/t`$. The integral $`𝑑\mathrm{}(\delta y_2^0)^{\prime \prime }(\mathrm{})`$ itself vanishes, and so the outgoing fast mode has an amplitude $`(L/t)(\delta x_1^0)^{}(\delta y_2^0)^{}0`$ as $`t\mathrm{}`$. We conclude that the two fast modes do not couple to a third fast mode, even if one of the colliding waveforms has a non-vanishing displacement (190).
Now let us examine the coupling to an outgoing Alfvén mode. Transforming to the light-cone coordinates $`z_\pm =z\pm t`$, and taking $`x=y=0`$, Eq. (VI.2) becomes
$`_+_{}(_xy_I^0`$ $``$ $`_yx_I^0)(0,0,z_+,z_{})=`$
$`{\displaystyle \frac{1}{4}}(1n_{2x})\{(\delta x_1^0)^{\prime \prime \prime }\left[{\displaystyle \frac{1}{2}}(z_{}z_+)\right](\delta y_2^0)^{}[{\displaystyle \frac{1}{2}}(n_{2z}+1)z_{}+{\displaystyle \frac{1}{2}}(n_{2z}1)z_+]`$
$``$ $`(n_{2x}^2+n_{2y}^2)(\delta x_1^0)^{}\left[{\displaystyle \frac{1}{2}}(z_{}z_+)\right](\delta y_2^0)^{\prime \prime \prime }[{\displaystyle \frac{1}{2}}(n_{2z}+1)z_{}+{\displaystyle \frac{1}{2}}(n_{2z}1)z_+]\}.`$
For example, the outgoing Alfvén wave propagating to large positive $`z`$ is
$`_{}(_xy_I^0`$ $``$ $`_yx_I^0)(0,0,zt)=`$
$`{\displaystyle \frac{1}{4}}(1n_{2x}){\displaystyle }dz_+\{(\delta x_1^0)^{\prime \prime \prime }\left[{\displaystyle \frac{1}{2}}(z_{}z_+)\right](\delta y_2^0)^{}[{\displaystyle \frac{1}{2}}(n_{2z}+1)z_{}+{\displaystyle \frac{1}{2}}(n_{2z}1)z_+]`$
$``$ $`(n_{2x}^2+n_{2y}^2)(\delta x_1^0)^{}\left[{\displaystyle \frac{1}{2}}(z_{}z_+)\right](\delta y_2^0)^{\prime \prime \prime }[{\displaystyle \frac{1}{2}}(n_{2z}+1)z_{}+{\displaystyle \frac{1}{2}}(n_{2z}1)z_+]\}.`$
This integral has a closed form solution when one of the wavepackets (say 2) is much shorter than the other and has a zero-frequency component. Then we can treat $`(\delta y_2^0)^{}`$ as a delta function, $`(\delta y_2^0)^{}(\mathrm{})=\mathrm{\Delta }y_2^0\delta (\mathrm{})`$. After restoring the dependence on the transverse coordinates $`x`$ and $`y`$, and integrating once over $`z_{}`$, our final expression for the outgoing wave is
$$\left(_xy_I^0_yx_I^0\right)(x,y,zt)=\frac{(1n_{2x})}{2n_{2z}}\mathrm{\Delta }y_2^0(\delta x_1^0)^{\prime \prime }\left[\frac{1n_{2x}n_{2z}}{1n_{2z}}x\frac{n_{2y}}{1n_{2z}}y\frac{n_{2z}}{1n_{2z}}(zt)\right].$$
(205)
The outgoing Alfvén wave is sheared, with wavevector proportional to
$$(\omega ,𝐤)(n_{2z},n_{2x}+n_{2z}1,n_{2y},n_{2z}).$$
(206)
Notice also that the amplitude of the outgoing wave involves a higher-order derivative of the ingoing wave, and so the higher-wavenumber fourier components are enhanced by the collision.
The frequency of the outgoing Alfvén wave is
$$\omega _A=\frac{n_{2z}n_{1z}}{n_{2z}1}\omega _{F1}$$
(207)
when the fast wave 1 is a pure fourier mode with frequency $`\omega _{F1}`$, and its wavevector has some component along the background magnetic field. The three-wave resonance condition
$$\omega _A=\omega _{F1}+\omega _{F2}=\omega _{F1};k_A^z=k_{F1}^z+k_{F2}^z=k_{F1}^z,$$
(208)
can be satisfied with $`\omega _{F2}=k_{F2}^z=0`$ if the propagation of fast wave 1 is aligned with the background field ($`n_{1z}=1`$). But, more generally, Eq. (207) is not consistent with such a three-wave resonance: the integral over $`z_+`$ in Eq. (VI.2) is non-vanishing only if
$$(n_{1z}1)\omega _{F1}+(n_{2z}1)\omega _{F2}=0.$$
(209)
A solution to this equation, with at least one non-vanishing frequency, requires that either $`n_{1z}=1`$ or $`n_{2z}=1`$. This illustrates why a net field line displacement across an MHD wave has only a limited description in terms of a zero-frequency component of the wave.
### VI.3 Collision between a fast wave and an Alfvén wave
The collision between a fast wave and an Alfvén wave can generate both an outgoing Alfvén mode and an outgoing fast mode through three-wave interactions. It is useful to compare the strength of this interaction with the interaction between two fast modes discussed in the previous section. We focus on the particular case where the Alfvén wave spectrum is highly anisotropic, and the coupling parameter $`k_yx^0k_xy^0(k_{}/\omega )(\delta B_{A,\omega }/B)`$ is independent of scale and close to unity. A key difference between these two types of wave collisions is that the sheared Alfvén waves are current-carrying, $`J_{A,\omega }^{(1)}k_{}\delta B_{A,\omega }`$ and there is a second-order Lorentz force involving the interaction between this current and the fluctuating field of the fast wave. The magnitude of this Lorentz force is
$$4\pi J_{A,\omega }^{(1)}\delta B_{F,\omega }k_{}\delta B_{A,\omega }\delta B_{F,\omega }.$$
(210)
Here $`\delta B_{A,\omega }`$ and $`\delta B_{F,\omega }`$ are the field perturbations of the Alfvén and fast waves at frequency $`\omega `$. We can compare this with the second order current generated in the fast-wave collision,
$$4\pi J_{FF}^{(2)}\omega \frac{\delta B_{F,\omega }^2}{B_0}.$$
(211)
The ratio is
$$\frac{J_{A,\omega }^{(1)}\delta B_{F,\omega }}{J_{FF}^{(2)}B_0}\left(\frac{k_{}\delta B_{A,\omega }}{\omega B_0}\right)\left(\frac{\delta B_{F,\omega }}{B_0}\right)^1\left(\frac{\delta B_{F,\omega }}{B_0}\right)^11.$$
(212)
Thus, if the Alfvén waves form an anisotropic cascade, then the fast wave spectrum evolves predominantly through collisions with Alfvén waves and not through self-collisions between fast waves.
###### Acknowledgements.
We thank the NSERC of Canada for its financial support, and Maxim Lyutikov for conversations.
## Appendix A
In this appendix, we show how the force-free equation may be derived from the action
$$S^{}=d^4x_0L^{}=d^4x\frac{1}{4}\sqrt{g}g_{\mu \rho }g_{\nu \sigma }\stackrel{~}{F}^{\mu \nu }\stackrel{~}{F}^{\rho \sigma }=d^4x_0\frac{J_4}{4}\sqrt{g}g_{\mu \rho }g_{\nu \sigma }\stackrel{~}{F}^{\mu \nu }\stackrel{~}{F}^{\rho \sigma },$$
(213)
using the magnetic lagrangian fluid variables (10). Varying Eq. (213),
$$\delta S^{}=d^4x_0\delta L^{}=d^4x_0\left[\frac{L^{}}{x^\mu }\delta x^\mu +\frac{L^{}}{(x^\mu /x_{0}^{}{}_{}{}^{\alpha })}\delta \left(\frac{x^\mu }{x_{o}^{}{}_{}{}^{\alpha }}\right)\right],$$
(214)
there are terms which arise from the explicit $`x^\mu `$-dependence of the metric,
$$\frac{L^{}}{x^\mu }=\left[\frac{(g_0)}{4J_4}\frac{x^\beta }{x_0^\kappa }\frac{x^\gamma }{x_0^\lambda }\frac{x^\delta }{x_0^\eta }\frac{x^\epsilon }{x_0^\zeta }\stackrel{~}{F}_{0}^{}{}_{}{}^{\kappa \lambda }\stackrel{~}{F}_{0}^{}{}_{}{}^{\eta \zeta }\right]\frac{}{x^\mu }\left[\frac{g_{\beta \delta }g_{\gamma \epsilon }}{\sqrt{g}}\right],$$
(215)
as well as from the dependence of $`\stackrel{~}{F}^{\mu \nu }`$ on $`x^\mu /x_0^\alpha `$,
$$\frac{L^{}}{(x^\mu /x_{0}^{}{}_{}{}^{\alpha })}=\left[\frac{g_{\beta \delta }g_{\gamma \epsilon }}{\sqrt{g}}\right]\frac{}{(x^\mu /x_{0}^{}{}_{}{}^{\alpha })}\left[\frac{(g_0)}{4J_4}\frac{x^\beta }{x_0^\kappa }\frac{x^\gamma }{x_0^\lambda }\frac{x^\delta }{x_0^\eta }\frac{x^\epsilon }{x_0^\zeta }\stackrel{~}{F}_{0}^{}{}_{}{}^{\kappa \lambda }\stackrel{~}{F}_{0}^{}{}_{}{}^{\eta \zeta }\right].$$
(216)
After extremizing, the usual Euler-Lagrange equations are obtained,
$$\frac{}{x_0^\alpha }\left[\frac{L^{}}{(x^\mu /x_0^\alpha )}\right]\frac{L^{}}{x^\mu }=0.$$
(217)
Substituting eqs. (215) and (216) then gives
$`{\displaystyle \frac{}{x_0^\alpha }}\left[g_{\beta \delta }g_{\gamma \mu }\stackrel{~}{F}^{\beta \gamma }\left(\sqrt{g_0}{\displaystyle \frac{x^\delta }{x_0^\eta }}\stackrel{~}{F}_0^{\eta \alpha }\right){\displaystyle \frac{J_4}{(x^\mu /x_0^\alpha )}}{\displaystyle \frac{\sqrt{g}}{4}}g_{\beta \delta }g_{\gamma \epsilon }\stackrel{~}{F}^{\beta \gamma }\stackrel{~}{F}^{\delta \epsilon }\right]`$
$`{\displaystyle \frac{(g)J_4}{4}}\stackrel{~}{F}^{\beta \gamma }\stackrel{~}{F}^{\delta \epsilon }{\displaystyle \frac{}{x^\mu }}\left[{\displaystyle \frac{g_{\beta \delta }g_{\gamma \epsilon }}{\sqrt{g}}}\right]=0.`$ (218)
Now the background field satisfies
$$\frac{}{x_0^\alpha }\left(\sqrt{g_0}\frac{x^\delta }{x_0^\eta }\stackrel{~}{F}_0^{\eta \alpha }\right)=0.$$
(219)
Making use of this relation, and the identities
$$\frac{}{x_0^\alpha }\left[\frac{J_4}{(x^\mu /x_0^\alpha )}\right]=0;J_4\frac{}{x^\mu }=\frac{J_4}{(x^\mu /x_0^\alpha )}\frac{}{x_0^\alpha }$$
(220)
in Eq. (A) gives
$$\stackrel{~}{F}^{\delta \epsilon }\frac{}{x^\epsilon }\left[g_{\beta \delta }g_{\gamma \mu }\stackrel{~}{F}^{\beta \gamma }\right]\frac{1}{2}\stackrel{~}{F}^{\delta \epsilon }\frac{}{x^\mu }\left[g_{\beta \delta }g_{\gamma \epsilon }\stackrel{~}{F}^{\beta \gamma }\right]=0.$$
(221)
To demonstrate the equivalence of this expression and the usual force free equation, one need only substitute $`F_{\alpha \beta }=\frac{1}{2}\sqrt{g}\epsilon _{\alpha \beta \rho \sigma }\stackrel{~}{F}^{\rho \sigma }`$ in
$$J^\mu F_{\mu \nu }=\frac{1}{\sqrt{g}}\frac{}{\rho }\left(\sqrt{g}g^{\mu \alpha }g^{\rho \beta }F_{\alpha \beta }\right)F_{\mu \nu }=0.$$
(222)
After making use of the contraction
$$g^{\mu \alpha }\epsilon _{\alpha \beta \gamma \delta }\epsilon _{\mu \nu \rho \sigma }=4\frac{g_{\beta \nu }g_{\gamma \rho }g_{\delta \sigma }}{(g)}\left[\delta _\beta ^\epsilon \delta _\gamma ^\kappa \delta _\delta ^\lambda \delta _\beta ^\epsilon \delta _\delta ^\kappa \delta _\gamma ^\lambda +\delta _\delta ^\epsilon \delta _\beta ^\kappa \delta _\gamma ^\lambda \delta _\delta ^\epsilon \delta _\gamma ^\kappa \delta _\beta ^\lambda +\delta _\gamma ^\epsilon \delta _\delta ^\kappa \delta _\beta ^\lambda \delta _\gamma ^\epsilon \delta _\beta ^\kappa \delta _\delta ^\lambda \right],$$
(223)
Eq. (13) is obtained.
## Appendix B
In this appendix we list, for reference, the axisymmetric perturbations of a background magnetofluid satisfying $`J^\mu =0`$, using the electric Lagrangian fluid variables (6).
The field strengths are given by
$$F_{r\varphi }=\left(\frac{r_0}{r}\frac{\varphi _0}{\varphi }\frac{\varphi _0}{r}\frac{r_0}{\varphi }\right)F_{r\varphi }^{(0)}+\left(\frac{r_0}{r}\frac{t_0}{\varphi }\frac{t_0}{r}\frac{r_0}{\varphi }\right)F_{rt}^{(0)},$$
(224)
where the background fields $`F_{r\varphi }^{(0)}`$ and $`F_{rt}^{(0)}`$ are
$$F_{r\varphi }^{(0)}=\frac{B_0g_{0\varphi \varphi }g_{0rr}}{\sqrt{g_0}R_0};$$
(225)
$$F_{rt}^{(0)}=\left(\frac{g_0^{t\varphi }}{g_0^{tt}}\right)F_{r\varphi }^{(0)}.$$
(226)
In a cylindrically symmetric spacetime with $`z_0=z`$, $`\varphi _0/\varphi =1`$, and $`r_0/\varphi =t_0/\varphi =0`$, the perturbations $`\delta F_{\mu \nu }=F_{\mu \nu }F_{\mu \nu }^{(0)}`$ have the form
$$\delta F_{r\varphi }=\left(\frac{r_0}{r}\right)F_{r\varphi }^{(0)}(r_0)F_{r\varphi }^{(0)}(r);$$
(227)
$$\delta F_{rt}=\left(\frac{r_0}{r}\frac{\varphi _0}{t}\frac{\varphi _0}{r}\frac{r_0}{t}\right)F_{r\varphi }^{(0)}(r_0)+\left(\frac{r_0}{r}\frac{t_0}{t}\frac{t_0}{r}\frac{r_0}{t}\right)F_{rt}^{(0)}(r_0)F_{rt}^{(0)}(r);$$
(228)
$$\delta F_{rz}=\left(\frac{r_0}{r}\frac{\varphi _0}{z}\frac{\varphi _0}{r}\frac{r_0}{z}\right)F_{r\varphi }^{(0)}(r_0)+\left(\frac{r_0}{r}\frac{t_0}{z}\frac{t_0}{r}\frac{r_0}{z}\right)F_{rt}^{(0)}(r_0);$$
(229)
$$\delta F_{t\varphi }=\left(\frac{r_0}{t}\right)F_{r\varphi }^{(0)}(r_0);$$
(230)
$$\delta F_{z\varphi }=\left(\frac{r_0}{z}\right)F_{r\varphi }^{(0)}(r_0);$$
(231)
and
$$\delta F_{tz}=\left(\frac{r_0}{t}\frac{\varphi _0}{z}\frac{\varphi _0}{t}\frac{r_0}{z}\right)F_{r\varphi }^{(0)}(r_0)+\left(\frac{r_0}{t}\frac{t_0}{z}\frac{t_0}{t}\frac{r_0}{z}\right)F_{rt}^{(0)}(r_0).$$
(232)
Substituting $`F_{rt}^{(0)}(r_0)=(g_0^{t\varphi }/g_0^{tt})F_{r\varphi }^{(0)}(r_0)`$, these equations reduce to
$$\delta F_{r\varphi }=\frac{r_0}{r}F_{r\varphi }^{(0)}(r_0)F_{r\varphi }^{(0)}(r)$$
(233)
$$\delta F_{rt}=\left[\left(\frac{r_0}{r}\frac{\varphi _0}{t}\frac{\varphi _0}{r}\frac{r_0}{t}\right)\left(\frac{r_0}{r}\frac{t_0}{t}\frac{t_0}{r}\frac{r_0}{t}\right)\left(\frac{g_0^{t\varphi }}{g_0^{tt}}\right)\right]F_{r\varphi }^{(0)}(r_0)\left(\frac{g^{t\varphi }}{g^{tt}}\right)F_{r\varphi }^{(0)}(r);$$
(234)
$$\delta F_{rz}=\left[\left(\frac{r_0}{r}\frac{\varphi _0}{z}\frac{\varphi _0}{r}\frac{r_0}{z}\right)\left(\frac{r_0}{r}\frac{t_0}{z}\frac{t_0}{r}\frac{r_0}{z}\right)\left(\frac{g_0^{t\varphi }}{g_0^{tt}}\right)\right]F_{r\varphi }^{(0)}(r_0);$$
(235)
$$\delta F_{t\varphi }=\left(\frac{r_0}{t}\right)F_{r\varphi }^{(0)}(r_0);$$
(236)
$$\delta F_{z\varphi }=\left(\frac{r_0}{z}\right)F_{r\varphi }^{(0)}(r_0);$$
(237)
and
$$\delta F_{tz}=\left[\left(\frac{r_0}{t}\frac{\varphi _0}{z}\frac{\varphi _0}{t}\frac{r_0}{z}\right)\left(\frac{r_0}{t}\frac{t_0}{z}\frac{t_0}{t}\frac{r_0}{z}\right)\left(\frac{g_0^{t\varphi }}{g_0^{tt}}\right)\right]F_{r\varphi }^{(0)}(r_0).$$
(238)
For example,
$$\delta F_{r\varphi }=\left[\left(\frac{r_0}{r}\right)\frac{g_{0rr}g_{0\varphi \varphi }}{\sqrt{g_0}}\frac{g_{rr}g_{\varphi \varphi }}{\sqrt{g}}\right]B_0.$$
(239)
## Appendix C
This appendix is devoted to a derivation of the force-free equation (100) in a non-static, axially symmetric spacetime. The force-free equation (94) is
$`\left[_r\left(\sqrt{g}g^{tt}g^{rr}\delta F_{tr}+\sqrt{g}g^{t\varphi }g^{rr}\delta F_{\varphi r}\right)+_z\left(\sqrt{g}g^{t\varphi }g^{zz}\delta F_{\varphi z}\right)\right]F_{tr}^{(0)}`$ $`+`$
$`[_r(\sqrt{g}g^{\varphi \varphi }g^{rr}\delta F_{\varphi r}+\sqrt{g}g^{\varphi t}g^{rr}\delta F_{tr})+_z\left(\sqrt{g}g^{\varphi \varphi }g^{zz}\delta F_{\varphi z}\right)`$ $`+`$
$`_t(\sqrt{g}g^{\varphi \varphi }g^{tt}\delta F_{\varphi t}+\sqrt{g}g^{\varphi t}g^{\varphi t}\delta F_{t\varphi })]F^{(0)}_{\varphi r}=0.`$ (240)
where the terms involving $`\delta F_{tz}`$ have canceled. The terms involving $`\delta F_{tr}`$ may be combined to give
$$\sqrt{g}g^{tt}g^{rr}\delta F_{tr}\frac{}{r}\left(\frac{g^{t\varphi }}{g^{tt}}\right)F_{r\varphi }^{(0)}.$$
(241)
The terms involving $`\delta F_{\varphi r}`$ combine to give
$$_r\left[\frac{\sqrt{g}}{g_{rr}g_{\varphi \varphi }}\frac{}{r}\left(\delta rF_{r\varphi }^{(0)}\right)\right]F_{r\varphi }^{(0)}\sqrt{g}g^{t\varphi }g^{rr}\frac{}{r}\left(\frac{g^{t\varphi }}{g^{tt}}\right)\frac{}{r}\left(\delta rF_{r\varphi }^{(0)}\right)F_{r\varphi }^{(0)}.$$
(242)
Here $`det(t,\varphi )=g_{tt}g_{\varphi \varphi }g_{t\varphi }^2`$. Lastly, the terms involving $`\delta F_{\varphi t}`$ and $`\delta F_{\varphi z}`$ yield
$$\frac{\sqrt{g}}{det(t,\varphi )}\left(\frac{^2\delta r}{t^2}+\frac{g^{zz}}{g^{tt}}\frac{^2\delta r}{z^2}\right)(F_{r\varphi }^{(0)})^2.$$
(243)
To first order in the coordinate perturbation, the perturbed Maxwell fields are
$$\delta F_{r\varphi }=\frac{}{r}\left[\delta rF_{r\varphi }^{(0)}(r)\right];$$
(244)
$$\delta F_{rt}=\frac{\delta \stackrel{~}{\varphi }}{t}F_{r\varphi }^{(0)}(r)\frac{}{r}\left[\delta r\frac{g^{t\varphi }(r)}{g^{tt}(r)}F_{r\varphi }^{(0)}(r)\right];$$
(245)
$$\delta F_{t\varphi }=\frac{\delta r}{t}F_{r\varphi }^{(0)}(r);$$
(246)
and
$$\delta F_{z\varphi }=\frac{\delta r}{z}F_{r\varphi }^{(0)}(r).$$
(247)
Substituting these expressions and collecting terms, we arrive at
$`{\displaystyle \frac{1}{det(t,\varphi )}}\left[{\displaystyle \frac{^2\delta r}{t^2}}+{\displaystyle \frac{g^{zz}}{g^{tt}}}{\displaystyle \frac{^2\delta r}{z^2}}\right](F_{r\varphi }^{(0)})^2+{\displaystyle \frac{1}{\sqrt{g}}}{\displaystyle \frac{}{r}}\left[{\displaystyle \frac{\sqrt{g}}{g_{\varphi \varphi }g_{rr}}}{\displaystyle \frac{}{r}}(\delta rF_{r\varphi }^{(0)})\right]F_{r\varphi }^{(0)}`$
$`g^{tt}g^{rr}{\displaystyle \frac{}{r}}\left({\displaystyle \frac{g^{t\varphi }}{g^{tt}}}\right)\left[{\displaystyle \frac{\stackrel{~}{\varphi }}{t}}+\delta r{\displaystyle \frac{}{r}}\left({\displaystyle \frac{g^{t\varphi }}{g^{tt}}}\right)\right](F_{r\varphi }^{(0)})^2`$ $`=`$ $`0.`$ (248)
If we multiply by $`det(\varphi ,t)g^{tt}`$ we obtain equation $`(\text{100})`$.
## Appendix D
This www.arxiv.org manuscript incorporates several minor corrections that will appear as an erratum to the paper as published in the December 2004 issue of Physical Review D.
1) The second term, RHS, Eq. (100) \[and the analogous term in Eq. (C9)\] was spurious and has been deleted.
2) An algebraic error was corrected in the RHS of Eqns. (181) and (182): the normalization increased by a factor 2; and $`r^4[r^4]r^1[r]`$ in front of \[inside\] the integral. The intermediate equation (179) also received a correction.
3) In Section V.B, a final step has been added to the derivation of the fast mode amplitude \[Eq. (166b)\]. The claim \[based on Eq. (157)\] that $`\delta R_{\mathrm{max}}`$ can be set to zero was incorrect: the integral for the background field energy in Eq. (157) in fact converges at large radius. As a result, the LHS of Eq. (166) gains a term $`\delta R_{\mathrm{max}}/R_{\mathrm{max}}`$, and the final step Eq. (166b) is required.
In addition there were a handful of (non-propagating) corrections to signs and indices. |
warning/0507/astro-ph0507056.html | ar5iv | text | # Dark and stellar matter in strong lensing galaxies from a joint lensing and stellar dynamics Based on data collected at Subaru Telescope, which is operated by the National Astronomical Observatory of Japan.
## 1 Introduction
Massive early type galaxies are predicted to form at the highest density peaks in the early universe and to evolve through subsequent mass assemblies. Measuring their dynamical structure and stellar content over the cosmic time provides important constraints not only on their formation history itself but also on models of structure formation scenario and cosmic star formation history.
Studying the internal structure of early type galaxies is, however, difficult because of lack of dynamical tracers at large radii such like H<sub>I</sub> gas in spirals, and of degeneracy between kinematic properties of dynamical tracers and mass distributions: The stellar dynamics suffers from the mass-anisotropy problem, and the gravitational lensing does from the mass-density profile problem as well.
Those degeneracies can be broken by combining those two probes, since they nicely complement each other (Treu & Koopmans 2002; 2004 (hereafter TK04); Koopmans & Treu 2003). A combined analysis thus places important limits on the distributions of luminous and dark matter in a lensing early type galaxy.
We select two lens systems, HST14113$`+`$5211 ($`z_L=0.464`$; Fischer, Schade & Barrientos 1998; Lubin et al. 2000) and B 2045$`+`$265 ($`z_L=0.868`$; Fassnacht et al. 1999), where the lensing galaxies are an early-type galaxy. We conducted the direct measurement of velocity dispersion of the lensing galaxies of those systems using the Faint Object Camera and Spectrograph (FOCAS, Kashikawa et al. 2002) mounted on the Subaru Telescope. We perform a joint gravitational lensing and stellar dynamics analysis to explore the luminous and dark matter distribution of lensing galaxies focusing on the slope of (dark and total matter) density profile and mass-to-light ratio. Also, we utilize the fundamental plane relation of early type galaxies for deriving an alternative estimate of the mass-to-light ratio, which, after combined with the joint analysis, provides a useful constraint on the mass distribution of the lensing galaxy.
The paper is organized as follows. In §2, we show the observations and data reduction. In §3, we derive the mass-to-light ratio of the lensing galaxies from the fundamental plane relation of early type galaxies. In §4, models for gravitational lensing and stellar dynamics are presented. §5 is devoted to the results of our model analysis, and discussion and concluding remarks are drawn in §6.
Throughout this paper, we adopt $`\mathrm{\Omega }_0=0.3`$, $`\lambda _0=0.7`$, and $`h=H_0/100`$ km s<sup>-1</sup> Mpc<sup>-1</sup> $`=0.65`$ for the relevant estimations.
## 2 VELOCITY DISPERSION MEASUREMENT
### 2.1 Observations
Spectroscopic observations were made with the Subaru 8.2m telescope (Iye et al. 2004). The FOCAS spectrograph (Kashikawa et al. 2002) was configured with a 0″.4-width slit, a 300 grooves mm<sup>-1</sup> grism which gives 1.40Å per pixel, and a Y47 order-cut filter to obtain optical spectra of the lensing galaxies of gravitational lens systems HST 14113$`+`$5211 and B 2045$`+`$265. The on-chip binning was set to 3 (along spatial direction to give 0″.3 per pixel) by 1 (along wavelength direction).
For HST 14113$`+`$5211, observation was made on 2002 June 13 and 14 (UT). The slit was placed along the major axis of the lensing galaxy, whose position angles (PAs) was PA$`=38.1^{}`$. We obtained ten 1800 seconds exposures, and the total exposure time was 5 hours. For B 2045$`+`$265, observation was made on 2002 June 13 (UT). The slit was placed along the major axis of the lensing galaxy (PA$`=120.1^{}`$). We obtained seven 1800 seconds exposures, and the total exposure time was 3.5 hours. The seeing conditions were around 0.4″ – 0.8″ during the two observing nights.
### 2.2 Data reduction
Basic data reductions were made following Ohyama et al. (2002). In the followings, we focus on the spectra of the lensing galaxies. Wavelength accuracy and resolution of the galaxy spectra were measured, within the wavelength region for the Fourier cross-correlation analyses (see section 2.4), by means of Gaussian fittings of several narrowest (unblended) sky OH lines. Here all measured values are shown in the rest frame of each galaxy, since all the following velocity dispersion measurements were made in their rest frames. For the spectrum of HST 14113$`+`$5211, we found that wavelength resolution is 4.5Å FWHM at $`40004450`$Å (the blue fitting region) and 3.2Å at $`47805170`$Å (the red fitting region. See below for details of the spectrum fitting regions). The wavelength accuracy is found to be typically $`0.09`$Å in RMS over $`38006500`$Å, and it gets slightly worse to $`0.28`$Å at the bluest wavelength ($`4000`$Å) for the cross-correlation analyses. For B 2045$`+`$265, wavelength resolution is 2.7Å FWHM at $`41004500`$Å, and wavelength accuracy is typically $`0.07`$Å in RMS over $`38004500`$Å. Aperture size to extract an 1-dimensional galaxy spectrum are 0″.9 (3-binned pixel, corresponding to 8.3 kpc at $`z=0.464`$ for the adopted cosmology) and 1″.2 (4-binned pixel, 19 kpc at $`z=0.868`$) for HST 14113$`+`$5211 and B 2045$`+`$265, respectively, to obtain spectra with maximum signal-to-noise quality.
### 2.3 Basic properties of the spectra
Reduced rest-frame spectra, after continuum normalization, are shown in Figures 1 and 2. For HST 14113$`+`$5211, deep Ca II H and K absorption lines and prominent G band feature are evidently seen in the spectrum, suggesting that major contribution in the observed wavelength region could be attributed to the late type giant stars (late G giant – early K giant stars). The redshift of the lensing galaxy is measured to be $`z=0.4644\pm 0.0002`$ from these prominent features, and is consistent with previous measurement (Lubin et al. 2000). In the redder part of the spectrum, several more features, including H$`\beta `$ and some Fe absorptions (Fe I 5270Å and 5406Å) are seen as well as Mg Ib 5172Å which is partly detected at the blue edge of the atmospheric absorption feature (the A band).
For B 2045$`+`$265, although signal-to-noise ratio of the observed spectrum is worse than that of HST 14113$`+`$5211 due to both fainter apparent brightness and shorter exposure time achieved for the galaxy, we have clearly detected Ca II H and K absorption lines and G band feature (Fig 2). We note, however, that these features are less prominent than those of HST 14113$`+`$5211. We also note that B 2045$`+`$265 shows another possible prominent absorption feature at just red edge of the A band feature. We checked this possibility by comparing the A band features of both galaxies in the observed frame (at 7660Å), and confirmed that B 2045$`+`$265 shows an excess absorption over expected A-band feature. This feature, at 4100Å in the rest frame, can be identified as H$`\delta `$ absorption, although another Balmer absorption line, H$`\gamma `$ near the G band, was not detected. Since H$`\gamma `$ emission is $`1.8`$ times brighter than H$`\delta `$ emission under the “case B” photoionization, which is typical for star-forming regions, difference in the observed properties of Balmer lines can be naturally understood. Note that more higher-order Balmer absorptions were identified by Fassnacht et al. (1999). Prominent \[OII\] emission is also detected, suggesting a star-forming activity in this galaxy. Fassnacht et al. (1999) classified the spectrum as the Sa type, and are best represented by giant stars of late F type. The redshift of the lensing galaxy is measured to be $`z=0.8682\pm 0.0001`$ by \[OII\] emission, which is consistent with that measured by prominent absorption lines, although it is slightly larger than the value measured by Fassnacht et al. (1999) ($`z=0.8673\pm 0.0005`$).
### 2.4 Fourier cross-correlation analysis
We basically followed the procedure of Ohyama et al. (2002), and the Fourier cross-correlation method (Tonry & Davis 1979) was used to measure the line-of-sight velocity dispersion of the lensing galaxies with the FXCOR task implemented in IRAF<sup>1</sup><sup>1</sup>1 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. (see also Falco et al. 1997). Here we outline our procedure applied for both galaxies. Firstly, velocity-resolution matched stellar spectra were created from the “Cóude feed spectral library” spectra (Leitherer et al. 1996), whose wavelength resolutions are 1.8Å FWHM for both red and blue spectra in the library, to match the resolution of the observed galaxy spectra, and were used as “templates” which are to be fitted with the observed galaxy spectra. Secondly, calibration curves, which relate the velocity dispersion of Gaussian broadening function applied to the templates to the width of the cross-correlation functions (CCF) peak, were created. Finally, curves were used to find the velocity dispersion of the galaxy from the CCF peak width calculated between the observed galaxy spectra and the templates. FXCOR parameters were kept unchanged from that used by Ohyama et al. (2002) except for template stars and fitting wavelength regions to match the spectral properties of the program galaxies in this work.
#### 2.4.1 HST 14113$`+`$5211
For HST 14113$`+`$5211, we choose two separate wavelength regions ($`40004450`$Å and $`47805170`$Å, hereafter blue and red fitting regions, respectively) to be used in the FXCOR, which include most of the prominent absorption features, such as G band, Fe I, and H$`\beta `$, while avoiding atmospheric absorption features (A and B bands). Fe I features at $`52505410`$Å are not included in the analyses because they are detected at redder part of the A band and available wavelength region around them is not wide enough. The Ca II H and K features were also excluded from the fitting regions, since the calculation with these features might give problematic result because of their intrinsically wider line width (Tonry 1998). We choose late type giant stars with sub-solar metallicity as spectral templates, since we found during the course of the spectrum fitting that stars with nearly solar metallicity show deeper Fe I absorptions than the observed ones. We choose three such stars, HD 206453, HD 1918, and HD 126778 (see Table 1 for their detailed properties)<sup>2</sup><sup>2</sup>2 Metallicity of the selected template stars (\[Fe/H\]$`=0.4`$$`0.5`$) is by no means the best estimated value for the galaxy. They are arbitrarily chosen from a small subsample of low-metal stars in the spectral library.. For each star, blue ($`38204500`$Å) and red ($`47805450`$Å) spectra are available in the spectral library, and each of them was used separately in the FXCOR for fitting blue ($`40004450`$Å) and red ($`47805170`$Å) fitting regions, respectively. Red and blue template spectra were processed separately in both the continuum normalization and velocity resolution matching procedures, and then each template was used separately in the FXCOR to make two calibration curves for a set of one template star and the galaxy.
We made FXCOR runs twelve times for each template (three stars), fitting region (two regions per star), and CCF peak width measurement lag (two lags, 30 and 35, per set of template and fitting region; see Ohyama et al. 2002 for details of the parameter) to measure velocity dispersion of HST 14113$`+`$5211. We found that all cross-correlations between the galaxy and templates gave good Tonry-Davis $`R`$ values ($`R=1418`$; Tonry and Davis 1979), suggesting that templates adopted for HST 14113$`+`$5211 could represent the galaxy spectrum well. We found that error of the galaxy velocity dispersion, originating from the error in measuring the CCF peak widths, was typically $`10`$ km s<sup>-1</sup> ($`1\sigma `$). Averaging all the FXCOR results with a weight of $`R`$s, we obtained $`\sigma =179\pm 9`$ km s<sup>-1</sup> and $`168\pm 7`$ km s<sup>-1</sup> in blue and red fitting regions, respectively. Here, errors represent $`1\sigma `$ scatters of the velocity dispersion values among all FXCOR runs. Another source of error, the velocity resolution matching error, estimated from the error in measuring the spectral resolution of the galaxy spectrum, could amount to $`14`$ km s<sup>-1</sup> and $`7`$ km s<sup>-1</sup> of galaxy velocity dispersion error ($`1\sigma `$) for blue and red fitting regions, respectively. Considering all sources of uncertainties (errors in velocity-resolution matching, CCF peak width measurement, and scattering of the velocity dispersion values among FXCOR runs), the overall $`1\sigma `$ uncertainties are estimated to be 19 ($`\sqrt{14^2+10^2+9^2}`$) and 13 ($`\sqrt{7^2+10^2+7^2}`$) km s<sup>-1</sup> for blue and red fitting regions, respectively. Note that results measured in two fitting regions are consistent to each other within expected errors. Therefore, averaging all twelve FXCOR results made in blue and red fitting regions, we adopt $`\sigma =174\pm 20`$ km s<sup>-1</sup> as the best estimated line-of-sight velocity dispersion of HST 14113$`+`$5211. Here, an error is estimated, based on velocity matching error of 14 km s<sup>-1</sup> (larger value out of two fitting regions), CCF fitting error of 10 km s<sup>-1</sup>, and scattering among all FXCOR results (10 km s<sup>-1</sup>; see Table 2), to be 20 ($`\sqrt{14^2+10^2+10^2}`$) km s<sup>-1</sup>.
Fig 1 compares the spectrum of HST 14113$`+`$5211 with the best template, HD 126778, giving the best $`R`$ values for both blue and red fitting regions, after Gaussian-broadened with velocity dispersion of $`\sigma =174`$ km s<sup>-1</sup>. We also show residual spectrum (galaxy $``$ Gaussian-broadened template) along with the sky spectrum. All the spectra are shown over wider wavelength region, covering other spectral features not used for the FXCOR analyses, to show the overall quality of the spectral fitting. One may find that the broadened template gives rather good representation of the observed galaxy spectrum over entire wavelength region, except for a large-scale residual pattern in the blue continuum. The residual feature likely comes from errors in fitting the continuum spectra of both the galaxy and the templates, used for continuum normalization. Since the continuum fittings were made over the wavelength regions where spectra show rather large-amplitude change in shape, an artificial large-scale spectral variation could be easily introduced into the continuum-normalized spectra. Note, however, that this error does not affect our cross-correlation results, since such a low-frequency variation in spectral shape is filtered out in the FXCOR before calculating the CCF (see Ohyama et al. 2002 for more details on the FXCOR parameters). In concluding, after considering the fitting quality as discussed above, we finally adopt $`\sigma =174\pm 20`$ km s<sup>-1</sup> as the best estimated velocity dispersion of HST 14113$`+`$5211 in the following lens analysis.
#### 2.4.2 B 2045$`+`$265
For B 2045$`+`$265, we followed the same procedure used for HST 14113$`+`$5211. However, because of relatively poor quality of B 2045$`+`$265 spectrum, only a narrow wavelength region, comprising two smaller wavelength regions of $`42804325`$Å and $`43554430`$Å, was chosen for the FXCOR analyses. The spectral regions were carefully chosen to include G band while avoiding both the strong sky emission regions and H$`\gamma `$, and these two smaller regions were fitted at a same time in FXCOR. Here, the H$`\gamma `$ was excluded in the fitting since H$`\gamma `$ emission may overlay underlying H$`\gamma `$ absorption. We choose two late F giant stars, BD+11 2998 and HD 136202, as spectral templates for the galaxy from the same spectral library (see Table 1 for their detailed properties)<sup>3</sup><sup>3</sup>3 Again, as in the case of the analyses of HST 14113$`+`$5211, metallicity of the selected template stars is by no means the best estimated value for the galaxy.. We ran FXCOR four times (for one fitting region, two template stars, and with two CCF peak width measurement lags), and found that $`R`$ values are rather large, $`3040`$, for all FXCOR runs. The larger $`R`$ values probably result from the fact that we have only one prominent spectral feature (G band) in a narrow wavelength region for fitting. The best estimated velocity dispersion is $`\sigma =213`$ km s<sup>-1</sup>, with an $`1\sigma `$ error of $`\pm 11`$ km s<sup>-1</sup> originating from scattering of the $`\sigma `$ values among all FXCOR runs (see Table 3). The velocity resolution-matching error could amount to $`5`$ km s<sup>-1</sup> of galaxy velocity dispersion error ($`1\sigma `$), which is smaller than in the case of HST 14113$`+`$5211 due to higher redshift of B 2045$`+`$265. We found that error of the galaxy velocity dispersion, originating from the error in measuring the CCF peak widths, was typically $`19`$ km s<sup>-1</sup> ($`1\sigma `$), which is larger than the value for HST 14113$`+`$5211 because of relatively poor CCF determination resulting from lower quality spectrum and narrower wavelength fitting regions available for B 2045$`+`$265. Considering all sources of uncertainties (errors in velocity-resolution matching, CCF peak width measurement, and scattering of the velocity dispersion values among FXCOR runs), the overall $`1\sigma `$ uncertainty is estimated to be 23 ($`\sqrt{5^2+19^2+11^2}`$) km s<sup>-1</sup>.
Fig 2 compares the observed galaxy spectrum with the Gaussian-broadened template spectrum of the best template, HD 136202, after Gaussian-broadened with the velocity dispersion of $`\sigma =213`$ km s<sup>-1</sup>, as in the same way for HST 14113$`+`$5211. One may find that the broadened template gives rather good representation of the galaxy spectrum over entire wavelength region, including H$`\delta `$, except for H$`\gamma `$ and Ca II H and K features. We expect that the residual emission of H$`\gamma `$ represents the real emission, overlaying on the absorption line, as expected from another emission line of \[OII\]. On the other hand, it seems difficult to attribute the shallower-than-observed Ca II absorptions in the broadened template spectrum to overlaying emissions. Although putative H$`ϵ`$ emission, whose wavelength is almost coincident with that of Ca II H absorption, might explain a part of the residual feature, there seems no emission lines which could be coincident with Ca II K feature neither in the spectra of the lens galaxy nor the lensed QSO. Therefore, the most simple and plausible explanation for the shallower observed Ca II absorption is that the lensing galaxy contains not only stars of late F type but also of even earlier types. Although neglecting the contribution of the earlier population in the FXCOR analyses should affect the velocity dispersion measurement, we expect that our result, made assuming a pure stellar population of late F type as spectral templates, is still reasonably reliable and is useful for our analyses on the mass model and the stellar dynamics in the following sections. This is because (1) stars of earlier types show much shallow G band feature, and they can not fit the observed spectrum alone, indicating that late F stars contribute most to the observed spectrum in flux, and (2) our measurement almost relies on the spectral fit of the G band feature and, hence, the velocity dispersion measured in our analyses lies larger weight on the later type stars, which contribute much more on the mass of the lensing galaxy than the case of stars of earlier types. Therefore, we adopt $`\sigma =213\pm 23`$ km s<sup>-1</sup> as the best estimated line-of-sight velocity dispersion of B 2045$`+`$265 in the following lens analysis.
## 3 Fundamental plane
Before going into a joint gravitational lensing and stellar dynamics, let us examine the Fundamental Plane (FP) for early type galaxies, which provides us with an alternative insight into the value of the mass-to-light ratio $`M_{}/L_B`$. The FP is defined by the effective radius ($`R_e`$ kpc), central velocity dispersion ($`\sigma _c`$ km s<sup>-1</sup>), and mean surface brightness inside $`R_e`$ ($`SB_e`$ mag arcsec<sup>-2</sup>):
$$\mathrm{log}R_e=\alpha _{FP}\mathrm{log}\sigma _c+\beta _{FP}SB_e+\gamma _{FP},$$
(1)
where $`\alpha _{FP}=1.25`$, $`\beta _{FP}=0.32`$, and $`\gamma _{FP}=8.895\mathrm{log}h_{50}`$ ($`h_{50}=H_0/50`$ km s<sup>-1</sup> Mpc<sup>-1</sup>) in the B band (Bender et al. 1998). The central velocity dispersion $`\sigma _c`$ is taken as the velocity dispersion measured inside $`R_e/8`$ (TK04).
For the lensing galaxy of HST 14113$`+`$5211, Fischer, Schade, & Barrientos (1998, hereafter F98) obtained the rest-frame central surface brightness of the lensing galaxy as $`\mu _{0,B}(AB)=13.57`$ mag arcsec<sup>-2</sup>. Then, as the surface brightness at $`R_e=0.\mathrm{}61\pm 0.\mathrm{}03`$ (F98) is $`\mu _{e,B}(AB)=\mu _{0,B}(AB)+8.33`$ and $`B(\mathrm{Vega})=B(AB)+0.12`$, we obtain the surface brightness at $`R_e`$ as $`\mu _{e,B}=22.02`$ mag arcsec<sup>-2</sup>. The mean surface brightness within $`R_e`$ is then estimated as $`SB_e=20.56`$ mag arcsec<sup>-2</sup> after correction for Galactic extinction $`E(BV)=0.016`$ mag (Schlegel, Finkbeiner, & Davis 1998) and an $`R_V=3.1`$ extinction curve. Alternatively, van de Ven, van Dokkum, & Franx (2003, hereafter vdV03) derived $`\mu _{e,B}=21.70`$ mag arcsec<sup>-2</sup> at the intermediate effective radius $`R_e`$ of $`0.\mathrm{}48`$ (Kochanek et al. 2000), giving $`SB_e=\mu _{e,B}1.39320.56`$ mag arcsec<sup>-2</sup>. As the small difference between these values results in only the difference of 0.2 to 0.3 in the mass-to-light ratio obtained below, we adopt the results of F98 in what follows.
For the lensing galaxy of B 2045$`+`$265, we adopt the values given in vdV03, i.e., $`R_e=0.\mathrm{}38_{0.\mathrm{}11}^{+0.\mathrm{}15}`$ and $`SB_e=\mu _{e,B}1.39319.11\pm 0.44`$ mag arcsec<sup>-2</sup>.
To derive a fiducial correction factor from the measured $`\sigma `$ in a specific aperture to $`\sigma _c`$ in equation (1), we utilize a fiducial model for internal mass distribution and stellar dynamics as explained in §4. Briefly saying, we adopt a single mass component represented by a power-law density profile with an index of $`\gamma ^{}`$ \[see equation (5)\] and the velocity dispersion of stars represented by a constant anisotropy parameter $`\beta `$ \[see equation (6)\]. For HST 14113$`+`$5211, we obtain $`\sigma _c=1.15\sigma `$ based on the most likely parameters of $`(\gamma ^{},\beta )=(1.93,0.14)`$ (see §5) thereby yielding $`\sigma _c=200.7`$ km s<sup>-1</sup>. For B 2045$`+`$265, we obtain $`\sigma _c=0.99\sigma `$ based on the most likely parameters of $`(\gamma ^{},\beta )=(1.66,0.10)`$, thereby yielding $`\sigma _c=210.3`$ km s<sup>-1</sup>.
Inserting into equation (1), we obtain $`\gamma _{FP}=8.87`$ for the lensing galaxy of HST 14113$`+`$5211 at $`z_L=0.464`$ and $`\gamma _{FP}=8.52`$ for the lensing galaxy of B 2045$`+`$265 at $`z_L=0.87`$. Comparison with $`\gamma _{FP}`$ at $`z=0`$ allows us to derive the difference in $`\gamma _{FP}`$, $`\mathrm{\Delta }\gamma _{FP}`$, and then $`\mathrm{\Delta }\mathrm{log}(M_{}/L_B)0.4\mathrm{\Delta }\gamma _{FP}/\beta _{FP}`$ provided $`\alpha _{FP}`$ and $`\beta _{FP}`$ are constant. We then obtain $`\mathrm{\Delta }\mathrm{log}(M_{}/L_B)=0.172`$ for the lensing galaxy of HST 14113$`+`$5211 and $`0.612`$ for the lensing galaxy of B 2045$`+`$265. The mass-to-light ratio at $`z>0`$ is given as,
$$\mathrm{log}(M_{}/L_B)_z=\mathrm{log}(M_{}/L_B)_0+\mathrm{\Delta }\mathrm{log}(M/L_B)$$
(2)
where $`(M_{}/L_B)_0=7.3\pm 2.1h_{65}`$ $`M_{}/L_{B,}`$ (Gerhard et al. 2001). We obtain $`(M_{}/L_B)_z=4.9\pm 1.4`$ $`M_{}/L_{B,}`$ for the lensing galaxy of HST 14113$`+`$5211 and $`1.8\pm 0.5`$ $`M_{}/L_{B,}`$ for the lensing galaxy of B 2045$`+`$265. As will be shown below, these values of $`M_{}/L_B`$ for both systems are in good agreement with those obtained from a joint lensing and dynamical analysis (see §5).
## 4 MASS MODEL AND STELLAR DYNAMICS
### 4.1 Lens models
Following TK04, we model the lensing galaxy as a singular isothermal ellipsoid (SIE: Kormann, Schneider & Bartelmann 1994). Note that the Einstein radius ($`R_E`$) and mass enclosed by the Einstein radius ($`M_E`$), both of which are the quantities required in the dynamical model of the lensing galaxy in the following sections, are very insensitive to the assumed mass model (Kochanek 1991; Koopmans & Treu 2004). We also allow for a constant external shear. The observed position of a lensing galaxy is taken by a lens position and we do not treat the lens position as a free parameter. Therefore, our lens model has five parameters (we follow the notations in TK04); the lensing strength ($`b_l=4\pi (\sigma _{\mathrm{SIE}}/c)^2D_{ls}/D_s`$, where $`\sigma _{\mathrm{SIE}}`$ is the one-dimensional velocity dispersion of SIE lens), axis ratio ($`q_l`$), position angle of the lens ($`\theta _l`$), strength of the external shear ($`\gamma _{\mathrm{ext}}`$), and its orientation ($`\theta _{\mathrm{ext}}`$). Also we treat the source position ($`\beta _x`$, $`\beta _y`$) as a free parameter.
We search for a set of model parameters that best reproduces the observed lens configuration. The best-fitting model parameters are summarized in Table 4. Figs 3 and 4 compare the observed image positions with the model images. The usual $`\chi ^2`$ values are 0.71 and 14 for HST14113$`+`$5211 and B 2045$`+`$265, respectively (for 1-$`\sigma `$ observational uncertainties of 0.03 arcsec) and the number of degrees of freedom is $`N_{dof}=1`$ (8 constraints and 7 parameters). As is evidently shown in Fig 3, for HST14113$`+`$5211, the model reproduces the lens configuration very well. The Einstein radius and mass enclosed by the Einstein radius are found to be $`R_E`$=5.29kpc and $`M(<R_E)=1.58\times 10^{11}M_{}`$ (for the adopted cosmological parameters), respectively. On the other hand, for B 2045$`+`$265, although the model nicely reproduces the cusp-lensing configuration (Schneider Ehlers & Falco 1992), the best-fitting model positions slightly deviate from the observed positions. Since the enclosed mass is not very sensitive to a detail lens model but is primarily determined by the image separation, the best-fitting model may give a good estimate of the the enclosed mass. The Einstein radius and mass enclosed are found to be $`R_E`$=9.19kpc and $`M(<R_E)=1.06\times 10^{12}M_{}`$, respectively, which we adopt in the following joint lensing and stellar dynamics analysis.
### 4.2 Mass model and stellar dynamics
Let the luminous-mass density and dark-matter density profiles be $`\rho _{lum}(r)`$ and $`\rho _{DM}(r)`$, respectively, provided that these are spherically symmetric. Two alternative mass models will be employed, following TK04.
First, we consider a two-component mass model, where the luminous mass and dark matter distribute differently:
$`\rho _{lum}(r)`$ $`=`$ $`{\displaystyle \frac{M_{}r_{}}{2\pi r(r+r_{})^3}}`$ (3)
$`\rho _{DM}(r)`$ $`=`$ $`{\displaystyle \frac{\rho _{DM,0}r_b^3}{r^\gamma (r^2+r_b^2)^{(3\gamma )/2}}},`$ (4)
where $`M_{}`$ is the total stellar mass and $`r_{}`$ denotes the scale length for the luminous matter. The profile $`\rho _{lum}(r)`$ corresponds to a Hernquist model (Hernquist 1990), reproducing the $`R^{1/4}`$ surface brightness profile with an effective radius of $`R_e=1.8153r_{}`$. For $`M_{}`$, we use the $`B`$-band total luminosity of the lensing galaxy, $`L_B`$, based on the mass-to-light ratio of the luminous matter, $`M_{}/L_B`$, as a model parameter. Our estimation of $`L_B`$ is given in Appendix, yielding $`L_B/L_{B,}=2.6\times 10^{10}`$ and $`1.3\times 10^{11}`$ ($`h=0.65`$) for the lensing galaxies of HST 14113$`+`$521 and B 2045$`+`$265, respectively. The dark-matter density profile, $`\rho _{DM}(r)`$, is determined by the scale length ($`r_b`$), density scale ($`\rho _{DM,0}`$), and inner slope ($`\gamma `$). The combination of all of these parameters will be constrained by the results of the lens fitting, $`R_E`$ and $`M(<R_E)`$, as obtained in §4.1.
Second, we consider a single component with power-law mass model:
$$\rho _{tot}(r)r^\gamma ^{},$$
(5)
where $`\gamma ^{}`$ denotes an effective slope.
The velocity distribution of the stars is also assumed to be spherically symmetric, such that the velocity dispersions in spherical coordinates $`(\sigma _r,\sigma _\theta ,\sigma _\varphi )`$ satisfy $`\sigma _\theta =\sigma _\varphi `$ (e.g., Binney & Tremaine 1987). We then use the parameter $`\beta (r)=1\sigma _\theta ^2/\sigma _r^2`$ describing the degree of velocity anisotropy. We employ the following two models for $`\beta (r)`$, the Osipkov-Merritt model with a parameter $`r_i`$ (referred to as Model A) and constant anisotropy model with a parameter $`\beta `$ (Model B).
$$\beta (r)=\{\begin{array}{cc}\frac{r^2}{r^2+r_i^2},\hfill & \text{for Model A}\hfill \\ const.(=\beta ),\hfill & \text{for Model B}\hfill \end{array}$$
(6)
Relevant parameters are $`r_i`$ and $`\beta `$.
## 5 Results of a Joint Lensing and Dynamical Analysis
Based on the models given in the previous section, we search for the best mass model for reproducing the observed velocity dispersion $`\sigma `$ using the joint lensing and dynamical analysis and also combining with the FP constraints. We especially focus on the best values of $`M_{}/L_B`$ for a two-component model, which are to be compared with those from the FP constraints. Also, the inner slope of the dark-matter halos, $`\gamma `$, are to be compared with the prediction of cosmological N-body simulations \[e.g., $`\gamma =11.5`$ (Navarro, Frenk, & White 1995; Moore et al. 1998; Fukushige & Makino 2003 and references therein)\]. For a single-component model, the derived slope $`\gamma ^{}`$ can be used to assess the isothermality of the lens density profile as usually adopted in other lensing work.
### 5.1 HST 14113$`+`$5211
First, we examine a two-component mass model. Figure 5 shows 68 % (1$`\sigma `$), 95 %, 99 %, and 99.9 % confidence levels of the likelihood contours in the plane of $`\gamma `$ and $`M_{}/L_B`$ when we adopt the velocity dispersion of $`\sigma =174\pm 20`$ km s<sup>-1</sup>. Dashed lines are derived from the joint lensing and dynamical analysis, while solid lines combine the additional constraints from the FP plane. For the lower panel (Model A) we set $`r_i=R_e`$ and $`r_b=R_E`$, and for the upper panel (Model B) we set $`\beta =0`$ and $`r_b=R_E`$. The likely $`\gamma `$ or $`M_{}/L_B`$ using these parameter sets will be utilized as a characteristic case in what follows, as the likelihood contours are found to remain basically the same in other parameter settings (TK04).
From the results based on the current joint analysis (dashed lines), the likely mass-to-light ratio appears to be in the range of $`4M_{}/L_B6`$ $`M_{}/L_{B,}`$: after marginalizing over $`\gamma `$, we obtain the most likely values of $`M_{}/L_B`$ ($`M_{}/L_{B,}`$), yielding $`4.4_{2.1}^{+2.0}`$ for Model A and $`5.6_{2.2}^{+2.2}`$ for Model B. As is evident, these values of $`M_{}/L_B`$ are virtually consistent with those obtained from the FP in §3 ($`4.9\pm 1.4`$ $`M_{}/L_{B,}`$). We also calculate the most likely fraction of the dark matter projected inside $`R_E`$ ($`1.4R_e`$), which is denoted as $`f_{DM}`$. Adopting the isotropic velocity model as a representative case, we obtain $`f_{DM}=0.47_{0.21}^{+0.21}`$, suggesting that about a half of the total mass derived from the lens model constitutes dark matter.
Turn next to the result from the joint analysis combined with the FP constraints (plotted by the solid lines). Since the above joint analysis gives $`M_{}/L_B`$ being in a good agreement with one from the FP, the additional constraint from the FP makes only a minor change in the preferred $`M_{}/L_B`$ value. It is found that marginalized constraints on $`M_{}/L_B`$ ($`M_{}/L_{B,}`$) are $`4.8_{1.2}^{+1.1}`$ for Model A and $`5.1_{1.1}^{+1.2}`$ for Model B. The most likely fraction of the dark matter projected inside $`R_E`$ (for the isotropic velocity case) reads $`f_{DM}=0.52_{0.12}^{+0.10}`$. The additional constraint from the FP improves the limit on $`\gamma `$. We obtained, after marginalizing over $`M_{}/L_B`$, a constraint on $`\gamma `$ to be $`\gamma <1.6`$ ($`\gamma <1.8`$) at 1-$`\sigma `$ for Model A (Model B).
Second, we examine a single-component mass model to set limits on the slope $`\gamma ^{}`$. Figure 6 shows 68 % (1$`\sigma `$), 95 %, 99 %, and 99.9 % confidence levels of the likelihood contours. We set $`r_b=3R_E`$ in this diagram, but the limits on $`\gamma ^{}`$ are found to be little affected as long as $`r_b3R_E`$. We estimate the most likely values of $`\gamma ^{}`$ for $`r_i=R_e`$ (Model A) and $`\beta =0`$ (Model B), yielding $`1.87_{0.09}^{+0.08}`$ and $`1.94_{0.09}^{+0.07}`$, respectively. Thus, the total density profile is well approximated as $`\rho _{tot}(r)r^2`$.
### 5.2 B 2045$`+`$265
For the lensing galaxy of B 2045$`+`$265, we show the results in Figure 7 and 8, which are to be compared with Figure 5 and 6, respectively, obtained for HST 14113$`+`$5211.
For a two-component mass model (Figure 7), the likely mass-to-light ratio appears to be in the range of $`0.7M_{}/L_B2`$ $`M_{}/L_{B,}`$: after marginalizing over $`\gamma `$, we obtain the most likely values of $`M_{}/L_B`$ ($`M_{}/L_{B,}`$) for Model A (Model B) as $`M_{}/L_B=0.8_{0.7}^{+0.5}`$ ($`1.2_{0.7}^{+0.5}`$) without the FP constraints. As is evident, these values of $`M_{}/L_B`$ are also consistent with those obtained from the FP in §3 ($`1.8\pm 0.5`$ $`M_{}/L_{B,}`$). Combining the FP constraint and after marginalizing over $`M_{}/L_B`$ we obtain $`\gamma <0.5`$ ($`\gamma <0.8`$) at $`1\sigma `$ for Model A (Model B). Thus the models prefer flatter dark matter inner slope than the case of HST 14113$`+`$5211.
We also obtain the most likely fraction of the dark matter projected inside $`R_E`$ ($`2.9R_e`$) for the isotropic velocity case, yielding $`f_{DM}=0.89_{0.04}^{+0.06}`$ (without the FP constraints) and $`0.86_{0.03}^{+0.05}`$ (with the FP constraints). Thus, in contrast to the case of HST 14113$`+`$5211, the total mass inside $`R_E`$ derived from the lens model is totally dominated by dark matter; some of this unseen mass component may be provided by the group of galaxies which the lensing galaxy belongs to.
For a single-component mass model (Figure 8), we obtain the most likely values of $`\gamma ^{}`$ for $`r_i=R_e`$ (Model A) and $`\beta =0`$ (Model B), yielding $`1.58_{0.09}^{+0.08}`$ and $`1.66_{0.08}^{+0.07}`$, respectively. Thus, the model prediction for the slope of the total density profile is systematically flatter than isothermal.
## 6 DISCUSSION AND CONCLUDING REMARKS
As explored in this work and also in TK04, the mass-to-light ratio for both of our targets based on the joint lensing and dynamical analysis is virtually consistent with that obtained from the FP constraint; the mass-to-light ratios at $`z_L=0.464`$ and 0.868 are systematically smaller than the current average value of $`(M_{}/L_B)_0=7.3\pm 2.1h_{65}`$ $`M_{}/L_{B,}`$ (Gerhard et al. 2001), implying the aging of stellar populations from these redshifts to the present day. For comparison with other sample lenses for which the similar analysis has been employed, we plot, in Figure 9, the redshift evolution of the stellar mass-to-light ratio for both of our targets (open circles) and the TK04 sample (solid circles). It is found that the current sample lenses follow the general redshift evolution of $`M_{}/L_B`$, as guided by the solid and dashed lines showing the average evolution of $`M_{}/L_B`$ derived from the FP of early-type lensing galaxies by TK04 and Rusin et al. (2003), respectively.
The inner slope of the dark-matter halos, $`\gamma `$, ranges from 0.0 to 1.8. More specifically, the most likely values of $`\gamma `$ (with the FP constraints) for HST 14113$`+`$5211 are $`0<\gamma <1.6`$ ($`0<\gamma <1.8`$) at $`1\sigma `$ for Model A (Model B), whereas for B 2045$`+`$265, we obtain $`0<\gamma <0.5`$ ($`0<\gamma <0.8`$) at $`1\sigma `$ for Model A (Model B). These values are consistent with the density slopes of dark halos predicted by cosmological N-body simulations, showing $`\gamma `$ of 1.0 to 1.5 (e.g., Navarro, Frenk, and White 1995; Moore et al. 1998; Fukushige & Makino 2003 and references therein), although the present-day inner slope of the dark-matter halos is affected by the effect of baryonic infall to some extent; the effect seems to be not so significant as obtained from other lens samples (TK04). It is also drawn that the inner slope of the dark-matter halos is systematically flatter than an isothermal profile.
The index for a single power-law model ($`\gamma ^{}`$) is almost 2 for HST 14113$`+`$5211, thereby suggesting that a singular isothermal model is a good fit to this lensing galaxy. For B 2045$`+`$265, $`\gamma ^{}`$ ranges from 1.5 to 1.7, which is systematically flatter than isothermal (Figure 8). This flat slope is related to the very small velocity dispersion of stars $`\sigma =213`$ km s<sup>-1</sup> compared with $`\sigma _{\mathrm{SIE}}=397`$ km s<sup>-1</sup>; a flatter slope of a dark halo than SIE is needed to reduce the radial gravitational force and thus the velocity dispersion of stars. Alternatively, some part of $`M(<R_E)`$ for B 2045$`+`$265 is provided by the group of galaxies, thereby causing a large value of $`\sigma _{\mathrm{SIE}}`$. This effect may partly contribute to a non-negligible scatter of $`\gamma ^{}`$ from an isothermal index (2) as is also reported in other lens samples (TK04).
## Acknowledgments
This work has been supported in part by a Grant-in-Aid for Scientific Research (15540241, 16740118 and 177401166827) of the Ministry of Education, Culture, Sports, Science and Technology in Japan.
## Appendix A Estimation of the Lens Luminosity in the B band
For the lensing galaxy of HST 14113$`+`$5211, F98 reported the F702W(AB) magnitude as $`20.78\pm 0.05`$ mag. Based on the K-correction using Coleman, Wu, & Weedman (1980) and the assumption of no galaxy evolution, F98 derived the absolute B magnitude as $`M_B(AB)=19.32+5\mathrm{log}h`$ for $`(\mathrm{\Omega }_0,\lambda _0)=(1,0)`$ and $`z_L=0.46`$. Then, we transform this magnitude into Vega-based B magnitude \[$`M_B(\mathrm{Vega})=M_B(AB)+0.12`$, Schmidt, Schneider, & Gunn 1995\] and consider the revised lens redshift of $`z_L=0.464`$ (Lubin et al. 2000) and cosmological parameters of $`(\mathrm{\Omega }_0,\lambda _0)=(0.3,0.7)`$ (Spergel et al. 2003). We thus obtain $`M_B=19.60+5\mathrm{log}h`$, giving the luminosity of $`L_B/L_{B,}=2.6\times 10^{10}`$ ($`h=0.65`$) for the lensing galaxy of HST 14113$`+`$5211.
For the lensing galaxy of B 2045$`+`$265, Fassnacht et al. (1999) reported the various infrared magnitudes in a $`1.\mathrm{}9`$ diameter aperture (corresponding to the size of the Einstein ring in their lens model). Based on the K-correction for an Sa galaxy as the lens appears to show its typical spectrum, they arrived at the rest-frame $`B`$-band luminosity of $`2.36\times 10^{10}h^2L_{}`$ in this aperture for $`(\mathrm{\Omega }_0,\lambda _0)=(1,0)`$. To derive the total $`B`$-band luminosity, we adopt the work of Rusin et al. (2003), where they obtained the intermediate-axis effective radius ($`R_e`$) determined by fitting the observed brightness distribution to a de Vaucouleurs profile. Using their $`R_e`$ of $`0.\mathrm{}38`$ and $`(\mathrm{\Omega }_0,\lambda _0)=(0.3,0.7)`$, we obtain the luminosity of $`L_B/L_{B,}=1.3\times 10^{11}`$ ($`h=0.65`$) for the lensing galaxy of B 2045$`+`$265. |
warning/0507/physics0507112.html | ar5iv | text | # Spectroscopic determination of the 𝑠-wave scattering lengths of 86Sr and 88Sr
## Abstract
We report the use of photoassociative spectroscopy to determine the ground state $`s`$-wave scattering lengths for the main bosonic isotopes of strontium, <sup>86</sup>Sr and <sup>88</sup>Sr. Photoassociative transitions are driven with a laser red-detuned by up to 1400 GHz from the $`{}_{}{}^{1}S_{0}^{}`$ \- $`{}_{}{}^{1}P_{1}^{}`$ atomic resonance at 461 nm. A minimum in the transition amplitude for <sup>86</sup>Sr at $`494\pm 5`$ GHz allows us to determine the scattering lengths $`610`$$`a_0<a_{86}<2300`$$`a_0`$ for <sup>86</sup>Sr and a much smaller value of $`1`$$`a_0<a_{88}<13`$$`a_0`$ for <sup>88</sup>Sr.
Photoassociative spectroscopy (PAS) of ultracold gases, in which a laser field resonantly excites colliding atoms to ro-vibrational states of excited molecular potentials, is a powerful probe of atomic cold collisions Weiner et al. (1999). Transition frequencies have been used to obtain dispersion coefficients of molecular potentials, which yield the most accurate value of the atomic excited-state lifetimeMcAlexander et al. (1995); Jones et al. (1996); Nagel et al. (2005). Transition amplitudes are related to the wave-function for colliding ground state atoms Napolitano et al. (1994); Julienne (1996), and can be used to determine the ground state $`s`$-wave scattering length Gardner et al. (1995); Abraham et al. (1997); Tiesinga et al. (1996); Williams et al. (1999); Degenhardt et al. (2003); Takasu et al. (2004).
The $`s`$-wave scattering length is a crucial parameter for determining the efficiency of evaporative cooling and the stability of a Bose-Einstein condensate (BEC). It also sets the scale for collisional frequency shifts, which can limit the accuracy and stability of atomic frequency standards.
The cold collision properties of alkaline-earth atoms like strontium, calcium, and magnesium, and atoms with similar electronic structure such as ytterbium, are currently the focus of intense study. These atoms posses narrow optical resonances that have great potential for optical frequency standards Katori et al. (2003); Takamoto and Katori (2003); Oates et al. (1999); Santra et al. (2004); Hong et al. (2004). Laser-cooling on narrow transitions is an efficient route to high phase-space density Ido et al. (2000); Mukaiyama et al. (2003), and a BEC was recently produced with ytterbium Takasu et al. (2003). Fundamental interest in alkaline-earth atoms is also high because their simple molecular potentials allow accurate tests of cold collision theory Machholm et al. (2001); Ciurylo et al. (2004); Montalvão and de Jesus Napolitano (2001).
PAS of calcium Degenhardt et al. (2003) and ytterbium Takasu et al. (2004) was recently used to determine $`s`$-wave scattering lengths of these atoms. This paper reports the use of PAS to determine the ground state $`s`$-wave scattering lengths for the main bosonic isotopes of strontium, <sup>86</sup>Sr and <sup>88</sup>Sr, which have relative abundances of 10% and 83% respectively. We find a huge scattering length for <sup>86</sup>Sr of $`610`$$`a_0<a_{86}<2300`$$`a_0`$. Appreciable uncertainty comes from the value of $`C_6`$ for the ground state potential. In contrast, for <sup>88</sup>Sr we find $`1`$$`a_0<a_{88}<13`$$`a_0`$. From the data, we also make an improved measurement Nagel et al. (2005) of the $`5s5p`$$`{}_{}{}^{1}P_{1}^{}`$ atomic lifetime ($`\tau =5.25\pm 0.01`$ ns). PAS results for <sup>88</sup>Sr, yielding a similar value of $`\tau `$ and a different value of $`a_{88}`$, were recently posted Yasuda et al. (2005).
For PAS of strontium, atoms are initially trapped in a magneto-optical trap (MOT) operating on the 461 nm $`{}_{}{}^{1}S_{0}^{}`$ \- $`{}_{}{}^{1}P_{1}^{}`$ transition, as described in Nagel et al. (2003, 2005). We are able to produce pure samples of each isotope from the same atomic beam due to the intrinsic isotope selectivity of a MOT. For <sup>86</sup>Sr and <sup>88</sup>Sr respectively, about $`7\times 10^7`$ and $`2.5\times 10^8`$ atoms are trapped and cooled to $`2`$ mK.
After this stage, the 461 nm laser-cooling light is extinguished, the field gradient is reduced to $`0.1`$ G/cm, and $`689`$ nm light for the $`{}_{}{}^{1}S_{0}^{}`$ \- $`{}_{}{}^{3}P_{1}^{}`$ intercombination-line MOT Katori et al. (1999) is switched on. This MOT consists of three retro-reflected beams, each with a diameter of 2 cm and intensity of $`400800`$$`\mu `$W/cm<sup>2</sup>. Initially, the frequency of the laser-cooling light is detuned from atomic resonance by about -1.3 MHz, and spectrally broadened with a $`\pm `$1.0 MHz sine-wave modulation.
During a 50 ms transfer and equilibration period, the field gradient and spectral modulation are linearly ramped to $`0.8`$ G/cm and $`\pm `$0.7 MHz. The detuning is ramped to about -0.9 MHz. This yields about $`4\times 10^7`$ <sup>86</sup>Sr atoms at a temperature of $`5\pm 1`$$`\mu `$K or $`1.5\times 10^8`$ <sup>88</sup>Sr atoms at a temperature of about $`8\pm 2`$$`\mu `$K. The peak density for both isotopes is about $`2\times 10^{11}`$ cm<sup>-3</sup>. The intercombination-line MOT parameters are then held constant during an adjustable hold time. In the absence of PAS, the lifetime of atoms in the trap is about 500 ms, limited by background gas collisions.
The size of the atom cloud, the number of atoms, and thus the peak density are primarily determined with absorption imaging using the $`{}_{}{}^{1}S_{0}^{}`$ \- $`{}_{}{}^{1}P_{1}^{}`$ transition. We image along the direction of gravity, and transverse $`1/\sqrt{e}`$ density radii are $`\sigma =250`$$`\mu `$m for <sup>86</sup>Sr and $`\sigma =400`$$`\mu `$m for <sup>88</sup>Sr. To obtain information on the cloud dimension along gravity Loftus et al. (2004), an additional camera monitors fluorescence perpendicular to this direction. The cloud is smaller by approximately a factor of two in this axis.
To excite photoassociative resonances, a PAS laser, tuned to the red of the atomic $`{}_{}{}^{1}S_{0}^{}`$ \- $`{}_{}{}^{1}P_{1}^{}`$ transition at 461 nm, is applied to the atoms during hold times of $`200800`$ ms. When the PAS laser is tuned to a molecular resonance, photoassociation provides a loss mechanism for the MOT, decreasing the number of atoms. The PAS laser has negligible effect on the atom cloud size or temperature. When the PAS light is tuned far from a molecular resonance, it has no effect on the rate of loss from the trap.
PAS light is generated from an extended cavity diode laser at 922 nm using second-harmonic generation in a linear enhancement cavity Bode et al. (1997). The laser linewidth is 80 MHz in the blue on a millisecond time scale, which contributes significantly to the observed PAS resonance linewidths. The laser frequency is measured with a wavemeter that is regularly calibrated with the cooling laser whose frequency is locked to the atomic $`{}_{}{}^{1}S_{0}^{}`$ \- $`{}_{}{}^{1}P_{1}^{}`$ transition. The resulting 1-sigma statistical uncertainty for frequency measurements of the PAS laser is 200 MHz. There is also a comparable systematic uncertainty due to the sensitivity of the wavementer to the alignment of the laser into the device.
The available PAS laser power is as high as 20 mW. Several beam geometries are used, but typically the PAS beam is retroreflected, with a 1/e<sup>2</sup> intensity radius of about $`w=1`$ mm, yielding a maximum intensity of 3 W/cm<sup>2</sup> on the atoms. Lower intensities are obtained using an AOM. For some data, a quarter-wave plate is inserted in the beam path after the first pass through the atoms to prevent the formation of a standing wave. This variation has no noticeable affect on the PAS intensity and frequency measurements.
The observed PAS spectrum is relatively simple because the bosonic isotopes of strontium lack hyperfine structure. In addition, at the low temperatures of the intercombination-line MOT, only $`s`$-wave collisions occur so only $`J=1`$ levels are excited. Ground state $`{}_{}{}^{1}S_{0}^{}`$ atoms collide on a $`{}_{}{}^{1}\mathrm{\Sigma }_{g}^{+}`$ potential, and, of the four states converging to the $`{}_{}{}^{1}S_{0}^{}`$ \+ $`{}_{}{}^{1}P_{1}^{}`$ asymptote Machholm et al. (2001), only the $`{}_{}{}^{1}\mathrm{\Sigma }_{u}^{+}`$ state is excited in this work. Figure 1 shows spectra recorded for <sup>86</sup>Sr in a region around 500 GHz to the red of the atomic transition. The variation in transition amplitude will be discussed below, but first we describe the method used to quantitatively analyze the spectra.
The atom density varies as $`\dot{n}=\beta (I,f)n^2\mathrm{\Gamma }n`$, where $`f`$ and $`I`$ are the laser frequency and intensity. This implies the number of atoms in the trap as a function of hold time, $`t`$, follows
$$N(t)=\frac{N_0\mathrm{e}^{\mathrm{\Gamma }\mathrm{t}}}{1+\frac{\mathrm{n}_0\beta (\mathrm{I}_{\mathrm{pk}},\mathrm{f})\zeta }{2\sqrt{2}\mathrm{\Gamma }}(1\mathrm{e}^{\mathrm{\Gamma }\mathrm{t}})},$$
(1)
where $`N_0`$ and $`n_0`$ are the number and peak density, respectively, at the beginning of the hold time. $`I_{pk}`$ is the peak laser intensity, and $`\zeta =w^2/(2\sigma ^2+w^2)`$ accounts for the laser-atom overlap. We approximate the density distribution as a Gaussian. The one-body loss rate ($`\mathrm{\Gamma }`$) and $`n_0`$ are fixed at values determined from independent measurements. Fit values of $`N_0`$ agree well with independent measurements and are a check of the method.
The photoassociative two-body loss rate, $`\beta `$, near resonance $`v`$, is approximated as
$$\beta (I_{pk},f)=\frac{2K_vI_{pk}\gamma _v}{\gamma }\frac{1}{1+4(ff_v)^2/\gamma ^2},$$
(2)
where $`f_v`$ is the center frequency for the transition. The experimental linewidth, $`\gamma 150`$ MHz, approximately equals the sum of the natural radiative linewidth for the transition, $`\gamma _v=61`$ MHz Nagel et al. (2005), and the measured laser linewidth. Including the factor $`\gamma _v/\gamma `$ in Eq. 2 accounts for broadening beyond $`\gamma _v`$McKenzie et al. (2002); Schloder et al. (2002). $`I_{pk}`$ is in units of mW/cm<sup>2</sup>, and $`K_v`$ is the collision rate constant, on resonance, for 1 mW/cm<sup>2</sup> intensity from an ideal laser with negligible linewidth. Difficulty in accurately determining the spatial dimensions of the atom cloud and in aligning the PAS laser beam on the atoms leads to systematic uncertainties of about a factor of two in measurements of $`K_v`$ (Figs. 2 and 3). Except for the most intense PAS transitions close to atomic resonance for <sup>86</sup>Sr, the linear variation with intensity used in Eq. 2 is a good approximation and $`K_v`$ is a constant for each transition.
From the observed transition frequencies we can obtain an accurate value of the atomic $`5s5p`$$`{}_{}{}^{1}P_{1}^{}`$ lifetime, $`\tau `$. For the region of the molecular potentials probed by PAS, the transition energies can be described with the semiclassical approximation Leroy and B.Bernstein (1970)
$`E(v)`$ $`=`$ $`DX_0(v_Dv)^6,`$
$`X_0`$ $`=`$ $`\left[{\displaystyle \frac{\mathrm{\Gamma }(4/3)}{2\sqrt{2\pi }\mathrm{\Gamma }(5/6)}}\right]^6{\displaystyle \frac{h^6}{\mu ^3C_3^2}},`$ (3)
where $`D`$ is the dissociation energy, $`\mu `$ the reduced mass, $`\lambda =461`$ nm, $`C_3=3\mathrm{}\lambda ^3/16\pi ^3\tau `$, and $`v_D`$ is a non-integer quantum number corresponding to a hypothetical level at the dissociation limit. Relativistic retardation, rotational energy, higher order dispersion terms, ground state potential curvature, thermal shifts, and all light-induced shifts are negligible at our level of accuracy. A fit to the <sup>86</sup>Sr data yields $`\tau =5.25\pm 0.01`$ ns. The <sup>88</sup>Sr data is not extensive enough to contribute to the determination.
While all the data in Figs. 2 and 3 are valuable, the detunings of the laser corresponding to minima of the PAS rate are particularly important. At these detunings, the Condon radius for the excitation matches a node of the ground-state wave function, which is equivalent to saying the overlap integral between the ground- and excited-state wavefunctions vanishes. Minima positions provide precise enough information about ground-state potentials and wavefunctions to determine $`s`$-wave scattering lengths to high accuracy. The minima positions are also independent of atom density and laser intensity calibrations, and for the ultracold gases used here, they are independent of $`T`$, the temperature of the atoms.
For excitation by a laser with negligible linewidth, the collision rate constant at ultra-low temperatures and on resonance is Napolitano et al. (1994)
$$K_v(T,I)=\frac{1}{hQ_T}_0^{\mathrm{}}𝑑ϵe^{ϵ/k_BT}\frac{\gamma _v\gamma _s(ϵ,J=0)}{(ϵ^2+(\gamma /2)^2)},$$
(4)
where $`Q_T=(2\pi \mu k_BT/h^2)^{3/2}`$, $`k_B`$ is the Boltzmann constant, $`\gamma =\gamma _v+\gamma _s(ϵ,J=0)`$ with $`\gamma _s(ϵ,J=0)`$ equal to the laser-stimulated width. At low laser intensities, $`\gamma _s(ϵ,J=0)`$ can be expressed using Fermi’s golden rule as $`\gamma _s(ϵ,J=0)=\pi Id^2/ϵ_0c`$. Here, $`ϵ_0`$ is the vacuum permittivity, $`c`$ the vacuum speed of light and $`d^2=|v|D(R)|ϵ|^2`$ where $`D(R)`$ is the molecular dipole transition moment connecting $`|v`$ and $`|ϵ`$, the excited vibrational wave function and the energy normalized ground continuum wave function respectively. Because only bound levels close to the potential dissociation limit are excited, $`\gamma _v`$ is independent of $`v`$ and equal to twice the atomic linewidth $`\gamma _{at}`$. By the same argument, $`D(R)`$ can be approximated as independent of $`R`$ Czuchaj et al. (2003) and is connected to the $`{}_{}{}^{1}S_{0}^{}`$ \- $`{}_{}{}^{1}P_{1}^{}`$ atomic dipole moment via $`D(R)=\sqrt{2}\alpha d_{at}`$. Using $`d_{at}=\sqrt{3\pi ϵ_0\mathrm{}\gamma _{at}\mathrm{\lambda ̄}^3}`$ and the line strength factor $`\alpha =\sqrt{2/3}`$ Machholm et al. (2001), we find $`D=3.6`$ a.u.
For our analysis, the inner part ($`R<19`$$`a_0`$) of ground- and excited-state potentials are formed by an experimental RKR potential Gerber et al. (1984) and ab initio potential Boutassetta et al. (1996) respectively. These are smoothly connected to the multipolar van der Waals expansion in $`C_n/R^n`$ at large internuclear separation. For the excited-state potential, only the $`C_3`$ term contributes, and we use the value determined above. For the ground-state potential, $`C_6`$, $`C_8`$ and $`C_{10}`$ terms are included Porsev and Derevianko (2002); Mitroy and Bromley (2003); c6n . Relativistic retardation effects in the asymptotic part of the excited-state potential are treated as described in Nagel et al. (2005). Wave functions are calculated using a full quantum calculation. For the bound vibrational levels, the Mapped Fourier Grid Method Kokoouline et al. (1999) has been used whereas the ground-state continuum wave function was calculated using a Numerov algorithm.
Large transition moments for <sup>86</sup>Sr allow us to experimentally characterize about 80 transitions for this isotope, extending to detunings as large as -1400 GHz (Fig. 2). This allows us to clearly identify minima in the transition moment at $`494\pm 5`$ GHz and $`1500`$ GHz, corresponding to Condon radii of $`62.6\pm 0.2`$$`a_0`$ and $`43`$$`a_0`$ respectively. To precisely determine the position of the minimum at $`494`$ GHz, we calculated a set of potential curves with different node positions by varying the position of the repulsive inner-wall. Using these curves we performed a least-squares fit to the data shown in the inset of Fig. 2. Quoted uncertainties are one standard deviation. An overall amplitude factor was also varied but the best values were always well within the experimental uncertainties discussed above. The determined position of the node was independent of the value of the ground state $`C_6`$ coefficient used. It also did not change significantly if only data from -600 to -400 GHz was fit.
The Sr ground state potential determined from the node position and the ground state $`C_6`$ coefficient accurately describe the collisional properties of the system. This allows us to determine the $`s`$-wave scattering length $`a_{86}`$ from the calculated zero-energy continuum wave function. However, we must carefully address the uncertainty in $`a_{86}`$ arising from uncertainty in $`C_6`$. The best values in the literature are a semiempirical method yielding $`C_6=3250`$ a.u. Mitroy and Bromley (2003), and a relativistic many-body calculation of $`C_6=3170`$ a.u. Porsev and Derevianko (2002), although a reevaluation of Porsev and Derevianko (2002) inputting a more recent value of the $`5s5p`$$`{}_{}{}^{1}P_{1}^{}`$ lifetime Yasuda et al. (2005) predicts $`C_6=3103(7)`$ a.u. c6n . If we take the latter value, and a minimum at $`494\pm 5`$ GHz, we find $`a_{86}=1450_{370}^{+850}`$$`a_0`$. ($`a_{86}`$ would diverge for a minimum position of $`482`$ GHz.) This large scattering length indicates a bound level in the ground state potential very near threshold, which implies that $`a_{86}`$ is also very sensitive to $`C_6`$. For $`C_6=3170`$ a.u. and $`C_6=3240`$ a.u. we calculate $`a_{86}=940_{170}^{+279}`$$`a_0`$ and $`a_{86}=700_{90}^{+140}`$$`a_0`$ respectively. The spread in $`C_6`$ values makes it difficult to quote a rigorous statistical best value and uncertainty for $`a_{86}`$. Taking the extremes of the values found above, including spread in $`C_6`$ and one-standard-deviation variation in node position, we find $`610`$$`a_0<a_{86}<2300`$$`a_0`$.
PAS transitions in <sup>88</sup>Sr are significantly weaker than in <sup>86</sup>Sr, and only 50 transitions were characterized, extending to binding energies of -174 GHz (Fig. 3). We lack the sensitivity required to observe transitions to the red of the first minimum in the transition amplitude and thus cannot independently determine its position with high accuracy. However, the potential found using <sup>86</sup>Sr should also determine the wavefunctions for <sup>88</sup>Sr. Because of the heavy mass of strontium, effects due to the breakdown of the Born-Oppenheimer approximation should be negligible as is the case for rubidium van Kempen et al. (2002). Predicted photoassociation rates agree with measurements within experimental uncertainties (Fig. 3), and we calculate $`a_{88}=10\pm 3`$ a<sub>0</sub> and a wave-function node at $`76\pm 0.4`$ a<sub>0</sub> for $`C_6=3103`$ a.u. For $`C_6=3240`$ a.u., $`a_{88}=2\pm 3`$ a<sub>0</sub>. This yields $`1`$ a$`{}_{0}{}^{}<a_{88}<13`$ a<sub>0</sub>.
In conclusion, we determined the scattering lengths of <sup>86</sup>Sr and <sup>88</sup>Sr from PAS spectra. Our analysis uses full quantum calculations and is not based on semi-classical arguments Boisseau et al. (2000); Yasuda et al. (2005) that can be invalid if the scattering length becomes divergent (as in <sup>86</sup>Sr), small (as in <sup>88</sup>Sr), or negative. The large positive value for <sup>86</sup>Sr is very promising for the realization of a BEC through evaporative cooling in an optical trap. Such a condensate would have the advantage that losses due to inelastic processes would be minimal since there is no hyperfine interaction and spin-exhange collisions would not take place. The lifetime of a <sup>86</sup>Sr condensate would be influenced mostly by three-body recombination. BEC of <sup>88</sup>Sr will be a significant challenge Ido et al. (2000) because of the small scattering length, but it may allow studies of a very weakly interacting quantum gas with behavior similar to a hydrogen condensate Fried et al. (1998).
This research was supported by the Welch Foundation (Grant # C-1579), Office for Naval Research, and David and Lucille Packard Foundation. The authors are grateful to A. R. Allouche for providing them the Sr<sub>2</sub> ab initio potentials,and to E. Tiesinga and P. Julienne for valuable discussions. |
warning/0507/math0507590.html | ar5iv | text | # Index in K-theory for Families of fibred cusp operators
## Introduction
The index theorem of Atiyah, Patodi and Singer gives a formula for the index of a Dirac operator on a compact manifold with boundary with boundary condition given by projection onto the range of the positive part of the boundary Dirac operator. Versions of this result for families, with the formula being in cohomology for the Chern character of the (virtual) index bundle, were given by Bismut-Cheeger and the first author and Piazza . Here we formulate an index theorem in K-theory in the wider context of the algebras of pseudodifferential of fibred-cusp type, so generalizing the K-theory formulation of Atiyah and Singer in the boundaryless case. Our result specializes to give an index theorem in K-theory for the families of Dirac operators in the earlier contexts cited above. In a subsequent paper we will show how to derive a formula for the Chern character of the index class, reducing to the known formula in the Dirac case. There is also a relation, discussed below, with the ‘direct’ proof of the theorem of Atiyah, Patodi and Singer given by Dai and Zhang in .
We consider a smooth fibration of compact manifolds where the fibres are manifolds with boundary, with the boundary possibly carrying a finer fibration. With the addition of a normal trivialization of the boundary along the fibres we call this a fibration with fibred cusp structure. For such objects we formulate a notion of K-theory, denoted $`\mathrm{K}_{\mathrm{\Phi }\mathrm{cu}}(\varphi )`$ (where $`\varphi `$ is the overall fibration and $`\mathrm{\Phi }`$ the boundary fibration) as the stable homotopy classes of the Fredholm families of corresponding ‘fibred-cusp’ pseudodifferential operators (introduced by Mazzeo and the first author in ) on the fibres. In the boundaryless case this reduces to the compactly supported K-theory of the fibre cotangent bundle as in . The analytic index arises from the realization as Fredholm operators either through perturbation to make the null spaces have constant rank or through Kasparov’s bivariant K-theory. In close analogy with the boundaryless case we define a topological index map by using an embedding of the fibration in the product of the base and a ball and then we show the equality of analytic and topological index homomorphisms
###### Theorem 1.
For a fibration with fibred cusp structure, the analytic and topological index maps, to K-theory of the base,
(1)
are equal.
We also give an analogue of Atiyah’s Poincaré duality in K-theory.
###### Theorem 2.
For a fibration with fibred cusp structure, there is a natural ‘quantization’ isomorphism
(2)
$$\mathrm{quan}:\mathrm{K}_{\mathrm{\Phi }\mathrm{cu}}(\varphi )\mathrm{KK}_B^0(𝒞_\mathrm{\Phi }(M),𝒞(B))$$
where $`𝒞_\mathrm{\Phi }(M)`$ is the $`C^{}`$ algebra of those continuous functions on the total space of the fibration which are constant on the fibres of $`\mathrm{\Phi }`$ on the boundary.
A brief review of $`KK`$-theory and a definition of $`\mathrm{KK}_B^0(𝒞_\mathrm{\Phi }(M),𝒞(B))`$ can be found in section 4 below. The ‘cusp’ case of these results (meaning $`\mathrm{\Phi }=\varphi =\varphi |_M`$ and in which case we write the K-group as $`\mathrm{K}_{\mathrm{cu}}(\varphi ))`$ applies to give a families index theorem in K-theory for problems of Atiyah-Patodi-Singer type.
###### Theorem 3.
Let $`ð`$ be a family of Dirac operators associated to a unitary Clifford module for a family of metrics of product type near the boundary of the fibres of a fibration and suppose $`P`$ is a spectral section for the boundary Dirac operator, then $`(ð,P)`$ defines a class $`[(ð,P)]\mathrm{K}_{\mathrm{cu}}(\varphi )`$ and
$$\mathrm{ind}(ð,P)=\mathrm{ind}_a([ð,P)]).$$
This is a refinement at a K-theoretic level of the index theorem of for families Dirac operators with Atiyah-Patodi-Singer boundary conditions. Note that *only* the cusp case is needed for this application, so the proof requires only a relatively small part of the discussion in the body of the paper, in particular Sections 68 are not required for this.
Since we generalize it here, let us briefly recall the families index in the boundaryless case. The index theorem in K-theory of Atiyah and Singer takes the form of the equality of an analytically defined and a topologically defined index for a family of elliptic pseudodifferential operators $`P\mathrm{\Psi }^m(M/B;𝔼)`$ on the fibres of a fibration
(3)
where $`𝔼=(E_+,E_{})`$ is a $`_2`$-graded complex vector bundle over the total space, $`M,`$ of the fibration and $`P`$ maps sections of $`E_+`$ to sections of $`E_{}.`$ The analytic index is the element in $`\mathrm{K}(B)`$ which is the formal difference of vector bundles
(4)
$$\mathrm{ind}_\text{a}(P)=[\mathrm{null}(P+A)\mathrm{null}(P^{}+A^{})]\mathrm{K}(B)$$
for a perturbation $`A\mathrm{\Psi }^{\mathrm{}}(M/B;𝔼)`$ such that the null spaces have constant rank (and for any choice of data leading to the adjoints). Such a perturbation always exists and $`\mathrm{ind}_\text{a}(P)\mathrm{K}(B)`$ is independent of choices. The vanishing of the index is equivalent to the existence of such a perturbation with $`P+A`$ invertible. The symbol of the family $`P`$ defines an element in the (compactly supported) K-theory of the fibre cotangent bundle, $`[\sigma (P)]\mathrm{K}_\text{c}(T^{}(M/B))`$ and the analytic index of $`P`$ depends only on $`[\sigma (P)].`$ Since all K-classes with compact support on $`T^{}(M/B)`$ arise in this way, the analytic index gives a map
(5)
$$\mathrm{ind}_\text{a}:\mathrm{K}_\text{c}(T^{}(M/B))\mathrm{K}(B).$$
Alternatively, the analytic index may be defined via the bivariant K-theory of Kasparov, consisting of equivalence classes of almost-unitary Fredholm modules. Thus, the choice of a selfadjoint family $`Q\mathrm{\Psi }^m(M/B;E_+),`$ such that $`Q^2P^{}P\mathrm{Id}`$ is smoothing, fixes a class
(6)
$$\left[\left(\begin{array}{cc}0& QP^{}\\ PQ& 0\end{array}\right)\right]\mathrm{KK}_B^0(𝒞(M),𝒞(B))$$
which also depends only on the class of the symbol, i.e. defines a homomorphism which combined with the natural push-forward map again gives the analytic index
(7)
$$\text{}.$$
In fact the map on the left is an isomorphism which is a realization of Atiyah’s map from elliptic pseudodifferential operators to K-homology, i.e. is a form of Poincaré duality.
The topological index is defined as a Gysin map, using Bott periodicity. By a standard generalization of Whitney’s embedding theorem , the fibration (3) may be embedded as a subfibration of a real vector bundle $`V`$ over $`B`$ (indeed the bundle may be taken to be a product); the K-theory of $`T^{}(V/B)`$ is then canonically isomorphic to the K-theory of the base. The composite map
(8)
$$\text{}.$$
is the ‘topological index’ and is again independent of choices. The index theorem of Atiyah and Singer , in K-theory is the equality of these two maps; the one obtained by ‘quantizing’ symbols by use of pseudodifferential operators, the other by ‘trivializing’ the symbols using embeddings.
In this paper the corresponding problem is considered for fibred-cusp pseudodifferential operators. The full case is discussed below, initially the discussion is restricted to the cusp algebra; this indeed is the special case which is most closely related to the index theorem of Atiyah, Patodi and Singer. This relationship itself is made precise below following the discussion of the pseudodifferential index theorem.
The algebra of cusp pseudodifferential operators $`\mathrm{\Psi }_{\mathrm{cu}}^m(Z)`$ on a compact manifold with boundary has properties similar to those of the usual algebra on a compact manifold without boundary $`Z;`$ it is discussed extensively in . The definition of cusp operators depends on the choice of a boundary defining function $`x𝒞^{\mathrm{}}(Z)`$, that is, a nonegative function which is zero on $`Z`$, positive everywhere else and such that $`dx`$ is non-zero on $`Z`$. Given such a defining function, consider the Lie algebra of cusp vector fields. These are arbitrary smooth vector fields in the interior which, near the boundary are of the form
$$ax^2_x+\underset{j=1}{\overset{m}{}}a_j_{z_j},a,a_j𝒞^{\mathrm{}}(Z)$$
where $`(x,z)`$ are coordinates near $`Z.`$ The cusp differential operatorsform the universal enveloping algebra of this Lie algebra (as a $`𝒞^{\mathrm{}}(Z)`$-module). By microlocalization this algebra is extended to the cusp pseudodifferential operators. A typical example of cusp differential operator is obtained by considering the Laplacian associated to a Riemannian metric $`g`$ which in a collar neighborhood of the boundary $`Z\times [0,1)_xZ`$ takes the form
$$g=\frac{dx^2}{x^4}+h,h𝒞^{\mathrm{}}(Z\times [0,1)_x;T^{}(Z)T^{}(Z)).$$
The algebra of cusp pseudodifferential operators is closely related to the b-pseudodifferential algebra but has the virtue of admitting a $`𝒞^{\mathrm{}}`$ functional calculus. The main difference, compared to the boundaryless case and as far as Fredholm properties are concerned, is that as well as a symbol map in the usual sense, taking values in functions homogeneous of degree $`m`$ on the cusp cotangent bundle, $`\sigma _m:\mathrm{\Psi }_{\mathrm{cu}}^m(Z)S_{\mathrm{cu}}^m(Z),`$ there is a non-commutative ‘indicial’ (or normal) homomorphism taking values in suspended families of pseudodifferential operators on the boundary
(9)
$$N:\mathrm{\Psi }_{\mathrm{cu}}^m(Z)\mathrm{\Psi }_{\mathrm{sus}}^m(Z);$$
the suspended algebra consists of pseudodifferential operators with a symbolic parameter representing the dual to the normal bundle to the boundary. As already noted, the cusp algebra is not quite naturally associated to a compact manifold with boundary but is fixed by the choice of a trivialization of the normal bundle to the boundary. This may be thought of as a residue of the product-type structure in the theorem of Atiyah, Patodi and Singer. A cusp pseudodifferential operator is Fredholm on the (weighted) cusp Sobolev spaces if and only if *both* the symbol and the normal operator are invertible; we describe such an operator as ‘fully elliptic.’ The data consisting of (compatible) pairs $`\sigma _m(P)`$ and $`N(P)`$ constitutes the *joint symbol*, so an operator is fully elliptic when its joint symbol is invertible.
Following the model of the theorem of Atiyah and Singer described above we consider a fibration as in (3) where now the model fibre, $`Z,`$ is a compact manifold with boundary. We define analogues of the objects described above in the boundaryless case including a ‘symbolic’ K-group $`\mathrm{K}_{\mathrm{cu}}(\varphi )`$ as the set of stable homotopy classes of compatible and invertible joint symbols. The analytic and topological indexes are homomorphisms from this group into the topological K-group of the base.
The definition of the analytic index is a straightforward extension of the boundaryless case. That is, for a fully elliptic family of cusp pseudodifferential operators $`P\mathrm{\Psi }_{\mathrm{cu}}^m(M/B;𝔼)`$ on the fibres of the fibration there exists a perturbation by a family of smoothing operators supported in the interior, such that the null bundle has constant rank and then (5) again fixes an element $`\mathrm{ind}_\text{a}(P)\mathrm{K}(B)`$ which is independent of the perturbation. In fact it only depends on the stable homotopy class of the joint symbol, within invertible symbols, and so defines the top map in (1) in this case.
The symbol, as opposed to the joint symbol, homomorphism leads to a short exact sequence
(10)
$$\text{}=\mathrm{K}_\text{c}(T^{}(M/B))$$
where $`\overline{{}_{}{}^{\mathrm{cu}}T_{}^{}(M/B)}`$ is the radial compactification of the fibrewise cusp cotangent bundle $`{}_{}{}^{\mathrm{cu}}T_{}^{}(M/B)`$(which is isomorphic to $`T^{}(M/B))`$ . The image group can also be interpreted as the compactly supported ‘absolute’ K-theory of $`{}_{}{}^{\mathrm{cu}}T_{}^{}(M/B)`$– the more intricate notation here emphasizes that it is ‘relative’ to fibre infinity but ‘absolute’ as regards the boundary of $`M.`$ The first map in (10) represents the inclusion of the fully elliptic operators of the form $`\mathrm{Id}+\mathrm{\Psi }_{\mathrm{cu}}^{\mathrm{}}(M/B;E).`$ The exactness of (10) follows from proposition 2.2 and theorem 5.2 in . In fact it is shown there that any family of elliptic operators $`P\mathrm{\Psi }_{\mathrm{cu}}^m(M/B;𝔼),`$ so only assuming the invertibility of the symbol family, $`\sigma (P),`$ can be perturbed by an element of $`\mathrm{\Psi }_{\mathrm{cu}}^{\mathrm{}}(M/B;𝔼)`$ to be invertible (hence of course fully elliptic). This leads to a splitting of the sequence (10) which we can write
(11)
In this sense the index element of $`\mathrm{K}(B)`$ is a ‘difference class’ measuring the twisting of the given fully elliptic family relative to the invertible perturbation; since the index map from homotopy classes of fully elliptic families with a fixed elliptic symbol to $`\mathrm{K}(B)`$ is an isomorphism, the K-theory of the base can be represented by the fully elliptic perturbations of any one elliptic family. Were it easy to determine the map $`\mathrm{inv},`$ i.e. to find an invertible family corresponding to a given symbol, the index problem would be much simpler!
As in the boundaryless case, this analytic index map can also be defined via bivariant K-theory. Let $`𝒞_{\mathrm{cu}}(M)𝒞(M)`$ be the $`C^{}`$ subalgebra of the continuous functions on $`M`$ consisting of those which are constant on the boundary of each fibre of $`\varphi ;`$ thus there is a short exact sequence of $`C^{}`$ algebras
(12)
where $`𝒞_0(M)`$ is the $`C^{}`$ algebra of continuous functions on $`M`$ vanishing on the boundary and $`R`$ is restriction to the boundary. The quantization of an invertible joint symbol to a Fredholm cusp pseudodifferential operator (and choice of parametrix) then gives an alternative definition of the analytic index as in (6), (7):
(13)
The quantization map here is an isomorphism which we can interpret as Poincaré duality. In fact we show that there is a commutative diagram
(14)
Here the left isomorphism is the natural identification of the KK group with K-theory and the right isomorphism is an absolute version of Atiyah’s isomorphism as discussed in ; it follows that the central map is also an isomorphism.
To complete the analogy with the Atiyah-Singer theorem we proceed to define a ‘topological’ index using embeddings of the fibration and then prove the equality of the two index maps. In this case, because of the non-commutative structure of the definition of elements of $`\mathrm{K}_{\mathrm{cu}}(\varphi ),`$ there is more of an analytic flavour to the definition of this ‘topological index’ than in the boundaryless case.
The main step in defining $`\mathrm{ind}_\text{t}`$ is to show that for an embedding of $`\varphi :MB`$ as a subfibration of $`\stackrel{~}{\varphi }:\stackrel{~}{M}B`$ there is a natural lifting map
(15)
$$(\stackrel{~}{\varphi }/\varphi )^!:\mathrm{K}_{\mathrm{cu}}(\varphi )\mathrm{K}_{\mathrm{cu}}(\stackrel{~}{\varphi })$$
corresponding to tensoring with the ‘Bott element’ for the normal bundle of the smaller fibration. Our construction is closely related to that of Atiyah and Singer in in the boundaryless case, but is of necessity more intricate since we need to preserve the invertibility of the indicial family, not just symbolic ellipticity. For this reason we replace the construction in by a slightly different one involving pseudodifferential operators ‘of product type.’ In view of the identification above, this can also be considered as a lifting construction for KK theory, giving a map, which we believe but do not show, is an explicit realization of the map dual to the restriction $`R:𝒞_{\mathrm{cu}}(\stackrel{~}{M})𝒞_{\mathrm{cu}}(M)`$
(16)
$$R^{}:\mathrm{KK}_B^0(𝒞_{\mathrm{cu}}(M),𝒞(B))\mathrm{KK}_B^0(𝒞_{\mathrm{cu}}(\stackrel{~}{M}),𝒞(B)).$$
It is important for the subsequent computation of the Chern character that this map is given by an explicit smooth construction with the corresponding $`\mathrm{K}_{\mathrm{cu}}(\varphi ).`$ For the proof of the index theorem it is essential that (15) be consistent both with the index and with the symbolic lifting construction of , which is to say that it leads to two commutative diagrams (one for the left-directed and one for the right-directed arrows)
(17)
There is always an embedding of the fibration of compact manifolds with boundary as a subfibration of the product $`\pi _N:B\times 𝔹^NB`$ of the base with a ball $`𝔹^N`$ of sufficiently large dimension $`N`$ and (15) allows $`\mathrm{K}_{\mathrm{cu}}(\varphi )`$ to be mapped into the group for such a product
(18)
$$\mathrm{K}_{\mathrm{cu}}(\varphi )\mathrm{K}_{\mathrm{cu}}(\pi _N),N\text{ large.}$$
The K-groups for these products may be computed directly (see section 12) and for odd dimensions there is a ‘Thom isomorphism’
$$\mathrm{Thom}:\mathrm{K}_{\mathrm{cu}}(\pi _{2N+1})\mathrm{K}(B).$$
Hence there is a well-defined ‘topological index map’
(19)
$$\mathrm{ind}_\text{t}=\mathrm{Thom}(\pi _{2N+1}/\varphi )^!:\mathrm{K}_{\mathrm{cu}}(\varphi )\mathrm{K}(B)$$
and this completes the constructions of the ingredients in the statement of Theorem 1 in the cusp case.
The more general case of *fibred cusp* operators is similar to, but a little more complicated than, the cusp operators discussed above. These algebras of operators on a compact manifold correspond to the choice of a fibration of the boundary and the behaviour of the boundary becomes more ‘commutative’ as the fibres become smaller – that is, the cusp case is the most non-commutative. Thus $`Z,`$ the typical fibre, is a compact manifold with boundary, $`Z,`$ which carries a distinguished fibration
(20)
where $`Y`$ and $`X`$ are compact manifolds without boundary. Associated with this structure, and a choice of trivialization, along the fibres of $`\psi ,`$ of the normal bundle to the boundary, is an algebra of ‘fibred cusp pseudodifferential operators’ $`\mathrm{\Psi }_{\psi \mathrm{cu}}^{}(Z)`$ introduced in . The cusp case corresponds to the extreme case of the one-fibre fibration $`Y=`${pt}. The other extreme case, choosing the point fibration of the boundary, $`Y=Z,`$ $`\mathrm{\Phi }=\mathrm{Id},`$ is the ‘scattering case’ in which the index theorem is reducible directly to the usual Atiyah-Singer setting by a doubling construction (briefly discussed in and below); this case is the simplest in most senses. It is discussed separately below, since a special case of the index theorem for scattering operators is used to handle the families index for perturbations of the identity in the general case. It is perhaps helpful to think of the fibred cusp operators as associated to the topological space $`Z/\psi `$ in which the boundary is smashed to the base $`Y.`$ Thus, for the cusp calculus, the boundary is smashed to a point whereas for the scattering calculus it is left unchanged.
Here we consider locally trivial families of such structures. Thus, suppose that $`M`$ is a manifold with boundary which admits a fibration over a base $`B`$ with typical fibre $`Z`$ which can be identified with the manifold in (20). We suppose that the boundary of $`M`$ has a finer fibration than over the base $`B`$ giving a commutative diagram
(21)
in which $`X`$ is the typical fibre of a fibration of $`M`$ over $`D`$ and $`Y`$ is the typical fibre of a fibration of $`D`$ over $`B.`$ We call such a pair of fibrations, together with a choice of normal trivialization, a fibration with fibred cusp structure. Associated to this geometry is an algebra of fibred cusp pseudodifferential operators acting on the fibres of $`M`$ over $`B,`$ which we shall denote $`\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^{}(M/B).`$ An element is a family parameterized by $`B`$ with each operator related to the fibre (in $`M)`$ above that point with its boundary smashed to the fibre of $`D`$ above that point of $`B`$ by the fibre of $`\mathrm{\Phi }.`$
In this more general setting we obtain similar results to those described above for the cusp calculus. First we define an abelian group, $`\mathrm{K}_{\mathrm{\Phi }\mathrm{cu}}(\varphi ),`$ with elements which are the stable homotopy classes of joint elliptic symbols, and a corresponding odd K-group, $`\mathrm{K}_{\mathrm{\Phi }\mathrm{cu}}^1(\varphi ).`$ The definition of analytic index maps, taking values respectively in $`\mathrm{K}(B)`$ and $`\mathrm{K}^1(B)`$ is essentially as above. We also give an analogue of (10) but now as a 6-term exact sequence
(22)
where $`\sigma _0,`$ $`\sigma _1`$ are maps related to the symbol, $`I_0,`$ $`I_1`$ are forms of the index map of Atiyah-Singer and $`i_0,`$ $`i_1`$ correspond to inclusion of perturbations of the identity by operators of order $`\mathrm{}.`$ We show below that this is isomorphic to the corresponding sequence in KK-theory arising from the short exact sequence of $`C^{}`$ algebras (replacing (12))
(23)
$$\begin{array}{c}𝒞_0(M)𝒞_\mathrm{\Phi }(M)𝒞(D),\\ 𝒞_\mathrm{\Phi }(M)=\left\{f𝒞(M);f|_M=\mathrm{\Phi }^{}g,g𝒞(D)\right\},\end{array}$$
namely
(24)
where the four isomorphisms between the corresponding spaces on the left and right in (24) and (22) are the Poincaré duality maps of Atiyah, as realized in Kasparov’s KK theory. The remaining two isomorphisms are given by quantization maps in the fibred-cusp calculus amounting as before to Poincaré duality; this is Theorem 2 in the general case. Note that the fact that the symbol map in (22) is not (in general) surjective shows that in some sense the ‘universal case’ is that of the cusp calculus since only through it can *every* elliptic symbol be quantized to a Fredholm family.
This universality appears explicitly in the form of a natural map in which the finer fibration of the boundary is ‘forgotten’
(25)
through which the index factors; it is defined through an adiabatic limit. We then define the topological index as the composite with the topological index in the cusp case. It is also possible to proceed more directly, through an appropriate embedding of the fibration.
Since we make substantial use below of various classes (‘calculi’) of pseudodifferential operators we have tried to use a uniform notation. In the families index theorem of Atiyah and Singer the quantization map giving the analytic index is in terms of families of pseudodifferential operators on the fibres of a fibration. We use what is the standard notation for these, $`\mathrm{\Psi }^m(M/B;𝔼),`$ except that $`𝔼=(E_+,E_{})`$ is a $`_2`$-graded bundle and the operators act from $`E_+`$ to $`E_{}.`$ We will sometime need to consider tensor products of the form
$$𝔼_+F=(E_+F,E_{})$$
where $`F`$ is a bundle which is not $`_2`$-graded. We also use $`_2`$-graded bundle notation for the operators in the quantization map (2), on a fibration with fibred cusp structure
$$\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^m(M/B;𝔼)$$
where the suffix alone indicates the ‘uniformity type’ at the boundary. Several more such classes arise below. First the indicial operator for families as in (9) takes values in the (singly) suspended algebra for which we use the suffix ‘sus.’ These are the pseudodifferential operators on $`\times Z,`$ or for families on the fibres of $`\times M,`$ which are translation-invariant and rapidly decreasing at infinity in the real factor – taking the Fourier transform therefore gives a parameterized family but the parameter enters as a symbolic variable. So more generally we use notation such as
$$\mathrm{\Psi }_{\psi \mathrm{sus}}^m(X),\mathrm{\Psi }_{\psi \mathrm{p}}^{m^{},m}(X),\mathrm{\Psi }_{\psi \mathrm{ad}}^m(X)$$
to denote, respectively, pseudodifferential operators which are suspended with respect to the cotangent variables of a fibration, pseudodifferential operators which are of product type with respect to a fibration and pseudodifferential operators which depend adiabatically on a parameter (and pass from operators on the total space to fibre operators in the limit in the parameter). The general approach to pseudodifferential operators, by defining them in terms of classical conormal distributions on some blown up version of the product space, allows these types to be combined where required. Thus we use adiabatic families of fibred cusp operators to pass from one boundary fibration to another in the definition of (25) and product-type cusp pseudodifferential operators in the lifting construction – these have product-type suspended operators as normal operators with corresponding notation.
We do not put the $`0`$’s at the end of short exact sequences.
In §1 the analytic index, in K-theory, of fully elliptic fibred cusp pseudodifferential operators is described as a map from the K-group associated to invertible full symbols. In the special case of the scattering structure, the index theorem is derived from the Atiyah-Singer index theorem in §2 and used to discuss families which are perturbations of the identity for general fibred cusp structures in §3. The quantization map, to KK-theory, is introduced in §4 and shown to be an isomorphism for the cusp structure in §5. The 6-term symbol sequences in K-theory and KK-theory are described in §6 and related in §7, resulting in the proof of Theorem 2. The map (25) is constructed in §8 and a result on the extension of fibred cusp structures is contained in §9 and the related multiplicativity and lifting properties are examined in §§10,11. In §12 the model cases are analyzed and used to define the topological index map and to prove Theorem 1 in §13. The application of the cusp case to families of Atiyah-Patodi-Singer type is made in §14 and Theorem 3 is proved there. In the appendices the various classes of pseudodifferential operators appearing in the body of the paper are described, including product-type, fibred cusp and adiabatic algebras and their combinations.
## 1. Analytic index
As already briefly described above, the general setting of this paper is a compact manifold with boundary, $`M,`$ with a fibration $`\varphi :MB`$ over a compact manifold usually without boundary; we also assume, without loss of generality, that the base is connected. In fact it is convenient at various points to allow the base (and hence also the total space) to be a manifold with corners, to allow especially products with intervals. We still treat $`M`$ as a manifold with boundary, in that ‘the’ boundary is then the union of the boundary faces of the fibres. Thus, $`\varphi `$ is a smooth surjective map with surjective differential at every point. It follows that the fibres, $`\varphi ^1(b)`$ for $`bB,`$ are compact manifolds with boundary, all diffeomorphic to a fixed manifold $`Z`$ for which we use the notation (3). It is often notationally convenient to assume that the fibres are also connected, but it is by no means necessary and we believe that the paper is written so that this assumption is not used. Such a fibration is always locally trivial.
In addition we assume that the boundary of the total space of the fibration, $`M,`$ carries a second, finer, fibration, giving a commutative diagram of fibrations
(1.1)
Thus the boundary of each fibre carries a fibration and the overall fibre, with this fibration of its boundary, is always diffeomorphic to the model fibre with its model boundary fibration, as in (20). The maps fit together as in (21) and it is always the case that there is a local trivialization of the overall fibration $`\varphi `$ in which the fibration of the boundary of the fibres is reduced to this normal form; in this sense the structure is locally trivial.
To associate with $`\varphi `$ and $`\mathrm{\Phi }`$ a class of pseudodifferential operators on the fibres of $`\varphi ,`$ of the type introduced in , we need one more piece of information. Namely we need to choose a trivialization of the normal bundle to the boundary of the fibres of $`\varphi `$ along the fibres of $`\mathrm{\Phi }.`$ This simply amounts to the choice of a boundary defining function $`x𝒞^{\mathrm{}}(M)`$ (so $`x0,`$ $`M=\{x=0\}`$ and $`dx0`$ on $`M).`$ Such a choice gives a trivialization of the normal bundle to $`M`$ everywhere (simply choose the inward pointing normal vector $`v_p`$ at $`pM`$ to satisfy $`v_px=1).`$ Two such choices $`x,`$ $`x^{},`$ are equivalent, in the sense that they give the same trivialization along the fibres of $`\mathrm{\Phi }`$ if and only if
(1.2)
$$x^{}=ax+bx^2,a,b𝒞^{\mathrm{}}(M),a|_M=\mathrm{\Phi }^{}a^{},a^{}𝒞^{\mathrm{}}(D).$$
Equivalent choices turn out to lead to the same algebra of pseudodifferential operators. Since even inequivalent choices are homotopic and lead to isomorphic structures the dependence on this choice of normal trivialization will not be emphasized; it should be fixed throughout but none of the results depend on which choice is made.
###### Definition 1.1.
A *fibration with fibred cusp structure* is a fibration of compact manifolds (3) with fibres modelled on a fixed compact manifold with boundary, a finer fibration as in (1.1) of the boundary of the total space and a choice of trivialization of the normal bundle to the boundary along the fibres as discussed above.
Notice that the extreme cases in which $`\mathrm{\Phi }=\varphi `$ is simply the restriction of the fibration to the boundary of $`M`$ is a particularly interesting case, the ‘cusp’ case, which includes the setting of the index theorem of Atiyah, Patodi and Singer. The other extreme, in which $`\mathrm{\Phi }`$ is the identity, is the essentially commutative ‘scattering’ case. In terms of , this corresponds to operators for which the obstruction in K-theory to the existence of local elliptic boundary conditions vanishes (in contrast with the global nature of the Atiyah-Patodi-Singer boundary condition). The general fibred cusp case is intermediate between these two extremes.
In general, even if we were assuming that $`M`$ is connected, it would be artificial to assume that the boundary $`M`$ is connected, since many interesting examples involve a disconnected boundary. We therefore avoid any such assumption. When the boundary is not connected, the cusp case is still interpreted as $`\mathrm{\Phi }=\varphi .`$ As discussed in and , this means one should allow terms of order $`\mathrm{}`$ acting between different components of the boundary and correspondingly in the definition of the indicial family.
Under these conditions we may associate to a fibration with fibred cusp structure an algebra of ‘fibred-cusp’ pseudodifferential operators, $`\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^{}(M/B).`$ The reader is referred to for the original definition of the algebra and to the discussion in . A generalization to product-type operators, used in the lifting construction below, is given in Appendix B. For our purposes here the main interest lies in the boundedness, compactness, Fredholm and related symbolic properties of these operators. Thus if $`𝔼=(E_+,E_{})`$ is a $`_2`$-graded complex vector bundle we will denote by $`\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^{}(M/B;𝔼)`$ the space of these operators acting from sections of $`E_+`$ to sections of $`E_{};`$ they always give continuous linear operators on weighted spaces of smooth sections
(1.3)
$$P\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^{}(M/B;𝔼):x^s𝒞^{\mathrm{}}(M;E_+)x^s𝒞^{\mathrm{}}(M;E_{}),s.$$
Both the symbol and the normal operator can be defined by appropriate ‘oscillatory testing’ of these maps. The symbol map gives a short exact sequence
(1.4)
with
(1.5)
$$𝒮_{\mathrm{\Phi }\mathrm{cu}}^m(M/B;𝔼)=𝒞^{\mathrm{}}({}_{}{}^{\mathrm{\Phi }\mathrm{cu}}S_{}^{}(M/B);\mathrm{hom}(𝔼)N_m)$$
where $`{}_{}{}^{\mathrm{\Phi }\mathrm{cu}}S_{}^{}(M/B)`$ is the sphere bundle of the fibred cusp cotangent bundle of the fibres (isomorphic, but not naturally so, to the ‘usual’ bundle $`T^{}(M/B))`$ and $`N_m=N^m`$ is the bundle with sections which are the functions homogeneous of degree $`m`$ on the fibres (then $`N`$ can be identified with the normal bundle to the boundary of the radial compactification) and $`\mathrm{hom}(𝔼)=\mathrm{hom}(E_+,E_{}).`$ The normal (or indicial) operator gives a short exact sequence
(1.6)
$$\text{}.$$
Here the image is the space of suspended pseudodifferential operators on the fibres of the fibration $`\mathrm{\Phi }`$ with symbolic parameters in
(1.7)
$${}_{}{}^{\mathrm{\Phi }\mathrm{cu}}T_{M}^{}(M/B)/T^{}(M/B)$$
which is to say the duals to the fibre variables of $`\mathrm{\Phi }`$ together with a dual to the normal variable. The symbol and normal operators are connected by the identity
(1.8)
$$\sigma |_M=\sigma N\text{ on }\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^m(M/B;𝔼).$$
This is the only constraint, so if we set
(1.9)
$$\begin{array}{c}A_\mathrm{\Phi }^m(\varphi ;𝔼)=\{(\sigma ,N)\hfill \\ \hfill 𝒮_{\mathrm{\Phi }\mathrm{cu}}^m(M/B;𝔼)\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{sus}}^m(M/D;𝔼);\sigma |_M=\sigma _m(N)\}\end{array}$$
the joint symbol sequence becomes
(1.10)
$$\text{}.$$
Such a family extends by continuity to an operator between the natural continuous families of weighted fibred-cusp Sobolev spaces
(1.11)
$$P:𝒞(B;x^sH_{\mathrm{\Phi }\mathrm{cu}}^M(M/B;E_+))𝒞(B;x^sH_{\mathrm{\Phi }\mathrm{cu}}^{Mm}(M/B;E_{})),M,s.$$
It is a Fredholm family on these spaces if and only if its image in $`A_\mathrm{\Phi }^m(\varphi ;𝔼)`$ is invertible, in which case we say that the family is fully elliptic.
###### Lemma 1.1.
If $`P\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^{}(M/B;𝔼)`$ is a fully elliptic family then there exists a smoothing perturbation in $`x^{\mathrm{}}\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^{\mathrm{}}(M/B;𝔼)`$ such that $`P+A`$ has null space in (1.11) forming a trivial smooth vector bundle, over $`B,`$ contained in $`\dot{𝒞}^{\mathrm{}}(M;E_+).`$
###### Proof.
The algebra is invariant under conjugation by $`x^s`$ and such conjugation affects neither the symbol nor the normal operator; thus we can take $`s=0.`$ For simplicity of notation we will also suppose that $`M=m`$ in (1.11); this is all that is used below and the general case follows easily. The full ellipticity and properties of the calculus allow us to construct a parametrix, which is to say a family $`Q\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^m(M/B;𝔼^{}),`$ where $`𝔼^{}=(E_{},E_+),`$ such that
(1.12)
$$PQ=\mathrm{Id}_+R_+,QP=\mathrm{Id}_{}R_{},R_\pm x^{\mathrm{}}\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^{\mathrm{}}(M/B;E_\pm ).$$
As a bundle of Hilbert spaces, $`𝒞(B;H_{\mathrm{\Phi }\mathrm{cu}}^m(M/B;E_+))`$ is necessarily trivial. Thus there is a sequence of orthogonal projections $`\mathrm{\Pi }_N,`$ tending strongly to the identity, corresponding to projection onto the first $`N`$ terms of an orthonormal basis in some trivialization. In particular for a compact operator such as $`R_+,`$ $`R_+\mathrm{\Pi }_NR_+`$ in the topology of norm continuous families of bounded operators.
Furthermore, $`\mathrm{\Pi }_N`$ may be approximated by smooth families of projections of the same rank. To see this, fixing $`N,`$ we may certainly find a sequence of smoothing operators with supports disjoint from the boundary such that $`W_j\mathrm{\Pi }_N`$ in the norm topology; it suffices to use a partition of unity and work locally. Then $`W_j^{}=W_j\mathrm{\Pi }_NW_j`$ is a continuous (in $`B)`$ family of smoothing operators, supported in the interior, which approximates $`\mathrm{\Pi }_N`$ and for large $`j`$ has rank $`N.`$ Taking $`j`$ sufficiently large the range of $`W_j^{}`$ is a trivial subbundle which has a basis $`e_{l,j}𝒞(B;𝒞^{\mathrm{}}(M/B;E_+))`$ with supports always in the interior. These sections can themselves be uniformly approximated by sections of the same form but smooth over $`B.`$ Replacing $`W_j^{}`$ by the orthogonal projection onto the span of these sections, for $`j`$ large, we have succeeded in approximating $`\mathrm{\Pi }_N`$ by smooth families of projections with kernels supported in the interior. Thus we may suppose that the $`\mathrm{\Pi }_N`$ are smooth families of smoothing operators with kernels vanishing to infinite order at both boundaries of the product.
Returning to (1.12) we may compose on the right with $`\mathrm{Id}\mathrm{\Pi }_N`$ and, using the fact that it is also a projection, deduce that
(1.13)
$$QP(\mathrm{Id}\mathrm{\Pi }_N)=(\mathrm{Id}B)(\mathrm{Id}\mathrm{\Pi }_N),B=R_+(\mathrm{Id}\mathrm{\Pi }_N).$$
For large $`N`$ it follows that $`B`$ has uniformly small norm. The inverse of $`\mathrm{Id}B`$ is then also of the form $`\mathrm{Id}+B^{}`$ with $`B^{}`$ a smoothing operator with kernel vanishing to infinite order at the boundary, and so is an element of $`x^{\mathrm{}}\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^{\mathrm{}}(M/B;E_+).`$ Replacing $`Q`$ by the new parametrix $`(\mathrm{Id}+B^{})Q`$ we have replaced the first identity in (1.12) by
(1.14)
$$Q(P+A)=\mathrm{Id}\mathrm{\Pi }_N,A=P\mathrm{\Pi }_Nx^{\mathrm{}}\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^{\mathrm{}}(M/B;𝔼).$$
This perturbs $`P`$ as desired. ∎
Now once the null space of $`P+A`$ is arranged to be a smooth bundle it follows that its range has a complement of the same type (but in general not trivial), namely the null bundle of $`P^{}+A^{}`$ for some choice of inner products and smooth density.
###### Proposition 1.2.
The element
$$\mathrm{null}(P+A)\mathrm{null}(P^{}+A^{})\mathrm{K}(B)$$
defined by any fully elliptic element of $`\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^m(M/B;E)`$ using Lemma 1.1 is independent of the choice of perturbation $`A,`$ is constant under smooth homotopy and is additive under direct sums.
###### Proof.
First suppose $`P`$ has been perturbed so that (1.14) holds for some parametrix $`Q`$ and some family of smooth projections $`\mathrm{\Pi }_N.`$ Changing to another family $`\mathrm{\Pi }_k^{}`$ we can choose $`k`$ so large that $`B^{}=\mathrm{\Pi }_N(\mathrm{Id}\mathrm{\Pi }_k^{}),`$ has small norm and then
(1.15)
$$(\mathrm{Id}B^{})^1Q(P+A^{})=\mathrm{Id}\mathrm{\Pi }_k^{},A^{}=P\mathrm{\Pi }_k^{}+A(\mathrm{Id}\mathrm{\Pi }_k^{}).$$
The null space of the new family is the range of $`\mathrm{\Pi }_k^{}`$ into which the range of $`\mathrm{\Pi }_N`$ is mapped isomorphically by $`\mathrm{Id}\mathrm{\Pi }_k^{}.`$ The range of $`P+A^{}`$ is then the direct sum of an isomorphic image of the complement of this image of $`\mathrm{\Pi }_N`$ plus the previous range. Thus the element in $`\mathrm{K}(B)`$ is unchanged.
To see the independence of the choice of stabilizing perturbation, suppose $`A`$ and $`A^{}`$ are two such perturbations. Consider the family depending on an additional parameter $`P+(\mathrm{cos}\theta A+\mathrm{sin}\theta A^{}),`$ $`\theta 𝕊.`$ The circle can be included in the base of the product fibration and then the argument above can be applied to stabilize the new family. This shows that the pairs of bundles resulting from different stabilizations are homotopic and so define the same element in $`\mathrm{K}(B).`$ Indeed the same argument applies to a homotopy of the operator itself, through fully elliptic operators.
It is then immediate that the index class is additive under direct sums. ∎
###### Remark 1.2.
Various ‘stabilization’ constructions like this are used below. For instance, if $`P_t`$ is a family of totally elliptic operators depending smoothly on an additional parameter $`t[0,1]`$ and it is invertible for $`t=0`$ then there is a finite rank smoothing perturbation, vanishing at $`t=0,`$ which makes the family invertible for all $`t[0,1].`$
To formulate the index as a map we now consider the K-group which arises from the symbol algebra.
###### Definition 1.3.
For a fibration with fibred cusp structure, $`\mathrm{K}_{\mathrm{\Phi }\mathrm{cu}}(\varphi )`$ denotes the set of equivalence classes of the collection of the invertible elements of the $`A_\mathrm{\Phi }^0(\varphi ,𝔼)`$ (with inverse in $`A_\mathrm{\Phi }^0(\varphi ,𝔼^{})`$, $`𝔼^{}=(E^{},E^+)`$) where the equivalence relation is a finite chain consisting of the following
(1.19)
$$\begin{array}{c}(\sigma ,N)A_\mathrm{\Phi }^0(\varphi ;𝔼)(\sigma _1,N_1)A_\mathrm{\Phi }^0(\varphi ;𝔽)\\ \text{ if there exist bundle isomorphisms over }M,a_\pm :E_\pm F_\pm \\ \text{ such that }\sigma =a_{}\sigma _1a_+\text{ and }N=(a_{}|_M)N_1(a_+|_M)\end{array}$$
(1.23)
$$\begin{array}{c}(\sigma ,N)A_\mathrm{\Phi }^0(\varphi ;𝔼)(\sigma _2,N_2)A_\mathrm{\Phi }^0(\varphi ;𝔼)\\ \text{ if there exists a homotopy }(\sigma (t),N(t)),t[0,1],(\sigma (t)^1,N(t)^1)A_\mathrm{\Phi }^0(\varphi ;𝔼^{})\\ \text{such that }\sigma (0)=\sigma ,\sigma (1)=\sigma _2,N(0)=N,N(1)=N_2,\end{array}$$
(1.26)
$$\begin{array}{c}(\sigma ,N)A_\mathrm{\Phi }^0(\varphi ;𝔼)(\sigma _3,N_3)A_\mathrm{\Phi }^0(\varphi ;𝔼F)\\ \text{ if }F\text{ is ungraded},\sigma _3=\sigma \mathrm{Id}_F\text{ and }N_3=N\mathrm{Id}_F.\end{array}$$
There is also a corresponding odd K-group. Given the diagram (21), let $`I=[0,1]`$ be the unit interval and consider the suspended version
(1.27)
where $`s\varphi =\varphi \times \mathrm{Id}`$ and $`s\mathrm{\Phi }=\mathrm{\Phi }\times \mathrm{Id}.`$ Let $`E`$ be an ungraded complex vector bundle. To the fibration (1.27) corresponds the space of joint symbols $`A_{s\mathrm{\Phi }}^m(s\varphi ;E).`$
###### Definition 1.4.
For a fibration with fibred cusp structure, $`\mathrm{K}_{\mathrm{\Phi }\mathrm{cu}}^1(\varphi )`$ denotes the set of equivalence classes of the collection of the invertible elements of the $`A_{s\mathrm{\Phi }}^0(s\varphi ,E)`$ (with $`E`$ ungraded and with inverse in $`A_{s\mathrm{\Phi }}^0(s\varphi ,E)`$) which are the identity when restricted to $`B\times \{0,1\}`$ under, the equivalence relation corresponding to a finite chain as in (1.19), (1.23) and (1.26) with bundle transformations and homotopies required to be the identity on $`B\times \{0,1\}.`$
###### Proposition 1.3.
For any fibration with fibred cusp structure, both $`\mathrm{K}_{\mathrm{\Phi }\mathrm{cu}}(\varphi )`$ and $`\mathrm{K}_{\mathrm{\Phi }\mathrm{cu}}^1(\varphi )`$ are abelian groups under direct sum (or equivalently stabilized product) which are naturally independent of the choice of boundary trivialization and the index construction of Lemma 1.1 defines group homomorphisms
(1.28)
$$\mathrm{ind}_\text{a}:\mathrm{K}_{\mathrm{\Phi }\mathrm{cu}}(\varphi )\mathrm{K}(B),\mathrm{ind}_\text{a}:\mathrm{K}_{\mathrm{\Phi }\mathrm{cu}}^1(\varphi )\mathrm{K}^1(B).$$
###### Proof.
The abelian group structure follows as in the boundaryless case. Since changing the boundary defining function only affects the calculus through a change of the trivialization of the normal bundle and all such trivializations are homotopic the resulting abelian groups are independent of this choice and Proposition 1.2 shows that the analytic index map, (1.28), is then well defined and additive. ∎
## 2. The scattering case
The scattering case, in which the fibres of the boundary fibration are reduced to points is effectively ‘commutative’ compared to the others. In particular it is very close to the setting of the original Atiyah-Singer index theorem and we show here that it is reducible to it. The resulting identification of the analytic and topological indexes allows us to derive, in the next section, an index theorem in the setting of a general boundary fibration but for perturbations of the identity of order $`\mathrm{}.`$
###### Lemma 2.1.
For the scattering structure on any compact fibration, (3),
(2.1)
$$\mathrm{K}_{\mathrm{sc}}(\varphi )=\mathrm{K}_{\mathrm{Id}\mathrm{cu}}(\varphi )\mathrm{K}_\text{c}(T^{}(M/B);T_M^{}(M/B)),$$
is identified with the compactly supported K-theory of the fibre cotangent bundle of the interior of $`M.`$
###### Proof.
In this case an element of $`A_{\mathrm{Id}}^0(\varphi ;𝔼)`$ is a pair $`(\sigma ,b)`$ each of which is a bundle isomorphism, taking values in the lift of $`\mathrm{hom}(𝔼).`$ The ‘symbolic part’ $`\sigma `$ is defined on the sphere bundle at infinity of the (radial compactification of the) appropriately rescaled fibre cotangent bundle $`{}_{}{}^{\mathrm{sc}}T_{}^{}(M/B)`$ and the boundary part is smooth on the radial compactification of the restriction, $`{}_{}{}^{\mathrm{sc}}T_{M}^{}(M/B)`$ of this bundle to the boundary. Since they are compatible at the intersection, which is to say the corner of $`\overline{{}_{}{}^{\mathrm{sc}}T_{}^{}}(M/B),`$ together this gives *precisely* a section of $`\mathrm{hom}(𝔼)`$ lifted to the boundary of $`\overline{{}_{}{}^{\mathrm{sc}}T_{}^{}}(M/B).`$ This is the data needed for the standard definition of a compactly supported K-class in the interior of a manifold with boundary (to which such a manifold with corners is homeomorphic) and all classes arise this way. This gives (2.1). ∎
Poincaré duality reduces to one of the cases discussed (for $`B=\{\text{pt}\})`$ in
(2.2)
$$\mathrm{K}_\text{c}(T^{}(M/B);T_M^{}(M/B))\mathrm{KK}_B^0(𝒞(M),𝒞(B)).$$
The index theorem in the scattering case has a ‘simple’ formulation and proof in the sense that it reduces directly to the Atiyah-Singer theorem through the following observation from .
###### Lemma 2.2.
In the scattering case $`\mathrm{K}_{\mathrm{sc}}(\varphi )`$ is generated by the equivalence classes of elements of the subset
(2.3)
$$\{(\sigma ,b)A_{\mathrm{Id}}^0(\varphi ;𝔼);\text{ near }M,E_+=E_{}=^N,\sigma =b=\mathrm{Id}\}A_{\mathrm{Id}}^0(\varphi ;𝔼).$$
###### Proof.
First we show that any invertible joint symbol is homotopic to an element in which, near the boundary, both $`\sigma `$ and $`b`$ are the lifts of some bundle isomorphism from $`E_+`$ to $`E_{}.`$ Given an element $`(s,b)A_{\mathrm{Id}}^0(\varphi ;𝔼)`$ the bundle isomorphism can be taken to be any extension off $`M`$ of the section $`b`$ restricted to the zero section (identified with $`M)`$ of $`{}_{}{}^{\mathrm{sc}}T_{M}^{}(M/B).`$ First perturb $`b`$ to be constant on the (linear) fibres of $`{}_{}{}^{\mathrm{sc}}T_{M}^{}(M/B)`$ near the zero section. An extension of the radial expansion of the vector bundle allows it to be deformed, with $`\sigma `$ to keep the consistency condition, to be fibre constant in this sense in a neighbourhood of the boundary.
Now, given that $`\sigma `$ and $`b`$ are identified with a bundle isomorphism near the boundary, this isomorphism can be used to modify $`E_{}`$ to be equal to $`E_+`$ close to the boundary so that both symbols become the identity on $`E_+`$ very close to the boundary without changing the equivalence class. Then $`E_+`$ can be complemented to a trivial bundle. ∎
Once the K-group is identified with the set of equivalence classes as in (2.3), the quantization map can also be arranged to yield pseudodifferential operators, in the ordinary sense, which are equal to the identity in a neighborhood of the boundary. Such operators can be extended to the double of $`M,`$ across the boundary, to be the identity on the additional copy of $`M`$ and the Atiyah-Singer index theorem then applies.
###### Theorem 2.3.
For the scattering structure on a fibration the analytic index for fully elliptic scattering operators on the fibres factors through the Atiyah-Singer index map for the double $`2M=MM^{}`$
(2.4)
$$\begin{array}{c}\mathrm{ind}_\text{a}:\mathrm{K}_{\mathrm{sc}}(\varphi )\stackrel{}{}\mathrm{K}_\text{c}(T^{}(M/B);T_M^{}(M/B))=\hfill \\ \hfill \mathrm{K}_\text{c}(T^{}(2M/B);T^{}(M^{}/B))\mathrm{K}_\text{c}(T^{}(2M/B))\stackrel{\mathrm{ind}_{\mathrm{AS}}}{}\mathrm{K}(B).\end{array}$$
###### Proposition 2.4.
Suppose that $`E`$ is a complex vector bundle over the total space of a fibration with scattering structure and $`b𝒮({}_{}{}^{\mathrm{sc}}T_{M}^{}(M/B);\mathrm{hom}(E))`$ is such that $`\mathrm{Id}+b`$ is invertible then any family $`\mathrm{Id}+B,`$ $`B\mathrm{\Psi }_{\mathrm{sc}}^{\mathrm{}}(M/B;E)`$ with $`N(\mathrm{Id}+B)=\mathrm{Id}+b`$ has analytic index equal to the image
(2.5)
$$[\mathrm{Id}+b]\mathrm{K}_\text{c}^1(\times T^{}(M/B))=\mathrm{K}_\text{c}(T^{}(M/B))\stackrel{\mathrm{ind}_{\mathrm{AS}}}{}\mathrm{K}(B)$$
under the Atiyah-Singer index map for the boundary.
###### Proof.
First we can complement $`E`$ to be trivial, stabilizing the symbol by the identity. By a small perturbation we can also arrange that the boundary symbol $`b`$ is compactly supported on $`{}_{}{}^{\mathrm{sc}}T_{M}^{}(M/B)=\times T^{}(M/B)`$ and is equal to a bundle map near the zero section. Thus, the varying part of $`b`$ can be confined to a compact subset of $`(0,\mathrm{})\times W,`$ where $`W`$ is the boundary of the radial compactification of the vector bundle $`_s\times T^{}(M/B)`$ and the first variable is the radial variable. In the deformation in the proof of Lemma 2.2 above, across the corner at infinity and into the interior, the radial variable becomes the normal variable to the boundary, $`x,`$ in a product decomposition near the boundary with the variation now in $`(0,1)`$ and $`W`$ is identified with the boundary of the radial compactification of $`T^{}(M/B)|_{x=\frac{1}{2}}.`$ Thus, $`\mathrm{Id}+b`$ has been identified with a symbol in the conventional sense on the sphere bundle of $`T^{}((0,1)_x\times M/B)`$ reducing to the identity near $`x=1`$ and some bundle isomorphism near $`x=0.`$ By Bott periodicity of the Atiyah-Singer index map this reduces to the identifications in (2.5). ∎
###### Remark 2.1.
From this result a families index theorem in K-theory in the setting of Callias’ index theorem follows. See also the discussion by Anghel and Kucerovsky for single operators.
## 3. Perturbations of the identity
The discussion in the previous section of the index in the scattering case allows us to compute the index of Fredholm perturbations of the identity by fibred cusp operators of order $`\mathrm{},`$ i.e. to give analogues of Proposition 2.4 for all fibred cusp structures. This was done in for the numerical index.
###### Proposition 3.1.
Suppose that $`E`$ is a complex vector bundle over the total space of a fibration with fibred cusp structure and $`b\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{sus}}^{\mathrm{}}(M/D;E)`$ is such that $`\mathrm{Id}+b`$ is invertible, then any family $`\mathrm{Id}+B,`$ $`B\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^{\mathrm{}}(M/B;E)`$ with $`N(\mathrm{Id}+B)=\mathrm{Id}+b`$ has analytic index equal to the image
(3.1)
$$[\mathrm{Id}+b]\mathrm{K}_\text{c}^1(\times T^{}(D/B))=\mathrm{K}_\text{c}(T^{}(D/B))\stackrel{\mathrm{ind}_{\mathrm{AS}}}{}\mathrm{K}(B)$$
under the Atiyah-Singer index map for the fibration of $`D`$ over $`B.`$
###### Proof.
First we may use an ‘excision’ construction to replace $`M`$ by the simpler manifold $`M\times [0,1]_x.`$ Indeed, taking a product decomposition of $`M`$ near the boundary and identifying it with a neighbourhood of $`x=0`$ in the product, the quantization of $`b`$ may be localized to vanish outside this neighbourhood, i.e. to have kernel vanishing outside the product of this neighbourhood with itself, and so can be identified with an operator on the model product with the same index. Thus it suffices to consider the product case $`M=M\times [0,1],`$ with a bundle lifted from the boundary and with $`b`$ trivial at the $`x=1`$ boundary.
Thus the boundary fibration extends to the whole space as a fibration over $`[0,1]\times D`$ and the strategy is to reduce the problem to the scattering calculus on this space. Consider a family of smoothing projections $`\mathrm{\Pi }_N`$ as in Section 1 for the bundle $`E`$ over the fibration of the boundary given by $`\mathrm{\Phi }`$ extended to act trivially in the variable $`x[0,1].`$ Then $`\mathrm{\Pi }_Nb\mathrm{\Pi }_Nb`$ as $`N\mathrm{}`$ uniformly on $`\times T^{}(D/B)`$ in view of the rapid decay. Thus we may replace $`b`$ by $`\mathrm{\Pi }_Nb\mathrm{\Pi }_N`$ for sufficiently large $`N`$ and hence assume that it acts on some finite rank subbundle of the smooth sections of $`𝒞^{\mathrm{}}(M/D)`$ (namely the range of $`\mathrm{\Pi }_N)`$ pulled back to $`\times T^{}(D/B).`$ Quantizing $`b`$ to an operator $`B\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^{\mathrm{}}(M;E)`$ we may arrange that $`\mathrm{\Pi }_NB=B=B\mathrm{\Pi }_N`$ by replacing $`B`$ by $`\mathrm{\Pi }_NB\mathrm{\Pi }_N.`$ Note that $`\mathrm{\Pi }_N`$ is not a fibred cusp pseudodifferential operator, because it kernel is singular on the fibre diagonal, however its composite with an element of $`\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^{\mathrm{}}(M;E)`$ is in the same space; this follows from an examination of the kernels, see Appendix A. Now in fact the same local analysis shows that $`\mathrm{\Pi }_NB\mathrm{\Pi }_N\mathrm{\Psi }_{\mathrm{sc}}^{\mathrm{}}([0,1]\times D/B;W_N)`$ has boundary symbol $`b,`$ where $`W_N`$ is the range of $`\mathrm{\Pi }_N.`$ Thus the result follows from Proposition 2.4. ∎
## 4. Analytic classes in $`\mathrm{KK}`$ theory
Baum, Douglas and Taylor, in , associate a $`\mathrm{KK}`$-class with a Dirac operator on a manifold with boundary with Atiyah-Patodi-Singer boundary condition. In this section, we extend, and refine, their construction to the situation of families of fully elliptic fibred cusp operators. This can also be thought as an adaptation to the context of fibred cusp operators of a similar discussion for b-pseudodifferential operators in , except that even in that special case, here we associate to a Fredholm b-pseudodifferential operator on $`X`$ a class in the K-homology of $`X/X,`$ the manifold with the boundary smashed to a point, rather than the absolute (or relative) K-homology (see the final remark of ). Although small, this is an important difference in that it is at the heart of our formulation of a families index theorem in K-theory, including the Atiyah-Patodi-Singer case.
For a quick review of $`\mathrm{KK}`$-theory and the relation with elliptic operators see and and the books and as well as the papers , , and where KK-theory was initially developed. To describe families of elliptic operators via KK-theory we recall the extra feature introduced in .
###### Definition 4.1.
If $`X`$ is a compact manifold and $`𝒜`$ a $`_2`$-graded $`C^{}`$ algebra then a $`𝒞(X)`$-algebra structure on $`𝒜`$ is a graded unital homomorphism
$$r:𝒞(X)Z((𝒜))$$
where $`Z((𝒜))`$ is the center of $`(𝒜),`$ the multiplier algebra of $`𝒜,`$ and where $`𝒞(X)`$ is trivially graded; in particular this gives $`𝒜`$ a $`𝒞(X)`$-module structure.
###### Definition 4.2.
Let $`(𝒜,r_𝒜)`$ and $`(𝒜^{},r_𝒜^{})`$ be graded $`𝒞(X)`$-algebras where $`X`$ is a compact manifold. Then $`𝔼_X(𝒜,𝒜^{})`$ is the set of all triples $`(E,\varphi ,F)`$ where $`E`$ is countably generated graded Hilbert module over $`𝒜^{},`$ $`\varphi `$ is a graded $``$-homomorphism from $`𝒜`$ to $`(E)`$ and $`F`$ is an operator in $`(E)`$ of degree 1, such that for all $`a𝒜,`$ $`b𝒜^{},`$ $`eE`$ and $`f𝒞(X),`$
(4.1) $`\text{(i) }[\varphi (a),F]𝒦(E),`$
$`\text{(ii) }\varphi (a)(F^2\mathrm{Id})𝒦(E),`$
$`\text{(iii) }\varphi (a)(FF^{})𝒦(E),`$
$`\text{(iv) }\varphi (ar_𝒜(f))(eb)=\varphi (a)(e(r_𝒜^{}(f)b)).`$
Here $`𝒦(E),`$ defined for instance in , is the analog of compact operators for Hilbert modules. The elements of $`𝔼_X(𝒜,𝒜^{})`$ are called Kasparov $`𝒞(X)`$-modules for $`(𝒜,𝒜^{}).`$ We denote by $`𝔻_X(𝒜,𝒜^{})`$ the set of triples in $`𝔼_X(𝒜,𝒜^{})`$ for which $`[F,\varphi (a)],`$ $`(FF^{})\varphi (a)`$ and $`(F^21)\varphi (a)`$ vanish for all $`a𝒜.`$ The elements of $`𝔻_X(𝒜,𝒜^{})`$ are called degenerate Kasparov modules.
Condition (iv) is the extra feature needed to deal with families of elliptic operators. It requires equivariance for the $`𝒞(X)`$-module structure of $`𝒜`$ and $`𝒜^{}.`$ One recovers the definition of standard Kasparov modules by dropping (iv).
An element $`(E,\varphi ,F)𝔼_X(𝒜,𝒞([0,1];𝒜^{}))`$ generates a family
$$\{(E_t,\varphi _t,F_t)𝔼_X(𝒜,𝒜^{});t[0,1]\}$$
obtained by evaluation at each $`t[0,1].`$ This family and the triple itself will be called a homotopy between $`(E_0,\varphi _0,F_0)`$ and $`(E_1,\varphi _1,F_1)`$ and these modules are then said to be homotopic with the relation written
$$(E_0,\varphi _0,F_0)_h(E_1,\varphi _1,F_1).$$
See for example for a proof that all degenerate Kasparov modules are homotopic to the trivial Kasparov $`𝒞(X)`$-module.
###### Definition 4.3.
We denote by $`\mathrm{KK}_X^0(𝒜,𝒜^{})`$ the set of equivalence classes of $`𝔼_X(𝒜,𝒜^{})`$ under the equivalence relation $`_h`$ and similarly define $`\mathrm{KK}_X^1(𝒜,𝒜^{})`$ by
$$\mathrm{KK}_X^1(𝒜,𝒜^{})=\mathrm{KK}_X^0(𝒜,𝒮𝒜^{})$$
where
$$𝒮𝒜^{}=\{f𝒞([0,1];𝒜^{});f(0)=f(1)=0\}$$
is the the suspension of the $`C^{}`$ algebra $`𝒜^{}.`$
In the paper of Kasparov , the notation $`\mathrm{KK}(X;𝒜,𝒜^{})`$ is used, we prefer a more compact notation.
As discussed in , there are various other useful equivalence relations which give $`\mathrm{KK}_X^0(𝒜,𝒜^{}).`$ The set of equivalent classes $`\mathrm{KK}_X^0(𝒜,𝒜^{})`$ is an abelian group with addition given by direct sum
$$[(E_0,\varphi _0,F_0)]+[(E_1,\varphi _1,F_1)]=[(E_0E_1,\varphi _0\varphi _1,F_0F_1)].$$
We are now in a position to define the $`\mathrm{KK}`$-classes associated to fully elliptic fibred cusp operators. In our situation, $`X=B`$ is the base of the fibration, while $`𝒜`$ will be variously $`𝒞_0(M),`$ $`𝒞_\mathrm{\Phi }(M),`$ $`𝒞(D),`$ etc., and $`𝒜^{}=𝒞(B).`$ In particular, we will only consider commutative $`C^{}`$ algebras which are trivially graded and the $`𝒞(B)`$-algebra structure will always be the obvious one.
Let $`P\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^m(M/B;𝔼)`$ be a family of fully elliptic fibred cusp operators of order $`m,`$ where $`𝔼`$ is a $`_2`$-graded complex vector bundle on $`M.`$ Introducing a graded inner product on $`𝔼`$ and family of metrics on $`M/B`$ it follows that $`P^{}P`$ is also a family of fully elliptic operators of order $`2m`$ with strictly positive symbol and indicial family. By standard pseudodifferential constructions there is an approximate inverse square-root $`Q\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^m(M/B;E_+)`$ which is invertible and positive definite such that
(4.2)
$$Q^2P^{}P\mathrm{Id}x^{\mathrm{}}\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^{\mathrm{}}(M/B;E_+),$$
where $`x`$ is a boundary defining function for $`M.`$ Consider the family of operators $`A=PQ\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^0(M/B;𝔼).`$ It is fully elliptic and by construction it is almost unitary in the sense that
(4.3)
$$A^{}A\mathrm{Id}x^{\mathrm{}}\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^{\mathrm{}}(M/B;E_+)\text{ and }AA^{}\mathrm{Id}x^{\mathrm{}}\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^{\mathrm{}}(M/B;E_{})$$
are families of compact operators. Let $`𝒞_\mathrm{\Phi }(M)`$ denote the space of continuous functions on $`M`$ which are constant along the fibres of $`\mathrm{\Phi }`$, that is
(4.4)
$$𝒞_\mathrm{\Phi }(M)=\{f𝒞(M);f|_M=g\mathrm{\Phi }\text{ for some }g𝒞(D)\}.$$
The injection $`𝒞(B)𝒞_\mathrm{\Phi }(M)`$ gives $`𝒞_\mathrm{\Phi }(M)`$ a $`𝒞(B)`$-algebra structure and $`𝒞(B)`$ itself has the $`𝒞(B)`$-algebra structure given by the identity map $`𝒞(B)𝒞(B).`$ Also let $`𝒞_\mathrm{\Phi }^{\mathrm{}}(M)𝒞_\mathrm{\Phi }(M)`$ denote the subspace of those functions which are smooth, thus
$$𝒞_\mathrm{\Phi }^{\mathrm{}}(M)=\{f𝒞^{\mathrm{}}(M);f|_M=g\mathrm{\Phi }\text{ for some }g𝒞^{\mathrm{}}(D)\}.$$
Let us denote by $`\mu `$ the action of the $`C^{}`$ algebra $`𝒞_\mathrm{\Phi }(M)`$ through multiplication
$$\mu (f)(),=\mathrm{L}^2(M/B;𝔼)=\mathrm{L}^2(M/B;E_+)\mathrm{L}^2(M/B;E_{}),f𝒞_\mathrm{\Phi }(M).$$
###### Lemma 4.1.
The triple $`(,\mu ,)`$ where
$$=\left(\begin{array}{cc}0& A^{}\\ A& 0\end{array}\right)()$$
gives rise to a well-defined Kasparov module in $`𝔼_B(𝒞_\mathrm{\Phi }(M),𝒞(B))`$ and hence a class
(4.5)
$$[P]=[(,\mu ,)]\mathrm{KK}_B^0(𝒞_\mathrm{\Phi }(M),𝒞(B)).$$
###### Proof.
By Kuiper’s theorem, $``$ is a countably generated $`_2`$-graded Hilbert module over $`𝒞(B).`$ To show that $`(,\mu ,)`$ is a module in the sense of Kasparov we need to check the following properties
(4.6)
$$\begin{array}{cc}& \text{(i) }[\mu (f),]𝒦(),\hfill \\ & \text{(ii) }\mu (f)(^2\mathrm{Id})𝒦(),\hfill \\ & \text{(iii) }\mu (f)(^{})𝒦(),\hfill \\ & \text{(iv) }\mu (b_1f)(hb_2)=\mu (f)(h(b_2b_1)),\hfill \end{array}f𝒞_\mathrm{\Phi }(M),b_1,b_2𝒞(B)\text{ and }h.$$
Property (iv) is immediate. Property (iii) follows directly from the fact that $`^{}=.`$ Property (ii) is a consequence of (4.3). To check property (i), we may restrict to $`f𝒞_\mathrm{\Phi }^{\mathrm{}}(M)`$ since these smooth functions are dense in $`𝒞_\mathrm{\Phi }(M)`$ and the map
$$[\mu (),]:𝒞_\mathrm{\Phi }(M)()$$
is continuous.
For any $`f𝒞_\mathrm{\Phi }^{\mathrm{}}(M),`$ $`\mu (f)\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^0(M/B;𝔼)`$ has symbol $`f`$ and indicial family which can be identified with $`f|_M`$ which is to say a constant multiple of the identity on each fibre, so commuting with the normal operator of any other element. Thus
$$[\mu (f),]x\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^1(M/B;𝔼)$$
is a family of compact operators. ∎
A fully elliptic operator also defines a class in the group $`\mathrm{KK}(𝒞_\mathrm{\Phi }(M),𝒞(B));`$ to get Poincaré duality, we need to take into account the $`𝒞(B)`$-algebra structure.
###### Lemma 4.2.
The class $`[P]\mathrm{KK}_B^0(𝒞_\mathrm{\Phi }(M),𝒞(B))`$ associated to a fully elliptic family of fibred cusp pseudodifferential operators by Lemma 4.1 does not depend on the choice of $`Q`$ in (4.2) and in fact only depends on the homotopy class of $`P`$ in the space of fully elliptic operators.
###### Proof.
Any two choices of a family of positive definite approximate square-roots differ by a family in $`x^{\mathrm{}}\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^{\mathrm{}}(M/B;E_+)`$ so the resulting families $``$ differ by compact families and hence define the same element in $`\mathrm{KK}_B^0(𝒞_\mathrm{\Phi }(M),𝒞(B)).`$
To prove the second part of the lemma, let $`p_t\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^m([0,1]\times M/B;𝔼)`$ be a smooth curve of families of fully elliptic operators, where $`t[0,1].`$ Then there exists a smooth curve $`Q_t\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^m([0,1]\times M/B;E_+)`$ of invertible approximate inverse square-roots such that
$$Q_t^2P_t^{}P_t\mathrm{Id}x^{\mathrm{}}\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^{\mathrm{}}([0,1]\times M/B;𝔼).$$
Hence, $`(,\mu ,_t)𝔼(𝒞_\mathrm{\Phi }(M),𝒞(B))`$ and if $`A_t=P_tQ_t`$ then
$$_t=\left(\begin{array}{cc}0& A_t^{}\\ A_t& 0\end{array}\right)(),$$
defines an operator homotopy between the modules $`(,\mu ,_0)`$ and $`(,\mu ,_1).`$ This implies that $`[P_0]=[P_1]`$ in $`\mathrm{KK}_B^0(𝒞_\mathrm{\Phi }(M),𝒞(B)).`$
This $`\mathrm{KK}`$-class also behaves in the expected manner under direct sums, so if $`P\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^m(M/B;𝔼)`$ and $`R\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^m(M/B;𝔽)`$ are families of fully elliptic operators, then
$$[PR]=[P]+[R]\text{ in }\mathrm{KK}_B^0(𝒞_\mathrm{\Phi }(M),𝒞(B)).$$
It follows that this construction defines a ‘quantization’ homomorphism of abelian groups
(4.7)
$$\mathrm{quan}:\mathrm{K}_{\mathrm{\Phi }\mathrm{cu}}(\varphi )\mathrm{KK}_B^0(𝒞_\mathrm{\Phi }(M),𝒞(B)).$$
The analytical index of the family $`P`$ factors through this map. Let $`c_\mathrm{\Phi }:𝒞(B)𝒞_\mathrm{\Phi }(M)`$ be the inclusion of constant functions along the fibres of $`\varphi :MB.`$ Then, at the level of $`\mathrm{KK}`$-theory, $`c_\mathrm{\Phi }`$ defines a contravariant functor
$$c_\mathrm{\Phi }^{}:\mathrm{KK}_B^0(𝒞_\mathrm{\Phi }(M),𝒞(B))\mathrm{KK}_B^0(𝒞(B),𝒞(B)).$$
###### Lemma 4.3.
Under the standard identification
$$\mathrm{KK}_B^0(𝒞(B),𝒞(B))\mathrm{KK}(,𝒞(B))\mathrm{K}^0(B),$$
there is a commutative diagram
###### Proof.
This follows from the discussion in , more precisely proposition 17.5.5, corollary 12.2.3 and paragraph 8.3.2. It is also a simple consequence of the stabilization of the null space of $`P,`$ by perturbation, of Lemma 1.1. ∎
It is also possible to define a quantization map for $`\mathrm{K}_{\mathrm{\Phi }\mathrm{cu}}^1(\varphi ).`$ Given a joint symbol $`(\sigma ,N)A_{s\mathrm{\Phi }}^0(s\varphi ;E_+)`$ representing a class in $`\mathrm{K}_{\mathrm{\Phi }\mathrm{cu}}^1(\varphi ),`$ let $`P_s\mathrm{\Psi }_{\mathrm{s}\mathrm{\Phi }\mathrm{cu}}^0(M\times I/B\times I;E)`$ be a family of fully elliptic fibred cusp operators with joint symbol $`(\sigma ,N)`$ such that $`P_s|_{B\times \{0,1\}}\mathrm{Id}.`$ Here, recall that $`s\mathrm{\Phi }:M\times ID\times I`$ is the boundary fibration of (1.27). Let $`Q_s\mathrm{\Psi }_{\mathrm{s}\mathrm{\Phi }\mathrm{cu}}^0(M\times I/B\times I;E_+)`$ be an approximate positive definite inverse square root
$$Q_s^2P_s^{}P_s\mathrm{Id}x^{\mathrm{}}\mathrm{\Psi }_{\mathrm{s}\mathrm{\Phi }\mathrm{cu}}^{\mathrm{}}(M\times I/B\times I,E_+)$$
such that $`Q|_{B\times \{0,1\}}\mathrm{Id},`$ where $`x`$ is the boundary defining function for $`M.`$ Consider the family of operators $`A=PQ\mathrm{\Psi }_{\mathrm{s}\mathrm{\Phi }\mathrm{cu}}^0(M\times I/B\times I;E_+).`$ It is fully elliptic, $`A|_{B\times \{0,1\}}\mathrm{Id},`$ and by construction it almost unitary in the sense that
$$A^{}A\mathrm{Id}x^{\mathrm{}}\mathrm{\Psi }_{\mathrm{s}\mathrm{\Phi }\mathrm{cu}}^{\mathrm{}}(M\times I/B\times I,E_+),AA^{}\mathrm{Id}x^{\mathrm{}}\mathrm{\Psi }_{\mathrm{s}\mathrm{\Phi }\mathrm{cu}}^{\mathrm{}}(M\times I/B\times I,E_{})$$
are families of compact operators which vanish on $`B\times \{0,1\}.`$ Notice that there is a natural inclusion $`𝒞_\mathrm{\Phi }(M)𝒞_{s\mathrm{\Phi }}(M\times I).`$ Let us denote by $`\mu _s`$ the action of the $`C^{}`$ algebra $`𝒞_\mathrm{\Phi }(M)`$ through multiplication
(4.8)
$$\begin{array}{c}\mu _s(f)(_s),_s=\mathrm{L}^2(M\times I/B\times I;𝔼)\hfill \\ \hfill =\mathrm{L}^2(M\times I/B\times I;E_+)\mathrm{L}^2(M\times I/B\times I;E_{})\end{array}$$
for $`f𝒞_\mathrm{\Phi }(M),`$ where in this odd context $`E_+=E_{}.`$ Let $`𝒮𝒞(B)`$ be the the $`C^{}`$ algebra of continuous functions
(4.9)
$$𝒮𝒞(B)=\{f𝒞(B\times I);f|_{B\times \{0,1\}}0\}.$$
###### Lemma 4.4.
The triple $`(_s,\mu _s,_s)`$ where
$$_s=\left(\begin{array}{cc}0& A^{}\\ A& 0\end{array}\right)(_s),$$
gives rise to a well-defined Kasparov module $`𝔼_B(𝒞_\mathrm{\Phi }(M),𝒮𝒞(B))`$ and a class
$$[(\sigma ,N)]=[P_s]\mathrm{KK}_B^0(𝒞_\mathrm{\Phi }(M),𝒮𝒞(B))$$
which only depends on the class $`[(\sigma ,N)]\mathrm{K}_{\mathrm{\Phi }\mathrm{cu}}^1(\varphi )`$
###### Corollary 4.5.
Under the standard identification
$$\mathrm{KK}_B^0(𝒞_\mathrm{\Phi }(M),𝒮𝒞(B))\mathrm{KK}_B^1(𝒞_\mathrm{\Phi }(M),𝒞(B)),$$
Lemma 4.4 gives us a well defined quantization map
(4.10)
$$\mathrm{quan}:\mathrm{K}_{\mathrm{\Phi }\mathrm{cu}}^1(\varphi )\mathrm{KK}_B^1(𝒞_\mathrm{\Phi }(M),𝒞(B)).$$
## 5. Poincaré duality for cusp operators
We shall prove Theorem 2 in the case of the cusp structure. To do so, we first discuss the short exact sequence (10).
Consider the collection of invertible joint symbols of cusp pseudodifferential operators,
(5.1)
$$\begin{array}{c}J_{\mathrm{cu}}^0(\varphi ,𝔼)=\{(\sigma ,N)S_{\mathrm{cu}}^0(M/B);𝔼)\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{sus}}^0(M/B;𝔼);\hfill \\ \hfill \sigma |_M=\sigma _0(N),(\sigma ^1,N^1)S_{\mathrm{cu}}^0(M/B);𝔼^{})\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{sus}}^0(M/B;𝔼^{})\}\end{array}$$
on $`_2`$-graded bundles where $`S_{\mathrm{cu}}^0(M/B;𝔼)=𝒞^{\mathrm{}}({}_{}{}^{\mathrm{cu}}S_{}^{}(M/B);\mathrm{hom}(𝔼)).`$ This naturally maps by restriction to the collection of invertible symbols
(5.2)
$$\begin{array}{c}G_{\mathrm{cu}}^0(\varphi ,𝔼)=\left\{\sigma S_{\mathrm{cu}}^0(M/B;𝔼);\sigma ^1S_{\mathrm{cu}}^0(M/B;𝔼^{})\right\},\\ \sigma :J_{\mathrm{cu}}^0(\varphi ,𝔼)G_{\mathrm{cu}}^0(\varphi ,𝔼).\end{array}$$
In fact this map is surjective. That is, for every family of elliptic symbols there does exist a family of invertible normal operators which is compatible with it. This is an aspect of the cobordism invariance of the index and is shown in this form in .
The clutching construction associates to each invertible symbol an element of the compactly supported $`\mathrm{K}`$ theory of $`{}_{}{}^{\mathrm{cu}}T_{}^{}(M/B)`$ which is ‘absolute’ with respect to the boundary of $`M.`$ As in the boundaryless case, the resulting element is stable under homotopy, bundle isomorphisms of $`E_\pm `$ and stabilization, so (5.2) descends to a surjective map
(5.3)
$$\mathrm{K}_{\mathrm{cu}}(\varphi )\mathrm{K}_\text{c}(T^{}(M/B))$$
where we use the fact that the cusp and standard cotangent bundles are isomorphic, with the isomorphism natural up to homotopy. Thus (5.3) is a surjective group homomorphism.
If $`E_+=E_{}`$ we can consider those invertible suspended families of pseudodifferential operators which are smoothing perturbations of the identity on a fixed bundle over $`M`$
(5.4)
$$G_{\mathrm{sus}}^{\mathrm{}}(\varphi ;E)=\left\{N\mathrm{Id}_E+\mathrm{\Psi }_{\mathrm{sus}}^{\mathrm{}}(M/B;E);N^1\mathrm{Id}_E+\mathrm{\Psi }_{\mathrm{sus}}^{\mathrm{}}(M/B;E)\right\}.$$
Let $`G_{\mathrm{sus}}^{\mathrm{}}(\varphi ;)`$ be the union over $`E`$ and let $`G_{\mathrm{e}\mathrm{sus}}^{\mathrm{}}(\varphi ;)`$ be the subset corresponding to bundles $`E`$ which bound a bundle over $`M.`$ Since we may complement a bundle to be trivial, and hence extendible, the stable homotopy classes of elements of $`G_{\mathrm{e}\mathrm{sus}}^{\mathrm{}}(\varphi ;)`$ and $`G_{\mathrm{sus}}^{\mathrm{}}(\varphi ;)`$ are the same.
###### Lemma 5.1.
Passing to the set of stable homotopy classes, with equivalence also under bundle isomorphisms, the inclusion and restriction maps give the split short exact sequence of Abelian groups (11):
(5.5)
###### Proof.
That the set of stable homotopy classes, also allowing smooth identification of bundles, of $`G_{\mathrm{e}\mathrm{sus}}^{\mathrm{}}(\varphi ,),`$ and hence also $`G_{\mathrm{sus}}^{\mathrm{}}(\varphi ,),`$ is canonically identified with $`\mathrm{K}(B)`$ is a standard result (see for instance ) when the fibration is trivial, $`M=Z\times B.`$ It remains true in the general case, this can be shown using the families of projections $`\mathrm{\Pi }_N`$ of Section 1, see also .
Since direct sums behave consistently, the resulting maps are group homomorphisms and form a complex, since $`G_{\mathrm{e}\mathrm{sus}}^{\mathrm{}}(\varphi ,)`$ clearly maps to the identity in $`G_{\mathrm{cu}}^0(\varphi ;).`$
We have already noted the surjectivity of the second map. To see exactness in the middle, suppose that a compatible pair $`(\sigma ,N)`$ induces a trivial class in $`\mathrm{K}_\text{c}(T^{}(M/B)).`$ Since stabilization and the action of bundle isomorphisms is the same on the full and symbolic data, we may suppose that $`\sigma `$ is homotopic to the identity through elliptic symbols. In particular, $`E_+=E_{}.`$ Adding the homotopy variable as an additional base variable, the surjectivity of the symbol map allows the homotopy of symbols to be lifted to an homotopy of joint symbols, see Remark 1.2. Thus $`(\sigma ,N)`$ may be deformed by homotopy to $`(\mathrm{Id},\mathrm{Id}+A)`$ where necessarily $`A\mathrm{\Psi }_{\mathrm{sus}}^1(M/B;E).`$ By a further small perturbation this is homotopic to $`\mathrm{Id}+AG_{\mathrm{e}\mathrm{sus}}^{\mathrm{}}(\varphi ,),`$ $`A\mathrm{\Psi }_{\mathrm{sus}}^{\mathrm{}}(M/B;E),`$ showing exactness at $`\mathrm{K}_{\mathrm{cu}}(\varphi ).`$ Injectivity of the first map follows from the fact that the index map provides a right inverse for it, which is a consequence of the families index of Proposition 3.1. Note that the existence of an invertible family with a given elliptic symbol defines the map $`\mathrm{inv}`$ which shows that the sequence splits. ∎
There is also a corresponding exact sequence at the level of KK-theory.
###### Lemma 5.2.
The short exact sequence of $`C^{}`$-algebras (12) leads to a split short exact sequence
(5.6)
$$\mathrm{KK}_B^0(𝒞(B),𝒞(B))\stackrel{\iota ^{}}{}\mathrm{KK}_B^0(𝒞_{\mathrm{cu}}(M),𝒞(B))\stackrel{s^{}}{}\mathrm{KK}_B^0(𝒞_0(M),𝒞(B)).$$
###### Proof.
From standard results in $`\mathrm{KK}`$-theory (cf. theorem 19.5.7 in ), the short exact sequence (12) leads to a six-term exact sequence
(5.7)
where both boundary homomorphisms, $`\delta ,`$ are obtained by multiplying by a specific element $`\delta _\iota \mathrm{KK}_B^1(𝒞(B),𝒞_0(M)).`$ More precisely, under the identification of $`\mathrm{KK}_B^1(𝒞(B),𝒞_0(M))`$ with $`\mathrm{KK}_B^0(S𝒞(B),𝒞_0(M)),`$ $`\delta _\iota =i^{}u`$ where $`u\mathrm{KK}_B^0(C_\iota ,𝒞_0(M))`$ and $`i:S𝒞(B)C_\iota `$ is the natural inclusion. Here, $`C_\iota `$ is the mapping cone
$$\begin{array}{c}C_\iota =\{(x,f)𝒞_{\mathrm{cu}}(M)𝒞_0([0,1)\times B);\iota (x)=f(0)\},S𝒞(B)=𝒞_0((0,1)\times B)\end{array}$$
and $`i(f)=(0,f)C_\iota `$ for $`fS𝒞(B).`$ In this situation there is an injective map
$$j:𝒞_0([0,1)\times B)f(\varphi ^{}(f(0)),f)C_\iota $$
so we may interpret $`i`$ as a map from $`S𝒞(B)`$ to $`𝒞_0([0,1)\times B).`$ Since $`𝒞_0([0,1)\times B)`$ is a contractible $`C^{}`$-algebra, the class $`[i]`$ of $`i`$ in $`\mathrm{KK}_B^0(S𝒞(B),C_\iota )`$ is zero. In particular, this means that $`i^{}u=0`$ since $`i^{}u`$ can be interpreted as the Kasparov product of $`[i]`$ with $`u`$ (see for example 18.4.2a in ). Thus $`\delta =0`$ and we get the short exact sequence (5.6).
To see that this short exact sequence splits, consider the natural injective $`C^{}`$-homomorphism $`i_\varphi :𝒞(B)𝒞_{\mathrm{cu}}(M).`$ This satisfies $`\iota i_\varphi =\mathrm{Id},`$ so $`i_\varphi ^{}`$ is a left inverse for $`\iota ^{}.`$
There is a correspondence between the short exact sequences of Lemmas 5.1 and 5.2. Consider the diagram
(5.8)
where the top row is the short exact sequence of Lemma 5.1 and the bottom row is the short exact sequence of Lemma 5.2. The central map is the quantization map of Lemma 4.3. Similarly, the map $`\mathrm{quan}_a`$ is the quantization map as discussed in .
The map $`\mathrm{quan}_{\mathrm{}}`$ on the left is essentially the same quantization map as in Lemma 4.3. Thus if $`NG_{\mathrm{e}\mathrm{sus}}^{\mathrm{}}(\varphi ,E_+)`$ there exists $`P=\mathrm{Id}+L,`$ $`L\mathrm{\Psi }_{\mathrm{cu}}^{\mathrm{}}(M/B;E_+)`$ with $`N(P)=N.`$ Choose an invertible, positive, approximate inverse square-root, $`Q\mathrm{Id}+\mathrm{\Psi }_{\mathrm{cu}}^{\mathrm{}}(M/B;E_+),`$
$$Q^2P^{}P\mathrm{Id}x^{\mathrm{}}\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^{\mathrm{}}(M/B;E_+)$$
and consider $`A=PQ.`$ Let $`_B`$ be the $`C(B)`$-Hilbert module
(5.9)
$$_B=_B^+_B^{}=\mathrm{L}^2(M;E)\mathrm{L}^2(M;E).$$
Then $`\mathrm{quan}_{\mathrm{}}(N)`$ is the $`\mathrm{KK}`$-class associated to the Kasparov module $`(_B,\mu _B,_B)`$ in $`𝔼_B(𝒞(B),𝒞(B))`$ where
(5.10)
$$_B=\left(\begin{array}{cc}0& A^{}\\ A& 0\end{array}\right)$$
and $`\mu _B:𝒞(B)(_B)`$ is just given by multiplication using the $`𝒞(B)`$-module structure.
###### Proposition 5.3.
The diagram (5.8) is commutative and $`\mathrm{quan}_{\mathrm{}},`$ $`\mathrm{quan}`$ and $`\mathrm{quan}_a`$ are isomorphisms.
###### Proof.
The fact that $`\mathrm{quan}_a`$ is an isomorphism is established in , where the Poincaré duality of is generalized. From the definition of $`\mathrm{quan}`$ and $`\mathrm{quan}_a,`$ it is straightforward to check that the right square of the diagram is commutative.
That $`\mathrm{quan}_{\mathrm{}}`$ is an isomorphism follows from the identification
$$\mathrm{KK}_B^0(𝒞(B),𝒞(B))\mathrm{KK}^0(,𝒞(B))\mathrm{K}(B)$$
and that under this $`\mathrm{quan}_{\mathrm{}}`$ gives the index map.
The commutativity of the left square is not quite obvious since it does not commute in terms of Kasparov modules. Indeed, if $`P\mathrm{Id}+K`$ is as discussed above then $`\iota ^{}\mathrm{quan}_{\mathrm{}}(P)`$ is represented by the Kasparov module $`(_B,\iota ^{}\mu _B,_B)`$ with $`_B`$ and $`_B`$ are as in (5.9) and (5.10) whereas $`\mathrm{quan}i_{}(P)`$ is represented by the Kasparov module $`(_B,\mu ,_B)`$ with $`\mu :𝒞_{\mathrm{cu}}(M)(_B)`$ given by multiplication by $`𝒞_{\mathrm{cu}}(M).`$ Since
$$\iota ^{}\mathrm{quan}_{\mathrm{}}(P)=\iota ^{}(i_\varphi )^{}\mathrm{quan}i_{}(P)$$
as Kasparov modules, the commutativity of the second square of the diagram, $`s^{}\mathrm{quan}i_{}(P)=0`$ shows that
$$\iota ^{}\mathrm{quan}_{\mathrm{}}(P)=\iota ^{}(i_\varphi )^{}\mathrm{quan}i_{}(P)=\mathrm{quan}i_{}(P)\text{ in }\mathrm{KK}_B^0(𝒞_{\mathrm{cu}}(M),𝒞(B))$$
and so these two modules give the same element in the K-group.
This implies that $`\mathrm{quan}`$ is also an isomorphism. ∎
## 6. The 6-term exact sequence
As noted above, it is always possible to perturb an elliptic family of cusp operators by a cusp operator of order $`\mathrm{}`$ so that it becomes invertible and this leads to the short exact sequence (11). In the general fibred cusp case there is an obstruction in K-theory to the existence of such a perturbation and this results in the 6-term exact sequence (22).
Consider a fibration with fibred cusp structure, as in (21), so that the algebra $`\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^{}(M/B)`$ of families of fibred cusp operators is well-defined. Let $`r_M`$ denote the inclusion
(6.1)
$$r_M:T_M^{}(M/B)T^{}(M/B).$$
If
(6.2)
$$d\mathrm{\Phi }:T_M(M/B)T(D/B)\times $$
is the (extended) differential of $`\mathrm{\Phi },`$ then identifying the tangent bundles with the cotangent bundles via some choice of metrics, there is a well-defined families index
$$\mathrm{ind}_{\mathrm{AS}}:K_c^0(T_M^{}(M/B))\mathrm{K}_\text{c}(T^{}(D/B)\times )\mathrm{K}_\text{c}^1(T^{}(D/B)).$$
###### Proposition 6.1.
An elliptic family, $`P\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^m(M/B;𝔼),`$ can be perturbed to be fully elliptic by the addition of some $`Q\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^{\mathrm{}}(M/B;𝔼)`$ if and only if the index of the class of its symbol at the boundary
(6.3)
$$\mathrm{ind}_{\mathrm{AS}}[r_M^{}\sigma (P)]=0\mathrm{K}_\text{c}^1(T^{}(D/B))$$
and then $`Q`$ can be chosen so that $`P+Q`$ is invertible.
###### Proof.
By the families index theorem of Atiyah and Singer, there is a suspended perturbation $`I(Q)`$ of order $`\mathrm{}`$ such that $`I(P)+I(Q)`$ is invertible if and only if (6.3) holds. With such a choice $`P+Q`$ is a fully elliptic family. Since it follows from Proposition 3.1 that the index map on perturbations of the identity is surjective, we may compose on the left with a Fredholm family of the form $`\mathrm{Id}+\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^{\mathrm{}}(M/B;E_{})`$ to get an operator of index zero and then perturb, as in Section 1 to make it invertible. ∎
In the particular case of a single elliptic cusp operator, the obstruction is in $`\mathrm{K}^1(\mathrm{pt})\{0\},`$ so there is no obstruction to such a perturbation. In the case of a family of elliptic cusp operators, the obstruction is in $`\mathrm{K}^1(D),`$ which is not the trivial group in general. However $`\mathrm{ind}_{\mathrm{AS}}r_M^{}\kappa (P)`$ is always zero in $`\mathrm{K}^1(D)`$ by the cobordism invariance of the index.
There is a parallel discussion in the odd case. The fibration (6.2) also induces an odd index
(6.4)
$$\mathrm{ind}_{\mathrm{AS}}:\mathrm{K}_\text{c}^1(T_M^{}(M/B))\mathrm{K}_\text{c}^1(T^{}(D/B)\times )\mathrm{K}_\text{c}(T^{}(D/B)).$$
###### Proposition 6.2.
Suppose $`P\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^0(M\times [0,1]/D\times [0,1];E)`$ is a family of elliptic operators which is the identity at $`t=0`$ and $`t=1`$ and let $`[\sigma (P)]\mathrm{K}_\text{c}^1(T^{}(M/B))`$ be the class of its symbol, then $`P`$ can be perturbed by $`Q\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^{\mathrm{}}(M\times [0,1]/D\times [0,1];E)`$ with $`Q|_{D\times \{0,1\}}=0`$ to be invertible if and only if
$$\mathrm{ind}_{\mathrm{AS}}r_M^{}[\sigma (P)]=0\mathrm{K}_\text{c}(T^{}(D/B)).$$
###### Proof.
One could proceed as in the proof of Proposition 6.1. However, there is an alternative proof which is more suggestive in this case. As discussed in , one can define the odd index (6.4) in the following way. Notice that $`r_M^{}\sigma (P),`$ which is the symbol of the indicial family, gives a class in
$$\mathrm{K}_\text{c}^1(T_M^{}(M/B))\mathrm{K}_\text{c}^1(T^{}(M/B)\times ).$$
If $`t[0,1]`$ denotes the suspension parameter, then by assumption $`r_M^{}\sigma (P)`$ is the identity at $`t=0`$ and $`t=1.`$ So there is no obstruction to perturb $`P`$ by smoothing operators so that one gets a 1-parameter family
(6.5)
$$tN(P_t)G_{\mathrm{\Phi }s(1)}^0(M/B;E)$$
such that $`N(P_0)=\mathrm{Id}`$ and $`N(P_1)G_{\mathrm{\Phi }s}^{\mathrm{}}(M/B;E),`$ where $`G_{\mathrm{\Phi }s(1)}^0(M/B;E)`$ is the group of invertible elliptic fibred suspended operators of order 0. Via the identification (if $`dimD=dimM,`$ this is only true after stabilization)
(6.6)
$$\pi _0(G_{\mathrm{\Phi }s}^{\mathrm{}}(M/B;E))\mathrm{K}_\text{c}^1(T^{}(D/B)),$$
the connected component in which $`N(P_1)`$ lies gives a class in $`\mathrm{K}_\text{c}^1(T^{}(D/B))`$ which is precisely $`\mathrm{ind}_{\mathrm{AS}}[r_M^{}\sigma (P)].`$ In particular, it does not depend on the choice of the 1-parameter family (6.5).
Now, if the index is zero, this means $`N(P_1)`$ is in the connected component of the identity in $`G_{\mathrm{\Phi }s}^{\mathrm{}}(M/B;E),`$ so via some smooth deformation, it can be arranged that $`N(P_1)=\mathrm{Id}`$ as well, so $`P`$ can be perturbed by $`Q\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^{\mathrm{}}(M/B;E)`$ with $`Q|_{D\times \{0,1\}}=0,`$ to become fully elliptic. Conversely, if such a perturbation $`Q`$ exists, then the index is given by the K-class corresponding to $`N(P_1+Q_1)=\mathrm{Id},`$ which is necessarily zero. ∎
The obstruction result of Proposition 6.1 and Proposition 6.2 indicates that the short exact sequence of Lemma 5.1 fails to be exact in the more general setting of fibred cusp operators. However, as in the case of the K-theory of a pair of spaces, there is a 6-term exact sequence given by
(6.7)
To define $`i_k`$ for $`k_2,`$ consider the group
$$G_{\mathrm{\Phi }s(1+k)}^{\mathrm{}}(M/B)\{\mathrm{Id}+Q;Q\mathrm{\Psi }_{\mathrm{\Phi }s(1+k)}^{\mathrm{}}(M/B),\mathrm{Id}+Q\text{ is invertible}\}.$$
Then using spectral sections techniques as in or the projections in Section 1, one has an identification
(6.8)
$$\mathrm{K}_\text{c}^\mathrm{k}(T^{}(D/B))\pi _0(G_{\mathrm{\Phi }s(1+k)}^{\mathrm{}}(M/B)).$$
Strictly speaking, the result is only true provided $`dimX>0,`$ but in the case $`dimX=0,`$ it is only necessary to allow some stabilization. Any element of $`G_{\mathrm{\Phi }s(1+k)}^{\mathrm{}}(M/B)`$ can be seen as the indicial family of a family of fully elliptic operators with symbol given by the identity. This gives a map
(6.9)
$$\pi _0(G_{\mathrm{\Phi }s(1+k)}^{\mathrm{}}(M/B))\mathrm{K}_{\mathrm{\Phi }\mathrm{cu}}^\mathrm{k}(\varphi )$$
and we define $`i_k`$ by composing (6.8) with (6.9). The symbol maps have already been defined and the boundary map $`I_0`$ is given by
$$I_0=\mathrm{ind}_{\mathrm{AS}}r_M^{}:\mathrm{K}_\text{c}(T^{}(M/B))\mathrm{K}_\text{c}^1(T^{}(D/B)),$$
while $`I_1`$ has a similar definition
$$I_1=\mathrm{ind}_{\mathrm{AS}}r_M^{}:\mathrm{K}_\text{c}^1(T^{}(M/B))\mathrm{K}_\text{c}^1(T^{}(D/B)\times )\mathrm{K}_\text{c}(T^{}(D/B)),$$
where the last identification is by Bott periodicity.
###### Theorem 6.3.
The diagram (6.7) is exact.
###### Proof.
The exactness at $`\mathrm{K}_{\mathrm{\Phi }\mathrm{cu}}(\varphi )`$ and $`\mathrm{K}_{\mathrm{\Phi }\mathrm{cu}}^1(\varphi )`$ follows rather directly from the definition since homotopies of the symbol can be lifted to homotopies of the joint symbol, see the discussion above in the cusp case. Exactness at $`\mathrm{K}_\text{c}(T^{}(M/B))`$ and $`\mathrm{K}_\text{c}^1(T^{}(M/B))`$ follows from Proposition 6.1 and its suspended version.
The exactness at $`\mathrm{K}_\text{c}(T^{}(D/B))`$ can be seen from the alternative definition of the odd index used in the proof of Proposition 6.2. Indeed, suppose that $`\alpha \mathrm{K}_\text{c}(T^{}(D/B))`$ is in the image of $`I_1.`$ According to the proof of Proposition 6.2, this means that there is a family of fully elliptic operators $`tP_t\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^0(M/B;E)`$ such that $`P_0=\mathrm{Id},`$ $`\sigma (P_1)=\mathrm{Id}`$ and
$$[N(P_1)]\pi _0(G_{\mathrm{\Phi }s}^{\mathrm{}}(M/B;E)\mathrm{K}_{\text{c}}^{}{}_{}{}^{0}(T^{}(D/B))$$
corresponds to $`\alpha .`$ The homotopy $`t(\sigma (P_t),N(P_t))`$ between the identity and $`(\mathrm{Id},N(P_1))`$ then shows that $`i(\alpha )=0`$ in $`\mathrm{K}_{\mathrm{\Phi }\mathrm{cu}}(\varphi ).`$
Conversely, suppose that $`i(\alpha )=0.`$ If $`N(P_1)G_{\mathrm{\Phi }s}^{\mathrm{}}(M/B;E)`$ represents $`\alpha ,`$ then after some stabilization, one can assume that there is a homotopy of invertible joint symbols $`t(\sigma (P_t),N(P_t))`$ between $`(\mathrm{Id},\mathrm{Id})`$ and $`(\mathrm{Id},N(P_1)).`$ The family symbol $`t\sigma (P_t)`$ then defines a class $`\beta \mathrm{K}_\text{c}^1(T^{}(M/B))`$ such that $`I_1(\beta )=\alpha ,`$ which establishes the exactness at $`\mathrm{K}_\text{c}(T^{}(D/B)).`$
To prove the exactness at $`\mathrm{K}_\text{c}^1(T^{}(D/B)),`$ one can proceed in a similar way. Indeed, the diagram
(6.10)
commutes, where the vertical arrows are given by Bott periodicity and $`I_2=\mathrm{ind}_{\mathrm{AS}}r_M^{}`$ in terms of the families index map
$$\mathrm{ind}_{\mathrm{AS}}:\mathrm{K}_\text{c}^2(T_M^{}(M/B))\mathrm{K}_\text{c}^2(T^{}(D/B)\times )\mathrm{K}_\text{c}^3(T^{}(D/B))$$
and $`j`$ is the map (6.9) obtained using the identification (when $`dimD=dimM,`$ this is only true after stabilization)
(6.11)
$$\mathrm{K}_\text{c}^3(T^{}(D/B))\pi _0(G_{\mathrm{\Phi }s(2)}^{\mathrm{}}(M/B)).$$
Notice that as opposed to (6.8), no Bott periodicity is involved in (6.11), hence the right triangle in (6.10) is commutative. The fact that the left square is commutative follows from the commutativity of the families index with Bott periodicity (and more generally with the Thom isomorphism).
The exactness of the bottom row of (6.10) can be proved by applying the proof of the exactness at $`\mathrm{K}_\text{c}(T^{}(D/B))`$ (really $`\mathrm{K}_\text{c}^2(T^{}(D/B)))`$ with one extra suspension parameter. Since the Bott periodicity maps in (6.10) are isomorphisms, this implies that
$$\mathrm{K}_\text{c}(T^{}(M/B))\stackrel{I_0}{}\mathrm{K}_\text{c}^1(T^{}(D/B))\stackrel{i_1}{}\mathrm{K}_{\mathrm{\Phi }\mathrm{cu}}^1(\varphi )$$
is also exact in the middle. ∎
In the scattering case (when stabilization is necessary in the arguments above), using Lemma 2.1, the 6-term exact sequence (6.7) reduces to the exact sequence in K-theory associated to the inclusion of the boundary of $`T^{}(M/B)`$
(6.12)
## 7. Poincaré duality, the general case
Consider the short exact sequence of $`C^{}`$ algebras (23) and the associated 6-term exact sequence
(7.1)
Each term of this sequence can be related to the corresponding term in $`(\text{6.7})`$ via a quantization map, namely the quantization maps of Section 4
(7.2)
$$\mathrm{quan}:\mathrm{K}_{\mathrm{\Phi }\mathrm{cu}}^\mathrm{k}(\varphi )\mathrm{KK}_B^k(𝒞_\mathrm{\Phi }(M),𝒞(B)),k_2,$$
and the quantization maps of and of Atiyah and Singer,
(7.3)
$$\mathrm{quan}_r:\mathrm{K}_\text{c}^\mathrm{k}(T^{}(M/B))\mathrm{KK}_B^k(𝒞_0(M),𝒞(B)),$$
(7.4)
$$\mathrm{quan}_{\mathrm{}}:\mathrm{K}_\text{c}^\mathrm{k}(T^{}(D/B))\mathrm{KK}_B^k(𝒞(D),𝒞(B)).$$
###### Theorem 7.1.
The quantization maps $`\mathrm{quan},`$ $`\mathrm{quan}_r,`$ and $`\mathrm{quan}_{\mathrm{}}`$ are isomorphisms giving a commutative diagram
between the 6-term exact sequences (6.7) and (7.1). In particular, this implies the Poincaré duality of Theorem 2.
The fact that $`\mathrm{quan}_r`$ and $`\mathrm{quan}_{\mathrm{}}`$ are isomorphisms follows from the Poincaré duality result of Kasparov and its extensions by the first author and Piazza in . Then, given the commutativity of the diagram between the two 6-term exact sequence, the fact that $`\mathrm{quan}`$ is an isomorphism follows from the fives lemma. So it remains to check that the diagram commutes.
For $`k_2,`$ the commutativity of
(7.5)
is clear from the definition of $`\mathrm{quan}`$ and $`\mathrm{quan}_r.`$ The proof of the commutativity of the four other squares of the diagram is more involved and will occupy the remainder of this section.
Let us first consider the commutativity of
(7.6)
We will only provide a proof in the even case since the odd case is similar. The first step is to describe the quantization map $`\mathrm{quan}_{\mathrm{}}`$ in terms of indicial families instead of symbols. Given a class $`\alpha \mathrm{K}_\text{c}(T^{}(D/B)),`$ let $`pG_{\mathrm{\Phi }s}^{\mathrm{}}(M/B;E_+)`$ be a representative of this class. Consider the manifold
$$M_c=M\times [0,1]$$
which can be seen as a collar neighborhood $`M`$ in $`M.`$ Consider the associated fibration
(7.7)
$$\varphi _c=\varphi \pi _1:M_cB,$$
where $`\pi _1:M\times [0,1]M`$ is the projection on the first factor. Since the boundary of $`M_c`$ has two parts $`M_0=M\times \{0\}`$ and $`M_1=M\times \{1\},`$ let
$$\mathrm{\Phi }_i:M_iD_i,i\{0,1\},$$
denote the two boundary fibration maps and let
$$\mathrm{\Phi }_c:M_cD_0D_1$$
be the total boundary map. Let $`P(\mathrm{Id}+\mathrm{\Psi }_{\mathrm{\Phi }_\mathrm{c}\mathrm{cu}}^{\mathrm{}}(M_c;E_+))`$ be a Fredholm operator with indicial family the identity at $`M_0`$ and by $`p`$ at $`M_1.`$ Then Lemma 4.1 gives an associated $`\mathrm{KK}`$-class
$$[P]\mathrm{KK}_B^0(𝒞_{\mathrm{\Phi }_c}(M_c),𝒞(B))$$
and the pull-back map
$$d^{}:\mathrm{KK}_B^0(𝒞_{\mathrm{\Phi }_c}(M_c),𝒞(B))\mathrm{KK}_B^0(𝒞(D),𝒞(B))$$
where $`d:𝒞(D)𝒞_{\mathrm{\Phi }_c}(M_c)`$ is the obvious inclusion, gives a KK-class
$$d^{}[P]\mathrm{KK}_B^0(𝒞(D),𝒞(B)).$$
Thus, this procedure gives a well-defined quantization map
$$\mathrm{quan}_{\mathrm{}}^{}:\mathrm{K}_\text{c}(T^{}(M/D))\mathrm{KK}_B^0(𝒞(D),𝒞(B)).$$
###### Lemma 7.2.
$`\mathrm{quan}_{\mathrm{}}^{}=\mathrm{quan}_{\mathrm{}}.`$
###### Proof.
Let $`\alpha \mathrm{K}_\text{c}(T^{}(M/D))`$ be given. Following the discussion in §3 we are reduced to the scattering case, so
$$M_c=D\times [0,1]$$
in our construction. Let $`p`$ be an indicial family representing the K-class $`\alpha `$ and let $`P`$ be as above. Let $`P_t\mathrm{\Psi }_{\mathrm{\Phi }_\mathrm{c}\mathrm{cu}}^0(M_c/B;E_+),`$ $`t[0,1]`$ be a homotopy through families of fully elliptic operators such that $`P_0=P`$ and $`P_1`$ has trivial indicial families both at $`M_0`$ and $`M_1`$ and so with symbol $`\sigma `$ having K-class
$$[\sigma ]\mathrm{K}_\text{c}(T^{}(M_c/B),T_{M_c}^{}(M_c/B))\mathrm{K}_\text{c}(T^{}(D\times (0,1)/B))$$
identifying via Bott periodicity with $`\alpha `$ (see for details). That such a homotopy exists is discussed in and §2. By considering a family of positive definite approximate inverse square roots for $`P_t,`$ we construct an operator homotopy $`(,\mu ,_t)`$ of Kasparov modules, which means that $`P_0`$ and $`P_1`$ define the same element in $`\mathrm{KK}_B^0(𝒞(D),𝒞(B)).`$
On the other hand, let us quantize partially the symbol $`\sigma `$ of $`P_1`$ in the $`T^{}[0,1]`$ direction to get a family of elliptic operators $`\widehat{p}_1`$ parameterized by $`T^{}(D/B)`$ and acting on the Hilbert bundle
$$\mathrm{L}^2(M_c/D;E_+)T^{}(D/B)$$
with typical fibre $`\mathrm{L}^2([0,1];E_+).`$ As discussed in , we can interpret
$$\widehat{p}_1:\mathrm{L}^2(M_c/D;E_+)\mathrm{L}^2(M_c/D;E_+)$$
as a symbol on a Hilbert bundle. The notion of ellipticity leads to very restrictive conditions in this context. But according to example 1.7 in , $`\widehat{p}_1`$ is elliptic and once it is quantized we get back $`P_1`$ modulo compact operators. Assume without loss of generality that $`\sigma `$ (and $`\widehat{p}_1`$) is homogeneous of degree zero in the fibres of $`T^{}(D/B)D`$ outside a compact neighborhood of the zero section. Assume also without loss of generality that $`E_+`$ is a trivial bundle on $`M_c.`$ Then as discussed in , it is possible to deform $`\widehat{p}_1`$ through a homotopy $`\widehat{p}_t,`$ $`t[1,2]`$ of elliptic symbols so that $`\widehat{p}_2`$ takes the form
$$\widehat{p}_2=\widehat{p}_2^{}\widehat{p}_2^{\prime \prime }:VV^{}\widehat{p}_1V(\widehat{p}_1V)^{}$$
where $`V`$ is a sub-bundle of $`\mathrm{L}^2(M_c/D;E)`$ of finite corank on which $`\widehat{p}_1`$ is injective and where $`\widehat{p}_2^{}`$ is invertible and constant along the fibres of $`T^{}(D/B)D.`$ Thus, when we quantize $`\widehat{p}_2`$ we get an operator of the form
$$P_2=P_2^{}P_2^{\prime \prime }:\mathrm{L}^2(D;V)\mathrm{L}^2(D;V^{})\mathrm{L}^2(D;\widehat{p}_1V)\mathrm{L}^2(D;(\widehat{p}_1V)^{})$$
with $`P_2^{}`$ invertible. Using positive definite approximate inverse square roots of $`(P_2^{})^{}P_2^{}`$ and $`(P_2^{\prime \prime })^{}P_2^{\prime \prime },`$ we can associate a KK-class $`\beta \mathrm{KK}_B^0(𝒞(D),𝒞(B)).`$ By homotopy invariance, $`P_1`$ gives the same K-class so $`\beta =\mathrm{quan}_{\mathrm{}}^{}(\alpha ).`$ On the other hand, the Kasparov module coming from $`P_2^{}`$ is degenerate, so $`\beta `$ can be defined using $`P_2^{\prime \prime }.`$ The K-class associated to the symbol of $`P_2^{\prime \prime }`$ is the families index of $`\widehat{p}_1`$ (cf. ) which by the families index of Atiyah-Singer should be precisely $`\alpha \mathrm{K}_\text{c}(T^{}(M/D)).`$ This means that
$$\mathrm{quan}_{\mathrm{}}^{}(\alpha )=\beta =\mathrm{quan}_{\mathrm{}}(\alpha ).$$
Thus we can use indicial families instead of symbols to define the quantization map $`\mathrm{quan}_{\mathrm{}}.`$ Consider again the manifold $`M_c.`$ Since the boundary of $`M_c`$ has two disconnected parts (which can themselves be disconnected), we can consider, instead of $`\mathrm{K}_{\mathrm{\Phi }_\mathrm{c}\mathrm{cu}}(\varphi _c),`$ the group $`\mathrm{K}_{\mathrm{\Phi }_1\mathrm{cu}}(\varphi _c)`$ of stabilized homotopy classes of invertible joint symbols with indicial family given by the identity at the boundary face $`M_0.`$ If we define $`𝒞_{\mathrm{\Phi }_1}(M_c)`$ to be the $`C^{}`$ algebra of continuous functions
(7.8)
$$𝒞_{\mathrm{\Phi }_1}(M_c)=\left\{f𝒞(M_c);f|_{M_1}=\mathrm{\Phi }_1^{}g\text{ for some }g𝒞(D_1)\right\},$$
then a quantization map
$$\mathrm{quan}_c:\mathrm{K}_{\mathrm{\Phi }_1\mathrm{cu}}(\varphi _c)\mathrm{KK}_B^0(𝒞_{\mathrm{\Phi }_1}(M_c),𝒞(B))$$
can be defined as follows. Given a joint symbol $`(\sigma ,N)`$ representing a class in $`\alpha \mathrm{K}_{\mathrm{\Phi }_1\mathrm{cu}}(\varphi _c),`$ one considers a family of fibred cusp operators $`P`$ of order zero with joint symbol given by $`(\sigma ,N).`$ Deforming $`\sigma `$ if needed, we can assume $`P`$ acts as the identity in a small collar neighborhood of $`M_0.`$ Then in the usual fashion, one can construct a Kasparov module in $`\mathrm{KK}_B^0(𝒞_{\mathrm{\Phi }_1}(M_c),𝒞(B))`$ by considering a positive definite approximate inverse square root to P which acts as the identity in a collar neighborhood of $`M_0.`$ As in the definition of $`\mathrm{quan},`$ one can check that the associate KK-class only depends on $`\alpha \mathrm{K}_{\mathrm{\Phi }_1\mathrm{cu}}(\varphi _c).`$
There is also a natural map
(7.9)
$$i_c:\mathrm{K}_\text{c}(T^{}(M/D))\mathrm{K}_{\mathrm{\Phi }_1\mathrm{cu}}(\varphi _c)$$
which to $`\alpha \mathrm{K}_\text{c}(T^{}(M/D))`$ associates the class $`i_c(\alpha )`$ represented by any joint symbol with trivial indicial family on $`M_0,`$ indicial family with K-class given by $`\alpha `$ on $`M_1`$ and with trivial symbol.
###### Lemma 7.3.
The diagram
is commutative, where $`\iota _c:𝒞_{\mathrm{\Phi }_1}(M_c)𝒞(D)`$ is the restriction to $`M_1.`$
###### Proof.
Let $`\alpha \mathrm{K}_\text{c}(T^{}D/B)`$ be given. Let $`e:𝒞(D)𝒞_{\mathrm{\Phi }_1}(M_c)`$ be the obvious injective map of $`C^{}`$ algebras so that $`\iota _ce=\mathrm{Id}.`$ Then
(7.10)
$$\iota _c^{}\mathrm{quan}_{\mathrm{}}(\alpha )=\iota _c^{}e^{}\mathrm{quan}_ci_c(\alpha ).$$
Consider the $`C^{}`$ algebra
$$𝒞_1(M_c)=\left\{f𝒞(M_c);f|_{M_1}=0\right\}.$$
From the commutativity of the diagram
(7.11)
and the exactness in the middle of
(7.12)
$$\mathrm{K}_\text{c}(T^{}(D/B))\stackrel{i_c}{}\mathrm{K}_{\mathrm{\Phi }_1\mathrm{cu}}(\varphi _c)\stackrel{𝜎}{}\mathrm{K}_\text{c}(T^{}(M_c/B)),$$
(7.13)
$$\mathrm{KK}_B^0(𝒞(D),𝒞(B))\stackrel{\iota _c}{}\mathrm{KK}_B^0(𝒞_{\mathrm{\Phi }_1}(M_c),𝒞(B))\mathrm{KK}_B^0(𝒞_1(M_c),𝒞(B)),$$
we deduce that there exists $`\beta \mathrm{KK}_B^0(𝒞(D),𝒞(B))`$ such that $`\iota _c^{}\beta =\mathrm{quan}_ci_c(\alpha ).`$ Since $`\iota _ce=\mathrm{Id},`$
$$\beta =e^{}\iota _c^{}\beta =e^{}\mathrm{quan}_ci_c(\alpha ),$$
which implies by (7.10) that
$$\mathrm{quan}_ci_c(\alpha )=\iota _c^{}\beta =\iota _c^{}(e^{}\mathrm{quan}_ci_c(\alpha ))=\iota _c^{}\mathrm{quan}_{\mathrm{}}(\alpha ).$$
###### Lemma 7.4.
For $`k_2,`$ the diagram (7.6) is commutative.
###### Proof.
As noted above, we will only provide a proof for the case $`k=0,`$ the case $`k=1`$ being similar. Think of $`M_c`$ as a collar neighborhood of $`M`$ where $`M_1`$ is identified with $`M.`$ Let
$$j:𝒞_\mathrm{\Phi }M)𝒞_{\mathrm{\Phi }_1}(M_c)$$
be the restriction map on $`M_cM.`$ Then consider the diagram
(7.14)
where $`e_c`$ is the extension of a representative in $`\mathrm{K}_{\mathrm{\Phi }_1\mathrm{cu}}(\varphi _c)`$ by the identity inside $`M.`$ Clearly, $`e_ci_c=i`$ and $`j^{}\iota _c^{}=\iota ^{},`$ so the lemma will be proven provided we show the diagram (7.14) is commutative. By the previous lemma, we only need to show that the right square is commutative. Consider the two graded Hilbert modules over $`𝒞(B)`$
(7.15)
$$_1=\mathrm{L}^2(M_c;𝔼)=\mathrm{L}^2(M_c;E_+)\mathrm{L}^2(M_c;E_{})$$
(7.16)
$$_2=\mathrm{L}^2(\overline{MM_c};𝔼)=\mathrm{L}^2(\overline{MM_c};E_+)\mathrm{L}^2(\overline{MM_c};E_{}).$$
Then given $`\alpha \mathrm{K}_{\mathrm{\Phi }_1\mathrm{cu}}(\varphi _c),`$ we can represent $`j^{}\mathrm{quan}_c(\alpha )`$ by a Kasparov module of the form $`(_1,j^{}\mu ,_1)`$ and at the same time $`\mathrm{quan}e_c(\alpha )`$ can be represented by a Kasparov module of the form $`(_1_2,j^{}\mu l^{}\nu ,_1_2)`$ where
$$l:𝒞_\mathrm{\Phi }M)𝒞(\overline{MM_c})$$
is the restriction map,
(7.17)
$$\mu :𝒞_{\mathrm{\Phi }_1}(M_c)(_1)$$
(7.18)
$$\nu :𝒞(\overline{MM_c})(_2)$$
are the obvious actions given by multiplication and
$$_2=\left(\begin{array}{cc}0& \mathrm{Id}\\ \mathrm{Id}& 0\end{array}\right).$$
Since $`(_2,\nu ,_2)`$ is a degenerate Kasparov module,
$$j^{}\mathrm{quan}_c(\alpha )=\mathrm{quan}e_c(\alpha ).$$
Finally we need to show that the diagram
(7.19)
is commutative for $`k_2,`$ which requires an understanding of the boundary homomorphism $`\delta .`$ Let us again limit our attention to the even case $`k=0,`$ the case $`k=1`$ being similar. Recall that in topological K-theory (see for instance ), the boundary homomorphism of the 6-term exact sequence associated to a pair of spaces $`(X,Y)`$ is defined via the cone space $`XCY`$ which is obtained from $`X`$ and
$$CY=Y\times [0,1]/Y\times \{1\}$$
by identifying $`YX`$ with $`Y\times \{0\}CY.`$ There is an exact sequence
$$\mathrm{K}(XCY,X)\stackrel{m^{}}{}\stackrel{~}{\mathrm{K}}(XCY)\stackrel{k^{}}{}\stackrel{~}{\mathrm{K}}(X).$$
Under the identifications
$$\stackrel{~}{\mathrm{K}}^1(Y)\mathrm{K}(XCY,X),\mathrm{K}(X,Y)\stackrel{~}{\mathrm{K}}(XCY),$$
this becomes
$$\stackrel{~}{\mathrm{K}}^1(Y)\stackrel{𝛿}{}\mathrm{K}(X,Y)\stackrel{~}{\mathrm{K}}(X)$$
where $`\delta `$ is the boundary homomorphism of the 6-term exact sequence associated to the pair $`(X,Y).`$
In $`\mathrm{KK}`$-theory, one can define the boundary homorphism in a similar way, introducing the mapping cone
(7.20)
$$𝒞_\iota =\left\{(x,f)𝒞_\mathrm{\Phi }(M)𝒞_0([0,1)\times D);\iota (x)=f(0)\right\},$$
where
$$𝒞_0([0,1)\times D)=\left\{f𝒞([0,1]\times D);f|_{\{1\}\times D}=0\right\}$$
is the $`C^{}`$ closure of $`𝒞_c([0,1)\times D).`$ There is a natural inclusion
(7.21)
$$e:𝒞_0(M)x(x,0)𝒞_\iota $$
which induces a map
(7.22)
$$e^{}:\mathrm{KK}_B^0(𝒞_\iota ,𝒞(B))\mathrm{KK}_B^0(𝒞_0(M),𝒞(B)).$$
This map is an isomorphism as can be seen by interpreting the map $`e^{}`$ as multiplication (on the left) by $`[e]\mathrm{KK}_B^0(𝒞_0(M),𝒞_\iota )`$ associated to the $`C^{}`$ homomorphism $`e`$ (see). Then $`[e]`$ is a $`\mathrm{KK}`$-equivalence between $`𝒞_0(M)`$ and $`𝒞_\iota ,`$ which is to say that it has an inverse $`u\mathrm{KK}_B^0(𝒞_\iota ,𝒞_0(M))`$ with respect to the Kasparov product.
Given this one can then define the boundary homomorphism as
(7.23)
$$\delta =j^{}(e^{})^1:\mathrm{KK}_B(𝒞_0(M),𝒞(B))\mathrm{KK}_B(𝒮𝒞(D),𝒞(B))\mathrm{KK}_B^1(𝒞(D),𝒞(B))$$
where $`j:𝒮𝒞(D)𝒞_\iota `$ is the natural inclusion. In terms of Kasparov product, $`\delta `$ is multiplication on the left by $`j^{}u\mathrm{KK}_B(𝒮𝒞(D),𝒞_0(M)).`$
In the present context, it is possible to give an alternative definition of the mapping cone which is especially useful. Consider as before a collar neighborhood $`M_c=M\times [0,1]`$ of $`M`$ with $`M`$ identifies with $`M_1=M\times \{1\}.`$ Then
(7.24)
$$𝒞_\iota \left\{f𝒞_0(M);f|_{M_c}=(\mathrm{\Phi }\times \mathrm{Id})^{}g\text{ for some }g𝒞_0(D\times [0,1))\right\}$$
by identifying $`M^{}=\overline{M/M_c}`$ with $`M.`$ The isomorphism (7.24) gives
(7.25)
$$\epsilon :𝒞_\iota 𝒞_0(M).$$
The map $`e`$ in (7.21) is then described by
$$e:𝒞_0(M)\stackrel{~}{}𝒞_0(M^{})𝒞_\iota .$$
###### Lemma 7.5.
The boundary homomorphism $`\delta `$ of the 6-term exact sequence (7.1) is given by the pull-back map
$$j^{}\epsilon ^{}:\mathrm{KK}_B^0(𝒞_0(M),𝒞(B))\mathrm{KK}_B^0(𝒮𝒞(D),𝒞(B)).$$
###### Proof.
If $`\sigma `$ is a symbol on $`T^{}(M/B)`$ then $`e^{}\epsilon ^{}\mathrm{quan}_r(\sigma )`$ can be obtained by quantizing the symbol
$$\sigma ^{}=\sigma |_{T^{}(M^{}/B)}$$
on $`T^{}(M^{}/B)`$ and making the identification $`𝒞_0(M^{})𝒞_0(M).`$ Since $`\sigma ^{}`$ and $`\sigma `$ are homotopic when $`M^{}`$ and $`M`$ are identified, they correspond to the same K-class. Since $`\sigma `$ was arbitrary, this shows $`\mathrm{quan}_r=e^{}\epsilon ^{}\mathrm{quan}_r.`$ Consequently,
$$(e^{})^1\mathrm{quan}_r=(e^{})^1e^{}\epsilon ^{}\mathrm{quan}_r=\epsilon ^{}\mathrm{quan}_r.$$
Since $`\mathrm{quan}_r`$ is an isomorphism, $`(e^{})^1=\epsilon ^{}`$ and it follows that $`\delta =j^{}(e^{})^1=j^{}\epsilon ^{}.`$
After identifying the tangent bundle with the cotangent bundle via some metric, The boundary fibration induces a fibration
(7.26)
$$\pi _\mathrm{\Phi }:T^{}(M_c/B)T^{}(D\times [0,1]/B).$$
Let
(7.27)
$$\mathrm{ind}_{\pi _\mathrm{\Phi }}:\mathrm{K}_\text{c}(T^{}(M_c/B))\mathrm{K}_\text{c}(T^{}(D\times [0,1]/B))$$
be the topological index family map associated to this fibration.
###### Lemma 7.6.
The diagram
is commutative, where
$$\epsilon _c:𝒞_0(M_c)𝒞_0(M),\epsilon _d:𝒮𝒞(D)𝒞_0(M_c),r:T^{}(M_c/B))T^{}(M/B),$$
are the natural inclusions.
###### Proof.
The commutativity of the first square is clear. The proof of the commutativity of the second square is essentially a consequence of the Atiyah-Singer families index theorem. Indeed, it allows us to define $`\mathrm{ind}_{\pi _\mathrm{\Phi }}`$ as an analytical families index. Then, using quantization of symbols over Hilbert bundles as in , one can represent (cf. the proof of Lemma 7.2) $`\epsilon _d^{}\mathrm{quan}_r(\alpha )`$ by a Kasparov module of the form
$$(_1_2,\mu ,_1_2)$$
where
$$\begin{array}{c}_1=\mathrm{L}^2(D\times [0,1];V_+)\mathrm{L}^2(D\times [0,1];V_{}),\\ _2=\mathrm{L}^2(D\times [0,1];V_+^{})\mathrm{L}^2(D\times [0,1];V_{}^{}),\end{array}$$
and $`V_\pm `$ is a sub-bundle of $`\mathrm{L}^2(M_c/(D\times [0,1]);E_\pm )`$ with $`V_+`$ and $`V_{}`$ of same finite corank. As usual,
$$\mu :𝒞_0(D\times [0,1])(_i)$$
denotes multiplication. One can do this in such a way that $`(_1,\mu ,_1)`$ is degenerate and $`(_2,\mu ,_2)`$ represents $`\mathrm{quan}_r\mathrm{ind}_{\pi _\mathrm{\Phi }}(\alpha ),`$ which establishes the commutativity of the left square. ∎
###### Lemma 7.7.
We have the identity $`\delta \mathrm{quan}_r=\mathrm{quan}_r\kappa I_0`$ where
$$\kappa :\mathrm{K}_\text{c}^1(T^{}(D/B))\mathrm{K}_\text{c}(T^{}(D\times [0,1]/B))$$
is the canonical isomorphism induced from the homotopy equivalence
$$(T^{}([0,1]),\mathrm{})(,\mathrm{}).$$
###### Proof.
Since clearly $`\epsilon _d^{}\epsilon _c^{}=j^{}\epsilon ^{}=\delta ,`$ we deduce from Lemma 7.6 that
$$\delta \mathrm{quan}_r=\epsilon _d^{}\epsilon _c^{}\mathrm{quan}_r=\mathrm{quan}_r\mathrm{ind}_{\pi _\mathrm{\Phi }}r^{},$$
so we can conclude from the fact that $`\mathrm{ind}_{\pi _\mathrm{\Phi }}r^{}=\kappa I_0,`$ which is a consequence of the commutativity of the topological index map with the Thom isomorphism. ∎
Thus, Lemma 7.7 reduces the proof of the commutativity of (7.19) in the even case to the following statement.
###### Lemma 7.8.
The diagram
is commutative, where the bottom arrow is Bott periodicity in KK-theory.
###### Proof.
This essentially follows from the family version of proposition 11.2.5 in and the alternative definition of odd quantization using self-adjoint operators. ∎
In the odd case, a similar argument reduces the proof of the commutativity of (7.19) to the following lemma.
###### Lemma 7.9.
The diagram
is commutative, where the top and bottom arrows are Bott periodicity in K-theory and KK-theory respectively.
###### Proof.
This essentially follows from the family version of proposition 11.2.5 in and the alternative definition of odd quantization using self-adjoint operators. ∎
## 8. Adiabatic passage to the cusp case
As noted in the introduction, and confirmed by the properties of the 6-term exact sequence above, it is the ‘cusp’ case amongst the possible fibred cusp structures on a give fibration which is universal.
###### Proposition 8.1.
For any fibration with fibred cusp structure there is a natural map, given by an adiabatic limit, $`q_{\mathrm{ad}}:\mathrm{K}_{\mathrm{\Phi }\mathrm{cu}}(\varphi )\mathrm{K}_{\mathrm{cu}}(\varphi )`$ into the cusp K-group which commutes with the analytic index and gives a commutative diagram
(8.1)
###### Proof.
Given a family $`P\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^0(M;𝔼)`$ we construct an adiabatic family $`P(ϵ)\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{ad},\mathrm{cu}}^0(M;𝔼)`$ which are cusp pseudodifferential operators for $`ϵ>0`$ but degenerate, in the adiabatic limit, to the fibred cusp operator $`P`$ at $`ϵ=0.`$ However, it is perhaps better to think of the construction as being in the opposite direction. The definition and properties of such adiabatic fibred cusp operators are discussed in Appendix C where they are defined directly through their Schwartz kernels which are defined on the space given by blow-ups in (C.10), in our case with the finer fibration of the boundary being the given one, $`\mathrm{\Phi },`$ and the coarser fibration being $`\varphi `$ corresponding to the cusp structure.
Thus there are three boundary faces of the double space in (C.10) which meet the diagonal (and the kernels are supposed to vanish rapidly at all other boundary faces). At the ‘old’ face $`ϵ=0`$ we wish to recover the given operator $`P`$ by the map (C.15); this fixes the kernel precisely on that face. The adiabatic front face can be constructed, as in (C.10), by the last blow-up and it is fibred over the lifted variable
(8.2)
$$\tau =\frac{xϵ}{x+ϵ}[1,1]$$
representing the adiabatic passage, with fibre which is just the front face for the fibred cusp calculus. On the fibre at $`\tau =1`$ the kernel is already fixed to be that of $`P.`$ Thus, we may simply choose to extend the kernel to the whole adiabatic face to be independent of the ‘angle’ $`\tau .`$ This fixes the kernel of the adiabatic normal operator and also fixes the boundary value, corresponding to $`\tau =1,`$ of the kernel on the front face for the cusp blow-up as $`ϵ0.`$ Simply extend it to that face as a conormal distribution with respect to the diagonal; in fact we also choose an extension which is conormal to the diagonal in the interior and consistent with these boundary values (and also the requirement of the vanishing at the ‘non-diagonal’ boundaries).
This fixes an element $`P(ϵ)\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{ad},\mathrm{cu}}^0(M;𝔼).`$ The choices ensure that in the sense of (C.15)
(8.3)
$$A(P(ϵ))=P\text{ and }\mathrm{ad}(P)^1\text{ exists,}$$
since the invertibility here comes from the invertibility of the normal operator of $`P.`$ Now, for $`ϵ\delta ,`$ for some $`\delta >0,`$ the hypotheses of Proposition C.1 apply and show that $`P(\delta )\mathrm{\Psi }_{\mathrm{cu}}^0(M;𝔼)`$ is fully elliptic. Thus we can define
(8.4)
$$q_{\mathrm{ad}}:\mathrm{K}_{\mathrm{\Phi }\mathrm{cu}}(\varphi )[P][P(\delta )]\mathrm{K}_{\mathrm{cu}}(\varphi )$$
provided the independence of choices is shown.
Clearly the image in (8.4) is independent of the choice of $`\delta `$ small and the choices in the construction can all be related by homotopies. Similarly an homotopy of $`P`$ through fully elliptic fibred cusp operators lifts to a homotopy of $`P(\delta )`$ in fully elliptic cusp operators and the behaviour under stabilization and composition with bundle isomorphism is appropriate to guarantee that (8.4) is well defined. The construction also ensures that this adiabatic limit commutes with the passage to the symbol, giving commutativity in (8.1). ∎
Under the Poincaré duality of Section 7, the corresponding adiabatic map in KK-theory is given by pull-back. Let $`\iota _{\mathrm{ad}}:𝒞_{\mathrm{cu}}(M)𝒞_\mathrm{\Phi }(M)`$ be the natural inclusion.
###### Proposition 8.2.
The diagram
is commutative.
###### Proof.
If $`s:𝒞_0(M)𝒞_{\mathrm{cu}}(M)`$ and $`i:𝒞(B)𝒞_{\mathrm{cu}}(M)`$ are the natural inclusions, then from the fact that $`q_{\mathrm{ad}}`$ preserves the index and the homotopy class of the symbol, we see that
$$i^{}\iota _{\mathrm{ad}}^{}\mathrm{quan}=i^{}\mathrm{quan}q_{\mathrm{ad}}\text{and}s^{}\iota _{\mathrm{ad}}^{}\mathrm{quan}=s^{}\mathrm{quan}q_{\mathrm{ad}}.$$
Thus, the result follows from the split short exact sequence of Lemma 5.2. ∎
## 9. The extension of fibred cusp structures
The definition of the topological index uses an embedding of a given family of cusp structures into a product setting. In fact we show how to embed a general fibred cusp structure (since this can also be used to define the index without using Proposition 8.1). We shall use as ‘model’ structure (for a given base manifold $`B)`$ the products
(9.1)
$$\begin{array}{c}\stackrel{~}{M}=B\times 𝔹^{p+1},p>0,\text{ for cusp structures and}\\ \stackrel{~}{M}=B\times 𝔹^{p+1}\times 𝕊^q,p,q>0\text{ for fibred cusp structures.}\end{array}$$
The model fibration, being just projection onto the first factor will be written
(9.2)
$$\pi :\stackrel{~}{M}B$$
and in the case of fibred cusp structures the model boundary fibration is the projection onto the first two factors (restricted to the boundary)
(9.3)
$$\pi ^{}:\stackrel{~}{M}\stackrel{~}{D}=B\times 𝕊^p.$$
###### Definition 9.1.
A fibration with fibred cusp structure $`\stackrel{~}{\varphi }:\stackrel{~}{M}B`$ is an *extension* of a given fibration with fibred cusp structure $`\varphi :MB`$ if $`i:M\stackrel{~}{M}`$ and $`j:D\stackrel{~}{D}`$ have the following properties
1. $`i`$ embeds $`M`$ as a ‘product type’ submanifold in the sense that it is an embedding and
(9.4)
$$i(M)=\stackrel{~}{M}i(M)$$
and the pull-back under $`i`$ of a defining function for $`\stackrel{~}{M}`$ is a defining function for $`M.`$
2. $`\varphi =\pi i.`$
3. $`j\mathrm{\Phi }=\pi ^{}i.`$
Notice that the inclusion $`j`$ is completly specified by the inclusion $`i`$.
###### Proposition 9.1.
For any family of fibred cusp structures as in (21) there is an extension
(9.5)
$$i:M\stackrel{~}{M}=B\times 𝔹^{p+1}\times 𝕊^q$$
provided that $`p`$ and $`q`$ are large enough. In the case of cusp structures there is an extension $`i:MB\times 𝔹^{p+1},`$ for $`p`$ sufficiently large.
###### Proof.
First consider the cusp case. By Whitney’s Embedding Theorem, we may embed $`M`$ in Euclidean space of sufficiently large dimension
(9.6)
$$i^{}:M^k.$$
To replace this by an embedding of the desired product type into the interior of a closed ball, consider the embedding
(9.7)
$$(x,i^{}):M[0,1)\times ^k,$$
where $`x𝒞^{\mathrm{}}M)`$ is the boundary defining function (it may be assumed without loss of generality that $`0x1.)`$ But $`[0,1)\times ^k`$ can be identified, using a smooth map $`e,`$ with an open neighbourhood of a piece of the boundary of $`𝔹^{p+1},`$ $`p=k.`$ It follows directly that this is an embedding of product type in the sense of (1). Then let
(9.8)
$$i=(\varphi ,e(x,i^{})):MB\times 𝔹^{p+1}$$
be the product of this map with $`M`$ as a map into $`B\times 𝔹^{p+1}.`$ This is still an embedding with property (1) and has the property (2) (and so (3) as well).
In the case of a fibred cusp structure, we can start with the constructions above for the embedding of the underlying cusp structure. Now, take an additional embedding of $`D`$ into a Euclidean space
$$d:D^k$$
for convenience again $`^k`$ by increasing $`k`$ if necessary. Using a product structure near the boundary this fibration can be extended inwards near the boundary to a fibration over $`[0,ϵ)_x\times D`$ and we can consider the smooth map
(9.9)
$$(\mathrm{Id},d\mathrm{\Phi }):[0,ϵ)_x\times M[0,ϵ)\times ^k.$$
Taking $`0<ϵ<1`$ small enough and using radial retraction on $`^k,`$ this can be extended to a smooth map
$$(x,d^{}):M[0,1)\times ^k.$$
For the overall embedding we can then take
(9.10)
$$i=(\varphi ,e^{}(x,d^{}),e^{\prime \prime }i^{}):MB\times 𝔹^{p+1}\times 𝕊^q,p=q=k,$$
with
$$\begin{array}{c}e^{}:[0,1)\times ^k\times ^k𝔹^{p+1},p=k,q=k\\ e\mathrm{"}:^k𝕊^q,\end{array}$$
being identifications of open sets. We can also take $`j:DB\times 𝕊^p`$ to be
$$j=(\varphi ,e^{\prime \prime }(0,d^{})).$$
Since $`i^{}`$ is an embedding, this is also and satisfies (1) – only the boundary is mapped into the boundary, since $`x`$ is a boundary defining function of $`M.`$ Now, (2) holds for the same reason as before and (3) follows from the definition of $`j.`$
## 10. Multiplicativity
Recall one form of the ‘multiplicativity property’ for the index of pseudodifferential operators as introduced by Atiyah and Singer. Consider an iterated fibration, for the moment of compact manifolds without boundary but later allowing $`Z`$ to have a boundary
(10.1)
Taking a connection on $`\varphi ^{}`$ allows the fibre cotangent bundle $`T^{}(M^{}/M)`$ to be identified (naturally up to homotopy) as a subbundle of $`T^{}(M^{}/B)`$ which then splits
(10.2)
$$T^{}(M^{}/B)=T^{}(M^{}/M)(\varphi ^{})^{}T^{}(M/B).$$
Choose cutoff functions $`\chi _1`$ and $`\chi _2`$ which are homogeneous of degree $`0,`$ smooth outside the zero section and are respectively supported outside the two summands and equal to one in a conic neighbourhood of the other with $`\chi _1+\chi _2=1.`$
Let $`A\mathrm{\Psi }^0(M/B;𝔼)`$ and $`B\mathrm{\Psi }^0(M^{}/M;𝔾)`$ be families of elliptic pseudodifferential, then the matrix
(10.3)
$$\left(\begin{array}{cc}\chi _1\sigma (B)& \chi _2\sigma (A)^{}\\ \chi _2\sigma (A)& \chi _1\sigma (B)\end{array}\right)$$
acting from sections of $`(𝔼𝔾)_+=E_+G_+E_{}G_{}`$ to $`(𝔼𝔾)_{}=E_{}G_+E_+G_{}`$ is a family of elliptic symbols for the overall fibration $`M^{}B.`$
###### Proposition 10.1.
(Atiyah and Singer) If the family $`B`$ has trivial one-dimensional index in $`\mathrm{K}(M)`$ then any operator in $`\mathrm{\Psi }^0(M^{}/B;𝔼𝔾)`$ with symbol (10.3) has index in $`\mathrm{K}(B)`$ equal to that of $`A.`$
###### Proof.
To prove this we construct a natural ‘product type’ calculus for the top fibration which includes the fibrewise pseudodifferential operators and an appropriate class of lifts of operators on the fibres of the lower fibration and then do a deformation to the ‘true’ pseudodifferential calculus.
To ease the notational burden, at least initially, we consider the case of the numerical index and suppose that the second fibration, $`\varphi `$ has just one fibre. The general case will be proved as part of the extension below to the cusp calculus. The product-type algebra is discussed in Appendix A. It contains the algebra of pseudodifferential operators on the total space, $`M^{}`$ but has a double order filtration and we denote the filtered spaces $`\mathrm{\Psi }_{\varphi ^{}\mathrm{p}}^{m,m^{}}(M^{};)`$ for a $`_2`$-graded vector bundle $`=(H_+,H_{})`$ over $`M^{};`$
(10.4)
$$\mathrm{\Psi }^m(M^{};)\mathrm{\Psi }_{\varphi ^{}\mathrm{p}}^{m,m}(M^{},)m$$
(there is no particular problem with real order, but since we do not need them, we shall not bother with them here.)
The basic properties of these operator are similar to those of regular pseudodifferential operators
(10.5)
$$\begin{array}{c}\mathrm{\Psi }_{\varphi ^{}\mathrm{p}}^{m_2,m_2^{}}(M^{},_1)\mathrm{\Psi }_{\varphi ^{}\mathrm{p}}^{m_1,m_1^{}}(M^{},_2)\mathrm{\Psi }_{\varphi ^{}\mathrm{p}}^{m_1+m_2,m_1^{}+m_2^{}}(M^{},),\\ \text{provided }(_1)_+=(_2)_{},_+=(_2)_+,_{}=(_1)_{}.\end{array}$$
The main difference between this product-type algebra and the usual one is that there are two related symbol maps. The ‘usual’ principal symbol map is modified to a short exact sequence
(10.6)
$$\begin{array}{c}\text{},\\ 𝒮_{\varphi ^{}\mathrm{p}}^{m,m^{}}(M^{},)=𝒞^{\mathrm{}}([S^{}M^{},\varphi ^{}S^{}(M)];\mathrm{hom}()_+N^m_+N_{\mathrm{ff}}^m^{}).\end{array}$$
Here, $`S^{}M^{}`$ is the sphere bundle of $`T^{}M^{}`$ (best thought of as the boundary of the radial compactification of this bundle) which is then blown up at the submanifold given by the corresponding sphere bundle at infinity of the cotangent bundle of the base. Thus $`[S^{}M^{},\varphi ^{}S^{}(M)]`$ is the ‘old’ boundary hypersurface of the manifold with corners $`[\overline{T^{}M^{}},\varphi ^{}S^{}(M)].`$ The bundle $`N`$ is the normal bundle in this sense and $`N_{\mathrm{ff}}`$ is the normal bundle to the front face, thought of as a trivial bundle over $`[S^{}M^{},\varphi ^{}S^{}M].`$ Thus these factors just represent the growth order of symbols at the boundaries. This ‘standard’ symbol is multiplicative in the obvious sense that
(10.7)
$$\sigma _{m_1+m_2,m_1^{}+m_2^{}}(BA)=\sigma _{m_2,m_2^{}}(B)\sigma _{m_1,m_1^{}}(A)$$
More interestingly, there is a non-commutative symbol, which has much in common with the indicial family for the fibred cusp calculus; it is called here the *base family.* Since $`S^{}MM`$ is a fibration we can lift the fibres of $`\varphi ^{}:M^{}M`$ to give a fibration $`S_\varphi ^{}^{}MS^{}M`$ which has the same typical fibre as $`\varphi ^{}.`$ This second symbol gives a short exact sequence
(10.8)
$$\mathrm{\Psi }_{\varphi ^{}\mathrm{p}}^{m,m^{}1}(M^{},)\mathrm{\Psi }_{\varphi ^{}\mathrm{p}}^{m,m^{}}(M^{},)\stackrel{L_{m,m^{}}}{}\mathrm{\Psi }^m^{}(S_\varphi ^{}^{}M/S^{}M;_+N^m).$$
Here $`N`$ is the inverse homogeneity bundle on $`T^{}M,`$ which is to say the normal bundle to the boundary of the radial compactification $`\overline{T^{}M}.`$ Again this sequence is multiplicative in the sense that
(10.9)
$$L_{m_1+m_2,m_1^{}+m_2^{}}(BA)=L_{m_2,m_2^{}}(B)L_{m_1,m_1^{}}(A)$$
under (10.5).
Although these two maps are separately surjective they are related through the symbol map on the image of (10.8). Namely the combination of the two maps gives a short exact sequence
(10.10)
$$\begin{array}{c}\text{}\hfill \\ \hfill A_{\varphi ^{}\mathrm{p}}^{m,m^{}}(M^{},)=\{(a,A)\mathrm{\Psi }^m^{}(S_\psi ^{}X/S^{}X;_+N^m)𝒮_{\varphi ^{}\mathrm{p}}^{m,m^{}}(M^{},);\\ \hfill \sigma _m^{}(A)=a|_{[S^{}M^{},\psi ^{}S^{}(M)]}\}.\end{array}$$
There are appropriate Sobolev spaces on which these operators are bounded. Since we are interested principally in the case of operators of order $`0`$ it suffices to note that
(10.11)
$$\begin{array}{c}\mathrm{\Psi }_{\varphi ^{}\mathrm{p}}^{0,0}(M^{};)A:L^2(M^{};H_+)L^2(M^{};H_{})\hfill \\ \hfill \text{ is bounded and is compact iff }\\ \hfill \sigma _{0,0}(A)=0,N_{0,0}(A)=0.\end{array}$$
From this and standard constructions it follows that $`A`$ in (10.11) is Fredholm iff
(10.12)
$$\sigma _{0,0}(A)^1𝒮_{\varphi ^{}\mathrm{p}}^{0,0}(M^{},^{})\text{ and }N_{0,0}(A)^1\mathrm{\Psi }^m^{}(S_\varphi ^{}^{}M^{}/S^{}M;^{}).$$
As well as (10.4) the fibrewise pseudodifferential operators may also be considered as product-type operators
(10.13)
$$\begin{array}{c}\mathrm{\Psi }^m(M^{}/M;)\mathrm{\Psi }_{\varphi ^{}\mathrm{p}}^{0,m}(M^{};),\\ \sigma _{0,m}(A)=\sigma _m(A),N_0(A)=AA\mathrm{\Psi }^m(M^{}/B;).\end{array}$$
In particular such an operator of order $`0`$ is Fredholm on $`L^2`$ if and only it is invertible, at least if the fibration has non-trivial base.
It is also important to see that pseudodifferential operators on the base are in a sense included in the product type pseudodifferential operators on the total space. Suppose that $`E`$ is a $`_2`$-graded vector bundle over $`M`$ which is embedded in $`𝒞^{\mathrm{}}(M^{};𝔽)`$ for some $`_2`$-bundle $`𝔽.`$ Thus there is a pair of families of finite rank, smoothing, projections $`\pi _\pm \mathrm{\Psi }^{\mathrm{}}(M^{}/M;F_\pm )`$ projecting onto the fibre of $`E_\pm `$ over each point of $`M.`$ Then suppose that $`A\mathrm{\Psi }^m(M;𝔼)`$ is some pseudodifferential operator. With the identification of bundles, we may identify $`A`$ as an operator from $`𝒞^{\mathrm{}}(M^{};F_+)`$ to $`𝒞^{\mathrm{}}(M^{};F_{})`$ and then
(10.14)
$$A=\pi _{}A\pi _+\mathrm{\Psi }_{\varphi ^{}\mathrm{p}}^{m,\mathrm{}}(M^{};𝔽).$$
In this case
(10.15)
$$N_m(A)=\pi _{}\sigma _m(A)\pi _+\mathrm{\Psi }^{\mathrm{}}(S_\varphi ^{}^{}M^{}/S^{}M;𝔽_+N_m).$$
Under the hypothesis of the Proposition that we are trying to prove – for the moment only in the case of a single operator – we have a fibre family of trivial index of rank one. So, by smoothing perturbation we may assume that this $`B\mathrm{\Psi }^0(M^{}/M;𝔾)`$ is surjective and has a null bundle which is a trivial line bundle. Taking the $`_2`$-bundle from the base we may form the extended operator which we denote
(10.16)
$$\begin{array}{c}\left(\begin{array}{cc}B& 0\\ 0& B^{}\end{array}\right)\mathrm{\Psi }^0(M^{}/M;𝔼𝔾)\mathrm{\Psi }_{\varphi ^{}\mathrm{p}}^{0,0}(M^{};𝔼𝔾),\hfill \\ \hfill 𝔼=(E_+G_+E_{}G_{},E_+G_{}E_{}G_+).\end{array}$$
Since the lifted bundles $`E_\pm `$ are trivial on each fibre, this has null bundle precisely $`E_+`$ as a subbundle of $`𝒞^{\mathrm{}}M;E_+H_+)`$ and cokernel bundle given by $`E_{}`$ as a subbundle of $`𝒞^{\mathrm{}}(E_+H_{}).`$ Thus we can interpret $`A`$ as mapping this null bundle into $`E_{}G_+`$ and so form the operator
(10.17)
$$\left(\begin{array}{cc}B& 0\\ A& B^{}\end{array}\right)\mathrm{\Psi }_{\varphi ^{}\mathrm{p}}^{0,0}(M^{};𝔼𝔾).$$
Directly from the definition, this is a Fredholm operator on $`L^2.`$ Since it has null space just the null space of $`A`$ as a subspace of the null space of $`B`$ and cokernel the complement of the range of $`A`$ as a subspace of the complement of the range of $`B^{}`$ we conclude directly that
(10.18)
$$\mathrm{ind}_\text{a}\left(\begin{array}{cc}B& 0\\ A& B^{}\end{array}\right)=\mathrm{ind}_\text{a}(A).$$
Now we proceed to deform this operator as an elliptic family in $`\mathrm{\Psi }_{\varphi ^{}\mathrm{p}}^{0,0}(M^{};𝔼𝔾).`$ First choose an element $`\stackrel{~}{A}\mathrm{\Psi }^0(M^{};𝔼G_+)`$ with symbol $`\chi _2\sigma (A)\mathrm{Id}_{G_+}.`$ As an element of $`\mathrm{\Psi }_{\varphi ^{}\mathrm{p}}^{0,0}(M^{};𝔼G_+)`$ its symbol is the lift of $`\chi _2\sigma (A)`$ and its base family is the lift of this symbol to the front face, interpreted as a bundle map. Consider the 1-parameter family
(10.19)
$$\left(\begin{array}{cc}B& s\stackrel{~}{A}^{}\\ s\stackrel{~}{A}+(1s)A& B^{}\end{array}\right)\mathrm{\Psi }_{\varphi ^{}\mathrm{p}}^{0,0}(M^{};𝔼𝔾),s[0,1].$$
This is fully elliptic, i.e. both symbol and base family are invertible for all $`s[0,1].`$ For the symbol itself this is rather clear since it is the family
(10.20)
$$\left(\begin{array}{cc}\sigma (b)& s\chi _2\sigma (a)^{}\\ s\chi _2\sigma (a)& \sigma (b)^{}\end{array}\right).$$
On the other hand the base family is
(10.21)
$$\left(\begin{array}{cc}B& s\sigma (A)^{}\\ s\sigma (A)+(1s)\pi _{}\sigma (A)\pi _+& B^{}\end{array}\right)$$
as an operator on each fibre of $`\varphi ^{},`$ lifted to $`S^{}M.`$ Thus, $`\sigma (A)`$ is simply being extended from the null space of $`B`$ to the whole bundle and this family is invertible for all $`s[0,1].`$
Then we may choose $`\stackrel{~}{B}\mathrm{\Psi }^0(M^{};E_+𝔾)`$ with symbol $`\chi _1\sigma (B)\mathrm{Id}_{E_+}`$ and similarly extend the operator(10.19) at $`s=1`$ to a 1-parameter elliptic family
(10.22)
$$\left(\begin{array}{cc}(1r)B+r\stackrel{~}{B}& \stackrel{~}{A}^{}\\ \stackrel{~}{A}& (1r)B^{}+r\stackrel{~}{B}^{}\end{array}\right)\mathrm{\Psi }_{\varphi ^{}\mathrm{p}}^{0,0}(M^{};𝔼𝔾),r[0,1].$$
Again this is a fully elliptic family and at $`r=1`$ reduces to an element of $`\mathrm{\Psi }^0(M^{};𝔼𝔾)`$ which has index equal to that of $`A`$ and symbol given by (10.3). ∎
###### Remark 10.1.
In this construction it is already clear that the K-class of the symbol of the resulting operator only depends on the K-classes of the symbols of $`A`$ and $`B`$ and so this defines a map
(10.23)
$$M_b:\mathrm{K}_\text{c}(T^{}(M/B))\mathrm{K}_\text{c}(T^{}(M^{}/B)).$$
It should also be noted that if the operator (or family of operators) $`A`$ is actually invertible then the lifted family is invertible and this invertibility can be maintained through the homotopy back to the standard algebra. This is used below for full ellipticity in the cusp setting.
Despite this, in general a fully elliptic family of product-type cannot be deformed through elliptics into the ‘standard’ subspace since this involves deforming the base family, through invertibles, to bundle maps and there may be an obstruction to this in K-theory.
Next we consider the corresponding construction in the cusp case; the main obstacle to this extension is notational! Thus we have an iterated fibration. Here $`\varphi :MB`$ is our usual fibration of compact manifolds with fibre a compact manifold with boundary. The ‘top’ fibration $`\varphi ^{}:M^{}M`$ is assumed to have compact fibre a manifold without boundary (in the application to lifting it is a sphere.) Again we are given an elliptic family $`B\mathrm{\Psi }^0(M^{}/M;𝔾)`$ which is surjective and has a trivial 1-dimensional null bundle where $`𝔾`$ is some $`_2`$-graded bundle over $`M^{}.`$
We refer to Appendix B for a discussion of product-type cusp operators in this setting. These are quite analogous to the product-type pseudodifferential operators discussed above. Such an operator $`P\mathrm{\Psi }_{\varphi ^{}\mathrm{p},\mathrm{cu}}^{m,m^{}}(M^{}/B;𝔽)`$ for any $`_2`$-graded bundle $`𝔽`$ over $`M^{}`$ acts on (weighted) smooth sections as in (1.3) but there are now *three* ‘symbol’ maps. The ‘commutative’ symbol corresponds to the usual cusp symbol modified in a way analogous to that of the symbol in the product-type pseudodifferential calculus in (10.6)
(10.24)
$$\begin{array}{c}\mathrm{\Psi }_{\varphi ^{}\mathrm{p},\mathrm{cu}}^{m1,m^{}}(M^{}/B;𝔽)\mathrm{\Psi }_{\varphi ^{}\mathrm{p},\mathrm{cu}}^{m,m^{}}(M^{}/B;𝔽)\stackrel{\sigma _{m,m^{}}}{}𝒮_{\varphi ^{}\mathrm{p},\mathrm{cu}}^{m,m^{}}(M^{}/B;𝔽),\hfill \\ \hfill 𝒮_{\varphi ^{}\mathrm{p},\mathrm{cu}}^{m,m^{}}(M^{}/B;𝔽)=𝒞^{\mathrm{}}([{}_{}{}^{\mathrm{cu}}S_{}^{}M^{},(\varphi ^{})^{}{}_{}{}^{\mathrm{cu}}S_{}^{}(M/B)];N^mN_{\mathrm{ff}}^m^{}\mathrm{hom}(𝔽)).\end{array}$$
Secondly is the indicial family, corresponding to the map (1.6) for the usual cusp calculus, but now taking values in the suspended version of the product-type calculus for the restriction of the fibration to the preimage of the boundary of $`M`$
(10.25)
$$\text{}.$$
Finally there is an analogue of the non-commutative symbol in (10.8), namely a short exact sequence
(10.26)
taking values in the pseudodifferential operators on the fibres of $`\varphi ^{}`$ but lifted to the fibrewise cusp cosphere bundle $`{}_{}{}^{\mathrm{cu}}S_{}^{}(M/B).`$ Whilst separately surjective these three maps are related and combine to give a ‘full symbol algebra’ consisting of the elements with the compatibility conditions
(10.27)
$$\begin{array}{c}A_{\varphi ^{}\mathrm{p},\mathrm{cu}}^{m,m^{}}(M^{}/B)=\{(a,I,\beta )\hfill \\ \hfill 𝒮_{\varphi ^{}\mathrm{p},\mathrm{cu}}^{m,m^{}}(M^{}/B;𝔽)\times \mathrm{\Psi }_{\mathrm{sus},\varphi ^{}\mathrm{p}}^{m,m^{}}(M^{}/B;𝔽)\times \mathrm{\Psi }^m^{}({}_{}{}^{\mathrm{cu}}S_{\varphi ^{}}^{}M/{}_{}{}^{\mathrm{cu}}S_{}^{}M;𝔽_+N^m);\\ \hfill a|_{}=\sigma (I),a|_{\mathrm{ff}}=\sigma (\beta ),N(\beta )=L(I)\}.\end{array}$$
This space itself corresponds to the short exact sequence
(10.28)
which captures compactness and Fredholm properties on the appropriate Sobolev spaces. For us it suffices to note that
(10.29)
$$\begin{array}{c}A\mathrm{\Psi }_{\varphi ^{}\mathrm{p},\mathrm{cu}}^{0,0}(M^{}/B;𝔽)A:L^2(M^{}/B;F_+)L^2(M^{}/B;F_{})\text{ is bounded and}\hfill \\ \hfill A\text{ is compact}\sigma (A)=0,N(A)=0,L(A)=0\\ \hfill A\text{ is Fredholm}(\sigma (A),N(A),L(A))^1A_{\varphi ^{}\mathrm{p},\mathrm{cu}}^{m,m^{}}(M^{}/B,𝔽^{}).\end{array}$$
In the latter case, we say $`A`$ is *fully elliptic*. As with the product-type pseudodifferential operators in the boundaryless case, it is important that the three different spaces of pseudodifferential operators, namely the ordinary pseudodifferential operators on the fibres of the smaller fibration, the ordinary cusp pseudodifferential operators on the fibres of the larger fibration and the cusp pseudodifferential operators on the base fibration all lift into this larger space:
(10.30)
$$\begin{array}{c}\mathrm{\Psi }^m(M^{}/M;𝔽)\mathrm{\Psi }_{\varphi ^{}\mathrm{p},\mathrm{cu}}^{0,m}(M^{}/B;𝔽)\\ \mathrm{\Psi }_{\mathrm{cu}}^m(M^{}/B;𝔽)\mathrm{\Psi }_{\varphi ^{}\mathrm{p},\mathrm{cu}}^{m,m}(M^{}/B;𝔽)\\ \mathrm{\Psi }_{\mathrm{cu}}^m(M/B;𝔼)\mathrm{\Psi }_{\varphi ^{}\mathrm{p},\mathrm{cu}}^{m,\mathrm{}}(M^{}/B;𝔽)\end{array}$$
where $`𝔼`$ is a $`_2`$ graded vector bundle embedded as a subbundle of $`𝒞^{\mathrm{}}(M^{}/M;𝔽)`$ as a bundle over $`M.`$
###### Proposition 10.2.
For an iterated fibration with cusp structure as in (10.1), if an elliptic family $`B\mathrm{\Psi }^0(M^{}/M;𝔽)`$ is surjective and has trivial one-dimensional null bundle $`H`$ then for any fully elliptic element $`A\mathrm{\Psi }_{\mathrm{cu}}^0(M/B;𝔼)`$ the operator
(10.31)
$$\left(\begin{array}{cc}B& 0\\ A& B^{}\end{array}\right)\mathrm{\Psi }_{\varphi ^{}\mathrm{p},\mathrm{cu}}^{0,0}(M^{}/B;𝔽𝔼),$$
where the inclusions are through (10.30), is a fully elliptic element with the same index as $`A`$ in $`\mathrm{K}(B)`$ and furthermore this element is deformable, through fully elliptic elements, into $`\mathrm{\Psi }_{\mathrm{cu}}^0(M^{}/B;𝔽𝔼).`$ This results again in a map
(10.32)
$$M_b:\mathrm{K}_{\mathrm{cu}}(\varphi )\mathrm{K}_{\mathrm{cu}}(\varphi \varphi ^{})$$
which only depends on $`b=[\sigma (B)]\mathrm{K}_\text{c}(T^{}(M^{}/M)).`$
###### Proof.
As noted above, the first part of the result, that (10.31) gives a fully elliptic element of the product-type calculus on the total fibration, and that (10.18) again holds, follows as in the proof of Proposition 10.1 with only minor notational changes.
For the second, deformation, part of the statement we need to proceed with more care. First, the arguments of Proposition 10.1 apply unchanged in the sense that (10.19) and (10.22) together give a symbolically elliptic deformation, indeed the computation of the symbol (although now for a family in the base) is the same as there, as is the computation of the base family – the latter is still given by (10.21) provided the symbol of $`A`$ is interpreted as the cusp symbol. Thus we have constructed a family $`P_t,`$ in the product-type cusp calculus, where we can relabel and change parameterization to make the family smooth in $`t[0,1].`$ This family is elliptic for all $`t,`$ has invertible base family for all $`t,`$ is Fredholm at $`t=0`$ and is in the ordinary cusp calculus at $`t=1.`$ So, consider the indicial family of $`P_t.`$ This is a suspended family of product-type pseudodifferential operators (acting on the fibres of a fibration) which is fully elliptic, i.e. is elliptic and has invertible base family, and which is invertible at $`t=0.`$ Since it is also invertible for large values of the suspending variable, it follow by standard arguments that it can be perturbed by a family of smoothing operators (see Remark 10.1) which vanishes near infinity in the suspension variable and vanishes near $`t=0`$ to be invertible for all values of $`t.`$ Modifying the family in this way results in a deformation as desired.
That this construction is symbolic as far as $`B`$ is concerned and ‘fully symbolic’ as far as $`A`$ is concerned, in the sense of (10.32), follows readily from the construction. Namely it certainly behaves well under stabilization and homotopy, with the homotopy parameter simply being added to the base variables. So it remains to show that it is stable under different regularizations of the null bundle of $`B;`$ in fact any two such stabilizations are homotopic. ∎
Finally we comment on the generalization of this result to the fibred cusp case although this is not used below. Thus we again consider an iterated fibration (10.1) where now the second fibration has a fibred cusp structure, $`\mathrm{\Phi }:MD.`$ There are two extreme possibilities for a fibred cusp structure for the overall fibration. First, it is always possible to ‘add’ the fibres of $`M^{}M`$ to the fibres of the boundary, i.e. to take the boundary fibration for the overall fibration to be
(10.33)
$$\mathrm{\Phi }^{}=\mathrm{\Phi }\varphi ^{}.$$
###### Proposition 10.3.
For an iterated fibration as in (10.1) where the second fibration has a fibred cusp structure and the top fibration has the fibred cusp structure (10.33), if an elliptic family $`B\mathrm{\Psi }^0(M^{}/M;𝔽)`$ is surjective and has trivial one-dimensional null bundle $`H`$ then for any fully elliptic element $`A\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^0(M/B;𝔼)`$ the operator
(10.34)
$$\left(\begin{array}{cc}B& 0\\ A& B^{}\end{array}\right)\mathrm{\Psi }_{\varphi ^{}\mathrm{p},\mathrm{\Phi }^{}\mathrm{cu}}^{0,0}(M^{}/B;𝔽𝔼),$$
where the inclusions are through (10.30), is a fully elliptic element with the same index as $`A`$ in $`\mathrm{K}(B)`$ and furthermore this element is deformable, through fully elliptic elements, into $`A_B\mathrm{\Psi }_{\mathrm{\Phi }^{}\mathrm{cu}}^0(M^{}/B;𝔽𝔼).`$ This results again in a map
(10.35)
$$M_b:\mathrm{K}_{\mathrm{\Phi }\mathrm{cu}}(\varphi )\mathrm{K}_{\mathrm{\Phi }\varphi ^{}\mathrm{cu}}(\varphi \varphi ^{})$$
which only depends on $`b=[\sigma (B)]\mathrm{K}_\text{c}(T^{}(M^{}/M)).`$
###### Proof.
The arguments in the proof of Proposition 10.2 go through essentially without change, the only difference being the appearance of more suspension parameters in the normal operator. ∎
There is a second extreme possibility, other than (10.33), for the fibred cusp structure on the top fibration in (10.1), corresponding to the fibres for $`\mathrm{\Phi }^{}:M^{}D^{}`$ being of the same dimension as for $`\mathrm{\Phi }`$ in which case all the ‘new’ boundary variables are in the base of the fibration. Then there is a commutative diagram of fibrations
(10.36)
In this case there is a difficulty in extending the construction above, in that it is not quite clear what the family $`B`$ should be. Approached directly $`B`$ would need to be a family which is adiabatic in the boundary variable. Since the index theory for such families has not been properly developed we have chosen to proceed more indirectly by using Proposition 8.1. However, it is possible instead to use Proposition 10.3 above and then make an adiabatic limit back to the case where the extra variables are in the base of the fibration. Since this involves a rather delicate homotopy, and is not used below, the details are omitted.
## 11. Lifting and excision
Next we consider the lifting construction for cusp K-theory with respect to an extension of a given fibration, see Definition 9.1.
###### Proposition 11.1.
If $`\stackrel{~}{\varphi }:\stackrel{~}{M}B`$ is an extension of a fibration $`\varphi :MB`$ with cusp structure in the sense of Definition 9.1 then there is a well-defined lifting homomorphism
(11.1)
$$(\stackrel{~}{\varphi }/\varphi )^!:\mathrm{K}_{\mathrm{cu}}(\varphi )\mathrm{K}_{\mathrm{cu}}(\stackrel{~}{\varphi })$$
which induces a commutative diagram (17) where the map on the right is an absolute form of the lifting map of Atiyah and Singer. When $`\stackrel{~}{M}=𝔹^{2N+1}\times B`$, this reduces to pull-back under Poincaré duality
(11.2)
where $`R:𝒞_{\mathrm{cu}}(\stackrel{~}{M})𝒞_{\mathrm{cu}}(M)`$ is the restriction map.
###### Proof.
The normal bundle to $`M`$ in $`\stackrel{~}{M}`$ is a real vector bundle which models the extended fibration near $`M.`$ Thus, if we take the one-point compactification of the fibres we obtain a bundle of spheres $`\pi :S_VM`$ which gives an extension of the original fibration but is also an iterated fibration in the sense of Proposition 10.2. Moreover, following the approach of Atiyah and Singer, there is a ‘Bott element’, which we can realize as a family
(11.3)
$$B\mathrm{\Psi }^0(S_V/M;𝕃)$$
for a $`_2`$-graded bundle $`𝕃`$ over $`S_V`$ with $`L_+`$ and $`L_{}`$ identified near the section at infinity and such that $`B\mathrm{Id}`$ has kernel with support disjoint from the section at infinity (in either factor). Now, we can apply Proposition 10.2 to ‘lift’ each fully elliptic element of $`\mathrm{\Psi }_{\mathrm{cu}}^0(M/B;𝔼)`$ to a fully elliptic element of $`\mathrm{\Psi }_{\mathrm{cu}}^0(S_V/B;𝔽).`$ A brief review of the construction shows that the condition that all operators be equal to the identity near the section at infinity can be maintained throughout. In view of this property of the kernel, including an identification of $`F_+`$ and $`F_{}`$ near the section at infinity, these operators may be trivially extended into $`\mathrm{\Psi }_{\mathrm{cu}}^0(\stackrel{~}{M}/B;𝔽)`$ to be fully elliptic. The nature of this construction shows that this does indeed construct a map (11.1) and that it leads to a commutative diagram (17). When $`\stackrel{~}{M}=𝔹^{2N+1}\times B`$, the commutativity of diagram (11.2) follows from proposition 12.1 and the fact that $`\stackrel{~}{\varphi }=\varphi R`$. ∎
Note that the Poincaré duality isomorphism and (11.2) show that the lifting map is independent of choices and is well-behaved under composition. This is also relatively easy to see directly.
## 12. Product with a ball and a sphere
We have shown in Proposition 9.1 that a fibred cusp structure over $`B`$ can always be ‘trivialized’ by embedding it in one of the product cases with both fibrations being projections, in which case the diagram (21) becomes
(12.1)
with $`\pi `$ being projection onto the leftmost factor and $`\pi ^{}`$ off the rightmost.
###### Proposition 12.1.
If $`M>0`$ and $`N0`$ in the product spaces in (12.1), then
(12.2)
$$\mathrm{K}_{\pi ^{}\mathrm{cu}}(\pi )\{\begin{array}{cc}\mathrm{K}(B)\hfill & M+N\text{ even}\hfill \\ \mathrm{K}(B)\mathrm{K}(B)\hfill & M+N\text{ odd}\hfill \end{array}.$$
In the cusp case with $`\pi :B\times 𝔹^{M+1}B`$, we have
$$\mathrm{K}_{\mathrm{cu}}(\pi )\{\begin{array}{cc}\mathrm{K}(B)\hfill & M\text{ even}\hfill \\ \mathrm{K}(B)\mathrm{K}(B)\hfill & M\text{ odd}\hfill \end{array}.$$
###### Proof.
From the Poincaré duality isomorphism, Theorem 2, we can use KK-theory to perform the computation. Now, for any product $`M=B\times Z`$ in which the overall fibration is the projection onto the left factor and the boundary fibration is the product with some fibration of the boundary of $`\psi :ZD,`$
(12.3)
$$𝒞_{\mathrm{Id}\times \psi }(B\times Z)=𝒞(B)\widehat{}𝒞_\psi (Z)$$
is the completed tensor product. From the general properties of KK-theory it follows that
(12.4) $`\mathrm{KK}_B(𝒞_{\mathrm{Id}\times \psi }(B\times Z),𝒞(B))`$ $`=\mathrm{KK}_B(𝒞(B)\widehat{}𝒞_\psi (Z),𝒞(B))`$
$`\mathrm{KK}(𝒞_\psi (Z),𝒞(B))`$
$`\mathrm{K}(B)\mathrm{KK}(𝒞_\psi (Z),),`$
where in the last step we used the Künneth formula for KK-theory (see for instance Theorem 23.1.3 in ) and we used the fact (proved below) that $`\mathrm{KK}(𝒞_\psi (Z),)`$ is a free $``$-module. Thus it suffices to consider the case in which the base in (12.1) is reduced to a point and to show then that $`\mathrm{K}_{\pi ^{}\mathrm{cu}}(\pi )`$ reduces to one or two copies of $``$ according to parity. In fact the general argument, with the base factor retained, is not much more complicated.
Thus we need only to consider
(12.5)
and compute the fibred-cusp K-theory in this case.
First consider the cusp case, where there is no factor of $`𝕊^N.`$ The same argument applies to reduce the computation to the case of $`𝔹^{M+1}.`$ So consider the (split) short exact sequence (10), proved in Section 5. Since $`K(\{\text{pt}\})=,`$ essentially by definition, it suffices to check that
(12.6)
$$\mathrm{K}_\text{c}(𝔹^{M+1}\times ^{M+1})=\{\begin{array}{cc}\{0\}\hfill & M\text{ even}\hfill \\ \hfill & M\text{ odd.}\hfill \end{array}$$
Here, recall that the ball is taken to be closed, so the K-theory is absolute in that factor and compactly supported in the Euclidean factor. Since the ball is contractible, the only source for K-classes is
$$\mathrm{K}_\text{c}(^{M+1})\mathrm{K}_\text{c}(𝔹^{M+1}\times ^{M+1})$$
so (12.6) follows.
In the fibred cusp case, set $`Z_{M,N}=𝔹^{M+1}\times 𝕊^N.`$ The obstruction to perturbing an elliptic operator so that it becomes fully elliptic lies in
$$\mathrm{K}(T^{}𝕊^M\times )\mathrm{K}(𝕊^M\times ^{M+1}).$$
Consider the 6-term exact sequence coming from the inclusion
$$r_{𝔹^{M+1}}:T^{}𝕊^M\times T^{}(𝔹^{M+1})|_{𝔹^{M+1}}T^{}(𝔹^{M+1}),$$
namely
(12.7)
using the identification
(12.8)
$$\mathrm{K}_\text{c}^\mathrm{k}(T^{}(𝔹^{M+1}),T^{}(𝕊^M)\times )\mathrm{K}_\text{c}^\mathrm{k}(^{2M+2})\{\begin{array}{cc}\hfill & k=0,\hfill \\ \{0\}\hfill & k=1.\hfill \end{array}$$
From (12.6) and (12.8) and using the fact that
(12.9)
$$\mathrm{K}_\text{c}^\mathrm{k}(^{2M+2})\mathrm{K}_\text{c}^\mathrm{k}(T^{}(𝔹^{M+1}))$$
maps to zero, we get that
(12.10)
$$K_c^0(T^{}𝕊^M)\{\begin{array}{cc}\hfill & M\text{ even}\hfill \\ \hfill & M\text{ odd}\hfill \end{array},\mathrm{K}_c^1(T^{}𝕊^M)\{\begin{array}{cc}\{0\}\hfill & M\text{ even}\hfill \\ \hfill & M\text{ odd.}\hfill \end{array}$$
Thus, in particular, when $`M`$ is even, there is no obstruction to the existence of a smoothing perturbation that makes an elliptic fibred cusp operator fully elliptic. So consider the image of
$$I_1:\mathrm{K}_\text{c}^1(T^{}(𝔹^{M+1}\times 𝕊^N))\mathrm{K}_\text{c}^0(T^{}(𝕊^M)).$$
Since $`\pi ^{}:𝕊^M\times 𝕊^N𝕊^M`$ extends to
$$P^{}:𝔹^{M+1}\times 𝕊^N𝔹^{M+1}$$
by projecting on the right factor
$$I_1=\mathrm{ind}_\pi ^{}r_{(𝔹^{M+1}\times 𝕊^N)}^{}=r_{(𝔹^{M+1})}^{}\mathrm{ind}_P^{}.$$
But $`\mathrm{ind}_P^{}`$ is clearly surjective, so the image of $`I_1`$ is the same as the image of $`r_{(𝔹^{M+1})}^{}.`$ Thus, we conclude from the 6-term exact sequence (12.7) that
$$I_1(\mathrm{K}_\text{c}^1(T^{}(𝔹^{M+1}\times 𝕊^N)))\{\begin{array}{cc}\hfill & M\text{ even}\hfill \\ \{0\}\hfill & M\text{ odd.}\hfill \end{array}$$
Thus, when $`M`$ is even, there is a short exact sequence
Since
$$K_c^0(T^{}(𝔹^{M+1}\times 𝕊^N))\{\begin{array}{cc}\{0\}\hfill & N\text{ even}\hfill \\ \hfill & N\text{ odd}\hfill \end{array}$$
is a free $``$-module, this sequence splits and hence
$$\mathrm{K}_{\pi ^{}\mathrm{cu}}(𝔹^{M+1}\times 𝕊^N)\mathrm{K}(T^{}(𝔹^{M+1}\times 𝕊^N))\{\begin{array}{cc}\hfill & N\text{ even}\hfill \\ \hfill & N\text{ odd.}\hfill \end{array}$$
When $`M`$ is odd, we have instead the short exact sequence
$$K_c^0(T^{}𝕊^M)\mathrm{K}_{\pi ^{}\mathrm{cu}}(𝔹^{M+1}\times 𝕊^N)\mathrm{ker}(I_0)$$
which splits by sending an element of $`\mathrm{ker}(I_0)`$ to a full symbol with null index. Thus we conclude that
$$\mathrm{K}_{\pi ^{}\mathrm{cu}}(𝔹^{M+1}\times 𝕊^N)\mathrm{ker}(I_0)K_c^0(T^{}(𝔹^{M+1}\times 𝕊^N))$$
since the space of obstruction is $`\mathrm{K}_\text{c}^1(T^{}𝕊^M)`$ and $`I_0`$ is surjective in this case, as one can see from the 6-term exact sequence (12.7). Using again (12.10)
(12.11) $`\mathrm{K}_{\pi ^{}\mathrm{cu}}(𝔹^{M+1}\times 𝕊^N)`$ $`\mathrm{K}_c(T^{}(𝔹^{M+1}\times 𝕊^N))`$
$`\mathrm{K}_c(T^{}𝕊^N\times ^{M+1})`$
$`\mathrm{K}_c(T^{}𝕊^N)\{\begin{array}{cc}\hfill & N\text{ even},\hfill \\ \hfill & N\text{ odd.}\hfill \end{array}`$
## 13. The topological index
Consider a fibration as in (21). Let
$$i:MB\times 𝔹^{p+1},$$
with $`p`$ even, be an embedding as in Proposition 9.1, where the fibration structure on $`B\times 𝔹^{p+1}`$ is given by the projection on the right factor
$$\pi :B\times 𝔹^{p+1}B.$$
Proposition 11.1 gives a well-defined lifting homomorphism
(13.1)
$$(\pi /\varphi )^!:\mathrm{K}_{\mathrm{cu}}(\varphi )\mathrm{K}_{\mathrm{cu}}(\pi ).$$
By the proof of Proposition 12.1,
$$\mathrm{K}_{\mathrm{cu}}(\pi )\stackrel{quan}{}\mathrm{KK}_B(𝒞_\pi (B\times 𝔹^{p+1}),𝒞(B))\stackrel{ind}{}\mathrm{K}(B)$$
is an isomorphism.
###### Definition 13.1.
the topological index map $`\mathrm{ind}_\text{t}:\mathrm{K}_{\mathrm{\Phi }\mathrm{cu}}\mathrm{K}(B)`$ is defined by
$$\mathrm{ind}_\text{t}=\mathrm{ind}\mathrm{quan}(\pi /\varphi )^!q_{\mathrm{ad}}.$$
The fact that it does not depend on the choice of the embedding follows from the stability of the lifting map under repeated embedding, but in any case will follow from Theorem 1!
###### Proof of Theorem 1.
The theorem follows from the commutativity of the diagram (17) stated in Proposition 11.1, and the commutativity of diagram (25) stated in Proposition 8.1. ∎
## 14. Families of Atiyah-Patodi-Singer type
The Atiyah-Patodi-Singer index theorem of was originally proved with the idea of obtaining a generalization of Hirzebruch’s signature theorem for the case of manifolds with boundary. Since then, variants of the proof of the theorem were obtained, for instance in and . An extension to the family case was discussed in , and in , . In all these proofs, including the original one, heat kernel techniques for Dirac operators play an important rôle. As opposed to the Atiyah-Singer index theorem, the Atiyah-Patodi-Singer index theorem was originally restricted to Dirac operators, but using trace functional techniques, a pseudodifferential generalization was obtained in in the setting of cusp operators. Such a generalization was also discussed by Piazza in for b-pseudodifferential operators.
In , Dai and Zhang provided an interesting proof of the Atiyah-Patodi-Singer index theorem by embedding the manifold with boundary into a large ball. Relating the Dirac operator of interest with one defined on the large ball via a careful analysis, they were able to take advantage of the simple topology of the ball to get the Atiyah-Patodi-Singer index theorem for the original operator. This is certainly closely related to our constructions above, however, the methods of are essentially analytical and no K-theory is involved.
We proceed to briefly recall the setting of Atiyah-Patodi-Singer boundary problem and its reformulation in terms of cusp pseudodifferential operators, leading to a K-theory index. Let $`Z`$ be an even dimensional Riemannian manifold with nonempty boundary $`Z=X`$ with a Riemannian metric which is of product type near the boundary, so there is a neighborhood $`X\times [0,1)Z`$ of the boundary in which the metric takes the form
(14.1)
$$g=du^2+h_X$$
where $`u[0,1)`$ is the coordinate normal to the boundary and $`h_X`$ is the pull-back of a metric on $`X`$ via the projection $`X\times [0,1)X.`$ Let $`𝔼`$ be a Hermitian vector bundle over $`Z`$ with a Clifford module structure for the metric (14.1) and with a unitary Clifford connection which is constant in the normal direction near the boundary under a product trivialization. This defines a generalized Dirac operator
$$ð:𝒞^{\mathrm{}}(Z,𝔼)𝒞^{\mathrm{}}(Z,𝔼).$$
In the neighborhood $`X\times [0,1)Z`$ of the boundary described above, it takes the form
(14.2)
$$ð^+=\gamma \left(\frac{}{u}+A\right)$$
where $`\gamma =cl(u):𝔼|_X𝔼|_X`$ is given by Clifford multiplication by the normal differential and $`A:𝒞^{\mathrm{}}(X,E_+|_X)𝒞^{\mathrm{}}(X,E_+|_X)`$ is a Dirac operator on $`X`$ such that
(14.3)
$$\gamma ^2=\mathrm{Id},\gamma ^{}=\gamma ,A\gamma =\gamma A,A^{}=A.$$
Consider the spectral boundary condition
(14.4)
$$\phi 𝒞^{\mathrm{}}(Z,𝔼),P(\phi |_X)=0,$$
where $`P`$ is the projection onto the nonnegative spectrum of $`A.`$ Then
(14.5)
$$ð^+:W_P^1\mathrm{L}^2(Z;E_{})$$
is a Fredholm operator, where
$$W_P^1=\left\{f\mathrm{H}^1(X;E_+);P(f|_X)=0\right\}$$
is a subspace of $`\mathrm{H}^1(X;E_+),`$ the Sobolev space of order 1.
Atiyah, Patodi and Singer show in that the index of $`ð^+`$ is given by
$$\mathrm{ind}(ð^+)=_Z\widehat{A}(Z)\mathrm{Ch}^{}(𝔼)\frac{h+\eta }{2}$$
where $`\mathrm{Ch}^{}(𝔼)`$ is the twisting Chern of $`𝔼,`$ $`\widehat{A}`$ is the $`\widehat{A}`$-genus, $`h=dim\mathrm{ker}A`$ and $`\eta `$ is the eta invariant of $`A.`$
As discussed in , one can alternatively describe the index problem by adding a cylindrical end to the manifold with boundary $`Z.`$ More precisely, $`Z`$ may be enlarged by attaching the half-cylinder $`(\mathrm{},0)\times X`$ to the boundary $`X`$ of $`Z.`$ Call the resulting manifold $`Z^{}.`$ The metric, being a product near the boundary, can be naturally extended to this half-cylinder, which makes the resulting manifold a complete Riemannian manifold. The bundle, Clifford structure, connection and hence the Dirac operator also have natural translation-invariant extensions to $`Z^{}`$ using the product structure near the boundary. On $`Z^{},`$ it is possible to think of $`ð`$ as a cusp operator by compactifying to a compact manifold with boundary (diffeomorphic to the original $`Z)`$ by replacing $`u`$ by the variable
(14.6)
$$x=\frac{1}{u}[0,1)$$
for $`u(\mathrm{},1).`$ The extension down to $`x=0,`$ gives the manifold with boundary $`\overline{Z^{}},`$ and $`x`$ is a boundary defining function for $`\overline{Z^{}}X`$ which defines a cusp structure. Let us denote by $`ð_{\mathrm{cu}}`$ the natural extension of $`ð`$ to $`\overline{Z^{}}.`$ Near the boundary of $`\overline{Z^{}},`$ $`ð_{\mathrm{cu}}`$ takes the form
(14.7)
$$ð_{\mathrm{cu}}=\gamma \left(x^2\frac{}{x}+A\right)$$
and so is clearly an elliptic cusp differential operator.
###### Lemma 14.1.
If $`A`$ is invertible, then
$$ð_{\mathrm{cu}}^+:\mathrm{H}^1(\overline{Z^{}};E_+)\mathrm{L}^2(\overline{Z^{}};E_{})$$
is Fredholm and has the same index has the operator (14.5).
###### Proof.
Recall () that a cusp operator is Fredholm if and only if it is elliptic and its indicial family is invertible. The indicial family of $`ð_{\mathrm{cu}}^+`$ is given by
$$e^{i\frac{\tau }{x}}ð_{\mathrm{cu}}^+e^{i\frac{\tau }{x}}|_{x=0}=\gamma (Ai\tau ),\tau .$$
Since $`A`$ is self-adjoint, this is invertible for all $`\tau `$ if and only if $`A`$ is invertible. Thus, $`ð_{\mathrm{cu}}`$ is Fredholm if and only if $`A`$ is invertible. That $`ð_{\mathrm{cu}}^+`$ has the same index as (14.5) then follows from Proposition 3.11 in and Proposition 9 in . ∎
When $`A`$ is not invertible, it is still possible to relate (14.5) with a Fredholm cusp operator. In fact, since this is basically the problem we encounter when we consider the family version, let us immediately generalize to this context. For the family case consider a fibration, (3), of a manifold with boundary and we assume now that $`ð`$ is a family of Dirac operators parameterized by $`B`$ which as before are of product near the boundary of $`M.`$ Thus, there is an associated family $`A`$ of self-adjoint Dirac operators on the boundary. The main difficulty in the family case is that interpreted directly, the spectral boundary condition need not be smooth. In , this difficulty was overcome by introducing the notion of a spectral section.
###### Definition 14.1.
A spectral section for a family of elliptic self-adjoint operators
$$A\mathrm{Diff}^1(M/B;𝔼|_M)$$
is a family of self-adjoint projections $`P\mathrm{\Psi }^0(M/B;𝔼|_M)`$ such that for some smooth function $`R:B[0,\mathrm{})`$ (depending on $`P`$) and every $`bB,`$
$$A_bf=\lambda f\{\begin{array}{cc}P_bf=f\hfill & \text{ if }\lambda >R(b),\hfill \\ P_bf=0\hfill & \text{ if }\lambda <R(b).\hfill \end{array}$$
Such a spectral section always exists for the boundary family and any such choice gives a smooth family of boundary problems problem
(14.8)
$$ð^+:W_P^1\mathrm{L}^2(M/B;E_{})$$
where
$$W_P^1=\left\{f\mathrm{H}^1(M/B;E_+);P(f|_M)=0\right\}.$$
The family $`ð^+`$ in (14.8) is Fredholm so has a well defined families index. As before, one can attach a cylindrical end and get a new fibration $`\varphi :M^{}B`$ where the family of operators $`ð`$ naturally extends. By making the change of variable $`x=\frac{1}{u},`$ one get the a family of cusp operators $`ð_{\mathrm{cu}}`$ which takes the form (14.7) near the boundary.
###### Lemma 14.2.
There exists $`Q\mathrm{\Psi }_{\mathrm{cu}}^{\mathrm{}}(M/B;E_+,E_{})`$ such that $`ð_{\mathrm{cu}}^++Q`$ is a fully elliptic (hence Fredholm) family with the same family index as (14.8).
###### Proof.
This is carried out in section 8 of in the context of b-pseudodifferential operators instead of cusp operators. Since the relationship corresponds to the introduction of the transcendental variable $`1/u`$ (for cusp) instead of $`e^u`$ (for b-) one can check that the argument continues to hold for cusp operators with only minor modifications. ∎
###### Proof of Theorem 3.
Let $`(ð,P)`$ be as in the proposition. Define $`[(ð,P)]\mathrm{K}_{\mathrm{cu}}(\varphi )`$ to be the K-class associated to the operator $`ð_{\mathrm{cu}}^++Q`$ of Lemma 14.2. Then by the lemma
$$\mathrm{ind}(ð,P)=\mathrm{ind}_\text{a}([(ð,P)])=\mathrm{ind}_\text{t}([(ð,P)]).$$
That $`[(ð,P)]`$ is canonically defined, that this, does not depend on the choice of $`Q`$ in Lemma 14.2, is a consequence of the relative families index theorem of . Thus, given the Poincaré duality of Theorem 2 and the commutativity of diagram (7), Theorem 3 follows. ∎
## Appendix A Product-type pseudodifferential operators
There are many variants of pseudodifferential operators ‘of product type’ to be found in the literature, see Shubin , Rodino , Melrose-Uhlmann , Hörmander . Here we describe, succinctly, a particularly natural algebra of such operators associated to a fibration (and for families to an iterated fibration). First consider the local setting.
On $`_y^{d_1}\times _z^{d_2}`$ we consider operators corresponding to this product thought of as a fibration over the first factor. The class of symbols admitted is determined by the $`𝒞^{\mathrm{}}`$ structure on the manifold with corners
(A.1)
$$X_{d_1,d_2}=^{d_1}\times ^{d_2}\times [\overline{^{d_1+d_2}};\overline{^{d_1}\times \{0\}}].$$
That is, take the radial compactification of $`^{d_1+d_2}`$ and blow up the boundary (at infinity) of the radial compactification of the subspace $`^{d_1}\times \{0\}.`$ Let $`\rho ,`$ $`\rho _{\mathrm{ff}}𝒞^{\mathrm{}}(X_{d_1.d_2})`$ be defining functions for the two boundary hypersurfaces, the first being the ‘old hypersurface’ at infinity and the second that produced by the blow up.
Now, if $`a\rho ^m\rho _{\mathrm{ff}}^m^{}𝒞_c^{\mathrm{}}(X_{d_1,d_2})`$ then it satisfies the estimates
(A.2)
$$|_y^\alpha _z^\beta _\eta ^\gamma _\zeta ^\delta a|C_{\alpha ,\beta ,\gamma ,\delta }(1+|\zeta |)^{mm^{}|\delta |}(1+|\eta |+|\zeta |)^{m^{}|\gamma |},$$
as follows by noting that one can take $`\rho =(1+|\zeta |^2)^{\frac{1}{2}}`$ and $`\rho _{\mathrm{ff}}=(1+|\zeta |^2)^{\frac{1}{2}}(1+|\eta |^2+|\zeta |^2)^{\frac{1}{2}}.`$ This gives the overall weight in (A.2) with no differentiation. The vector fields
$$_{y_j},_{z_k},\eta _i_{\eta _j},\zeta _l_{\zeta _k},\zeta _k_{\eta _j}$$
all lift to be smooth on $`^{d_1+d_2}\times \overline{^{d_1+d_2}}`$ and tangent to the boundary and within the boundary to the submanifold blown up in (A.1) so
(A.3)
$$_y^\alpha _z^\beta _\eta ^\gamma _\zeta ^\delta a\rho ^{m+\gamma +\delta }\rho _{\mathrm{ff}}^{m^{}+\gamma }𝒞_c^{\mathrm{}}(X_{d_1,d_2});$$
this leads directly to the estimates (A.2). Consider the kernels on $`^{2d_1+2d_2}`$ defined by Weyl quantization of symbols
(A.4)
$$\begin{array}{c}A(y,z,y^{},z^{})=(2\pi )^{d_1d_2}e^{i(yy^{})\eta +(zz^{})\zeta }a(\frac{y+y^{}}{2},\frac{z+z^{}}{2},\eta ,\zeta )𝑑\eta 𝑑\zeta \text{ where }\hfill \\ \hfill a=a_1+a_2+a_3,a_1\rho ^m\rho _{\mathrm{ff}}^m^{}𝒞_c^{\mathrm{}}(X_{d_1,d_2}),\\ \hfill a_2\rho ^{\mathrm{}}𝒮(^{2d_2};\rho _\eta ^m^{}𝒞_c^{\mathrm{}}(^{d_1}\times \overline{^{d_1}})),a_3𝒮(^{2d_1+2d_2}).\end{array}$$
Note that the three terms in the amplitude in (A.4) are really of the same type and the second and third can be included in the first, except that the support conditions are relaxed to rapid decay at infinity. Thus, the third class of symbols corresponds to Schwartz kernels. The second class corresponds to Schwartz functions in $`z,z^{}`$ with values in the the classical pseudodifferential operators of order $`m^{}`$ on $`^{d_1}`$ and with kernels having bounded support in $`y+y^{}.`$ Since these kernels are actually Schwartz if the singularity at $`y=y^{}`$ is cut out, the effect of the second two terms is simply to admit the kernels which are Schwartz functions of $`z,z^{}`$ with values in the pseudodifferential kernels of order $`m^{}`$ on $`^{d_1}`$ with bounded singular support (in $`y)`$ and Schwartz tails. Similarly addition of these two terms ‘completes’ the first term in admitting appropriate tails at infinity to ensure that
###### Proposition A.1.
The operators with kernels as in (A.4) act on $`𝒮(^{d_1+d_2})`$ and form a bifiltered algebra with the orders $`m,m^{};`$ omitting the first term in (A.4) gives an ideal, as does omitting the first two terms. The filtration is delineated by two symbol maps
(A.5)
$$\begin{array}{c}\sigma _{m,m^{}}(A)=\rho ^m\rho _{\mathrm{ff}}^m^{}a_1|_{\rho =0}𝒞_c^{\mathrm{}}(^{d_1+d_2}\times [𝕊^{d_1+d_21};𝕊^{d_11}];N_{m,m^{}})\\ \begin{array}{cc}\hfill L(A)=(2\pi )^{d_2}e^{i(zz^{})\zeta }(b_1& +b_2)(y,\frac{z+z^{}}{2},\widehat{\eta },\zeta )d\zeta ,\hfill \\ \hfill b_1& =(\rho _{\mathrm{ff}}^ma_1)|_{\rho _{\mathrm{ff}}=0},b_2=a_2|_{𝕊^{d_11}}\hfill \end{array}\end{array}$$
which are homomorphisms into the algebras of functions and parameterized pseudodifferential operators on $`^{d_2}`$ respectively.
###### Proof.
All these conclusions follow from the standard methods for proving the composition formula for pseudodifferential operators on Euclidean space, i.e. some form of stationary phase. The fact that the two symbol maps are homomorphism follows by oscillatory testing. ∎
Now, these objects can be transferred to a general fibration of compact manifolds by localization. Thus, the kernels in (A.4) are smooth outside the *two* submanifolds
(A.6)
$$\{y=y^{},z=z^{}\}\{y=y^{}\}$$
and the singularity is determined by the Taylor series of $`a_1`$ at the boundary of $`X_{d_1,d_2}`$ and the Taylor series of $`a_2`$ at the boundary of the ball. Furthermore these singularities are locally determined in the sense that the singularity on the diagonal near a point $`(\overline{y},\overline{z})`$ is determined by $`a_1`$ near $`(\overline{y},\overline{z},\eta ,\zeta )`$ and by $`a_2`$ near $`(\overline{z},\overline{z},\overline{y},\eta ).`$ The singularity near a point on $`y=y^{}`$ away from the diagonal is determined by $`a_1`$ and $`a_2`$ near that point $`y=\overline{y}.`$ So, for a fibration we can associate an algebra of operators using such localizations.
###### Definition A.1.
If $`\stackrel{~}{\varphi }:\stackrel{~}{M}M`$ is a fibration of compact manifolds without boundary then $`\mathrm{\Psi }_{\stackrel{~}{\varphi }\mathrm{p}}^{m^{},m}(\stackrel{~}{M};𝔼)`$ for a $`_2`$-graded vector bundle over $`\stackrel{~}{M}`$ consists of those operators $`A:𝒞^{\mathrm{}}(\stackrel{~}{M};E_+)𝒞^{\mathrm{}}(\stackrel{~}{M};E_{})`$ which have Schwartz kernels on $`\stackrel{~}{M}^2`$ which under local trivializations of the bundles and densities are matrices of distributions which are
1. Smooth away from the fibre diagonal.
2. Near a point of the complement of the diagonal in the fibre diagonal are given by smooth functions of the fibre variables with values in the classical pseudodifferential kernels of order $`m^{}`$ on the base.
3. Near a point of the diagonal are of the form (A.4).
As a consequence of Proposition A.1 we then conclude that the scalar operators in $`\mathrm{\Psi }_{\stackrel{~}{\varphi }\mathrm{p}}^{m^{},m}(\stackrel{~}{M})`$ form a bigraded algebra for $`m,m^{}`$ and that there are corresponding global symbol homomorphisms giving surjective, and multiplicative, maps
(A.7)
$$\begin{array}{c}\sigma :\mathrm{\Psi }_{\stackrel{~}{\varphi }\mathrm{p}}^{m^{},m}(\stackrel{~}{M})𝒞^{\mathrm{}}([\overline{S^{}\stackrel{~}{M}};\stackrel{~}{\varphi }^{}S^{}M];N_{m^{},m})\\ L:\mathrm{\Psi }_{\stackrel{~}{\varphi }\mathrm{p}}^{m^{},m}(\stackrel{~}{M})\mathrm{\Psi }^m(\pi ^{}(\stackrel{~}{M}/M);𝔼_+N_m^{}).\end{array}$$
In the first map, $`N_{m^{},m}`$ is a trivial real bundle corresponding to the coefficients $`\rho ^m^{}`$ and $`\rho ^m`$ in the symbols and in the second sequence, the pseudodifferential operators act on the fibres of the lift of the fibration from $`M`$ to $`S^{}M`$ using the projection $`\pi :S^{}MM`$ and there is an additional trivial line bundle corresponding to the factor $`\rho _{\mathrm{ff}}^m^{}.`$ As noted, both maps are surjective, but together they are constrained precisely by the fact that the symbol of $`L(A),`$ as a family, is given by $`\sigma (A)`$ evaluated at the front face of the blow up of $`S^{}\stackrel{~}{M}.`$
Apart from these composition properties, and their natural generalizations to the case of families over a second fibration, it is important to note certain inclusions.
First,
(A.8)
$$\mathrm{\Psi }^m(\stackrel{~}{M};𝔼)\mathrm{\Psi }_{\stackrel{~}{\varphi }\mathrm{p}}^{m,m}(\stackrel{~}{M};𝔼).$$
Indeed, the definition of $`\mathrm{\Psi }_{\stackrel{~}{\varphi }\mathrm{p}}^{m,m}(\stackrel{~}{M};𝔼)`$ is modelled on one of the standard definitions of the usual pseudodifferential operators, so it is enough to refer back to the model case. Directly from the definition of the symbols in (A.4), the classical symbols of order $`m,`$ which are just elements of $`\rho ^m𝒞_c^{\mathrm{}}(^{d_1+d_2}\times \overline{^{d_1+d_2}}),`$ lift to product type symbols of order $`(m,m)`$ under the blow up. This leads immediately to (A.8). Note that in this case, the symbol in the product-type sense is just the lift of the symbol in the usual sense (actually loosing no information by continuity) and the base family is simply again the symbol, evaluated on the lift of the cosphere bundle from the base and interpreted as acting as bundle isomorphisms (so local operators) on the fibres.
The second inclusion is of pseudodifferential operators acting on the fibres of the fibration. By definition the kernels of these operators are smooth families in the base variables, with values in the classical pseudodifferential operators on the fibres; as operators on smooth sections over the total space they are therefore of the form of a product of a classical conormal distribution with a delta section on the fibre diagonal. Working locally this reduces to (A.4) with $`a_1`$ or $`a_2`$ actually independent of $`\eta ;`$ it follows directly that
(A.9)
$$\mathrm{\Psi }^m(\stackrel{~}{M}/M;𝔼)\mathrm{\Psi }_{\stackrel{~}{\varphi }\mathrm{p}}^{0,m}(\stackrel{~}{M};𝔼).$$
The symbol map reduces to the lift of the symbol on the fibres and the base family of such an operator is the operator itself.
The third inclusion is simply of pseudodifferential operators on the base but acting on a bundle $`𝔼`$ which is embedded as a subbundle of $`𝒞^{\mathrm{}}(\stackrel{~}{M}/M;𝔽)`$ for some bundle $`𝔽`$ over $`\stackrel{~}{M}.`$ This embedding corresponds to a family of smoothing projections of finite rank $`\pi _\pm \mathrm{\Psi }^{\mathrm{}}(\stackrel{~}{M}/M;𝔽)`$ and the kernel can then be written, somewhat formally, as $`\pi _{}K(A)\pi _+.`$ Again this is everywhere locally of the form (A.4), with the fibre part of order $`\mathrm{}`$ and it follows that
(A.10)
$$\mathrm{\Psi }^m(M;𝔼)\mathrm{\Psi }_{\stackrel{~}{\varphi }\mathrm{p}}^{m,\mathrm{}}(\stackrel{~}{M};𝔽).$$
In this case, the operator being of principal order $`\mathrm{},`$ the symbol is zero at any order, but the base family is simply $`\pi _{}\sigma (A)\pi _+`$ acting on sections of $`𝔽.`$
We pass over without extensive comment the extension of this construction to define families of operators, with respect to an overall fibration, and more generally suspended families in which the cotangent variables of the base are symbolic variables in the operators. These latter parameters can always be incorporated into the operators as the duals of additional Euclidean (base) variables in which the operators are translation invariant.
## Appendix B Product-type fibred cusp operators
The discussion above of operators of product-type can be extended to fibred cusp pseudodifferential operators. In such an extension, as in the subsequent one to an adiabatic limit, the extension is based on the principle that the product-type operators above are defined through a geometric class of distributions, the product-type conormal distributions for the pair of embedded submanifolds
$$\mathrm{Diag}\mathrm{Diag}_{\stackrel{~}{\varphi }}.$$
So, to generalize these operators to another setting it is only necessary to start with a space of operators associated with the conormal distributions on an embedded submanifold, replacing $`\mathrm{Diag},`$ and to replace these in turn by an appropriate class of product-type distributions.
This is precisely the case with the fibred cusp operators defined and discussed in for a compact manifold with boundary $`M`$ with a given fibration of the boundary $`\mathrm{\Phi }:MD`$ and a choice of normal trivialization of the fibres. The latter choice can be represented by a choice of boundary defining function $`x𝒞^{\mathrm{}}(M).`$ Then the product $`M^2`$ on which kernels are normally defined, is replace by a blown-up version of it. Namely first the corner is blown up
(B.1)
$$M_\mathrm{b}^2=[M^2;(M)^2]$$
(when the boundary is not connected all products of pairs of boundary components should be blown up.) Within the new, or front, face of this manifold with corners the fibration and choice of normal trivialization combine to define a submanifold
(B.2)
$$\mathrm{\Gamma }_{\mathrm{\Phi }\mathrm{cu}}\mathrm{ff}(M_\mathrm{b}^2).$$
Namely $`\mathrm{\Gamma }_{\mathrm{\Phi }\mathrm{cu}}`$ is the fibre diagonal in the boundary variables intersected with the submanifold $`\frac{xx^{}}{x+x^{}}=0`$ where it should be observed that this *is* a smooth function on $`M_\mathrm{b}^2`$ and that the resulting submanifold does only depend on the data giving the fibred cusp structure. Then the kernels for fibred cusp pseudodifferential operators are simply the standard conormal distributional sections with respect to the lifted diagonal of an appropriate smooth bundle over
(B.3)
$$\mathrm{Diag}_{\mathrm{\Phi }\mathrm{cu}}M_{\mathrm{\Phi }\mathrm{cu}}^2=[M_\mathrm{b}^2;\mathrm{\Gamma }_{\mathrm{\Phi }\mathrm{cu}}].$$
The lifted diagonal is an embedded p-submanifold, i.e. has a product type decomposition at the boundaries. Thus the definition of the kernels as the conormal distributions (which are also required to vanish to infinite order at boundary faces not meeting the diagonal) is meaningful.
Here we consider the case of a fibration over the manifold with boundary, $`\stackrel{~}{\varphi }:\stackrel{~}{M}M`$ which has compact fibres without boundary. We suppose that $`M`$ has a fibred cusp structure as above and that the fibres of $`\stackrel{~}{\varphi }`$ are to be treated as part of the boundary fibres, i.e. we take the fibred cusp structure on $`\stackrel{~}{M}`$ given by the fibration
(B.4)
$$\stackrel{~}{\mathrm{\Phi }}:MD,\stackrel{~}{\mathrm{\Phi }}=(\stackrel{~}{\varphi })\mathrm{\Phi }$$
and with the normal trivialization given by lifting an admissible defining function on $`M.`$
In the construction of $`M_{\mathrm{\Phi }\mathrm{cu}}^2,`$ each blow up is near the corners of $`M^2`$ and the procedure is local with respect to the fibration of the boundary. That is, if $`\mathrm{\Phi }`$ is locally trivialized to a product $`O\times X,`$ consistent with the fibred cusp structure, where $`O`$ is an open set in $`[0,\mathrm{})\times ^l`$ then over the preimage of this set the stretched product is just
(B.5)
$$M_{\mathrm{\Phi }\mathrm{cu}}^2O_{\mathrm{sc}}^2\times X^2.$$
Here, $`O_{\mathrm{sc}}^2=O_{\mathrm{Id}\mathrm{cu}}^2`$ is the stretched product in the scattering case, that is, the case where the boundary map is the identity. Thus, when the extra fibration is added it follows that, again locally near boundary points and in small open sets $`U,U^{}X`$ (so that $`\stackrel{~}{\varphi }`$ is also locally trivialized)
(B.6)
$$\stackrel{~}{M}_{\stackrel{~}{\mathrm{\Phi }}\mathrm{cu}}^2O_{\mathrm{sc}}^2\times U\times U^{}\times Z^2.$$
In fact $`\stackrel{~}{M}_{\stackrel{~}{\mathrm{\Phi }}\mathrm{cu}}^2`$ fibres over $`M_{\mathrm{\Phi }\mathrm{cu}}^2`$ with fibre which is $`Z^2.`$ This shows that the geometric situation of the lifted diagonal and the lifted fibre diagonal, which is just the diagonal in $`M_{\mathrm{\Phi }\mathrm{cu}},`$ is completely analogous to the product-type setting discussed above.
###### Definition B.1.
If $`\stackrel{~}{\varphi }:\stackrel{~}{M}M`$ is a fibration with fibres compact manifolds without boundary over a compact manifold with fibred cusp structure then the space $`\mathrm{\Psi }_{\stackrel{~}{\varphi }\mathrm{p},\mathrm{\Phi }\mathrm{cu}}^{m^{},m}(\stackrel{~}{M};𝔼)`$ is given by the space of product-type conormal distributions, as in Definition A.1, with respect to the lifted diagonal and fibred diagonal, and vanishing to infinite order at all boundary faces which do not meet these p-submanifolds.
As shown in , the composition properties of fibred cusp operators follow from geometric considerations. Namely if the product is interpreted as a push-forward for an appropriately defined triple space (as in ) then locally for the fibred cusp calculus the problem is the same uniformly up to the boundary, and hence follows from the discussion above. Thus we conclude that
(B.7)
$$\mathrm{\Psi }_{\stackrel{~}{\varphi }\mathrm{p},\mathrm{\Phi }\mathrm{cu}}^{m^{},m}(\stackrel{~}{M};𝔼)\mathrm{\Psi }_{\stackrel{~}{\varphi }\mathrm{p},\mathrm{\Phi }\mathrm{cu}}^{p^{},p}(\stackrel{~}{M};𝔽)\mathrm{\Psi }_{\stackrel{~}{\varphi }\mathrm{p},\mathrm{\Phi }\mathrm{cu}}^{m^{}+p^{},m+p}(\stackrel{~}{M};𝔾)$$
provided the product makes sense, i.e. $`E_+=F_{},`$ $`G_+=E_+,`$ $`G_{}=F_{}.`$ Furthermore there are now three ‘symbol homomorphisms’. Two of these are modified versions of the corresponding homomorphism for the fibred cusp calculus. Thus, the symbol map now takes values, as in (A.7), in sections of the appropriate bundle over a blown-up version of the fibred-cusp cosphere bundle
(B.8)
$$\sigma :\mathrm{\Psi }_{\stackrel{~}{\varphi }\mathrm{p},\mathrm{\Phi }\mathrm{cu}}^{m^{},m}(\stackrel{~}{M})𝒞^{\mathrm{}}([\overline{{}_{}{}^{\mathrm{cu}}S_{}^{}\stackrel{~}{M}};\stackrel{~}{\varphi }^{}{}_{}{}^{\mathrm{cu}}S_{}^{}M];N_{m^{},m}).$$
Similarly the indicial operator, which corresponds geometrically to the restriction of the kernel to the final front face in the blown-up space, now takes values in product-type and suspended operators on the boundary
(B.9)
$$N:\mathrm{\Psi }_{\stackrel{~}{\varphi }\mathrm{p},\mathrm{\Phi }\mathrm{cu}}^{m^{},m}(\stackrel{~}{M};𝔼)\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{sus},\stackrel{~}{\varphi }\mathrm{p}}^{m,m^{}}(\stackrel{~}{M};𝔼).$$
The base map can also be defined by oscillatory testing
(B.10)
$$L:\mathrm{\Psi }_{\stackrel{~}{\varphi }\mathrm{p},\mathrm{\Phi }\mathrm{cu}}^{m^{},m}(\stackrel{~}{M};𝔼)\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^m(\pi ^{}(\stackrel{~}{M}/M);𝔼_+N_m^{}).$$
Each of these maps delineates a filtration of the algebra, corresponding to the order $`m,`$ the degree of boundary vanishing $`x`$ and the order $`m^{}.`$ They are separately surjective and jointly, in pairs or all together, subject only to the natural compatibility conditions
(B.11)
$$\sigma (N)=\sigma _{},\sigma (L)=\sigma _{\mathrm{ff}},N(L)=L(N).$$
## Appendix C Adiabatic limit in the fibres
A notion of the adiabatic limit of pseudodifferential operators was introduced in . Namely, for a fibration of compact manifolds, for the moment without boundary, $`\stackrel{~}{\varphi }:\stackrel{~}{M}M,`$ one can consider a sense in which pseudodifferential operators on $`\stackrel{~}{M}`$ degenerate, by localizing in the base variables, to become families of pseudodifferential operators on the fibres of $`\stackrel{~}{\varphi }.`$
To do so, let $`ϵ[0,1]`$ be the ‘adiabatic parameter.’ If we consider kernels given by conormal distributions with respect to the submanifold
(C.1)
$$\mathrm{Diag}\times [0,1]\stackrel{~}{M}^2\times [0,1]$$
we simply arrive at the $`ϵ`$-parameterized pseudodifferential operators on $`\stackrel{~}{M}.`$ On the other hand, if we blow up the fibre diagonal at $`ϵ=0,`$ introducing
(C.2)
$$\stackrel{~}{M}_{\stackrel{~}{\varphi }\mathrm{ad}}^2=[\stackrel{~}{M}^2\times [0,1];\mathrm{Diag}_{\stackrel{~}{\varphi }}\times \{0\}]\stackrel{𝛽}{}\stackrel{~}{M}^2\times [0,1]$$
we obtain a manifold with corners with two important boundary faces (we ignore $`ϵ=1`$ as being ‘regular’), the ‘old boundary’ being the proper lift, $`\beta ^\mathrm{\#}\{ϵ=0\}`$ and the new ‘front face’ produced by the blow up. The diagonal has proper lift to a smooth p-submanifold
(C.3)
$$\beta ^\mathrm{\#}(\mathrm{Diag}\times [0,1])\stackrel{~}{M}_{\stackrel{~}{\varphi }\mathrm{ad}}^2$$
which meets the boundary only in the front face. The space of adiabatic pseudodifferential operators is then
(C.4)
$$\begin{array}{c}\mathrm{\Psi }_{\stackrel{~}{\varphi }\mathrm{ad}}^m(\stackrel{~}{M};𝔼)=\{KI^{m\frac{1}{4}}(\stackrel{~}{M}_{\stackrel{~}{\varphi }\mathrm{ad}}^2,\beta ^\mathrm{\#}(\mathrm{Diag}\times [0,1]);\mathrm{hom}(𝔼)\mathrm{\Omega }_{\stackrel{~}{\varphi }\mathrm{ad}});\hfill \\ \hfill K0\text{ at }\beta ^\mathrm{\#}\{ϵ=0\}\}.\end{array}$$
In fact a neighbourhood of $`\beta ^\mathrm{\#}(\mathrm{Diag}\times [0,1])`$ in $`\stackrel{~}{M}_{\stackrel{~}{\varphi }\mathrm{ad}}^2`$ is diffeomorphic to a neighbourhood of $`\mathrm{Diag}\times [0,1]`$ in $`\stackrel{~}{M}^2\times [0,1]`$ so we may legitimately think of these kernels as having the ‘same singularities’ as the ordinary pseudodifferential families but with a different action. The (trivial) density line bundle in (C.4) takes care of factors that arise even for the identity.
These operators compose in the expected way and have two symbols. The first is the usual symbol, now taking values in sections of the appropriate bundle over the ‘adiabatic cosphere bundle’ (which is a bundle over $`\stackrel{~}{M}\times [0,1])`$ and giving a short exact sequence
(C.5)
Secondly there is a symbol representing the limit at $`ϵ=0.`$ It is a suspended family of pseudodifferential operators on the fibres of the fibration with (symbolic) parameters in the rescaled cotangent bundle of the base
(C.6)
Now, in the multiplicativity construction in Section 10 we use a quite analogous construction to pass from fibred cusp operators with respect to a fibration and a boundary fibration to fibred cusp operators for the same fibration and a finer boundary fibration, that is with smaller fibres. Thus we are ‘converting’ some fibre variables in the boundary to base variables. The main step is to carry this out on one fibre, so we can consider the model case of a compact manifold with boundary $`Z`$ with an iterated fibration of its boundary giving a commutative diagram
(C.7)
so the map from $`\stackrel{~}{D}`$ to $`D`$ is also a fibration.
The construction of the fibred cusp calculus on $`X`$ is briefly discussed above. Adding an adiabatic parameter $`ϵ[0,1]`$ we can consider the $`ϵ`$-parameterized fibred cusp calculus with respect to $`\stackrel{~}{\mathrm{\Phi }}`$ (and some choice of fibred cusp structure) by replacing (B.3) by
(C.8)
$$Z_{\mathrm{\Phi }\mathrm{cu}}^2\times [0,1]=[Z_\mathrm{b}^2\times [0,1];\mathrm{\Gamma }_{\mathrm{\Phi }\mathrm{cu}}\times [0,1]].$$
Now, at $`ϵ=0`$ we consider the finer boundary fibration given by $`\stackrel{~}{\mathrm{\Phi }},`$ with a consistent fibred cusp structure (meaning such that there is a global boundary defining function which induces both). Note that the corresponding ‘lifted’ fibre diagonal in the boundary is then a p-submanifold
(C.9)
$$\mathrm{\Gamma }_{\stackrel{~}{\mathrm{\Phi }}\mathrm{cu}}\mathrm{\Gamma }_{\mathrm{\Phi }\mathrm{cu}}\mathrm{\Gamma }_{\stackrel{~}{\mathrm{\Phi }}\mathrm{cu}}\{ϵ=0\}\mathrm{\Gamma }_{\mathrm{\Phi }\mathrm{cu}}\times [0,1].$$
In particular this means that the proper lift of this manifold under the blow up in (C.8) is again a p-submanifold which can be blown up, giving the new compact manifold with corners
(C.10)
$$\begin{array}{c}Z_{\mathrm{\Phi }\mathrm{cu},\stackrel{~}{\mathrm{\Phi }}\mathrm{ad}}^2=[Z_{\mathrm{\Phi }\mathrm{cu}}^2\times [0,1];\mathrm{\Gamma }_{\stackrel{~}{\mathrm{\Phi }}\mathrm{cu}}\{ϵ=0\}]\hfill \\ \hfill =[Z_\mathrm{b}^2\times [0,1];\mathrm{\Gamma }_{\mathrm{\Phi }\mathrm{cu}}\times [0,1],\mathrm{\Gamma }_{\stackrel{~}{\mathrm{\Phi }}\mathrm{cu}}\{ϵ=0\}].\end{array}$$
The kernels of the operators we consider will be defined on this manifold. Ignoring $`ϵ=1`$ as we shall, there are three boundary faces which meet the proper lift of the diagonal (with a factor of $`[0,1]),`$ which as usual is a p-submanifold. Two of these are the boundary hypersurfaces produced by the blow-ups in (C.10), the first is essentially the front face of the $`\mathrm{\Phi }`$-fibred cusp calculus with an extra factor of $`[0,1]`$ but also blown up at $`ϵ=0`$ corresponding to the $`\stackrel{~}{\mathrm{\Phi }}`$-fibred cusp calculus. The second is the front face produced by the last blow-up over $`ϵ=0`$ and the third is the proper lift of $`ϵ=0,`$ this is simply the manifold with corners $`X_{\stackrel{~}{\mathrm{\Phi }}\mathrm{cu}}^2.`$
Now, the space of $`\stackrel{~}{\mathrm{\Phi }}`$-adiabatic, $`\mathrm{\Phi }`$-fibred cusp pseudodifferential operators on $`Z,`$ $`\mathrm{\Psi }_{\stackrel{~}{\mathrm{\Phi }}\mathrm{ad},\mathrm{\Phi }\mathrm{cu}}^m(Z)`$ is identified with the conormal distributions, of an appropriate density bundle, on $`Z_{\mathrm{\Phi }\mathrm{cu},\stackrel{~}{\mathrm{\Phi }}\mathrm{ad}}^2`$ with respect to the proper lift of the diagonal and vanishing to infinite order at all boundary faces which do not meet this lift. These operators map $`𝒞^{\mathrm{}}(\stackrel{~}{M}_{\mathrm{ad}};E_+),`$ to $`𝒞^{\mathrm{}}(\stackrel{~}{M}_{\mathrm{ad}};E_{})`$ where
(C.11)
$$\stackrel{~}{M}_{\mathrm{ad}}=[\stackrel{~}{M}\times [0,1];\stackrel{~}{M}\times \{0\}]$$
and compose in the usual way.
Corresponding to this definition and the discussion above of boundary hypersurfaces, there are four ‘symbol maps’, the symbol, the indicial operator, the adiabatic symbol and the adiabatic operator.
The first two of these correspond to the symbol for the $`\mathrm{\Phi }`$-fibred cusp calculus and its indicial operator, with dependence on the parameter $`ϵ`$ with a change of uniformity at $`ϵ=0.`$ The third symbol is the real transition between the $`\mathrm{\Phi }`$\- and $`\stackrel{~}{\mathrm{\Phi }}`$-fibred cusp calculi, and the last is simply the limiting $`\stackrel{~}{\mathrm{\Phi }}`$-fibred cusp operator itself. More precisely, the ‘usual’ symbol becomes
(C.12)
Note that the symbols here are section of the tensor product of $`\mathrm{hom}(𝔼)`$ with a density bundle over the modified cosphere bundle to $`\stackrel{~}{M}_{\mathrm{ad}}.`$ This coshere bundle is associated to the cotangent bundle dual to the tangent bundle of $`\stackrel{~}{M},`$ rescaled near the boundary, with locally generating sections over $`\stackrel{~}{M}\times [0,1]`$ near a boundary point of the corner given by the vector fields (which lift to be smooth on $`\stackrel{~}{M}_{\mathrm{ad}})`$
$$x^2_x,x_y,(x^2+ϵ^2)^{\frac{1}{2}}_y^{},_z$$
where $`x`$ is the normal variable, the $`y`$’s are base variables for $`\mathrm{\Phi },`$ the $`y^{}`$’s are the extra base variables for $`\stackrel{~}{\mathrm{\Phi }}`$ (so fibre variables for $`\mathrm{\Phi })`$ and the $`z`$’s are fibre variables for $`\stackrel{~}{\mathrm{\Phi }}.`$
The normal, or indicial operator corresponds to the restriction of the kernel to the front face of $`\stackrel{~}{M}_{\mathrm{\Phi }\mathrm{cu}}^2\times [0,1]`$ after part of its boundary is blown up in the last step in the construction, (C.10). Since this can be related directly to the adiabatic construction for the fibre calculus of $`\mathrm{\Phi }`$ with respect to $`\stackrel{~}{\mathrm{\Phi }}`$ it becomes a map into the corresponding adiabatic (and suspended) calculus
(C.13)
Note that the factor $`\stackrel{~}{x}𝒞^{\mathrm{}}(\stackrel{~}{M}_{\mathrm{ad}})`$ is a defining function for the boundary after the last blow-up, it can be taken to be $`x(x^2+ϵ^2)^{\frac{1}{2}}.`$
The transitional, adiabatic normal operator corresponds to the restriction of the kernel to the front face produced in the final blow-up in (C.10) and takes values in a suspended space of pseudodifferential operators on the fibres of $`\stackrel{~}{\mathrm{\Phi }}`$ giving a short exact sequence
(C.14)
$$\text{},$$
where $`V[1,1]_\tau \times \stackrel{~}{D}`$ is is the null bundle of the restriction $`{}_{}{}^{\stackrel{~}{\mathrm{\Phi }}}T_{\stackrel{~}{M}}^{}MT_{\stackrel{~}{M}}\stackrel{~}{M}`$ and
$$\tau =\frac{xϵ}{x+ϵ}[1,1]$$
is a variable on the adiabatic front face. Finally, the fourth map, into the finer fibred cusp calculus gives a short exact sequence
(C.15)
Here $`\stackrel{~}{ϵ}=ϵ(x^2+ϵ^2)^{\frac{1}{2}}`$ is a defining function for the limiting boundary in $`\stackrel{~}{M}_{\mathrm{ad}}.`$
Notice that
(C.16)
$$P\mathrm{\Psi }_{\stackrel{~}{\mathrm{\Phi }}\mathrm{ad},\mathrm{\Phi }\mathrm{cu}}^m(\stackrel{~}{M}/B;𝔼),N(P)=0,\mathrm{ad}(P)=0Px\mathrm{\Psi }_{\stackrel{~}{\mathrm{\Phi }}\mathrm{ad},\mathrm{\Phi }\mathrm{cu}}^m(\stackrel{~}{M}/B;𝔼).$$
The various compatibility conditions are then given by
(C.17)
$$\begin{array}{c}A\sigma (P)=\sigma A(P),\sigma N(P)=N\sigma (P),\mathrm{ad}\sigma (P)=\sigma \mathrm{ad}(P),\\ \mathrm{ad}N(P)=\mathrm{ad}(P)|_{\tau =1}\text{and}NA(P)=\mathrm{ad}(P)|_{\tau =1}.\end{array}$$
In particular there is no compatibility condition between $`A(P)`$ and $`N(P).`$
As noted above there is in fact a fifth map corresponding to restriction to $`ϵ=1`$ (or really any positive value of $`ϵ)`$
(C.18)
In some sense, $`P\mathrm{\Psi }_{\stackrel{~}{\mathrm{\Phi }}\mathrm{ad},\mathrm{\Phi }\mathrm{cu}}^m(\stackrel{~}{M};𝔼)`$ should be interpreted as a homotopy between $`A(P)\mathrm{\Psi }_{\stackrel{~}{\mathrm{\Phi }}\mathrm{cu}}^m(\stackrel{~}{M};𝔼)`$ and $`P|_{ϵ=1}\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^m(\stackrel{~}{M};𝔼).`$ When both $`A(P)`$ and $`P|_{ϵ=1}`$ are Fredholm operators, one would expect them to have the same index provided $`P`$ is a homotopy through Fredholm operators. The next proposition makes this statement more precise.
###### Proposition C.1.
If $`P\mathrm{\Psi }_{\stackrel{~}{\mathrm{\Phi }}\mathrm{ad},\mathrm{\Phi }\mathrm{cu}}^m(\stackrel{~}{M};𝔼)`$ exists with $`\sigma (P),`$ $`N(P)`$ and $`\mathrm{ad}(P)`$ invertible in their pseudodifferential (or symbol) calculi then $`P|_{ϵ=1}`$ and $`A(P)`$ are both fully elliptic and have the same (families) index.
###### Proof.
The compatibility conditions between $`\sigma (P),`$ $`N(P)`$ and $`\mathrm{ad}(P)`$ are such that if each is invertible within the calculus of pseudodifferential operators then the inverses are compatible. Thus, under this condition a parametrix can be constructed for $`P.`$ As usual, a symbolic parametrix can be improved to a full parametrix in the sense that $`Q\mathrm{\Psi }_{\stackrel{~}{\mathrm{\Phi }}\mathrm{ad},\mathrm{\Phi }\mathrm{cu}}^m(\stackrel{~}{M};𝔼^{})`$ satisfies
(C.19)
$$QP\mathrm{Id}x^{\mathrm{}}\mathrm{\Psi }_{\stackrel{~}{\mathrm{\Phi }}\mathrm{ad},\mathrm{\Phi }\mathrm{cu}}^{\mathrm{}}(\stackrel{~}{M};E_+),PQ\mathrm{Id}x^{\mathrm{}}\mathrm{\Psi }_{\stackrel{~}{\mathrm{\Phi }}\mathrm{ad},\mathrm{\Phi }\mathrm{cu}}^{\mathrm{}}(\stackrel{~}{M};E_{}).$$
Note that we do *not* achieve vanishing at $`ϵ=0`$ since we have not assumed that $`A(P)`$ is invertible. In fact the errors in (C.19) are just smoothing operators, smooth in $`ϵ`$ and with kernels vanishing to infinite order at the boundary. Following the discussion of the index in Section 1 the index bundle (for a family of such operators) can be stabilized to a bundle over $`[0,1]\times B.`$ Indeed, this follows from the fact that
$$x^{\mathrm{}}\mathrm{\Psi }_{\stackrel{~}{\mathrm{\Phi }}\mathrm{ad},\mathrm{\Phi }\mathrm{cu}}^{\mathrm{}}(\stackrel{~}{M};E_+)=𝒞^{\mathrm{}}([0,1],\dot{\mathrm{\Psi }}^{\mathrm{}}(\stackrel{~}{M};E_+))$$
where $`\dot{\mathrm{\Psi }}^{\mathrm{}}(\stackrel{~}{M};E_+)=x^{\mathrm{}}\mathrm{\Psi }_{\mathrm{\Phi }\mathrm{cu}}^{\mathrm{}}(\stackrel{~}{M};E_+)`$ does not depend on the choice of the boundary fibration $`\mathrm{\Phi }.`$ It follows from this that the families index at $`ϵ=0`$ and $`ϵ=1`$ are the same and the former is the index bundle for $`A(P),`$ the latter for $`P|_{ϵ=1}.`$ |
warning/0507/quant-ph0507006.html | ar5iv | text | # Properties of Fermion Spherical Harmonics
## 1 Introduction
Our previous derivation and presentation of the Fermion Spherical Harmonics (i.e. $`Y_{\mathrm{}}^m(\theta ,\varphi )`$ for $`\mathrm{}`$ and $`m`$ half odd-integers) was motivated by the idea that the half-odd-integer spherical harmonics could be a useful representation of particle spin; in particular for the electron its spin is associated with a magnetic moment having an orientation in space specifiable by the spherical polar angles $`\theta `$ and $`\varphi `$, and the two Fermion spherical harmonics for $`\mathrm{}=\frac{1}{2}`$ and $`m=\pm \frac{1}{2}`$ correspond to the experimentally established spin states of the electron. This explicit dependence upon the orientation in space of the electron’s magnetic moment is a distinct advantage over the commonly used abstract spin eigenfunctions $`\alpha `$ and $`\beta `$ \[2, §10.1,p.282\], because the latter have no dependence upon any coordinates.
The investigation reported here was initiated by the observation that the Fermion Spherical Harmonics are not always inter-related by the well-known ladder operators \[2, §5.4,p.115\].
Our objectives in this research were threefold:
* to confirm that these functions of $`\theta `$ and $`\varphi `$ are eigenfunctions of the angular momentum operators, $`𝐌_𝐳`$ and $`𝐌^\mathrm{𝟐}`$, with the expected eigenvalues,<sup>1</sup><sup>1</sup>1The notation $`𝐌_𝐳`$, $`𝐌^\mathrm{𝟐}`$, is taken from Levine \[2, p.116,¶ 1\] as denoting any kind of angular momentum, whereas spin angular momentum is usually denoted by $`𝐒_𝐳`$, $`𝐒^\mathrm{𝟐}`$, orbital angular momentum by $`𝐋_𝐳`$, $`𝐋^\mathrm{𝟐}`$, and total angular momentum by $`𝐉_𝐳`$, $`𝐉^\mathrm{𝟐}`$.
* to determine to what extent the ladder operators operating upon these functions produce the usual, expected results (since they do fail in some cases),
* to consider if there are reasons (failure of the ladder operators or otherwise) that make these functions unsatisfactory as a representation of spin angular momentum.
### 1.1 Angular Momentum Operators
The following definitions of the cartesian-components of the angular momentum operators (in spherical polar angles, $`\theta ,\varphi `$) are taken from McQuarrie \[3, eq.6-85, page 217\]
$`𝐌_𝐱`$ $`=`$ $`{\displaystyle \frac{\mathrm{}}{i}}\left(\mathrm{sin}\varphi {\displaystyle \frac{}{\theta }}\mathrm{cot}\theta \mathrm{cos}\varphi {\displaystyle \frac{}{\varphi }}\right)`$ (1)
$`𝐌_𝐲`$ $`=`$ $`{\displaystyle \frac{\mathrm{}}{i}}\left(\mathrm{cos}\varphi {\displaystyle \frac{}{\theta }}\mathrm{cot}\theta \mathrm{sin}\varphi {\displaystyle \frac{}{\varphi }}\right)`$ (2)
$`𝐌_𝐳`$ $`=`$ $`{\displaystyle \frac{\mathrm{}}{i}}{\displaystyle \frac{}{\varphi }}`$ (3)
From these definitions the form of the operator for the square of the total angular momentum, $`𝐌^\mathrm{𝟐}`$ is \[3, eq.6-45, page 207\].
$`𝐌^\mathrm{𝟐}=𝐌_𝐱^\mathrm{𝟐}+𝐌_𝐲^\mathrm{𝟐}+𝐌_𝐳^\mathrm{𝟐}=\mathrm{}^2({\displaystyle \frac{^2}{\theta ^2}}+\mathrm{cot}\theta {\displaystyle \frac{}{\theta }}+{\displaystyle \frac{1}{\mathrm{sin}^2\theta }}{\displaystyle \frac{^2}{\varphi ^2}})`$ (4)
### 1.2 Ladder Operators
The Ladder Operators, $`𝐌_+`$ and $`𝐌_{}`$ are defined by (\[2, p.116\]):
$`𝐌_+=𝐌_𝐱+i𝐌_𝐲𝐌_{}=𝐌_𝐱i𝐌_𝐲`$ (5)
whose form in terms of $`\theta `$ and $`\varphi `$ is \[4, p.182\]:
$`𝐌_+=\mathrm{}e^{(i\varphi )}\left({\displaystyle \frac{}{\theta }}+i\mathrm{cot}\theta {\displaystyle \frac{}{\varphi }}\right)𝐌_{}=\mathrm{}e^{(i\varphi )}\left({\displaystyle \frac{}{\theta }}i\mathrm{cot}\theta {\displaystyle \frac{}{\varphi }}\right)`$ (6)
which are the appropriate forms for our purposes.
## 2 Spherical Harmonics
It is true in general (for integer and half-odd integer values of $`\mathrm{}`$ and $`m`$) that a spherical harmonic has the form:
$`Y_{\mathrm{}}^m=e^{(im\varphi )}(\mathrm{sin}\theta )^{|m|}P_{\mathrm{}}^{|m|}(\mathrm{cos}\theta )`$ (7)
where $`P_{\mathrm{}}^{|m|}(\mathrm{cos}\theta )`$ is a polynomial in $`\mathrm{cos}\theta `$ of order $`(\mathrm{}|m|)`$ (this order is always an integer even when $`\mathrm{}`$ and $`m`$ are half-odd-integers). Some of the Legendre functions are shown explicitly in Table 1 of our previous paper \[1, p.796\]; from this table the first twelve Fermion spherical harmonics have the explicit (unnormalized) forms:
$`Y_{\frac{1}{2}}^{\frac{1}{2}}=\sqrt{\mathrm{sin}\theta }e^{(\frac{1}{2}i\varphi )}`$ $`Y_{\frac{1}{2}}^{\frac{1}{2}}=\sqrt{\mathrm{sin}\theta }e^{(\frac{1}{2}i\varphi )}`$ (8)
$`Y_{\frac{3}{2}}^{\frac{3}{2}}=(\mathrm{sin}\theta )^{\frac{3}{2}}e^{(\frac{3}{2}i\varphi )}`$ $`Y_{\frac{3}{2}}^{\frac{3}{2}}=(\mathrm{sin}\theta )^{\frac{3}{2}}e^{(\frac{3}{2}i\varphi )}`$ (9)
$`Y_{\frac{3}{2}}^{\frac{1}{2}}=\mathrm{cos}\theta \sqrt{\mathrm{sin}\theta }e^{(\frac{1}{2}i\varphi )}`$ $`Y_{\frac{3}{2}}^{\frac{1}{2}}=\mathrm{cos}\theta \sqrt{\mathrm{sin}\theta }e^{(\frac{1}{2}i\varphi )}`$ (10)
$`Y_{\frac{5}{2}}^{\frac{5}{2}}=(\mathrm{sin}\theta )^{\frac{5}{2}}e^{(\frac{5}{2}i\varphi )}`$ $`Y_{\frac{5}{2}}^{\frac{5}{2}}=(\mathrm{sin}\theta )^{\frac{5}{2}}e^{(\frac{5}{2}i\varphi )}`$ (11)
$`Y_{\frac{5}{2}}^{\frac{3}{2}}=\mathrm{cos}\theta (\mathrm{sin}\theta )^{\frac{3}{2}}e^{(\frac{3}{2}i\varphi )}`$ $`Y_{\frac{5}{2}}^{\frac{3}{2}}=\mathrm{cos}\theta (\mathrm{sin}\theta )^{\frac{3}{2}}e^{(\frac{3}{2}i\varphi )}`$ (12)
$`Y_{\frac{5}{2}}^{\frac{1}{2}}=(14\mathrm{cos}^2\theta )\sqrt{\mathrm{sin}\theta }e^{(\frac{1}{2}i\varphi )}`$ $`Y_{\frac{5}{2}}^{\frac{1}{2}}=(14\mathrm{cos}^2\theta )\sqrt{\mathrm{sin}\theta }e^{(\frac{1}{2}i\varphi )}`$ (13)
## 3 The Eigenfunction Properties
In this section we explicitly confirm that the Fermion Spherical Harmonics are eigenfunctions of $`𝐌_𝐳`$ and $`𝐌^\mathrm{𝟐}`$ with the expected eigenvalues. This was implicit in the derivation of these functions presented in our previous paper .
### 3.1 $`𝐌_𝐳`$ Operator
From the general form of the Spherical Harmonics (7) it follows immediately that $`Y_{\mathrm{}}^m`$ is always an eigenfunction of $`𝐌_𝐳`$ (3) with eigenvalue $`m\mathrm{}`$:
$`M_zY_{\mathrm{}}^m={\displaystyle \frac{\mathrm{}}{i}}{\displaystyle \frac{Y_{\mathrm{}}^m}{\varphi }}={\displaystyle \frac{\mathrm{}}{i}}(\mathrm{sin}\theta )^{|m|}P_{\mathrm{}}^{|m|}(\mathrm{cos}\theta ){\displaystyle \frac{e^{(im\varphi )}}{\varphi }}=m\mathrm{}Y_{\mathrm{}}^m`$ (14)
and hence this eigenfunction property is proven for all cases and requires no further consideration; the eigenvalue, $`m\mathrm{}`$, is the physically expected value for the $`z`$-component of angular momentum both for integral and half-odd-integral values of $`m`$.
### 3.2 $`𝐌^\mathrm{𝟐}`$ Operator
One conclusion from the general form of the Spherical Harmonics (7) and from the form of the $`𝐌^\mathrm{𝟐}`$ operator (4) is that operation on $`Y_{\mathrm{}}^{+|m|}`$ produces the same result as on $`Y_{\mathrm{}}^{|m|}`$; i.e. the same result for the two harmonics that differ only in the sign of $`m`$. This occurs because the only differentiation with respect to $`\varphi `$ in $`𝐌^\mathrm{𝟐}`$ is the second derivative with respect to $`\varphi `$, and in view of the simple exponential dependence of every harmonic on $`\varphi `$, (7), differentiation of this exponential factor twice produces the same result for positive $`m`$ as for negative $`m`$; in the latter case the multiplying negative sign introduced by the first differentiation is canceled by the second differentiation. Thus it is only necessary to consider operation of the $`𝐌^\mathrm{𝟐}`$ operator on $`Y_{\mathrm{}}^{+|m|}`$.
#### 3.2.1 Some Special Cases
In view of the polynomial dependence of $`Y_{\mathrm{}}^m`$ on $`\mathrm{cos}\theta `$, it isn’t possible to infer the general result as we did for $`𝐌_𝐳`$; nevertheless some especially simple cases and the following six examples indicate that $`Y_{\mathrm{}}^m`$ is an eigenfunction of $`𝐌^\mathrm{𝟐}`$ with the physically expected eigenvalue for both integral and half-odd-integral values of $`\mathrm{}`$. These results were obtained manually and were checked using the Maple computer-algebra program.
$`𝐌^\mathrm{𝟐}Y_{\frac{1}{2}}^{\frac{1}{2}}=𝐌^\mathrm{𝟐}e^{(\frac{1}{2}i\varphi )}\mathrm{sin}^{\frac{1}{2}}\theta `$ $`=`$ $`\frac{3}{4}\mathrm{}^2e^{(\frac{1}{2}i\varphi )}(1\mathrm{cos}^2\theta )/\mathrm{sin}^{\frac{3}{2}}\theta `$
$`=`$ $`\frac{3}{4}\mathrm{}^2e^{(\frac{1}{2}i\varphi )}\mathrm{sin}^{\frac{1}{2}}\theta =\frac{1}{2}\left(\frac{1}{2}+1\right)\mathrm{}^2Y_{\frac{1}{2}}^{\frac{1}{2}}`$
$`𝐌^\mathrm{𝟐}Y_{\frac{3}{2}}^{\frac{3}{2}}=𝐌^\mathrm{𝟐}e^{(\frac{3}{2}i\varphi )}\mathrm{sin}^{\frac{3}{2}}\theta `$ $`=`$ $`\frac{15}{4}\mathrm{}^2e^{(\frac{3}{2}i\varphi )}(1\mathrm{cos}^2\theta )/\mathrm{sin}^{\frac{1}{2}}\theta `$
$`=`$ $`\frac{15}{4}\mathrm{}^2e^{(\frac{3}{2}i\varphi )}\mathrm{sin}^{\frac{3}{2}}\theta =\frac{3}{2}\left(\frac{3}{2}+1\right)\mathrm{}^2Y_{\frac{3}{2}}^{\frac{3}{2}}`$
$`𝐌^\mathrm{𝟐}Y_{\frac{3}{2}}^{\frac{1}{2}}=𝐌^\mathrm{𝟐}\mathrm{cos}\theta \mathrm{sin}^{\frac{1}{2}}\theta e^{(\frac{1}{2}i\varphi )}`$ $`=`$ $`\frac{15}{4}\mathrm{}^2\mathrm{cos}\theta e^{(\frac{1}{2}i\varphi )}(1\mathrm{cos}^2\theta )/\mathrm{sin}^{\frac{3}{2}}\theta `$
$`=`$ $`\frac{15}{4}\mathrm{}^2\mathrm{cos}\theta \mathrm{sin}^{\frac{1}{2}}\theta e^{(\frac{1}{2}i\varphi )}=\frac{3}{2}\left(\frac{3}{2}+1\right)\mathrm{}^2Y_{\frac{3}{2}}^{\frac{1}{2}}`$
$`𝐌^\mathrm{𝟐}Y_{\frac{5}{2}}^{\frac{5}{2}}=𝐌^\mathrm{𝟐}\mathrm{sin}^{\frac{5}{2}}\theta e^{(\frac{5}{2}i\varphi )}`$ $`=`$ $`\frac{35}{4}\mathrm{}^2\mathrm{sin}^{\frac{1}{2}}\theta e^{(\frac{5}{2}I\varphi )}(1\mathrm{cos}^2\theta )`$
$`=`$ $`\frac{35}{4}\mathrm{}^2\mathrm{sin}^{\frac{5}{2}}\theta e^{(\frac{5}{2}i\varphi )}=\frac{5}{2}\left(\frac{5}{2}+1\right)\mathrm{}^2Y_{\frac{5}{2}}^{\frac{5}{2}}`$
$`𝐌^\mathrm{𝟐}Y_{\frac{5}{2}}^{\frac{3}{2}}=𝐌^\mathrm{𝟐}\mathrm{cos}\theta \mathrm{sin}^{\frac{3}{2}}\theta e^{(\frac{3}{2}i\varphi )}`$ $`=`$ $`\frac{35}{4}\mathrm{}^2\mathrm{cos}\theta e^{(\frac{3}{2}I\varphi )}(1\mathrm{cos}^2\theta )/\mathrm{sin}^{\frac{1}{2}}\theta `$
$`=`$ $`\frac{35}{4}\mathrm{}^2\mathrm{cos}\theta \mathrm{sin}^{\frac{3}{2}}\theta e^{(\frac{3}{2}i\varphi )}=\frac{5}{2}\left(\frac{5}{2}+1\right)\mathrm{}^2Y_{\frac{5}{2}}^{\frac{3}{2}}`$
$`𝐌^\mathrm{𝟐}Y_{\frac{5}{2}}^{\frac{1}{2}}=𝐌^\mathrm{𝟐}e^{(\frac{1}{2}i\varphi }\mathrm{sin}^{\frac{1}{2}}\theta (14\mathrm{cos}^2\theta )`$ $`=`$ $`\frac{35}{4}\mathrm{}^2e^{(\frac{1}{2}i\varphi )}(15\mathrm{cos}^2\theta +4\mathrm{cos}^4\theta )/\mathrm{sin}^{\frac{3}{2}}\theta `$
$`=`$ $`\frac{35}{4}\mathrm{}^2e^{(\frac{1}{2}i\varphi )}(14\mathrm{cos}^2\theta )(1\mathrm{cos}^2\theta )/\mathrm{sin}^{\frac{3}{2}}\theta `$
$`=`$ $`\frac{35}{4}\mathrm{}^2e^{(\frac{1}{2}i\varphi )}\mathrm{sin}^{\frac{1}{2}}\theta (14\mathrm{cos}^2\theta )=\frac{5}{2}\left(\frac{5}{2}+1\right)\mathrm{}^2Y_{\frac{5}{2}}^{\frac{1}{2}}`$
#### 3.2.2 Some General Cases: $`\mathrm{}=|m|`$, $`\mathrm{}=|m|+1`$, $`\mathrm{}=|m|+2`$ and $`\mathrm{}=|m|+3`$
##### In the case $`\mathrm{}=|m|`$
Table 1 of (the column for $`i=0`$) shows that the polynomial $`P_{\mathrm{}}^{|m|}`$ is simply a constant (equal to 1 in Table 1 of ) and hence the spherical harmonic has the simple form:
$$Y_{\mathrm{}}^m=e^{im\varphi }(\mathrm{sin}\theta )^{|m|}$$
(21)
From this simple form it follows that $`Y_{\mathrm{}}^m`$ is an eigenfunction of $`𝐌^\mathrm{𝟐}`$ for all values of $`\mathrm{}`$:
$`𝐌^\mathrm{𝟐}Y_{\mathrm{}}^m=𝐌^\mathrm{𝟐}e^{(im\varphi )}(\mathrm{sin}\theta )^{|m|}=|m|(|m|+1)\mathrm{}^2Y_{\mathrm{}}^m=\mathrm{}(\mathrm{}+1)\mathrm{}^2Y_{\mathrm{}}^m`$ (22)
##### In the case $`\mathrm{}=|m|+1`$
Table 1 of (the column for $`i`$=$`\mathrm{\hspace{0.17em}1}`$) shows that the polynomial $`P_{\mathrm{}}^{|m|}`$ is simply $`\mathrm{cos}\theta `$ and hence the spherical harmonic has the form:
$$Y_{\mathrm{}}^m=e^{im\varphi }(\mathrm{sin}\theta )^{|m|}\mathrm{cos}\theta $$
(23)
From this simple form Maple deduced that $`Y_{\mathrm{}}^m`$ is an eigenfunction of $`𝐌^\mathrm{𝟐}`$ for all values of $`m`$:
$`𝐌^\mathrm{𝟐}Y_{\mathrm{}}^m=𝐌^\mathrm{𝟐}e^{(im\varphi )}(\mathrm{sin}\theta )^{|m|}\mathrm{cos}\theta =(|m|+1)(|m|+2)\mathrm{}^2Y_{\mathrm{}}^m=\mathrm{}(\mathrm{}+1)\mathrm{}^2Y_{\mathrm{}}^m`$ (24)
##### In the case $`\mathrm{}=|m|+2`$
Table 1 of (the column for $`i`$=$`\mathrm{\hspace{0.17em}2}`$) shows that the polynomial $`P_{\mathrm{}}^{|m|}`$ is $`\left[1(2m+3)\mathrm{cos}^2\theta \right]`$ and hence the spherical harmonic has the form:
$$Y_{\mathrm{}}^m=e^{im\varphi }(\mathrm{sin}\theta )^{|m|}\left[1(2m+3)\mathrm{cos}^2\theta \right]$$
(25)
From this form Maple proved that $`Y_{\mathrm{}}^m`$ is an eigenfunction of $`𝐌^\mathrm{𝟐}`$ for all values of $`m`$:
$`𝐌^\mathrm{𝟐}Y_{\mathrm{}}^m`$ $`=`$ $`𝐌^\mathrm{𝟐}e^{(im\varphi )}(\mathrm{sin}\theta )^{|m|}\left[1(2m+3)\mathrm{cos}^2\theta \right]`$ (26)
$`=`$ $`(|m|+2)(|m|+3)\mathrm{}^2Y_{\mathrm{}}^m=\mathrm{}(\mathrm{}+1)\mathrm{}^2Y_{\mathrm{}}^m`$ (27)
##### In the case $`\mathrm{}=|m|+3`$
Table 1 of (the column for $`i`$=$`\mathrm{\hspace{0.17em}3}`$) shows that the polynomial $`P_{\mathrm{}}^{|m|}`$ is $`\mathrm{cos}\theta \left[3(2m+5)\mathrm{cos}^2\theta \right]`$ and hence the spherical harmonic has the form:
$$Y_{\mathrm{}}^m=e^{im\varphi }(\mathrm{sin}\theta )^{|m|}\mathrm{cos}\theta \left[3(2m+5)\mathrm{cos}^2\theta \right]$$
(28)
From this form Maple proved that $`Y_{\mathrm{}}^m`$ is an eigenfunction of $`𝐌^\mathrm{𝟐}`$ for all values of $`m`$:
$`𝐌^\mathrm{𝟐}Y_{\mathrm{}}^m`$ $`=`$ $`𝐌^\mathrm{𝟐}e^{(im\varphi )}(\mathrm{sin}\theta )^{|m|}\mathrm{cos}\theta \left[3(2m+5)\mathrm{cos}^2\theta \right]`$ (29)
$`=`$ $`(|m|+3)(|m|+4)\mathrm{}^2Y_{\mathrm{}}^m=\mathrm{}(\mathrm{}+1)\mathrm{}^2Y_{\mathrm{}}^m`$ (30)
#### 3.2.3 Summary for $`𝐌^\mathrm{𝟐}`$
The above results prove that $`Y_{\mathrm{}}^m`$ is an eigenfunction of $`𝐌^\mathrm{𝟐}`$ when $`\mathrm{}=|m|`$, $`\mathrm{}=|m|+1`$, $`\mathrm{}=|m|+2`$ and $`\mathrm{}=|m|+3`$, for all values of $`\mathrm{}`$; i.e. for all the Legendre Functions in Table (1 of ) except those in the last two columns, for all values of $`|m|`$ from 0 to $`\mathrm{}`$ including the half-odd-integer values; i.e. for all rows of Table 1 of extended to $`|m|\mathrm{}`$.
We didn’t prove any more results explicitly, but it is a plausible induction from the proven results that $`Y_{\mathrm{}}^m`$ is always an eigenfunction of $`𝐌^\mathrm{𝟐}`$ for all values of $`\mathrm{}`$ and $`m`$ including the functions for which $`\mathrm{}`$ and $`m`$ have half-odd-integer values.
## 4 Ladder Operations
Since the ladder operators, $`𝐌_+`$ and $`𝐌_{}`$, normally transform $`Y_{\mathrm{}}^m`$ into $`Y_{\mathrm{}}^{m+1}`$ and $`Y_{\mathrm{}}^{m1}`$ respectively, it is appropriate to work with the operators defined by (6) divided by $`\mathrm{}`$ in order to avoid multiplying the result of the transformation by a factor of $`\mathrm{}`$ ; these renormalized ladder operators are distinguished by primes, $`𝐌_{}^{}{}_{+}{}^{}`$ and $`𝐌_{}^{}{}_{}{}^{}`$:
$`𝐌_{}^{}{}_{+}{}^{}=𝐌_+/\mathrm{}=e^{(i\varphi )}\left({\displaystyle \frac{}{\theta }}+i\mathrm{cot}\theta {\displaystyle \frac{}{\varphi }}\right)𝐌_{}^{}{}_{}{}^{}=𝐌_{}/\mathrm{}=e^{(i\varphi )}\left({\displaystyle \frac{}{\theta }}i\mathrm{cot}\theta {\displaystyle \frac{}{\varphi }}\right)`$ (31)
### 4.1 $`\varphi `$ Dependence
A general conclusion about operating on a spherical harmonic, $`Y_{\mathrm{}}^m`$, with the ladder operators, $`𝐌_{}^{}{}_{+}{}^{}`$ and $`𝐌_{}^{}{}_{}{}^{}`$, is that the $`\varphi `$ dependence of the result is always the normal, expected result for both integer and half-odd integer harmonics. This conclusion is inferred by observing that the differentiation w.r.t. $`\varphi `$ in $`𝐌_{}^{}{}_{+}{}^{}`$ and $`𝐌_{}^{}{}_{}{}^{}`$ leaves the $`\varphi `$ dependence of $`Y_{\mathrm{}}^m`$ unchanged because the $`\varphi `$ dependence of $`Y_{\mathrm{}}^m`$ is the exponential function $`\mathrm{exp}\{im\varphi \}`$. In addition, $`𝐌_{}^{}{}_{+}{}^{}`$ and $`𝐌_{}^{}{}_{}{}^{}`$ multiply $`Y_{\mathrm{}}^m`$ by $`\mathrm{exp}(i\varphi )`$ and $`\mathrm{exp}(i\varphi )`$ respectively; this multiplication has the effect of increasing and decreasing the exponent by 1 respectively; i.e.:
$$𝐌_{}^{}{}_{+}{}^{}e^{\{im\varphi \}}=m\mathrm{cot}\theta e^{\{i(m+1)\varphi \}}𝐌_{}^{}{}_{}{}^{}e^{\{im\varphi \}}=+m\mathrm{cot}\theta e^{\{i(m1)\varphi \}}$$
(32)
### 4.2 $`\theta `$ Dependence
The differentiation w.r.t. $`\theta `$ in $`𝐌_{}^{}{}_{+}{}^{}`$ and $`𝐌_{}^{}{}_{}{}^{}`$ of $`\mathrm{sin}^{|m|}\theta P_{\mathrm{}}^{|m|}`$ will produce:
$`{\displaystyle \frac{d}{d\theta }}\mathrm{sin}^{|m|}\theta P_{\mathrm{}}^{|m|}=\mathrm{sin}^{|m|}\theta \left\{|m|{\displaystyle \frac{\mathrm{cos}\theta }{\mathrm{sin}\theta }}P_{\mathrm{}}^{|m|}+{\displaystyle \frac{dP_{\mathrm{}}^{|m|}}{d\theta }}\right\}`$ (33)
Recognition of $`P_{\mathrm{}}^{|m|}`$ as a polynomial in $`x=\mathrm{cos}\theta `$ (as shown in Table 1 of ) suggests that some simplification will result from expressing the differentiation of $`P_{\mathrm{}}^{|m|}`$ w.r.t. $`x`$ rather than $`\theta `$, since:
$`{\displaystyle \frac{d}{d\theta }}P_{\mathrm{}}^{|m|}=\mathrm{sin}\theta {\displaystyle \frac{dP_{\mathrm{}}^{|m|}}{dx}}`$ (34)
Thus the general results are:
$`𝐌_{}^{}{}_{+}{}^{}Y_{\mathrm{}}^m`$ $`=`$ $`𝐌_{}^{}{}_{+}{}^{}\left\{e^{(im\varphi )}sin^{|m|}\theta P_{\mathrm{}}^{|m|}\right\}`$ (35)
$`=`$ $`e^{(i[m+1]\varphi )}\mathrm{sin}^{(|m|+1)}\theta \left\{(|m|m){\displaystyle \frac{\mathrm{cos}\theta }{\mathrm{sin}^2\theta }}P_{\mathrm{}}^{|m|}{\displaystyle \frac{dP_{\mathrm{}}^{|m|}}{dx}}\right\}`$
$`=`$ $`e^{(i[m+1]\varphi )}\mathrm{sin}^{(|m|1)}\theta \left\{(|m|m)\mathrm{cos}\theta P_{\mathrm{}}^{|m|}\mathrm{sin}^2\theta {\displaystyle \frac{dP_{\mathrm{}}^{|m|}}{dx}}\right\}`$ (36)
$`𝐌_{}^{}{}_{}{}^{}Y_{\mathrm{}}^m`$ $`=`$ $`𝐌_{}^{}{}_{}{}^{}\left\{e^{(im\varphi )}sin^{|m|}\theta P_{\mathrm{}}^{|m|}\right\}`$ (37)
$`=`$ $`e^{(i[m1]\varphi )}\mathrm{sin}^{(|m|+1)}\theta \left\{(|m|+m){\displaystyle \frac{\mathrm{cos}\theta }{\mathrm{sin}^2\theta }}P_{\mathrm{}}^{|m|}{\displaystyle \frac{dP_{\mathrm{}}^{|m|}}{dx}}\right\}`$
$`=`$ $`e^{(i[m1]\varphi )}\mathrm{sin}^{(|m|1)}\theta \left\{(|m|+m)\mathrm{cos}\theta P_{\mathrm{}}^{|m|}\mathrm{sin}^2\theta {\displaystyle \frac{dP_{\mathrm{}}^{|m|}}{dx}}\right\}`$ (38)
Which of the two alternative, equivalent expressions for $`𝐌_{}^{}{}_{+}{}^{}Y_{\mathrm{}}^m`$ (35,36) and the two for $`𝐌_{}^{}{}_{}{}^{}Y_{\mathrm{}}^m`$ (37,38) produces the simpler result depends upon whether $`m`$ is positive or negative; the term in (35) and in (36) having a factor of $`(|m|m)`$ will be zero when $`m`$ is positive; likewise the term in (37) and in (38) having a factor of $`(|m|+m)`$ will be zero when $`m`$ is negative.
#### 4.2.1 The Case of Positive $`m`$
In the cases where $`𝐌_{}^{}{}_{+}{}^{}`$ and $`𝐌_{}^{}{}_{}{}^{}`$ operate on a $`Y_{\mathrm{}}^m`$ with $`m`$ positive, $`m=|m|`$, and hence the expressions (35) and (38) produce simpler results than (36) and (37) respectively:
$`𝐌_{}^{}{}_{+}{}^{}Y_{\mathrm{}}^m`$ $`=`$ $`e^{(i[m+1]\varphi )}\mathrm{sin}^{(|m|+1)}\theta \left\{{\displaystyle \frac{dP_{\mathrm{}}^{|m|}}{dx}}\right\}`$ (39)
$`𝐌_{}^{}{}_{}{}^{}Y_{\mathrm{}}^m`$ $`=`$ $`e^{(i[m1]\varphi )}\mathrm{sin}^{(|m|1)}\theta \left\{2m\mathrm{cos}\theta P_{\mathrm{}}^{|m|}\mathrm{sin}^2\theta {\displaystyle \frac{dP_{\mathrm{}}^{|m|}}{dx}}\right\}`$ (40)
#### 4.2.2 The Case of Negative $`m`$
In the cases where $`𝐌_{}^{}{}_{+}{}^{}`$ and $`𝐌_{}^{}{}_{}{}^{}`$ operate on a $`Y_{\mathrm{}}^m`$ with $`m`$ negative, $`m=|m|`$, and hence (36) and (37) produce simpler results than (35) and (38) respectively:
$`𝐌_{}^{}{}_{+}{}^{}Y_{\mathrm{}}^m`$ $`=`$ $`e^{(i[m+1]\varphi )}\mathrm{sin}^{(|m|1)}\theta \left\{2\right|m|\mathrm{cos}\theta P_{\mathrm{}}^{|m|}\mathrm{sin}^2\theta {\displaystyle \frac{dP_{\mathrm{}}^{|m|}}{dx}}\}`$ (41)
$`𝐌_{}^{}{}_{}{}^{}Y_{\mathrm{}}^m`$ $`=`$ $`e^{(i[m1]\varphi )}\mathrm{sin}^{(|m|+1)}\theta \left\{{\displaystyle \frac{dP_{\mathrm{}}^{|m|}}{dx}}\right\}`$ (42)
#### 4.2.3 The Case $`\mathrm{}=|m|`$
In this case the polynomial $`P_{\mathrm{}}^{|m|}`$ is simply a constant (the column for $`i=0`$ in Table 1 of ), and since the spherical harmonic has the simple form of (21) we can derive the effect of operating with $`𝐌_{}^{}{}_{+}{}^{}`$ and $`𝐌_{}^{}{}_{}{}^{}`$ for all values of $`m`$; the derivatives $`dP_{\mathrm{}}^{|m|}/dx`$ in (39,40,41,42) are zero, and hence from the fact that (39) and (42) both have this zero derivative as a factor it follows that:
$`𝐌_{}^{}{}_{+}{}^{}Y_{\mathrm{}}^{\mathrm{}}=0\mathrm{and}𝐌_{}^{}{}_{}{}^{}Y_{\mathrm{}}^{\mathrm{}}=0`$ (43)
which are the usual results that the ladder operators produce a zero result (anihilation) when $`𝐌_{}^{}{}_{+}{}^{}`$ operates on $`Y_{\mathrm{}}^{\mathrm{}}`$, and when $`𝐌_{}^{}{}_{}{}^{}`$ operates on $`Y_{\mathrm{}}^{\mathrm{}}`$.
The other two results for $`\mathrm{}=|m|`$ are obtained from (40) and (41) respectively:
$`𝐌_{}^{}{}_{}{}^{}Y_{\mathrm{}}^{\mathrm{}}`$ $`=`$ $`e^{[i(\mathrm{}1)\varphi ]}\mathrm{sin}^{(\mathrm{}1)}\theta \left\{2\mathrm{}\mathrm{cos}\theta \right\}`$
$`=`$ $`2\mathrm{}Y_{\mathrm{}}^{(\mathrm{}1)}\mathrm{if}\mathrm{}1`$
$`𝐌_{}^{}{}_{+}{}^{}Y_{\mathrm{}}^{\mathrm{}}`$ $`=`$ $`e^{[i(\mathrm{}1)\varphi ]}\mathrm{sin}^{(\mathrm{}1)}\theta \left\{2\mathrm{}\mathrm{cos}\theta \right\}`$
$`=`$ $`2\mathrm{}Y_{\mathrm{}}^{(\mathrm{}+1)}\mathrm{if}\mathrm{}1`$
These are the expected results as long as $`\mathrm{}1`$; i.e. the functions in the column for $`i=1`$ in Table 1 of excluding the row for $`|m|=\frac{1}{2}`$.<sup>2</sup><sup>2</sup>2In the case of the first row of Table 1 of (the case $`|m|=0`$) equations (4.2.3) and (4.2.3) produce the expected anihilation results because of the factor of $`\mathrm{}=0`$ on their right hand sides. The exceptional cases are:
$`𝐌_{}^{}{}_{}{}^{}Y_{\frac{1}{2}}^{\frac{1}{2}}=e^{(i\frac{\varphi }{2})}\mathrm{sin}^{\frac{1}{2}}\theta \left\{2\mathrm{}\mathrm{cos}\theta \right\}`$ $`=`$ $`Y_{\frac{1}{2}}^{\frac{1}{2}}\times \mathrm{cot}\theta \mathrm{const}\times Y_{\frac{1}{2}}^{\frac{1}{2}}`$ (46)
$`𝐌_{}^{}{}_{+}{}^{}Y_{\frac{1}{2}}^{\frac{1}{2}}=e^{(i\frac{\varphi }{2})}\mathrm{sin}^{\frac{1}{2}}\theta \left\{2\mathrm{}\mathrm{cos}\theta \right\}`$ $`=`$ $`Y_{\frac{1}{2}}^{\frac{1}{2}}\times \mathrm{cot}\theta \mathrm{const}\times Y_{\frac{1}{2}}^{\frac{1}{2}}`$ (47)
These exceptional results for $`\mathrm{}=|m|=\frac{1}{2}`$ have a $`\theta `$ factor of $`\mathrm{cos}\theta /\sqrt{\mathrm{sin}\theta }`$, whereas in the expected result this factor would be $`\sqrt{\mathrm{sin}\theta }`$.
### 4.3 Cases: $`\mathrm{}=\frac{1}{2}`$, $`\mathrm{}=\frac{3}{2}`$ and $`\mathrm{}=\frac{5}{2}`$
Operation on the 12 half-odd integer functions defined in (8)-(13) with each of $`𝐌_{}^{}{}_{+}{}^{}`$ and $`𝐌_{}^{}{}_{}{}^{}`$ defined in (31) produced the following results; these results were (like the eigenvalue results) obtained manually and then checked using Maple computer algebra.<sup>3</sup><sup>3</sup>3some of the Maple results were simplified manually.
#### 4.3.1 Cases for $`\mathrm{}=\frac{1}{2}`$
The results of applying the ladder operators $`𝐌_{}^{}{}_{+}{}^{}`$ and $`𝐌_{}^{}{}_{}{}^{}`$ \[i.e. (31)\] to the two spherical harmonics for $`\mathrm{}=\frac{1}{2}`$ \[i.e. (8)\] are explicated by equations (43), (46) and (47) above.<sup>4</sup><sup>4</sup>4since in all four of these cases $`\mathrm{}=|m|`$.
##### The first two results (43) for $`\mathrm{}=\frac{\mathrm{𝟏}}{\mathrm{𝟐}}`$ show:
that Merzbacher’s inference \[5, p.241,col.2\] that “the ladder does not terminate” is incorrect; he made this inference by operating with the square of $`𝐌_{}^{}{}_{}{}^{}`$ on $`Y_{\frac{1}{2}}^{\frac{1}{2}}`$:
$`\left(𝐌_{}^{}{}_{}{}^{}\right)^2Y_{\frac{1}{2}}^{\frac{1}{2}}0`$ (48)
which is true because in the first application of $`𝐌_{}^{}{}_{}{}^{}`$ on $`Y_{\frac{1}{2}}^{\frac{1}{2}}`$ the result is the abnormal (46), and the second application of $`𝐌_{}^{}{}_{}{}^{}`$ on this abnormal result is indeed not zero. However, Merzbacher is incorrect in inferring from this result (via the abnormal intermediate result (46)) that “the ladder does not terminate” \[5, p.241,col.2\]; equations (43) demonstrate that the ladder does indeed terminate as it is expected to do. Fortissimo, this general result demonstrates that the ladder terminates at both ends for all values of $`\mathrm{}`$, whereas Merzbacher was only concerned with the case $`\mathrm{}=\frac{1}{2}`$.
#### 4.3.2 Cases for $`\mathrm{}=\frac{3}{2}`$
Application of $`𝐌_{}^{}{}_{+}{}^{}`$ and $`𝐌_{}^{}{}_{}{}^{}`$ \[defined by (31)\] to the two spherical harmonics defined by (9) produces four results:
$`𝐌_{}^{}{}_{+}{}^{}Y_{\frac{3}{2}}^{\frac{3}{2}}=0𝐌_{}^{}{}_{}{}^{}Y_{\frac{3}{2}}^{\frac{3}{2}}=0`$ (49)
which are instances of the general anihilation results (43), and
$`𝐌_{}^{}{}_{}{}^{}Y_{\frac{3}{2}}^{\frac{3}{2}}`$ $`=`$ $`3e^{(\frac{1}{2}i\varphi )}\sqrt{\mathrm{sin}\theta }\mathrm{cos}\theta =3Y_{\frac{3}{2}}^{\frac{1}{2}}`$ (50)
$`𝐌_{}^{}{}_{+}{}^{}Y_{\frac{3}{2}}^{\frac{3}{2}}`$ $`=`$ $`3e^{(\frac{1}{2}i\varphi )}\sqrt{\mathrm{sin}\theta }\mathrm{cos}\theta =3Y_{\frac{3}{2}}^{\frac{1}{2}}`$ (51)
which are instances of the generally expected results (4.2.3) and (4.2.3) for $`\mathrm{}1`$.
Application of $`𝐌_{}^{}{}_{+}{}^{}`$ and $`𝐌_{}^{}{}_{}{}^{}`$ to the two spherical harmonics defined by (10) produces four results:
$`𝐌_{}^{}{}_{+}{}^{}Y_{\frac{3}{2}}^{\frac{1}{2}}`$ $`=`$ $`e^{(3/2i\varphi )}\mathrm{sin}\theta ^{(3/2)}=Y_{\frac{3}{2}}^{\frac{3}{2}}`$ (52)
$`𝐌_{}^{}{}_{}{}^{}Y_{\frac{3}{2}}^{\frac{1}{2}}`$ $`=`$ $`e^{(3/2i\varphi )}\mathrm{sin}\theta ^{(3/2)}=Y_{\frac{3}{2}}^{\frac{3}{2}}`$ (53)
which are instances of the usually expected results, and
$`𝐌_{}^{}{}_{}{}^{}Y_{\frac{3}{2}}^{\frac{1}{2}}`$ $`=`$ $`e^{(\frac{1}{2}i\varphi )}{\displaystyle \frac{(2\mathrm{cos}^2\theta 1)}{\sqrt{\mathrm{sin}\theta }}}=2\mathrm{cot}(2\theta )Y_{\frac{3}{2}}^{\frac{1}{2}}\mathrm{const}\times Y_{\frac{3}{2}}^{\frac{1}{2}}`$ (54)
$`𝐌_{}^{}{}_{+}{}^{}Y_{\frac{3}{2}}^{\frac{1}{2}}`$ $`=`$ $`e^{(\frac{1}{2}i\varphi )}{\displaystyle \frac{(2\mathrm{cos}^2\theta 1)}{\sqrt{\mathrm{sin}\theta }}}=2\mathrm{cot}(2\theta )Y_{\frac{3}{2}}^{\frac{1}{2}}\mathrm{const}\times Y_{\frac{3}{2}}^{\frac{1}{2}}`$ (55)
which are not the usually expected results because the multiplier of the expected function is not a constant; i.e.
$`Y_{\frac{3}{2}}^{\pm \frac{1}{2}}=e^{(\pm \frac{1}{2}i\varphi )}\sqrt{\mathrm{sin}\theta }\mathrm{cos}\theta `$ (56)
from Table 1 of (the entry for $`|m|=\frac{1}{2}`$, $`i=1`$). Results (54) and (55) are distinct from the $`\mathrm{}=|m|`$ results of (4.2.3) and (4.2.3).
#### 4.3.3 Cases for $`\mathrm{}=\frac{5}{2}`$
Application of $`𝐌_{}^{}{}_{+}{}^{}`$ and $`𝐌_{}^{}{}_{}{}^{}`$ to the two spherical harmonics defined by (11) produces four results:
$`𝐌_{}^{}{}_{+}{}^{}Y_{\frac{5}{2}}^{\frac{5}{2}}=0𝐌_{}^{}{}_{}{}^{}Y_{\frac{5}{2}}^{\frac{5}{2}}=0`$ (57)
which are instances of the general anihilation results (43), and
$`𝐌_{}^{}{}_{}{}^{}Y_{\frac{5}{2}}^{\frac{5}{2}}=5e^{(3/2i\varphi )}\mathrm{sin}^{(3/2)}\theta \mathrm{cos}\theta =5Y_{\frac{5}{2}}^{\frac{3}{2}}`$ (58)
$`𝐌_{}^{}{}_{+}{}^{}Y_{\frac{5}{2}}^{\frac{5}{2}}=5e^{(3/2i\varphi )}\mathrm{sin}^{(3/2)}\theta \mathrm{cos}\theta =5Y_{\frac{5}{2}}^{\frac{3}{2}}`$ (59)
which are instances of the general results (4.2.3) and (4.2.3) for $`\mathrm{}1`$.
Application of $`𝐌_{}^{}{}_{+}{}^{}`$ and $`𝐌_{}^{}{}_{}{}^{}`$ to the two spherical harmonics defined by (12) produces four results:
$`𝐌_{}^{}{}_{+}{}^{}Y_{\frac{5}{2}}^{\frac{3}{2}}=e^{(5/2i\varphi )}\mathrm{sin}\theta ^{(5/2)}=Y_{\frac{5}{2}}^{\frac{5}{2}}`$ (60)
$`𝐌_{}^{}{}_{}{}^{}Y_{\frac{5}{2}}^{\frac{3}{2}}=e^{(5/2i\varphi )}\mathrm{sin}\theta ^{(5/2)}=Y_{\frac{5}{2}}^{\frac{5}{2}}`$ (61)
which are instances of the usually expected results, and
$`𝐌_{}^{}{}_{}{}^{}Y_{\frac{5}{2}}^{\frac{3}{2}}=e^{(\frac{1}{2}i\varphi )}\sqrt{\mathrm{sin}\theta }(4\mathrm{cos}\theta ^21)=Y_{\frac{5}{2}}^{\frac{1}{2}}`$ (62)
$`𝐌_{}^{}{}_{+}{}^{}Y_{\frac{5}{2}}^{\frac{3}{2}}=e^{(\frac{1}{2}i\varphi )}\sqrt{\mathrm{sin}\theta }(4\mathrm{cos}\theta ^21)=Y_{\frac{5}{2}}^{\frac{1}{2}}`$ (63)
which are also instances of the usually expected results.
Application of $`𝐌_{}^{}{}_{+}{}^{}`$ and $`𝐌_{}^{}{}_{}{}^{}`$ to the two spherical harmonics defined by (13) produces four results:
$`𝐌_{}^{}{}_{+}{}^{}Y_{\frac{5}{2}}^{\frac{1}{2}}=8\mathrm{sin}^{(3/2)}\theta \mathrm{cos}\theta e^{(3/2i\varphi )}=8Y_{\frac{5}{2}}^{\frac{3}{2}}`$ (64)
$`𝐌_{}^{}{}_{}{}^{}Y_{\frac{5}{2}}^{\frac{1}{2}}=8\mathrm{sin}\theta ^{(3/2)}\mathrm{cos}\theta e^{(3/2i\varphi )}=8Y_{\frac{5}{2}}^{\frac{3}{2}}`$ (65)
which are also instances of the usually expected results, and
$`𝐌_{}^{}{}_{}{}^{}Y_{\frac{5}{2}}^{\frac{1}{2}}`$ $`=`$ $`3e^{(\frac{1}{2}i\varphi )}{\displaystyle \frac{\mathrm{cos}\theta (4\mathrm{cos}^2\theta 3)}{\sqrt{\mathrm{sin}\theta }}}`$ (66)
$`=`$ $`3e^{(\frac{1}{2}i\varphi )}\sqrt{\mathrm{sin}\theta }{\displaystyle \frac{\mathrm{cos}(3\theta )}{\mathrm{sin}\theta }}=3\mathrm{cot}(3\theta )Y_{\frac{5}{2}}^{\frac{1}{2}}\mathrm{const}\times Y_{\frac{5}{2}}^{\frac{1}{2}}`$
$`𝐌_{}^{}{}_{+}{}^{}Y_{\frac{5}{2}}^{\frac{1}{2}}`$ $`=`$ $`3e^{(\frac{1}{2}i\varphi )}{\displaystyle \frac{\mathrm{cos}\theta (4\mathrm{cos}^2\theta 3)}{\sqrt{\mathrm{sin}\theta }}}`$ (67)
$`=`$ $`3e^{(\frac{1}{2}i\varphi )}\sqrt{\mathrm{sin}\theta }{\displaystyle \frac{\mathrm{cos}(3\theta )}{\mathrm{sin}\theta }}=3\mathrm{cot}(3\theta )Y_{\frac{5}{2}}^{\frac{1}{2}}\mathrm{const}\times Y_{\frac{5}{2}}^{\frac{1}{2}}`$
since:<sup>5</sup><sup>5</sup>5from Table 1 of (the entry for $`|m|=\frac{1}{2}`$, $`i=2`$)
$`Y_{\frac{5}{2}}^{\pm \frac{1}{2}}=e^{(\pm \frac{1}{2}i\varphi )}\sqrt{\mathrm{sin}\theta }(14\mathrm{cos}^2\theta )`$ (68)
These results, (66) and (67), are not the usually expected results because the multiplier of the expected function is not a constant.
### 4.4 Overall Conclusion
##### The ladder operators produce the usually expected result for all values of $`\mathrm{}`$
except in two cases:
$`𝐌_{}^{}{}_{+}{}^{}Y_{\mathrm{}}^{\frac{1}{2}}\mathrm{const}\times Y_{\mathrm{}}^{\frac{1}{2}}`$ (69)
$`𝐌_{}^{}{}_{}{}^{}Y_{\mathrm{}}^{\frac{1}{2}}\mathrm{const}\times Y_{\mathrm{}}^{\frac{1}{2}}`$ (70)
From the computed instances of these unusual results (46,47,54,55,66,67) the general non-constant multiplier appears to be:
$`𝐌_{}^{}{}_{+}{}^{}Y_{\mathrm{}}^{\frac{1}{2}}`$ $`=`$ $`(\mathrm{}+\frac{1}{2})\mathrm{cot}[(\mathrm{}+\frac{1}{2})\theta ]Y_{\mathrm{}}^{\frac{1}{2}}`$ (71)
$`𝐌_{}^{}{}_{}{}^{}Y_{\mathrm{}}^{\frac{1}{2}}`$ $`=`$ $`(\mathrm{}+\frac{1}{2})\mathrm{cot}[(\mathrm{}+\frac{1}{2})\theta ]Y_{\mathrm{}}^{\frac{1}{2}}`$ (72)
for all values of $`\mathrm{}`$, which are, of course, half-odd-integer values.
These failures in the Schrödinger represention are not present in the abstract theory of angular momentum based upon the fundamental commutation relations \[2, §5.4,pp.119-120\].
The algebraic origin of the failures can be understood by reference to the general results (40) and (41). For the simplest case of $`\mathrm{}=|m|`$; $`P_{\mathrm{}}^{|m|}`$ is a constant, which makes $`dP_{\mathrm{}}^{|m|}/d\theta =0`$ in (40) and (41), but $`|m|`$ multiplies the $`\mathrm{cos}\theta `$ in these formulae:
* For integer $`\mathrm{}=m=1`$ the $`\mathrm{cos}\theta `$ is exactly the expected, usual result, $`Y_1^0`$, but
* For $`\mathrm{}=|m|=\frac{1}{2}`$ the $`\mathrm{cos}\theta `$ multiplies the expected result, $`Y_{\frac{1}{2}}^{\pm \frac{1}{2}}`$ divided by $`\mathrm{sin}\theta `$
which is the algebraic reason for the failure.
## 5 Double-Valuedness
A distinctive feature of the Fermion Spherical Harmonics is that they are double-valued functions of $`\varphi `$; this arises because:
$$\mathrm{exp}[i\frac{n}{2}(\varphi +2\pi )]=\mathrm{exp}[i\frac{n}{2}\varphi ]$$
(73)
with $`n`$ an odd integer. The angle $`\varphi `$ must transit two complete circles for the wavefunction to return to its original value:
$$\mathrm{exp}[i\frac{n}{2}(\varphi +4\pi )]=+\mathrm{exp}[i\frac{n}{2}\varphi ]$$
(74)
This property of Fermion wavefunctions is well known . The double-valuedness of the wavefunction nevertheless leaves the probability single-valued, because the probability is computed as the product of the $`\varphi `$ factor on the R.H.S. of (74) with its own complex conjugate, and thus the imaginery exponent produces a probability that is independent of the angle $`\varphi `$ for all values of the exponent; i.e. of $`n`$ in (73) and (74).
## 6 The Validity of the Fermion Harmonics
The existence of Spherical Harmonics having half-odd integer quantum numbers, $`\mathrm{}`$ and $`m`$, has been known to some scientists for many years ; these and other authors have been concerned with finding theoretical reasons for only using the spherical harmonics having integer values of $`\mathrm{}`$ and $`m`$. A related question is why a given physical system has angular momentum states that all have integer values of $`\mathrm{}`$ and $`m`$, while other systems have states that all have half-odd-integer values of $`\mathrm{}`$ and $`m`$; a simple proof that a particle cannot have both integer and half-odd-integer values of $`\mathrm{}`$ and $`m`$, has been given by Bohm \[8, p.389\]. However, Merzbacher has shown that the orbital angular momentum of a rigid body may be quantized in half-odd-integer values of $`\mathrm{}`$ and $`m`$.
Merzbacher \[4, p.174\] taking the single-valuedness of the wavefunction as axiomatic advanced the argument that the eigenfunction<sup>6</sup><sup>6</sup>6of $`𝐌^\mathrm{𝟐}`$ and $`𝐌_𝐳`$ $`\sqrt{sin\theta }\mathrm{exp}(i\varphi /2)`$, can be made single-valued by limiting the range of $`\varphi `$ to 0 to $`2\pi `$; he then notes \[using (73)\] that with this restriction the $`\varphi `$ factor is discontinuous at $`\varphi =2\pi `$, and is therefore not differentiable at $`\varphi =2\pi `$, which makes it invalid because valid wavefunctions must be differentiable everywhere.
Buchdal questions the validity of Merzbacher’s argument as being too restrictive, and he quotes Bohm’s argument \[8, p.389-390\] that it is only physically observable quantities that must be single valued. This accords with our discussion above \[after (73) and (74)\], that the double valuedness nevertheless leaves the probability singled-valued.
Buchdal quotes Pauli and notes that Pauli’s “not entirely simple” argument has been misrepresented. However, it should be noted that dismissing Merzbacher’s argument and being sceptical about Pauli’s “not entirely simple” argument, are peripheral to Buchdal’s main purpose, which is to advance an alternative argument (to the single-valuedness of the wavefunction) to infer that the orbital angular momentum quantum numbers must be integers. Thus Buchdal only infers that the half-integral spherical harmonics are invalid representations of orbital angular momentum.
Likewise, the note published by Gray is only concerned with why half-integral angular momenta are invalid representations of the orbital angular momentum of a particle. His argument is based upon that of Bohm ; that a system cannot have both integer and half-integer quantum numbers, and since $`\mathrm{}=m=0`$ is known to be physically valid for orbital angular momentum (the non-rotating state), and since the ladder operations (relating different eigenfunctions having the same $`\mathrm{}`$ but different $`m`$) step through the eigenfunctions in integer steps of $`m`$, he concludes that all the values (including zero) of $`m`$ (and therefore of $`\mathrm{}`$) must be integers; it is important to realize that Gray proposes this argument as an alternative to Merzbacher’s (single-valuedness of the wavefunction) argument for the orbital angular momentum of a particle; Gray is not concerned with spin angular momentum.
Blatt and Weisskopf \[11, p.783\] quote the unpublished argument of Nordsieck that dismisses single-valuedness as unnecessary because only probability densities and expectation values must be singled-valued; his argument is reinforced by noting that double-valued wave functions are used in the theory of particles with spin.
## 7 Summary and Conclusions
Spectroscopic results indicate that orbital angular momentum eigenstates always correspond to integer values of $`\mathrm{}`$, whereas the intrinsic angular momentum (spin) eigenstates of elementary particles (notably the electron), correspond to half-integer values of $`\mathrm{}`$ and $`m`$.
This difference is rooted in the diference between the orbital rotation of a body (which is classical apart from quantization of the angular momentum in integer multiples of $`\mathrm{}`$), and the intrinsic rotation of an elementary particle whose nature is not really understood. Attempts to construct coherent models of the electron have not yet yielded a physical understanding of the nature of spin . Dahl’s paper “The Spinning Electron” promises (in its Abstract and Introduction) to present a physical model as a “3-dimensional rotor governed by relativistic quantum mechanics”, but the body of the paper leaves the reader still groping for a tangible physical model of what an electron is, and in particular what gives rise to its spin and associated magnetic moment.
Including the Fermion Spherical Harmonics (for $`\mathrm{}=\frac{1}{2}`$) as part of a coherent model of the electron implies adopting an interpretation of the angles $`\theta `$ and $`\varphi `$; the obvious interpretation is that they represent the orientation in space of the particle’s intrinsic angular momentum and its magnetic moment vector. This explicit description of the orientation is, of course, limited by the uncertainty principle; i.e. while the $`z`$-component is defined by the value of $`m`$ (to be $`m\mathrm{}`$), the total angular momentum is always larger ($`\mathrm{}\sqrt{\mathrm{}(\mathrm{}+1)}`$), and hence the angular momentum has a component in the $`x`$-$`y`$ plane whose direction in space remains undetermined. The explicit angular description may allow for a more precise theoretical description of physical systems such as the Stern-Gerlach experiment \[15, p.141-148\].
The failure of the ladder operations in the cases where $`m`$ changes sign, is not, in itself, a convincing argument that the Fermion Spherical Harmonics are not valid as eigenfunctions of spin (and total) angular momentum. They are (as we have explicitly shown here) eigenfunctions of $`M_z`$ and $`M^2`$ with the expected eigenvalues in all cases, and they are normalizable as shown in our previous paper \[1, Table 2\]. The eigenfunction property is essential for a valid quantum mechanical state, whereas ladder operators relating states with different eigenvalues are only known for a few physical systems;<sup>7</sup><sup>7</sup>7well-known for angular momentum and the one-dimensional harmonic oscillator; also known for the hydrogen atom and for the electronic motion in the $`H_2^+`$ ion. that this coordinate representation of spin angular momentum differs from the abstract theory of angular momentum in this respect, is an interesting curiosity worthy of further investigation.
That the double-valued factors of the Fermion Spherical Harmonics ($`\mathrm{exp}\{in\varphi /2\}`$) occur in the accepted Dirac and Pauli theories of spin supports their validity; however, in these established theories the wavefunction has 4 and 2 components respectively, which seems to reduce the validity question to whether a scalar wavefunction (with a Fermion Spherical Harmonic factor) can be an acceptable representation of spin angular momentum. |
warning/0507/hep-th0507031.html | ar5iv | text | # Oscillator quantization of the massive scalar particle dynamics on AdS spacetime
## 1 Introduction
$`AdS`$ spacetimes have an isometry group of the same dimension as the corresponding Minkowski case. Therefore, quantization of particle dynamics within standard Hamiltonian reduction techniques is possible. This has been done before for lower dimensional cases . Our present paper continues a systematic study of both the classical and quantum particle dynamics on $`AdS_{N+1}`$ for generic $`N`$ started in . With these investigations we find new explicit realizations of UIR’s of $`SO(2,N)`$.
Another motivation for this program comes from possible interrelations with the corresponding quantum field theories and string theories on $`AdS_{N+1}`$, which play a crucial role in the $`AdS`$/CFT correspondence, see e.g. . In this context one of the most challenging open problems is the quantization of strings on $`AdS_5\times S^5`$. Certainly, the full understanding of quantized particle dynamics on such backgrounds could be a useful warm-up.
There are well-known explicit expressions for field theoretical propagators on $`AdS`$ spacetime. They are crucial tools for the use of the $`AdS`$/CFT correspondence in application to the large $`N`$ and strong ’t Hooft coupling limit on the gauge field theory side. Less is known about explicit expressions for field theoretical propagators on $`AdS\times S`$ backgrounds . Since they could be very helpful to understand the fate of the holographic picture in the BMN limit , any technique for constructing propagators should be of interest. In we have explicitely shown that at least some of the standard $`AdS`$ propagators can be obtained as propagation kernels of the quantized particle.
Of particular interest is also the case of massless particles , which in field theory have been related to singleton representations of the conformal group. For them it is natural to relate the original spacetime to the conformal boundary of a spacetime with just one more dimension.
In , quantizing the dynamics of a massive scalar particle in $`AdS_{N+1}`$ space-time along the lines of geometric quantization, we have constructed the following representations of the isometry group generating $`so(2,N)`$ algebra
$`E=\alpha +\zeta _l_{\zeta _l},J_{mn}=i(\zeta _n_{\zeta _m}\zeta _m_{\zeta _n}),`$ (1.1)
$`z_n^{}=2\alpha \zeta _n+(2\zeta _n\zeta _l\zeta ^2\delta _{nl})_{\zeta _l},z_n=_{\zeta _n},(l,m,n=1,\mathrm{},N).`$
Here $`E`$ and $`J_{mn}`$ are the generators of the compact subgroup $`SO(2)\times SO(N)`$, while $`z_n^{}`$ and $`z_n`$ stand for the two kind of boosts of the theory. The representations (1.1) are characterized by the parameter $`\alpha `$, which corresponds to the lowest eigenvalue of the energy operator $`E`$. The Hilbert space is spanned by the holomorphic functions of $`N`$ complex variables $`\zeta _n`$ $`(n=1,\mathrm{},N)`$ defined inside some bounded domain and the measure for the scalar product depends on the parameter $`\alpha `$ and the dimension $`N`$. The case $`N=1`$ corresponds to Bargmann’s representation of the $`so(2,1)`$ algebra given on holomorphic functions inside the unit disk $`|\zeta |<1`$ (see e.g. ), and the regularity of the scalar product requires $`\alpha >1/2`$. For $`N>1`$ the integration domain is more complicated and the integration measure turns out to be regular for $`\alpha >N1`$ only.
Since the constructed Hilbert space seems to have some similarities to the Fock space of a $`N`$-dimensional oscillator, it is natural to ask for a corresponding explicit construction in terms of oscillator variables and its canonical quantization. From there remained also another puzzling question. The allowed values of $`\alpha `$ do not exhaust the well-known unitarity bound, see e.g. and refs. therein,
$$\alpha (N2)/2\text{for}N2\text{ and}\alpha 0\text{for}N=1.$$
(1.2)
In addition, within the representations (1.1) one cannot reach values for $`\alpha `$, corresponding to the Casimir number $`C=(1N^2)/4`$ related to a conformal invariant setting in $`AdS_{N+1}`$.
The formulation in terms of oscillator variables for the simplest case $`N=1`$ can be obtained by the Holstein-Primakoff method in the form
$$z=\sqrt{H+2\alpha }a,z^{}=a^{}\sqrt{H+2\alpha },E=H+\alpha ,$$
(1.3)
where $`H`$ is the normal ordered oscillator Hamiltonian $`H=a^{}a`$ and the square root of the positive operator $`H+2\alpha `$ is defined in standard manner via spectral representation. The representation (1.3) is unitary and irreducible for $`\alpha >0`$. Thus, the canonical scheme covers all lowest weight UIR’s of the $`so(2,1)`$ algebra. The aim of the present paper is to generalize this result to arbitrary $`N`$.
## 2 Classical theory
The $`N+1`$ dimensional $`AdS`$ space can be represented as the universal covering of the hyperboloid of radius $`R`$
$$X_0^2+X_0^{}^2\underset{n=1}{\overset{N}{}}X_n^2=R^2$$
(2.1)
embedded in the $`N+2`$ dimensional flat space $`_N^2`$ with coordinates $`X_A,`$ $`A=(0,0^{},1,\mathrm{},N)`$ and the metric tensor $`G_{AB}=\text{diag}(+,+,,\mathrm{},)`$. The induced metric tensor on the hyperboloid has Lorentzian signature and the polar angle $`\theta `$ in the $`(X_0,X_0^{})`$ plane can be taken as the time coordinate: $`X_0=r\mathrm{cos}\theta `$, $`X_0^{}=r\mathrm{sin}\theta `$ ($`r^2=X_nX_n+R^2`$).
The dynamics of a massive particle moving on the hyperboloid we describe by the action
$$S=𝑑\tau \left[m\sqrt{\dot{X}^A\dot{X}_A}+\frac{\mu }{2}(X^AX_AR^2)\right],$$
(2.2)
where $`m>0`$ is the particle mass, $`\mu `$ is a Lagrange multiplier and $`\tau `$ is an evolution parameter. The time direction we fix in a $`SO(2,N)`$-invariant way by requiring $`\dot{\theta }>0`$, which is equivalent to $`X_0\dot{X}_0^{}X_0^{}\dot{X}_0>0.`$ The space-time isometry group $`SO(2,N)`$ provides the Noether conserved quantities
$$J_{AB}=P_AX_BP_BX_A,$$
(2.3)
where $`P_A`$ are the canonical momenta. Since $`\theta `$ is the time coordinate, $`J_{00^{}}`$ is associated with the particle energy. We denote it by $`E`$ and due to the choice of the time direction it is positive
$$E=P_0X_0^{}P_0^{}X_0=\frac{m}{\sqrt{\dot{X}^A\dot{X}_A}}(X_0\dot{X}_0^{}X_0^{}\dot{X}_0)>0.$$
(2.4)
The conservation of $`J_{AB}`$ allows to represent the set of trajectories geometrically without solving the dynamical equations. From Eq. (2.3) we find N equations as identities in the variables $`(P,X)`$
$$EX_n=J_{0n}X_0^{}J_{0^{}n}X_0,(n=1,\mathrm{},N).$$
(2.5)
Since $`E`$, $`J_{0^{}n}`$, $`J_{0^{}n}`$ are constants, Eq. (2.5) defines a 2-dimensional plane in the embedding space $`_N^2`$ and the intersection of this plane with the hyperbola (2.1) is a particle trajectory.
The action (2.2) is invariant under the reparametrizations $`\tau f(\tau )`$ with the corresponding transformations of the Lagrange multiplier ($`\mu \mu /f^{}`$). The gauge symmetry, as usual, leads to the dynamical constraints. Applying Dirac’s procedure, we find three constraints
$$X^AX_AR^2=0,P_AP^Am^2=0,P_AX^A=0,$$
(2.6)
which fix the quadratic Casimir number of the symmetry group
$$C=\frac{1}{2}J_{AB}J^{AB}=m^2R^2.$$
(2.7)
For further calculations it is convenient to introduce complex valued dynamical integrals
$$z_n=J_{0^{}n}iJ_{0n},z_n^{}=J_{0^{}n}+iJ_{0n},$$
(2.8)
and the scalar variables $`\lambda ^2=z_n^{}z_n,`$ $`\rho ^4=z_{}^{}{}_{}{}^{2}z^2`$ (with $`z^2=z_nz_n`$). Then (2.7) becomes
$$E^2+J^2=\lambda ^2+\alpha ^2,$$
(2.9)
where
$$J^2=\frac{1}{2}J_{mn}J_{mn}\text{and}\alpha =mR.$$
(2.10)
A set of other quadratic relations follows from (2.3) as identities in the variables $`(P,X)`$
$$J_{AB}J_{A^{}B^{}}=J_{AA^{}}J_{BB^{}}J_{AB^{}}J_{BA^{}}.$$
(2.11)
These equations are nontrivial in terms of the dynamical integrals, if all indices $`A`$, $`B`$, $`A^{}`$, $`B^{}`$ are different. Taking $`A=0`$, $`B=0^{}`$, $`A^{}=m`$ and $`B^{}=n`$ $`(mn)`$ we obtain
$$2iEJ_{mn}=z_m^{}z_nz_n^{}z_m,$$
(2.12)
and its square yields $`4E^2J^2=\lambda ^4\rho ^4`$. Then, together with (2.9) we conclude that $`E^2`$ and $`J^2`$ are roots of the quadratic equation $`\mathrm{\hspace{0.17em}4}x^24(\lambda ^2+\alpha ^2)x+\lambda ^4\rho ^4=0`$ and find
$`E^2={\displaystyle \frac{1}{2}}\left(\lambda ^2+\alpha ^2+\sqrt{\alpha ^4+2\alpha ^2\lambda ^2+\rho ^4}\right).`$ (2.13)
We neglect the second root as unphysical, since it does not provide positivity of the energy (see for more details). According to (2.13) $`\alpha `$ is the lowest value of energy. Eqs. (2.13) and (2.12) define $`E`$ and $`J_{mn}`$ as functions of ($`z_n,z_n^{}`$) and, therefore, ($`z_n,z_n^{}`$) are global coordinates on the space of dynamical integrals, which is the physical phase space $`\mathrm{\Gamma }_{ph}`$ of the system.
Based on the canonical brackets between the variables $`(P,X)`$ the $`so(2,N)`$ Poisson bracket algebra of the symmetry generators can be written as
$`\{z_m^{},z_n\}=2J_{mn}2i\delta _{mn}E,\{z_m,z_n\}=0=\{z_m^{},z_n^{}\},`$ (2.14)
$`\{J_{lm},z_n\}=z_l\delta _{mn}z_m\delta _{ln},\{E,z_n\}=iz_n,\{E,J_{mn}\}=0,`$ (2.15)
and $`J_{mn}`$’s form the $`so(N)`$ algebra. These relations are preserved after reduction to $`\mathrm{\Gamma }_{ph}`$, since the constraints (2.6) are $`SO(2,N)`$ scalars. But (2.14)-(2.15) become essentially non-linear in terms of the independent variables $`z_n`$, $`z_n^{}`$ and their quantum realization is a nontrivial issue.
Our aim is to find a canonical parametrization of $`\mathrm{\Gamma }_{ph}`$ and then to quantize the system canonically. For this purpose we introduce a set of $`2N`$ variables ($`a_n`$, $`a_n^{};n=1,\mathrm{},N`$) with the canonical Poisson brackets
$$\{a_n,a_m\}=0=\{a_n^{},a_m^{}\},\{a_n,a_m^{}\}=i\delta _{mn},$$
(2.16)
and assume that the generators of the compact transformations are given by
$`E=H+\alpha ,J_{mn}=i(a_n^{}a_ma_m^{}a_n),`$ (2.17)
where $`H=a_n^{}a_n`$ is the oscillator Hamiltonian. Note that Eqs. (2.16)-(2.17) immediately provide the $`so(2)\times so(N)`$ part of $`so(2,N)`$ algebra and the $`so(N)`$ scalar (2.10) constructed from $`J_{mn}`$ becomes $`J^2=H^2a_{}^{}{}_{}{}^{2}a^2`$, with $`a^2=a_na_n`$ and $`a_{}^{}{}_{}{}^{2}=a_n^{}a_n^{}`$.
We look now for a parametrization of the generators $`z_n`$ and $`z_n^{}`$ in the form
$$z_n=X(H,J^2)a_n+Y(H,J^2)a^2a_n^{},z_n^{}=X(H,J^2)a_n^{}+Y(H,J^2)a_{}^{}{}_{}{}^{2}a_n,$$
(2.18)
where $`X`$ and $`Y`$ are real functions of the scalar variables ($`H,J^2)`$. This ansatz guarantees the correct commutation relations between compact and non-compact generators. The Casimir condition (2.9) and the quadratic relation (2.12) yield two equations for the functions $`X`$, $`Y`$
$`HX^2+(H^2J^2)(2XY+HY^2)=H^2+J^2+2\alpha H,`$ (2.19)
$`X^2(H^2J^2)Y^2=2(H+\alpha ).`$
After elimination of $`X`$, (2.19) reduces to a quadratic equation for $`Y^2`$. Choosing the root which is nonsingular for $`J^2=0`$ and putting $`Y=\sqrt{Y^2}`$, we obtain
$`X={\displaystyle \frac{1}{2Y}}[1+2HY^2],Y=\left(2H+4\alpha +2\sqrt{(H+2\alpha )^2J^2}\right)^{\frac{1}{2}}.`$ (2.20)
Note that the choice $`Y=\sqrt{Y^2}`$ leads to a canonically equivalent answer, reproduced from (2.18) by the inversion $`a_na_n`$.
Due to (2.18) and (2.20) the $`SO(N)`$ scalars $`z^2`$ and $`a^2`$ are related by
$$z^2=Fa^2,$$
(2.21)
where $`F`$ is the following function of real scalar variables
$$F=\sqrt{(H+2\alpha )^2J^2}=\sqrt{(E+\alpha )^2J^2}.$$
(2.22)
This function also played an important role in the scheme of geometric quantization . Note that the parametrization of $`\mathrm{\Gamma }_{ph}`$ given by (2.18) and (2.20) can be written as
$`z_n={\displaystyle \frac{1}{\sqrt{2(H+2\alpha +F)}}}\left((2H+2\alpha +F)a_na^2a_n^{}\right).`$ (2.23)
The relation (2.21) helps to invert (2.18) and we find
$$a_n=\frac{1}{2E}\left(Xz_n\frac{Y}{F}z^2z_n^{}\right),a_n^{}=\frac{1}{2E}\left(Xz_n^{}\frac{Y}{F}z_{}^{}{}_{}{}^{2}z_n\right),$$
(2.24)
where now $`X`$, $`Y`$ and $`F`$ are treated as functions of $`(E,J^2)`$ and they are obtained from (2.20) replacing $`H`$ by $`E\alpha `$.
The direct calculation shows that the canonical brackets (2.16) lead to the $`so(2,N)`$ algebra (2.14)-(2.15). To our knowledge, the constructed canonical parametrization and its quantum realization for generic $`N`$ was not discussed in the literature before.
## 3 Quantum theory
We quantize the $`AdS`$ particle dynamics using the canonical coordinates constructed in the previous section. We will use for quantum operators the same letters as for the corresponding classical variables. Canonical quantization assumes a realization of the canonical commutation relations
$$[a_n,a_m]=0=[a_n^{},a_m^{}],[a_n,a_m^{}]=\delta _{mn},$$
(3.1)
and a construction of the symmetry generators $`E`$, $`J_{mn}`$, $`z_n`$, $`z_n^{}`$ on the basis of the parametrization (2.17), (2.18) and (2.20). Due to the non-linearity of the parametrization we face with the problem of ordering ambiguity.
There is not such a problem for the generators of $`SO(N)`$ rotations
$$J_{mn}=i(a_n^{}a_ma_m^{}a_n),$$
(3.2)
and the operator $`J^2`$ in terms of the creation-annihilation operators becomes
$$J^2=\frac{1}{2}J_{mn}J_{mn}=H^2+(N2)Ha^{}{}_{}{}^{2}a_{}^{2},$$
(3.3)
where $`H=a_n^{}a_n`$ is the normal ordered oscillator Hamiltonian, $`a^2=a_na_n`$ and $`a^{}{}_{}{}^{2}=a_n^{}a_n^{}`$.
Classically we had for the quadratic Casimir $`C=m^2R^2=\alpha ^2`$. In the quantum case we continue to denote the lowest energy value by $`\alpha `$, i.e.
$$E=H+\alpha ,$$
(3.4)
but we have to expect a renormalization of the relation of $`\alpha `$ to the quadratic Casimir, which defines the squared mass of the quantum particle. Since we are interested in representations which are unitary equivalent to those constructed by geometric quantization we take the corresponding renormalized relation from
$$C=E^2+\frac{1}{2}J_{mn}J_{mn}\frac{1}{2}(z_n^{}z_n+z_nz_n^{})=\alpha (\alpha N).$$
(3.5)
Note in addition, that the quadratic relation (2.12) due to (1.1) is deformed into
$`z_m^{}z_nz_n^{}z_m=2i(E1)J_{mn},\text{or}z_nz_m^{}z_mz_n^{}=2i(E+1)J_{mn}.`$ (3.6)
The ordering problem is more complicated for $`z_n`$ and $`z_n^{}`$. We look for them in the form
$$z_n=X_N(H,J^2)a_n+Y_N(H,J^2)a^2a_n^{},z_n^{}=a_n^{}X_N(H,J^2)+a_na_{}^{}{}_{}{}^{2}Y_N(H,J^2),$$
(3.7)
where $`X_N`$ and $`Y_N`$ are real unknown functions of the commuting operators $`H`$ and $`J^2`$, similarly to (2.18). Due to the ordering ambiguity, the functions $`X_N`$, $`Y_N`$ are quantum mechanical deformations of $`X`$, $`Y`$ and the index $`N`$ indicates that the deformations can depend on the dimension $`N`$. To fix these functions we use the quantum versions of the Casimir condition (3.5) and the quadratic relation (3.6), which can be written as
$$z_nz_n^{}=H^2+J^2+(2\alpha +N)H+2N\alpha \text{and}z_nz_m^{}z_mz_n^{}=2(H+\alpha +1)(a_m^{}a_na_n^{}a_m),$$
(3.8)
respectively. Taking into account the commutation relations between the scalar operators
$$[H,a^2]=2a^2,[H,a_{}^{}{}_{}{}^{2}]=2a_{}^{}{}_{}{}^{2},[a^2,a_{}^{}{}_{}{}^{2}]=4H+2N,$$
(3.9)
from (3.8) we find two equations for $`X_N,Y_N`$
$`(H+N)X_N^2+\left(H^2J^2+(N+2)H+2N\right)\left(2X_NY_N+(H+2)Y_N^2\right)=`$ (3.10)
$`H^2+J^2+(2\alpha +N)H+2\alpha N,`$
$`X_N^2\left(H^2J^2+(N+2)H+2N\right)Y_N^2=2(H+\alpha +1).`$
These equations contain only commuting operators and they can be solved as algebraic equations like (2.19) in the classical case. After elimination of $`X_N`$, (3.10) reduces to a quadratic equation for $`Y_N^2`$. Neglecting the root, which corresponds to the singular solution in the classical limit, and choosing $`Y_N=\sqrt{Y_N^2}`$, we obtain (see (2.20))
$`X_N={\displaystyle \frac{1}{2Y_N}}[1+(2H+N+2)Y_N^2],`$ (3.11)
$`Y_N=\left(2H+4\alpha N+2+2\sqrt{(H+2\alpha )^2(N2)(H+2\alpha )J^2}\right)^{\frac{1}{2}}.`$
These operator expressions are naturally defined on the eigenstates of $`H`$ and $`J^2`$ as multiplication operators. The choice $`Y_N=\sqrt{Y_N^2}`$ leads to an unitary equivalent answer, since the change of sign of $`Y_N`$ corresponds to the inversion $`a_na_n`$.
Eqs. (3.7) and (3.11) provide a deformed version of (2.23)
$`z_n={\displaystyle \frac{1}{\sqrt{2(H_N+2\alpha _N+F_N)}}}\left((2H_N+2\alpha _N+F_N)a_na^2a_n^{}\right),\text{with}`$ (3.12)
$`F_N=\sqrt{(H_N+2\alpha _N)^2J_N^2},\alpha _N=\alpha {\displaystyle \frac{N}{2}},`$ (3.13)
$`H_N=H+{\displaystyle \frac{N+2}{2}},J_N^2=J^2+\left({\displaystyle \frac{N2}{2}}\right)^2.`$
The notation in the above formulas has been chosen in a form to make both the structural similarities as well as the modifications relative to their classical counterparts (2.22)-(2.23) manifest. One can check that for $`N=1`$ Eq. (3.12) reproduce the operator $`z`$ in (1.3).
The ansatz (3.7) obviously satisfies the commutation relations of $`z_n`$ with $`E`$ and $`J_{mn}`$ for arbitrary $`X_N`$ and $`Y_N`$
$$[E,z_n]=z_n,[J_{lm},z_n]=i(\delta _{ln}z_m\delta _{mn}z_l).$$
(3.14)
To calculate the commutation relations between the operators (3.7) we use the exchange relations of the creation-annihilation operators with the scalar operators. These relations are derived in the Appendix and by (A.10)-(A.17) we find an alternative form of (3.7)
$$z_n=a_n\stackrel{~}{X}_N+a_n^{}a^2\stackrel{~}{Y}_N,z_n^{}=\stackrel{~}{X}_Na_n^{}+\stackrel{~}{Y}_Na_{}^{}{}_{}{}^{2}a_n,\text{where}$$
(3.15)
$`\stackrel{~}{X}_N={\displaystyle \frac{1}{2\stackrel{~}{Y}_N}}\left(1+(2H+N2)\stackrel{~}{Y}_N^2\right),`$ (3.16)
$`\stackrel{~}{Y}_N=\left(2H+4\alpha N2+2\sqrt{(H+2\alpha )^2(N+2)(H+2\alpha )+2NJ^2}\right)^{\frac{1}{2}}.`$
Note that the functions (3.11) and (3.16) are related by
$`\stackrel{~}{X}_N(H,J^2)=X_N(H2,J^2),\stackrel{~}{Y}_N(H,J^2)=Y_N(H2,J^2).`$ (3.17)
The operators (3.15)-(3.16) satisfy the equations
$$z_n^{}z_n=H^2+J^2+(2\alpha N)H\text{and}z_m^{}z_nz_n^{}z_m=2(H+\alpha 1)(a_m^{}a_na_n^{}a_m),$$
(3.18)
which are equivalent to (3.5) and (3.6), respectively.
It remains to check the commutation relations
$$[z_m,z_n]=0,[z_m^{},z_n]=2(a_n^{}a_ma_m^{}a_n)2\delta _{mn}(H+\alpha ).$$
(3.19)
Using the two representations of $`z_n`$ operators (3.7) and (3.15), we obtain (see Appendix)
$`[z_m,z_n]=(a_ma_n^{}a_na_m^{})a^2U(H,J^2),\text{where}`$ (3.20)
$`U=X_N(H2,J^2)\stackrel{~}{Y}_N(H,J^2)Y_N(H2,J^2)\stackrel{~}{X}_N(H,J^2),`$
and due to (3.17), the commutator $`[z_m,z_n]`$ vanishes.
Calculating $`[z_m^{},z_n]`$ in a similar way we get the following structure (see Appendix)
$`[z_m^{},z_n]=\delta _{mn}U_0+a_m^{}a_nU_1+a_n^{}a_mU_2+a_m^{}a_n^{}a^2U_3+a_ma_na_{}^{}{}_{}{}^{2}U_4,`$ (3.21)
where $`U_0,\mathrm{},U_4`$ are functions of the scalar variables $`H`$, $`J^2`$ like $`U`$ in (3.20). The exchange relations (A.10) and (A.19) provide $`U_3=U_4=0,U_2=U_1=2,U_0=2(H+\alpha )`$ and this completes the proof of (3.19).
The unitarity of our representations is guaranteed as long as $`z_n`$ and $`z_n^{}`$ expressed by (3.7) in terms of $`a_n`$ and $`a_n^{}`$ are adjoint to each other. This in turn implies selfadjoint $`X_N`$ and $`Y_N`$, hence positivity of the operator expression under the square root in (3.11). With the help of (3.3) this condition can be written as $`a^{}{}_{}{}^{2}a_{}^{2}+2(H+\alpha )(2\alpha N+2)0`$ and it reproduces the unitarity bound (1.2).
Irreducibility of the representations is certainly given as long as the creation-annihilation operators $`a_n^{}`$ and $`a_n`$ can be expressed in terms of the symmetry generators. This requires an inversion of (3.7). As a first step for this inversion we need the quantum analog of (2.21), which is obtained from (3.7) and (3.15)
$$z^2=F_N(H,J^2)a^2.$$
(3.22)
Based on this relation we get finally
$$a_n=\frac{1}{2(E+1)}\left(X_Nz_n\frac{Y_N}{F_N}z^2z_n^{}\right),a_n^{}=\frac{1}{2E}\left(z_n^{}X_Nz_nz_{}^{}{}_{}{}^{2}\frac{Y_N}{F_N}\right).$$
(3.23)
This inversion formula is well-defined as long as $`\alpha `$ is above the unitarity bound. But note also that for $`\alpha `$ just on the unitarity bound $`F_N=\sqrt{a_{}^{}{}_{}{}^{2}a^2}`$, which cannot be inverted within the full oscillator space. Therefore irreducibility is lost at the unitarity bound.
## 4 Conclusions
Our main result is the realization of spin zero $`so(2,N)`$ representations in terms of just one set of $`N`$-dimensional oscillator operators. The formulas defining this representation are (3.2), (3.4), (3.7) and (3.11) (or equivalently (3.12) and (3.13)) and they reproduces the well-known unitarity bound (1.2).
The whole construction was based on an one to one map of the space of dynamical integrals of a scalar massive particle in $`AdS_{N+1}`$ to the phase space of a $`N`$-dimensional oscillator. This distinguishes our approach from the oscillator like representations of non-compact groups, developed in , where, using more oscillators, all symmetry generators are represented bilinearly in the oscillators operators and the representation space is selected out from the oscillator Fock space by some conditions, which can be interpreted as coherent state constraints.
Our $`so(2,N)`$ representations arose as a by-product of scalar particle dynamics. It would be interesting to extend these considerations to particles with spin and to construct the corresponding $`so(2,N)`$ representations with non-zero $`so(N)`$ weights.
In a forthcoming paper we will analyze in detail the quantization of massless particles. This will correspond to the special case $`\alpha =(N\pm 1)/2`$ in which the symmetry group is enlarged to the conformal group of $`AdS_{N+1}`$. There we further comment on the reducibility of our representation of $`so(2,N)`$ at the unitarity bound and discuss its relation to the singleton representations and to massless particle dynamics in one dimension less, i.e. in $`AdS_N`$.
Acknowledgments. We thank H. Nicolai for discussions. G.J. is grateful to Humboldt University and AEI Potsdam for hospitality. His research was supported by grants from the DFG, GRDF and GAS. H.D. was supported in part by DFG with the grant DO 447-3/3.
Appendix
Let us consider the operators $`H_N`$ and $`J_N^2`$ given by (3.13), (3.3) and calculate their exchange relations with the operators $`a_n`$ and $`b_n=a^2a_n^{}`$. The canonical commutators (3.1) provide
$`H_Na_n=a_n(H_N1),J_N^2a_n=a_n(J_N^22H_N+1)+2b_n,`$ (A.1)
$`H_Nb_n=b_n(H_N1),J_N^2b_n=b_n(J_N^2+2H_N+1)2a_n(H_N^2J_N^2).`$ (A.2)
These relations are diagonalized by the operators
$$c_n=b_na_n(H_N+J_N),d_n=b_na_n(H_NJ_N),$$
(A.3)
in the following form
$`H_Nc_n=c_n(H_N1),J_N^2c_n=c_n(J_N1)^2,`$ (A.4)
$`H_Nd_n=d_n(H_N1),J_N^2d_n=d_n(J_N+1)^2.`$ (A.5)
Here $`J_N=\sqrt{J_N^2}`$ and the square root from the positive operator is defined in a standard way. Introducing new scalar variables
$$u=H_N+J_N\text{and}v=H_NJ_N,$$
(A.6)
the functions (3.11) can be written as
$$X_N=\frac{u\sqrt{u+\beta }v\sqrt{v+\beta }}{uv},Y_N=\frac{\sqrt{u+\beta }\sqrt{v+\beta }}{uv},\text{with}\beta =2\alpha N.$$
(A.7)
Due to (A.4)-(A.5), a function $`f(u,v)`$ satisfies the exchange relations
$`f(u,v)c_n=c_nf(u2,v),f(u,v)d_n=d_nf(u,v2).`$ (A.8)
Inverting (A.3)
$$a_n=d_n\frac{1}{uv}c_n\frac{1}{uv},b_n=d_n\frac{u}{uv}c_n\frac{v}{uv},$$
(A.9)
and using (A.8), we obtain
$`f(u,v)a_n=a_n{\displaystyle \frac{uf(u2,v)vf(u,v2)}{uv}}b_n{\displaystyle \frac{f(u2,v)f(u,v2)}{uv}},`$ (A.10)
$`f(u,v)b_n=a_n{\displaystyle \frac{uvf(u2,v)uvf(u,v2)}{uv}}b_n{\displaystyle \frac{vf(u2,v)uf(u,v2)}{uv}}.`$
Applying these exchange relations to the functions (A.7) we get
$`X_Na_n=a_nX_{(1)}+b_nY_{(1)},Y_Nb_n=a_nX_{(2)}+b_nY_{(2)},`$ (A.11)
$`X_Nb_n=a_nX_{(3)}+b_nY_{(3)},Y_Na_n=a_nX_{(4)}+b_nY_{(4)},`$ (A.12)
where
$`X_{(1)}={\displaystyle \frac{u(u2)\sqrt{u2+\beta }uv\sqrt{v+\beta }}{(uv)(u2v)}}{\displaystyle \frac{uv\sqrt{u+\beta }v(v2)\sqrt{v2+\beta }}{(uv)(uv+2)}},`$
$`Y_{(1)}={\displaystyle \frac{(u2)\sqrt{u2+\beta }v\sqrt{v+\beta }}{(uv)(u2v)}}+{\displaystyle \frac{u\sqrt{u+\beta }(v2)\sqrt{v2+\beta }}{(uv)(uv+2)}};`$ (A.13)
$`X_{(2)}={\displaystyle \frac{uv\sqrt{u2+\beta }uv\sqrt{v+\beta }}{(uv)(u2v)}}+{\displaystyle \frac{uv\sqrt{u+\beta }uv\sqrt{v2+\beta }}{(uv)(uv+2)}},`$
$`Y_{(2)}={\displaystyle \frac{v\sqrt{u2+\beta }v\sqrt{v+\beta }}{(uv)(u2v)}}{\displaystyle \frac{u\sqrt{u+\beta }u\sqrt{v2+\beta }}{(uv)(uv+2)}}.`$ (A.14)
and the functions $`X_{(3)}`$, $`Y_{(3)}`$, $`X_{(4)}`$, $`Y_{(4)}`$ are expressed in a similar way. Since $`b_n=a_n^{}a^2+2a_n`$, the operator $`z_n=X_Na_n+Y_Nb_n`$ can be rewritten in the form (3.15) with
$`\stackrel{~}{X}_N=X_{(1)}+X_{(2)}+2(Y_{(1)}+Y_{(2)}),\stackrel{~}{Y}_N=Y_{(1)}+Y_{(2)}.`$ (A.15)
Then, Eqs. (A.13)-(A.14) yield
$`\stackrel{~}{X}_N={\displaystyle \frac{(u2)\sqrt{u2+\beta }(v2)\sqrt{v2+\beta }}{uv}},`$ (A.16)
$`\stackrel{~}{Y}_N={\displaystyle \frac{\sqrt{u2+\beta }\sqrt{v2+\beta }}{uv}}.`$ (A.17)
and passing back from $`(u,v)`$ to $`(H,J)`$, we arrive at (3.16).
The commutator $`[z_m,z_n]`$ is obtained by anti-symmetrization of
$`z_mz_n=X_Na_ma_n\stackrel{~}{X}_N+X_Na_ma_n^{}a^2\stackrel{~}{Y}_N+Y_Na^2a_m^{}a_n\stackrel{~}{X}_N+Y_Na^2a_m^{}a_n^{}a^2\stackrel{~}{Y}_N,`$ (A.18)
where the representations (3.7) and (3.15) are used for $`z_m`$ and $`z_n`$ respectively. Taking into account that the operator $`J_{mn}=i(a_n^{}a_ma_m^{}a_n)`$ commutes with scalar operators and that $`X_N(H,J^2)a^2=a^2X_N(H2,J^2)`$, the commutator $`[z_m,z_n]`$ reduces to (3.20).
To represent the commutator $`[z_m^{},z_n]`$ in the form (3.21) we also use the exchange relations of $`a_n^{}`$ and $`b_n^{}`$ with scalar operators
$`f(u,v)a_n^{}=a_n^{}{\displaystyle \frac{uf(u+2,v)vf(u,v+2)}{uv}}b_n^{}{\displaystyle \frac{f(u+2,v)f(u,v+2)}{uv}},`$ (A.19)
$`f(u,v)b_n^{}=a_n^{}{\displaystyle \frac{uvf(u+2,v)uvf(u,v+2)}{uv}}b_n^{}{\displaystyle \frac{vf(u+2,v)uf(u,v+2)}{uv}},`$
which can be derived similarly to (A.10). Writing the terms $`z_m^{}z_n`$ and $`z_nz_m^{}`$ as
$`z_m^{}z_n=(a_m^{}X_N+b_m^{}Y_N)(a_n(\stackrel{~}{X}_N2\stackrel{~}{Y}_N)+b_n\stackrel{~}{Y}_N),\text{and}`$ (A.20)
$`z_nz_m^{}=(a_n\stackrel{~}{X}_N+a_n^{}a^2\stackrel{~}{Y}_N)(a_m^{}X_N+b_m^{}Y_N)`$
respectively, and applying (A.10), (A.19) and then (3.1), we obtain the commutator $`[z_m^{},z_n]`$ in the form (3.21). |
warning/0507/math0507101.html | ar5iv | text | # Bost-Connes-Marcolli systems for Shimura varieties. I. Definitions and formal analytic properties.
## 1 Introduction
A few years ago, Bost and Connes \[BC95\] discovered a surprising relationship between the class field theory of $``$ and quantum statistical mechanics.
Mathematically, a quantum statistical mechanical system consists of a pair $`(𝒜,\sigma _t)`$, where $`𝒜`$ is a C\*-algebra and $`\sigma _t`$ is a one-parameter group of automorphisms of $`𝒜`$; physically, $`𝒜`$ is the algebra of observables and $`\sigma _t`$ is the time evolution of the physical system. The physical states of the system are given by certain linear functionals on $`𝒜`$.
The analogy between classical and quantum statistical mechanics can be described by the following array:
| | Classical | Quantum |
| --- | --- | --- |
| Observables | $`aC^{\mathrm{}}(X)`$ | $`a𝒜`$ |
| | $`(X,\omega )`$ $`2n`$-dim symplectic manifold | $`𝒜`$: C\*-algebra, $`a`$=$`a`$* |
| | (phase space) | |
| Bracket | Poisson bracket | Commutator |
| | $`\{a_1,a_2\}=\omega (\xi _{a_1},\xi _{a_2})`$ | $`[a_1,a_2]`$ |
| | with $`da_i+\omega (\xi _{a_i},\mathrm{\_})=0`$ | |
| Hamiltonian | $`H:X`$ | $`H`$ unbounded selfadjoint on $``$ |
| | | Representation $`\pi :𝒜()`$ |
| Time | Solution of | $`\sigma :\mathrm{Aut}(𝒜)`$ |
| evolution | $`\{H,a\}(x)=(\frac{d}{dt})_{t=0}a(\sigma _t(x))`$ | $`e^{itH}\pi (a)e^{itH}=\pi (\sigma _t(a))`$ |
| States | Probability measure $`\mu `$ on $`X`$ | Linear functional of norm $`1`$ |
| | $`\mathrm{\Phi }(a)=_Xad\mu `$ | $`\mathrm{\Phi }:𝒜`$ |
| Partition | $`\zeta (\beta )=_Xe^{\beta H}𝑑\mathrm{\Omega }`$ | $`\zeta (\beta )=\mathrm{Tr}(e^{\beta H})`$ |
| function | with $`\mathrm{\Omega }=\omega ^n`$ the volume form | |
| Equilibrium | Canonical ensemble | KMS condition: |
| States | $`d\mu =\frac{e^{\beta H}d\mathrm{\Omega }}{\zeta (\beta )}`$ | $`\mathrm{\Phi }(ab)=\mathrm{\Phi }(\sigma _{i\beta }(b)a)`$ |
| | | example: $`\mathrm{\Phi }(a)=\frac{\mathrm{Tr}(ae^{\beta H})}{\zeta (\beta )}`$ |
The statistical content means that one singles out the equilibrium states at a given temperature $`T=1/\beta `$ on $`𝒜`$, and these are characterized by the so called $`\mathrm{KMS}_\beta `$ condition. The set of these equilibrium states may have symmetries. Changing the temperature of a system can produce a phase transition phenomenon with spontaneous symmetry breaking, meaning that the symmetry changes radically with an arbitrary small change of temperature. For example, the formation at zero temperature of a snowflake from water is a phase transition, for which we can observe a symmetry breaking phenomenon: a snowflake has much more symmetry (it has crystal structure) than a drop of water (which consists of a random collection of molecules).
The Bost-Connes system $`(𝒜,\sigma _t)`$ also exhibits a phase transition phenomenon with symmetry breaking at $`\beta =1`$. For $`\beta <1`$, i.e., at high temperature, there is enough disorder so that the symmetry is trivial. For $`\beta >1`$, the set of equilibrium states “freezes” and has as symmetry the Galois group of the maximal abelian extension of $``$. Bost and Connes also defined explicitly a rational subalgebra of $`A_{}𝒜`$ such that the evaluation of KMS states on $`A_{}`$, at small temperature, generate $`^{\mathrm{ab}}`$. This system is related to $`\mathrm{GL}_{1,}`$.
Much more recently, Connes and Marcolli \[CM04\] defined an analogous system for $`\mathrm{GL}_{2,}`$, and overcame extreme technicalities to give in this case a meaning to all prominent features of the Bost-Connes system (symmetries, rational subalgebra, zeta function as partition function, relation to the Galois group of the modular field and its modular reciprocity law). One of the key points in their new approach is that their system is related to the study of the “noncommutative space” of $``$-lattices up to commensurability:
$$\mathrm{GL}_2()\backslash \mathrm{M}_2(𝔸_f)\times ^\pm ;$$
and that the set of $`\mathrm{KMS}_\beta `$ states at small temperature is in natural bijection with the Shimura variety
$$\mathrm{Sh}(\mathrm{GL}_2,^\pm )=\mathrm{GL}_2()\backslash \mathrm{GL}_2(𝔸_f)\times ^\pm .$$
The C\*-algebra corresponding to the “noncommutative space” of $``$-lattices up to commensurability is a groupoid C\*-algebra. This also gives a nice explanation for the origin of the Bost-Connes system.
We choose one direction of generalizing this work of Connes and Marcolli by replacing their basic Shimura datum $`(\mathrm{GL}_2,^\pm )`$ by a general Shimura datum $`(G,X)`$. In order to deal with the technical issue of defining the partition function, the construction of the Connes-Marcolli system involves a groupoid $`𝒰`$, corresponding to the commensurability relation on $``$-lattices, and the quotient of $`𝒰`$ by the arithmetic subgroup $`\mathrm{SL}_2()\mathrm{GL}_2()`$. We start by defining an algebra in more adelic terms, meaning that we use a quotient by the compact open subgroup $`\mathrm{GL}_2(\widehat{})\mathrm{GL}_2(𝔸_f)`$. The motivation for this construction comes from the fact that for number fields, one wants the partition function to be the Dedekind zeta function, and this is easier to obtain in the adelic language (as pointed out by Paula Cohen \[Coh99\]).
The Connes-Marcolli algebra is not exactly a groupoid algebra, because the quotient of the groupoid $`𝒰`$ by $`\mathrm{SL}_2()`$ is no longer a groupoid, since $`\mathrm{SL}_2()`$ does not act freely on $``$. In fact, if we use the stacky quotient, then this is a groupoid, but one cannot define an associated convolution C\*-algebra because there is no good notion of functions on stacks. There are two solutions to this problem, corresponding to two resolutions of the stack’s singularities. The first is to choose a smaller $`\mathrm{\Gamma }\mathrm{SL}_2()`$ that acts freely on $``$. This gives a finite resolution of the stack singularities. However, this first method works only for classical Shimura varieties, which does not include the case of a general number field. The second solution is to identify functions on $`\mathrm{SL}_2()\backslash =\mathrm{GL}_2()\backslash \mathrm{GL}_2()/^\times `$ to functions on $`\mathrm{GL}_2()\backslash \mathrm{GL}_2()`$ (which is another infinite resolution of stack singularities) which are invariant for the scaling action of $`^\times `$. This allows one to define a convolution algebra. This second method was the one chosen by Connes and Marcolli.
The role of $`\mathrm{M}_{2,}`$ in the $`\mathrm{GL}_{2,}`$ case is, in the case of general Shimura data $`(G,X)`$, played by a multiplicative semigroup $`\mathrm{M}`$ such that $`\mathrm{M}^\times =G`$. We learned a lot about such semigroups from N. Ramachandran and L. Lafforgue. Their main properties are given in the appendix.
This article describes the first steps in our work on these Bost-Connes-Marcolli systems for general Shimura data.
We solve along the way the problem of defining a Bost-Connes system for general number fields, which has the Dedekind zeta function as partition function *and* the group of connected components of the idele class group as symmetry group. For imaginary quadratic fields, this problem was very recently solved by Connes-Marcolli-Ramachandran \[CMR05\]. Previous works were either restricted to class number one, or did not have the right symmetry, or did not have the right partition function. All these interesting works however gradually improved and simplified the techniques involved in the study of Bost-Connes systems, and we also use methods from this litterature to prove some of our results. A generalization of the work \[BC95\] to the case of arbitrary global fields was proposed by Harari and Leichtnam \[HL97\]. A Hecke algebra construction using semi-group crossed products was proposed in the number field case by Arledge-Laca-Raeburn \[ALR97\], see also \[LR99\] and \[Lac98\]. Van Frankenhuijsen and Laca \[LvF04\] defined a system with Galois group as symmetry group for totally imaginary fields of class number one. P. Cohen \[Coh99\] constructed adelically a system with the right partition function in the number field case. For a nice and more complete survey of known results, see \[CM04\], Section 1.4.
We also study the explicit example of the Hilbert-Blumenthal modular varieties.
## Acknowledgments
The authors would like to thank the following persons for useful discussions in the preparation of this paper: A. Connes, G. Harder, C. Kaiser, M. Laca, L. Lafforgue, V. Lafforgue, B. Noohi, D. Panov, N. Ramachandran, B. Toen, D. Zagier. In particular, N. Ramachandran and L. Lafforgue gave us references and methods to construct enveloping semigroups, that are key objects in our formalism. B. Toen gave us a simplicial definition of stack groupoids.
After writing this paper, we learned from V. Lafforgue another construction of the Bost-Connes algebra for number fields that will certainly be useful to study finer aspects of those systems in dimension 1.
We thank the Max-Planck-Institut of Bonn for its hospitality and excellent working conditions during the preparation of this article. We also thank the Institut des Hautes Études Scientifiques for hospitality during the finalization of this article.
We especially thank Matilde Marcolli for suggesting us to work on the Connes-Marcolli system in the Hilbert modular case, for answering all our questions about the article \[CM04\], and for freely sharing with us her insights.
## 2 Another take on the Bost-Connes system
Before describing the general setting, we would like to present the illustrating example of the Bost-Connes system, which illuminates our general constructions.
### 2.1 The classical system
The *Bost-Connes groupoid* is given by the partially defined action of $`_+^\times `$ on $`\widehat{}`$. More precisely, it is given by
$$Z_{BC}=\{(g,\rho )_+^\times \times \widehat{}g\rho \widehat{}\}.$$
The unit space is $`\widehat{}`$, and the source and target maps are given by $`s(g,\rho )=\rho `$ and $`t(g,\rho )=g\rho `$. Composition is given by $`(g_2,\rho _2)(g_1,\rho _1)=(g_2g_1,\rho _1)`$ if $`g_1\rho _1=\rho _2`$.
The Bost-Connes Hecke algebra is simply $`:=C_c(Z_{BC})`$, equiped with the convolution product
$$(f_1f_2)(g,\rho )=\underset{h_+^\times ,h\rho \widehat{}}{}f_1(gh^1,h\rho )f_2(h,\rho ).$$
The time evolution on this algebra is given by
$$\sigma _t(f)(g,\rho )=g^{it}f(g,\rho ).$$
(2.1)
For each $`\rho _0\widehat{}^\times `$, we define a representation $`\pi _0:(\mathrm{}^2(^\times ))`$ by
$$(\pi _0(f)(\xi ))(n)=\underset{h^\times }{}f(nh^1,h\rho _0)\xi (h).$$
To finish, the Hamiltonian of this system is given by
$$H:\mathrm{}^2(^\times )\mathrm{}^2(^\times ),f(n)log(n)f(n).$$
By definition, the partition function
$$\zeta _{BC}(s)=\mathrm{Tr}(e^{sH})=\underset{n^\times }{}n^s=\zeta (s)$$
is exactly Riemann’s zeta function.
### 2.2 The same system in adelic terms
We will now give a more complicated description of the Bost-Connes groupoid, that has the advantage of admitting a direct generalization to other number fields, whose partition function is the Dedekind zeta function. This is the first step to be carried out in constructing Bost-Connes systems for number fields. Moreover, the advantage of this adelic formulation is that it also makes sense for general Shimura varieties.
We first remark that the quotient of $`\widehat{}`$ by the partially defined action of $`_+^\times `$ is the same as the quotient of $`\widehat{}\times \{\pm 1\}`$ by the partially defined action of $`^\times `$. In fact, this equality of quotient spaces can be described at the level of groupoids.
Let $`U^{\mathrm{princ}}^\times \times \widehat{}\times \{\pm 1\}`$ be the groupoid of elements $`(g,\rho ,z)`$ such that $`g\rho \widehat{}`$. This groupoid encodes the partially defined action of $`^\times `$ on $`\widehat{}\times \{\pm 1\}`$.
Now consider the quotient $`Z^{\mathrm{princ}}`$ of $`U^{\mathrm{princ}}`$ by the action of $`(^\times )^2=\{\pm 1\}^2`$ given by
$$(\gamma _1,\gamma _2).(g,\rho ,z):=(\gamma _1g\gamma _2^1,\gamma _2\rho ,\gamma _2z).$$
This is also a groupoid.
There is a natural morphism of groupoids $`Z_{BC}Z^{\mathrm{princ}}`$ given by $`(r,\rho )(r,\rho ,1)`$, which is in fact an isomorphism (cf. 5.1.2).
This new descrition of the Bost-Connes groupoid is nicer because it clearly relates the Bost-Connes system with the pair $`(𝔾_m,\{\pm 1\})`$, which is called the *multiplicative Shimura datum*.
To make this connexion clearer, it is natural to seek a fully adelic description of the Bost-Connes groupoid. This is because Shimura varieties are defined adelically. The adelic framework also facilitates the definition of Bost-Connes systems for number fields with Dedekind zeta function as partition function.
Recall that $`𝔸_f:=\widehat{}_{}`$. The strong approximation property for the multiplicative group $`𝔾_m`$ (which in this case is simply the chinese reminder theorem) tells us that
$$𝔸_f^\times =_+^\times .\widehat{}^\times =\widehat{}^\times ._+^\times .$$
We will now denote
$$\mathrm{Sh}(𝔾_m,\{\pm 1\}):=^\times \backslash \{\pm 1\}\times 𝔸_f^\times $$
and
$$Y:=\widehat{}\times \mathrm{Sh}(𝔾_m,\{\pm 1\}).$$
Consider the partially defined action of $`𝔸_f^\times `$ on $`Y`$ given by
$$g.(\rho ,[z,l]):=(g\rho ,[z,lg^1])$$
and let
$$U𝔸_f^\times \times Y$$
be the corresponding groupoid of elements $`(g,y)`$ such that $`gyY`$.
Now consider the quotient $`Z`$ of $`U`$ by the action of $`(\widehat{}^\times )^2`$ given by
$$(\gamma _1,\gamma _2).(g,y):=(\gamma _1g\gamma _2^1,\gamma _2y).$$
The strong approximation theorem for $`𝔾_m`$ implies that the natural map
$$\begin{array}{ccc}^\times \times \widehat{}\times \{\pm 1\}& & 𝔸_f^\times \times \widehat{}\times \mathrm{Sh}(𝔾_m,\{\pm 1\})\\ (g,\rho ,z)& & (g,\rho ,[z,1])\end{array}$$
induces an isomorphism of groupoids $`Z^{\mathrm{princ}}Z`$ (cf. 5.1.3).
The Bost-Connes algebra can thus be described as the algebra $`=C_c(Z)`$ with the convolution product
$$(f_1f_2)(g,y)=\underset{h\widehat{}^\times \backslash 𝔸_f^\times ,hyY}{}f_1(gh^1,hy)f_2(h,y).$$
Using the isomorphism $`d:\widehat{}^\times \backslash 𝔸_f^\times _+^\times `$, we can define the time evolution by
$$\sigma _t(f)(g,y)=d(g)^{it}f(g,y).$$
This is exactly the time evolution we defined in 2.1.
Let $`\widehat{}^{\mathrm{}}:=𝔸_f^\times \widehat{}`$ and $`^{\mathrm{}}:=\{0\}`$. The strong approximation theorem gives us that
$$\widehat{}^\times \backslash \widehat{}^{\mathrm{}}^\times \backslash ^{\mathrm{}}^\times .$$
Let $`_0:=\mathrm{}^2(\widehat{}^\times \backslash \widehat{}^{\mathrm{}})\mathrm{}^2(^\times ).`$ For each $`\rho _0\widehat{}^\times `$, we define a representation $`\pi _0:(_0)`$ by
$$(\pi _0(f)(\xi ))(n)=\underset{h^\times }{}f(nh^1,h\rho _0)\xi (h).$$
To finish, the Hamiltonian of this system is given by
$$H:_0_0,f(n)log(d(n))f(n).$$
This adelic system is perfectly identical to the original Bost-Connes system. We will however see in the sequel that essentially the same definitions of algebra, time evolution, representations and Hamiltonian now work for general Shimura data.
## 3 Background material
In this paper we draw upon the theory of Shimura varieties and operator algebras. Since these fields have traditionally had little to do with each other, we review for the convenience of the reader some of the basic (well-known) results that we shall need. This also allows us to establish notation. We stress that our definition of a Shimura variety is a slight variation on the usual one given by Deligne \[Del79\], 2.1.
### 3.1 Shimura varieties
If $`G`$ is a reductive group over $``$, $`G()^+`$ will denote the connected component of identity in the real Lie groups of its real points and $`G()^+:=G()G()^+`$. First recall briefly the definition of Shimura data. We will use a mix of Deligne’s definition (see \[Del79\], 2.1) and Pink’s definition (see \[Pin90\], 2.1). Let $`𝕊:=\mathrm{Res}_/𝔾_m`$.
###### Definition 3.1.1.
A *Shimura datum* is a triple $`(G,X,h)`$, with $`G`$ a connected reductive group over $``$, $`X`$ a left homogeneous space under $`G()`$ and $`h:X\mathrm{Hom}(𝕊,G_{})`$ a $`G()`$-equivariant map <sup>1</sup><sup>1</sup>1for the natural conjugation action of $`G()`$ on $`\mathrm{Hom}(𝕊,G_{})`$ with finite fibres such that:
1. For $`h_xh(X)`$, $`\mathrm{Lie}(G_{})`$ is of type $`\{(1,1),(0,0),(1,1)\}`$;
2. The involution $`\mathrm{int}h_x(i)`$ is a Cartan involution of the adjoint group $`G_{}^{ad}`$;
3. The adjoint group has no factor $`G^{}`$ defined over $``$ on which the projection of $`h_x`$ is trivial.
A Shimura datum is called *classical* if it moreover fulfils the axiom
1. Let $`Z_0(G)`$ be the maximal split subtorus of the center of $`G`$; then $`\mathrm{int}h_x(i)`$ is a Cartan involution of $`G/Z_0(G)`$.
###### Example 3.1.2.
Let $`F`$ be a number field, $`T=\mathrm{Res}_{F/}𝔾_{m,F}`$ and $`X_F=T()/T()^+`$. We have $`F_{}^i\times ^j`$. We put on $`F_{}`$ the Hodge structure that is trivial on $`^j`$ and given by the choice of a complex structure on $`^i`$ (among the $`2^i`$ possibilities). This gives a morphism $`h_1:𝕊T_{}`$. The triple $`(\mathrm{Res}_{F/}𝔾_{m,F},X_F,h_1)`$ is called the *multiplicative Shimura datum of the field $`F`$*. This Shimura datum is classical if and only if $`F=`$ of $`F`$ is imaginary quadratic.
We will often denote a Shimura data just by a couple $`(G,X)`$ when the morphism $`h`$ is clear from the situation.
###### Example 3.1.3.
Let $`h:𝕊\mathrm{GL}_{2,}`$ be the morphism given by $`h(a+ib)=\left(\begin{array}{cc}a& b\\ b& a\end{array}\right)`$. Let $`^\pm `$ be the $`\mathrm{GL}_2()`$-conjugacy class of $`h`$. It identifies with the Poincaré double half plane with action of $`\mathrm{GL}_2()`$ by homographies. Then $`(\mathrm{GL}_2,^\pm )`$ is called the *modular Shimura datum*.
###### Definition 3.1.4.
Let $`(G,X)`$ be a Shimura datum. Let $`KG(𝔸_f)`$ be a compact open subgroup. The *level $`K`$ Shimura variety* is
$$\mathrm{Sh}_K(G,X):=G()\backslash X\times G(𝔸_f)/K$$
and the *Shimura variety* is the projective limit
$$\mathrm{Sh}(G,X):=\underset{}{\mathrm{lim}}_K\mathrm{Sh}_K(G,X)$$
over all compact open subgroups $`KG(𝔸_f)`$.
A *topological stack* will be for us a stack on the site (Top) of topological spaces with usual open coverings, i.e., a category fibered in groupoids fulfilling some descent condition. See appendix A for more details.
We first remark that, from the modular viewpoint, it is more natural to study the *level K Shimura stack*, given by the topological stacky quotient
$$𝔖𝔥_K(G,X):=[G()\backslash X\times G(𝔸_f)/K],$$
and the *Shimura stack*, given by the 2-projective limit
$$𝔖𝔥(G,X):=\underset{}{\mathrm{lim}}_K𝔖𝔥_K(G,X).$$
In the case of the multiplicative Shimura datum of a number field, i.e., $`G=\mathrm{Res}_{F/}𝔾_{m,F}`$, the level $`K`$ Shimura stack can have infinite isotropy groups given by $`G()^+K`$. These isotropy groups are given by generalized congruence relations on the group of units $`𝒪_F^\times `$. We will have to keep track of (some of) these isotropy groups in the case of non-classical Shimura data.
There is a natural right action of $`G(𝔸_f)`$ on $`\mathrm{Sh}(G,X)`$ given, for each $`gG(𝔸_f)`$ and $`KG(𝔸_f)`$ compact open, by an isomorphism
$$\begin{array}{cccc}(.g):& 𝔖𝔥_K(G,X)& & 𝔖𝔥_{g^1Kg}(G,X)\\ & [x,l]& & [x,lg].\end{array}$$
Under the hypothesis that $`(G,X)`$ is classical (see also \[Del79\], 2.1.1.4, 2.1.1.5), there is an easier description of the Shimura variety (see \[Del79\], Corollaire 2.1.11):
$$\mathrm{Sh}(G,X)G()\backslash X\times G(𝔸_f)$$
and the Shimura stacks $`𝔖𝔥_K(G,X)`$ are in fact algebraic stacks over $``$.
Unfortunately, these hypothesis are not always fulfilled in the case of the multiplicative Shimura datum of a general number field $`F`$. In fact, the quotient $`G()\backslash X\times G(𝔸_f)`$ is not always Hausdorff in this case.
For example, if $`F=(\sqrt{2})`$,
$$\mathrm{Sh}(\mathrm{Res}_{F/}𝔾_{m,F},X_F)F^\times \backslash X_F\times 𝔸_{f,F}^\times $$
(this is essentially due to the fact that the group of units $`𝒪_F^\times `$ is infinite).
Points in a Shimura variety will be denoted by pairs $`[z,l]`$. If the Shimura datum is classical, this means that $`zX`$ and $`lG(𝔸_f)`$. Otherwise, $`[z,l]=[z_K,l_K]_{KG(𝔸_f)}`$ is a family of points in $`\mathrm{Sh}_K(G,X)`$ indexed by the set of compact open subgroups in $`G(𝔸_f)`$.
###### Definition 3.1.5.
Let $`(G,X)`$ be a Shimura datum. A compact open subgroup $`KG(𝔸_f)`$ is called *neat* if it acts freely on $`G()\backslash X\times G(𝔸_f)`$.
We would like to be able to define *natural* algebras of continuous “functions” on the finite Shimura varieties in play. In order to do that, we have to resolve their stack singularities.
We remark that if $`K`$ is neat, then the quotient analytic stack
$$𝔖𝔥_K(G,X)=[G()\backslash X\times G(𝔸_f)/K]$$
is a usual analytic space, but otherwise it is worthwhile from the moduli viewpoint to keep track of the nontrivial stack structure. For classical Shimura data, one can resolve the stack singularities by choosing a smaller compact open subgroup $`K^{}G(𝔸_f)`$ that acts freely. This is what we will usually do in order to be able to define continuous “functions” on the stack $`𝔖𝔥_K(G,X)`$.
However, this finite resolution of the stack singularities is usually not possible for nonclassical Shimura data, as we can see on the following example. Let $`F=(\sqrt{2})`$ and $`(\mathrm{Res}_{F/}𝔾_{m,F},X_F)`$ be the corresponding Shimura datum (here $`X_F\{\pm 1\}^2`$). Let $`K=\widehat{𝒪}_{F}^{}{}_{}{}^{\times }`$ and consider the stack $`𝔖𝔥_K(\mathrm{Res}_{F/}𝔾_{m,F},X_F)`$. Its coarse quotient is the ideal class group of $`F`$, i.e., the trivial group $`\{1\}`$. Since $`F`$ has class number one, this coarse quotient can also be described as $`𝒪_F^\times \backslash X_F`$. In this case, $`𝒪_F^\times `$ is infinite, so that we can not choose a smaller $`K^{}K`$ that acts freely on $`F^\times \backslash X_F\times 𝔸_{f,F}^\times `$. In fact, the unit group is finite if and only if $`F=`$ or $`F`$ is imaginary quadratic, i.e., if and only if $`(\mathrm{Res}_{F/}𝔾_{m,F},X_F)`$ is classical in our language.
If we want to resolve the stack singularities, we can use the quotient map
$$F^\times \backslash 𝔸_F^\times /K𝔖𝔥_K(\mathrm{Res}_{F/}𝔾_{m,F},X_F)$$
for the scaling action of the connected component of identity $`D_F`$ in the full idele class group $`C_F:=F^\times \backslash 𝔸_F^\times `$.
###### Remark 3.1.6.
From the viewpoint of moduli spaces, it is important that the coarse space $`\mathrm{Sh}_K(\mathrm{Res}_{F/}𝔾_{m,F},X_F)`$, i.e., the big ideal class group, be replaced by the corresponding group stack with infinite stabilizers (given by groups of units with congruence conditions):
$$𝔖𝔥_K(\mathrm{Res}_{F/}𝔾_{m,F},X_F).$$
This “equivariant viewpoint” of the finite level Shimura variety could also be important to understand geometrically the definition of Stark’s zeta functions, and also for the understanding of Manin’s real multiplication program \[Man\].
### 3.2 C\*-algebras and quantum statistical mechanics
We review here basic definitions from the theory of C\*-algebras, emphasising those parts relevant to quantum statistical mechanics. Good references for the material in this section are \[BR87\] and \[BR97\]. For an overview of the grand physical picture, see \[Haa96\].
###### Definition 3.2.1.
A *C\*-algebra* is a (not necessarily unital) complex algebra $`A`$ endowed with a conjugate-linear involutive anti-automorphism $`{}_{}{}^{}:AA`$, and a norm $``$, satisfying the following conditions: For every $`a`$, $`bA`$ we have
1. $`A`$ is complete with respect to the norm, and $`abab`$ (i.e., $`A`$ is a *Banach algebra*); and
2. $`a^{}a=a^2`$, the crucial *C\*-condition*.
Actually a C\*-algebra is not as abstract as it may seem, because every C\*-algebra can be realized as a norm-closed sub-\*-algebra of the algebra of bounded operators on a Hilbert space (Theorem of Gelfand-Naimark \[BR87\], Theorem 2.1.10), and every such subalgebra is a C\*-algebra.
The operator algebraic formulation of quantum statistical mechanics (see the introduction to \[BR87\]) consists of a C\*-algebra $`A`$ together with a 1-parameter group of automorphism $`\sigma _t:AA`$, which is continuous in the sense that $`t\sigma _t(a)`$ is continous for every $`aA`$. The algebra $`A`$ is then the algebra of quantum observables, while $`\sigma _t`$ is the time evolution. The pair $`(A,\sigma _t)`$ is an example of a *C\*-dynamical system*. The *states* of the C\*-algebra $`A`$ are the continuous complex-linear functionals $`\mathrm{\Phi }`$ of norm $`1`$ which are positive, i.e., $`\mathrm{\Phi }(a^{}a)0`$ for every $`aA`$. The number $`\mathrm{\Phi }(a)`$ is then the expectation value of the observable $`a`$ in the physical state $`\mathrm{\Phi }`$.
To regard the pair $`(A,\sigma _t)`$ as a statistical mechanical system we need an appropriate notion of an “equilibrium state” at temperature $`T=1/\beta `$. This is provided by the KMS condition.
###### Definition 3.2.2.
The *KMS-$`\beta `$ condition* ($`0<\beta <\mathrm{}`$) for a state $`\mathrm{\Phi }`$ is the condition: For every pair of elements $`a`$, $`bA`$, there is a complex-valued function $`F`$ on the closed strip $`\mathrm{\Omega }=\{z0\mathrm{im}z\beta \}`$ such that
$$F(t)=\mathrm{\Phi }\left(a\sigma _t(b)\right),F(t+i\beta )=\mathrm{\Phi }\left(\sigma _t(b)a\right);$$
furthermore, the function $`F`$ is required to be bounded and continuous on $`\mathrm{\Omega }`$, and analytic on its interior.
This is the definition one often sees in the literature, although in practice it is easier to use the following equivalent characterization.
###### Proposition 3.2.3.
Let $`(A,\sigma _t)`$ be a C\*-dynamical system, and let $`\mathrm{\Phi }`$ be a state of $`A`$.
1. (\[BR87\], Corollary 2.5.23) There is a norm-dense \*-subalgebra $`A^{\mathrm{an}}`$ of $`A`$ such that for every $`aA^{\mathrm{an}}`$, the function $`t\sigma _t(a)`$ can be analytically continued to an entire function.
2. (\[BR87\], Definition 5.3.1 and Corollary 5.3.7) The state $`\mathrm{\Phi }`$ is a KMS-$`\beta `$ state if and only if
$$\mathrm{\Phi }\left(a\sigma _{i\beta }(b)\right)=\mathrm{\Phi }(ba)$$
for all $`a`$, $`b`$ in a norm-dense $`\sigma _t`$-invariant \*-subalgebra of $`A^{\mathrm{an}}`$.
We now proceed to a description of the structure of the set of KMS-$`\beta `$ states. But before doing so, we need to explain the GNS construction, which is a method of getting representations of a C\*-algebra from its states; it is a basic, widely used result in the theory of operator algebras. We also need to define the notion of a factor state. We shall use standard notation: given a Hilbert space $``$, we denote the C\*-algebra of all bounded operators on $``$ by $`B()`$, and the inner product on $``$ by $`,`$.
###### Proposition 3.2.4 (GNS construction; \[BR87\], 2.3.16).
Let $`\mathrm{\Phi }`$ be a state of a C\*-algebra $`A`$. Then there is a triple $`(_\mathrm{\Phi },\pi _\mathrm{\Phi },\xi _\mathrm{\Phi })`$ consisting of a representation $`\pi _\mathrm{\Phi }`$ of $`A`$ on a Hilbert space $`_\mathrm{\Phi }`$ and a unit vector $`\xi _\mathrm{\Phi }_\mathrm{\Phi }`$ such that:
1. $`\mathrm{\Phi }(a)=\pi _\mathrm{\Phi }(a)\xi _\mathrm{\Phi },\xi _\mathrm{\Phi }`$ for all $`aA`$; and
2. The orbit $`\pi _\mathrm{\Phi }(A)\xi _\mathrm{\Phi }`$ is norm-dense in $`B(_\mathrm{\Phi })`$.
The triple $`(_\mathrm{\Phi },\pi _\mathrm{\Phi },\xi _\mathrm{\Phi })`$ is unique up to unitary equivalence.
The states of particular relevance to the KMS theory are the *factor states*. These are the states $`\mathrm{\Phi }`$ for which the corresponding GNS representation $`\pi _\mathrm{\Phi }`$ generates a *factor*, which is to say that the weak closure of $`\pi _\mathrm{\Phi }(A)`$ in $`B(_\mathrm{\Phi })`$ has centre consisting of the scalar operators. (This weak closure is an example of a *Von Neumann algebra*.)
We can now state the main structure theorem for the set of KMS-$`\beta `$ states.
###### Proposition 3.2.5 (Structure of KMS states; \[BR97\], Theorem 5.3.30).
The set $`_\beta `$ of KMS-$`\beta `$ states is a convex, weak\*-compact simplex. The extreme points of $`_\beta `$ are precisely those KMS-$`\beta `$ states that are factor states.
## 4 Abstract Bost-Connes-Marcolli systems
The aim of this section is to define Bost-Connes-Marcolli systems for general Shimura data $`(G,X)`$ and study their basic formal properties. A better understanding of the general setup might be gained by looking at section 7 where we specialise to the case of multiplicative Shimura datum (the case relevant for number fields).
### 4.1 BCM data
In order to define a generalization of the Connes-Marcolli algebra to general Shimura data, we want to make clear the separation between algebraic and level structure data, which is already implicit in the construction of Connes and Marcolli.
Algebraic data. We first need to consider a semigroup $`M`$ which plays the role for a general reductive group $`G`$ that $`\mathrm{M}_{2,}`$ plays for $`\mathrm{GL}_{2,}`$.
###### Definition 4.1.1.
Let $`G`$ be reductive group over a field. An *enveloping semigroup* for $`G`$ is a multiplicative semigroup $`M`$ which is irreducible and normal, and such that $`M^\times =G`$.
###### Definition 4.1.2.
A *BCM datum* is a tuple $`𝒟=(G,X,V,M)`$ with $`(G,X)`$ a Shimura datum, $`V`$ a faithful representation of $`G`$ and $`M`$ an enveloping semigroup for $`G`$ contained in $`\mathrm{End}(V)`$.
The faithful representation will often be denoted $`\varphi :G\mathrm{GL}(V)`$.
Level structure data. Every Shimura datum $`(G,X)`$ comes implicitly with a family of level structures given by the family of compact open subgroups $`KG(𝔸_f)`$. Connes and Marcolli fixed the full level structure $`\mathrm{GL}_2(\widehat{})\mathrm{GL}_2(𝔸_f)`$ as starting datum for their construction. To avoid the problem they had with stack singularities of their groupoid, we will fix a neat level structure as part of the datum.
The level structure also plays a role in defining the partition function of our system. Consideration of maximal level structures then yields standard zeta functions as partition functions, for example, the Dedekind zeta function of a number field. A technical requirement in the definition of the partition function is the choice of a lattice in the representation of $`G`$, which enables us to define a rational determinant for the adelic matrices in play.
###### Definition 4.1.3.
Let $`𝒟=(G,X,V,M)`$ be a BCM datum. A *level structure on $`𝒟`$* is a triple $`=(L,K,K_M)`$, with $`LV`$ a lattice, $`KG(𝔸_f)`$ a compact open subgroup, and $`K_MM(𝔸_f)`$ a compact open subsemigroup, such that
* $`K_M`$ stabilizes $`L_{}\widehat{}`$,
* $`\varphi (K)`$ is contained in $`K_M`$.
The pair $`(𝒟,)`$ will be called *a BCM pair*.
We can summarize the relation between $`L`$, $`K`$ and $`K_M`$ by the following diagram:
###### Definition 4.1.4.
The *maximal level structure* $`_0=(L,K_0,K_{M,0})`$ associated with a datum $`𝒟=(G,X,V,M)`$ and a lattice $`LV`$ is defined by setting
$$\begin{array}{ccc}K_{M,0}& :=& M(𝔸_f)\mathrm{End}(L_{}\widehat{}),\\ K_0& :=& \varphi ^1(K_{M,0}^\times ).\end{array}$$
###### Definition 4.1.5.
The level structure $``$ on $`𝒟`$ is called *fine* if $`K`$ acts freely on $`G()\backslash X\times G(𝔸_f)`$.
The maximal level structure is usually not neat enough to avoid stack singularity problems in the generalization of the Connes-Marcolli algebra. This is why we introduce the additional data of a compact open subgroup $`KK_M`$. For example, for the Connes-Marcolli case, one takes $`K=\mathrm{GL}_2(\widehat{})`$, $`K_M=\mathrm{M}_2(\widehat{})`$, but the fact that this choice of $`K`$ is not neat implies that the groupoid we introduce in the next section has stack singularities. Thus we instead choose a smaller $`K=K(N)\mathrm{GL}_2(\widehat{})`$ given by the kernel of the mod $`N`$ reduction of matrices.
Symmetries and zeta function. The symmetries of the Connes-Marcolli system play an important role in its relations with arithmetic. The analogous symmetry in our generalization is the following (which will be justified in Subsection 4.5).
###### Definition 4.1.6.
The semigroup $`\mathrm{Sym}_f(𝒟,):=\varphi ^1(K_M)`$ is called the *finite symmetry semigroup of the BCM pair $`(𝒟,)`$*. We will denote by $`\mathrm{Sym}_f^\times (𝒟,)`$ the group of invertible elements in $`\mathrm{Sym}_f(𝒟,)`$.
We included in $``$ the datum of a lattice in the representation $`\varphi `$ in order to define a determinant map.
###### Lemma 4.1.7.
The determinant $`det:\mathrm{GL}(L)𝔾_m`$ induces a natural map,
$$(det\varphi ):K\backslash G(𝔸_f)/K_+^\times .$$
The image of $`\mathrm{Sym}_f(𝒟,)`$ under this map is contained in $`^\times `$.
###### Proof.
Since $`\varphi (K)K_M^\times \mathrm{GL}(L)(\widehat{})`$, the representation $`\varphi :G\mathrm{GL}(L_{})`$ induces a map
$$\varphi :K\backslash G(𝔸_f)/K\mathrm{GL}(L)(\widehat{})\backslash \mathrm{GL}(L)(𝔸_f)/\mathrm{GL}(L)(\widehat{}).$$
The determinant map $`\mathrm{GL}(L)𝔾_m`$ induces a natural map
$$det:\mathrm{GL}(L)(\widehat{})\backslash \mathrm{GL}(L)(𝔸_f)/\mathrm{GL}(L)(\widehat{})\widehat{}^\times \backslash 𝔸_f^\times /\widehat{}^\times \widehat{}^\times \backslash 𝔸_f^\times ^\times \backslash ^\times _+^\times .$$
The composition $`det\varphi `$ gives us the desired map. The image of $`\mathrm{Sym}_f`$ under this map is contained in the image of $`\mathrm{GL}(L)(𝔸_f)\mathrm{End}(L)(\widehat{})`$ under the determinant map, which is exactly $`\widehat{}^{\mathrm{}}:=𝔸_f^\times \widehat{}`$. The quotient $`\widehat{}^\times \backslash \widehat{}^{\mathrm{}}`$ is identified with $`^\times \backslash ^\times ^\times `$. ∎
###### Definition 4.1.8.
The *zeta function* of the BCM pair $`(𝒟,)`$ is the complex valued series
$$\zeta _{𝒟,}(\beta ):=\underset{g\mathrm{Sym}_f^\times \backslash \mathrm{Sym}_f}{}det(\varphi (g))^\beta .$$
The BCM pair $`(𝒟,)`$ is called *summable* if there exists $`\beta _0`$ such that $`\zeta _{𝒟,}(\beta )`$ converges in the right plane $`\{\beta \text{Re}(\beta )>\beta _0\}`$ and extends to a meromorphic function on the full complex plane.
### 4.2 The BCM groupoid
Let $`(𝒟,)=((G,X,V,M),(L,K,K_M))`$ be a BCM pair. There are left and right actions of $`G(𝔸_f)`$ on $`M(𝔸_f)`$.
#### 4.2.1 Definition
Connes and Marcolli remarked in \[CM04\] that, if we want to take a quotient of a groupoid by a group action, it is essential that the action is free on the unit space of the groupoid. If we take the usual quotient set of a groupoid by an action that is not free on the unit space, this will not give a groupoid. We are thus obliged to use unit spaces that are in fact stacks. Some of them have nice singularities (i.e., those with finite stabilizers). Others don’t, but the language of stacks allows one to work in full generality without bothering about the freeness of actions in play.
We will denote the stacks by german letters; the corresponding coarse spaces will be denoted by right letters.
Let
$$Y_{𝒟,}=K_M\times \mathrm{Sh}(G,X).$$
We denote points of $`Y_{𝒟,}`$ by triples $`y=(\rho ,[z,l])`$ with $`\rho K_M`$, $`[z,l]\mathrm{Sh}(G,X)`$.
We want to study the equivalence relation on $`Y_{𝒟,}`$ given by the following partially defined action of $`G(𝔸_f)`$:
$$g.y=(g\rho ,[z,lg^1]),\text{where }y=(\rho ,[z,l]).$$
This equivalence relation will be called the *commensurability relation*. This terminology is derived from the notion of commensurability for $``$-lattices, cf. \[CM04\].
Consider the subspace
$$U_{𝒟,}G(𝔸_f)\times Y_{𝒟,}$$
of pairs $`(g,y)`$ such that $`gyY_{𝒟,}`$, i.e. $`g\rho K_M`$.
The space $`U_{𝒟,}`$ is a groupoid with unit space $`Y_{𝒟,}`$. The source and target maps $`s:U_{𝒟,}Y_{𝒟,}`$ and $`t:U_{𝒟,}Y_{𝒟,}`$ are given by $`s(g,y)=y`$ and $`t(g,y)=gy`$. The composition is given, for $`y_1=g_2y_2`$, by $`(g_1,y_1)(g_2,y_2)=(g_1g_2,y_2)`$. Notice that the groupoid obtained by restricting this groupoid to the $`(g,(\rho ,[z,l]))`$ such that $`\rho `$ is invertible is free, i.e., the equality $`t(g,y)=s(g,y)`$ implies $`g=1`$.
There is a natural action of $`K^2`$ on the groupoid $`U_{𝒟,}`$, given by
$$(g,y)(\gamma _1g\gamma _2^1,\gamma _2y),$$
and the induced action on $`Y_{𝒟,}`$ is given by
$$y\gamma _2y.$$
There are two motivations for quotienting $`U_{𝒟,}`$ by this action. The first one is physical: it is necessary to obtain a reasonable partition function for our system. The second is moduli theoretic: $`U_{𝒟,}`$ is only a pro-analytic groupoid and the quotient by $`K^2`$ is fibered over the Shimura variety $`𝔖𝔥_K(G,X)`$ which is an algebraic moduli stack of finite type whose definition could be made over $`\overline{}`$, at least when $`(G,X)`$ is classical and the Shimura variety has a canonical model.
Let $`_{𝒟,}`$ be the quotient stack $`[K^2\backslash U_{𝒟,}]`$ and $`𝔖_{𝒟,}`$ be the quotient stack $`[K\backslash Y_{𝒟,}]`$. The natural maps
$$s,t:_{𝒟,}𝔖_{𝒟,}$$
define a stack-groupoid structure (see appendix A) on $`_{𝒟,}`$ with unit stack $`𝔖_{𝒟,}`$.
###### Definition 4.2.2.
The stack-groupoid $`_{𝒟,}`$ is called the *Bost-Connes-Marcolli*<sup>2</sup><sup>2</sup>2We will often call it the BCM groupoid, for short. groupoid.
Let $`Z_{𝒟,}:=K^2\backslash U_{𝒟,}`$ be the (classical, i.e., coarse) quotient of $`U_{𝒟,}`$ by the action of $`K^2`$. If $`K`$ is small enough, i.e., if $`K`$ acts freely on $`G()\backslash X\times G(𝔸_f)`$, then $`_{𝒟,}`$ is equal to the classical quotient $`Z_{𝒟,}`$, which is a groupoid in the usual sense, with units $`S=K\backslash Y_{𝒟,}`$. Otherwise, suppose that there exists a compact open subgroup $`K^{}K`$ that acts freely on $`G()\backslash X\times G(𝔸_f)`$ and choose on $`𝒟`$ the level structure $`^{}=(L,K^{},K_M)`$. The stack $`_{𝒟,^{}}`$ is a usual topological space that is a finite covering of the coarse space $`Z_{𝒟,}`$ and such that the stack $`_{𝒟,}`$ is the stacky quotient of $`Z_{𝒟,^{}}`$ by the projection equivalence relation to $`Z_{𝒟,}`$.
The reader who prefers to work with usual analytic spaces will thus suppose that $`K`$ is small enough, but as we remarked before, our basic examples (number fields) do not fulfil this hypothesis. We have also to recall that for nonclassical Shimura data $`(G,X)`$ in the sense of definition 3.1.1, there exists no such small enough $`KG(𝔸_f)`$. This is essentially due to the fact that the “unit group” $`C()K`$ (where $`C`$ denotes the center of $`G`$) can be infinite.
#### 4.2.3 The commensurability class map
Recall that $`\varphi `$ is the representation of $`G`$, which we view as the natural inclusion $`GM`$.
For classical Shimura data. We want to give an explicit description of the quotient of $`Y_{𝒟,}`$ by the commensurability equivalence relation, in the case where $`(G,X)`$ is classical, i.e., when
$$\mathrm{Sh}(G,X)=G()\backslash X\times G(𝔸_f).$$
Let $`K_𝔸^M=\varphi (G)(𝔸_f).K_MM(𝔸_f)`$. There is a natural surjective map of sets
$$\pi :Y_{𝒟,}G()\backslash X\times K_𝔸^M$$
given by $`\pi (\rho ,[z,l])=[z,l\rho ]`$.
Let $`Y_{𝒟,}^\times =K_M^\times \times (G()\backslash X\times G(𝔸_f))`$ be the invertible part of $`Y_{𝒟,}`$ and let $`Z_{𝒟,}^\times _{𝒟,}`$ be the corresponding subspace (which is a groupoid in the usual sense because $`K`$ acts freely on $`K_M^\times `$); that is, $`Z_{𝒟,}^\times `$ is defined just as $`_{𝒟,}`$ is, but with $`Y_{𝒟,}^\times `$ in place of $`Y_{𝒟,}`$. Let $`S_{𝒟,}^\times :=K\backslash Y_{𝒟,}^\times `$ be the unit space of $`Z_{𝒟,}^\times `$. Since $`K_M^\times M^\times (𝔸_f)=\varphi (G(𝔸_f))`$, the map $`\pi `$ induces a natural map
$$\begin{array}{cccc}\pi ^\times :& Y_{𝒟,}^\times & & G()\backslash X\times G(𝔸_f)\\ & (\rho ,[z,l])& & [z,l\varphi ^1(\rho )],\end{array}$$
which is complex analytic (for the natural analytic structures induced by the complex structure on $`X`$) and surjective. Both $`\pi `$ and $`\pi ^\times `$ factor through the quotient of their sources by the left action of $`K`$. We will continue to denote this factorisation by $`\pi `$ and $`\pi ^\times `$.
###### Definition 4.2.4.
The maps $`\pi `$ and $`\pi ^\times `$ are called the *commensurability class maps*.
The last definition is justified by the following lemma. The notion of coarse quotient can be found in Definition A.1.1.
###### Lemma 4.2.5.
The maps $`\pi `$ and $`\pi ^\times `$ are in fact the coarse quotient maps for the groupoids $`_{𝒟,}`$ and $`Z_{𝒟,}^\times `$ acting on their unit spaces $`𝔖_{𝒟,}`$ and $`S_{𝒟,}^\times `$.
###### Proof.
If $`(g,\rho ,[z,l])U_{𝒟,}`$, then $`\pi (g\rho ,[z,lg^1])=[z,l\rho ]=\pi (\rho ,[z,l])`$ which proves that $`\pi `$ factors through
$$|𝔖_{𝒟,}/_{𝒟,}|G()\backslash X\times K_𝔸^M.$$
This surjective map is in fact an isomorphism. Indeed, if $`(\rho ,[z,l]),(\rho ^{},[z^{},l^{}])Y_{𝒟,}`$ have same image under $`\pi `$, then there exists $`gG()`$ such that $`gl\rho =l^{}\rho ^{}`$ and $`gz=z^{}`$. We then know that in the quotient space $`|𝔖_{𝒟,}/_{𝒟,}|`$,
$$\begin{array}{ccc}(\rho ,[z,l])& =& (l^1g^1gl\rho ,[z,l])\hfill \\ & =& (l^1g^1l^{}\rho ^{},[z,l])\hfill \\ & & (l_{}^{}{}_{}{}^{1}gll^1g^1l^{}\rho ^{},[z,ll^1g^1l^{}])\hfill \\ & =& (\rho ^{},[z,g^1l^{}])\hfill \\ & =& (\rho ^{},[gz,l^{}])\hfill \\ & =& (\rho ^{},[z^{},l^{}]).\hfill \end{array}$$
This proves injectivity of $`\pi `$ and surjectivity was already known. The argument for $`\pi ^\times `$ is similar. ∎
For commutative Shimura data. Commutative Shimura data form another family of examples for which we can construct the commensurability class map in simple terms. The multiplicative datum of a number field is in this familly. Thus we now suppose that $`𝒟=(G,X,V,M)`$ is a BCM datum such that $`G`$ and $`M`$ are commutative, and we let $``$ be a level structure on $`𝒟`$. For each $`K^{},KG(𝔸_f)`$ compact open, there is a natural map
$$𝔜_K^{}:=K_M\times 𝔖𝔥_K^{}(G,X)[G()\backslash X\times M(𝔸_f)/K^{}]$$
given by $`(\rho ,[z,l])[z,l\rho ]`$. This map is $`K`$-equivariant for the trivial action of $`K`$ on the range because the image of $`k.(\rho ,[z,l])=(k\rho ,[z,lk^1])`$ is equal to the image of $`(\rho ,[z,l])`$. Recall that $`𝔖_{𝒟,}:=[K\backslash Y_{𝒟,}]`$ and $`S_{𝒟,}^\times =K\backslash Y_{𝒟,}^\times `$. If we pass to the limit on $`K^{}G(𝔸_f)`$, and then to the quotient by $`K`$, we obtain natural maps
$$\pi :𝔖_{𝒟,}\underset{}{\mathrm{lim}}_K^{}[G()\backslash X\times M(𝔸_f)/K^{}]$$
and
$$\pi ^\times :S_{𝒟,}^\times \mathrm{Sh}(G,X)$$
that will be called as before the *commensurability class maps*.
The image of the map $`\pi `$ is as before the coarse quotient for the action of the groupoid $``$ on its unit space $`𝔖`$.
Denote $`𝔖_K^{}:=K\backslash (K_M\times 𝔖𝔥_K^{}(G,X))`$. We should remark here that in this commutative case, the space $`𝔖_K^{}`$ is the unit space of a well defined groupoid $`_K^{}`$ because the $`G(𝔸_f)`$ action on $`𝔖𝔥_K^{}(G,X)`$ is well defined. This shows that
$$=\underset{}{\mathrm{lim}}_K^{}_K^{},$$
(4.1)
which will be useful for the description of the symmetries of Bost-Connes systems for number fields.
### 4.3 Defining BCM algebras
#### 4.3.1 Functions on BCM stacks?
Let $`𝒟=(G,X,V,M)`$ be a BCM datum, $`V`$ be a representation of $`G`$, and let $`L_0`$ be the associated maximal level structure (Def. 4.1.4). We would like to define the BCM algebra of $`(𝒟,_0)`$ as a groupoid algebra. Unfortunately, the corresponding groupoid is usually only a stack and there is no canonical notion of continuous functions on such a space. More precisely, if a Stack has some nontrivial isotropy group, Connes’ philosophy of noncommutative geometry tells us that the “algebra of functions” on it should include this isotropy information in a nontrivial way, and this algebra depends on a presentation of the stack.
If $`(G,X)`$ is classical, there is a very natural way to resolve the stack singularities of $`_{𝒟,_0}`$ by choosing a neat level structure $``$, for which the projection map
$$Z_{𝒟,}_{𝒟,_0}$$
is such a resolution. The corresponding convolution algebra of the groupoid $`Z_{𝒟,}`$ is a completely natural replacement for the groupoid algebra of the stack-groupoid $`_{𝒟,_0}`$.
If $`(G,X)`$ is nonclassical, there is no nice resolution of the stack singularities of $`_{𝒟,_0}`$. We will thus work with the algebra of functions on the coarse quotient $`Z_{𝒟,_0}`$. However $`Z_{𝒟,_0}`$ is *not* a groupoid, and so to define a convolution algebra from the function algebra $`C_c(Z_{𝒟,_0})`$, we use the trick used by Connes and Marcolli in \[CM04\], 1.83, which consists in introducing $`G()`$. Namely, we introduce the groupoid
$$_{𝒟,_0}K\backslash G(𝔸_f)\underset{𝐾}{\times }(K_M\times \underset{}{\mathrm{lim}}_K^{}G()\backslash G(𝔸)/K^{}),$$
where $`K^{}`$ runs over compact open subgroups of $`G(𝔸_f)`$; and identify $`C_c(Z_{𝒟,_0})`$ with the subalgebra of $`C_c(_{𝒟,})`$ obtained by composing by the projection map $`_{𝒟,_0}Z_{𝒟,_0}`$. Since $`_{𝒟,_0}`$ is a groupoid, convolution can be defined on $`C_c(Z_{𝒟,})`$.
Remark that this solution, even if not completely satisfactory from the geometrical viewpoint (because we work on coarse quotients), suffices (and seems to be necessary) for the physical interpretation, i.e., analysis of KMS states.
#### 4.3.2 BCM algebras
Now we give the precise definition of the algebra alluded to in the previous paragraph. Let $`(𝒟,)=((G,X,V,M)(L,K,K_M))`$ be a BCM pair.
Let
$$(𝒟,):=C_c(Z_{𝒟,})$$
be the algebra of compactly supported continuous functions on $`Z_{𝒟,}`$. As in \[CM04\], p44, in order to define the convolution of two functions, we consider functions on $`Z_{𝒟,}`$ as functions on $`U_{𝒟,}`$ satisfying the following properties:
$$f(\gamma g,y)=f(g,y)f(g\gamma ,y)=f(g,\gamma y),\gamma K,gG(𝔸_f),yY_{𝒟,}.$$
The convolution product on $`(𝒟,)`$ is then defined by the expression
$$(f_1f_2)(g,y):=\underset{hK\backslash G(𝔸_f),hyY_{𝒟,}}{}f_1(gh^1,hy)f_2(h,y),$$
and the adjoint by
$$f^{}(g,y):=\overline{f(g^1,gy)}.$$
The fact that we consider functions with compact support implies that the sum defining the convolution product is finite.
###### Definition 4.3.3.
The algebra $`(𝒟,)`$ (under the convolution product) is called the *Bost-Connes-Marcolli algebra* of the pair $`(𝒟,)`$.
###### Remark 4.3.4.
We proved in Lemma 4.2.5 that, if $`(G,X)`$ is classical, the quotient of $`Y_{𝒟,}`$ by the commensurability equivalence relation (encoded by the action of the groupoid $`Z_{𝒟,}`$) does not depend on the choice of $`K`$. This implies that in the classical case, the Morita equivalence class of $`(𝒟,)`$ is independent of the choice of neat level structure $`K`$. More precisely, all these algebras are in fact Morita equivalent to the algebra corresponding to the “noncommutative quotient”
$$G()\backslash X\times K_𝔸^M,\text{where }K_𝔸^M=G(𝔸_f)K_M\text{.}$$
### 4.4 Time evolution, Hamiltonian and partition function
Let $`(𝒟,)=((G,X,V,M),(L,K,K_M))`$ be a BCM pair with neat level.
###### Definition 4.4.1.
The *time evolution* on $`(𝒟,)`$ is defined by
$$\sigma _t(f)(g,y)=det(\varphi (g))^{it}f(g,y).$$
(4.2)
Let $`y=(\rho ,[z,l])`$ be in $`Y_{𝒟,}`$ and let $`G_y=\{gG(𝔸_f)g\rho K_M\}`$. Let $`_y`$ be the Hilbert space $`\mathrm{}^2(K\backslash G_y)`$.
###### Definition 4.4.2.
The *representation $`\pi _y:(𝒟,)(_y)`$ of the Hecke algebra on $`_y`$* is defined by
$$(\pi _y(f)\xi )(g):=\underset{hK\backslash G_y}{}f(gh^1,hy)\xi (h),gG_y,$$
for $`f(𝒟,)`$ and $`\xi _y`$.
###### Lemma 4.4.3.
The representation $`\pi _y`$ is well defined, i.e., $`\pi _y(f)`$ is bounded for each $`f(𝒟,)`$.
###### Proof.
For $`f(𝒟,)`$, We want to prove that the norm
$$\pi _y(f):=\underset{\xi =1}{sup}\pi _y(f)\xi _2$$
is bounded. This follows from the fact that the functions we consider are with compact support. More precisely, denote $`Z:=Z_{𝒟,}`$. Given $`f(𝒟,)=C_c(Z)`$, we need to show that there is a bound $`C>0`$ such that for every pair of vectors $`\xi ,\eta _y`$ we have
$$|\pi _y(f)\xi ,\eta |C\xi \eta .$$
To this end, we introduce the following notation. We set
$$S_y=\{[gh^1,hy]Zg,hK\backslash G_y\},$$
and for each $`\gamma S_y`$ we set
$$R_y(\gamma )=\{\gamma ^{}Z_ys(\gamma ^{})=t(\gamma )\}.$$
These are discrete sets. Here we use the usual notation for groupoids, namely $`Z_y=t^1\{y\}`$, which we shall identify with $`K\backslash G_y`$.
Using the Cauchy-Schwarz inequality, we now get a bound on $`|\pi _y(f)\xi ,\eta |`$ as follows:
$`|\pi _y(f)\xi ,\eta |`$ $`{\displaystyle \underset{\gamma _1Z_y}{}}\left|\left(\pi _y(f)\xi \right)(\gamma _1)\overline{\eta (\gamma _1)}\right|`$
$`{\displaystyle \underset{\gamma _1,\gamma _2Z_y}{}}\left|f(\gamma _1\gamma _2^1)\xi (\gamma _2)\overline{\eta (\gamma _1)}\right|`$
$`={\displaystyle \underset{\gamma S_y}{}}|f(\gamma )|{\displaystyle \underset{\gamma ^{}R_y(\gamma )}{}}|\xi (\gamma ^{})\eta (\gamma \gamma ^{})|`$
$`{\displaystyle \underset{\gamma S_y}{}}|f(\gamma )|\left({\displaystyle \underset{\gamma ^{}R_y(\gamma )}{}}|\xi (\gamma ^{})|^2\right)^{\frac{1}{2}}\left({\displaystyle \underset{\gamma ^{}R_y(\gamma )}{}}|\eta (\gamma \gamma ^{})|^2\right)^{\frac{1}{2}}`$
$`\xi \eta {\displaystyle \underset{\gamma S_y}{}}|f(\gamma )|.`$
Because $`f`$ has compact support, the sum $`_{\gamma S_y}|f(\gamma )|`$ is finite, and we thereby get the desired bound. ∎
Let $`K_0=\varphi ^1(K_M^\times )`$. We view the Hamiltonian as a virtual operator on $`\mathrm{}^2(K_0\backslash G_y)`$. By this we mean that the Hamiltonian does not depend on the choice of $`K`$ and there is a minimal space on which it is defined: the space $`\mathrm{}^2(K_0\backslash G_y)`$. Consequently, its trace must be computed as a virtual (i.e., equivariant) trace, i.e., must be divided by $`\mathrm{card}(K\backslash K_0)`$. These considerations are related to the fact that, if $`(G,X)`$ is classical, we prefer to define BCM algebras using neat level structures to resolve the stack singularities of $`_{𝒟,}`$.
###### Proposition 4.4.4.
The operator on $`_y`$ given by
$$(H_y\xi )(g)=\mathrm{log}det(\varphi (g))\xi (g)$$
is the *Hamiltonian*, i.e., the infinitesimal generator of the time evolution, meaning that we have the equality
$$\pi _y(\sigma _t(f))=e^{itH_y}\pi _y(f)e^{itH_y}$$
(4.3)
for all $`f(𝒟,)`$.
###### Proof.
This is just a matter of unwinding the definitions. Let $`\xi _y`$, and let $`gG_y`$. On the one hand we have
$`\left(\pi _y(\sigma _tf)\xi \right)(g)`$ $`={\displaystyle \underset{hK\backslash G_y}{}}(\sigma _tf)(gh^1,hy)\xi (h)`$
$`={\displaystyle \underset{hK\backslash G_y}{}}det(\varphi (g))^{it}det(\varphi (h))^{it}f(gh^1,hy)\xi (h),`$
while on the other hand we have
$`\left(e^{itH_y}(\pi _yf)e^{itH_y}\xi \right)(g)`$ $`=det(\varphi (g))^{it}\left((\pi _yf)e^{itH_y}\xi \right)(g)`$
$`=det(\varphi (g))^{it}{\displaystyle \underset{hK\backslash G_y}{}}f(gh^1,hy)(e^{itH_y}\xi )(h)`$
$`=det(\varphi (g))^{it}{\displaystyle \underset{hK\backslash G_y}{}}f(gh^1,hy)det(\varphi (h))^{it}\xi (h).`$
We thereby obtain the desired equality. ∎
###### Definition 4.4.5.
Let $`yY_{𝒟,}`$ and $`\beta >0`$. The *partition function* of the system $`((𝒟,),\sigma _t,_y,H_y)`$, is
$$\zeta _y(\beta ):=\frac{1}{\mathrm{card}(K\backslash K_0)}\mathrm{Trace}(e^{\beta H_y}).$$
Let $`Y_{𝒟,}^\times Y_{𝒟,}`$ be the set of invertible $`y=(\rho ,[z,l])`$, i.e., $`\rho K_M^\times `$.
###### Proposition 4.4.6.
Suppose that $`yY_{𝒟,}^\times `$. Then $`G_y=\mathrm{Sym}_f:=\varphi ^1(K_M)`$. The partition function of the system $`((𝒟,),\sigma _t,_y,H_y)`$, coincides with the zeta function $`\zeta _{𝒟,}(\beta )`$ of $`(𝒟,)`$ (see Definition 4.1.8).
Moreover, it follows from 4.1.7 that the Hamiltonian has positive energy in the representation $`\pi _y`$.
### 4.5 Symmetries
Let $`(𝒟,)=((G,X,V,M),(L,K,K_M))`$ be a BCM pair with neat level. We will denote the center of $`G`$ by $`C`$.
Recall that $`\mathrm{Sym}_f`$ is the semigroup $`\varphi ^1(K_M)`$. For $`m\mathrm{Sym}_f`$ and $`cC()`$, we define
$$\theta _{(m,c)}(f)(g,\rho ,[z,l]):=f(g,\rho \varphi (m),[cz,l]).$$
###### Lemma 4.5.1.
This gives a well defined right action of
$$\mathrm{Sym}(𝒟,):=\mathrm{Sym}_f(𝒟,)\times C()$$
on $`(𝒟,)`$ which moreover commutes with the time evolution.
###### Proof.
The action is well-defined because $`K`$ acts on $`Y_{𝒟,}`$ on the left, while $`\mathrm{Sym}`$ acts on the right. Recalling that the time evolution is given by the formula $`(\sigma _tf)(g,y)=\left(det\varphi (g)\right)^{it}f(g,y)`$, it is clear that the action of $`\mathrm{Sym}`$ commutes with $`\sigma _t`$. ∎
Let $`CK_M`$ be the center of $`K_M`$.
###### Definition 4.5.2.
Let $`\mathrm{Inn}(𝒟,)`$ be the subsemigroup of $`\mathrm{Sym}`$ defined by
$$\mathrm{Inn}(𝒟,):=C()\varphi ^1(CK_M).$$
###### Remark 4.5.3.
There is a (diagonal) inclusions of semigroups
$$\mathrm{Inn}(𝒟,)\mathrm{Sym}(𝒟,).$$
This gives a natural action of $`\mathrm{Inn}(𝒟,)`$ on $`(𝒟,)`$.
###### Definition 4.5.4.
The semigroup $`\mathrm{Out}(𝒟,):=\mathrm{Inn}(𝒟,)\backslash \mathrm{Sym}(𝒟,)`$ is called the *outer symmetry semigroup of the BCM system* $`((𝒟,),\sigma _t)`$.
In practical situations, the following hypotheses will often be fulfilled (see Propositions 7.3.3 and 9.2.1).
###### Definition 4.5.5.
The level structure $`=(L,K,K_M)`$ is called *faithful* if the image $`\varphi (C())`$ of the center of $`G`$ commutes with $`K_M`$, i.e., $`\varphi (C())CK_M`$. The level structure $``$ is called *full* if the natural morphism $`\mathrm{Out}C()\backslash G(𝔸_f)`$ is surjective; if this morphism is an isomorphism, the $``$ is called *fully faithful*.
These symmetries are symmetries up to inner automorphisms.
###### Proposition 4.5.6.
There is a morphism
$$\mathrm{Out}(𝒟,)\mathrm{Out}((𝒟,),\sigma _t)$$
to the quotient of the automorphism group of the BCM system by inner automorphisms of the algebra.
###### Proof.
We have to prove that $`\mathrm{Inn}`$ acts by inner automorphisms. For $`n\mathrm{Inn}`$, we let $`\mu _n`$ be
$$\mu _n(g,y)=1\text{ if }gK.n^1,\mu _n(g,y)=0\text{ if }gK.n^1.$$
We will show that
$$\theta _{(n,n)}(f)=\mu _nf\mu _n^{},$$
i.e., the action of $`\theta _{(n,n)}`$ is given by the inner automorphism corresponding to $`\mu _n`$.
We have, for all $`yY_{𝒟,}`$,
$`(\mu _nf\mu _n^{})(g,y)`$ $`={\displaystyle \underset{hK\backslash G(𝔸_f),hyY}{}}\mu _n(gh^1,hy)(f\mu _n^{})(h,y),`$
$`={\displaystyle \underset{hK\backslash G(𝔸_f),hyY}{}}\mu _n(gh^1,hy){\displaystyle \underset{kK\backslash G(𝔸_f),kyY}{}}f(hk^1,ky)\mu _n^{}(k,y),`$
$`={\displaystyle \underset{h,kK\backslash G(𝔸_f),hy,kyY}{}}\mu _n(gh^1,hy)f(hk^1,ky)\mu _n(k^1,ky).`$
Now, by definition of $`\mu _n`$, the only nontrivial term of this sum is obtained when $`k^1=n^1`$ and $`gh^1=n^1`$, i.e., $`k=n`$ and $`h=ng`$. Since $`n`$ is central,
$`(\mu _nf\mu _n^{})(g,y)`$ $`=f(ngn^1,ny),`$
$`=f(g,n\rho ,[z,ln^1]),`$
$`=f(g,\rho n,[nz,l]),`$
$`=\theta _{(n,n)}(f)(g,y).`$
## 5 Comparison with the original Bost-Connes-Marcolli systems
We want to understand how our systems are related with the usual Bost-Connes-Marcolli systems in the class number one case. These class number one systems are called principal BCM systems. They are directly related to Connes-Marcolli systems defined in \[CM04\].
### 5.1 Principal BCM systems
Let $`(𝒟,)=((G,X,V,M),(L,K,K_M))`$ be a BCM pair with $`(G,X)`$ classical.
Let $`\mathrm{\Gamma }:=G()K`$ and
$$U^{\mathrm{princ}}:=\{(g,\rho ,z)G()\times K_M\times Xg\rho K_M\}.$$
Let $`X^+`$ be a connected component of $`X`$, $`G()^+`$ be $`G()G()^+`$ (where $`G()^+`$ is the identity component of $`G()`$) and $`\mathrm{\Gamma }_+=G()^+K`$. Let
$$U^+:=\{(g,\rho ,z)G()^+\times K_M\times X^+g\rho K_M\}.$$
We have a natural action of $`\mathrm{\Gamma }^2`$ (resp. $`\mathrm{\Gamma }_+^2`$) on $`U^{\mathrm{princ}}`$ (resp. $`U^+`$) given by $`(g,\rho ,z)(\gamma _1g\gamma _2^1,\gamma _2\rho ,\gamma _2z)`$. Let $`_{𝒟,}^{\mathrm{princ}}`$ (resp. $`_{𝒟,}^+`$) be the stacky quotient of $`U^{\mathrm{princ}}`$ (resp. $`U^+`$) by $`\mathrm{\Gamma }^2`$ (resp. $`\mathrm{\Gamma }_+^2`$).
###### Definition 5.1.1.
The stack groupoid $`_{𝒟,}^{\mathrm{princ}}`$ is called the *principal BCM groupoid* for the pair $`(𝒟,)`$.
###### Proposition 5.1.2.
Suppose that the natural map $`\mathrm{\Gamma }G()/G()^+`$ is surjective. Then the natural map
$$_{𝒟,}^+_{𝒟,}^{\mathrm{princ}}$$
is an isomorphism.
###### Proof.
*Surjectivity*: Let $`u=(g,\rho ,z)U^{\mathrm{princ}}`$. We want to show that there exists $`\gamma _1,\gamma _2\mathrm{\Gamma }`$ such that $`(\gamma _1,\gamma _2).u=(\gamma _1g\gamma _2^1,\rho ,\gamma _2z)U^+.`$ There exists $`\gamma _2\mathrm{\Gamma }`$ with $`\gamma _2zX^+`$ because: 1) the definition (3.1.1) of a Shimura datum implies that $`\pi _0(X)`$ is a $`\pi _0(G())`$-homogeneous space; and 2) from our hypothesis and the theorem of real approximation, we get a surjection $`\mathrm{\Gamma }G()/G()^+G()/G()^+.`$ Our hypothesis now implies that there exists $`\gamma _1\mathrm{\Gamma }`$ such that $`\gamma _1g\gamma _2^1G()^+`$. This proves surjectivity.
*Injectivity*: Now suppose that two points $`(g_1,\rho _1,z_1)`$ and $`(g_2,\rho _2,z_2)`$ have the same image in the quotient. Then there exists $`\gamma _1,\gamma _2\mathrm{\Gamma }`$ such that $`(g_1,\rho _1,z_1)=(\gamma _1g_2\gamma _2^1,\gamma _2\rho _2,\gamma _2z_2)`$. Since $`\gamma _2`$ stabilizes $`X^+`$, it is in $`G()^+`$, and therefore also in $`\mathrm{\Gamma }_+`$. This implies that $`\gamma _1`$ is in $`\mathrm{\Gamma }_+`$. This proves injectivity. ∎
We denote by $`h(G,K)`$ the cardinality of the finite set $`G()\backslash G(𝔸_f)/K`$.
###### Proposition 5.1.3.
If $`h(G,K)=1`$ then the principal and the full BCM groupoids are the same, i.e., the natural map
$$_{𝒟,}^{\mathrm{princ}}_{𝒟,}$$
is an isomorphism.
###### Proof.
There is a natural map
$$\begin{array}{cccc}\psi :& (\mathrm{\Gamma }\backslash G())\times K_M\times X& & (K\backslash G(𝔸_f))\times K_M\times G()\backslash (X\times G(𝔸_f))\\ & (g,\rho ,z)& & (g,\rho ,[z,1])\end{array}.$$
The action of $`\gamma _2\mathrm{\Gamma }`$ on the source is given by $`(g,\rho ,z)(g\gamma _2^1,\gamma _2\rho ,\gamma _2z)`$ and on the range by $`(g,\rho ,[z,l])(g\gamma _2^1,\gamma _2\rho ,[z,l\gamma _2^1])`$. Since $`\mathrm{\Gamma }=KG()`$, we have
$$\begin{array}{cc}\psi (\gamma _2(g,\rho ,z))& =(g\gamma _2^1,\gamma _2\rho ,[\gamma _2z,1]),\hfill \\ & =(g\gamma _2^1,\gamma _2\rho ,[z,\gamma _2^1]),\hfill \\ & =\gamma _2\psi (g,\rho ,z).\hfill \end{array}$$
This proves that $`\psi `$, being $`\mathrm{\Gamma }`$-equivariant, induces a well defined map
$$\overline{\psi }:(\mathrm{\Gamma }\backslash G())\underset{\Gamma }{\times }[K_M\times X](K\backslash G(𝔸_f))\underset{𝐾}{\times }[K_M\times G()\backslash (X\times G(𝔸_f))].$$
Let us prove that $`\overline{\psi }`$ is surjective. This will essentially follow from the equalities $`G(𝔸_f)=K.G()=G().K`$ (the class number one hypothesis $`h(G,K)=1`$).
For $`(g,\rho ,[z,l])(K\backslash G(𝔸_f))\underset{𝐾}{\times }[K_M\times G()\backslash (X\times G(𝔸_f))]`$, there exists $`\gamma _2K`$ and $`l_2G()`$ such that $`l=l_2\gamma _2`$. Then, we have the equalities in our quotient space
$$\begin{array}{ccc}(g,\rho ,[z,l])& =& (g,\rho ,[z,l_2\gamma _2])\hfill \\ & =& (g\gamma _2^1,\gamma _2\rho ,[z,l_2])\hfill \\ & =& (g\gamma _2^1,\gamma _2\rho ,[l_2^1z,1]).\hfill \end{array}$$
There exists $`\gamma _1K`$ and $`g_1G()`$ such that $`\gamma _1g_1=g\gamma _2^1`$ and we have the following equalities in our quotient space
$$\begin{array}{ccc}(g,\rho ,[z,l])& =& (g\gamma _2^1,\gamma _2\rho ,[l_2^1z,1])\hfill \\ & =& (\gamma _1g_1,\gamma _2\rho ,[l_2^1z,1])\hfill \\ & =& \psi (g_1,\gamma _2\rho ,l_2^1z).\hfill \end{array}$$
Thus $`\overline{\psi }`$ is surjective.
Now we prove that $`\overline{\psi }`$ is injective. Suppose that
$$\overline{\psi }(g_1,\rho _1,z_1)=\overline{\psi }(g_2,\rho _2,z_2).$$
Then there exists $`\gamma _1K`$, $`\gamma _2K`$, $`\gamma _3G()`$ such that
$$(\gamma _1g_1\gamma _2^1,\gamma _2\rho _1,[\gamma _3z_1,\gamma _3\gamma _2^1])=(g_2,\rho _2,[z_2,1]).$$
This implies $`\gamma _3=\gamma _2`$ and then $`\gamma _2KG()=\mathrm{\Gamma }`$. But we also have $`\gamma _1=g_2\gamma _2g_1^1G()K=\mathrm{\Gamma }`$. This shows that
$$(g_2,\rho _2,z_2)=(\gamma _1g_1\gamma _2^1,\gamma _2\rho _1,\gamma _2z_1)$$
with $`\gamma _1,\gamma _2\mathrm{\Gamma }`$, i.e., $`(g_2,\rho _2,z_2)`$ and $`(g_1,\rho _1,z_1)`$ are the same in $`(\mathrm{\Gamma }\backslash G())\underset{\Gamma }{\times }[K_M\times X]`$. This proves injectivity.
To finish, we prove that the bijection $`\overline{\psi }:_{𝒟,Lc}^{\mathrm{princ}}_{𝒟,}`$ is compatible with the groupoid structures. Let $`Y^{\mathrm{princ}}=K_M\times X`$, and $`Y=K_M\times \mathrm{Sh}(G,X)`$. If $`(g,\rho ,z)^{\mathrm{princ}}`$, the image of $`(\rho ,z)Y^{\mathrm{princ}}`$ under $`gG()`$ is given by $`(g\rho ,gz)Y^{\mathrm{princ}}`$. The image of $`(\rho ,[z,1])Y`$ under $`g`$ is given by $`(g\rho ,[z,g^1])Y`$, which is equal to $`(g\rho ,[gz,1])`$. This finishes the proof. ∎
###### Definition 5.1.4.
Let $`(𝒟,)`$ be a BCM pair with neat level. The algebra $`_{\mathrm{princ}}(𝒟,)=C_c(Z_{𝒟,}^{\mathrm{princ}})`$ is called the *principal BCM algebra* for $`(𝒟,)`$.
### 5.2 The Bost-Connes system
Let $`F/`$ be a number field. Let $`G=\mathrm{Res}_{F/}𝔾_{m,F}`$, $`X_F=G()/G()^+\{\pm 1\}^{\mathrm{Hom}(F,)}`$, $`V=F`$ and $`M=\mathrm{Res}_{F/}\mathrm{M}_{1,F}`$. Let $`K=\widehat{𝒪}_{F}^{}{}_{}{}^{\times }`$, $`L=𝒪_F`$ and $`K_M=\widehat{𝒪}_F=\mathrm{M}_1(\widehat{𝒪}_F)`$.
###### Definition 5.2.1.
The pair
$$𝒫(\mathrm{Res}_{F/}𝔾_{m,F},X_F)=(𝔾_{m,F},X_F,F,\mathrm{Res}_{F/}\mathrm{M}_{1,F}),(𝒪_F,\widehat{𝒪}_{F}^{}{}_{}{}^{\times },\widehat{𝒪}_F))$$
is called the *Bost-Connes pair* for $`F`$. The corresponding algebra $`(\mathrm{Res}_{F/}𝔾_{m,F},X_F)`$ is called the *Bost-Connes algebra* for $`F`$.
###### Proposition 5.2.2.
In the case $`F=`$, $`(𝔾_{m,},\{\pm 1\})`$ is the original Bost-Connes algebra.
###### Proof.
Recall from Section 2.1 that the Bost-Connes algebra is the convolution algebra of the groupoid $`Z_{BC}_+^\times \times \widehat{}`$ of pair $`(g,\rho )`$ with $`g\rho \widehat{}`$; thus we need only to show that $`Z_{BC}`$ coincides with the BCM groupoid $`Z`$ of the Bost-Connes pair. Indeed, in the notation of Section 5.1, we have
$$U^+=\{(g,\rho ,1)_+^\times \times \widehat{}\times \{1\}g\rho \widehat{}\},\mathrm{\Gamma }=\{\pm 1\},\text{ and }\mathrm{\Gamma }_+=1.$$
Therefore $`^+:=\mathrm{\Gamma }^+\backslash U^+=Z_{BC}`$; the map $`\mathrm{\Gamma }G()/G()^+`$ is an isomorphism of $`\{\pm 1\}`$; and $`h(𝔾_{m,},\widehat{}^\times )=1`$, since it is the usual class number of $``$. The proposition follows from Propositions 5.1.2 and 5.1.3. ∎
### 5.3 The Connes-Marcolli system
We now show that in the $`\mathrm{GL}_{2,}`$ case, we obtain exactly the same groupoid as Connes and Marcolli \[CM04\]. This groupoid is only a stack-groupoid, not a usual groupoid. This restriction was circumvented by Connes and Marcolli using functions of weight 0 for the scaling action (see \[CM04\], remark shortly preceeding 1.83). Such a scaling action is not canonically defined in the general case we consider. As explained before, we deliberately chose to view this groupoid as a stack-groupoid in order to define a natural groupoid algebra for it that depends on the resolution of stack singularities given by the choice of $`K`$.
Consider the Shimura datum $`(\mathrm{GL}_{2,},^\pm )`$, $`V=^2`$, and $`M=\mathrm{M}_{2,}`$. let $`L=^2`$, $`K=\mathrm{GL}_2(\widehat{})`$, and $`K_M=\mathrm{M}_2(\widehat{})`$.
###### Definition 5.3.1.
The pair
$$𝒫(\mathrm{GL}_2,^\pm ):=((\mathrm{GL}_2,^\pm ,^2,\mathrm{M}_{2,}),(^2,\mathrm{GL}_2(\widehat{}),\mathrm{M}_2(\widehat{})))$$
is called *the modular BCM pair*. The corresponding BCM stack-groupoid is denoted by $`_{\mathrm{GL}_2,^\pm }`$.
The stack-groupoid $`_{\mathrm{GL}_2,^\pm }^+`$ is defined as in Section 5.1. This is exactly the groupoid studied by Connes and Marcolli in \[CM04\].
###### Lemma 5.3.2.
Our BCM stack-groupoid is the same as Connes and Marcolli’s one. In other words, the natural map
$$_{\mathrm{GL}_2,}^+_{\mathrm{GL}_2,^\pm }$$
is an isomorphism.
###### Proof.
We have in this case $`h(G,K)=1`$ so by Proposition 5.1.3, we have $`[Z_{\mathrm{GL}_2,^\pm }^{\mathrm{princ}}][Z_{\mathrm{GL}_2,^\pm }]`$. The map $`\mathrm{GL}_2()\mathrm{GL}_2()/\mathrm{GL}_2()^+`$ is surjective, so that we can apply proposition 5.1.2, which tells us that $`_{\mathrm{GL}_2,^{pm}}^+_{\mathrm{GL}_2,^\pm }^{\mathrm{princ}}`$. ∎
## 6 Operator theoretic results on BCM algebras
### 6.1 The C\*-algebra associated to a BCM datum
Let $`(𝒟,)=((G,X,V,M),(L,K,K_M))`$ be a BCM pair with neat level. On the algebra $`(𝒟,)`$, we put the following norm: for every $`f(𝒟,)`$,
$$f=\underset{yY_{𝒟,}}{sup}\pi _y(f).$$
###### Lemma 6.1.1.
This defines a C\*-norm on $`(𝒟,)`$, i.e., $`f^{}f=f^2`$.
###### Proof.
Indeed, it is easy to check that this is a seminorm satisfying the C\*-condition (Definition 3.2.1): observe that for arbitarily small $`ϵ>0`$ there is a $`y`$ such that $`f^2ϵ=\pi _y(f)^2`$. We then have
$$f^{}f\pi _y(f^{}f)=\pi _y(f)^2=f^2ϵ,$$
which of course means that $`f^{}ff^2`$. This inequality is easily shown to imply the C\*-condition.
That we get a norm (i.e., $`f=0`$ only when $`f=0`$), and not just a seminorm, follows from the fact that $`f(g,y)0`$ implies that $`\pi _y(f)0`$:
$$\pi _y(f)\epsilon _g,\epsilon _g=f(1,gy)=f(g,y)0.$$
Here $`ϵ_g_y`$ is the unit vector which takes value $`1`$ at $`g`$, and $`0`$ elsewhere. ∎
###### Definition 6.1.2.
The completion of $`(𝒟,)`$ under the norm $`.`$ is denoted $`𝒜(𝒟,)`$ and called the *BCM C\*-algebra*.
### 6.2 Construction of extreme KMS states at small temperature
Let $`(𝒟,)=((G,X,V,M),(L,K,K_M))`$ be a summable BCM pair. Recall that $`Y_{𝒟,}^\times =\{(g,\rho ,[z,l])Y_{𝒟,}\rho \text{ invertible}\}`$.
###### Lemma 6.2.1.
Let $`yY_{𝒟,}^\times `$. Let $`\beta `$ be such that the zeta function $`\zeta _{𝒟,}(\beta )`$ converges. The state
$$\mathrm{\Phi }_{\beta ,y}(f):=\frac{\mathrm{Trace}(\pi _y(f)e^{\beta H_y})}{\zeta _{𝒟,}(\beta )}$$
is a $`\mathrm{KMS}_\beta `$ state for the system $`(𝒜(𝒟,),\sigma _t)`$ (by Lemma 4.1.7, $`\zeta _{𝒟,}(\beta )0`$).
###### Proof.
By construction, the algebra $`(𝒟,)`$ is a norm-dense subalgebra of $`𝒜(𝒟,)`$, which is also $`\sigma _z`$-invariant. Thus, to verify the $`\mathrm{KMS}_\beta `$ condition, it is enough to show that
$$\mathrm{\Phi }_{\beta ,y}(f_1\sigma _{i\beta }(f_2))=\mathrm{\Phi }_{\beta ,y}(f_2f_1)$$
for every pair of functions $`f_1,f_2(𝒟,)`$; see Proposition 3.2.3. The convergence of the zeta function implies that the operator $`e^{\beta H_y}`$ is trace class. The invariance of the trace under cyclic permutations implies that
$$\begin{array}{ccc}\zeta _{𝒟,}(\beta ).\mathrm{\Phi }_{\beta ,y}(f_1\sigma _{i\beta }(f_2))& =& \mathrm{Trace}(f_1e^{\beta H_y}f_2e^{\beta H_y}e^{\beta H_y}),\hfill \\ & =& \mathrm{Trace}(f_1e^{\beta H_y}f_2),\hfill \\ & =& \mathrm{Trace}(f_2f_1e^{\beta H_y}),\hfill \\ & =& \zeta _{𝒟,}(\beta ).\mathrm{\Phi }_{\beta ,y}(f_2f_1),\hfill \end{array}$$
which finishes the proof of the KMS condition. ∎
The commutant of a subset $`S(_y)`$ is by definition $`S^{}=\{a(_y)as=sa,sS\}`$.
###### Lemma 6.2.2.
If $`yY_{𝒟,}^\times `$, then the commutant $`\pi _y(𝒜(𝒟,))^{}`$ consists only of scalar operators.
###### Proof.
In general, if $`yY_{𝒟,}`$, then the Von Neumann algebra $`\pi _y(𝒜)^{}`$ is generated by the right regular representation of the isotropy group $`Z_{y,y}:=\{[g,y]Zs[g,y]=[y]=[gy]=t[g,y]\}`$ (cf. \[Con79\] Proposition VII.5). If $`y`$ is now in $`Y_{𝒟,}^\times `$, then the isotropy group $`Z_{y,y}`$ is trivial. Therefore, the commutant $`\pi _y(𝒜)^{}`$ consists only of scalar operators. ∎
Recall that the set of $`\mathrm{KMS}_\beta `$ states is a convex simplex (see Proposition 3.2.5), whose extreme points are called *extreme $`\mathrm{KMS}_\beta `$ states*.
###### Proposition 6.2.3.
Let $`yY_{𝒟,}^\times `$ be an invertible element of $`Y_{𝒟,}`$. Let $`\beta `$ be such that the zeta function $`\zeta _{𝒟,}(\beta )`$ converges. The $`\mathrm{KMS}_\beta `$ state
$$\mathrm{\Phi }_{\beta ,y}(f):=\frac{\mathrm{Trace}(\pi _y(f)e^{\beta H_y})}{\zeta _{𝒟,}(\beta )}$$
is extremal of type $`\mathrm{I}_{\mathrm{}}`$.
###### Proof.
By Proposition 3.2.5, the property, for $`\mathrm{\Phi }_{\beta ,y}`$, of being extreme is equivalent to the property of being a factor state, i.e., the algebra $`𝒜(𝒟,)`$ generates a factor in the GNS representation of $`\mathrm{\Phi }_{\beta ,y}`$. Following Harari-Leichtnam, \[HL97\], proof of Theorem 5.3.1, the GNS representation is (up to unitary equivalence)
$$\stackrel{~}{\pi }_y=\pi _y\mathrm{id}__y:𝒜(𝒟,)(_y_y),$$
and the associated cyclic vector is
$$\mathrm{\Omega }_{\beta ,y}=\zeta _{𝒟,}(\beta )^{1/2}\underset{hK\backslash G_y}{}det(\varphi (h))^{1/2}ϵ_hϵ_h,$$
where $`ϵ_h`$ is the basis vector of $`_y`$ that takes values $`1`$ at $`h`$, and $`0`$ elsewhere.
The properties that characterize the triple $`(_y_y,\stackrel{~}{\pi }_y,\mathrm{\Omega }_{\beta ,y})`$ as the GNS representation of $`\mathrm{\Phi }_{\beta ,y}`$ are precisely:
1. $`\mathrm{\Phi }_{\beta ,y}(f)=\stackrel{~}{\pi }_y(f)\mathrm{\Omega }_{\beta ,y},\mathrm{\Omega }_{\beta ,y}`$, for every $`f𝒜(𝒟,)`$; and
2. The orbit $`\stackrel{~}{\pi }_y(𝒜(𝒟,))\mathrm{\Omega }_{\beta ,y}`$ is dense in the Hilbert space $`_y_y`$.
These two properties are verified by direct calculation. For example, to verify the second condition first observe that
$$\pi _y(f)ϵ_h=\underset{gK\backslash G_y}{}f(gh^1,hy)ϵ_g,$$
and so
$$\stackrel{~}{\pi }_y(f)\mathrm{\Omega }_{\beta ,y}=\zeta _{𝒟,}(\beta )^{1/2}\underset{g,hK\backslash G_y}{}det(\varphi (h))^{\beta /2}f(gh^1,hy)ϵ_gϵ_h.$$
But since $`G_y=\mathrm{Sym}_f`$, every $`det(\varphi (h))`$ is positive, and we can choose $`f`$ to have sufficiently small support about $`(gh^1,hy)`$ to see that the basis vector $`ϵ_gϵ_h`$ lies in the closure of $`\stackrel{~}{\pi }_y(𝒜(𝒟,))\xi _{\beta ,y}`$.
By Lemma 6.2.2, we know that the commutant $`\pi _y(𝒜(𝒟,))^{}`$ consists of scalar operators. It is then clear that $`\stackrel{~}{\pi }_y(𝒜(𝒟,))^{}=\pi _y(𝒜(𝒟,))^{}(_y)=(_y)`$, and so
$$\stackrel{~}{\pi }_y(𝒜(𝒟,))^{\prime \prime }=(_y)\mathrm{id}__yB(_y).$$
This proves that $`\mathrm{\Phi }_{\beta ,y}`$ is a Type $`\mathrm{I}_{\mathrm{}}`$ factor state. ∎
###### Question 6.2.4.
Let $`(𝒟,)=((G,X,V,M),(L,K,K_M))`$ be a BCM pair. Is is true that for $`\beta >>0`$, the map $`y\mathrm{\Phi }_{\beta ,y}`$ induces a bijection from the Shimura variety $`\mathrm{Sh}(G,X)`$ to the space $`_\beta `$ of extremal $`\mathrm{KMS}_\beta `$ states on $`((𝒟,),\sigma _t)`$?
## 7 A Bost-Connes system for number fields
### 7.1 Reminder of Dedekind zeta functions
A first step in understanding what a good analogue of the Bost-Connes algebra may be is to find a nice description of the partition functions. This was done first by Harari and Leichtnam in \[HL97\] in the class number one case and by Paula Cohen in \[Coh99\] for general number fields, where she used an adelic description of the Dedekind zeta function.
Let $`F`$ be a number field. There is a multiplicative semigroup injection $`𝒪_F^\times 𝒪_F`$. Let $`\widehat{𝒪}_{F}^{}{}_{}{}^{\mathrm{}}:=\widehat{𝒪}_F𝔸_{f,F}^\times `$ and $`\widehat{}^{\mathrm{}}:=\widehat{}𝔸_{f,}^\times `$. Then the space $`\widehat{𝒪}_{F}^{}{}_{}{}^{\times }\backslash \widehat{𝒪}_{F}^{}{}_{}{}^{\mathrm{}}`$ is identified with the multiplicative semigroup $`I_F`$ of integral ideals in $`F`$. The norm map induces a natural map
$$\mathrm{Nm}:\widehat{𝒪}_{F}^{}{}_{}{}^{\times }\backslash \widehat{𝒪}_{F}^{}{}_{}{}^{\mathrm{}}\widehat{}^\times \backslash \widehat{}^{\mathrm{}}^\times \backslash \{0\}^\times .$$
This is the usual norm on ideal classes.
Now the Dedekind zeta function of $`F`$ can be expressed as
$$\zeta _F(s)=\underset{n\widehat{𝒪}_{F}^{}{}_{}{}^{\times }\backslash \widehat{𝒪}_{F}^{}{}_{}{}^{\mathrm{}}}{}\frac{1}{\mathrm{Nm}(n)^s}.$$
### 7.2 The adelic Bost-Connes algebra
We recall from Subsection 5.2 the definition of the Bost-Connes datum for number fields. Let $`F/`$ be a number field. Let $`G=\mathrm{Res}_{F/}𝔾_{m,F}`$ and $`X_F=G()/G()^+\{\pm 1\}^{\mathrm{Hom}(F,)}`$.
Following Definition 5.2.1, the Bost-Connes pair for $`F`$ is
$$𝒫_F:=𝒫(\mathrm{Res}_{F/}𝔾_{m,F},X_F)=((\mathrm{Res}_{F/}𝔾_{m,F},X_F,F=\mathrm{End}_F(F)),(𝒪_F,\widehat{𝒪}_{F}^{}{}_{}{}^{\times },\widehat{𝒪}_F)).$$
Remark that in this case, we have
$$\mathrm{Sh}(\mathrm{Res}_{F/}𝔾_{m,F},X_F)\pi _0(C_F),$$
where $`C_F`$ is the idele class group of $`F`$.
Then $`Y_F=\widehat{𝒪}_F\times \mathrm{Sh}(\mathrm{Res}_{F/}𝔾_{m,F},X_F)`$ and
$$U_F𝔸_{f,F}^\times \times Y_F$$
is the subspace of tuples $`(g,\rho ,[z,l])`$ such that $`g\rho \widehat{𝒪}_F`$. We let $`\gamma _1,\gamma _2\widehat{𝒪}_{F}^{}{}_{}{}^{\times }\times \widehat{𝒪}_{F}^{}{}_{}{}^{\times }`$ act on $`(g,y=(\rho ,[z,l]))U_F`$ by
$$(g,\rho ,[z,l])(\gamma _1g\gamma _2^1,\gamma _2\rho ,[z,l\gamma _2^1]).$$
Let $`Z_F:=(\widehat{𝒪}_{F}^{}{}_{}{}^{\times }\times \widehat{𝒪}_{F}^{}{}_{}{}^{\times })\backslash U_F`$ and let $`_F=(\mathrm{Res}_{F/}𝔾_{m,F},X_F):=C_c(Z_F)`$ be the corresponding Bost-Connes algebra for $`F`$.
### 7.3 Partition function and symmetries
###### Lemma 7.3.1.
Let $`yY_F`$ and $`_y=\mathrm{}^2(K\backslash G_y)`$. The time evolution on $`_F`$ is given by
$$\sigma _t(f)(g,y)=\mathrm{Nm}(g)^{it}f(g,y).$$
The Hamiltonian $`H_y`$ in $`_y`$ is given by
$$(H_y\xi )(g)=\mathrm{log}(\mathrm{Nm}(g))\xi (g).$$
###### Proof.
Notice that for $`aF`$, the determinant of the $``$-linear map $`xa.x`$ on $`F`$ is the norm $`\mathrm{Nm}(a)`$. The lemma then follows from Definitions 4.4.1 and 4.4.4. ∎
###### Lemma 7.3.2.
The finite symmetry semigroup $`\mathrm{Sym}_f(\mathrm{Res}_{F/}𝔾_{m,F},X_F)`$ of $`𝒫_F`$ is $`\widehat{𝒪}_{F}^{}{}_{}{}^{\mathrm{}}`$. Its zeta function $`\zeta _{𝒫_F}`$ is the Dedekind zeta function $`\zeta _F`$ of $`F`$.
###### Proof.
The description of the symmetry semigroup follows from its Definition 4.1.6. The description of the zeta function follows from Definition 4.1.8 and Subsection 7.1. ∎
We now identify the action of the full symmetry semigroup $`\mathrm{Sym}(\mathrm{Res}_{F/}𝔾_{m,F},X_F)=\widehat{𝒪}_{F}^{}{}_{}{}^{\mathrm{}}\times \mathrm{Res}_{F/}𝔾_{m,F}()`$, which contains archimedean information.
###### Proposition 7.3.3.
We have $`\mathrm{Inn}(\mathrm{Res}_{F/}𝔾_{m,F},X_F)=𝒪_F^{\mathrm{}}:=𝒪_F\{0\}`$ and the outer symmetry semigroup $`\mathrm{Out}(\mathrm{Res}_{F/}𝔾_{m,F},X_F)`$ acts on the BCM algebra $`_F`$ through
$$\pi _0(F^\times \backslash 𝔸_F^\times ).$$
###### Proof.
Recall from Eq. 4.1 that $`_F`$ can be written as a projective limit of groupoids $`_K^{}`$ for $`K^{}G(𝔸_f)`$ compact open. The $`\mathrm{Sym}`$-action can thus be enhanced to an action of the projective limit semigroup $`\underset{}{\mathrm{lim}}_K^{}\mathrm{Sym}/K^{}`$ over all compact open $`K^{}G(𝔸_f)`$. We know from \[Del79\], 2.2.3, that
$$\underset{}{\mathrm{lim}}_K^{}F^\times \backslash (𝔸_{f,F}^\times /K^{}\times \pi _0(\mathrm{Res}_{F/}𝔾_{m,F}())):=\mathrm{Sh}(\mathrm{Res}_{F/}𝔾_{m,F},X_F)\pi _0(F^\times \backslash 𝔸_F^\times ).$$
It thus remains to prove that the natural map
$$𝒪_F^{\mathrm{}}\backslash (\widehat{𝒪}_{F}^{}{}_{}{}^{\mathrm{}}\times \pi _0(\mathrm{Res}_{F/}𝔾_{m,F}()))F^\times \backslash (𝔸_{f,F}^\times \times \pi _0(\mathrm{Res}_{F/}𝔾_{m,F}()))$$
is an isomorphism. The injectivity of this map is clear because $`𝒪_F^{\mathrm{}}:=𝒪_F\{0\}=\widehat{𝒪}_{F}^{}{}_{}{}^{\mathrm{}}F^\times `$. Since $`F^\times `$ acts transitively on $`\pi _0(\mathrm{Res}_{F/}𝔾_{m,F}())`$, to prove surjectivity it suffices to prove surjectivity of the upper map of the following diagram:
The lower arrow is an isomorphism because these two groups are equal to the ideal class group of $`F`$. Let $`g𝔸_{f,F}^\times `$ be a finite idele. Then its image by the vertical projection gives an ideal class, which is the image of some $`m\widehat{𝒪}_{F}^{}{}_{}{}^{\mathrm{}}`$. We have $`[m]=[g]`$ in the right quotient so that there exists $`k\widehat{𝒪}_{F}^{}{}_{}{}^{\times }`$ such that $`g=mkmodF^\times `$. Then $`mk\widehat{𝒪}_{F}^{}{}_{}{}^{\mathrm{}}`$ is in the preimage of the upper arrow of the diagram, which proves surjectivity. ∎
###### Remark 7.3.4.
Analogous results were already obtained for $`F`$ imaginary quadratic by Connes-Marcolli-Ramachandran (see \[CMR05\]). In this case, the datum $`(\mathrm{Res}_{F/}𝔾_{m,F},X_F)`$ is classical so that the system is simpler.
## 8 A Bost-Connes system for Dirichlet characters
### 8.1 Reminder of zeta functions of Dirichlet characters
We here recall from Neukirch’s book \[Neu92\], p. 501, some facts about characters.
###### Definition 8.1.1.
A *Hecke character* is a character of the idele class group $`C_F:=𝔸_F^\times /F^\times `$, i.e., a continuous homomorphism $`\chi :C_FS^1`$ to the group $`S^1`$ of complex numbers of norm $`1`$. A *Dirichlet character* is a Hecke character that factors through the quotient group $`(F_{})_+^\times \backslash 𝔸_F^\times /F^\times `$ where $`+`$ denotes the connected component for the real topology.
Let $`𝔪=_𝔭𝔭^n`$ be a full ideal of $`𝒪_F`$ and let $`K(𝔪)`$ be the kernel of the natural map
$$\widehat{𝒪}_{F}^{}{}_{}{}^{\times }(\widehat{𝒪}_F/𝔪)^\times .$$
We say that $`𝔪`$ is *a module of definition* for the Dirichlet character $`\chi `$ if $`\chi (K(𝔪))=1`$. We then call $`K(𝔪)`$ *a subgroup of definition* for $`\chi `$.
Each Dirichlet character has a module of definition and for such an $`𝔪`$, we have a factorisation $`\chi :C(𝔪)S^1`$, where $`C(𝔪)=((F_{})_+^\times \times K(𝔪))\backslash 𝔸_F^\times /F^\times `$ is the big ray class group modulo $`𝔪`$. Such an $`𝔪`$ that is moreover minimal (among the modules of definition) is called *the conductor of the Dirichlet character*.
Recall that $`\widehat{𝒪}_{F}^{}{}_{}{}^{\mathrm{}}=𝔸_{f,F}^\times \widehat{𝒪}_F`$. If $`\chi :𝔸_F^\times S^1`$ is a Dirichlet character, we factor it through $`(F_{})_+^\times \backslash 𝔸_F^\times `$, and thus restrict it to $`\pi _0(F_{}^\times )\times \widehat{𝒪}_{F}^{}{}_{}{}^{\mathrm{}}`$. Let $`K(𝔪)\widehat{𝒪}_{F}^{}{}_{}{}^{\times }`$ be a primitive subgroup of definition for $`\chi `$ and let $`K^{\mathrm{}}(𝔪):=\{n\widehat{𝒪}_{F}^{}{}_{}{}^{\mathrm{}}\overline{n}=1\widehat{𝒪}_F/𝔪\}`$.
There is an injective map $`K(𝔪)\backslash K^{\mathrm{}}(𝔪)\widehat{𝒪}_{F}^{}{}_{}{}^{\times }\backslash \widehat{𝒪}_{F}^{}{}_{}{}^{\mathrm{}}`$ whose image is the semigroup of all ideals of $`F`$ prime to $`𝔪`$.
At least if $`\chi `$ is trivial at infinity, it induces $`\chi :K(𝔪)\backslash K^{\mathrm{}}(𝔪)S^1.`$ Now, we can define the $`L`$-function of our Dirichlet character $`\chi `$ as
$$L_F(s,\chi )=\underset{nK(𝔪)\backslash K^{\mathrm{}}(𝔪)}{}\frac{\chi (n)}{\mathrm{Nm}(n)^s},$$
where $`\mathrm{Nm}`$ was defined in section 7.1. In the particular case of a class character, we have
$$L_F(s,\chi )=\underset{n\widehat{𝒪}_{F}^{}{}_{}{}^{\times }\backslash \widehat{𝒪}_{F}^{}{}_{}{}^{\mathrm{}}}{}\frac{\chi (n)}{\mathrm{Nm}(n)^s}.$$
### 8.2 A Bost-Connes algebra for Dirichlet characters
Let $`\chi :𝔸_FS^1`$ be a Dirichlet character that is supposed to be trivial at infinity. Let $`G=\mathrm{Res}_{F/}𝔾_{m,F}`$ and $`X:=G()/G()^+\{\pm 1\}^{\mathrm{Hom}(F,)}`$. Let $`𝔪`$ be the conductor of $`\chi `$ and $`K_M(𝔪)\widehat{𝒪}_F`$ be the multiplicative semigroup defined by
$$K_M(𝔪)=\mathrm{Ker}_{\mathrm{mult}}(\widehat{𝒪}_F\widehat{𝒪}_F/𝔪):=\{n\widehat{𝒪}_F\overline{n}=1\widehat{𝒪}_F/𝔪\}.$$
Recall that we denoted $`K(𝔪)\widehat{𝒪}_{F}^{}{}_{}{}^{\times }`$ the subgroup $`K(𝔪)=\mathrm{Ker}(\widehat{𝒪}_{F}^{}{}_{}{}^{\times }(\widehat{𝒪}_F/𝔪)^\times ).`$ Let $`L=𝒪_F`$ and $`\varphi :G\mathrm{GL}_{}(F)`$ be the regular representation.
###### Definition 8.2.1.
The tuple $`𝒟_{F,𝔪}:=((\mathrm{Res}_{F/}𝔾_{m,F},X,K(𝔪)),(K_M(𝔪),\varphi ,L))`$ is called the *Bost-Connes datum of conductor $`𝔪`$*.
The time evolution and Hamiltonian are the same as in the Bost-Connes case studied in Subsection 7.3.
Let $`a_\chi `$ be the operator on $`_y`$ defined by
$$(a_\chi \xi )(g)=\chi (g).\xi (g).$$
###### Definition 8.2.2.
The *$`\chi `$-twisted trace* $`\mathrm{Trace}_\chi `$ on $`(_y)`$ is defined by
$$\mathrm{Trace}_\chi (D)=\mathrm{Trace}(a_\chi .D).$$
###### Definition 8.2.3.
The *$`\chi `$-twisted partition function* of $`𝒟_{F,𝔪}`$ is defined as
$$\zeta _{𝒟_{F,𝔪},\chi }(s)=\mathrm{Trace}_\chi (e^{\beta H_y}).$$
###### Lemma 8.2.4.
The $`\chi `$-twisted partition function of $`𝒟_{F,𝔪}`$ is equal to the Dirichlet $`L`$-function $`L_F(s,\chi )`$.
###### Proof.
This follows from the definition and Subsection 8.1. ∎
Notice that in this case, the symmetry semigroup is not full in the sense of Definition 4.5.5.
###### Remark 8.2.5.
If we want to treat Dirichlet characters with nontrivial infinite component, it could be useful to construct the groupoid given by the partial action of $`𝔸_F^\times `$ on the space $`𝔸_F\times \pi _0(C_F)`$ where $`C_F:=F^\times \backslash 𝔸_F^\times `$. If we do the construction as before, using a quotient by $`(\widehat{𝒪}_{F}^{}{}_{}{}^{\times })^2`$, the partition function will not be reasonable. It could be interesting to use Tate’s thesis \[Tat67\], that expresses the Dedekind zeta function as an integral, to deal with this problem. It is not clear to us if a meaningful physical system can be constructed this way.
## 9 The Hilbert modular BCM system
We now specialize the general formalism of Section 4 to the case of Hilbert modular Shimura data. This is a good training ground for the case of a general Shimura datum.
### 9.1 Construction
Let $`F`$ be a totally real number field. Let $`G:=\mathrm{Res}_{F/}\mathrm{GL}_2`$, $`X=(^\pm )^{\mathrm{Hom}(F,)}`$. The Shimura datum $`(G,X)`$ is the *Hilbert modular Shimura datum*. Let $`V`$ be the $``$-vector space $`F^2`$ with the natural action $`\varphi `$ of $`G`$. Let $`M:=\mathrm{Res}_{F/}\mathrm{M}_{2,F}`$. Let $`LV`$ be $`𝒪_F^2`$. Let $`K_0=\mathrm{GL}_2(\widehat{𝒪}_F)G(𝔸_f)`$ and $`K_M=\mathrm{M}_2(\widehat{𝒪}_F)\mathrm{M}_2(𝔸_{f,F})`$. Choose a neat subgroup $`KK_0`$.
###### Definition 9.1.1.
The pair $`𝒫(G,X,K):=((G,X,V,M),(L,K,K_M))`$ is called a *Hilbert modular BCM pair* for $`F`$. The BCM algebra $`(𝒫)`$ is called a *Hilbert modular BCM algebra*.
###### Lemma 9.1.2.
If we suppose that $`F`$ has class number one, then the natural morphism
$$(\mathrm{GL}_{2,F},X,K)_{\mathrm{princ}}(\mathrm{GL}_{2,F},X,K)$$
from the principal to the full Bost-Connes-Marcolli algebra is an isomorphism.
###### Proof.
The hypothesis implies (in fact is equivalent to) $`h(G,K)=1`$. The result then follows from proposition 5.1.3. ∎
We now describe more explicitly the time evolution whose construction was made in Subsection 4.4.
Let $`C:=\mathrm{Res}_{F/}𝔾_m`$, which is the center of $`G=\mathrm{Res}_{F/}\mathrm{GL}_2`$. The natural determinant map $`det:GC`$ induces $`det:K\backslash G(𝔸_f)C(\widehat{})\backslash C(𝔸_f)`$. The norm map $`\mathrm{Nm}:C𝔾_{m,}`$ induces
$$\mathrm{Nm}:C(\widehat{})\backslash C(𝔸_f)\widehat{}^\times \backslash 𝔸_f^\times ^\times \backslash ^\times _+^\times _+^\times .$$
###### Lemma 9.1.3.
The time evolution on the Hilbert modular BCM algebra $`(G,X,K)`$ is equal to
$$\sigma _t(f)(g,y)=\mathrm{Nm}(det(g))^{it}f(g,y).$$
### 9.2 Symmetries
We apply the general definitions of Subsection 4.5 to this case. We see that
$$\mathrm{Sym}_f=\mathrm{M}_2(\widehat{𝒪}_F)^{\mathrm{}}:=\mathrm{GL}_2(𝔸_{f,F})\mathrm{M}_2(\widehat{𝒪}_F).$$
The center of $`G=\mathrm{Res}_{F/}\mathrm{GL}_{2,F}`$ is $`C=\mathrm{Res}_{F/}𝔾_m`$ and the center of $`\mathrm{M}_2(\widehat{𝒪}_F)`$ is $`\widehat{𝒪}_F`$ as a diagonal subsemigroup. We also have $`\mathrm{Inn}=𝒪_F^{\mathrm{}}:=𝒪_FF^\times `$ and an inclusion of semigroups $`𝒪_F^{\mathrm{}}\mathrm{M}_2(\widehat{𝒪}_F)^{\mathrm{}}.`$
The following lemma explains what the symmetries are in the case of Hilbert modular BCM systems.
###### Proposition 9.2.1.
The outer symmetry semigroup $`\mathrm{Out}`$ of the Hilbert modular BCM system is isomorphic to $`F^\times \backslash \mathrm{GL}_2(𝔸_{f,F})\times \mathrm{Res}_{F/}𝔾_{m,F}()`$, more precisely, the natural map
$$\mathrm{Sym}_f:=𝒪_F^{\mathrm{}}\backslash \mathrm{M}_2(\widehat{𝒪}_F)^{\mathrm{}}F^\times \backslash \mathrm{GL}_2(𝔸_{f,F})$$
is an isomorphism.
###### Proof.
The injectivity of this map is clear because,
$$𝒪_F^{\mathrm{}}=F^\times \mathrm{M}_2(\widehat{𝒪}_F)^{\mathrm{}}.$$
Let $`(\mathrm{M}_2(\widehat{𝒪}_F)^{\mathrm{}})^1=\{mG(𝔸_f)m^1\mathrm{M}_2(\widehat{𝒪}_F)\}`$ be the semigroup of inverses of elements in $`\mathrm{M}_2(\widehat{𝒪}_F)^{\mathrm{}}`$. We then have
$$\mathrm{M}_2(\widehat{𝒪}_F)^{\mathrm{}}.(\mathrm{M}_2(\widehat{𝒪}_F)^{\mathrm{}})^1=\mathrm{GL}_2(𝔸_{f,F}).$$
Let $`m𝒪_F^{\mathrm{}}\backslash \mathrm{M}_2(\widehat{𝒪}_F)^{\mathrm{}}`$. We only need to prove that $`m^1𝒪_F^{\mathrm{}}\backslash \mathrm{M}_2(\widehat{𝒪}_F)^{\mathrm{}}`$. Moreover, to invert a matrix it is enough to prove that its determinant is invertible. We have $`det(m)𝒪_F^{\mathrm{}}\backslash \widehat{𝒪}_{F}^{}{}_{}{}^{\mathrm{}}`$. The nonarchimedean part of Proposition 7.3.3 gives $`𝒪_F^{\mathrm{}}\backslash \widehat{𝒪}_{F}^{}{}_{}{}^{\mathrm{}}F^\times \backslash 𝔸_{f,F}^\times `$, which implies that $`det(m)^1𝒪_F^{\mathrm{}}\backslash \widehat{𝒪}_{F}^{}{}_{}{}^{\mathrm{}}𝒪_F^{\mathrm{}}\backslash \mathrm{M}_2(\widehat{𝒪}_F)^{\mathrm{}}`$. This finishes the proof. ∎
## Appendix A Stack groupoids
### A.1 Topological stacks and stack groupoids
We refer to \[LMB00\] for the theory of stacks and to \[Noo05\] for the theory of topological stacks.
A topological stack will be for us a stack on the site (Top) of topological spaces with usual open coverings, i.e., a category fibered in groupoids fulfilling some descent condition (which is precisely described in \[LMB00\], Définition 3.1):
* isomorphisms between two given objects form a sheaf,
* every descent condition with respect to an open covering is effective.
We remark that B. Noohi gave in \[Noo05\] a more restrictive condition (saying that the stack admits a covering by a topological space that is some kind of local fibration), but since we do not use fine stacky geometry, we will ignore this.
Now we would like to define a stack groupoid. Since both groupoids and stacks are categories, one needs the language of 2-categories to describe stack groupoids. We could do this in a way analogous to the spaces in groupoids (“espaces en groupoïdes”) of \[LMB00\], 2.4.3. The theory of Picard stacks, exposed in \[SGA73\], EXP. XVIII, 1.4 and in \[LMB00\], 14.4, is also an inspiring reference. Our references for the definition of a weak 2-category are Kapranov-Voevodsky \[KV94\], Tamsamani’s thesis \[Tam95\], 1.4, and Simpson \[Sim97\].
Roughtly speaking, a *stack groupoid* is a groupoid in the category of topological stacks, i.e., the datum of a tuple $`(𝔛_1,𝔛_0,s,t,ϵ,m)`$ composed of two stacks $`𝔛_1`$ and $`𝔛_0`$, equiped with 1-morphisms source $`s:𝔛_1𝔛_0`$, target $`t:𝔛_1𝔛_0`$, unit $`ϵ:𝔛_0𝔛_1`$, and composition $`m:𝔛_1\underset{s,𝔛_0,t}{\times }𝔛_1𝔛_1`$:
$$\text{},𝔛_1\underset{s,𝔛_0,t}{\times }𝔛_1\stackrel{𝑚}{}𝔛_1.$$
The 1-morphism
$$(\mathrm{Id}_{𝔛_1}\times m):𝔛_1\underset{𝔛_0}{\times }𝔛_1𝔛_1\underset{𝔛_0}{\times }𝔛_1,$$
that sends morally a pair $`(a,b)`$ of composable morphisms to the pair $`(a,ab)`$, is supposed to be an equivalence (which implies the existence of an inverse for the composition law). This tuple should be equiped with the additional data of an associator
$$\mathrm{\Phi }:m(m\times \mathrm{Id}_{𝔛_1})\stackrel{}{}m(\mathrm{Id}_{𝔛_1}\times m),$$
and two unity constraints
$$U:m(\mathrm{Id}_{𝔛_1}\times ϵ)\stackrel{}{}\mathrm{Id}_{𝔛_1}\text{ and }V:m(ϵ\times \mathrm{Id}_{𝔛_1})\stackrel{}{}\mathrm{Id}_{𝔛_1},$$
fulfilling some higher coherence (or cocycle) conditions: pentagon, …
Rather than writing explicitly the coherence conditions, we prefer to use Toen’s viewpoint of Segal groupoid stacks, which allows one to forget these conditions by including them in the choice of inverses for some equivalences in a simplicial diagram.
###### Definition A.1.1.
Let $`(𝔛_1,𝔛_0,s,t,ϵ,m)`$ be a tuple as before. Its coarse quotient is by definition the quotient of the coarse moduli space $`|𝔛_0|`$ (space of isomorphism classes of objects in $`𝔛_0`$) by the equivalence relation generated by
$$x_0x_0^{}x_1|𝔛_1|\text{ such that }s(x_1)=x_0\text{ and }t(x_1)=x_0^{}.$$
### A.2 Groupoids in the category of spaces with group operations
Let (OSpace) be the category of “spaces with group operation”, i.e., pairs $`(G,X)`$ composed of a topological space $`X`$ and a group $`G`$ that acts on $`X`$. A morphism $`\varphi =(\varphi _1,\varphi _2):(G_1,X_1)(G_2,X_2)`$ between two such pairs is a pair composed of a group morphism $`\varphi _1:G_1G_2`$ and a space morphism $`\varphi _2:X_1X_2`$ such that
$$\varphi _2(g_1.x_1)=\varphi _1(g_1).\varphi _2(x_1),(g_1,x_1)G_1\times X_1.$$
One can define the notion of groupoid in the category (OSpace). This is the datum of a tuple $`((G_1,X_1),(G_0,X_0),s,t,ϵ,m)`$ fulfilling some natural conditions that we will not write explicitly here, because we prefer the geometrical language of stacks. There is a relation between these two languages, which is given by a natural functor called “stacky quotient”. Thus one can naturally associate to a groupoid in (OSpace) a stack groupoid.
The reason for introducing the category (OSpace) is to provide an economical description of the notion of quotient of a groupoid by a group action as a stack groupoid.
###### Example A.2.1.
Let $`(𝒟,)`$ be a BCM pair and let $`(U_{𝒟,},Y_{𝒟,},s,t,ϵ,m)`$ be the groupoid defined in Subsection 4.2. There is a natural action of $`K^2`$ on the groupoid $`U_{𝒟,}`$, given by
$$(g,y)(\gamma _1g\gamma _2^1,\gamma _2y).$$
There is also a natural action of $`K`$ on $`Y_{𝒟,}`$ given by
$$y\gamma y.$$
Let $`s_K:K^2K,(\gamma _1,\gamma _2)\gamma _2`$ and $`t_K:K^2K,(\gamma _1,\gamma _2)\gamma _1`$ be the two projections. Then the morphisms in (OSpace) given by $`(s,s_K),(t,t_K):(K^2,U_{𝒟,})(K,Y_{𝒟,})`$ are called the equivariant source and target respectively. The fiber product
$$(K^2,U_{𝒟,})\underset{(s,s_K),(K,Y_{𝒟,}),(t,t_K)}{\times }(K^2,U_{𝒟,})$$
is naturally isomorphic to the OSpace
$$(K^2\underset{s_K,K,t_K}{\times }K^2,U_{𝒟,}\underset{s,Y_{𝒟,},t}{\times }U_{𝒟,}).$$
It means that $`m`$ induces a natural multiplication map
$$m_e=(m_K,m):(K^2,U_{𝒟,})\underset{(s,s_K),(K,Y_{𝒟,}),(t,t_K)}{\times }(K^2,U_{𝒟,})(K^2,U_{𝒟,})$$
in this equivariant setting. The map
$$m_K:K^2\underset{s_K,K,t_K}{\times }K^2K^2$$
is given by $`m(\gamma _1,\gamma _2,\gamma _2,\gamma _3)=(\gamma _1,\gamma _3)`$.
Passing to the stacky quotient, we obtain the multiplication map
$$m:\underset{𝔖}{\times }$$
on the stack groupoid of Subsection 4.2. Remark that if the action of $`K`$ on $`Y_{𝒟,}`$ is free, then we obtain a standard groupoid.
### A.3 Toen’s Segal groupoid stacks
Following the referee’s suggestion, we give the general definition of a stack groupoid. B. Toen proposed us an elegant definition of the 1-category of stack groupoids which has the advantage of being very concise. It is based on the simplicial point of view of 2-categories as explained in Tamsamani’s thesis \[Tam95\] and Simpson \[Sim97\]. Analogous constructions can also be found in \[TV\], 1.3.4 and \[Lei00\].
Recall that $`\mathrm{\Delta }`$ is the category whose objects are totally ordered sets $`[n]=\{0,\mathrm{},n\}`$ and whose morphisms are increasing maps.
The category of topological stacks can be seen as a 1-category (Stacks) with a notion of equivalences.
###### Definition A.3.1.
A *Segal stack category* is a simplicial stack $`𝔛_{}:\mathrm{\Delta }^{op}\text{(Stacks)}`$ such that the *Segal morphisms*
$$𝔛_n𝔛_1\underset{𝔛_0}{\times }\mathrm{}\underset{𝔛_0}{\times }𝔛_1$$
(given by the $`n`$ morphisms in $`\mathrm{\Delta }`$, $`[1][n]`$ that send $`0`$ to $`i`$ and $`1`$ to $`i+1`$) are stack equivalences. A Segal stack category is a *Segal stack groupoid* if the *right multiplication morphism*
$$𝔛_2𝔛_1\underset{𝔛_0}{\times }𝔛_1$$
given by the two morphisms in $`\mathrm{\Delta }`$
* $`[1][2]`$ such that $`00`$, $`11`$ and
* $`[1][2]`$ such that $`00`$, $`12`$,
is a stack equivalence.
Remark that if we replace the category (Stacks) by the category of sets, we find back the usual notions of category and groupoid.
The stacks $`𝔛_n`$ can be thought as families of $`n`$ composable morphisms and the groupoid condition is that the map $`(a,b)(a,ab)`$ is an equivalence, which implies that each $`a`$ is an isomorphism.
The source and target maps $`s,t:𝔛_1𝔛_0`$ are induced by the morphisms $`s=[0][1]:00`$ and $`t=[0][1]:01`$.
The choice of an inverse $`\varphi `$ for the Segal morphism $`𝔛_2𝔛_1\underset{𝔛_0}{\times }𝔛_1`$ allows one to define a *composition* $`\mu :𝔛_1\underset{s,𝔛_0,t}{\times }𝔛_1𝔛_2𝔛_1`$ given by composing $`\varphi `$ with the morphism induced by $`[1][2]:00,12`$.
The increasing map $`[1][0]:00,10`$ induces a map $`ϵ:𝔛_0𝔛_1`$ called the unit map.
Now, choose an inverse $`\psi `$ to the right multiplication morphism
$$𝔛_2𝔛_1\underset{𝔛_0}{\times }𝔛_1,$$
and compose it with
$$\begin{array}{c}\mathrm{Id}_{𝔛_1}\times _{𝔛_0}ϵ:𝔛_1𝔛_1\times _{𝔛_0}𝔛_1\text{and}\\ d_2:𝔛_2𝔛_1,\end{array}$$
where $`d_2`$ is induced by the map $`[1][2]:01,12`$. Morally, these successive maps send an arrow $`a𝔛_1`$ to the pair $`(a,1)`$, then to $`(a,a^1)`$ and finally to $`a^1`$. Let $`i:𝔛_1𝔛_1`$ be this composition.
To conclude, up to the two additional choices of $`\varphi `$ and $`\psi `$, we have obtained a tuple $`(𝔛_1,𝔛_0,s,t,ϵ,i,m)`$ giving a diagram
and a multiplication $`m:𝔛_1\underset{s,𝔛_0,t}{\times }𝔛_1𝔛_1`$, which is the basic datum necessary to define any notion of stack groupoid. The problem is now to define an associativity 2-isomorphism and some other 2-conditions, and to find the right notion of coherence conditions for them. The point of Toen’s construction is that these coherence conditions are already encoded in the simplicial structure. Let’s be more explicit.
First remark that since we work in a 2-category, the inverse of a 1-isomorphism is supposed to be defined up to a *unique* 2-isomorphism. This implies that the multiplication
$$m:𝔛_1\underset{𝔛_0}{\times }𝔛_1𝔛_1$$
is well-defined up to a unique 2-isomorphism.
To define the associator, we use the following strictly commutative diagrams (products are done over $`𝔛_0`$)
and
whose vertical arrows are equivalences.
The uniqueness of inverses of equivalences up to unique 2-isomorphisms gives natural 2-isomorphisms between the multiplication maps
$$(𝔛_1\times 𝔛_1)\times 𝔛_1𝔛_1$$
and
$$𝔛_1\times (𝔛_1\times 𝔛_1)𝔛_1$$
and the morphism obtained by choosing an inverse of the equivalence
$$𝔛_3𝔛_1\times 𝔛_1\times 𝔛_1.$$
This gives the associator 2-isomorphism. To check that the associator fulfils the desired 2-cocycle condition (pentagon), it is necessary to use the simplicial diagram up to $`𝔛_4`$. An explanation of the argument is given in \[Lei00\].
## Appendix B Enveloping semigroups
In this appendix, we explain what enveloping semigroups are, as they are a key ingredient in our formalism (cf. 4.1).
### B.1 Drinfeld’s classification
All groups will be over a field of characteristic $`0`$. Recall from Subsection 4.1 the following definition.
###### Definition B.1.1.
Let $`G`$ be reductive group over a field. An *enveloping semigroup* for $`G`$ is a multiplicative semigroup $`M`$ such that $`M^\times =G`$, $`M`$ is irreducible and normal.
Such semigroups were classified by V. Drinfeld in private notes \[Dri\] that were kindly given to us by L. Lafforgue. Suppose that the base field is algebraically closed. Choose a maximal torus $`TG`$ and a Borel subgroup $`BT`$. Let $`W`$ denote the Weyl group for $`(G,B,T)`$, and let $`X=\mathrm{Hom}(T,𝔾_m)`$.
###### Theorem B.1.2 (Drinfeld).
There exists a bijection between
1. the set of normal affine irreducible semigroups $`M`$ containing $`G`$ as their group of units, and
2. the set of $`W`$-invariant rational polyhedral convex cones $`KX_{}`$ which contain zero and are non-degenerate, i.e., not contained in a hyperplane.
This classification implies that a semisimple group $`G`$ has only one enveloping semigroup, namely $`G`$ itself. This case is for us not very interesting (because a BCM system with such an enveloping semigroup has a trivial zeta function) and we would like to construct more interesting semigroups, in particular, we would like to construct cartesian diagrams
for some fixed representation $`\varphi :G\mathrm{GL}(V)`$.
For example, for an adjoint Shimura datum $`(G,X)`$ (i.e., $`ZG=\{1\}`$), the triviality of the enveloping semigroup implies that the BCM systems we construct have a trivial partition function. It is then interesting to construct another Shimura datum with adjoint datum $`(G,X)`$ and such that the enveloping semigroup is no longer trivial.
There is a natural method due to Vinberg to “enlarge the center” of a given semisimple simply connected group $`G`$ in order to have an enveloping semigroup that is universal in a certain sense. For the sake of brevity, we do not discuss this construction here.
### B.2 Ramachandran’s construction of Chevalley semigroups
There is another way to construct enveloping semigroups quite explicitly, which was communicated to us by N. Ramachandran. The construction of N. Ramachandran uses the following theorem of Chevalley (see \[Spr98\], 5.1).
###### Theorem B.2.1 (Chevalley).
Let $`G`$ be an algebraic group over $``$, and let $`\varphi :G\mathrm{GL}(V)`$ be a faithful representation of $`G`$. There is a tensor construction $`T^{i,j}:=V^iV^{,j}`$ and a line $`DT^{i,j}`$ such that $`\varphi (G)\mathrm{GL}(V)`$ is the stabilizer of this line.
###### Definition B.2.2.
Let $`G`$ be an algebraic group over $``$, $`\varphi :G\mathrm{GL}(V)`$ be a faithful representation of $`G`$. Let $`T`$ and $`D`$ be as in Chevalley’s theorem. Suppose that $`T=V^i`$ (resp. $`T=V^{,i}`$) contains no (resp. only) dual tensor factors. The multiplicative semigroup
$$\begin{array}{c}M(G,V,T,D):=\{m\mathrm{End}(V)m.DD\}\\ (\text{resp. }M(G,V,T,D):=\{m\mathrm{End}(V)^tm.DD\})\end{array}$$
is called a *Chevalley enveloping semigroup* for $`G`$ in $`\mathrm{End}(V)`$.
###### Example B.2.3.
If $`G=\mathrm{GL}_2`$ and $`V`$ is the standard representation, then $`D=^2VV^2`$ is a line as in Chevalley’s theorem, and $`\mathrm{M}_2`$ is the corresponding Chevalley enveloping semigroup.
###### Example B.2.4.
Let $`(V,\psi )`$ be a $`2g`$-dimensional $``$-vector space equipped with an alternating form $`\psi ^2(V^{})`$. Then the line $`D=\psi V^{,2}`$ is a line as in Chevalley’s theorem for the group $`\mathrm{GSp}_{2g}`$ with its standard representation, and the points of the corresponding semigroup in a commutative $``$-algebra $`A`$ are given by
$$\mathrm{MSp}_{2g}(A):=\{m\mathrm{End}(V)(A)\mu (m)A,\psi (m.x,m.y)=\mu (m)\psi (x,y),x,yV_A\}.$$ |
warning/0507/astro-ph0507701.html | ar5iv | text | # Acknowledgments
## Acknowledgments
There are so many people to acknowledge, so many reasons to do it, and it is so easy to let me take by the excitement of the moment, that I will try to be as short and objective as I can.
* To my wife Luz Yasmid, who has shared with me all those beautiful moments in England, and all her efforts, constancy, and dreams. She is an essential part of this success.
* To all my relatives, specially my fathers Ramón and Rosalba, my siblings Yohany and Yamile, my grandmother Leonilde, and my little niece Jinneth. Without them, no PhD, no thesis, no anything. I love them all.
* To my friend David, who has guided me through the physics paths. In absence of my father during these three years, David has reminded me many times of him.
* To all our friends throughout these three years, specially to Margareth, Rebecca, Mati, Mila, Flora, Terry, Tessa, Julieta, Juan, José Roberto, Patricia, José Roberto Jr., Francis, José Daniel, Santiago, Laura, Carlos Miguel, María Isabel, Angélica, José, Karla, Jeannette, Juan Carlos, Mauricio, Pablo, Pili, Luis, John Jairo, Marcos, Simone, Isadora, Carlos, Julia, Abel, Margarita, Sofía, Daniel, Gerardo, Ana Cecilia, Iván, Rosario, Vladimir, Raúl, Oliverio, José Juan, Dulce, Federico, Florencia, Agustín, Santiago, Ignacio, Ricardo, Alejandra, María Francisca, Manuel, and Maribel.
* To Elspeth and Andy, and the Across Cultures group. They have made a great effort to make all the international students and their families feel like at home. We will never forget such a praiseworthy action.
* To the fathers Paul and Hugh, and the sister Ella, for being such excellent friends, and for helping my wife and creating such a wonderful environment for her.
* To the members of the Cosmology and Astroparticle Physics Group: David, Kostas, John, Lotfi, Karim, Ignacio, Juan Carlos, Leila, and Chia-Min.
* To the United Kingdom, our second fatherland.
* To my late uncle Julio, who always supported me and who died days before my travel to the UK.
* To my fathers and to my cousin Julio, who have served as debtors of my loan-scholarships.
* To Carlos, who I began my physics career with.
* To the Centro de Investigaciones at the Universidad Antonio Nariño, which supported me in my application to the COLCIENCIAS loan-scholarship.
* To the Colombian Institute for the Science and Technology Development “Francisco José de Caldas” COLCIENCIAS, for its full postgraduate loan-scholarship.
* To the Foundation for the Future of Colombia COLFUTURO, Universities UK, and the Department of Physics of Lancaster University, for their partial financial support.
Lancaster UK, June 28th 2005.
## Abstract
We review some theoretical and statistical aspects of the origin of the large-scale structure in the Universe, in view of the two most widely known and accepted scenarios: the inflaton scenario (primordial curvature perturbation $`\zeta `$ generated by the quantum fluctuations of the light scalar field $`\phi `$ that drives inflation, named the inflaton), and the curvaton scenario ($`\zeta `$ generated by the quantum fluctuations of a weakly coupled light scalar field $`\sigma `$ that does not drive inflation, named the curvaton). Among the theoretical aspects, we point out the impossibility of having a low inflationary energy scale in the simplest curvaton model. A couple of modifications to the simplest setup are explored, corresponding to the implementation of a second (thermal) inflationary period whose end makes the curvaton field ‘heavy’, triggering either its oscillations or immediate decay. Low scale inflation is then possible to attain with $`H_{}`$ (the Hubble parameter a few Hubble times after horizon exit) being as low as $`1`$ TeV. Among the statistical aspects, we study the bispectrum $`B_\zeta (k_1,k_2,k_3)`$ of $`\zeta `$ whose normalisation $`f_{\mathrm{NL}}`$ gives information about the level of non-gaussianity in the primordial curvature perturbation. In connection with $`f_{\mathrm{NL}}`$, several conserved and/or gauge invariant quantities described as the second-order curvature perturbation have been given in the literature. We review each of these quantities showing how to interpret one in terms of the others, and analyze the respective expected $`f_{\mathrm{NL}}`$ in both the inflaton and the curvaton scenarios as well as in other less known models for the generation of primordial perturbations and/or non-gaussianities. The $`\delta N`$ formalism turns out to be a powerful technique to compute $`f_{\mathrm{NL}}`$ in multi-component slow-roll inflation, as the knowledge of the evolution of some family of unperturbed universes is the only requirement. We present for the first time this formalism and apply it to selected examples.
## Chapter 1 Introduction
The Friedmann-Robertson-Walker (FRW) cosmological model (known also as the standard or Big-Bang cosmological model) is the successful framework that describes the observed properties of the Universe: homogeneity and isotropy at large scales, Hubble expansion, and almost 14 billion years of evolution in agreement with globular clusters and radioactive isotopes dating. The additional predictions of the cosmic background radiation, confirmed by Penzias and Wilson’s discovery in 1965 , and the relative abundances of light elements in full agreement with observation, have established a solid base for the study of the Universe throughout its history, turning the old speculative cosmology into well established and experimentally supported science .
The introduction of a period of exponential expansion (called inflationary) , prior to the Big-Bang, brought an elegant solution to the horizon, flatness, and unwanted relics problems that were present in the original standard cosmological model . The horizon problem, or why is the Cosmic Microwave Background radiation (CMB) temperature highly isotropic?, was solved as the accelerated expansion blows up a region initially in thermal equilibrium to a much bigger size, making the horizon at the end of the matter dominated era be still inside that region; as a result all regions in the sky appear today at the same background temperature. The same accelerated expansion makes the comoving horizon shrink so rapidly that our local patch of the universe becomes extremely flat despite its actual topology, solving this way the flatness problem, or why is our observable Universe almost perfectly flat?. The huge dilution of the abundances of unwanted relics (e.g. topological defects: magnetic monopoles, cosmic strings, and domain walls) caused by the exponential grow of the size of the Universe during inflation and the huge production of entropy by the decay of the scalar field that drives inflation, gave solution to the unwanted relics problem, or where are the topological defects (and some other troublesome stuff) predicted by the standard cosmology? <sup>1</sup><sup>1</sup>1Be aware however that a minimum of inflationary expansion (at least 70 e-folds which correspond to a minimum of $`10^{36}`$ seconds of inflation) is required to successfully solve the horizon, flatness, and unwanted relics problems, assuming standard evolution..
In spite of its success at solving the above mentioned problems, the inflationary period became perhaps more important because of its ability to stretch the quantum fluctuations of the fields living in the FRW spacetime , making them classical and almost constant soon after horizon exit. They correspond to small inhomogeneities in the energy density and are responsible, via gravitational attraction, of the large-scale structure seen today in the Universe (see Figs. 1.1 and 1.2). If this scenario turned to be correct, the energy density inhomogeneities should have left their trace in the CMB released at the time of recombination. Indeed, the Cosmic Background Explorer (COBE) in 1992 found small anisotropies in the CMB temperature of the order of 1 part in $`10^5`$ (with average temperature $`T_0=2.725\pm 0.002`$ K ), on scales of order thousands of Megaparsecs. With 30 times better angular resolution and sensitivity than COBE, the Wilkinson Microwave Anisotropy Probe (WMAP) confirmed this picture in 2003 (see Fig. 1.3), measuring in turn the cosmological parameters with a $`1\%`$ order precision on scales of order tens of Megaparsecs. The PLANCK satellite , due to be launched in 2007, will be able to refine these observations (see Fig. 1.4). With 10 times better angular resolution and sensitivity than WMAP, PLANCK promises to determine the temperature anisotropies with a resolution of the order of 1 part in $`10^6`$, and the cosmological parameters with a $`0.1\%`$ order precision.
The anisotropies in the CMB temperature<sup>2</sup><sup>2</sup>2From now on, and unless otherwise stated, the perturbation $`\delta y`$ in any quantity $`y`$ will be regarded as first-order in cosmological perturbation theory. Unperturbed quantities will be denoted by a subscript 0 unless otherwise stated. $`\delta T/T_0`$ are directly related to the perturbations in the energy density $`\delta \rho /\rho _0`$ at the time of recombination (Sachs-Wolfe effect), whose primarily origin is the stretched quantum fluctuations of one or several scalar fields $`\varphi _i`$ that fill the Universe during inflation <sup>3</sup><sup>3</sup>3In Eqs. (1.1) and (1.2) all the quantities are evaluated at time of last scattering, being $`a`$ the global expansion parameter and $`H\dot{a}/a`$ the global Hubble parameter. A dot means derivative with respect to the cosmic time. The subscripts $`k`$ stand for the fourier modes with comoving wavenumber $`k`$.:
$$\left(\frac{\delta T}{T_0}\right)_k=\frac{1}{2}\left(\frac{aH}{k}\right)^2\left(\frac{\delta \rho }{\rho _0}\right)_k.$$
(1.1)
The perturbations in the energy density at the time of recombination can in turn be quantified by the gauge-invariant primordial curvature perturbation $`\zeta `$ :
$$\left(\frac{\delta \rho }{\rho _0}\right)_k=\frac{2}{5}\left(\frac{k}{aH}\right)^2\zeta _k,$$
(1.2)
which, on flat slices<sup>4</sup><sup>4</sup>4A choice of coordinates defines a threading of spacetime into lines of fixed spatial coordinates $`x_i`$, and a slicing into constant time $`t`$ hypersurfaces. The flat slices are defined such that the intrinsic spatial curvature $`{}_{}{}^{(3)}R`$ vanishes in those hypersurfaces., can be expressed in terms of only the fluctuations in the fields $`\varphi _i`$. For instance, in the case of only one scalar field $`\phi `$ present during inflation, $`\zeta `$ is given by
$$\zeta =H_{\mathrm{inf}}\frac{\delta \phi }{\dot{\phi }_0},$$
(1.3)
where $`H_{\mathrm{inf}}`$ is the global Hubble parameter during inflation. The curvature perturbation $`\zeta `$ is a convenient quantity to describe the primordial perturbations since it is conserved on superhorizon scales ($`kaH_{\mathrm{inf}}`$), as long as the pressure is a unique function of the energy density , and it is well defined even after the scalar fields $`\varphi _i`$ have decayed. We will define $`\zeta `$ in a rigorous way in Chapter 2.
The scalar fields $`\varphi _i`$ we have been talking about might or might not have dominated the energy density during inflation and, therefore, driven the inflationary stage prior to the Big-Bang<sup>5</sup><sup>5</sup>5Although the waterfall field in hybrid inflation dominates the energy density, it does not drive inflation. This is not in contradiction with our previous statement since, in the strict sense, the waterfall field fluctuations are suppressed as the waterfall mass is much bigger than $`H_{}`$ (the star ‘$``$’ denoting the global Hubble parameter evaluated a few Hubble times after horizon exit) .. Our adopted definition for the inflaton field in this thesis will be the light field $`\phi `$ that dominates the energy density during inflation<sup>6</sup><sup>6</sup>6By light field we mean a field $`\varphi `$ whose mass $`m_\varphi `$ is much less than $`H_{}`$. and drives the exponential expansion. This field in most cases parameterises the distance along the inflationary trajectories. Until recently, the most widely known and accepted scenario for the origin of the density perturbations identified the inflaton with the scalar field whose fluctuations were responsible for the primordial density perturbations. This scenario, called the inflaton scenario<sup>7</sup><sup>7</sup>7For simplicity we will just consider single-component inflationary models of the slow-roll variety in the present introduction and in the following chapter. The multi-component case, which contains one inflaton and one or more ‘light non-inflaton fields’, will be considered in Chapters 5 and 6. , describes very well the properties of $`\zeta `$, leading to an almost scale invariant power spectrum
$$𝒫_\zeta (k)A_\zeta ^2\left(\frac{k}{aH_{\mathrm{inf}}}\right)^{n_\zeta },$$
(1.4)
which is defined by the statistical average
$$\zeta _{𝐤_1}\zeta _{𝐤_2}\frac{2\pi ^2}{k^3}\delta ^3(𝐤_1+𝐤_2)𝒫_\zeta (k),$$
(1.5)
calculated over an ensemble of universes. The amplitude $`A_\zeta `$ and spectral index $`n_\zeta `$ in Eq. (1.4) <sup>8</sup><sup>8</sup>8In this thesis we will use natural units such that $`c=\mathrm{}=k_B=1`$, and Newton’s gravitational constant given by $`8\pi Gm_P^2`$, with $`m_P=2.436\times 10^{18}`$ GeV being the reduced Planck mass.:
$`A_\zeta `$ $``$ $`{\displaystyle \frac{H_{}}{\sqrt{8\epsilon }\pi m_P}},`$ (1.6)
$`n_\zeta `$ $`=`$ $`2\eta _\phi 6\epsilon ,`$ (1.7)
are functions of the slow-roll parameters
$`\epsilon `$ $``$ $`\left({\displaystyle \frac{\dot{H}_{\mathrm{inf}}}{H_{\mathrm{inf}}^2}}\right)_{},`$ (1.8)
$`\eta _\phi `$ $``$ $`\epsilon {\displaystyle \frac{\ddot{\phi }_0}{H_{\mathrm{inf}}\dot{\phi }_0}},`$ (1.9)
that characterize the inflationary behaviour and satisfy the slow-roll conditions $`\epsilon 1`$ and $`|\eta _\phi |1`$ . The scale of inflation, given by $`H_{}`$, is not completely determined in the inflaton scenario by the already measured amplitude $`|A_\zeta |`$ , since the slow-roll parameter $`\epsilon `$ is unknown (except for the upper bound $`\epsilon \text{ }\stackrel{<}{}\text{ }0.01`$ ).
The next statistical significant quantity after $`𝒫_\zeta (k)`$, the bispectrum $`B_\zeta (k_1,k_2,k_3)`$ defined by the statistical average
$$\zeta _{𝐤_1}\zeta _{𝐤_2}\zeta _{𝐤_3}(2\pi )^{3/2}B_\zeta (k_1,k_2,k_3)\delta ^3(𝐤_1+𝐤_2+𝐤_3),$$
(1.10)
is also well described in the inflaton scenario as its normalisation $`f_{\mathrm{NL}}`$, defined by
$$B_\zeta (k_1,k_2,k_3)\frac{6}{5}f_{\mathrm{NL}}\left[\frac{2\pi ^2}{k_1^3}𝒫_\zeta (k_1)\frac{2\pi ^2}{k_2^3}𝒫_\zeta (k_2)+\mathrm{cyclic}\mathrm{permutations}\right],$$
(1.11)
is suppressed by $`\epsilon `$ and $`\eta _\phi `$ describing a highly gaussian set of perturbations:
$$f_{\mathrm{NL}}=\frac{5}{12}[2\eta _\phi 6\epsilon 2ϵf(k_1,k_2,k_3)].$$
(1.12)
The value of the scale dependent function $`f`$ in the previous expression lies in the range $`0f5/6`$, being precisely determined by the respective wavevector configuration .
We make a couple of observations which are valid not only for the inflaton scenario but also for the curvaton one discussed below. First, the amplitude $`A_T`$ of the power spectrum of gravitational waves
$$𝒫_T(k)A_T^2\left(\frac{k}{aH_{\mathrm{inf}}}\right)^{n_T},$$
(1.13)
which is defined by the statistical average
$$\underset{ij}{}h_{𝐤_1}^{ij}h_{𝐤_2}^{ij}\frac{2\pi ^2}{k^3}\delta ^3(𝐤_1+𝐤_2)𝒫_T(k),$$
(1.14)
where $`h_{ij}`$ is the tensor perturbation in the perturbed metric tensor, depends only on the scale of inflation (given by $`H_{}`$):
$$A_T\frac{\sqrt{2}H_{}}{\pi m_P}.$$
(1.15)
Second, its tilt $`n_T`$ is a function only of $`\epsilon `$:
$$n_T=2\epsilon .$$
(1.16)
In view of our previous discussion about $`𝒫_\zeta (k)`$, we can write a consistency relation involving $`A_\zeta `$, $`A_T`$, and $`n_T`$, which presents itself as a nice prediction of the inflaton scenario:
$$r_{T\zeta }\frac{A_T^2}{A_\zeta ^2}=8n_T.$$
(1.17)
Its confirmation, as well as a negative detection of non-gaussianity by both WMAP and PLANCK satellites, would give strong support to the inflaton scenario as the correct framework to understand the origin of the large-scale structure in the Universe. We will discuss in more detail the inflaton scenario in Chapter 2.
An alternative to the inflaton scenario is when the weakly coupled light field $`\sigma `$, whose fluctuations are responsible for the primordial density perturbations, does not dominate the energy density, and therefore does not drive inflation. This scenario is called the curvaton scenario (see also Refs. ), and the field $`\sigma `$ receives the name curvaton. Introduced in 2002, this scenario describes also very well the properties of $`\zeta `$, with an almost scale invariant power spectrum $`𝒫_\zeta (k)A_\zeta ^2(k/aH_{\mathrm{inf}})^{n_\zeta }`$ whose spectral index $`n_\zeta `$ written as
$$n_\zeta =2\eta _\sigma 2\epsilon ,$$
(1.18)
is function of $`\epsilon `$ and
$$\eta _\sigma \frac{m_P^2}{V}\frac{^2V}{\sigma _0^2}\frac{m_\sigma ^2}{3H_{}^2}1,$$
(1.19)
being $`V`$ the scalar potential, and whose amplitude $`A_\zeta `$ is given by $`H_{}`$, the unperturbed component of the curvaton field during inflation $`\sigma _{}`$, and the fractional global curvaton energy density $`\mathrm{\Omega }_{\mathrm{dec}}\rho _{\sigma _0}/\rho _{\mathrm{total}_0}`$ just before the curvaton decay:
$$A_\zeta \frac{H_{}\mathrm{\Omega }_{\mathrm{dec}}}{3\pi \sigma _{}}.$$
(1.20)
The scale of inflation in the curvaton scenario, given by $`H_{}`$, does not depend on $`\epsilon `$ nor on $`\eta _\sigma `$, instead it depends on the free parameters $`\sigma _{}`$ and $`\mathrm{\Omega }_{\mathrm{dec}}`$. A consequence from the previous expression is that there is no analogous to the consistency relation \[c.f. Eq. (1.17)\] for the curvaton scenario. However, there are distinctive non-gaussian signatures that can allow us to distinguish this model from other scenarios for the origin of the large-scale structure in the Universe (see below) . The curvaton scenario will be studied in detail in Chapter 2.
Perhaps the main motivation for the introduction of the curvaton scenario is that it liberates the inflaton field from the generation of the primordial perturbations . This is particularly good from the particle physics point of view since the intrinsic difficulty at embedding the inflaton scenario in a particle physics model is greatly alleviated. Indeed, in the inflaton scenario, the energy scale of inflation is likely to be quite high<sup>9</sup><sup>9</sup>9The upper bound on $`H_{}`$ can be obtained from Eq. (1.6) and the current bound on the slow-roll parameter $`\epsilon \text{ }\stackrel{<}{}\text{ }0.01`$. ($`H_{}\text{ }\stackrel{<}{}\text{ }10^{14}`$ GeV) in order to produce the required level of primordial perturbations $`|A_\zeta |5\times 10^5`$ <sup>10</sup><sup>10</sup>10Nevertheless there are examples where an $`\epsilon `$ parameter of order $`10^{24}`$ is naturally obtained, so that the right amount of primordial perturbations is generated for $`H_{}10^3`$ GeV (see e.g. Ref. ).. This makes very difficult the identification of the inflaton field with one of the scalar fields present in the supersymmetry (SUSY) breaking sector or with one of the Minimal Supersymmetric Standard Model (MSSM) flat directions. This is because the characteristic SUSY energy scale, which depends on the symmetry breaking scheme, goes typically from $`H_{}10^2`$ GeV to $`H_{}10^5`$ GeV . In contrast, in the curvaton scenario, the energy scale has to be much lower, satisfying $`H_{}\text{ }\stackrel{<}{}\text{ }10^{12}`$ GeV , so that the inflaton field does not contribute to the curvature perturbation \[c.f. Eq. (1.6)\]<sup>11</sup><sup>11</sup>11An interesting scenario is when both the inflaton and the curvaton fields contribute to $`\zeta `$ . This is the case where the curvaton starts oscillating around the minimum of its potential when it already contributes significantly to the total energy density $`\rho _{\mathrm{total}}`$. The upper bound $`H_{}\text{ }\stackrel{<}{}\text{ }10^{12}`$ GeV is, in this case, therefore relaxed. In this thesis we will consider only the standard curvaton scenario, where the inflaton field does not contribute to $`\zeta `$ and the curvaton oscillations begin when the curvaton energy density $`\rho _\sigma `$ is still subdominant.. The last bound can be taken as an anti-smoking gun for the curvaton scenario because a positive gravitational wave signal would require $`H_{}\text{ }\stackrel{>}{}\text{ }10^{12}`$ GeV , ruling out the curvaton as the source of primordial density fluctuations. Some people can see this as a bad feature of the scenario in question as it sends an unpromising message to all those who are making big efforts to detect gravitational waves. However, as described in Ref. , the curvaton scenario may well be consistent with detectable gravitational waves as far as the inflaton is a ‘heavy’ field ($`m_\phi \text{ }\stackrel{>}{}\text{ }H_{}`$), violating this way the slow-roll condition $`|\eta _\phi |1`$<sup>12</sup><sup>12</sup>12Of course, inflation in this case is not of the slow-roll variety. Some possibilities are fast-roll or hilltop inflation ., and suppressing the amplitude of the curvature perturbation produced by the inflaton itself.
One of our main concerns in this thesis is the possibility to accommodate low-scale inflation in the curvaton scenario. Unfortunately, even when the energy scale may in principle be greatly reduced compared to the inflaton scenario, the simplest curvaton model still requires a quite high inflationary energy scale satisfying $`H_{}>10^7`$ GeV . This of course makes impossible the embedding of the inflaton field within the framework of a SUSY particle physics model. In Chapters 3 and 4 we explore two modifications to the curvaton model which can instead allow inflation at a low scale . In the first modification the end of a second (thermal) inflationary stage , driven by the rolling of a flaton field $`\chi `$ coupled to the curvaton, makes the curvaton mass $`m_\sigma `$ increase suddenly at some moment after the end of inflation but before the onset of the curvaton oscillations. This proposal can work but not in a completely natural way. Nevertheless, we show that inflation with $`H_{}`$ as low as 1 TeV or lower is possible to attain. In the second modification the increment in $`m_\sigma `$ at the end of the thermal inflation era is so huge that the decay rate overtakes the Hubble parameter and the curvaton field decays immediately. The advantage of this second modification is that low scale inflation is achieved for more natural values in the relevant parameter space.
Aside from the previously introduced theoretical aspects of the origin of the large-scale structure in the Universe, we also study some of its statistical aspects. Conversely to the inflaton scenario \[c.f. Eq. (1.12)\], $`\zeta `$ in the simplest curvaton model may present a sizable non-gaussian component if the curvaton does not dominate the energy density before decaying . More specifically, according to the expression
$$f_{\mathrm{NL}}=\frac{5}{3}+\frac{5}{6}r\frac{5}{4r},$$
(1.21)
which is valid in the curvaton scenario, $`|f_{\mathrm{NL}}|1`$ is obtained if $`r`$ is very small. In the previous expression $`r`$ is defined by
$$r\frac{3\rho _{\sigma _0}}{4\rho _{r_0}+3\rho _{\sigma _0}},$$
(1.22)
and is evaluated just before the curvaton decay (being $`\rho _r`$ the global radiation energy density). This is of extreme importance since the next WMAP data release , or in its defect the future PLANCK satellite data , will either detect non-gaussianity or put strong constraints on $`f_{\mathrm{NL}}`$, offering the possibility of successfully discriminating among the different inflaton and curvaton models. The current constraint on $`f_{\mathrm{NL}}`$, according to the first-year WMAP data, is $`|f_{\mathrm{NL}}|\text{ }\stackrel{<}{}\text{ }10^2`$ . The next data release is expected to lower this bound by one order of magnitude or so .
The non-linearity parameter $`f_{\mathrm{NL}}`$, if independent of position, is closely related to the second-order curvature perturbation $`\zeta _2`$ defined by
$`\zeta (𝐱)`$ $``$ $`\zeta _1(𝐱)+{\displaystyle \frac{1}{2}}\zeta _2(𝐱)`$ (1.23)
$``$ $`\zeta _g(𝐱){\displaystyle \frac{3}{5}}f_{\mathrm{NL}}(\zeta _g^2(𝐱)\zeta _g^2),`$
where $`\zeta _g`$ is gaussian with $`\zeta _g=0`$. In connection with this issue we point out that several conserved and/or gauge invariant quantities described as the second-order curvature perturbation have been given in the literature . In Chapter 5 we revisit various scenarios for the generation of second-order non-gaussianity in $`\zeta `$, employing for the first time a unified notation and focusing on $`f_{\mathrm{NL}}`$ . When $`\zeta `$ first appears a few Hubble times after horizon exit, $`|f_{\mathrm{NL}}|`$ is much less than $`1`$ and is, therefore, negligible. Thereafter $`\zeta `$ (and hence $`f_{\mathrm{NL}}`$) is conserved as long as the pressure is a unique function of the energy density (adiabatic pressure) . Non-adiabatic pressure comes presumably only from the effect of fields, other than the one pointing along the inflationary trajectory, which are light during inflation (light non-inflaton fields) . Our expectation is that, although during single-component inflation $`f_{\mathrm{NL}}`$ is constant, multi-component inflation might generate $`|f_{\mathrm{NL}}|1`$ or bigger. We mention some recent proposals where non-gaussianity can be generated during the preheating stage following inflation , and conjecture that preheating can affect $`f_{\mathrm{NL}}`$ only in atypical scenarios where it involves light non-inflaton fields . We also study the curvaton scenario and derive Eq. (1.21), showing that the simplest model typically gives $`f_{\mathrm{NL}}1`$ or $`f_{\mathrm{NL}}=+5/4`$. The inhomogeneous reheating scenario (see also Refs. ), where $`\zeta `$ is generated by the inhomogeneities in the inflaton decay rate during reheating, is quickly reviewed showing that it can give a wide range of values for $`f_{\mathrm{NL}}`$ . One important conclusion from this chapter is that it will be crucial to calculate the precise observational limit on $`f_{\mathrm{NL}}`$ using second order theory in case that, unless there is a detection, observation could eventually provide a limit $`|f_{\mathrm{NL}}|\text{ }\stackrel{<}{}\text{ }1`$ .
A new and interesting proposal is the extension to second order of the $`\delta N`$ formalism (see also Refs. ), used initially to calculate $`\zeta `$ at first order and the spectral index in multi-component slow-roll models of inflation <sup>13</sup><sup>13</sup>13For an extension to multi-component non slow-roll models see Ref. .. In this formalism $`\zeta (t,𝐱)`$ is expressed as the perturbation in the amount of expansion
$$N(t,𝐱)\mathrm{ln}\left[\frac{\stackrel{~}{a}(t,𝐱)}{a(t_{\mathrm{in}})}\right],$$
(1.24)
from an initially flat slice at time $`t_{in}`$ to a final slice of uniform energy density at time $`t`$:
$$\zeta (t,𝐱)=\delta NN(t,𝐱)N_0(t).$$
(1.25)
Here
$$N_0(t)\mathrm{ln}\left[\frac{a(t)}{a(t_{\mathrm{in}})}\right],$$
(1.26)
is the unperturbed amount of expansion, and $`\stackrel{~}{a}(t,𝐱)`$ is the local expansion parameter. Thus, $`\zeta `$ up to second order is given by
$$\zeta (t,𝐱)=\underset{i}{}N_{,i}(t)\delta \varphi _i+\frac{1}{2}\underset{ij}{}N_{,ij}(t)\delta \varphi _i\delta \varphi _j,$$
(1.27)
where
$`N_{,i}`$ $`=`$ $`{\displaystyle \frac{N}{\varphi _{i_0}}},`$ (1.28)
$`N_{,ij}`$ $`=`$ $`{\displaystyle \frac{^2N}{\varphi _{i_0}\varphi _{j_0}}},`$ (1.29)
and the fields $`\varphi _i`$, evaluated a few Hubble times after horizon exit, are those relevant for the generation of $`\zeta `$. In view of Eq. (1.23), this presents as a powerful method to calculate $`f_{\mathrm{NL}}`$ in any multi-component slow-roll inflationary model for the generation of $`\zeta `$ . In Chapter 6 we give for the first time this formalism, which allows us to extract all the stochastic properties of $`\zeta `$ if the initial field perturbations are gaussian<sup>14</sup><sup>14</sup>14In the case that the initial field perturbations are non-gaussian, there is an additional contribution to Eq. (1.30) which is strongly wavevector dependent . This contribution is in any case very small , being $`\text{ }\stackrel{<}{}\text{ }10^2`$.. The elegance and power of this method lies in the fact that the calculation requires only the knowledge of the evolution of some family of unperturbed universes. The following formula is given for $`f_{\mathrm{NL}}`$ in terms of $`N(t,\varphi _{i_0})`$ and its derivatives :
$$\frac{3}{5}f_{\mathrm{NL}}=\frac{_{ij}N_{,i}N_{,j}N_{,ij}}{2\left[_iN_{,i}^{}{}_{}{}^{2}\right]^2}+\mathrm{ln}(kL)\frac{A_\zeta ^2}{2}\frac{_{ijk}N_{,ij}N_{,jk}N_{,ki}}{\left[_iN_{,i}^{}{}_{}{}^{2}\right]^3},$$
(1.30)
where $`k^1`$ is a typical cosmological scale and $`L`$ is the size of the region within which the stochastic properties are specified, so that $`\mathrm{ln}(kL)1`$. We apply the above formula to the Kadota and Stewart modular inflation model , the curvaton scenario , and the multi-component ‘hybrid’ inflation model of Enqvist and V$`\ddot{\mathrm{a}}`$ihk$`\ddot{\mathrm{o}}`$nen . The relation of this formula to cosmological perturbation theory is also explained.
The conclusions of this thesis are drawn in Chapter 7.
## Chapter 2 Two mechanisms for the origin of the large-scale structure
### 2.1 Introduction
In the standard inflationary scenario a single scalar field, named the inflaton, is responsible for the solution of the horizon, flatness, and unwanted relics problems, as well as for the origin of the large scale structure seen in the observable Universe. This double mission for the inflaton field imposes strong constraints on the parameters of the inflationary models, leading to big intrinsic difficulties at building successful and realistic models of inflation . To rescue the well motivated inflationary models that fail at generating the required level of primordial perturbations , the inflaton field is left in charge of driving inflation only. The other task, the generation of the primordial perturbations, is assigned to a weakly coupled light field $`\sigma `$ different from the inflaton. This is the curvaton scenario (see also Refs. ), where the original curvature perturbation $`\zeta _\sigma `$, associated to and produced by $`\sigma `$ during inflation, is gradually transformed into the total curvature perturbation $`\zeta `$ <sup>1</sup><sup>1</sup>1During inflation and until the start of the radiation dominated epoch just after reheating, $`\zeta _\sigma `$ is actually an isocurvature perturbation. The reason of the name isocurvature is because $`\zeta _\sigma `$ during that time does not contribute at all to the total curvature perturbation $`\zeta `$.. The conversion process starts during the radiation dominated epoch that follows the reheating stage produced by the inflaton decay<sup>2</sup><sup>2</sup>2The cause of the conversion process is the relative redshifting between the radiation and the curvaton fluid energy densities. The curvaton at this stage is considered a matter fluid since the period of its oscillations around the minimum of its potential is much less than the characteristic expansion time scale . This makes the average curvaton pressure essentially zero..
This chapter will give some preliminary definitions and basic facts about the inflaton and the curvaton scenarios, such as the first-order perturbations in the metric, the precise definition of $`\zeta `$, the slow-roll conditions, the characteristics of the spectrum of $`\zeta `$ in both the inflaton and the curvaton scenarios, and the spectrum of gravitational waves and the consequences derived from its possible detection. Having this information at hand, it will be easier to follow the main discussions of this thesis that are exposed in Chapters 3 to 6.
### 2.2 Metric perturbations and the primordial curvature perturbation $`\zeta `$
Our observable expanding Universe is well described as homogeneous and isotropic, down to scales of order of tens of Megaparsecs. The FRW line element
$$ds^2=dt^2+a^2(t)\left(\frac{dr^2}{1\kappa r^2}+r^2d\theta ^2+r^2\mathrm{sin}^2\theta d\varphi ^2\right),$$
(2.1)
describes such an Universe , assuming that the galaxies are always at the same comoving spherical coordinates $`r`$, $`\theta `$, and $`\varphi `$, and that the intergalactic space is continually increasing with time due to the expansion parameter (scale factor) $`a(t)`$. The proper time, once synchronized in all the galaxies, runs at the same rate everywhere, so the metric is well described by a global cosmic time $`t`$. Finally, the topology of the Universe can be classified by the parameter $`\kappa `$ as closed ($`\kappa >0`$), open ($`\kappa =0`$), or flat ($`\kappa <0`$), according to whether the space is finite but unbounded, infinite with certain curvature, or infinite but strictly flat a la Minkowski <sup>3</sup><sup>3</sup>3The actual value of $`\kappa `$ is conventional since it depends on the chosen value for the present expansion parameter $`a_{\mathrm{today}}`$.. The inflationary stage however shrinks the comoving horizon so rapidly that the present Universe looks so extremely flat. This makes the parameter $`\kappa `$ be completely irrelevant when going back in time, so that we can safely discard it when studying the inflationary and early Universe processes. Thus, and switching to cartesian coordinates $`x^i`$ and conformal time $`\eta `$ defined by $`d\eta dt/a`$, the FRW line element looks
$$ds^2=a^2(\eta )\left(d\eta ^2+\delta _{ij}dx^idx^j\right).$$
(2.2)
#### 2.2.1 First-order perturbations in the FRW line element: classification and number of degrees of freedom
The homogeneity and isotropy assumptions make a good description of the Universe, but they are not in any case perfect conditions. The observed galaxy distribution and the temperature fluctuations in the CMB (see Figs. 1.1 \- 1.4) reveal the importance of modifying the FRW line element so that it accounts for the small deviations from homogeneity and isotropy required to correctly describe such large-scale structures. The introduction of small inhomogeneities in the FRW line element, and the truncation at first-order of the perturbed Einstein equations, are well justified since the galaxy density contrast $`\delta \rho /\rho _0`$ and the CMB temperature anisotropies $`\delta T/T_0`$ are observed to be of order $`10^5`$ . In consequence, considering only scalar degrees of freedom, the most generic first-order perturbed line element reads
$$ds^2=a^2(\eta )\left[(1+2\varphi _G)d\eta ^2+2_iBd\eta dx^i+((12\psi )\delta _{ij}+D_{ij}E)dx^idx^j\right],$$
(2.3)
where $`D_{ij}_i_j(1/3)\delta _{ij}^2`$, and $`\varphi _G`$, $`B`$, $`\psi `$, and $`E`$ are scalar quantities.
The peculiar form of the perturbed line element in Eq. (2.3), as well as the right number of scalar degrees of freedom in that expression, are justified as follows:
* The full perturbed metric tensor is generically described by scalar, vector, and tensor perturbations.
* The total number of degrees of freedom for a symmetric tensor in an $`(n+1)`$-dimensional spacetime is $`(n+2)(n+1)/2`$.
* The $`g_{00}`$ entry can be written as $`(1+2\varphi _G)`$, where $`\varphi _G`$ corresponds to a purely scalar perturbation (1 scalar degree of freedom).
* The $`g_{0i}`$ entries can be parameterised by $`B_i_iB+\upsilon _i`$, where $`\stackrel{}{}\stackrel{}{\upsilon }=0`$. The term $`_iB`$ corresponds to a purely scalar perturbation (1 scalar degree of freedom). The term $`\upsilon _i`$ corresponds to a purely vector perturbation ($`n1`$ vector degrees of freedom because of the $`\stackrel{}{}\stackrel{}{\upsilon }=0`$ constraint).
* The $`g_{ij}`$ entries can be written as $`(12\psi )\delta _{ij}+\mathrm{\Pi }_{ij}`$, where $`\psi `$ is a purely scalar perturbation (1 scalar degree of freedom) that accounts for the trace of $`g_{ij}`$, and $`\mathrm{\Pi }_{ij}`$ is a symmetric traceless tensor.
* $`\mathrm{\Pi }_{ij}=\mathrm{\Pi }_{ij}^S+\mathrm{\Pi }_{ij}^V+\mathrm{\Pi }_{ij}^T`$ is expressed in terms of a purely scalar perturbation (encoded in $`\mathrm{\Pi }_{ij}^S`$), a purely vector perturbation (encoded in $`\mathrm{\Pi }_{ij}^V`$), and a purely tensor perturbation (encoded in $`\mathrm{\Pi }_{ij}^T`$).
* $`\mathrm{\Pi }_{ij}^S`$ can be written as $`D_{ij}E`$ because it is a symmetric traceless tensor corresponding to the purely scalar perturbation $`E`$ (1 scalar degree of freedom).
* $`\mathrm{\Pi }_{ij}^V`$ can be written as $`(_i\mathrm{\Pi }_j+_j\mathrm{\Pi }_i)/2`$, where $`\mathrm{\Pi }_i`$ is a purely vector perturbation satisfying $`\stackrel{}{}\stackrel{}{\mathrm{\Pi }}=0`$ ($`n1`$ vector degrees of freedom).
* $`\mathrm{\Pi }_{ij}^T`$ corresponds to a purely tensor perturbation such that $`_i\mathrm{\Pi }_{ij}^T=0`$. Considering the previous three items, and the fact that the number of degrees of freedom coming from a traceless symmetric tensor in an $`n`$-dimensional space is $`(n+1)n/21`$, the number of tensor degrees of freedom is $`(n2)(n+1)/2`$.
* The total number of scalar degrees of freedom (4), vector degrees of freedom ($`2(n1)`$), and tensor degrees of freedom ($`(n2)(n+1)/2`$), add up to reproduce the total number of degrees of freedom of the metric tensor $`g_{\mu \nu }`$ given in the second item: $`(n+2)(n+1)/2`$.
Two very important facts follow from the first-order perturbed Einstein equations . First, at first order the vector perturbations are decoupled from the scalar perturbations being anyway usually neglected due to their rapid decrease with time. Second, the tensor perturbations at first order also decouple from their scalar counterparts. These two facts, along with the parameterisation of the metric tensor discussed in the above items, justify the use of the Eq. (2.3) as the most generic first-order metric perturbed line element that describes the energy density (scalar) perturbations in the FRW spacetime.
#### 2.2.2 The curvature perturbation $`\psi `$ and its non-invariance under infinitesimal coordinate transformations
In the perturbed line element of Eq. (2.3), $`\psi `$ represents the intrinsic spatial curvature $`{}_{}{}^{(3)}R`$ on hypersurfaces of constant conformal time $`\eta `$ :
$${}_{}{}^{(3)}R=\frac{4}{a^2}^2\psi ,$$
(2.4)
where the operator $`^2`$ is the comoving Laplacian operator. For this reason the quantity $`\psi `$ is usually referred to as the curvature perturbation. The curvature perturbation $`\psi `$, as well as the other scalar perturbations $`\varphi _G`$, $`B`$, and $`E`$, are however not invariant under a coordinate transformation. Indeed, from the most general infinitesimal coordinate transformation<sup>4</sup><sup>4</sup>4The vector component that should appear in the transformation of $`x_i`$ has been discarded. This is because the infinitesimal first-order vector shifts contribute to the transformation rules of the vector perturbations only.
$`\eta `$ $``$ $`\eta +\xi (x^\mu ),`$ (2.5)
$`x_i`$ $``$ $`x_i+_i\beta (x^\mu ),`$ (2.6)
the scalar perturbations $`\varphi _G`$, $`B`$, $`\psi `$, and $`E`$ transform as
$`\varphi _G`$ $``$ $`\varphi _G\xi ^{}\xi ,`$ (2.7)
$`B`$ $``$ $`B+\xi \beta ^{},`$ (2.8)
$`\psi `$ $``$ $`\psi +{\displaystyle \frac{1}{3}}^2\beta +\xi ,`$ (2.9)
$`E`$ $``$ $`E2\beta ,`$ (2.10)
where a prime means derivation with respect to the conformal time $`\eta `$, and $``$ is the global conformal Hubble parameter defined by $`a^{}/a`$. The above transformation rules are under the physical proviso that the perturbed line element $`ds^2`$ in Eq. (2.3) should remain invariant.
The parameters $`\xi `$ and $`\beta `$ that give account of the infinitesimal coordinate transformation in Eqs. (2.5) and (2.6) can be adjusted (fixing the slicing and the threading) so that two of the four scalar perturbations in Eqs. (2.7) to (2.10) vanish. The longitudinal gauge, for instance, corresponds to choose $`B`$ and $`E`$ equal to zero. In this specific gauge the gravitational potential $`\varphi _G`$ becomes equal to the curvature perturbation $`\psi `$ up to first order as long as the considered fluid is described by a perfect isotropic stress, examples of such a fluid being the inflaton and/or the curvaton fields. All of this leads us to say with confidence that the curvature perturbation $`\psi `$ really represents the effect of the inhomogeneities in the FRW spacetime and it is, therefore, the quantity to study.
#### 2.2.3 The gauge-invariant curvature perturbation $`\zeta `$
To parameterise adequately the inhomogeneities in the FRW spacetime, we need a quantity invariant under the coordinate (gauge) transformations in Eqs. (2.5) and (2.6). This is not completely possible to do, but we may define a quantity which is invariant under transformations in time only. This is done taking into account that, for any scalar quantity $`f(x^\mu )`$ different to $`\varphi _G`$, $`B`$, $`\psi `$, and $`E`$, the transformation law in the associated perturbation $`\delta f(x^\mu )`$ is given by
$$\delta f\delta ff_0^{}\xi .$$
(2.11)
The first gauge invariant quantity that we may define is that which represents the curvature perturbation $`\psi `$ in the comoving slices, defined them as the constant time hypersurfaces where there is no flux of energy. Considering the inflaton field $`\phi `$, the comoving slices coincide with those where $`\phi `$ is uniform ($`\delta \phi _{\mathrm{com}}=0`$) so that the time translation $`\xi _{\mathrm{com}}(x^\mu )`$ required to go from a generic slice to the comoving slice is:
$$\xi _{\mathrm{com}}=\frac{1}{\phi _0^{}}[\delta \phi \delta \phi _{\mathrm{com}}]=\frac{\delta \phi }{\phi _0^{}}.$$
(2.12)
Therefore the comoving curvature perturbation $`\psi _{\mathrm{com}}`$, denoted from now on as $`\zeta `$, is written in terms of $`\psi `$ and $`\phi `$ in the generic slice as
$`\zeta `$ $`=`$ $`\psi \xi _{\mathrm{com}}`$ (2.13)
$`=`$ $`\psi {\displaystyle \frac{\delta \phi }{\phi _0^{}}}.`$
The overall minus sign is just a convention, chosen in this thesis to match the agreed definition of $`\zeta `$ by most of the authors.
The second gauge invariant quantity that we may define represents the curvature perturbation $`\psi `$ in the slices of uniform energy density. Again, and following the same steps as before, we have to consider the infinitesimal time translation $`\xi _{\mathrm{uni}}(x^\mu )`$ required to go from a generic slice to the uniform energy density slice where $`\delta \rho _{\mathrm{uni}}=0`$:
$$\xi _{\mathrm{uni}}=\frac{1}{\rho _0^{}}[\delta \rho \delta \rho _{\mathrm{uni}}]=\frac{\delta \rho }{\rho _0^{}}.$$
(2.14)
Thus, the curvature perturbation in the uniform density slice $`\psi _{\mathrm{uni}}`$, denoted from now on as $`\zeta `$, is given by
$`\zeta `$ $`=`$ $`\psi \xi _{\mathrm{uni}}`$ (2.15)
$`=`$ $`\psi {\displaystyle \frac{\delta \rho }{\rho _0^{}}}.`$
We have chosen to denote both the comoving and the uniform density slice curvature perturbations with the same letter $`\zeta `$ because they are equivalent on superhorizon scales ($`kaH_{\mathrm{inf}}`$) . The superhorizon scale region is of great importance because $`\zeta `$ is conserved in that region if the adiabatic condition, described below, is satisfied. In addition, at superhorizon scales $`\zeta `$ becomes truly gauge-invariant because the contribution of $`\beta `$ in Eq. (2.9) gets suppressed by the spatial derivatives, so that $`\psi `$ does not depend on the changes in the threading<sup>5</sup><sup>5</sup>5 Notice that if $`\phi _0^{}0`$ or $`\rho _0^{}0`$ the curvature perturbations in Eqs. (2.13) and (2.15) blow up. To avoid such a disaster we need very small values for $``$ so as to generate the observed value for $`\zeta `$. If for some reason $`\phi _0^{}=0`$ or $`\rho _0^{}=0`$, $`\zeta `$ as given in Eq. (2.13) or (2.15) becomes ill defined and we would need to look for a better well defined gauge invariant quantity that represents the intrinsic curvature perturbation $`\psi `$..
The conservation of $`\zeta `$ is guaranteed as long as the pressure $`P`$ is a unique function of the energy density $`\rho `$ (the adiabatic condition). The last statement follows from the local energy conservation equation in the uniform density slicing at large scales :
$$\dot{\rho }(t)=3(H+\dot{\zeta })[\rho (t)+P(t,𝐱)].$$
(2.16)
If $`P`$ satisfies the adiabatic condition then it becomes spatially uniform ($`\zeta =\zeta `$), and so does $`\dot{\zeta }`$. As a result
$$\dot{\zeta }=0,$$
(2.17)
because
$$\dot{\zeta }=\dot{\zeta }=\frac{d}{dt}\zeta =0.$$
(2.18)
If $`P`$ does not satisfy the adiabatic condition, i.e. if it has a non-adiabatic component $`P_{\mathrm{nad}}`$, Eq. (2.16) implies
$$\dot{\zeta }=\frac{H}{\rho _0+P_0}\delta P_{\mathrm{nad}}.$$
(2.19)
The curvaton scenario is one example where the existence of a non-adiabatic pressure perturbation makes $`\zeta `$ evolve from the negligible curvature perturbation produced by the inflaton to the right value observed today. The non-adiabatic pressure perturbation is, in this case, the result of the presence of two weakly interacting fluids, the curvaton matter fluid and the radiation fluid, in the period that follows the inflaton decay and reheating.
### 2.3 Inflation and its effect on the spectrum of perturbations of a non-dominating massless scalar field during a de Sitter stage
Any period in the history of the Universe during which the expansion is accelerated is denominated as inflationary . The inflationary stage prior to the Hot Big-Bang has the nice property to stretch the quantum fluctuations of the scalar fields living in the FRW spacetime , so that they become classical and almost constant, sourcing the primordial density inhomogeneities responsible for the presently observed large-scale structure. The amplitude of the spectrum of the classical field perturbations (for light fields) is generically the same for all kinds of quasi de Sitter models. However, the spectral index is written down in a certain way for fields that dominate the energy density and in another way for fields that do not. In this section we will describe inflation, paying special attention to the constraint imposed by it on the slow-roll parameter $`\epsilon `$, and review the properties of the power spectrum of perturbations of a massless scalar field during a de Sitter stage, characterized the latter by a constant Hubble parameter $`H_{\mathrm{inf}}=H_{}`$ during inflation.
#### 2.3.1 Inflation
Inflation can be rigorously defined as the period when the global expansion parameter $`a`$ satisfies the condition
$$\ddot{a}>0.$$
(2.20)
Such a condition translates into a definite requirement on the global Hubble parameter during inflation $`H_{\mathrm{inf}}`$. From the definition of $`H_{\mathrm{inf}}`$ in terms of $`a`$, which can be written alternatively as the following evolution equation:
$$a(t)=a_{\mathrm{ini}}\mathrm{exp}\left(_{t_{\mathrm{ini}}}^tH_{\mathrm{inf}}𝑑t\right),$$
(2.21)
being $`a_{\mathrm{ini}}`$ the expansion parameter at some time $`t_{\mathrm{ini}}`$, the inflationary condition in Eq. (2.20) is satisfied while
$$\dot{H}_{\mathrm{inf}}>H_{\mathrm{inf}}^2.$$
(2.22)
The last expression reduces to an upper bound on the slow-roll parameter $`\epsilon `$ defined in Eq. (1.8):
$$\epsilon \left(\frac{\dot{H}_{\mathrm{inf}}}{H_{\mathrm{inf}}^2}\right)_{}<1.$$
(2.23)
Inflation is held as long as $`\epsilon <1`$, being one possibility when $`H_{\mathrm{inf}}`$ is constant in time. This scenario is commonly recognized as the de Sitter stage, and it will be useful at studying the spectrum of perturbations of a massless scalar field. Other possibilities correspond to nonvanishing values for $`\dot{H}_{\mathrm{inf}}`$, being the quasi de Sitter case ($`\epsilon 1`$) the most popular. Indeed, the quasi de Sitter stage supplemented by the condition $`|\eta _\phi |1`$, being $`\eta _\phi `$ the slow-roll parameter defined in Eq. (1.9), is what is known as slow-roll inflation. This variety of inflation will be discussed in Subsection 2.5.1. Meanwhile in the following subsection, we will study the fluctuations of a non-dominating massless scalar field during a de Sitter stage. Later on we will generalise these results to the fluctuations of non-dominating and dominating massive scalar fields in a quasi de Sitter stage.
#### 2.3.2 Spectrum of perturbations of a non-dominating massless scalar field during a de Sitter stage
In this subsection we will consider the effects of a de Sitter inflationary stage on the fluctuations of a massless scalar field that does not dominate the energy density. Let’s call this field $`\varphi `$. During inflation, and before horizon exit, the fluctuations in $`\varphi `$ can still be regarded as quantum operators. If we further assume that $`\varphi `$ is almost a non-interacting field, we can write down the field perturbation operator $`\delta \widehat{\varphi }(𝐱,t)`$ in terms of the usual creation and annihilation operators $`\widehat{a}_𝐤^{}`$ and $`\widehat{a}_𝐤`$:
$$\delta \widehat{\varphi }(𝐱,t)=\frac{d^3k}{(2\pi )^{3/2}}\mathrm{exp}(i𝐤𝐱)\delta \widehat{\varphi }_𝐤(t),$$
(2.24)
where
$$\delta \widehat{\varphi }_𝐤(t)\omega _k(t)\widehat{a}_𝐤+\omega _k^{}(t)\widehat{a}_𝐤^{}.$$
(2.25)
As it was discussed in the introduction of this thesis, the properties of the primordial curvature perturbation $`\zeta `$ are specified by the spectrum $`𝒫_\zeta (k)`$, which is defined by the statistical average $`\zeta _{𝐤_1}\zeta _{𝐤_2}`$ over an ensemble of universes. The same definition can be applied for the perturbations in $`\varphi `$, so that the only information we need to know to calculate $`𝒫_{\delta \varphi }(k)`$ is the quantum state of the Universe during inflation, being the most reasonable choice the vacuum state<sup>6</sup><sup>6</sup>6The vacuum state guarantees the homogeneity and isotropy in the whole space.. Since the universes in the ensemble are all in the vacuum state during inflation, the statistical average $`\delta \varphi _{𝐤_1}\delta \varphi _{𝐤_2}`$ is now very easy to calculate corresponding to the expectation value $`0|\delta \widehat{\varphi }_{𝐤_1}\delta \widehat{\varphi }_{𝐤_2}|0`$. It is straightforward then to recognize that the spectrum $`𝒫_{\delta \varphi }(k)`$ of the $`\delta \varphi `$ perturbations, defined by
$$\delta \varphi _{𝐤_1}\delta \varphi _{𝐤_2}\frac{2\pi ^2}{k^3}\delta ^3(𝐤_1+𝐤_2)𝒫_{\delta \varphi }(k),$$
(2.26)
is given by the simple formula
$$𝒫_{\delta \varphi }(k)=\frac{k^3}{2\pi ^2}|\omega _k|^2.$$
(2.27)
To calculate $`|\omega _k|`$ we need to solve the Klein-Gordon equation for the fluctuations in $`\varphi `$. The usual Klein-Gordon equation
$$\ddot{\omega }_k+3H_{}\dot{\omega }_k+\frac{k^2}{a^2}\omega _k=0,$$
(2.28)
is only applicable if the inflaton field (that which dominates the energy density) is unable to generate significant primordial perturbations. That guarantees that the line element in Eq. (2.2) is not modified by any scalar perturbation as $`\varphi `$ is assumed not to dominate the energy density during inflation. As a consequence, the usual structure of the Klein-Gordon equation is unmodified.
Eq. (2.28) is easily solved by making the following change of variables:
$$\omega _k\frac{\lambda _k}{a},$$
(2.29)
and going to conformal time where the expansion parameter in a de Sitter stage ($`H_{\mathrm{inf}}`$ being constant)
$$a(t)=a_{\mathrm{ini}}\mathrm{exp}[H_{}(tt_{\mathrm{ini}})],$$
(2.30)
is given by
$$a(\eta )=\frac{1}{H_{}\eta },$$
(2.31)
with conformal time $`\eta `$ taking negative values. Thus, the Klein-Gordon equation in Eq. (2.28) reduces to
$$\lambda _k^{\prime \prime }+\left(k^2\frac{2}{\eta ^2}\right)\lambda _k=0,$$
(2.32)
whose exact solution is
$$\lambda _k=\frac{\mathrm{exp}(ik\eta )}{\sqrt{2k}}\left(1\frac{i}{k\eta }\right).$$
(2.33)
The reader might worry about the fact that Eq. (2.33) is valid up to a multiplicative constant. To solve this ambiguity we note that in the subhorizon regime ($`kaH_{\mathrm{inf}}`$) Eq. (2.32) reproduces the Klein-Gordon equation for a massless scalar field in Minkowski spacetime. The solution for $`\lambda _k`$ in Eq. (2.33) satisfies the Bunch and Davies normalisation on subhorizon scales, so that no integration constant should amplify it.
The subhorizon regime is useful in the sense that we can find the adequate solution and normalisation for $`\lambda _k`$. However, the superhorizon regime ($`kaH_{\mathrm{inf}}`$) is much more interesting because $`\zeta `$ defined in that regime is truly gauge invariant and conserved as long as the pressure satisfies the adiabatic condition. From Eqs. (2.29), (2.31), and (2.33), the magnitude of the mode function $`\omega _k`$ for superhorizon scales turns out to be constant and it is given by
$$|\omega _k|\frac{H_{}}{\sqrt{2k^3}}.$$
(2.34)
The latter expression allows us to obtain the spectrum of perturbations in $`\varphi `$, given by Eq. (2.27):
$$𝒫_{\delta \varphi }(k)\left(\frac{H_{}}{2\pi }\right)^2.$$
(2.35)
As we can see, the spectrum $`𝒫_{\delta \varphi }(k)`$ of the perturbations of a massless scalar field $`\varphi `$ in a de Sitter stage is constant and scale invariant. We will see later on that the previous amplitude is generically the same for any kind of quasi de Sitter model as far as we deal with light fields. A possible scale dependence will arise if the field under consideration has a finite mass and/or if the inflationary stage is not completely de Sitter ($`\dot{H}_{\mathrm{inf}}0`$).
### 2.4 The curvaton scenario: an example of a non-dominating light scalar field during a quasi de Sitter stage
In the previous section we described the spectrum of perturbations of a non-dominating massless scalar field during an inflationary period with $`H_{\mathrm{inf}}`$ being constant (de Sitter stage). Now we move a step ahead considering a non-dominating scalar field $`\sigma `$ whose mass $`m_\sigma `$ satisfies $`m_\sigma H_{}`$ (light field), during an inflationary period where the Hubble parameter $`H_{\mathrm{inf}}`$ is not constant but evolves slowly satisfying the condition $`\dot{H}_{\mathrm{inf}}/H_{\mathrm{inf}}^21`$ so that $`\epsilon 1`$ (quasi de Sitter stage). We will see that the introduction of a slowly varying $`H_{\mathrm{inf}}`$ and a small mass $`m_\sigma `$ for the field $`\sigma `$ results in a small scale dependence for $`𝒫_{\delta \sigma }(k)`$ which is parameterised by $`\epsilon `$ and $`\eta _\sigma m_\sigma ^2/3H_{}^2`$. The amplitude of $`𝒫_{\delta \sigma }(k)`$ will turn out to be the same as that for a non-dominating massless scalar field in a de Sitter stage, which was the case described in Subsection 2.3.2. This example will help us to understand the properties of the curvature perturbation produced in one of the most satisfying models proposed to explain the origin of the large-scale structure in the Universe: the curvaton scenario.
#### 2.4.1 Spectrum of perturbations of a non-dominating light scalar field during a quasi de Sitter stage
The calculation of the spectrum of perturbations of a non-dominating light scalar field $`\sigma `$ during a quasi de Sitter stage closely resembles that done in Subsection 2.3.2. We will now have to consider the mass $`m_\sigma `$ of $`\sigma `$ and the running of $`H_{\mathrm{inf}}`$ given by the slow-roll parameter $`\epsilon `$. Since the inflaton field is supposed to produce negligible curvature perturbation, and $`\sigma `$ does not dominate the energy density during inflation, the line element is still given by Eq. (2.2). This means that the Klein-Gordon equation for the mode functions $`\omega _k`$ does not change compared with that for the mode functions of the background component of $`\sigma `$:
$$\ddot{\omega }_k+3H_{\mathrm{inf}}\dot{\omega }_k+\left(\frac{k^2}{a^2}+m_\sigma ^2\right)\omega _k=0.$$
(2.36)
In the quasi de Sitter stage the expansion is almost exponential; nevertheless it is better described by the following evolution equation:
$$a(t)=a_{\mathrm{ini}}[1+H_{\mathrm{inf}}(t_{\mathrm{ini}})\epsilon (tt_{\mathrm{ini}})]^{1/\epsilon },$$
(2.37)
where $`H_{\mathrm{inf}}(t_{\mathrm{ini}})`$ is the Hubble parameter at the time $`t_{\mathrm{ini}}`$. The last expression can be written down using the conformal time $`\eta `$ in an easier way:
$$a(\eta )=\frac{1}{H_{\mathrm{inf}}(\eta )\eta (1\epsilon )},$$
(2.38)
having in mind that $`\eta `$ takes negative values. Going to conformal time and making the change of variables
$$\omega _k\frac{\lambda _k}{a},$$
(2.39)
the following equation of motion for $`\lambda _k`$ is obtained:
$$\lambda _k^{\prime \prime }+\left[k^2\frac{1}{\eta ^2}\left(\upsilon _\sigma ^2\frac{1}{4}\right)\right]\lambda _k=0,$$
(2.40)
where
$`\upsilon _\sigma `$ $``$ $`\left[{\displaystyle \frac{1}{4}}{\displaystyle \frac{3\eta _\sigma 2+\epsilon }{(1\epsilon )^2}}\right]^{1/2}`$ (2.41)
$``$ $`{\displaystyle \frac{3}{2}}+\epsilon \eta _\sigma .`$
The solution for such an equation is given in terms of the Hankel’s functions of the first and second kind $`H_{\upsilon _\sigma }^{(1)}`$ and $`H_{\upsilon _\sigma }^{(2)}`$ :
$$\lambda _k=\sqrt{\eta }[c_1(k)H_{\upsilon _\sigma }^{(1)}(k\eta )+c_2(k)H_{\upsilon _\sigma }^{(2)}(k\eta )],$$
(2.42)
where $`c_1(k)`$ and $`c_2(k)`$ are integration constants that are determined going to the subhorizon regime ($`kaH_{\mathrm{inf}}`$), which corresponds to $`k\eta 1`$, and normalising the solution according to Bunch and Davies <sup>7</sup><sup>7</sup>7The requirement of a solution for $`\lambda _k`$ normalised a la Bunch and Davis is justified because Eq. (2.40) on subhorizon scales reduces to the Klein-Gordon equation for a massless scalar field in Minkowski spacetime.. Indeed, taking into account that in the subhorizon regime the Hankel’s functions are well approximated by
$`H_{\upsilon _\sigma }^{(1)}(k\eta 1)`$ $``$ $`\sqrt{{\displaystyle \frac{2}{\pi k\eta }}}\mathrm{exp}\left[i\left(k\eta +{\displaystyle \frac{\pi }{2}}\upsilon _\sigma +{\displaystyle \frac{3\pi }{4}}\right)\right],`$ (2.43)
$`H_{\upsilon _\sigma }^{(2)}(k\eta 1)`$ $``$ $`\sqrt{{\displaystyle \frac{2}{\pi k\eta }}}\mathrm{exp}\left[i\left(k\eta +{\displaystyle \frac{\pi }{2}}\upsilon _\sigma +{\displaystyle \frac{3\pi }{4}}\right)\right],`$ (2.44)
the Bunch and Davies normalisation
$$\lambda _k=\frac{\mathrm{exp}(ik\eta )}{\sqrt{2k}},$$
(2.45)
is obtained in the subhorizon regime by choosing the following values for the integration constants:
$`c_1(k)`$ $`=`$ $`{\displaystyle \frac{\sqrt{\pi }}{2}}\mathrm{exp}\left[i\left(\upsilon _\sigma +{\displaystyle \frac{1}{2}}\right){\displaystyle \frac{\pi }{2}}\right],`$ (2.46)
$`c_2(k)`$ $`=`$ $`0.`$ (2.47)
Thus, the exact solution in Eq. (2.42) is rewritten as
$$\lambda _k=\frac{\sqrt{\pi }}{2}\mathrm{exp}\left[i\left(\upsilon _\sigma +\frac{1}{2}\right)\frac{\pi }{2}\right]\sqrt{\eta }H_{\upsilon _\sigma }^{(1)}(k\eta ).$$
(2.48)
To find out the spectrum of $`\delta \sigma `$ on superhorizon scales ($`kaH_{\mathrm{inf}}`$), corresponding to $`k\eta 1`$, we need to consider the behaviour of $`\lambda _k`$ in that regime. This is given by the following approximation for the Hankel’s function of the first kind on superhorizon scales
$$H_{\upsilon _\sigma }^{(1)}(k\eta 1)\sqrt{\frac{2}{\pi }}\mathrm{exp}\left(i\frac{\pi }{2}\right)2^{\upsilon _\sigma \frac{3}{2}}\frac{\mathrm{\Gamma }(\upsilon _\sigma )}{\mathrm{\Gamma }(3/2)}(k\eta )^{\upsilon _\sigma }.$$
(2.49)
The magnitude of the mode function $`\omega _k`$ on superhorizon scales is then almost constant and approximately given by
$`|\omega _k|`$ $``$ $`[2(1\epsilon )]^{\upsilon _\sigma \frac{3}{2}}(1\epsilon ){\displaystyle \frac{\mathrm{\Gamma }(\upsilon _\sigma )}{\mathrm{\Gamma }(3/2)}}{\displaystyle \frac{H_{\mathrm{inf}}}{\sqrt{2k^3}}}\left({\displaystyle \frac{k}{aH_{\mathrm{inf}}}}\right)^{\frac{3}{2}\upsilon _\sigma }`$ (2.50)
$``$ $`{\displaystyle \frac{H_{}}{\sqrt{2k^3}}}\left({\displaystyle \frac{k}{aH_{\mathrm{inf}}}}\right)^{\eta _\sigma \epsilon }.`$
where we have used the relation involving $`\upsilon _\sigma `$, $`\epsilon `$, and $`\eta _\sigma `$, established in Eq. (2.41).
Making use of the expression in Eq. (2.27), the spectrum of perturbations in $`\sigma `$ is finally written down as
$$𝒫_{\delta \sigma }(k)\left(\frac{H_{}}{2\pi }\right)^2\left(\frac{k}{aH_{\mathrm{inf}}}\right)^{n_{\delta \sigma }},$$
(2.51)
whose spectral index $`n_{\delta \sigma }`$ is given by
$$n_{\delta \sigma }=2\eta _\sigma 2\epsilon .$$
(2.52)
Comparing the previous result with that found for the case of a non-dominating massless scalar field during a de Sitter stage \[c.f. Eq. (2.35)\], we see that the amplitude of $`𝒫_{\delta \sigma }(k)`$ is exactly the same for both cases, but now we have a small scale dependence parameterised by the slow evolution of $`H_{\mathrm{inf}}`$ and the mass $`m_\sigma `$ of the light field $`\sigma `$, characterized by the smallness of the parameters $`\epsilon `$ and $`\eta _\sigma `$.
#### 2.4.2 The curvaton scenario
In the basic curvaton setup the Hubble parameter during inflation $`H_{\mathrm{inf}}`$ is assumed to be slowly varying ($`\epsilon 1`$), the curvaton energy density $`\rho _\sigma `$ during inflation is assumed to be negligible, and the inflaton field is supposed to produce a negligible curvature perturbation $`\zeta _r`$ which is imprinted to the radiation fluid while the inflaton decays. After the end of inflation the Universe is composed by the almost unperturbed radiation fluid that originates from the reheating process following the inflaton decay, and the weakly coupled light curvaton field $`\sigma `$ <sup>8</sup><sup>8</sup>8The curvaton field must be weakly coupled to avoid premature thermalisation, which would erase all the information about the generatated curvature perturbation . whose unperturbed component is kept frozen at a value $`\sigma _{}`$ until the Hubble parameter $`H`$ becomes of the order of the curvaton mass $`m_\sigma `$ during the radiation dominated epoch. Once $`\sigma `$ is unfrozen, it begins oscillating around the minimum of its potential, which is taken to be quadratic, with an oscillation period which rapidly becomes much less than the characteristic expansion time scale. This ensures that the average curvaton pressure vanishes and, therefore, $`\sigma `$ may be considered as a matter fluid . During the oscillatory period, the curvaton energy density $`\rho _\sigma `$ decreases with time according to $`\rho _\sigma a^3`$, while the radiation energy density $`\rho _r`$ decreases with time faster than $`\rho _\sigma `$ according to $`\rho _ra^4`$. Eventually the curvaton will decay, but by that time the contribution of $`\rho _\sigma `$ to the total energy density will be big enough for the original isocurvature perturbation $`\zeta _\sigma `$, generated by $`\sigma `$ during inflation and which is not negligible, to become the total curvature perturbation $`\zeta `$.
Since the oscillations of $`\sigma `$ around the minimum of its potential are so fast, we can approximate its energy density by
$$\rho _\sigma (t,𝐱)\frac{1}{2}m_\sigma ^2\sigma _a^2(t,𝐱),$$
(2.53)
where $`\sigma _a(t,𝐱)`$ is the amplitude of the oscillations. Notice that, under these circumstances, the expression for the curvaton energy density in Eq. (2.53) corresponds also to the expression for the curvaton potential $`V(\sigma )`$. The no appearance of quartic or higher order terms in the potential is essential for the success of the model because, otherwise, the density ratio $`\rho _\sigma /\rho _r`$ would not increase with time .
Making use of the curvature perturbation definition in Eq. (2.15), we can write down the total $`\zeta `$ in the curvaton scenario as:
$$\zeta \psi H\left(\frac{\delta \rho }{\dot{\rho }_0}\right)_{\mathrm{total}},$$
(2.54)
where the total energy density $`\rho _{\mathrm{total}}`$ is simply the addition of the curvaton and radiation energy densities $`\rho _\sigma `$ and $`\rho _r`$, that define the conserved curvaton and radiation curvature perturbations $`\zeta _\sigma `$ and $`\zeta _r`$ <sup>9</sup><sup>9</sup>9The curvature perturbations $`\zeta _\sigma `$ and $`\zeta _r`$, associated to the curvaton and the radiation fluids respectively, are conserved since the adiabatic condition is satisfied separately being both fluids non-interacting.:
$`\zeta _\sigma `$ $``$ $`\psi H{\displaystyle \frac{\delta \rho _\sigma }{\dot{\rho }_{\sigma _0}}}=\psi +{\displaystyle \frac{1}{3}}{\displaystyle \frac{\delta \rho _\sigma }{\rho _{\sigma _0}}},`$ (2.55)
$`\zeta _r`$ $``$ $`\psi H{\displaystyle \frac{\delta \rho _r}{\dot{\rho }_{r_0}}}=\psi +{\displaystyle \frac{1}{4}}{\displaystyle \frac{\delta \rho _r}{\rho _{r_0}}}.`$ (2.56)
In the above expressions we have employed the background continuity equation
$$\dot{\rho }_0+3H(\rho _0+P_0)=0,$$
(2.57)
where $`P_0=0`$ for a matter fluid, and $`P_0=\rho _0/3`$ for a radiation fluid. Combining Eqs. (2.54), (2.55), and (2.56), the total curvature perturbation $`\zeta `$ can then be written down as the weighted sum
$$\zeta =(1r)\zeta _r+r\zeta _\sigma ,$$
(2.58)
with modulation factor
$$r\frac{3\rho _{\sigma _0}}{4\rho _{r_0}+3\rho _{\sigma _0}}.$$
(2.59)
Notice that just at the beginning of the radiation dominated epoch that follows the reheating stage produced by the inflaton decay, $`r`$ is almost zero since $`\rho _\sigma `$ is negligible by that time; therefore $`\zeta \zeta _r`$ which is negligible. However, $`r`$ grows in time due to the relative redshifting between $`\rho _\sigma `$ and $`\rho _r`$ until when eventually $`\sigma `$ decays. In view of Eqs. (2.58) and (2.59), the total $`\zeta `$ grows then in time approaching more and more to the curvaton curvature perturbation $`\zeta _\sigma `$. One extreme example is when $`\sigma `$ has dominated the energy density before decaying; in that case $`r1`$ and therefore $`\zeta \zeta _\sigma `$. When $`\sigma `$ decays, $`\zeta `$ is imprinted in remaining radiation fluid starting this way the gravitational instability process that ends up with the presently observed large-scale structure.
As we have already pointed out, one of the requirements of the curvaton scenario is that the curvature perturbation produced by the inflaton during inflation $`\zeta _r`$ is completely negligible compared with that produced by the curvaton $`\zeta _\sigma `$ during the same period: $`\zeta _r\zeta _\sigma `$. Under that assumption, the expression for the total $`\zeta `$ after $`\sigma `$ decays comes from Eq. (2.58) as
$$\zeta r\zeta _\sigma ,$$
(2.60)
in the sudden decay approximation . For a model that goes beyond this approximation the expression for the total $`\zeta `$ in terms of $`\zeta _\sigma `$ is only obtained by means of numerical calculations, the result being in that case
$$\zeta \mathrm{\Omega }_{\mathrm{dec}}\zeta _\sigma ,$$
(2.61)
where $`\mathrm{\Omega }_{\mathrm{dec}}`$ is the fractional global curvaton energy density just before the curvaton decay:
$$\mathrm{\Omega }_{\mathrm{dec}}=\left(\frac{\rho _{\sigma _0}}{\rho _{\mathrm{total}_0}}\right)_{\mathrm{dec}}.$$
(2.62)
As we mentioned in Chapter 1, and will explain in Chapter 5, the normalisation $`f_{\mathrm{NL}}`$ of the bispectrum $`B_\zeta (k_1,k_2,k_3)`$ of $`\zeta `$ in the curvaton scenario is directly related to $`\mathrm{\Omega }_{\mathrm{dec}}`$ if the latter is not so close to 1 :
$$f_{\mathrm{NL}}\frac{5}{4\mathrm{\Omega }_{\mathrm{dec}}}.$$
(2.63)
The parameter $`f_{\mathrm{NL}}`$ gives information about the level of non-gaussianity present in $`\zeta `$, and the actual bound on it, coming from WMAP data , is $`|f_{\mathrm{NL}}|\text{ }\stackrel{<}{}\text{ }10^2`$. This bound translates into a lower bound for $`\mathrm{\Omega }_{\mathrm{dec}}`$, that combined with the obvious energy density condition $`\mathrm{\Omega }_{\mathrm{dec}}1`$, gives the allowed range
$$0.01\text{ }\stackrel{<}{}\text{ }\mathrm{\Omega }_{\mathrm{dec}}1.$$
(2.64)
The present lower bound on $`\mathrm{\Omega }_{\mathrm{dec}}`$ is likely to be increased by the next WMAP data release or the future PLANCK satellite data if non-gaussianity effects are not detected.
Once we have studied how the curvature perturbation is produced in the curvaton scenario, we proceed now to study the spectrum $`𝒫_\zeta (k)`$ of $`\zeta `$. In view of Eq. (2.53), and having in mind that the equation of motion for $`\delta \sigma _a`$ is the same as that for the background field $`\sigma _{a_0}`$, throughout inflation and during the post-inflationary period, as long as the non-gauge invariant curvature perturbation $`\psi `$ is fixed to be zero<sup>10</sup><sup>10</sup>10That is indeed the case while the curvature perturbation $`\zeta _r`$ in the radiation fluid is taken to be negligible , which is one of the key assumptions in the curvaton scenario., we can relate the contrast in the energy density of $`\sigma `$ at any time $`t`$ with the contrast in $`\sigma `$ some time after horizon exit but before the onset of the curvaton oscillations:
$$\frac{\delta \rho _\sigma }{\rho _{\sigma _0}}2\left(\frac{\delta \sigma _a}{\sigma _{a_0}}\right)2\frac{\delta \sigma }{\sigma _{}}.$$
(2.65)
From Eqs. (2.55), (2.61), and (2.65), $`\zeta `$ is expressed in terms of the perturbations in $`\sigma `$ a few Hubble times after horizon exit:
$$\zeta \frac{2}{3}\mathrm{\Omega }_{\mathrm{dec}}\frac{\delta \sigma }{\sigma _{}},$$
(2.66)
and in consequence the spectrum $`𝒫_\zeta (k)`$ is given by
$$𝒫_\zeta (k)\frac{4}{9}\mathrm{\Omega }_{\mathrm{dec}}^2\frac{𝒫_{\delta \sigma }(k)}{\sigma _{}^2}.$$
(2.67)
The curvaton field $`\sigma `$ is a light field whose energy density is negligible during inflation; therefore the discussion and results of Subsection 2.4.1 apply \[c.f. Eqs. (2.51) and (2.52)\], giving as a result
$$𝒫_\zeta (k)A_\zeta ^2\left(\frac{k}{aH_{\mathrm{inf}}}\right)^{n_\zeta }\left[\frac{H_{}\mathrm{\Omega }_{\mathrm{dec}}}{3\pi \sigma _{}}\right]^2\left(\frac{k}{aH_{\mathrm{inf}}}\right)^{2\eta _\sigma 2\epsilon }.$$
(2.68)
The spectral index $`n_\zeta `$ is in good agreement with observation, which requires an almost-scale invariant power spectrum :
$$0.048<n_\zeta <0.016.$$
(2.69)
Unfortunately the Hubble parameter a few Hubble times after horizon exit $`H_{}`$, which gives information about the inflationary energy scale, is not predicted by the amplitude of $`𝒫_\zeta (k)`$ since $`\sigma _{}`$ is an unknown and unbounded parameter. Nevertheless, a lower bound for $`H_{}`$ will be found in Chapter 3 by taking into consideration other effects that give a different relation between $`H_{}`$ and $`\sigma _{}`$, and the amplitude $`A_\zeta `$ in Eq. (2.68) once the WMAP normalisation ($`|A_\zeta |5\times 10^5`$) is taken into account:
$$\sigma _{}(1.5\pi \times 10^4)^1\mathrm{\Omega }_{\mathrm{dec}}H_{}.$$
(2.70)
What is important however to emphasise at this point is that the biggest possible value for $`H_{}`$ in the curvaton scenario is for sure $`10^{12}`$ GeV. Otherwise the curvature perturbation $`\zeta _r`$, produced by the inflaton field during inflation, would contribute significantly to the total $`\zeta `$, spoiling the main motivation for the proposal of the curvaton scenario<sup>11</sup><sup>11</sup>11The only way to have $`H_{}>10^{12}`$ GeV in the curvaton scenario while making $`\zeta _r`$ negligible is by requiring the inflaton field not to be light during inflation ($`m_\phi H_{}`$) . A non slow-roll inflationary model is in that case compulsory.. The justification of this assertion will be given in the following section.
### 2.5 The inflaton scenario
In this section we will discuss the main facts about the inflaton scenario where inflation is assumed to be of the slow-roll variety . Slow-roll inflation corresponds to the case where the inflaton field $`\phi `$ slowly-roll down towards the minimum of its potential. We will specify the slow-roll conditions and see what their consequences are on the shape of the inflaton potential as well as on the value and structural form of the slow-roll parameters $`\epsilon `$ and $`\eta _\phi `$. Being $`\phi `$ in the inflaton scenario the responsible of driving inflation and also of generating the curvature perturbation $`\zeta `$, the power spectrum $`𝒫_\zeta (k)`$ of $`\zeta `$ presents definite signatures that are expressed in terms of $`\epsilon `$ and $`\eta _\phi `$. We will calculate $`𝒫_\zeta (k)`$ and see what the constraints on the inflaton potential are in order to produce enough primordial perturbations. The Hubble parameter during inflation $`H_{}`$ will turn out to be likely quite high ($`H_{}\text{ }\stackrel{<}{}\text{ }10^{14}`$ GeV) for the inflaton scenario to be consistent with the amplitude of perturbations observed by WMAP. The scale of inflation is, therefore, likely high enough to impose severe constraints on concrete inflation models .
#### 2.5.1 The slow-roll conditions
We begin by considering the Friedmann and continuity equations, derived from the background Einstein equations for the FRW cosmological model , that relate the Hubble parameter at any time with the global energy density and pressure of the fluid that fills the Universe:
$`H^2`$ $`=`$ $`{\displaystyle \frac{\rho _0}{3m_P^2}},`$ (2.71)
$`\dot{\rho }_0`$ $`=`$ $`3H(\rho _0+P_0).`$ (2.72)
A direct consequence of both equations is that the second derivative of the global expansion parameter $`a`$ with respect to the cosmic time is given by a simple relation involving $`\rho _0`$ and $`P_0`$:
$$\frac{\ddot{a}}{a}=\frac{\rho _0+3P_0}{6m_P^2}.$$
(2.73)
This expression tells us that, to satisfy the inflationary condition $`\ddot{a}>0`$, the pressure of the fluid that fills the Universe must be negative satisfying
$$\rho _0+3P_0<0.$$
(2.74)
As an application of the above formula we may study the dynamics of the inflaton field $`\phi `$ knowing that, from the energy momentum tensor for a homogeneous scalar field , the unperturbed energy density and pressure of $`\phi `$ are given by
$`\rho _{\phi _0}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\dot{\phi }_0^2+V(\phi _0),`$ (2.75)
$`P_{\phi _0}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\dot{\phi }_0^2V(\phi _0).`$ (2.76)
The inflationary condition in Eq. (2.74) is then satisfied provided that
$$\dot{\phi }_0^2<V(\phi _0).$$
(2.77)
The most popular type of inflationary models assume that the kinetic energy density of the inflaton field is much less than the potential energy density:
$$\frac{1}{2}\dot{\phi }_0^2V(\phi _0),$$
(2.78)
which corresponds intuitively to a very flat potential along which the inflaton field $`\phi `$ slowly-roll down during inflation towards the minimum of its potential . If the expression in Eq. (2.78) is supplemented by the condition
$$|\ddot{\phi }_0||3H_{\mathrm{inf}}\dot{\phi }_0|,$$
(2.79)
the inflaton field $`\phi `$ satisfies what is known as the slow-roll conditions . As we will see, these conditions can be expressed in terms of the parameters $`\epsilon `$ and $`\eta _\phi `$ that parameterise the spectral index and amplitude of $`𝒫_\zeta (k)`$ in the inflaton scenario. Notice that, under the slow-roll conditions, the background field $`\phi _0`$ follows the slow-roll equation of motion
$$3H_{\mathrm{inf}}\dot{\phi _0}\frac{V}{\phi _0},$$
(2.80)
which corresponds to the background Klein-Gordon equation under the condition given by Eq. (2.79).
As discussed in Subsection 2.3.1, the requirement to have a period of accelerated expansion is easily expressed as an upper bound on the slow-roll parameter $`\epsilon `$ that describes the rate of change of the Hubble parameter a few Hubble times after horizon exit:
$$\epsilon \left(\frac{\dot{H}_{\mathrm{inf}}}{H_{\mathrm{inf}}^2}\right)_{}<1.$$
(2.81)
The true reason why $`\epsilon `$ is called a slow-roll parameter is because it is constrained to be much less than 1 under the slow-roll conditions in Eqs. (2.78) and (2.79), being easily expressed in terms of the unperturbed inflaton potential $`V(\phi _0)`$ and its derivative with respect to $`\phi `$ :
$$\epsilon \frac{m_P^2}{2V^2}\left(\frac{V}{\phi _0}\right)^21.$$
(2.82)
The flatness condition on the potential $`V(\phi )`$ required for $`\phi `$ to slowly-roll during inflation is evident from the above expression.
Two slow-roll conditions (Eqs. (2.78) and (2.79)) require constraints on two slow-roll parameters. One of them is that given in Eq. (2.82) in terms of $`\epsilon `$; the other one is given in terms of the parameter $`\eta _\phi `$ already defined in Eq. (1.9):
$$\eta _\phi \epsilon \frac{\ddot{\phi }_0}{H_{\mathrm{inf}}\dot{\phi }_0}.$$
(2.83)
The respective constraint on $`\eta _\phi `$ and its relation with $`V(\phi )`$ are obtained once we take into consideration the slow-roll conditions in Eqs. (2.78) and (2.79) :
$$|\eta _\phi |\left|\frac{m_P^2}{V}\frac{^2V}{\phi _0^2}\right|1.$$
(2.84)
This relation again shows how flat the potential of the inflaton field ought to be to drive inflation. This is particularly good in order to generate enough inflation as $`\phi `$ spends a lot of time rolling along the flat part of its potential, which is in turn perhaps the main motivation to have an inflationary slow-roll model.
From the practical point of view, inflation is said to start when $`V(\phi )`$ satisfies both Eqs. (2.82) and (2.84), and ends when any of them is violated. Let’s however remember that, in any case, the slow-roll conditions are sufficient but not necessary to drive inflation. Strictly speaking inflation may end some time after the slow-roll conditions are violated, but this time is very small compared with the 70 e-folds or so of inflation required to solve the horizon, flatness, and unwanted relics problems, under standard evolution.
#### 2.5.2 Spectrum of perturbations of a dominating light scalar field during a quasi de Sitter stage
When we consider a scalar field $`\phi `$ that dominates the energy density during inflation and whose perturbations are sizable enough, the spacetime stops being perfectly smooth so that we have to leave the comfortable unperturbed metric line element in Eq. (2.2) to adopt the perturbed line element described in Eq. (2.3). As a result the Klein-Gordon equation for the Fourier modes $`\omega _k`$ of the perturbations in $`\phi `$ is modified to take into account the backreaction of the metric :
$$\ddot{\omega }_k+3H_{\mathrm{inf}}\dot{\omega }_k+\left(\frac{k^2}{a^2}+\frac{^2V}{\phi _0^2}\right)\omega _k=2\varphi _{G_k}\frac{V}{\phi _0}+\dot{\varphi }_{G_k}\dot{\phi }_0+3\dot{\psi }_k\dot{\phi }_0k^2B_k\dot{\phi }_0.$$
(2.85)
The above equation looks quite difficult to manage but, fortunately, we can eliminate some of the scalar degrees of freedom by fixing the gauge and using the perturbed Einstein equations for the inflaton field $`\phi `$ . For instance, going to the longitudinal gauge, we can fix the scalar perturbations $`B`$ and $`E`$ to be zero in the metric line element of Eq. (2.3), whereas the non-diagonal part of the $`ij`$ component of the perturbed Einstein equations requires $`\varphi _G=\psi `$ being the stress associated to $`\phi _0`$ completely isotropic. The modified Klein-Gordon equation reduces in this case to
$$\ddot{\omega }_k+3H_{\mathrm{inf}}\dot{\omega }_k+\left(\frac{k^2}{a^2}+\frac{^2V}{\phi _0^2}\right)\omega _k=2\psi _k\frac{V}{\phi _0}+4\dot{\psi }_k\dot{\phi }_0.$$
(2.86)
To solve the previous equation we still require to know the behaviour of $`\psi `$. To that aim we take advantage of the $`00`$, $`0i`$, and the diagonal part of the $`ij`$ components of the perturbed Einstein equations in the longitudinal gauge :
$`3H_{\mathrm{inf}}(\dot{\psi }+H_{\mathrm{inf}}\psi )+{\displaystyle \frac{^2\psi }{a^2}}`$ $`=`$ $`{\displaystyle \frac{1}{2m_P^2}}\left(\dot{\phi }_0\delta \dot{\phi }\dot{\phi }_0^2\psi +{\displaystyle \frac{V}{\phi _0}}\delta \phi \right),`$ (2.87)
$`\dot{\psi }+H_{\mathrm{inf}}\psi `$ $`=`$ $`{\displaystyle \frac{1}{2m_P^2}}\dot{\phi }_0\delta \phi ,`$ (2.88)
$`\left(2{\displaystyle \frac{\ddot{a}}{a}}+H_{\mathrm{inf}}^2\right)\psi +4H_{\mathrm{inf}}\dot{\psi }+\ddot{\psi }`$ $`=`$ $`{\displaystyle \frac{1}{2m_P^2}}\left(\dot{\phi }_0\delta \dot{\phi }\dot{\phi }_0^2\psi {\displaystyle \frac{V}{\phi _0}}\delta \phi \right),`$ (2.89)
which combined give the following equation for $`\psi _k`$ in terms of the slow-roll parameters $`\epsilon `$ and $`\eta `$:
$$\ddot{\psi }_k+H_{\mathrm{inf}}(1+2\eta 2\epsilon )\dot{\psi }_k+2H_{\mathrm{inf}}^2(\eta 2\epsilon )\psi _k+\frac{k^2}{a^2}\psi _k=0.$$
(2.90)
A quick look at the previous expression reveals that, on superhorizon scales, $`\psi `$ behaves as $`\dot{\psi }_k2(2\epsilon \eta )H_{\mathrm{inf}}\psi _k`$ so that $`|4\dot{\psi }_k\dot{\phi }_0||2\psi _kV/\phi _0|`$, whereas on subhorizon scales $`\psi _k0`$. In view of this, and by making use of the $`0i`$ component of the perturbed Einstein equations \[c.f. Eq. (2.88)\], which may also be written down as
$$\dot{\psi }+H_{\mathrm{inf}}\psi =\epsilon H_{\mathrm{inf}}^2\frac{\delta \phi }{\dot{\phi }_0},$$
(2.91)
we conclude that, on superhorizon scales,
$$\psi _k\frac{\epsilon H_{\mathrm{inf}}\omega _k}{\dot{\phi }_0},$$
(2.92)
so that the equation of motion for $`\omega _k`$ is finally given by
$`\ddot{\omega }_k+3H_{\mathrm{inf}}\dot{\omega }_k+\left({\displaystyle \frac{k^2}{a^2}}+{\displaystyle \frac{^2V}{\phi _0^2}}\right)\omega _k=\left\{\begin{array}{c}0,\mathrm{for}kaH_{\mathrm{inf}},\\ 6\epsilon H_{\mathrm{inf}}^2\omega _k,\mathrm{for}kaH_{\mathrm{inf}}.\end{array}\right\}`$ (2.95)
This kind of differential equation is much more familiar to us, and we know that it can be solved going to conformal time and making the usual change of variables
$$\lambda _k\frac{\omega _k}{a}.$$
(2.96)
The resultant equation of motion for $`\lambda _k`$ is then
$$\lambda _k^{\prime \prime }+\left[k^2\frac{1}{\eta ^2}\left(\upsilon _\phi ^2\frac{1}{4}\right)\right]\lambda _k=0,$$
(2.97)
with $`\upsilon _\phi `$ defined by
$`\upsilon _\phi `$ $``$ $`\left[{\displaystyle \frac{1}{4}}{\displaystyle \frac{3\eta _\phi 25\epsilon }{(1\epsilon )^2}}\right]^{1/2}`$ (2.98)
$``$ $`{\displaystyle \frac{3}{2}}+3\epsilon \eta _\phi .`$
The solution for this equation is immediate, based on the results found in Subsection 2.4.1. The magnitude of the mode function $`\omega _k`$ on superhorizon scales is then almost constant and given by
$`|\omega _k|`$ $``$ $`[2(1\epsilon )]^{\upsilon _\phi \frac{3}{2}}(1\epsilon ){\displaystyle \frac{\mathrm{\Gamma }(\upsilon _\phi )}{\mathrm{\Gamma }(3/2)}}{\displaystyle \frac{H_{\mathrm{inf}}}{\sqrt{2k^3}}}\left({\displaystyle \frac{k}{aH_{\mathrm{inf}}}}\right)^{\frac{3}{2}\upsilon _\phi }`$ (2.99)
$``$ $`{\displaystyle \frac{H_{}}{\sqrt{2k^3}}}\left({\displaystyle \frac{k}{aH_{\mathrm{inf}}}}\right)^{\eta _\phi 3\epsilon },`$
which is used to calculate the spectrum $`𝒫_{\delta \phi }(k)`$ of perturbations in the inflaton field $`\phi `$ by means of Eq. (2.27):
$$𝒫_{\delta \phi }(k)\left(\frac{H_{}}{2\pi }\right)^2\left(\frac{k}{aH_{\mathrm{inf}}}\right)^{n_{\delta \phi }}.$$
(2.100)
The spectrum of perturbations in $`\phi `$ is almost scale-invariant with spectral index $`n_{\delta \phi }`$ given in terms of the slow-roll parameters $`\epsilon `$ and $`\eta _\phi `$:
$$n_{\delta \phi }=2\eta _\phi 6\epsilon .$$
(2.101)
Comparing the spectrum obtained \[c.f. Eqs. (2.100) and (2.101)\] with that for a non-dominating scalar field \[c.f. Eqs. (2.51) and (2.52)\], we see that the backreaction of the metric only affects the spectral index of the spectrum of perturbations. The amplitude remains the same either the respective field dominates the energy density or not.
#### 2.5.3 The spectrum of $`\zeta `$ in the inflaton scenario
Now we are in position to calculate the spectrum of the curvature perturbation $`\zeta `$ in the inflaton scenario, based on the results found in the previous subsection. We first begin by invoking the definition of $`\zeta `$ given in Eq. (2.13) in terms of the $`\phi `$ field:
$$\zeta =\psi H_{\mathrm{inf}}\frac{\delta \phi }{\dot{\phi }_0},$$
(2.102)
which, on superhorizon scales, reduces to
$$\zeta _k=(1+\epsilon )H_{\mathrm{inf}}\frac{\omega _k}{\dot{\phi }_0}H_{\mathrm{inf}}\frac{\omega _k}{\dot{\phi }_0},$$
(2.103)
where the expression in Eq. (2.92) has been used.
The spectrum $`𝒫_\zeta (k)`$ of $`\zeta `$ is, in view of the latter, given in terms of the spectrum $`𝒫_{\delta \phi }(k)`$ of the perturbations in $`\phi `$:
$$𝒫_\zeta (k)\left(\frac{H_{}}{\dot{\phi }_0}\right)^2𝒫_{\delta \phi }(k),$$
(2.104)
which, according to Eq. (2.100), gives the final expression
$`𝒫_\zeta (k)A_\zeta ^2\left({\displaystyle \frac{k}{aH_{\mathrm{inf}}}}\right)^{n_\zeta }`$ $``$ $`\left[{\displaystyle \frac{H_{}^2}{2\pi \dot{\phi }_0}}\right]^2\left({\displaystyle \frac{k}{aH_{\mathrm{inf}}}}\right)^{2\eta _\phi 6\epsilon }`$ (2.105)
$`=`$ $`\left[{\displaystyle \frac{H_{}}{\sqrt{8\epsilon }\pi m_P}}\right]^2\left({\displaystyle \frac{k}{aH_{\mathrm{inf}}}}\right)^{2\eta _\phi 6\epsilon },`$
where, in the second line, the amplitude $`A_\zeta `$ is written down in terms of $`\epsilon `$.
In the inflaton scenario $`\phi `$ is responsible of driving inflation and also of generating the required level of primordial perturbations measured by WMAP ($`|A_\zeta |5\times 10^5`$ ). That imposes the following constraint on the Hubble parameter during inflation $`H_{}`$ in terms of $`\epsilon `$:
$$H_{}10^{15}\sqrt{\epsilon }\mathrm{GeV},$$
(2.106)
that combined with the present bound $`\epsilon \text{ }\stackrel{<}{}\text{ }0.01`$ coming from spectral index and gravitational waves constraints requires $`H_{}\text{ }\stackrel{<}{}\text{ }10^{14}`$ GeV . Despite the fact that the inflationary energy scale, directly related to the value of $`H_{}`$, is in the inflaton scenario regulated by the parameter $`\epsilon `$, low-scale inflation may well be obtained but only at the expense of a very small $`\epsilon `$, which in turn requires a high level of fine-tuning (see however Ref. ). As a consequence serious problems appear when trying to build successful particle physics inflationary models . The relevance of finding such a kind of low-scale inflation models is evident since the inflaton field could be identified with one of the MSSM flat directions or one of the scalar fields in the SUSY breaking sector (see for example Refs. ).
Going back to the curvaton scenario, we said that one of the assumptions of the model was a negligible curvature perturbation generated by the inflaton. This assumption may be quantified by requiring $`\zeta _r`$ to be, say, at most $`1\%`$ of the total $`\zeta `$, which means, from Eq. (2.105), that $`H_{}`$ in the curvaton scenario must satisfy
$$H_{}\text{ }\stackrel{<}{}\text{ }10^{13}\sqrt{\epsilon }\mathrm{GeV}\text{ }\stackrel{<}{}\text{ }10^{12}\mathrm{GeV}.$$
(2.107)
In the following section we will show how such an upper bound on $`H_{}`$ makes the detection of gravitational waves an anti-smoking gun for the curvaton scenario.
### 2.6 Gravitational waves
Primordial tensor-type perturbations in spacetime are regarded as gravitational waves, being unsourced during inflation and susceptible to be decomposed in a polarisation tensor basis. They also propagate in a way that each subhorizon mode function follows a harmonic wave equation of motion, i.e. the Klein-Gordon equation associated to a massless scalar field in Minkowski spacetime.
The relevant quantity to discuss now is $`\mathrm{\Pi }_{ij}^T`$, which is the tensor component of the full perturbed metric tensor, and that, as shown in Subsection 2.2.1, is characterized by two degrees of freedom in our three dimensional space as well as by satisfying the transversality condition $`_i\mathrm{\Pi }_{ij}^T=0`$. Hereafter we will write $`\mathrm{\Pi }_{ij}^T`$ as $`2h_{ij}`$ as it is done in most of the literature. Displaying only the tensor perturbation, the metric line element in conformal time reads then
$$ds^2=a^2(\eta )[d\eta ^2+(\delta _{ij}+2h_{ij}(\eta ,𝐱))dx^idx^j].$$
(2.108)
Tensor perturbations are decoupled from their scalar and vector counterparts. The line element in Eq. (2.108) shows then that $`h_{ij}`$ is a gauge invariant quantity.
The Einstein-Hilbert action involving $`h_{ij}`$, and given in a general way by
$$S_E\frac{m_P^2}{2}d^4x(g)^{1/2}R,$$
(2.109)
being $`g`$ the determinant of the $`g_{\mu \nu }`$ metric tensor and $`R`$ the Ricci scalar, is given as a function of the kinetic term associated to $`h_{ij}`$ as obtained from Eq. (2.108):
$$S_E=\frac{m_P^2}{2}d^4x(g)^{1/2}\frac{1}{2}_\mu h_{ij}^\mu h_{ij}.$$
(2.110)
Notice that no more terms have been added to Eqs. (2.109) and (2.110) because no tensor-type contributions to the energy-momentum tensor exist during the inflationary period. The primordial tensor perturbations $`h_{ij}`$ are in consequence unsourced so that they propagate freely throughout space following a harmonic (on subhorizon scales) wave propagation pattern.
To clearly show how the $`h_{ij}`$ perturbations propagate, we apply to them the same kind of treatment we do to the scalar perturbations in previous sections. We begin by decomposing $`h_{ij}`$ in canonically normalised Fourier modes $`h_k^p`$:
$$h_{ij}(\eta ,𝐱)=\frac{\sqrt{2}}{m_P}\frac{d^3k}{(2\pi )^{3/2}}\mathrm{exp}(i𝐤𝐱)\underset{p}{}\epsilon _{ij}(p,𝐤)h_k^p(\eta )+h.c.,$$
(2.111)
where $`p=+,\times `$ are the two degrees of freedom (polarisation states), and the factors $`\epsilon _{ij}(p,𝐤)`$ are the polarisation tensors that satisfy
$`{\displaystyle \underset{i}{}}k_i\epsilon _{ij}(p,𝐤)`$ $`=`$ $`0,`$ (2.112)
$`{\displaystyle \underset{ij}{}}\epsilon _{ij}^{}(p,𝐤)\epsilon _{ij}(p^{},𝐤)`$ $`=`$ $`2\delta _{pp^{}},`$ (2.113)
$`{\displaystyle \underset{ijl}{}}\epsilon ^{ilk}\epsilon _{ij}^{}(+,𝐤)\epsilon _{jl}(\times ,𝐤)`$ $`=`$ $`{\displaystyle \underset{ijl}{}}\epsilon ^{ilk}\epsilon _{ij}^{}(\times ,𝐤)\epsilon _{jl}(+,𝐤)=2{\displaystyle \frac{k_k}{\left|𝐤\right|}},`$ (2.114)
$`{\displaystyle \underset{ijl}{}}\epsilon ^{ilk}\epsilon _{ij}^{}(+,𝐤)\epsilon _{jl}(+,𝐤)`$ $`=`$ $`{\displaystyle \underset{ijl}{}}\epsilon ^{ilk}\epsilon _{ij}^{}(\times ,𝐤)\epsilon _{jl}(\times ,𝐤)=0,`$ (2.115)
according to the transversality condition and the properties of the rotational transformations <sup>12</sup><sup>12</sup>12In the expressions of Eqs. (2.114) and (2.115), $`\epsilon ^{ijk}`$ is the totally antisymmetric Levi-Civita tensor.. Next, we recognize that the mode functions $`h_k^p`$, which satisfy the Klein-Gordon equation of motion derived from the Einstein action in Eq. (2.110) :
$$\ddot{h}_k^p+3H_{\mathrm{inf}}\dot{h}_k^p+\frac{k^2}{a^2}h_k^p=0,$$
(2.116)
are better handled if we rescale them as
$$h_k^p\frac{z_k^p}{a}.$$
(2.117)
Thus, the equation of motion for $`z_k^p`$ is the same as that for a massless scalar field<sup>13</sup><sup>13</sup>13That property reflects the masslessness of the graviton (the gravity messenger particle).:
$$z_k^{\prime \prime }+\left[k^2\frac{1}{\eta ^2}\left(\upsilon _h^2\frac{1}{4}\right)\right]z_k=0,$$
(2.118)
and reduces in the subhorizon limit ($`k\eta 1`$) to the Klein-Gordon equation in Minkowski spacetime . We point out that, in deriving the previous expression, we have worked in conformal time during a quasi de Sitter stage. The expansion parameter $`a(\eta )`$ is in this case given by
$$a(\eta )=\frac{1}{H_{\mathrm{inf}}(\eta )\eta (1\epsilon )},$$
(2.119)
where the conformal time $`\eta `$ takes negative values, and the parameter $`\upsilon _h`$ is
$`\upsilon _h`$ $`=`$ $`\left[{\displaystyle \frac{1}{4}}+{\displaystyle \frac{2\epsilon }{(1\epsilon )^2}}\right]^{1/2}`$ (2.120)
$``$ $`{\displaystyle \frac{3}{2}}+\epsilon .`$
The solution to Eq. (2.118) is well known from previous sections (see specifically Eq. (2.50) in Subsection 2.4.1), and reduces to the almost time-independent value
$`|h_k^p|`$ $``$ $`[2(1\epsilon )]^{\upsilon _h\frac{3}{2}}(1\epsilon ){\displaystyle \frac{\mathrm{\Gamma }(\upsilon _h)}{\mathrm{\Gamma }(3/2)}}{\displaystyle \frac{H_{\mathrm{inf}}}{\sqrt{2k^3}}}\left({\displaystyle \frac{k}{aH_{\mathrm{inf}}}}\right)^{\frac{3}{2}\upsilon _h}`$ (2.121)
$``$ $`{\displaystyle \frac{H_{}}{\sqrt{2k^3}}}\left({\displaystyle \frac{k}{aH_{\mathrm{inf}}}}\right)^\epsilon ,`$
for the magnitude of the mode function $`h_k^p`$ in the superhorizon regime.
We are interested in the statistical properties of the gravitational waves which are well described by the spectrum $`𝒫_T(k)`$ defined by the statistical average
$$\underset{ij}{}h_{𝐤_1}^{ij}h_{𝐤_2}^{ij}\frac{2\pi ^2}{k^3}\delta ^3(𝐤_1+𝐤_2)𝒫_T(k),$$
(2.122)
over an ensemble of universes. Here $`h_𝐤^{ij}`$ stands for
$$h_𝐤^{ij}\frac{\sqrt{2}}{m_P}\underset{p}{}\epsilon _{ij}(p,𝐤)h_k^p+h.c..$$
(2.123)
To calculate the statistical average during inflation, we must promote the gravitational wave amplitude to an operator $`\widehat{h}_{ij}`$ by introducing the creation and annihilation operators $`\widehat{a}_𝐤^p`$ and $`\widehat{a}_𝐤^p`$ that depend on the polarisation $`p`$ and wave vector $`𝐤`$, and satisfy the commutation relation
$$[\widehat{a}_𝐤^p,\widehat{a}_𝐤^{}^p^{}]=\delta ^3(𝐤𝐤^{})\delta _{pp^{}}.$$
(2.124)
The gravitational amplitude operator $`\widehat{h}_{ij}(\eta ,𝐱)`$ that generalizes Eq. (2.111) is then given by
$$\widehat{h}_{ij}(\eta ,𝐱)=\frac{d^3k}{(2\pi )^{3/2}}\mathrm{exp}(i𝐤𝐱)\widehat{h}_𝐤^{ij}(\eta ),$$
(2.125)
with
$$\widehat{h}_𝐤^{ij}(\eta )\frac{\sqrt{2}}{m_P}\underset{p}{}\left[\epsilon _{ij}(p,𝐤)h_k^p(\eta )\widehat{a}_𝐤^p+\epsilon _{ij}^{}(p,𝐤)h_k^p(\eta )\widehat{a}_𝐤^p\right].$$
(2.126)
Being all the universes in the ensemble in the vacuum state during inflation, the statistical average $`h_{𝐤_1}^{ij}h_{𝐤_2}^{ij}`$ is easily identified with the expectation value $`0|\widehat{h}_{𝐤_1}^{ij}\widehat{h}_{𝐤_2}^{ij}|0`$. The spectrum of gravitational perturbations $`𝒫_T(k)`$, defined by Eq. (2.122), is then
$`𝒫_T(k)`$ $`=`$ $`{\displaystyle \frac{k^3}{\pi ^2m_P^2}}{\displaystyle \underset{ij}{}}{\displaystyle \underset{pp^{}}{}}\epsilon _{ij}(p,𝐤)\epsilon _{ij}^{}(p^{},𝐤)h_k^p(\eta )h_k^p^{}(\eta )`$ (2.127)
$`=`$ $`{\displaystyle \frac{4k^3}{\pi ^2m_P^2}}|h_k^p|^2,`$
where one of the properties of the polarisation tensors \[c.f. Eq. (2.113)\] has been used. Now we can make use of the result in Eq. (2.121) to finally arrive to a definite expression for $`𝒫_T(k)`$ on superhorizon scales in terms of $`H_{}`$ and $`\epsilon `$ :
$$𝒫_T(k)A_T^2\left(\frac{k}{aH_{\mathrm{inf}}}\right)^{n_T}\left[\frac{\sqrt{2}H_{}}{\pi m_P}\right]^2\left(\frac{k}{aH_{\mathrm{inf}}}\right)^{2\epsilon }.$$
(2.128)
This nice result shows that the inflationary energy scale, given by $`H_{}`$, can be known from a direct measurement of the amplitude $`A_T`$. Unfortunately, at the moment all the efforts to detect gravity waves have been fruitless, leaving only the upper bound $`H_{}\text{ }\stackrel{<}{}\text{ }10^{14}`$ GeV . In addition, technological restrictions impose the lower bound $`H_{}\text{ }\stackrel{>}{}\text{ }10^{12}`$ GeV if gravity waves may some day be detected . A positive detection would kill then the curvaton scenario because, as we had discussed in Subsection 2.5.3, the inflationary energy scale in this scenario is required to satisfy $`H_{}\text{ }\stackrel{<}{}\text{ }10^{12}`$ GeV to make the inflaton field $`\phi `$ not to generate enough curvature perturbation. Only in non-slow roll inflationary models (specifically if $`\eta _\phi m_\phi ^2/3H_{}^2>1`$), the energy scale during inflation could be high enough to allow the detection of gravity waves consistent with a negligible contribution of $`\phi `$ to $`\zeta `$.
We end up this section by reporting the existence of a consistency relation between the curvature perturbation spectrum $`𝒫_\zeta (k)`$ and the gravitational waves one $`𝒫_T(k)`$ in the inflaton scenario . The ratio between the amplitudes of both spectra \[c.f. Eqs. (2.105) and (2.128)\] is given by the slow-roll parameter $`\epsilon `$:
$$r_{T\zeta }\frac{A_T^2}{A_\zeta ^2}=16\epsilon ,$$
(2.129)
which in turn gives information about the spectral index of $`𝒫_T(k)`$:
$$n_T=2\epsilon .$$
(2.130)
The ratio $`r_{T\zeta }`$ is then consistently related to the spectral index $`n_T`$, this relation being given by the expression
$$r_{T\zeta }=8n_T.$$
(2.131)
No consistency relation of this type is encountered in the curvaton scenario or in other scenarios for the origin of the large-scale structure in the Universe, although it is true that $`r_{T\zeta }`$ is always smaller in the multi-component inflationary case , so that its future confirmation would mean good news for the inflaton scenario. Nevertheless, if the consistency relation turns out to be experimentally wrong, that does not mean necessarily that the inflationary paradigm is wrong, just that the single-field variant is not nature’s choosing. Anyway, the non-gaussianity signatures associated to each model would add up to the consistency relation in Eq. (2.131), to act as powerful discriminators for models that give origin to the primordial energy density perturbations (see Chapters 5 and 6).
### 2.7 Conclusions
The curvature perturbation $`\zeta `$ is a well defined quantity, gauge-invariant and conserved on large scales (if the adiabatic condition is satisfied), that allows us to quantify the primordial energy density inhomogeneities produced during inflation. The statistical properties of $`\zeta `$ are given by the spectrum $`𝒫_\zeta (k)`$ whose amplitude and spectral index strongly depend on the specific mechanism of production of density inhomogeneities. This makes $`𝒫_\zeta (k)`$ act as a discriminator for the different production mechanisms, at least as the spectral index $`n_\zeta `$ and the possible relation of $`A_\zeta `$ with the gravitational waves spectrum $`𝒫_T(k)`$ are concerned . The inflationary energy scale, given by $`H_{}`$, is well determined by the amplitude of $`𝒫_T(k)`$ so that the current upper bound on $`A_T`$ leads to $`H_{}\text{ }\stackrel{<}{}\text{ }10^{14}`$ GeV . The energy scale may be well below $`10^{14}`$ GeV, but only at the expense of a high level of fine tuning to make the slow-roll parameter $`\epsilon `$ be extremely below 1. This reflects how constrained is the inflaton potential in the inflaton scenario, in order to produce enough curvature perturbation while driving inflation. As a result the particle physics motivated inflationary models are quite unrealistic if we insist that the inflaton field $`\phi `$ has to produce the energy density inhomogeneities . It is here where the curvaton scenario comes to the rescue: by requiring $`\phi `$ just to drive inflation, the weakly coupled curvaton field $`\sigma `$ is in charge of giving origin to $`\zeta `$ . The inflationary energy scale is in this case easily lowered so as to possibly associate $`\phi `$ with one of the fields appearing in supersymmetric extensions of the Standard Model of particle physics. Gravitational waves are in this case so tiny to ever be detected, since the current detection technology restricts $`H_{}`$ to be above $`10^{12}`$ GeV . Which scenario for the generation of $`\zeta `$ is correct will be determined by future observations. At the moment, we will just try to do our best to successfully integrate cosmology and particle physics, being the first step the determination of the lower bound for $`H_{}`$ in the simplest curvaton model . As we will see in the next chapter, $`H_{}`$ in such a model is still quite high, being $`H_{}>10^7`$ GeV, so that a modification to the basic setup is urgently needed. Two modifications to the simplest curvaton model will be explored in Chapters 3 and 4 , showing that low scale inflation with $`H_{}1`$ TeV or lower is possible to be obtained.
## Chapter 3 Low scale inflation and the curvaton mechanism
### 3.1 Introduction
The primordial curvature perturbation $`\zeta `$ is generated presumably from the perturbation of some scalar field, which in turn is generated from the vacuum fluctuation during inflation. The scalar field responsible for the primordial curvature perturbation is traditionally supposed to be the inflaton field $`\phi `$, i.e. the field responsible for the dynamics and, in particular, the end of inflation . This ‘inflaton hypothesis’ is economical, but it is quite difficult to implement and, if many scalar fields exist, it presumably is not particularly likely. An alternative is that the curvature perturbation is generated by the weakly coupled curvaton field $`\sigma `$, which could dominate (though not necessarily) the energy density before it decays (see also Refs. ). According to this ‘curvaton hypothesis’, the contribution of the inflaton to the curvature perturbation is negligible. This is especially true if the energy scale of inflation is much lower than the scale of grand unification, the latter scale being the typical requirement of the traditional inflaton hypothesis<sup>1</sup><sup>1</sup>1Although some exceptions exist (see for example Ref. ).. In fact, one of the advantages of the curvaton scenario is the relaxation of the constraints on the inflationary energy scale, which alleviates many tuning problems in inflation model–building and allows for the construction of realistic, theoretically well-motivated inflation models .
In the simplest version of the curvaton model though, the scale of inflation is still required to be quite high corresponding to Hubble parameter $`H_{}>10^7`$ GeV . The purpose of this chapter is to systematically explore a modification of the curvaton model which can instead allow inflation at an even lower scale <sup>2</sup><sup>2</sup>2Low scale inflation has also been studied in the context of the ‘inhomogeneous reheating scenario’ (see also Refs. ), where the inhomogeneities in the inflaton decay rate during inflation give origin to $`\zeta `$.. To be specific, we aim for $`H_{}10^3`$ GeV, which holds if the inflationary potential is generated by some mechanism of gravity-mediated supersymmetry breaking which holds also in the vacuum.
We begin by presenting some known bounds in a unified notation. Then we consider the possibility that the curvaton mass increases suddenly at some moment after the end of inflation but before the onset of the curvaton oscillations .
### 3.2 The bounds on the scale of inflation
In this section we present four bounds on the scale of inflation, in terms of three parameters which encode possible modifications of the simplest curvaton scenario. These bounds have been presented at least implicitly in earlier works but not in the unified notation that we employ. The advantage of this notation is that it will allow us to compare the bounds in various situations, establishing with ease which is the most crucial. The three parameters are
* The ratio $`ϵ\sigma _{}/\sigma _{\mathrm{osc}}`$, where $`\sigma _{}`$ is the global value of the curvaton field at horizon exit and $`\sigma _{\mathrm{osc}}`$ is its global value when it starts to oscillate.
* The ratio $`fH_{\mathrm{osc}}/\stackrel{~}{m}_\sigma `$, where $`H_{\mathrm{osc}}`$ is the Hubble parameter at the start of the oscillations and $`\stackrel{~}{m}_\sigma `$ is the effective curvaton mass after the onset of the oscillations.
* The ratio $`\delta \sqrt{H_{\mathrm{osc}}/H_{}}`$ where $`H_{}`$ is the Hubble parameter a few Hubble times after horizon exit.
#### 3.2.1 Curvaton physics considerations
The observed value of the nearly scale invariant spectrum of curvature perturbations, parameterised by the amplitude $`A_\zeta `$, is $`|A_\zeta |5\times 10^5`$ . In the curvaton scenario $`\zeta `$ is given by \[c.f. Eq. (2.62)\]
$$\zeta \mathrm{\Omega }_{\mathrm{dec}}\zeta _\sigma ,$$
(3.1)
where $`\mathrm{\Omega }_{\mathrm{dec}}1`$ is the density fraction of the global curvaton energy density $`\rho _{\sigma _0}`$ over the global total energy density of the Universe $`\rho _{\mathrm{total}_0}`$ at the time of the decay of the curvaton:
$$\mathrm{\Omega }_{\mathrm{dec}}\left(\frac{\rho _{\sigma _0}}{\rho _{\mathrm{total}_0}}\right)_{\mathrm{dec}},$$
(3.2)
and $`\zeta _\sigma `$ is the curvature perturbation of the curvaton field $`\sigma `$, which is \[c.f. Eq. (2.55)\]
$$\zeta _\sigma \left(\frac{\delta \sigma }{\sigma }\right)_{\mathrm{dec}}\left(\frac{\delta \sigma }{\sigma }\right)_{\mathrm{osc}},$$
(3.3)
where ‘osc’ denotes the time when the curvaton oscillations begin and ‘dec’ denotes the time of curvaton decay.
In all the cases which we consider,
$$\left(\frac{\delta \sigma }{\sigma }\right)_{}\left(\frac{\delta \sigma }{\sigma }\right)_{\mathrm{osc}},$$
(3.4)
where ‘\*’ denotes the epoch when the cosmological scales exit the horizon during inflation. The above typically holds true because the curvaton (being a light field) is frozen during and after inflation until the onset of its oscillations. However, this does not mean that $`\sigma _{}\sigma _{\mathrm{osc}}`$ necessarily. Indeed, in the case of a pseudo Nambu-Goldstone boson (PNGB) curvaton with a varying order parameter $`v`$, the curvaton field is associated with the angular displacement $`\theta `$ from the minimum of its potential as
$$\sigma \sqrt{2}v\theta .$$
(3.5)
Therefore, even though after the end of inflation, $`\theta `$ remains approximately frozen (the angular motion is over damped), we may have $`ϵ1`$, where
$$ϵ\frac{\sigma _{}}{\sigma _{\mathrm{osc}}},$$
(3.6)
because \[cf. Eq. (3.5)\] $`v_{}=ϵv_{\mathrm{osc}}v_{\mathrm{osc}}`$. However, in this case too, for the curvaton fractional perturbation we find
$$\left(\frac{\delta \sigma }{\sigma }\right)_{}=\left(\frac{\delta \theta }{\theta }\right)_{}\left(\frac{\delta \sigma }{\sigma }\right)_{\mathrm{osc}},$$
(3.7)
which agrees nicely with Eq. (3.4).
Now, for the perturbation of the curvaton we have the following value for the amplitude $`A_{\delta \sigma _{}}`$ of the spectrum of perturbations \[c.f. Eq. (2.51)\]
$$A_{\delta \sigma _{}}\frac{H_{}}{2\pi }.$$
(3.8)
Combining Eqs. (3.6) and (3.8) we find
$$A_{\delta \sigma _{\mathrm{osc}}}\frac{H_{}}{2\pi ϵ},$$
(3.9)
which means that, if the order parameter of a PNGB curvaton grows, the curvaton perturbation is amplified by a factor $`ϵ^1`$ .
From Eqs. (3.1) and (3.3) we have
$$\sigma _{\mathrm{osc}}\mathrm{\Omega }_{\mathrm{dec}}\frac{\delta \sigma _{\mathrm{osc}}}{\zeta }=\mathrm{\Omega }_{\mathrm{dec}}\frac{A_{\delta \sigma _{\mathrm{osc}}}}{A_\zeta }.$$
(3.10)
Using Eq. (3.9), we can recast the above as
$$\sigma _{\mathrm{osc}}\frac{H_{}\mathrm{\Omega }_{\mathrm{dec}}}{\pi ϵA_\zeta }.$$
(3.11)
#### 3.2.2 The main bound on the scale of inflation
For the density fraction at the onset of the curvaton oscillations we have:
$$\left(\frac{\rho _{\sigma _0}}{\rho _{\mathrm{total}_0}}\right)_{\mathrm{osc}}f^2\left(\frac{\sigma _{\mathrm{osc}}}{m_P}\right)^2,$$
(3.12)
where
$$f\frac{H_{\mathrm{osc}}}{\stackrel{~}{m}_\sigma },$$
(3.13)
and we used that $`(\rho _{\sigma _0})_{\mathrm{osc}}\frac{1}{2}\stackrel{~}{m}_\sigma ^2\sigma _{\mathrm{osc}}^2`$ and $`(\rho _{\mathrm{total}_0})_{\mathrm{osc}}=3H_{\mathrm{osc}}^2m_P^2`$. Here, $`\stackrel{~}{m}_\sigma `$ denotes the effective mass of the curvaton after the onset of its oscillations. In the basic setup of the curvaton hypothesis this effective mass is the bare mass $`m_\sigma `$. If this is the case then $`\stackrel{~}{m}_\sigma =m_\sigma H_{\mathrm{osc}}`$ (i.e. $`f1`$). However, in the heavy curvaton scenario, the mass of the curvaton is supposed to be suddenly incremented at some time after the end of the inflationary epoch due to a coupling of the form $`\lambda \chi ^2\sigma ^2`$ with a field $`\chi `$ which acquires a large vacuum expectation value (VEV) at some time after the end of inflation . In this case $`\stackrel{~}{m}_\sigma ^2=m_\sigma ^2+\lambda \chi ^2\lambda \chi ^2H_{\mathrm{osc}}^2`$ (i.e. $`f1`$).
Now, we need to consider separately the cases when the curvaton decays before it dominates the Universe ($`\mathrm{\Omega }_{\mathrm{dec}}1`$) or after it does so ($`\mathrm{\Omega }_{\mathrm{dec}}1`$). Note, that the WMAP constraints on non-gaussianity in the CMB impose a lower bound on $`\mathrm{\Omega }_{\mathrm{dec}}`$, which allows the range \[c.f. Eqs. (2.63) and (2.64)\]
$$0.01\text{ }\stackrel{<}{}\text{ }\mathrm{\Omega }_{\mathrm{dec}}1.$$
(3.14)
Because of the above bound we might require that the density ratio $`\rho _\sigma /\rho _{\mathrm{total}}`$ grows substantially after the end of inflation. Typically, in the curvaton scenario this does indeed take place after the curvaton begins oscillating, but only if the curvaton oscillates in a quadratic potential during the radiation era. As it was shown in Ref. , if the curvaton oscillates in a quartic or even higher order potential, its density ratio does not increase with time (it may well decrease instead) and satisfying the bound in Eq. (3.14) might be very hard. Due to this fact, in the following, we assume that the period of oscillations occurs in the radiation era with a quadratic potential. Hence, we consider that $`H_{\mathrm{osc}}\mathrm{\Gamma }_{\mathrm{inf}}`$, being $`\mathrm{\Gamma }_{\mathrm{inf}}`$ the inflaton decay rate.
Suppose, at first, that the curvaton decays before dominating the density of the Universe so that $`\mathrm{\Omega }_{\mathrm{dec}}1`$. Assuming that the curvaton oscillates in a quadratic potential, during the radiation epoch, its density fraction grows as $`\rho _{\sigma _0}/\rho _{\mathrm{total}_0}H^{1/2}`$. Therefore, at curvaton decay we have
$$\mathrm{\Omega }_{\mathrm{dec}}\frac{\stackrel{~}{m}_\sigma ^2\sigma _{\mathrm{osc}}^2}{T_{\mathrm{dec}}H_{\mathrm{osc}}^{3/2}m_P^{3/2}},$$
(3.15)
where we used Eq. (3.12) and also that $`(\rho _{\mathrm{total}_0})_{\mathrm{dec}}T_{\mathrm{dec}}^4`$, with $`T_{\mathrm{dec}}`$ being the radiation temperature just after the curvaton decay. Using Eq. (3.11) the above can be recast as
$$H_{}\pi ϵA_\zeta f\frac{m_P}{\sqrt{\mathrm{\Omega }_{\mathrm{dec}}}}\left(\frac{H_{\mathrm{dec}}}{H_{\mathrm{osc}}}\right)^{1/4},$$
(3.16)
where we used that $`T_{\mathrm{dec}}^2H_{\mathrm{dec}}m_P`$.
Now, suppose that the curvaton decays after it dominates the Universe so that $`\mathrm{\Omega }_{\mathrm{dec}}1`$. Since $`(\rho _\sigma /\rho _{\mathrm{total}})_{\mathrm{dom}}1`$ by definition, using again that, during the radiation epoch, $`\rho _{\sigma _0}/\rho _{\mathrm{total}_0}H^{1/2}`$ and in view of Eq. (3.12), we obtain
$$H_{\mathrm{dom}}H_{\mathrm{osc}}f^4\left(\frac{\sigma _{\mathrm{osc}}}{m_P}\right)^4,$$
(3.17)
where ‘dom’ denotes the time of curvaton domination<sup>3</sup><sup>3</sup>3Here we define $`H_{\mathrm{dom}}`$ by $`H_{\mathrm{dom}}=H_{\mathrm{eq}}`$, where $`H_{\mathrm{eq}}`$ is the Hubble parameter at the time when the global curvaton energy density $`\rho _{\sigma _0}`$ makes equal to the global radiation energy density $`\rho _{r_0}`$.. Employing again Eq. (3.11), the above can be written as
$$H_{}\pi ϵA_\zeta fm_P\left(\frac{H_{\mathrm{dom}}}{H_{\mathrm{osc}}}\right)^{1/4}.$$
(3.18)
Combining Eqs. (3.16) and (3.18) we find that, in all cases
$$H_{}\pi ϵA_\zeta f\frac{m_P}{\sqrt{\mathrm{\Omega }_{\mathrm{dec}}}}\left(\frac{\mathrm{max}\{H_{\mathrm{dom}},H_{\mathrm{dec}}\}}{H_{\mathrm{osc}}}\right)^{1/4}.$$
(3.19)
This can be rewritten as
$$H_{}\mathrm{\Omega }_{\mathrm{dec}}^{2/5}\left(\frac{H_{}}{H_{\mathrm{osc}}}\right)^{1/5}\left(\frac{\mathrm{max}\{H_{\mathrm{dom}},H_{\mathrm{dec}}\}}{H_{\mathrm{BBN}}}\right)^{1/5}(\pi ϵA_\zeta f)^{4/5}(T_{\mathrm{BBN}}^2m_P^3)^{1/5},$$
(3.20)
or equivalently (using $`V_{}^{1/4}\sqrt{H_{}m_P}`$)
$$V_{}^{1/4}\mathrm{\Omega }_{\mathrm{dec}}^{1/5}\left(\frac{H_{}}{H_{\mathrm{osc}}}\right)^{1/10}\left(\frac{\mathrm{max}\{H_{\mathrm{dom}},H_{\mathrm{dec}}\}}{H_{\mathrm{BBN}}}\right)^{1/10}(\pi ϵA_\zeta f)^{2/5}(T_{\mathrm{BBN}}m_P^4)^{1/5},$$
(3.21)
where ‘BBN’ denotes the epoch of Big-Bang Nucleosynthesis (BBN) ($`T_{\mathrm{BBN}}1`$ MeV). Now, according to Eq. (3.14) we have $`\mathrm{\Omega }_{\mathrm{dec}}1`$. Also, we require that the curvaton decays before BBN, i.e. $`H_{\mathrm{dec}}>H_{\mathrm{BBN}}`$. Hence, the above provides the following bounds
$`H_{}>(\pi ϵA_\zeta f)^{4/5}(T_{\mathrm{BBN}}^2m_P^3)^{1/5}(ϵf)^{4/5}\times 10^7\mathrm{GeV}`$, $`V_{}^{1/4}>(\pi ϵA_\zeta f)^{2/5}(T_{\mathrm{BBN}}m_P^4)^{1/5}(ϵf)^{2/5}\times 10^{12}\mathrm{GeV}`$. (3.22)
In the standard setup of the curvaton scenario $`ϵ=f=1`$ and the above bounds do not allow inflation at low energy scales to take place . However, we see that if either $`ϵ`$ or $`f`$ are much smaller than unity the lower bound on the inflationary scale can be substantially relaxed and low scale inflation can be accommodated<sup>4</sup><sup>4</sup>4The relevance of a low $`ϵ`$ makes evident in Ref. where the scenario of the curvaton as a PNGB is studied. The specific explored model refers to a PNGB whose order parameter $`v`$ is increased after the cosmological scales exit the horizon during inflation, but before the onset of the curvaton oscillations. That makes $`ϵ`$ very small.. Still, though, there are more bounds to be considered.
#### 3.2.3 Other bounds related to curvaton decay
Firstly, let us consider the bound coming from the fact that the decay rate of the curvaton field cannot be arbitrarily small. Indeed, in view of the fact that the curvaton interactions are at least of gravitational strength, we find the following decay rate for the curvaton
$$\mathrm{\Gamma }_\sigma \gamma _\sigma \frac{\stackrel{~}{m}_\sigma ^3}{m_P^2}\stackrel{~}{m}_\sigma ,$$
(3.23)
where $`\gamma _\sigma \text{ }\stackrel{>}{}\text{ }1`$.
Suppose, at first, that the curvaton decays after the onset of its oscillations, as in the basic setup of the curvaton scenario. In this case, $`\mathrm{\Gamma }_\sigma H_{\mathrm{osc}}`$ and $`H_{\mathrm{dec}}=\mathrm{\Gamma }_\sigma `$. Using the fact that max$`\{H_{\mathrm{dom}},\mathrm{\Gamma }_\sigma \}\mathrm{\Gamma }_\sigma `$, Eq. (3.23) suggests
$$\frac{\mathrm{max}\{H_{\mathrm{dom}},H_{\mathrm{dec}}\}}{H_{\mathrm{osc}}}\gamma _\sigma f^1\left(\frac{\stackrel{~}{m}_\sigma }{m_P}\right)^2.$$
(3.24)
Including the above into Eq. (3.19) the latter becomes
$$H_{}\sqrt{\gamma _\sigma }(\pi ϵA_\zeta )^2\sqrt{f}\frac{m_P}{\mathrm{\Omega }_{\mathrm{dec}}}\left(\frac{H_{\mathrm{osc}}}{H_{}}\right),$$
(3.25)
which results in the bounds
$`H_{}(\pi ϵA_\zeta )^2\sqrt{f}\delta ^2m_Pϵ^2\sqrt{f}\delta ^2\times 10^{11}\mathrm{GeV}`$, $`V_{}^{1/4}\pi ϵA_\zeta f^{1/4}\delta m_Pϵf^{1/4}\delta \times 10^{14}\mathrm{GeV}`$, (3.26)
where we have defined
$$\delta \sqrt{\frac{H_{\mathrm{osc}}}{H_{}}},$$
(3.27)
which must be really small in order to reduce the bounds in Eq. (3.22) to satisfactory levels. In the case of a PNGB curvaton we see that the bounds in Eq. (3.26) are drastically reduced with $`ϵ`$, compared with the bounds in Eq. (3.22).
Now, provided we demand that the curvaton field does not itself result in a period of inflation, we see that the curvaton cannot dominate the Universe before the onset of its oscillations. This results into the constraint
$$\left(\frac{\rho _{\sigma _0}}{\rho _{\mathrm{total}_0}}\right)_{\mathrm{osc}}1\stackrel{~}{m}_\sigma \pi ϵA_\zeta \delta ^2\frac{m_P}{\mathrm{\Omega }_{\mathrm{dec}}}f\frac{\mathrm{\Omega }_{\mathrm{dec}}H_{}}{(\pi ϵA_\zeta )m_P},$$
(3.28)
where we used Eqs. (3.11), (3.12), (3.13) and (3.27). Inserting the above into Eq. (3.25) we obtain
$$H_{}\gamma _\sigma (\pi ϵA_\zeta )^3\delta ^4\frac{m_P}{\mathrm{\Omega }_{\mathrm{dec}}},$$
(3.29)
which results in the bounds
$`H_{}(\pi ϵA_\zeta )^3\delta ^4m_Pϵ^3\delta ^4\times 10^7\mathrm{GeV}`$, $`V_{}^{1/4}(\pi ϵA_\zeta )^{3/2}\delta ^2m_Pϵ^{3/2}\delta ^2\times 10^{12}\mathrm{GeV}`$. (3.30)
A similar bound is reached with the use of the upper bound on $`\stackrel{~}{m}_\sigma `$
$$\stackrel{~}{m}_\sigma \gamma _\sigma ^{1/3}(H_{\mathrm{osc}}m_P^2)^{1/3},$$
(3.31)
which comes from $`\mathrm{\Gamma }_\sigma H_{\mathrm{osc}}`$ and Eq. (3.23), instead of the bound in Eq. (3.28). Inserting the above into Eq. (3.25) one finds \[cf. Eq. (3.29)\]
$$H_{}\gamma _\sigma (\pi ϵA_\zeta )^3\delta ^4\frac{m_P}{\mathrm{\Omega }_{\mathrm{dec}}^{3/2}},$$
(3.32)
which, again, results in the bound in Eq. (3.30), as it was suggested in Ref. .
In the heavy curvaton scenario we have $`ϵ=1`$ and also $`H_{\mathrm{osc}}\mathrm{min}\{H_{\mathrm{pt}},\stackrel{~}{m}_\sigma \}`$, where $`H_{\mathrm{pt}}`$ corresponds to the phase transition which increases the effective mass of the curvaton. Then, if $`\delta 1`$, the bounds in Eq. (3.30) are not possible to be relaxed below the standard case discussed in Ref. despite the fact that we may have $`f1`$ in Eqs. (3.22) and (3.26). Therefore, in the heavy curvaton scenario we require $`\delta 1`$, i.e. the onset of the curvaton oscillations has to occur much later than the end of inflation so that $`H_{}H_{\mathrm{osc}}\mathrm{\Gamma }_\sigma `$ . In this case, as can be seen in Eq. (3.30), it is easy to lower the bound on the inflationary scale even for a not-so-small value of $`\delta `$. This is a very nice feature of this scenario. Note also, that in the case of a PNGB curvaton $`H_{\mathrm{osc}}m_\sigma H_{}`$ and $`\delta `$ is very small necessarily. Because, in this case, $`ϵ1`$, it is straightforward to see that the bounds in Eq. (3.30) are much weaker than the bounds in Eq. (3.22).
As it was pointed out in Ref. , the sudden increment in the curvaton mass might lead to a growth in the curvaton decay rate enough for $`\mathrm{\Gamma }_\sigma >H_{\mathrm{pt}}`$. This would force the curvaton to decay immediately and we can write $`H_{\mathrm{osc}}H_{\mathrm{pt}}H_{\mathrm{dec}}`$. Obviously, in this case we cannot have $`H_{\mathrm{dec}}<H_{\mathrm{dom}}`$ and there is no period when $`\rho _{\sigma _0}/\rho _{\mathrm{total}_0}H^{1/2}`$. This means that $`(\rho _{\sigma _0}/\rho _{\mathrm{total}_0})_{\mathrm{osc}}\mathrm{\Omega }_{\mathrm{dec}}`$. Using Eqs. (3.11) and (3.12) it is easy to find
$$H_{}\pi ϵA_\zeta f\frac{m_P}{\sqrt{\mathrm{\Omega }_{\mathrm{dec}}}},$$
(3.33)
which results in the following bounds
$`H_{}\pi ϵA_\zeta fm_Pϵf\times 10^{14}\mathrm{GeV}`$, $`V_{}^{1/4}\sqrt{\pi ϵA_\zeta f}m_P(ϵf)^{1/2}\times 10^{16}\mathrm{GeV}`$. (3.34)
It is evident that the above bounds may challenge the WMAP constraint for the curvaton scenario leading to excessive curvature perturbations from the inflaton field if $`\epsilon `$ and/or $`f`$ are not much smaller than unity.
The bounds in Eqs. (3.22), (3.26), and (3.30) provide the basis for our investigation , leaving the fourth bound in Eq. (3.34) to be considered in the next chapter . As a matter of completeness we have considered all the other possible bounds coming from the requirements that $`\mathrm{\Gamma }_\sigma <\stackrel{~}{m}_\sigma `$ and $`H_{\mathrm{dec}}H_{\mathrm{BBN}}`$. We have found that these bounds lead to consistent and/or weaker constraints than the above four.
### 3.3 The case of a heavy curvaton
In this section we are going to consider the so called ‘heavy curvaton scenario’ where an increment in the curvaton mass, at some moment after the end of inflation but before the onset of the curvaton oscillations, leads to a huge decrease of the inflationary scale through the attainment of a very small parameter $`\delta `$ \[cf. Eq. (3.30)\]. We will do so by the implementation of a second inflationary period following the idea first presented in Ref. <sup>5</sup><sup>5</sup>5Note however that any post-inflationary phase transition could serve the purpose of giving an effective mass to the curvaton field.. We identify this second inflationary period as the thermal inflation one which triggers the increment in the curvaton mass when the flaton field, that responsible for the generation of the thermal inflation era, rolls down towards the minimum of the potential.
#### 3.3.1 The thermal inflation model
Thermal inflation was introduced as a very nice mechanism to get rid of some unwanted relics that the main inflationary epoch is not able to dilute, without affecting the density perturbations generated during ordinary inflation. As its name suggests, thermal inflation relies on the finite-temperature effects on the flaton scalar potential. A flaton field $`\chi `$ could be defined as a field with mass $`m_\chi `$ and vacuum expectation value $`Mm_\chi `$ . More specifically, a flaton field is a MSSM flat direction lifted by non-renormalisable terms. SUSY breaking provides soft terms which create a large vacuum expectation value because the absence of quartic terms in the potential. The possible candidates for a flaton field within particle physics are either one of the many expected gauge singlets in string theory or the GUT Higgs (which is a scalar field charged under the GUT gauge symmetry but neutral under the Standard Model one) with $`m_\chi 10^3`$ GeV and $`M10^{16}`$ GeV <sup>6</sup><sup>6</sup>6Note, though, that in some GUT models there are additional Higgs fields with much smaller vevs .. After the period of reheating following the main inflationary epoch, the thermal background modifies the flaton potential $`V`$ trapping the flaton field at the origin and preventing it to roll-down towards $`M`$ . At this stage the total energy density $`\rho _{\mathrm{total}}`$ and pressure $`P_{\mathrm{total}}`$ are
$`\rho _{\mathrm{total}}`$ $`=`$ $`V+\rho _r,`$
$`P_{\mathrm{total}}`$ $`=`$ $`V+{\displaystyle \frac{1}{3}}\rho _r,`$ (3.35)
making the condition for thermal inflation, $`\rho _{\mathrm{total}_0}+3P_{\mathrm{total}_0}<0`$, valid when the global thermal energy density $`\rho _{r_0}`$ falls below the height of the potential $`V_h`$, which corresponds to a temperature of roughly $`V_h^{1/4}`$. Thermal inflation ends when the finite temperature becomes ineffective at confining the field, at a temperature of order $`m_\chi `$, so the number of e-folds this inflationary period lasts is
$$N=\mathrm{ln}\left(\frac{a_{\mathrm{end}}}{a_{\mathrm{start}}}\right)=\mathrm{ln}\left(\frac{T_{\mathrm{start}}}{T_{\mathrm{end}}}\right)\mathrm{ln}\left(\frac{V_h^{1/4}}{m_\chi }\right)\frac{1}{2}\mathrm{ln}\left(\frac{M}{m_\chi }\right)10.$$
(3.36)
Here we have used the fact that, in a flaton potential of the form
$$V=V_h(m_\chi ^2gT^2)|\chi |^2+\underset{n=1}{\overset{\mathrm{}}{}}\lambda _nm_P^{2n}|\chi |^{2n+4},$$
(3.37)
where the $`n`$th term dominates:
$`\stackrel{~}{m}_\chi ^2`$ $`=`$ $`2(n+1)m_\chi ^2,`$ (3.38)
$`M^{2n+2}m_P^{2n}`$ $`=`$ $`[2(n+1)(n+2)\lambda _n]^1\stackrel{~}{m}_\chi ^2,`$ (3.39)
$`V_h`$ $`=`$ $`[2(n+2)]^1\stackrel{~}{m}_\chi ^2M^2.`$ (3.40)
Note that the $`gT^2`$ contribution to the effective mass of the flaton field stands for the effect of the thermal background, which changes the slope of the potential in the $`\chi `$ direction and traps the flaton field at the origin of the potential . It is worthwhile to mention that the potential is stabilized by non-renormalisable terms, with dimensionless couplings $`\lambda _n1`$ to make the theory valid up to the Planck scale. Notice also that the $`\lambda _4|\chi |^4`$ term is absent in the potential; otherwise, the vacuum expectation value $`M`$ would not be much bigger than $`\stackrel{~}{m}_\chi `$, spoiling the suppression of the unwanted relics.
Before embedding the thermal inflation epoch and the curvaton mechanism into a single model, we want to clarify some issues about the nature of the interactions that produce the thermal background. If the flaton is a GUT Higgs, it is coupled with those fields charged under the GUT gauge symmetry, in particular with those the inflaton field decays into. That collection of particles makes the thermal background, and its interaction with the flaton field produces the thermal correction. If the flaton field is a gauge singlet it still can be coupled, via Yukawa coupling terms, with some other fields, possibly in a hidden sector, that the inflaton field decays into. Again, a thermal correction is generated. The actual interactions and decay rate are not important as the main objective of this chapter and the next one is to obtain some particle physics model-independent information about the possibility of reconciling low scale inflation with the curvaton mechanism, in a scenario that involves a second period of inflation (thermal inflation), without going into the details of the identification of all the relevant fields (inflaton, flaton, and curvaton) in the framework of a particle physics model (GUT theories, MSSM, etc.), which would make the results highly particle physics model-dependent. The flaton could be either a gauge singlet or the GUT Higgs. In the former case the flaton can be coupled with some other fields that the inflaton field decays into, via Yukawa coupling terms, and the specific interactions would be known once we choose what of the many gauge singlets expected in string theory is the flaton. In the latter case the interactions in the GUT models are already known. The specific interactions are important of course, but there are so many possibilities that the general result would be hidden behind the characteristics associated to any definite particle physics model.
Having discussed the nature of the flaton interactions, and guided by the result in Ref. , we proceed to implement a second inflationary stage into the curvaton scenario in order to lower the main inflationary energy scale. If this second epoch of inflation is the thermal inflation one devised in Refs. we would be solving not only the issue of the ordinary inflation energy scale but also the moduli problem present in the standard cosmology .
In the curvaton model supplemented by a thermal inflation epoch two fields $`\chi `$ and $`\sigma `$, which we identify as the flaton and the curvaton fields respectively, are embedded into the radiation background left by the inflaton decay. It is assumed that the curvaton field could be either a gauge singlet , the Peccei-Quinn field , a PNGB , or a MSSM flat direction , and has just a quadratic interaction with the flaton one so that the unperturbed component is frozen at some value $`\sigma _{}`$ until the time when the flaton field is released from the origin and rolls down towards the minimum of the potential. This in turn signals the end of the thermal inflation era and the beginning of the oscillations of the curvaton field around the minimum of its quadratic potential . The flaton field, in addition to the non renormalisable terms with $`\lambda _n1`$ that stabilize the potential and make its slope in the $`\chi `$ direction be very flat, presents a quadratic interaction with the curvaton field. The complete expression for the potential is
$$V(\chi ,\sigma )=V_h(m_\chi ^2gT^2)|\chi |^2+m_\sigma ^2|\sigma |^2+\lambda |\chi |^2|\sigma |^2+\underset{n=1}{\overset{\mathrm{}}{}}\lambda _nm_P^{2n}|\chi |^{2n+4},$$
(3.41)
where $`m_\chi 10^3\mathrm{GeV}`$ due to the soft SUSY contributions in a gravity mediated SUSY breaking scheme. Under these circumstances the condition for an inflationary period, $`\rho _{\mathrm{total}_0}+3P_{\mathrm{total}_0}<0`$, is satisfied when the global thermal energy density $`\rho _{r_0}`$ falls below $`V_h`$. Of course, this period of thermal inflation ends when the effect of the thermal background becomes unimportant, at a temperature $`Tm_\chi `$, liberating the flaton field to roll down towards the minimum of the potential and letting it acquire a large vacuum expectation value $`M`$ given by:
$$M\frac{V_h^{1/2}}{m_\chi }.$$
(3.42)
The evolution of the energy densities associated to the different fluids in this case are sketched in Fig. 3.1.
Let’s assume that the usual inflation and its corresponding reheating have already happened, so that the flaton and the curvaton fields are embedded into a radiation bath. Therefore, even when the minimum of the potential is located at $`\chi _0=M_\chi (\sigma _{})0`$ and $`\sigma _0=0`$, $`\chi `$ is trapped at the origin because of the finite-temperature effects and $`|\sigma _0|=\sigma _{}0`$ because $`m_\sigma <H<H_{}`$. Thus, the value of the scalar potential at this stage is:
$$V(\chi _0=0,\sigma _0=\sigma _{})=V_h+m_\sigma ^2\sigma _{}^2,$$
(3.43)
with
$`\stackrel{~}{m}_\chi ^2`$ $`=`$ $`2(n+1)(m_\chi ^2\lambda |\sigma _0|^2),`$ (3.44)
$`M_\chi ^{2n+2}m_P^{2n}`$ $`=`$ $`[(n+2)\lambda _n]^1(m_\chi ^2\lambda |\sigma _0|^2),`$ (3.45)
$`V_h`$ $`=`$ $`[2(n+2)]^1(\stackrel{~}{m}_\chi ^2M_\chi ^2)_{\sigma _0=0}.`$ (3.46)
When the thermal energy density falls below $`V_h+m_\sigma ^2\sigma _{}^2`$ thermal inflation begins. This period lasts until the temperature is of the order the effective mass of the flaton field which is $`\stackrel{~}{m}_\chi =(m_\chi ^2\lambda \sigma _{}^2)^{1/2}`$. Note that $`\lambda \sigma _{}^2<m_\chi ^2`$ because otherwise there is no thermal inflation. Then, we obtain a first constraint on the value of the parameter $`\lambda `$:
$$\lambda <\frac{m_\chi ^2}{\sigma _{}^2}\frac{10^2\mathrm{GeV}^2}{H_{}^2\mathrm{\Omega }_{\mathrm{dec}}^2},$$
(3.47)
where we have used the Eq. (3.11) and focused on $`m_\chi 10^3`$ GeV which comes from the gravity-mediated SUSY breaking contributions.
When thermal inflation ends the thermal energy density is no longer dominant. The Hubble parameter at the end of thermal inflation is then associated to the energy density coming from the curvaton and the flaton fields:
$$H_{\mathrm{osc}}^2=\frac{\rho _T+V(\chi _0=0,\sigma _0=\sigma _{})}{3m_P^2}\frac{m_\chi ^2M^2}{3m_P^2},$$
(3.48)
so that
$$H_{\mathrm{osc}}10^{16}M,$$
(3.49)
and therefore the parameter $`f`$ \[cf. Eq. (3.13)\] is
$$f\frac{H_{\mathrm{osc}}}{\stackrel{~}{m}_\sigma }10^{16}\frac{M}{\stackrel{~}{m}_\sigma },$$
(3.50)
where $`MM_\chi _{\sigma _0=0}`$ is somewhere in the range $`10^3\mathrm{GeV}M\text{ }\stackrel{<}{}\text{ }10^{18}\mathrm{GeV}`$.
With this so-low value for the Hubble parameter at the end of thermal inflation, the parameter $`\delta `$ \[cf. Eq. (3.27)\] is
$$\delta 10^8\sqrt{\frac{M}{H_{}}},$$
(3.51)
so that the bounds in Eqs. (3.22) and (3.26) become<sup>7</sup><sup>7</sup>7The bound in Eq. (3.30) is consistent with low scale inflation in view of $`M\text{ }\stackrel{<}{}\text{ }10^{18}`$ GeV. Notice also that, in the heavy curvaton mechanism, $`ϵ=1`$ because there is no amplification of the curvaton perturbations.:
$`H_{}`$ $`>`$ $`10^6\mathrm{GeV}{\displaystyle \frac{M^{4/5}}{\stackrel{~}{m}_\sigma ^{4/5}\mathrm{\Omega }_{\mathrm{dec}}^{2/5}}},`$ (3.52)
$`H_{}`$ $`>`$ $`10^7\mathrm{GeV}^{1/2}{\displaystyle \frac{M^{3/4}}{\stackrel{~}{m}_\sigma ^{1/4}\mathrm{\Omega }_{\mathrm{dec}}^{1/2}}}.`$ (3.53)
The effective mass of the curvaton field after the end of thermal inflation, i.e., when $`\overline{\chi }=M_\chi `$ and $`\overline{\sigma }=0`$ are the average over the oscillations of the flaton and the curvaton fields, is
$$\stackrel{~}{m}_\sigma =(m_\sigma ^2+\lambda M^2)^{1/2}.$$
(3.54)
Note that we are focusing in the case of a final curvaton decay rate $`\mathrm{\Gamma }_\sigma `$ smaller than the Hubble parameter at the beginning of the oscillations $`H_{\mathrm{osc}}`$. This is to allow the curvaton field to decay after the flaton field so that we can keep working in the simplest curvaton scenario where the curvaton field oscillates in a radiation background (see Fig. 3.1).
Making use of the constraint in Eq. (3.47) and the expression in Eq. (3.54), and taking into account that the bare curvaton mass $`m_\sigma `$ is smaller than the Hubble parameter $`H_{\mathrm{osc}}`$ at the end of thermal inflation, we obtain an upper bound on the effective mass of the curvaton field:
$$\stackrel{~}{m}_\sigma <10^1\mathrm{GeV}\frac{M}{H_{}\mathrm{\Omega }_{\mathrm{dec}}}.$$
(3.55)
This bound, when applied to Eq. (3.52), is consistent with low scale inflation. When Eq. (3.55) is applied to Eq. (3.53), we obtain a lower bound for $`H_{}`$ which is consistent too with low-energy scale inflation since $`M\text{ }\stackrel{<}{}\text{ }10^{18}`$ GeV:
$$H_{}>10^9\mathrm{GeV}^{1/3}M^{2/3}\mathrm{\Omega }_{\mathrm{dec}}^{1/3}.$$
(3.56)
The last inequality is stronger than that of Eq. (3.52) only while the effective mass of the curvaton field is
$$\stackrel{~}{m}_\sigma >10^2\mathrm{GeV}^{10/11}M^{1/11}\mathrm{\Omega }_{\mathrm{dec}}^{2/11}.$$
(3.57)
Otherwise, we still need to consider the expression in Eq. (3.52).
#### 3.3.2 Required parameter space
Once we have checked the viability of a low-energy scale inflation we proceed to investigate the required range of values for the parameters of the Lagrangian. Remember that we are going to focus on the gravity-mediated SUSY breaking scheme where the Hubble parameter during inflation is $`H_{}m_{3/2}10^3`$ GeV. After thermal inflation has ended, the flaton and curvaton fields start to oscillate, eventually decaying into thermalised radiation (see Fig. 3.1). The decay process is distinguished by the decay rate. The field with the biggest decay rate will decay first. The flaton and curvaton decay rates are given by
$`\mathrm{\Gamma }_\chi \gamma _\chi {\displaystyle \frac{m_\chi ^3}{M^2}}`$ $`\mathrm{and}`$ $`\mathrm{\Gamma }_\sigma \gamma _\sigma {\displaystyle \frac{\stackrel{~}{m}_\sigma ^3}{m_P^2}},`$ (3.58)
with $`\gamma _\chi \text{ }\stackrel{<}{}\text{ }1`$ and $`\gamma _\sigma \text{ }\stackrel{>}{}\text{ }1`$. Since we like the curvaton mechanism not to be modified, the flaton field must decay well before the curvaton decay. This requires
$$\stackrel{~}{m}_\sigma ^3m_\chi ^3\frac{m_P^2}{M^2}\frac{10^{46}\mathrm{GeV}^5}{M^2}.$$
(3.59)
Now, using the expression in Eq. (3.52), which is relevant for $`\stackrel{~}{m}_\sigma 10^2\mathrm{GeV}^{10/11}M^{1/11}\mathrm{\Omega }_{\mathrm{dec}}^{2/11}`$ \[cf. Eq. (3.57)\], we require
$$\stackrel{~}{m}_\sigma >10^{11}M\mathrm{\Omega }_{\mathrm{dec}}^{1/2},$$
(3.60)
in order to obtain low-energy scale inflation. Note that, combining the above with Eq. (3.50), we find
$$f<10^5\sqrt{\mathrm{\Omega }_{\mathrm{dec}}}1,$$
(3.61)
as required by the heavy curvaton scenario. Similarly to the above, using the expression in Eq. (3.53), which is relevant for $`\stackrel{~}{m}_\sigma >10^2\mathrm{GeV}^{10/11}M^{1/11}\mathrm{\Omega }_{\mathrm{dec}}^{2/11}`$ \[cf. Eq. (3.57)\], we require
$$\stackrel{~}{m}_\sigma >10^{40}\mathrm{GeV}^2M^3\mathrm{\Omega }_{\mathrm{dec}}^2.$$
(3.62)
Thus, for values of $`\stackrel{~}{m}_\sigma `$ less than $`10^2\mathrm{GeV}^{10/11}M^{1/11}\mathrm{\Omega }_{\mathrm{dec}}^{2/11}`$ the required range of values for $`\stackrel{~}{m}_\sigma `$ is<sup>8</sup><sup>8</sup>8The bound in Eq. (3.59) is weaker than $`\stackrel{~}{m}_\sigma <10^2\mathrm{GeV}^{10/11}M^{1/11}`$ within the allowed range for $`M`$ (see Fig. 3.2).:
$$10^{11}M<\stackrel{~}{m}_\sigma <10^2\mathrm{GeV}^{10/11}M^{1/11},$$
(3.63)
where the lower bound comes from Eq. (3.60). The vacuum expectation value $`M`$ is in the range
$$10^{12}\mathrm{GeV}\text{ }\stackrel{<}{}\text{ }M\text{ }\stackrel{<}{}\text{ }10^{14}\mathrm{GeV},$$
(3.64)
where the lower bound comes from the solution to the moduli problem as we will see later, and the upper bound comes from Eq. (3.63). On the other hand, for values of $`\stackrel{~}{m}_\sigma `$ bigger than $`10^2\mathrm{GeV}^{10/11}M^{1/11}\mathrm{\Omega }_{\mathrm{dec}}^{2/11}`$ the required range of values for $`\stackrel{~}{m}_\sigma `$ is:
$$\mathrm{max}\{10^2\mathrm{GeV}^{10/11}M^{1/11},10^{40}\mathrm{GeV}^2M^3\}<\stackrel{~}{m}_\sigma <10^{15}\mathrm{GeV}^{5/3}/M^{2/3},$$
(3.65)
where we have used Eqs. (3.59) and (3.62), and $`M`$ can be, a priori, up to $`m_P`$. We have considered all the other possible constraints on $`\stackrel{~}{m}_\sigma `$ and found they are irrelevant compared with those in Eq. (3.63) and Eq. (3.65).
Fig. 3.2 shows the required parameter space $`\lambda `$ vs $`M`$ (grey region) as a logarithmic plot. We have made use of the definition of the curvaton effective mass $`\stackrel{~}{m}_\sigma `$ in terms of the coupling constant $`\lambda `$ and the vacuum expectation value $`M`$ \[c.f. Eq. (3.54)\]:
$$\stackrel{~}{m}_\sigma ^2\lambda M^2,$$
(3.66)
and the required parameter space $`\stackrel{~}{m}_\sigma `$ vs $`M`$ studied before. Note that for values of $`M`$ higher than $`10^{15}`$ GeV it is impossible to satisfy Eq. (3.65), so our final range for $`M`$ is
$$10^{12}\mathrm{GeV}\text{ }\stackrel{<}{}\text{ }M\text{ }\stackrel{<}{}\text{ }10^{15}\mathrm{GeV}.$$
(3.67)
The required values for $`\lambda `$, according to Fig. 3.2:
$$10^{22}\text{ }\stackrel{<}{}\text{ }\lambda \text{ }\stackrel{<}{}\text{ }10^{10},$$
(3.68)
are in agreement with the upper bound in the Eq. (3.47):
$$\lambda <\frac{10^2\mathrm{GeV}^2}{H_{}^2}10^4,$$
(3.69)
and with the lower bound
$$\lambda >\frac{H_{\mathrm{osc}}^2}{M^2}\frac{m_\chi ^2}{3m_P^2}10^{31},$$
(3.70)
which follows from $`\stackrel{~}{m}_\sigma >H_{\mathrm{osc}}`$.
In view of the allowed range of values for $`M`$ \[c.f. Eq. (3.67)\], we conclude that our flaton field cannot be the GUT Higgs field investigated in Ref. . We must remember however that in some other GUT models there are additional Higgs fields with much smaller vevs so they are still good flaton candidates. The flaton field as a gauge singlet in string theory remains as a viable option.
Once we have found the required parameter space for $`\lambda `$ we must do the same for the other relevant parameter of the Lagrangian: the bare mass of the curvaton $`m_\sigma `$. The only bound on $`m_\sigma `$ is
$$m_\sigma <H_{\mathrm{osc}}10^{16}M,$$
(3.71)
which is related to the fact that the oscillations of the curvaton around the minimum begin due to the sudden increment in the curvaton mass at the end of thermal inflation. That means, in view of Eq. (3.67), that
$$m_\sigma \text{ }\stackrel{<}{}\text{ }10^1\mathrm{GeV}.$$
(3.72)
Such a small value for $`m_\sigma `$, taking into account the soft supersymmetric contributions of order the gravitino mass for any scalar field which is not protected by a global symmetry, leads us to point a PNGB as a viable curvaton candidate .
Finally, we still need to understand the lower bound $`M\text{ }\stackrel{>}{}\text{ }10^{12}`$ GeV. To do that, we must study the solution to the moduli problem<sup>9</sup><sup>9</sup>9In the following subsection we correct one mistake in Ref. which led to a reduced parameter space for $`m_\sigma `$. Conclusions are different of course, but they are now more positive than before..
#### 3.3.3 Solution to the moduli problem
Among the unwanted relics that the inflationary epoch is not able to dilute are the moduli . Moduli fields are flaton fields with a vacuum expectation value of order the Planck mass. The decays of the flaton and the curvaton fields increment the entropy, so that the big-bang moduli abundance, defined as that produced before thermal inflation and given by
$$\frac{n_\mathrm{\Phi }}{s}\frac{\mathrm{\Phi }^2}{10m_P^{3/2}m_\mathrm{\Phi }^{1/2}},$$
(3.73)
where $`\mathrm{\Phi }`$ is the vacuum expectation value of the moduli fields, gets suppressed by three factors. One is
$$\mathrm{\Delta }_{PR}\frac{g_{}(T_{PR})}{g_{}(T_C)}\frac{T_{PR}^3}{T_C^3},$$
(3.74)
due to the parametric resonance process following the end of the thermal inflation era, where the $`g_{}`$ are the total internal particle degrees of freedom, $`T_{PR}`$ is the temperature just after the period of preheating, and $`T_C`$ is the temperature at the end of thermal inflation; another is
$$\mathrm{\Delta }_\chi \frac{4\beta V_h/3T_\chi }{(2\pi ^2/45)g_{}(T_{PR})T_{PR}^3},$$
(3.75)
due to the flaton decay, where $`T_\chi `$ is the temperature just after the decay<sup>10</sup><sup>10</sup>10This is assuming that the flaton has come to dominate the energy density just before decaying (see Fig. 3.1)., and $`\beta `$ is the fraction of the total energy density left in the flatons by the parametric resonance process and the increment in the energy density of the curvaton ($`\beta \text{ }\stackrel{<}{}\text{ }1`$); the other is
$$\mathrm{\Delta }_\sigma \frac{4\stackrel{~}{m}_\sigma ^2\sigma _{\mathrm{osc}}^2/3\mathrm{\Omega }_{\mathrm{dec}}T_{\mathrm{dec}}}{(2\pi ^2/45)g_{}(T_\chi )T_\chi ^3},$$
(3.76)
due to the curvaton decay, where $`T_{\mathrm{dec}}`$ is the associated reheating temperature which must be bigger than $`1`$ MeV not to disturb the nucleosynthesis process<sup>11</sup><sup>11</sup>11We have assumed that $`\rho _\sigma `$ does not change appreciably from the time when $`T=T_C`$ to the time when $`T=T_\chi `$. This is a good approximation since $`\mathrm{\Gamma }_\chi \mathrm{\Gamma }_\sigma `$.. This enhancement in the entropy depends on the temperature just after the flaton decay
$$T_\chi \frac{10^{13}\mathrm{GeV}^2}{M}\gamma _\chi ^{1/2},$$
(3.77)
which is obtained by setting $`\mathrm{\Gamma }_\chi H`$ and assuming that the flaton decay products thermalise promptly. Thus, the abundance of the big-bang moduli after thermal inflation is:
$`{\displaystyle \frac{n_\mathrm{\Phi }}{s}}`$ $``$ $`{\displaystyle \frac{\mathrm{\Phi }^2}{10m_P^{3/2}m_\mathrm{\Phi }^{1/2}\mathrm{\Delta }_{PR}\mathrm{\Delta }_\chi \mathrm{\Delta }_\sigma }}{\displaystyle \frac{\mathrm{\Phi }^2T_\chi ^4T_{\mathrm{dec}}T_C^3}{10^5\beta V_hm_\mathrm{\Phi }^{1/2}\stackrel{~}{m}_\sigma ^2\mathrm{\Omega }_{\mathrm{dec}}m_P^{3/2}H_{}^2}}`$ (3.78)
$`\stackrel{>}{}`$ $`10^{48}\mathrm{GeV}^8\lambda ^1M^8\gamma _\chi ^2\left({\displaystyle \frac{\mathrm{\Phi }}{m_P}}\right)^2\left({\displaystyle \frac{T_{\mathrm{dec}}}{1\mathrm{MeV}}}\right)\left({\displaystyle \frac{T_C}{m_\mathrm{\Phi }}}\right)^3\times `$
$`\times \left({\displaystyle \frac{m_\mathrm{\Phi }}{10^3\mathrm{GeV}}}\right)^{1/2}{\displaystyle \frac{1}{\beta }}\left({\displaystyle \frac{m_\mathrm{\Phi }^2M^2}{V_h}}\right){\displaystyle \frac{1}{\mathrm{\Omega }_{\mathrm{dec}}}}\left({\displaystyle \frac{10^3\mathrm{GeV}}{H_{}}}\right)^2.`$
The lower bound
$$\lambda \text{ }\stackrel{>}{}\text{ }\frac{10^{60}\mathrm{GeV}^8}{M^8}\gamma _\chi ^2,$$
(3.79)
is obtained when taking into account the restriction $`n_\mathrm{\Phi }/s\text{ }\stackrel{<}{}\text{ }10^{12}`$ coming from nucleosynthesis . This is a weaker bound on $`\lambda `$ than those presented in Fig. 3.2.
Let’s have a look at the thermal inflation moduli abundance defined as that produced during the preheating stage following the end of the thermal inflation era
$`{\displaystyle \frac{n_{\mathrm{\Phi }_T}}{s}}`$ $``$ $`{\displaystyle \frac{\mathrm{\Phi }_T^2V_h^2/10m_\mathrm{\Phi }^3m_P^4}{(2\pi ^2/45)g_{}(T_{PR})T_{PR}^3\mathrm{\Delta }_\chi \mathrm{\Delta }_\sigma }}{\displaystyle \frac{\mathrm{\Phi }_T^2V_hT_\chi ^4T_{\mathrm{dec}}}{10^7\beta m_\mathrm{\Phi }^3\stackrel{~}{m}_\sigma ^2\mathrm{\Omega }_{\mathrm{dec}}m_P^4H_{}^2}}`$ (3.80)
$`\stackrel{>}{}`$ $`10^4\mathrm{GeV}^4\lambda ^1M^4\gamma _\chi ^2\left({\displaystyle \frac{\mathrm{\Phi }_T}{m_P}}\right)^2\left({\displaystyle \frac{T_{\mathrm{dec}}}{1\mathrm{MeV}}}\right){\displaystyle \frac{1}{\beta }}\times `$
$`\times \left({\displaystyle \frac{10^3\mathrm{GeV}}{m_\mathrm{\Phi }}}\right)\left({\displaystyle \frac{V_h}{m_\mathrm{\Phi }^2M^2}}\right){\displaystyle \frac{1}{\mathrm{\Omega }_{\mathrm{dec}}}}\left({\displaystyle \frac{10^3\mathrm{GeV}}{H_{}}}\right)^2.`$
Here $`\mathrm{\Phi }_T`$ corresponds to the vacuum expectation value of the thermal moduli fields. To suppress the thermal inflation moduli at the required level $`n_{\mathrm{\Phi }_T}/s\text{ }\stackrel{<}{}\text{ }10^{12}`$ we require
$$\lambda \text{ }\stackrel{>}{}\text{ }\frac{10^8\mathrm{GeV}^4}{M^4}\gamma _\chi ^2.$$
(3.81)
Again this is a weaker bound on $`\lambda `$ than those in Fig. 3.2.
The Eqs. (3.78) and (3.80) give us information about the necessary conditions for the suppression of the big-bang and thermal inflation moduli, but they are based on the unknown parameters $`M`$ and $`\lambda `$. Since we still need to know if the range $`M\text{ }\stackrel{<}{}\text{ }10^{15}\mathrm{GeV}`$, required to obtain a low-energy scale inflation, is not forbidden by the requirements coming from the solution to the moduli problem, we must find a $`\lambda `$-independent relation on $`M`$. This relation can be found noting that the increment in the entropy due to the curvaton decay \[c.f. Eq. (3.76)\] can be written in an alternative way:
$$\mathrm{\Delta }_\sigma \left[\frac{g_{}(T_{\mathrm{dec}})}{g_{}(T_\chi )(1\mathrm{\Omega }_{\mathrm{dec}})^3}\right]^{1/4},$$
(3.82)
so the abundance of big-bang moduli after thermal inflation is:
$`{\displaystyle \frac{n_\mathrm{\Phi }}{s}}`$ $``$ $`{\displaystyle \frac{\mathrm{\Phi }^2}{10m_P^{3/2}m_\mathrm{\Phi }^{1/2}\mathrm{\Delta }_{PR}\mathrm{\Delta }_\chi \mathrm{\Delta }_\sigma }}{\displaystyle \frac{10\mathrm{\Phi }^2T_\chi T_C^3(1\mathrm{\Omega }_{\mathrm{dec}})^{3/4}}{\beta V_hm_\mathrm{\Phi }^{1/2}m_P^{3/2}}}`$ (3.83)
$`\stackrel{>}{}`$ $`10^{24}\mathrm{GeV}^3M^3\gamma _\chi ^{1/2}(1\mathrm{\Omega }_{\mathrm{dec}})^{3/4}\left({\displaystyle \frac{\mathrm{\Phi }}{m_P}}\right)^2\times `$
$`\times \left({\displaystyle \frac{T_C}{m_\mathrm{\Phi }}}\right)^3\left({\displaystyle \frac{m_\mathrm{\Phi }}{10^3\mathrm{GeV}}}\right)^{1/2}{\displaystyle \frac{1}{\beta }}\left({\displaystyle \frac{m_\mathrm{\Phi }^2M^2}{V_h}}\right).`$
This means that
$$M\text{ }\stackrel{>}{}\text{ }10^{12}\mathrm{GeV},$$
(3.84)
to satisfy $`n_\mathrm{\Phi }/s\text{ }\stackrel{<}{}\text{ }10^{12}`$. This is the lower bound on $`M`$ we have used throughout this chapter. A similar treatment to the abundance of thermal inflation moduli \[c.f. Eq. (3.80)\] leads to the bound $`M\text{ }\stackrel{<}{}\text{ }10^{16}`$ GeV, which is weaker than that obtained in Fig. 3.2.
Of course we might have considered the scenario where there are no moduli fields at all. Without the introduction of the moduli problem Eq. (3.84) becomes unnecessary. This does not help for the improvement of the required range of values for $`m_\sigma `$ but it does for $`\lambda `$ as the lower bound on $`M`$ in Eq. (3.84) becomes replaced by $`M10^3`$ GeV, which comes from the definition of the flaton fields. In this way the range of values for $`M`$ extends to smaller values well below $`10^{12}`$ GeV until the coupling constant $`\lambda `$ eventually reaches the lower bound $`10^4`$.
The introduction of a period of thermal inflation into our curvaton scenario, sketched in Fig. 3.1, has helped us not only to lower the energy scale of the main inflationary epoch, but also to solve the moduli problem still present after ordinary inflation. The required parameter space for $`\lambda `$ has been plotted in Fig. 3.2, and the vacuum expectation value for the flaton field has been showed to be in the range $`10^{12}\mathrm{GeV}\text{ }\stackrel{<}{}\text{ }M\text{ }\stackrel{<}{}\text{ }10^{15}\mathrm{GeV}`$. Our flaton field, in view of the allowed range of values for $`M`$, should be one of the gauge singlets present in string theory . The upper bound on $`m_\sigma `$ \[c.f. Eq. (3.72)\], $`m_\sigma \text{ }\stackrel{<}{}\text{ }10^1`$ GeV, suggests the curvaton field could be a PNGB . This is because in the presence of supergravity all the scalar fields, whose masses are not protected by a global symmetry, acquire soft masses of the order of the gravitino mass if $`H\text{ }\stackrel{<}{}\text{ }m_{3/2}`$, and contributions contributions to the squared mass of order $`H^2`$ if $`H\text{ }\stackrel{>}{}\text{ }m_{3/2}`$ except during the radiation dominated era . The smallness of the curvaton mass is in turn because of the very small value for $`H_{\mathrm{osc}}`$. The parameter $`H_{\mathrm{osc}}`$ is directly proportional to $`M`$, so the bigger $`M`$ is, the more possible to obtain a range of values for $`m_\sigma `$ compatible with the soft supersymmetric contributions. We, in the next chapter, will look for a mechanism to improve the required range of values for the bare mass $`m_\sigma `$ and the coupling constant $`\lambda `$ in presence of the moduli problem .
### 3.4 Conclusions
We have presented a different type of curvaton scenario , in which the scale of inflation can be much lower than $`H_{}10^7`$ GeV, which is the default lower bound for the standard curvaton model . This scenario considers a curvaton, whose mass, being appropriately Higgsed, is substantially enlarged at a phase transition after the end of inflation (‘heavy curvaton’). We have shown that this mechanism is indeed able to accommodate inflation scales as low as $`H_{}`$ 1 TeV or even lower.
We have implemented the idea of a thermal inflation epoch, introduced in Refs. to solve the moduli problem, as a second inflationary period necessary to lower the energy scale of the main inflationary stage. In our model, a flaton field $`\chi `$ with bare mass coming from soft supersymmetric contributions and vacuum expectation value in the range $`10^{12}\mathrm{GeV}\text{ }\stackrel{<}{}\text{ }M\text{ }\stackrel{<}{}\text{ }10^{15}\mathrm{GeV}`$ (i.e. one of the gauge singlets in string theory ), is held at the origin of the scalar potential by finite-temperature effects. These effects are associated to the thermal background created by the main reheating epoch. When temperature falls below $`V_h`$ thermal inflation begins. This period of thermal inflation lasts around ten e-folds until the temperature falls below $`m_\chi `$ liberating the flaton field to roll away towards the minimum of the potential. The curvaton field is coupled to the flaton one, through a coupling constant $`\lambda `$ in the range $`10^{22}\text{ }\stackrel{<}{}\text{ }\lambda \text{ }\stackrel{<}{}\text{ }10^{10}`$, so its mass is largely increased at the end of thermal inflation. This increment is enough to lower the bound on $`H_{}`$ to satisfactory levels, without sending the non-gaussianity constraint to the limit. However, the energy scale of the thermal inflation epoch is very small, requiring in turn a bare mass for the curvaton field of at most $`10^410^1`$ GeV. Taking into account the soft supersymmetric contributions to $`m_\sigma `$, the required smallness of $`m_\sigma `$ points toward using a PNGB curvaton to achieve low-scale inflation.
The type of mechanism that we presented is not completely compelling. It suffers from the problem that the mass of the curvaton before oscillation, as well as its coupling, have to be much smaller than one would expect. In the next chapter it will be shown how this tuning problem can be at least alleviated.
## Chapter 4 Low scale inflation and the immediate heavy curvaton decay
### 4.1 Introduction
Low scale inflation is desirable in order to identify the inflaton field with one of the MSSM flat directions or with one of the fields appearing in the SUSY breaking sector, giving the inflaton a much deeper particle physics root. In contrast low scale inflation is not desirable because it makes very difficult the generation of the adiabatic perturbations by the inflaton, leading to multiple fine-tuning and model-building problems, unless the curvaton mechanism is invoked (see also Refs. ). With the aim of generating the curvature perturbation that gives origin to the large-scale structure in the observable universe, the curvaton mechanism has appeared as a nice and plausible option and a lot of research has been devoted to its study. Making the curvaton mechanism viable in a low energy inflationary framework would be the ideal situation but, unfortunately, the simplest curvaton model has shown to be incompatible with low enough values for the Hubble parameter during inflation . Some general proposals to make the curvaton paradigm accommodate low scale inflation have recently appeared and specific models have been studied too . In the previous chapter, a thermal inflation epoch was attached to the general curvaton mechanism making the curvaton field gain a huge increment in the mass at the end of the thermal inflationary period, triggering this way a period of curvaton oscillations, and lowering the main inflationary scale to satisfactory levels . However, the parameters of the model required for this effect to take place showed to be extremely small to affect the reliability of the model. The purpose of this chapter is to study the same mechanism but in the case where the increment in the mass is so huge that the decay rate becomes bigger than the Hubble parameter and the curvaton decays immediately . The results are very positive, offering a more natural parameter space.
### 4.2 Thermal inflation and the immediate heavy curvaton decay
The thermal inflation model has been investigated before and found to be a very efficient mechanism to dilute the abundance of some unwanted relics, like the moduli fields, that the main inflationary epoch is not able to get rid of (see Refs. ). We will constrain the available parameter space for $`\lambda `$ and $`m_\sigma `$ in the scalar potential of Eq. (3.41) so that enough dilution of the moduli abundance is obtained. In Chapter 3 this was done for the case in which the flaton-curvaton coupling term gives a huge contribution to the mass of the curvaton when the flaton field is released and gets its vacuum expectation value $`M`$. In that case the effective curvaton mass $`\stackrel{~}{m}_\sigma `$ may become bigger than the Hubble parameter giving birth to a period of curvaton oscillations and making the scale of the main inflationary period low enough ($`H_{}m_{3/2}10^3\mathrm{GeV}`$) to think about the inflaton as a field associated to the SUSY breaking sector . The evolution of the energy densities associated to the different fluids in that case are sketched in Fig. 3.1.
The purpose of this chapter is to analyse the scenario where there are no oscillations of the curvaton field. As it was pointed out in Ref. both the curvaton decay rate and the associated lower bound \[c.f. Eq. (3.23)\]
$$\mathrm{\Gamma }_\sigma \frac{\stackrel{~}{m}_\sigma ^3}{m_P^2},$$
(4.1)
are also increased when the flaton field acquires its vacuum expectation value so that, if this increment is big enough for the curvaton decay rate to be bigger than the Hubble parameter, the curvaton field may decay immediately rather than oscillating for some time. Low scale inflation in this case is also possible to be attained , but the lower bound on $`H_{}`$ changes with respect to the case when the curvaton oscillatory process is triggered. The evolution of the energy densities associated to the different fluids in this case are sketched in Fig. 4.1.
In the scenario where curvaton oscillations are allowed, corresponding to $`\mathrm{\Gamma }_\sigma <H_{\mathrm{pt}}`$, the lower bound on $`H_{}`$ is \[c.f. Eqs. (3.22), (3.26), and (3.30)\]
$$H_{}\mathrm{max}\{f^{4/5}\times 10^7\mathrm{GeV},\sqrt{f}\delta ^2\times 10^{11}\mathrm{GeV},\delta ^4\times 10^7\mathrm{GeV}\},$$
(4.2)
where $`f`$ and $`\delta `$, given by Eqs. (3.13) and (3.27), are less than 1, and with $`H_{\mathrm{osc}}=H_{\mathrm{pt}}`$ being the Hubble parameter at the end of the thermal inflation period (which also corresponds to the beginning of the curvaton oscillations). In contrast, the lower bound in the scenario where the curvaton field decays immediately, corresponding to $`\mathrm{\Gamma }_\sigma >H_{\mathrm{pt}}`$, is \[c.f. Eqs. (3.22) and (3.34)\]
$$H_{}\mathrm{max}\{f^{4/5}\times 10^7\mathrm{GeV},f\times 10^{14}\mathrm{GeV}\},$$
(4.3)
which may challenge the WMAP constraint for the curvaton scenario leading to excessive curvature perturbations from the inflaton field if $`f`$ is not much smaller than unity.
The lower bound in Eq. (4.2), for $`H_{}10^3`$ GeV, was shown in the previous chapter to be satisfied for very small values for the flaton-curvaton coupling constant, $`\lambda 10^{22}10^{10}`$ (see Fig. 3.2), and very small values for the bare mass of the curvaton field, $`m_\sigma \text{ }\stackrel{<}{}\text{ }10^1\mathrm{GeV}`$ \[c.f. Eq. (3.72)\], which suggests that the curvaton field could be a PNGB . This is, in any case, a quite negative result due to the required smallness of the parameters $`\lambda `$ and $`m_\sigma `$. However, when taking into account the lower bound in Eq. (4.3), corresponding to the case when the decay rate $`\mathrm{\Gamma }_\sigma `$ becomes bigger than $`H_{\mathrm{pt}}`$, things change appreciably.
#### 4.2.1 The flaton-curvaton coupling constant $`\lambda `$
Thermal inflation ends when the thermal energy density is no longer dominant; thus, the Hubble parameter at the end of thermal inflation is associated to the energy density coming from the curvaton and the flaton fields:
$$H_{\mathrm{pt}}^2=\frac{\rho _T+V(\chi =0,\sigma _0=\sigma _{})}{3m_P^2}\frac{m_\chi ^2M^2}{3m_P^2},$$
(4.4)
so that
$$H_{\mathrm{pt}}10^{16}M.$$
(4.5)
Since the effective mass of the curvaton field after the end of thermal inflation, i.e., when $`\overline{\chi }=M_\chi `$ is the average over oscillations of the flaton field and $`\sigma _0=0`$, is
$$\stackrel{~}{m}_\sigma =(m_\sigma ^2+\lambda M^2)^{1/2}\sqrt{\lambda }M,$$
(4.6)
the parameter $`f`$ \[cf. Eq. (3.13)\] becomes
$$f\frac{H_{\mathrm{pt}}}{\stackrel{~}{m}_\sigma }10^{16}\frac{1}{\sqrt{\lambda }}.$$
(4.7)
In view of the Eqs. (4.3) and (4.7) the smallest possible value for $`\lambda `$, compatible with $`H_{}10^3\mathrm{GeV}`$, becomes $`\lambda 10^{10}`$, which is very good because this already improves the results found in the previous chapter. Moreover, the effective flaton mass during thermal inflation $`\stackrel{~}{m}_\chi =(m_\chi ^2\lambda \sigma _{}^2)^{1/2}`$ must be positive to trap the flaton field at the origin of the potential. Thus, $`\lambda \sigma _{}^2<m_\chi ^2`$, and the biggest possible value for $`\lambda `$ becomes \[c.f. Eq. (3.11)\]
$$\lambda <\frac{m_\chi ^2}{\sigma _{}^2}\frac{10^2\mathrm{GeV}^2}{\mathrm{\Omega }_{\mathrm{dec}}^2H_{}^2}\text{ }\stackrel{<}{}\text{ }10^4,$$
(4.8)
which is already a small value but much bigger and more natural than that found in the case where curvaton oscillations are allowed. The lower bound on $`\lambda `$ vs $`\mathrm{\Omega }_{\mathrm{dec}}`$ is depicted in Fig. 4.2. Note that a small value for $`\mathrm{\Omega }_{\mathrm{dec}}`$, which is restricted to be $`\mathrm{\Omega }_{\mathrm{dec}}0.01`$ in order to satisfy the WMAP constraints on non gaussianity , is desirable to obtain a higher value for $`\lambda `$, so the biggest possible value $`\lambda 10^4`$ is at the expense of a high level of non gaussianity. A smaller upper bound for $`\lambda `$ could be a possibility, according to Eq. (4.8) and Fig. 4.2, by increasing $`\mathrm{\Omega }_{\mathrm{dec}}`$. Nevertheless $`\mathrm{\Omega }_{\mathrm{dec}}`$, in the scenario studied in this chapter, must satisfy $`\mathrm{\Omega }_{\mathrm{dec}}<1`$ to avoid a period of inflation driven by the curvaton field.
Recalling, in the scenario where curvaton oscillations are allowed the coupling constant $`\lambda `$ is in the range (see Fig. 3.2)
$$10^{22}\text{ }\stackrel{<}{}\text{ }\lambda \text{ }\stackrel{<}{}\text{ }10^{10},$$
(4.9)
whereas in the scenario where the curvaton decays immediately the range is
$$10^{10}\text{ }\stackrel{<}{}\text{ }\lambda \text{ }\stackrel{<}{}\text{ }10^4.$$
(4.10)
It is easy to see that the allowed range of values for $`\lambda `$ in Eq. (4.10), valid for the case where the curvaton field decays immediately at the end of the thermal inflation era, is complementary to the allowed range for $`\lambda `$ in Eq. (4.9), valid when the curvaton has some time to oscillate before decaying (see Subsection 3.3.2).
#### 4.2.2 The bare curvaton mass $`m_\sigma `$
The only bound on $`m_\sigma `$ is given by the fact that in the heavy curvaton scenario the bare mass must be smaller than the Hubble parameter at the end of the thermal inflation era, so that the sudden increment in the mass and the decay rate leads to the immediate decay of the field avoiding in this case the oscillations. Thus,
$$m_\sigma <H_{\mathrm{pt}}10^{16}M,$$
(4.11)
so we need to worry about the possible values for $`M`$. In the scenario where the curvaton field decays immediately the flaton field is left immersed in a background of radiation, so it must decay before the time of nucleosynthesis in order not to disturb the abundances of the light elements. By setting $`\mathrm{\Gamma }_\chi H`$ we get the temperature just after the flaton decay
$$T_\chi 10^{13}\mathrm{GeV}^2\frac{1}{M},$$
(4.12)
which must be bigger than $`1`$ MeV to satisfy the nucleosynthesis constraint. Therefore
$$M\text{ }\stackrel{<}{}\text{ }10^{16}\mathrm{GeV},$$
(4.13)
leading to an upper bound on the bare curvaton mass given by $`m_\sigma \text{ }\stackrel{<}{}\text{ }1\mathrm{GeV}`$, which is again a more relaxed constraint than that found in the previous chapter for the case of an oscillating curvaton, but that still reduces the number of possible curvaton candidates, leaving essentially the PNGB . Recalling, in the scenario where curvaton oscillations are allowed the bare curvaton mass $`m_\sigma `$ is in the range \[c.f. Eq. (3.72)\]
$$m_\sigma \text{ }\stackrel{<}{}\text{ }10^1\mathrm{GeV},$$
(4.14)
whereas in the scenario where the curvaton decays immediately the range is
$$m_\sigma \text{ }\stackrel{<}{}\text{ }1\mathrm{GeV}.$$
(4.15)
Some important constraints might come from the solution to the moduli problem and could limit the reliability of the Eqs. (4.13) and (4.15). Moduli fields are flaton fields with a vacuum expectation value $`\mathrm{\Phi }`$ of order the Planck mass. The decay of the flaton field increments the entropy density $`s`$, so that the big-bang moduli abundance, defined as that produced before thermal inflation and given by
$$\frac{n_\mathrm{\Phi }}{s}\frac{\mathrm{\Phi }^2}{10m_P^{3/2}m_\mathrm{\Phi }^{1/2}},$$
(4.16)
where $`m_\mathrm{\Phi }`$ is the mass of the moduli fields, gets suppressed by three factors<sup>1</sup><sup>1</sup>1Eqs. (4.17) and (4.18) correct a mistake in Ref. . However Eqs. (4.20) and (4.22) are not affected by that mistake and, therefore, the conclusions in Ref. about the lower and upper bounds on $`M`$ remain unchanged.. One is
$$\mathrm{\Delta }_\sigma \frac{g_{}(T_\sigma )}{g_{}(T_C)}\frac{T_\sigma ^3}{T_C^3},$$
(4.17)
due to the curvaton decay, where the $`g_{}`$ are the total internal degrees of freedom, $`T_\sigma `$ is the temperature just after the curvaton decay, and $`T_Cm_\chi `$ is the temperature at the end of thermal inflation; another is
$$\mathrm{\Delta }_{PR}\frac{g_{}(T_{PR})}{g_{}(T_\sigma )}\frac{T_{PR}^3}{T_\sigma ^3},$$
(4.18)
due to the parametric resonance process following the end of the thermal inflation era, where $`T_{PR}`$ is the temperature just after the period of preheating; and the other is
$$\mathrm{\Delta }_\chi \frac{4\beta V_h/3T_\chi }{(2\pi ^2/45)g_{}(T_{PR})T_{PR}^3},$$
(4.19)
due to the flaton decay, where $`T_\chi `$ is the temperature just after the decay<sup>2</sup><sup>2</sup>2This is assuming for simplicity that the flaton has come to dominate the energy density just before decaying (see Fig. 4.1)., and $`\beta `$ is the fraction of the total energy density left in the flatons by the parametric resonance process ($`\beta \text{ }\stackrel{<}{}\text{ }1`$). Thus, the abundance of the big-bang moduli after thermal inflation is:
$`{\displaystyle \frac{n_\mathrm{\Phi }}{s}}`$ $``$ $`{\displaystyle \frac{\mathrm{\Phi }^2}{10m_P^{3/2}m_\mathrm{\Phi }^{1/2}\mathrm{\Delta }_\sigma \mathrm{\Delta }_{PR}\mathrm{\Delta }_\chi }}{\displaystyle \frac{10\mathrm{\Phi }^2T_\chi T_C^3}{\beta V_hm_\mathrm{\Phi }^{1/2}m_P^{3/2}}}`$ (4.20)
$`\stackrel{>}{}`$ $`10^6\mathrm{GeV}^2M^2\left({\displaystyle \frac{\mathrm{\Phi }}{m_P}}\right)^2\left({\displaystyle \frac{T_\chi }{1\mathrm{MeV}}}\right)\left({\displaystyle \frac{T_C}{m_\mathrm{\Phi }}}\right)^3\times `$
$`\times \left({\displaystyle \frac{m_\mathrm{\Phi }}{10^3\mathrm{GeV}}}\right)^{1/2}\left({\displaystyle \frac{1}{\beta }}\right)\left({\displaystyle \frac{m_\mathrm{\Phi }^2M^2}{V_h}}\right).`$
which must be suppressed enough ($`n_\mathrm{\Phi }/s\text{ }\stackrel{<}{}\text{ }10^{12}`$) so that the nucleosynthesis constraints studied in Ref. are satisfied. This is easily achieved by imposing a lower bound on $`M`$:
$$M\text{ }\stackrel{>}{}\text{ }10^9\mathrm{GeV},$$
(4.21)
which does not affect the upper bounds on $`M`$ and $`m_\sigma `$ in Eqs. (4.13) and (4.15).
We also have to take care about the abundance of the thermal inflation moduli, defined as that produced during the preheating stage following the end of the thermal inflation era:
$`{\displaystyle \frac{n_{\mathrm{\Phi }_T}}{s}}`$ $``$ $`{\displaystyle \frac{\mathrm{\Phi }_T^2V_h^2/10m_\mathrm{\Phi }^3m_P^4}{(2\pi ^2/45)g_{}(T_{PR})T_{PR}^3\mathrm{\Delta }_\chi }}{\displaystyle \frac{\mathrm{\Phi }_T^2V_hT_\chi }{10\beta m_\mathrm{\Phi }^3m_P^4}}`$ (4.22)
$`\stackrel{>}{}`$ $`10^{44}\mathrm{GeV}^2M^2\left({\displaystyle \frac{\mathrm{\Phi }_T}{m_P}}\right)^2\left({\displaystyle \frac{T_\chi }{1\mathrm{MeV}}}\right)\times `$
$`\times \left({\displaystyle \frac{1}{\beta }}\right)\left({\displaystyle \frac{10^3\mathrm{GeV}}{m_\mathrm{\Phi }}}\right)\left({\displaystyle \frac{V_h}{m_\mathrm{\Phi }^2M^2}}\right).`$
Here $`\mathrm{\Phi }_T`$ corresponds to the vacuum expectation value of the thermal moduli fields. To suppress the thermal inflation moduli at the required level $`n_{\mathrm{\Phi }_T}/s\text{ }\stackrel{<}{}\text{ }10^{12}`$ we require
$$M\text{ }\stackrel{<}{}\text{ }10^{16}\mathrm{GeV},$$
(4.23)
which is precisely the same bound as in Eq. (4.13). Recalling, the allowed range of values for the vacuum expectation value of the flaton field is
$$10^9\mathrm{GeV}\text{ }\stackrel{<}{}\text{ }M\text{ }\stackrel{<}{}\text{ }10^{16}\mathrm{GeV},$$
(4.24)
so that the moduli problem is solved and, in the best case, $`m_\sigma 1`$ GeV. The latter allowed range for $`M`$ means that, unlike the case where the curvaton has some time to oscillate before decaying, the flaton field could be the GUT Higgs field studied in Ref. .
### 4.3 Some useful remarks
Before concluding, we want to stress some points that can help to avoid possible confusion. The parameter space compatible with low scale inflation is a feature of the specific model studied, and we cannot say it is the same for all classes of models in the basis of Eqs. (3.22), (3.26), and (3.30), which provide just some general bounds. That is why specific models have been studied (see Refs. ), even when the general bounds were already known from Refs. . Although the claim, that the available parameter space is bigger for the immediate curvaton decay, was given before in Ref. , we again cannot say that the available parameter space is the same for all classes of models in the basis of the bounds required to have low energy scale inflation. For example, from Eq. (4.5), $`H_{\mathrm{pt}}`$ depends on $`M`$ so there is no direct bound on it unless we know the bound on $`M`$ <sup>3</sup><sup>3</sup>3Notice that the bounds required to have low energy scale inflation \[c.f. Eq. (4.3)\] depend only on the ratio $`f=H_{\mathrm{pt}}/\stackrel{~}{m}_\sigma `$, and not exclusively on $`H_{\mathrm{pt}}`$.. The bound on $`M`$ comes in turn from the requirement that the flaton decays before nucleosynthesis \[c.f. Eqs. (4.12) and (4.13)\] and must be consistent with the adequate suppression of the thermal inflation moduli \[c.f. Eqs. (4.22) and (4.23)\]. These are, of course, features specific only to the model we are studying, and are therefore not present in Ref. .
Naively, one would think that the bounds on $`\lambda `$ and $`m_\sigma `$ are found from that on $`H_{\mathrm{pt}}`$ only through a mere change of variables. This is of course not true as the bound on $`H_{\mathrm{pt}}`$ is a very sensitive quantity that has to avoid disturbing the nucleosynthesis process and the adequate moduli abundance suppression. It is worth mentioning that the scenario discussed in this chapter differs appreciably from that studied in Chapter 3, due to the immediate curvaton decay, so that the conditions to satisfy the nucleosynthesis and thermal inflation moduli constraints are completely different<sup>4</sup><sup>4</sup>4For example, the expressions for the big-bang and thermal inflation moduli abundances in Chapter 3 \[c.f. Eqs. (3.78) and (3.80)\] are different from those in this chapter \[c.f. Eqs. (4.20) and (4.22)\]..
Finally, the agreement between the bounds found in Ref. (which are supposed to be general) and those found in this chapter is apparent and corresponds just to a mere coincidence. We justify this observation by noting that Eq. (6) in Ref. is essentially the same as our Eq. (3.16), the latter being generalized to give Eq. (3.22), except for $`\mathrm{\Gamma }_\sigma `$ which in our Eq. (3.16) appears to be $`H_{\mathrm{dec}}`$. The expressions in the previous chapter were carefully derived so that the correct expression is that given there . In contrast, Eq. (6) in Ref. is just valid for the standard case where the curvaton field has some time to oscillate before decaying, so we can identify $`\mathrm{\Gamma }_\sigma `$ with $`H_{\mathrm{dec}}`$. However, for the immediate decay case, $`\mathrm{\Gamma }_\sigma >H_{\mathrm{dec}}=H_{\mathrm{pt}}`$, which renders Eq. (6) in Ref. invalid. Based on the previous discussion we claim that the bound $`H_{\mathrm{pt}}<1`$ GeV, as are those on $`\lambda `$ and $`m_\sigma `$, is presented in this thesis for the first time in a correct way.
### 4.4 Conclusions
In this chapter we have investigated the required parameter space compatible with low scale inflation in the thermal inflation curvaton scenario where there are no oscillations of the curvaton field . We have shown that the parameter space is greatly enhanced when the increment in the curvaton decay rate is big enough for the curvaton field to decay immediately at the end of the thermal inflation era. The best case corresponds to a flaton-curvaton coupling constant $`\lambda 10^4`$ and a bare curvaton mass $`m_\sigma 1\mathrm{GeV}`$, which are much bigger and more natural than the ranges $`10^{22}\text{ }\stackrel{<}{}\text{ }\lambda \text{ }\stackrel{<}{}\text{ }10^{10}`$ and $`m_\sigma \text{ }\stackrel{<}{}\text{ }10^1\mathrm{GeV}`$ found previously in Chapter 3 for the case where the curvaton oscillates for some time before decaying . In addition we have found $`10^9\mathrm{GeV}\text{ }\stackrel{<}{}\text{ }M\text{ }\stackrel{<}{}\text{ }10^{16}\mathrm{GeV}`$ for the vacuum expectation value $`M`$ of the flaton field. Therefore, our flaton field as the GUT Higgs field discussed in Ref. is a viable option in this scenario.
## Chapter 5 Non-gaussianity from the second-order cosmological perturbation
### 5.1 Introduction
In chapters 2, 3, and 4, we discussed some of the theoretical aspects of the origin of the large-scale structure in the Universe, emphasising the possibility to achieve low scale inflation in the curvaton scenario. We begin now the discussion of the statistical aspects, specifically the presence of non-gaussianities in the fields responsible for the origin of the curvature perturbation $`\zeta `$ and/or in $`\zeta `$ itself. Chapters 5 and 6 will deal with such an interesting subject.
Cosmological scales leave the horizon during inflation and re-enter it after Big Bang Nucleosynthesis. Throughout the super-horizon era it is very useful to define a primordial cosmological curvature perturbation, which is conserved if and only if pressure throughout the Universe is a unique function of energy density (the adiabatic pressure condition) (see Subsection 2.2.3) . Observation directly constrains the curvature perturbation at the very end of the super-horizon era, a few Hubble times before cosmological scales start to enter the horizon, when it apparently sets the initial condition for the subsequent evolution of all cosmological perturbations. As discussed in Chapter 2, the observed curvature perturbation is almost Gaussian with an almost scale-invariant spectrum.
Cosmological perturbation theory expands the exact equations in powers of the perturbations and keeps terms only up to the $`n`$th order. Since the observed curvature perturbation is of order $`10^5`$, one might think that first-order perturbation theory will be adequate for all comparisons with observation. That may not be the case however, because the PLANCK satellite and its successors may be sensitive to non-gaussianity of the curvature perturbation at the level of second-order perturbation theory .
Several authors have treated the non-gaussianity of the primordial curvature perturbation in the context of second-order perturbation theory. They have adopted different definitions of the curvature perturbation and obtained results for a variety of situations. In this chapter we revisit the calculations, using a single definition of the curvature perturbation which we denote by $`\zeta `$ . In some cases we disagree with the findings of the original authors.
The outline of this chapter is the following: in Section 5.2 we review two definitions of the curvature perturbation found in the literature, which are valid during and after inflation, and establish definite relationships between them; in section 5.3 we review a third curvature perturbation definition, which applies only during inflation, and study it in models of inflation of the slow-roll variety; in Section 5.4 we describe the present framework for thinking about the origin and evolution of the curvature perturbation; in Section 5.5 we see how non-gaussianity is defined and constrained by observation; in Section 5.6 we study the initial non-gaussianity of the curvature perturbation, a few Hubble times after horizon exit; in Section 5.7 we study its subsequent evolution according to some different models. The conclusions are summarised in Section 5.8.
We shall denote unperturbed quantities by a subscript $`0`$, and generally work with conformal time $`\eta `$. Sometimes though we revert to physical time $`t`$. We shall adopt the convention that a generic perturbation $`g`$ is split into a first- and second-order part according to the formula
$$g=g_1+\frac{1}{2}g_2.$$
(5.1)
### 5.2 Two definitions of the curvature perturbation
#### 5.2.1 Preliminaries
Cosmological perturbations describe small departures of the actual Universe, away from some perfect homogeneous and isotropic universe with the line element in Eq. (2.2). For a generic perturbation it is convenient to make the Fourier expansion
$$g(𝐱,\eta )=\frac{1}{(2\pi )^{3/2}}d^3kg_𝐤(\eta )e^{i𝐤𝐱},$$
(5.2)
where the spacetime coordinates are those of the unperturbed Universe. The inverse of the comoving wavenumber, $`k^1`$, is often referred to as the scale. Except where otherwise stated, our discussion applies only to the super-horizon regime ($`kaH_{\mathrm{inf}}`$).
When evaluating an observable quantity only a limited range of scales will be involved. The largest scale, relevant for the low multipoles of the Cosmic Microwave Background anisotropy, is $`k^1H_{\mathrm{today}}^1`$ where $`H_{\mathrm{today}}`$ is the present Hubble parameter. The smallest scale usually considered is the one enclosing matter with mass $`10^6M_{}`$, which corresponds to $`k^110^2\text{Mpc}10^6H_{\mathrm{today}}^1`$. The cosmological range of scales therefore extends over only six orders of magnitude or so.
To define cosmological perturbations in general, one has to introduce in the perturbed Universe a coordinate system $`(t,x^i)`$, which defines a slicing of spacetime (fixed $`t`$) and a threading (fixed $`x^i`$). To define the curvature perturbation it is enough to define the slicing .
#### 5.2.2 Two definitions of the curvature perturbation
In this chapter we take as our definition of $`\zeta `$ the following expression for the spatial metric which applies non-perturbatively:
$$g_{ij}=a^2(\eta )\stackrel{~}{\gamma }_{ij}e^{2\zeta }.$$
(5.3)
Here $`\stackrel{~}{\gamma }_{ij}`$ has unit determinant, and the time-slicing is one of uniform energy density<sup>1</sup><sup>1</sup>1It is proved in Ref. that this definition of $`\zeta `$ coincides with that of Lyth and Wands , provided that their slices of uniform coordinate expansion are taken to correspond to those on which the line element has the form Eq. (5.3) without the factor $`e^{2\zeta }`$ (this makes the slices practically flat if $`\stackrel{~}{\gamma }_{ij}\delta _{ij}`$)..
It has been shown under weak assumptions that this defines $`\zeta `$ uniquely, and that $`\zeta `$ is conserved as long as the pressure is a unique function of energy density. Also, it has been shown that the uniform density slicing practically coincides with the comoving slicing (orthogonal to the flow of energy), and with the uniform Hubble slicing (corresponding to uniform proper expansion, that expansion being practically independent of the threading which defines it) . The coincidence of these slicings is important since all three have been invoked by different authors.
Since the matrix $`\stackrel{~}{\gamma }`$ has unit determinant it can be written $`\stackrel{~}{\gamma }=Ie^h`$, where $`I`$ is the unit matrix and $`h`$ is traceless . Assuming that the initial condition is set by inflation, $`h`$ corresponds to a tensor perturbation (gravitational wave amplitude) which will be negligible unless the scale of inflation is very high. As we shall see later (see footnote 12 in this chapter), the results we are going to present are valid even if $`h`$ is not negligible, but to simplify the presentation we drop $`h`$ from the equations. Accordingly, the space part of the metric in the super-horizon regime is supposed to be well approximated by
$$g_{ij}=a^2(\eta )\delta _{ij}e^{2\zeta }.$$
(5.4)
At first order, Eq. (5.4) corresponds to
$$g_{ij}=a^2(\eta )\delta _{ij}(1+2\zeta ).$$
(5.5)
Up to a sign, this is the definition of the first-order curvature perturbation adopted by all authors \[c.f. Eqs. (2.3) and (2.15)\]. There is no universally agreed convention for the sign of $`\zeta `$. Ours coincides with the convention of most of the papers to which we refer, and we have checked carefully that the signs in our own set of equations are correct.
At second order we have
$$g_{ij}=a^2(\eta )\delta _{ij}(1+2\zeta +2\zeta ^2).$$
(5.6)
This is our definition of $`\zeta `$ at second order .
Malik and Wands instead defined $`\zeta `$ by Eq. (5.5) even at second order. Denoting their definition by a subscript MW,
$$\zeta ^{\mathrm{MW}}=\zeta +\zeta ^2,$$
(5.7)
or equivalently
$$\zeta _2^{\mathrm{MW}}=\zeta _2+2\left(\zeta _1\right)^2,$$
(5.8)
where $`\zeta _1`$ is the first-order quantity whose definition Eq. (5.5) is agreed by all authors.
To make contact with calculations of the curvature perturbation during inflation, we need some gauge-invariant expressions for the curvature perturbation. As stated in Subsections 2.2.2 and 2.2.3, ‘gauge-invariant’ means that the definition is valid for any choice of the coordinate system which defines the slicing and threading<sup>2</sup><sup>2</sup>2In the unperturbed limit the slicing has to be the one on which all quantities are uniform and the the threading has to be orthogonal to it..
We shall write gauge-invariant expressions in terms of $`\zeta `$ and $`\zeta ^{\mathrm{MW}}`$. First we consider a quantity $`\psi ^{\mathrm{MW}}`$, defined even at second order by
$$g_{ij}=a^2(\eta )\delta _{ij}(12\psi ^{\mathrm{MW}}).$$
(5.9)
This definition, which is written in analogy to Eq. (5.5), applies to a generic slicing. Analogously to Eq. (5.4) we can consider a quantity $`\psi `$, valid also in a generic slicing, defined by
$$g_{ij}=a^2(\eta )\delta _{ij}e^{2\psi }.$$
(5.10)
On uniform-density slices, $`\psi _1=\psi _1^{\mathrm{MW}}=\zeta _1`$, $`\psi _2^{\mathrm{MW}}=\zeta _2^{\mathrm{MW}}`$, and $`\psi _2=\zeta _2`$. We shall also need the energy density perturbation $`\delta \rho `$, defined on the generic slicing, as well as the unperturbed energy density $`\rho _0`$.
At first order, the gauge-invariant expression for $`\zeta `$ has the well-known form \[c.f. Eq. (2.15)\]
$$\zeta _1=\psi _1\frac{\delta \rho _1}{\rho _0^{}},$$
(5.11)
where $`=a^{}/a`$, and the unperturbed energy density satisfies $`\rho _0^{}=3(\rho _0+P_0)`$ with $`P_0`$ being the unperturbed pressure. This expression obviously is correct for the uniform density slicing, and it is correct for all slicings because the changes in the first and second terms induced by a change in the slicing cancel .
At second order, Malik and Wands show that
$`\zeta _2^{\mathrm{MW}}`$ $`=`$ $`\psi _2^{\mathrm{MW}}{\displaystyle \frac{\delta \rho _2}{\rho _0^{}}}+2{\displaystyle \frac{\delta \rho _1}{\rho _0^{}}}{\displaystyle \frac{\delta \rho _1^{}}{\rho _0^{}}}+2{\displaystyle \frac{\delta \rho _1}{\rho _0^{}}}\left(\psi _1^{}+2\psi _1\right)`$ (5.12)
$`\left({\displaystyle \frac{\delta \rho _1}{\rho _0^{}}}\right)^2\left({\displaystyle \frac{\rho _0^{\prime \prime }}{\rho _0^{}}}{\displaystyle \frac{^{}}{^2}}2\right),`$
which is, again and for the same reason as before, obviously correct for all the slices. Accordingly, from Eq. (5.8), we can write a gauge invariant definition for our second-order $`\zeta `$: <sup>3</sup><sup>3</sup>3This relation has recently been confirmed in Ref. (see also Ref. ) using a nonlinear coordinate-free approach.
$$\zeta _2=\psi _2\frac{\delta \rho _2}{\rho _0^{}}+2\frac{\delta \rho _1}{\rho _0^{}}\frac{\delta \rho _1^{}}{\rho _0^{}}+2\frac{\delta \rho _1}{\rho _0^{}}\psi _1^{}\left(\frac{\delta \rho _1}{\rho _0^{}}\right)^2\left(\frac{\rho _0^{\prime \prime }}{\rho _0^{}}\frac{^{}}{^2}\right),$$
(5.13)
where the relation
$$\psi _2^{\mathrm{MW}}=\psi _22(\psi _1)^2,$$
(5.14)
coming from Eqs. (5.9) and (5.10), has been used.
### 5.3 Slow-roll inflation and a third definition
Now we specialize to the era of slow-roll inflation . We consider single-component inflation, during which the curvature perturbation $`\zeta `$ is conserved, and multi-component inflation during which it varies. After defining both paradigms, we give a third definition of the curvature perturbation which applies only during inflation.
#### 5.3.1 Single-component inflation
In a single-component inflation model the inflaton trajectory is by definition essentially unique. The inflaton field $`\phi `$ parameterises the distance along the inflaton trajectory. In terms of the field variation, slow-roll inflation (see Subsection 2.5.1) is characterised by the slow-roll conditions
$`\epsilon `$ $``$ $`\left({\displaystyle \frac{\dot{H}_{\mathrm{inf}}}{H_{\mathrm{inf}}^2}}\right)_{}1,`$ (5.15)
$`|\eta _\phi \epsilon |`$ $``$ $`\left|{\displaystyle \frac{\ddot{\phi }_0}{H_{\mathrm{inf}}\dot{\phi }_0}}\right|1.`$ (5.16)
The inflaton field can be taken to be canonically normalised, in which case these definitions are equivalent to conditions on the potential $`V`$
$`\epsilon `$ $``$ $`{\displaystyle \frac{m_P^2}{2V^2}}\left({\displaystyle \frac{V}{\phi _0}}\right)^2,`$ (5.17)
$`\eta _\phi `$ $``$ $`{\displaystyle \frac{m_P^2}{V}}{\displaystyle \frac{^2V}{\phi _0^2}},`$ (5.18)
which, together with the slow-roll approximation, lead to the slow-roll behaviour
$$3H_{\mathrm{inf}}\dot{\phi }_0\frac{V}{\phi _0}.$$
(5.19)
Even without the slow-roll approximation, slices of uniform $`\phi `$ correspond to comoving slices because a spatial gradient of $`\phi `$ would give non-vanishing momentum density. Since comoving slices coincide with slices of uniform energy density, the slices of uniform $`\phi `$ coincide also with the latter. Also, since $`\phi `$ is a Lorentz scalar, its gauge transformation is the same as that of $`\rho `$. It follows that we can replace $`\rho `$ by $`\phi `$ in the above expressions:
$`\zeta _1`$ $`=`$ $`\psi _1_{\mathrm{inf}}{\displaystyle \frac{\delta \phi _1}{\phi _0^{}}},`$ (5.20)
$`\zeta _2^{\mathrm{MW}}`$ $`=`$ $`\psi _2^{\mathrm{MW}}_{\mathrm{inf}}{\displaystyle \frac{\delta \phi _2}{\phi _0^{}}}+2_{\mathrm{inf}}{\displaystyle \frac{\delta \phi _1}{\phi _0^{}}}{\displaystyle \frac{\delta \phi _1^{}}{\phi _0^{}}}+2{\displaystyle \frac{\delta \phi _1}{\phi _0^{}}}\left(\psi _1^{}+2_{\mathrm{inf}}\psi _1\right)`$ (5.21)
$`\left(_{\mathrm{inf}}{\displaystyle \frac{\delta \phi _1}{\phi _0^{}}}\right)^2\left({\displaystyle \frac{\phi _0^{\prime \prime }}{_{\mathrm{inf}}\phi _0^{}}}{\displaystyle \frac{_{\mathrm{inf}}^{}}{_{\mathrm{inf}}^2}}2\right),`$
$`\zeta _2`$ $`=`$ $`\psi _2_{\mathrm{inf}}{\displaystyle \frac{\delta \phi _2}{\phi _0^{}}}+2_{\mathrm{inf}}{\displaystyle \frac{\delta \phi _1}{\phi _0^{}}}{\displaystyle \frac{\delta \phi _1^{}}{\phi _0^{}}}+2{\displaystyle \frac{\delta \phi _1}{\phi _0^{}}}\psi _1^{}`$ (5.22)
$`\left(_{\mathrm{inf}}{\displaystyle \frac{\delta \phi _1}{\phi _0^{}}}\right)^2\left({\displaystyle \frac{\phi _0^{\prime \prime }}{_{\mathrm{inf}}\phi _0^{}}}{\displaystyle \frac{_{\mathrm{inf}}^{}}{_{\mathrm{inf}}^2}}\right).`$
#### 5.3.2 Multi-component inflation
Now consider the case of multi-component inflation, where there is a family of inequivalent inflationary trajectories lying in an $`N`$-dimensional manifold of field space. If the relevant part of the manifold is not too big it will be a good approximation to take the fields to be canonically normalised. Then the inequivalent trajectories will be curved in field space<sup>4</sup><sup>4</sup>4More generally they will be non-geodesics, the geodesics being the trajectories which the background fields could follow if there was no potential term in the scalar Lagrangian .. To define the trajectories one can choose a fixed basis in field space corresponding to fields $`\varphi _1,\mathrm{},\varphi _N`$.
Assuming canonical normalisation, multi-component slow-roll inflation is characterised by the conditions
$`{\displaystyle \frac{m_P^2}{2V^2}}\left({\displaystyle \frac{V}{\varphi _{n_0}}}\right)^2`$ $``$ $`1,`$ (5.23)
$`\left|{\displaystyle \frac{m_P^2}{V}}{\displaystyle \frac{^2V}{\varphi _{n_0}\varphi _{m_0}}}\right|`$ $``$ $`1,`$ (5.24)
$`3H_{\mathrm{inf}}\dot{\varphi }_{n_0}`$ $``$ $`{\displaystyle \frac{V}{\varphi _{n_0}}}.`$ (5.25)
The procedure of choosing a fixed basis is quite convenient for calculations, but a different procedure leads to a perhaps simpler theoretical description. This is to take $`\phi `$ to parameterise the distance along the inflaton trajectories, just as in single-component inflation, but now with the proviso that uniform $`\phi `$ corresponds to uniform field potential (since we work in the slow-roll approximation, this means that the slices in field space of uniform $`\phi `$ are orthogonal to the trajectories). Then, in the slow-roll approximation, slices of spacetime with uniform $`\phi `$ will again coincide with slices of uniform density (see Fig. 5.1a). Since $`\phi `$ is a scalar, Eqs. (5.20) and (5.21) will then be valid. This is the simplest form of the gauge-invariant expression, though for a practical calculation it may be better to write it in terms of a fixed basis.
There is a subtlety here. For the first-order case we could define $`\phi `$ in a different way; around a given point on the unperturbed trajectory we could choose a fixed field basis, with one of the basis vectors pointing along the trajectory, and define $`\phi `$ as the corresponding field component. Then we could choose $`\phi `$ to be canonically normalised in the vicinity of the chosen point in field space. That would not work at second order though, because at that order it makes a difference whether $`\phi `$ is the appropriate parameterisation of the distance along the trajectories (our adopted definition) or the distance along a tangent vector to the trajectory (the alternative definition) (see Fig. 5.1b). Only our adopted one will make Eqs. (5.21) and (5.22) valid.
#### 5.3.3 A third definition of the curvature perturbation
The third definition in the literature applies only during inflation. It was given originally by Acquaviva et. al. for the single-component case, and the generalization to the multi-component case was noted by Rigopoulos . We shall denote this definition by $`\zeta ^\mathrm{A}`$.
The definition of Acquaviva et. al. and Rigopoulos is
$$\zeta _2^\mathrm{A}=\psi _2^{\mathrm{MW}}_{\mathrm{inf}}\frac{\delta \phi _2}{\phi _0^{}}\frac{(\psi _1^{}+2_{\mathrm{inf}}\psi _1+_{\mathrm{inf}}\delta \phi _1^{}/\phi _0^{})^2}{_{\mathrm{inf}}^{}+2_{\mathrm{inf}}^2_{\mathrm{inf}}\phi _0^{\prime \prime }/\phi _0^{}}.$$
(5.26)
This is gauge-invariant by construction, with $`\phi `$ defined as in Figure 1(a).
It was pointed out by Vernizzi (actually in the context of single-component inflation) that comparing this definition with Eq. (5.21) gives simply
$$\zeta _2^\mathrm{A}=\zeta _2^{\mathrm{MW}}\frac{4_{\mathrm{inf}}^2(\zeta _1)^2}{_{\mathrm{inf}}^{}+2_{\mathrm{inf}}^2_{\mathrm{inf}}\phi _0^{\prime \prime }/\phi _0^{}}.$$
(5.27)
In the limit of slow-roll the denominator of the last term becomes just $`2_{\mathrm{inf}}^2`$, and then
$`\zeta _2^\mathrm{A}`$ $`=`$ $`\zeta _2.`$ (5.28)
In other words, this third definition coincides with our adopted one in the slow-roll limit.
Making use of the slow-roll parameters in Eqs. (5.15) and (5.16), the expression in Eq. (5.27) gives to first-order in the slow-roll approximation
$$\zeta _2^\mathrm{A}=\zeta _2(2\epsilon \eta _\phi )(\zeta _1)^2.$$
(5.29)
### 5.4 The evolution of the curvature perturbation
The simplest possibility for the origin of the observed curvature perturbation is that it comes from the vacuum fluctuation of the inflaton field in a single-component model (see Section 2.5). More recently other possibilities were recognised and we summarise the situation now. Although the discussion is usually applied to the magnitude of the curvature perturbation, it applies equally to the non-gaussianity.
#### 5.4.1 Heavy, light and ultra-light fields
On each scale the initial epoch, as far as classical perturbations are concerned, should be taken to be a few Hubble times after horizon exit during inflation. The reason is that all such perturbations are supposed to originate from the vacuum fluctuation of one or more light scalar fields, the fluctuation on each scale being promoted to a classical perturbation around the time of horizon exit .
Considering a fixed basis with canonical normalisation, a light field is roughly speaking one satisfying the flatness condition in Eq. (5.24). The terminology is suggested by the important special case that the effective potential during inflation is quadratic. Then, a light field is roughly speaking that whose effective mass during inflation is less than the value $`H_{}`$ of the Hubble parameter. More precisely, the condition that the vacuum fluctuation be promoted to a classical perturbation is
$$m<\frac{3}{2}H_{}.$$
(5.30)
From now on we focus on the quadratic potential, and take this as the the definition of a light field. Conversely a heavy field may be defined as one for which the condition in Eq. (5.30) is violated.
During inflation light fields slowly roll according to Eq. (5.25) (with the vacuum fluctuation superimposed) while the heavy fields presumably are pinned down at an instantaneous minimum of the effective potential. As we have seen, multi-component inflation takes place in a subspace of field space. The fields in this subspace are light, but their effective masses are sufficient to appreciably curve the inflationary trajectories. In the case of both multi-component and single-component inflation, there could also be ‘ultra-light’ fields, which do not appreciably curve the inflationary trajectory and which therefore have practically no effect on the dynamics of inflation.
#### 5.4.2 The evolution of the curvature perturbation
To describe the behaviour of perturbations during the super-horizon era, without making too many detailed assumptions, it is useful to invoke the separate universe hypothesis after smoothing on a given comoving scale much bigger than the horizon<sup>5</sup><sup>5</sup>5 When considering linear equations, smoothing is equivalent to dropping short wavelengths fourier components. In the nonlinear situation the smoothing procedure could be in principle ambiguous. In a given situation one should state explicitly which quantities are being smoothed.. According to this hypothesis the local evolution at each position is that of some unperturbed universe (separate universe). Of course the separate universe hypothesis can and should be checked where there is a sufficiently detailed model. However, it should be correct on cosmological scales for a very simple reason. The unperturbed Universe may be defined as the one around us, smoothed on a scale a bit bigger than the present Hubble distance. In other words, the separate universe hypothesis is certainly valid when applied to that scale. But the whole range of cosmological scales spans only a few orders of magnitude. This means that cosmological scales are likely to be huge compared with any scale that is relevant in the early Universe, and accordingly that the separate universe hypothesis should be valid when applied to cosmological scales even though it might fail on much smaller scales (this expectation was verified in a preheating example to which we return later).
We are concerned with the curvature perturbation, which during the super-horizon era is conserved as long as the pressure is a unique function of the energy density (the adiabatic pressure condition) (see Subsection 2.2.3). The adiabatic pressure condition will be satisfied if and only if the separate universes are identical (at least as far as the relation between pressure and energy density is concerned) <sup>6</sup><sup>6</sup>6Of course the identity will only hold after making an appropriate synchronization of the clocks at different positions. Having made that synchronization, horizon entry will occur at different times in different positions, which can be regarded as the origin of the curvature perturbation.. The condition to have identical universes after a given epoch is that the specification of a single quantity at that epoch is sufficient to determine the entire subsequent evolution.
In the case of single-component inflation, the initial condition may be supplied by the local value of the inflaton field, at the very beginning of the super-horizon era when it first becomes classical. Given the separate universe hypothesis, that is the only possibility if the inflaton is the only light field ever to play a significant dynamical role. This means that the curvature perturbation generated at horizon exit during single-component inflation will be equal to the one observed at the approach of horizon entry, provided that the inflaton is the only light field ever to play a dynamical role.
If inflation is multi-component, more than one field is by definition relevant during inflation. Then the curvature perturbation cannot be conserved during inflation. The variation of the curvature perturbation during multi-component inflation is caused by the vacuum fluctuation orthogonal to the unperturbed inflationary trajectory, which around the time of horizon exit kicks the trajectory onto a nearby one so that the local trajectory becomes position-dependent. After inflation is over, the curvature perturbation will be conserved if the local trajectories lead to practically identical universes. In other words it will be conserved if the light (and ultra-light) fields, orthogonal to the trajectory at the end of inflation, do not affect the subsequent evolution of the Universe.
The curvature perturbation after inflation will vary if some light or ultra-light field, orthogonal to the trajectory at the end of inflation, affects the subsequent evolution of the Universe (to be precise, affects the pressure) . As we shall describe in Section 5.7, three types of scenario have been proposed for this post-inflationary variation of the curvature perturbation.
### 5.5 Non-gaussianity
#### 5.5.1 Defining the non-gaussianity
A gaussian perturbation is one whose Fourier components are uncorrelated . All of its statistical properties are defined by its spectrum, and the spectrum $`𝒫_g(k)A_g^2(k/aH_{\mathrm{inf}})^{n_g}`$ of generic perturbation is conveniently defined by<sup>7</sup><sup>7</sup>7Technically the expectation values in this and the following expressions refer to an ensemble of universes but, because the stochastic properties of the perturbations are supposed to be invariant under translations, the expectation values can also be regarded as averages over the location of the observer who defines the origin of coordinates.
$$g_{𝐤_1}g_{𝐤_2}=\frac{2\pi ^2}{k^3}\delta ^3(𝐤_1+𝐤_2)𝒫_g(k),$$
(5.31)
the normalisation being chosen so that
$$g^2(𝐱)=_0^{\mathrm{}}𝒫_g(k)\frac{dk}{k}.$$
(5.32)
On cosmological scales a few Hubble times before horizon entry, observation shows that the curvature perturbation is almost Gaussian with $`|A_\zeta |5\times 10^5`$ .
The simplest kind of non-gaussianity that the curvature perturbation could possess is of the form
$`\zeta (𝐱)=\zeta _\mathrm{g}(𝐱){\displaystyle \frac{3}{5}}f_{\mathrm{NL}}\left(\zeta _\mathrm{g}^2(𝐱)\zeta _\mathrm{g}^2\right),`$ (5.33)
where $`\zeta _\mathrm{g}`$ is Gaussian with $`\zeta _\mathrm{g}=0`$, and the non-linearity parameter $`f_{\mathrm{NL}}`$ is independent of position. We will call this correlated $`\chi ^2`$ non-gaussianity. Note that this definition assumes that $`\zeta =0`$, which means that the zero Fourier mode (spatial average) is dropped.
Following Maldacena , we have inserted the prefactor $`(3/5)`$ so that in first-order perturbation theory our definition agrees with that of Komatsu and Spergel , which is generally the definition people use when comparing theory with observation. Working in first-order perturbation theory, these authors write $`\mathrm{\Phi }(𝐱)=\mathrm{\Phi }_g(𝐱)+f_{\mathrm{NL}}\left(\mathrm{\Phi }_g^2(𝐱)\mathrm{\Phi }_g^2\right)`$, and their $`\mathrm{\Phi }`$ is equal to $`3/5`$ times our $`\zeta `$ <sup>8</sup><sup>8</sup>8The actual quantity constrained by observational data is $`f_{\mathrm{NL}}^T`$, which is the non-linearity parameter for the CMB temperature anisotropies:
$$\frac{\delta T}{T}(𝐱)=\left(\frac{\delta T}{T}\right)_g(𝐱)+f_{\mathrm{NL}}^T\left[\left(\frac{\delta T}{T}\right)_g^2(𝐱)\left(\frac{\delta T}{T}\right)_g^2\right].$$
(5.34) At first order $`f_{\mathrm{NL}}^T=3f_{\mathrm{NL}}`$ because $`\delta T/T_0=(1/5)\zeta `$ \[c.f. Eqs. (1.1) and (1.2)\]. However, to compare adequately the observational data with our $`f_{\mathrm{NL}}`$, we must calculate $`f_{\mathrm{NL}}^T`$ in terms of $`f_{\mathrm{NL}}`$ at second order (see e.g. Refs. ). .
One of the most powerful observational signatures of non-gaussianity is a nonzero value for the three-point correlator, specified by the bispectrum $`B`$ defined by
$$\zeta _{𝐤_1}\zeta _{𝐤_2}\zeta _{𝐤_3}=(2\pi )^{3/2}B(k_1,k_2,k_3)\delta ^3(𝐤_1+𝐤_2+𝐤_3).$$
(5.35)
For correlated $`\chi ^2`$ non-gaussianity (with the gaussian term dominating)
$$B(k_1,k_2,k_3)=\frac{6}{5}f_{\mathrm{NL}}\left[P_\zeta (k_1)P_\zeta (k_2)+\mathrm{cyclic}\mathrm{permutations}\right],$$
(5.36)
where $`P_\zeta (k)=2\pi ^2𝒫_\zeta (k)/k^3`$. For any kind of non-Gaussianity one may use the above expression to define a function $`f_{\mathrm{NL}}(k_1,k_2,k_3)`$.
Given a calculation of $`f_{\mathrm{NL}}`$ using first-order perturbation theory, one expects in general that going to second order will change $`f_{\mathrm{NL}}`$ by an amount of order 1. On this basis, one expects that a first-order calculation is good enough if it yields $`|f_{\mathrm{NL}}|1`$, but that otherwise a second-order calculation will be necessary.
The definition Eq. (5.36) of $`f_{\mathrm{NL}}`$ is made using our adopted definition of $`\zeta `$. If $`\zeta `$ in the definition is replaced by $`\zeta ^{\mathrm{MW}}`$ (with the zero Fourier mode dropped) then $`f_{\mathrm{NL}}`$ should be replaced by
$$f_{\mathrm{NL}}^{\mathrm{MW}}f_{\mathrm{NL}}\frac{5}{3}.$$
(5.37)
To obtain this expression we used Eq. (5.7) and dropped terms higher than second order<sup>9</sup><sup>9</sup>9Obviously the parameter $`f_{\mathrm{NL}}^T`$, which is the important one to make comparison with observational data, does not depend on the chosen definition for $`f_{\mathrm{NL}}`$. .
All of this assumes that the non-gaussian component of $`\zeta `$ is fully correlated with the gaussian component. An alternative possibility that will be important for us is if $`\zeta `$ has the form
$`\zeta (𝐱)=\zeta _g(𝐱){\displaystyle \frac{3}{5}}\stackrel{~}{f}_{\mathrm{NL}}\left(\zeta _\vartheta ^2(𝐱)\zeta _\vartheta ^2\right),`$ (5.38)
where $`\zeta _g`$ and $`\zeta _\vartheta `$ are uncorrelated Gaussian perturbations, normalised to have equal spectra, and the parameter $`\stackrel{~}{f}_{\mathrm{NL}}`$ is independent of position. We will call this uncorrelated $`\chi ^2`$ non-gaussianity. It can be shown that in this case, $`f_{\mathrm{NL}}`$ as defined by Eq. (5.36) is given by
$$f_{\mathrm{NL}}\left(\frac{\stackrel{~}{f}_{\mathrm{NL}}}{653}\right)^3.$$
(5.39)
#### 5.5.2 Observational constraints on the non-gaussianity
Taking $`f_{\mathrm{NL}}`$ to denote the non-linearity parameter at the primordial era, let us consider the observational constraints. Detailed calculations have so far been made only with $`f_{\mathrm{NL}}`$ independent of the wavenumbers, and only by using first-order perturbation theory for the evolution of the cosmological perturbations after horizon entry. It is found that present observation requires $`|f_{\mathrm{NL}}|\text{ }\stackrel{<}{}\text{ }10^2`$ making the non-gaussian fraction at most of order $`10^3`$. The use of first-order perturbation theory in this context is amply justified. Looking to the future though, it is found that the PLANCK satellite will either detect non-gaussianity or reduce the bound to $`|f_{\mathrm{NL}}|\text{ }\stackrel{<}{}\text{ }5`$ , and that foreseeable future observations can reach a level $`|f_{\mathrm{NL}}|3`$ .
Although the use of first-order perturbation theory is not really justified for the latter estimates, we can safely conclude that it will be difficult for observation ever to detect a value $`|f_{\mathrm{NL}}|1`$. That is a pity because, as we shall see, such a value is predicted by some theoretical scenarios. On the other hand, other scenarios predict $`|f_{\mathrm{NL}}|`$ roughly of order 1. It will therefore be of great interest to have detailed second-order calculations, to establish precisely the level of sensitivity that can be achieved by future observations. A step in this direction has been taken in Ref. (see also Refs. ), where a non-linear expression for the large-scale CMB anisotropy is given in terms of only the curvature perturbation (generalizing the first-order Sachs-Wolfe effect ).
### 5.6 The initial non-gaussianity
#### 5.6.1 Single-component inflation
At first order, the curvature perturbation during single-component inflation is Gaussian. The amplitude of its time-independent spectrum is given by \[c.f. Eq. (2.105)\]
$$A_\zeta =\frac{H_{}^2}{2\pi \dot{\phi }_0},$$
(5.40)
and its spectral index $`n_\zeta d\mathrm{ln}𝒫_\zeta (k)/d\mathrm{ln}k`$ is given by
$$n_\zeta =2\eta _\phi 6\epsilon .$$
(5.41)
The squared amplitude of the spectrum $`r_{T\zeta }`$ of the tensor perturbation, defined as a fraction of $`A_\zeta ^2`$, is also given in terms of the slow-roll parameter $`\epsilon `$ \[c.f. Eq. (2.129)\]:
$$r_{T\zeta }=16\epsilon .$$
(5.42)
If the curvature perturbation does not evolve after single-component inflation is over observation constrains $`n_\zeta `$ and $`r_{T\zeta }`$, and hence the slow-roll parameters $`\eta _\phi `$ and $`\epsilon `$. A current bound is $`0.048<n_\zeta <0.016`$ and $`r_{T\zeta }<0.46`$. The second bound gives $`\epsilon <0.029`$, but barring an accurate cancellation the first bound gives $`\epsilon \text{ }\stackrel{<}{}\text{ }0.003`$. In most inflation models $`\epsilon `$ is completely negligible and then the first bound gives $`0.024<\eta _\phi <0.008`$ (irrespective of slow-roll inflation models, the upper bound in this expression holds generally, and the lower bound is badly violated only if there is an accurate cancellation). The bottom line of all this is that $`\epsilon `$ and $`|\eta _\phi |`$ are both constrained to be $`\text{ }\stackrel{<}{}\text{ }10^2`$.
Going to second order, Maldacena has calculated the bispectrum during single-component inflation (see also Refs. ) <sup>10</sup><sup>10</sup>10In Ref. (see also Ref. ) Rigopoulos et. al. calculated the three-point correlator in single-component slow-roll inflation using a stochastic approach. Their result agrees with Maldacena’s in the squeezed limit (where one of the scales $`k^1`$ crosses the horizon much earlier than the other two, $`k_1k_2,k_3`$), but disagrees in the limit where the $`\stackrel{}{k}_i`$’s form an equilateral triangle. Calcagni in Ref. extended this stochastic approach to calculate the non-gaussianity originated from a Dirac-Born-Infeld tachyonic inflaton and in braneworld scenarios, finding results identical to Maldacena’s one. The three-point correlators calculated for both cases were found identical to that calculated in Ref. .. His result may be written in the form
$$f_{\mathrm{NL}}=\frac{5}{12}\left[2\eta _\phi 6\epsilon 2\epsilon f(k_1,k_2,k_3)\right],$$
(5.43)
with $`0f5/6`$. By virtue of the slow-roll conditions, $`|f_{\mathrm{NL}}|1`$ <sup>11</sup><sup>11</sup>11Near a maximum of the potential ‘fast-roll’ inflation can take place with $`|\eta _\phi |`$ somewhat bigger than 1. Maldacena’s calculation does not apply to that case but, presumably, it gives initial non-gaussianity $`|f_{\mathrm{NL}}|1`$. Although the corresponding initial spectral index is far from $`1`$, which means that the initial curvature perturbation produced by $`\phi `$ must be negligible, the precise initial value of $`f_{\mathrm{NL}}`$ may in this case be important as long as another field (like the curvaton) be in charge of generating the observed curvature perturbation.. In other words, the curvature perturbation $`\zeta `$, as we have defined it, is almost Gaussian during single-component inflation.
From Eq. (5.29) $`\zeta ^\mathrm{A}`$ is also practically gaussian, but this quantity is defined only during inflation and therefore could not be considered as a replacement for $`\zeta `$. More importantly, $`\zeta ^{\mathrm{MW}}`$ has significant non-gaussianity because, from Eq. (5.37), it corresponds to $`f_{\mathrm{NL}}^{\mathrm{MW}}5/3`$.
One may ask why it is our $`\zeta `$ and not $`\zeta ^{\mathrm{MW}}`$ which is gaussian in the slow-roll limit. One feature that distinguishes our $`\zeta `$, is that any part of it can be absorbed into the scale factor without altering the rest; indeed
$$g_{ij}=\delta _{ij}a^2(\eta )e^{2\zeta _1+\zeta _2}=\delta _{ij}\stackrel{~}{a}^2(\eta )e^{\zeta _2},$$
(5.44)
with $`\stackrel{~}{a}=ae^{\zeta _1}`$ (if we tried to do that with $`\zeta _{\mathrm{MW}}`$, the part of $`\zeta `$ not absorbed would have to be re-scaled). This means that an extremely long-wavelength and possible large part of $`\zeta `$ has no local significance. It also means, in the context of perturbation theory, that the first-order part of $`\zeta `$ can be absorbed into the scale factor when discussing the second-order part. However, the gaussianity of $`\zeta `$ does not seem to be related directly to this feature. Rather, it has to do with the gauge transformation, relating quantities $`\psi _A`$ and $`\psi _B`$ defined on different slicings.
With our definition , the gauge transformation is
$$\psi _A(t,𝐱)\psi _B(t,𝐱)=\mathrm{\Delta }N_{AB}(t,𝐱),$$
(5.45)
where $`\mathrm{\Delta }N_{AB}`$ is the number of $`e`$-folds of expansion going from a slice $`B`$ to a slice $`A`$, both of them corresponding to time $`t`$ <sup>12</sup><sup>12</sup>12This expression is valid even when the tensor perturbation is included . As a result, the gauge-invariant expressions mentioned earlier are still valid in that case, as are the results based on them including the present discussion. . In writing this expression we used physical time $`t`$ instead of conformal time, the two related by $`dt=ad\eta `$. Along a comoving worldline, the number of $`e`$-folds of expansion is defined as $`N\stackrel{~}{H}𝑑\tau `$ where $`\stackrel{~}{H}`$ is the local Hubble parameter and $`d\tau `$ is the proper time interval .
To understand the relevance of this result, take $`\psi _B=0`$ and $`\psi _A=\zeta `$. The pressure is adiabatic during single-component inflation, which means that $`dt`$ can be identified with the proper time interval $`d\tau `$, and the proper expansion rate on slicing $`A`$ is uniform . As a result, to second order,
$`\zeta `$ $`=`$ $`H_{\mathrm{inf}}(t)\mathrm{\Delta }t(t,𝐱)+{\displaystyle \frac{1}{2}}\dot{H}_{\mathrm{inf}}(t)\left(\mathrm{\Delta }t(t,𝐱)\right)^2`$ (5.46)
$``$ $`H_{\mathrm{inf}}\mathrm{\Delta }t(t,𝐱)+{\displaystyle \frac{1}{2}}{\displaystyle \frac{\dot{H}_{\mathrm{inf}}}{H_{\mathrm{inf}}^2}}\left(H_{\mathrm{inf}}\mathrm{\Delta }t(t,𝐱)\right)^2`$
$``$ $`H_{\mathrm{inf}}\mathrm{\Delta }t(t,𝐱).`$
In the last line we made the slow-roll approximation, and from the second line we can see that the error in $`f_{\mathrm{NL}}`$ caused by this approximation is precisely $`\epsilon `$.
We also need the gauge transformation for the inflaton field $`\phi `$ in terms of $`\mathrm{\Delta }t`$. Since the slices correspond to the same coordinate time, the unperturbed inflaton field can be taken to be the same on each of them which means that the gauge transformation for $`\delta \phi `$ is
$$\delta \phi _A(t,𝐱)\delta \phi _B(t,𝐱)=\mathrm{\Delta }\phi _{AB}(t,𝐱),$$
(5.47)
where $`\mathrm{\Delta }\phi _{AB}`$ is the change in $`\phi `$ going from slice $`B`$ to slice $`A`$. But slice $`A`$ corresponds to uniform $`\phi `$, which means that on slice $`B`$ to second order
$`H_{\mathrm{inf}}(t){\displaystyle \frac{\delta \phi _B(t,𝐱)}{\dot{\phi }_0}}`$ $`=`$ $`H_{\mathrm{inf}}(t)\mathrm{\Delta }t(t,𝐱){\displaystyle \frac{1}{2}}H_{\mathrm{inf}}(t){\displaystyle \frac{\ddot{\phi }_0}{\dot{\phi }_0}}(\mathrm{\Delta }t(t,𝐱))^2`$ (5.48)
$``$ $`H_{\mathrm{inf}}\mathrm{\Delta }t(t,𝐱){\displaystyle \frac{1}{2}}{\displaystyle \frac{\ddot{\phi }_0}{H_{\mathrm{inf}}\dot{\phi }_0}}(H_{\mathrm{inf}}\mathrm{\Delta }t(t,𝐱))^2`$
$``$ $`H_{\mathrm{inf}}\mathrm{\Delta }t(t,𝐱),`$
where in the last line we used the slow-roll approximation. We can see that the fractional error caused by this approximation is $`\ddot{\phi }_0/H_{\mathrm{inf}}\dot{\phi }_0=\epsilon \eta _\phi `$.
Combining Eqs. (5.46) and (5.48) we have in the slow-roll approximation
$$\zeta H_{\mathrm{inf}}(t)\frac{\delta \phi _B(t,𝐱)}{\dot{\phi }_0},$$
(5.49)
with fractional error of order $`\mathrm{max}\{\eta _\phi ,\epsilon \}`$ (this can also be seen directly from Eqs. (5.20) and (5.22) evaluated with $`\psi =0`$, but we give the above argument because it explains why the result is valid for $`\zeta `$ as opposed to $`\zeta ^{\mathrm{MW}}`$).
The final and crucial step is to observe that in the slow-roll approximation $`\phi _B`$ is gaussian, with again a fractional error of order $`\mathrm{max}\{\eta _\phi ,\epsilon \}`$. This was demonstrated by Maldacena but the basic reason is very simple. The non-gaussianity of $`\phi `$ comes either from third and higher derivatives of $`V`$ (through the field equation in unperturbed spacetime) or else through the back-reaction (the perturbation of spacetime); but the first effect is small by virtue of the flatness requirements on the potential, and the second effect is small because $`\dot{\phi }_0/H_{\mathrm{inf}}^2`$ is small . This explains why $`\zeta `$ with our adopted definition is practically Gaussian by virtue of the slow-roll approximation.
#### 5.6.2 Multi-component inflation
The flatness and slow-roll conditions Eqs. (5.23), (5.24), and (5.25) ensure that the curvature of the inflationary trajectories is small during the few Hubble times around horizon exit, during which the quantum fluctuation is promoted to a classical perturbation. As a result, the initial curvature perturbation in first-order perturbation theory is still given by the amplitude in Eq. (5.40) and the spectral index in Eq. (5.41) in terms of the field $`\phi `$ that we defined earlier.
What about the initial non-gaussianity generated at second order? In the approximation that the curvature of the trajectories around horizon exit is completely negligible, we can safely say that the initial non-gaussianity corresponds to $`|f_{\mathrm{NL}}|1`$. Confirming this expectation, Seery and Lidsey have calculated the three-point correlator of the perturbations in the fields involved in multi-component slow-roll inflation. Their result is given by
$$\delta \varphi _{𝐤_1}^i\delta \varphi _{𝐤_2}^j\delta \varphi _{𝐤_3}^k\left(\frac{H_{}}{2\pi }\right)^3(2\pi )^{3/2}B_{ijk}(k_1,k_2,k_3)\delta ^3(𝐤_1+𝐤_2+𝐤_3),$$
(5.50)
with
$$B_{ijk}(k_1,k_2,k_3)\frac{6}{5}f_{ijk}\left[\frac{4\pi ^4}{k_1^3k_2^3}+\mathrm{cyclic}\mathrm{permutations}\right],$$
(5.51)
and<sup>13</sup><sup>13</sup>13The cyclic permutations in $`k`$ and $`\varphi `$ in Eq. (5.52) must be simultaneous, i.e. when exchanging indices $`i`$ and $`j`$, for example, $`k_1`$ and $`k_2`$ must also be exchanged. Notice also that the calculation of Seery and Lidsey’s is only valid when the magnitudes of the three wavevectors are roughly comparable, so that they exit the horizon at similar epochs.
$$f_{ijk}=\frac{5}{12}\left[\frac{\dot{\varphi }_{}^i}{2\pi m_P^2}\delta _{jk}f_{SL}(k_1,k_2,k_3)+\mathrm{cyclic}\mathrm{permutations}\mathrm{in}k\mathrm{and}\varphi \right].$$
(5.52)
As we will discuss in Chapter 6, where the $`\delta N`$ formalism is used to calculate the stochastic properties of $`\zeta `$ , the contribution $`\mathrm{\Delta }f_{\mathrm{NL}}`$ of the wavevector dependent parameter $`f_{ijk}`$ to the total $`f_{\mathrm{NL}}`$ is in any case very small, being $`\mathrm{\Delta }f_{\mathrm{NL}}`$ generically below $`(15/24)f_{SL}\sqrt{r_{T\zeta }\epsilon }\text{ }\stackrel{<}{}\text{ }10^2`$ where $`f_{SL}`$ is in the range $`1/3f_{SL}11/18`$.
### 5.7 The evolution after horizon exit
#### 5.7.1 Single-component inflation and $`\zeta _2^\mathrm{A}`$
During single-component inflation the curvature perturbation $`\zeta `$, as we have defined it, does not evolve. From its definition Eq. (5.7), the same is true of $`\zeta ^{\mathrm{MW}}`$.
In contrast $`\zeta _2^\mathrm{A}`$, given by Eq. (5.29), will have the slow variation
$$\dot{\zeta }_2^\mathrm{A}(2\dot{\epsilon }\dot{\eta }_\phi )(\zeta _1)^2.$$
(5.53)
This variation has no physical significance, being an artifact of the definition.
Using a particular gauge, Acquaviva et. al. have calculated $`\dot{\zeta }_2^\mathrm{A}`$ in terms of first-order quantities $`\psi _1`$, $`\delta \phi _1`$, and their derivatives, and they have displayed the result as an indefinite integral
$$\zeta _2^\mathrm{A}(t)=^tA(t)𝑑t+B(t).$$
(5.54)
Inserting an initial condition, valid a few Hubble times after horizon exit, this becomes
$$\zeta _2^\mathrm{A}(t)=\zeta _2^\mathrm{A}(t_{\mathrm{ini}})+_{t_{\mathrm{ini}}}^tA(t)𝑑t+B|_{t_{\mathrm{ini}}}^t.$$
(5.55)
In view of our discussion, it is clear that these equations will, if correctly evaluated, just reproduce the time dependence of Eq. (5.53).
The authors of Ref. also present the respective equation for $`\dot{\zeta }_2^\mathrm{A}`$, involving only first-order quantities, which is valid also before horizon entry. Contrary to the claim of the authors, this classical equation cannot by itself be used to calculate the initial value (more precisely, the stochastic properties of the initial value) of $`\zeta _2^\mathrm{A}`$. In particular, it cannot by itself reproduce Maldacena’s calculation of the bispectrum.
It is true of course that in the Heisenberg picture the quantum operators satisfy the classical field equations. In first-order perturbation theory, where the equations are linear, this allows one to calculate the curvature perturbation without going to the trouble of calculating the second-order action (at the $`n`$th order of perturbation theory the action has to be evaluated to order $`n+1`$ if it is to be used). At second order in perturbation theory it remains to be seen whether the Heisenberg picture can provide a useful alternative to Maldacena’s calculation, who adopted the interaction picture and calculated the action to third order.
#### 5.7.2 Multi-component inflation
During multi-component inflation the curvature perturbation by definition varies significantly along a generic trajectory, which means that non-gaussianity is generated at some level. So far only a limited range of models has been investigated . To keep the spectral tilt within observational bounds, the unperturbed trajectory in these models has to be specially chosen, but the choice might be justified by a suitable initial condition.
We shall consider here a calculation by Enqvist and V$`\ddot{\mathrm{a}}`$ihk$`\ddot{\mathrm{o}}`$nen in Refs. . Following the same line as Acquaviva et. al. , they study a two-component inflation model, in which the only important parts of the potential are
$$V(\phi ,\vartheta )=V_h+\frac{1}{2}m_\vartheta ^2\vartheta ^2+\frac{1}{2}m_\phi ^2\phi ^2.$$
(5.56)
The masses are both supposed to be less than $`(3/2)H_{}`$, so that this is a two-component inflation model, and the above form of the potential is supposed to hold for some number $`\mathrm{\Delta }N`$ of $`e`$-folds after cosmological scales leave the horizon. They take the unperturbed inflation trajectory to have $`\vartheta _0=0`$, and the idea is to calculate the amount of non-gaussianity generated after $`\mathrm{\Delta }N`$ $`e`$-folds. Irrespective of any later evolution, this calculated non-gaussianity will represent the minimal observed one (unless non-gaussianity generated later happens to cancel it).
It is supposed that the condition $`\vartheta _0=0`$, as well as the ending of inflation, will come from a tree-level hybrid potential,
$$V(\phi ,\vartheta )=V_h\frac{1}{2}m_\vartheta ^2\vartheta ^2+\frac{1}{4}\lambda \vartheta ^4+\frac{1}{2}m_\phi ^2\phi ^2+\frac{1}{2}g^2\vartheta ^2\phi ^2.$$
(5.57)
Like the original authors though, we shall not investigate the extent to which Eq. (5.57) can reproduce Eq. (5.56) for at least some number of $`e`$-folds. We just focus on Eq. (5.56), with the assumption $`\vartheta _0=0`$ for the unperturbed trajectory.
Because $`\vartheta _0=0`$, the unperturbed trajectory is straight, and at first order the curvature perturbation $`\zeta `$ is conserved. This is not the case though at second order. Adopting the definition $`\zeta ^\mathrm{A}`$, the authors of Ref. give an expression for $`\zeta _2^\mathrm{A}`$ similar to that in Eq. (5.55) describing the evolution of the second-order curvature perturbation on superhorizon scales<sup>14</sup><sup>14</sup>14The fields $`\phi `$ and $`\vartheta `$ in Eq. (5.56) are supposed to be canonically normalised, which means that $`\phi `$ is not the field appearing in the Rigopoulos definition Eq. (5.26) of $`\zeta ^\mathrm{A}`$. Instead the authors of Ref. give an equivalent definition in terms of the canonically normalised fields.. This equation, in the generalized longitudinal gauge, reads (from Eq. (67) in Ref. ):
$`\zeta _2^\mathrm{A}(t)\zeta _2^\mathrm{A}(t_\mathrm{i})`$ $`=`$ $`{\displaystyle \frac{1}{\stackrel{~}{\epsilon }H_{inf}m_P^2}}\{{\displaystyle _{t_\mathrm{i}}^t}[6H_{\mathrm{inf}}^2_i(\delta \dot{\vartheta }_1^i\delta \vartheta _1)+4^2_i(\delta \dot{\vartheta }_1^i\delta \vartheta _1)^{}`$
$`2(\delta \dot{\vartheta }_1)^2+m_\vartheta ^2(\delta \vartheta _1)^2+(\stackrel{~}{\epsilon }\eta _\phi )6H_{\mathrm{inf}}^4_i(_k^k\delta \vartheta _1^i\delta \vartheta _1)^{}`$
$`+(\stackrel{~}{\epsilon }\eta _\phi )H_{\mathrm{inf}}^4_i^i(_k\delta \vartheta _1^k\delta \vartheta _1)^{}3^4_i(_k^k\delta \vartheta _1^i\delta \vartheta _1)^{}`$
$`{\displaystyle \frac{1}{2}}^4_i^i(_k\delta \vartheta _1^k\delta \vartheta _1)^{}]dt+[^2_i(\delta \dot{\vartheta }_1^i\delta \vartheta _1)`$
$`+3^4_i(_k^k\delta \vartheta _1^i\delta \vartheta _1)^{}+{\displaystyle \frac{1}{2}}^4_i^i(_k\delta \vartheta _1^k\delta \vartheta _1)^{}`$
$`+3\stackrel{~}{\epsilon }H_{\mathrm{inf}}^4_i(_k^k\delta \vartheta _1^i\delta \vartheta _1)+{\displaystyle \frac{\stackrel{~}{\epsilon }H_{\mathrm{inf}}}{2}}^4_i^i(_k\delta \vartheta _1^k\delta \vartheta _1)]|_{t_\mathrm{i}}^t\},`$
where $`^2`$ is the inverse of the Laplacian operator, $`\eta _\phi m_\phi ^2/3H_{}^2`$, and $`\stackrel{~}{\epsilon }`$ is defined by
$$\stackrel{~}{\epsilon }\frac{\dot{\phi }_0^2(t)}{2m_P^2H_{\mathrm{inf}}^2},$$
(5.59)
which reduces to the $`\epsilon `$ parameter in Eq. (5.17) for $`t=t_{}`$, being $`t_{}`$ the time when cosmological scales exit the horizon.
Assuming that this expression is correct, we consider the non-gaussianity it may generate. Reviewing what it was done in Ref. , we note first that at $`t=t_{}`$
$$\delta \vartheta _1(t_{})\delta \phi _1(t_{}),$$
(5.60)
which is a good approximation since at that time the amplitude of the spectrum of perturbations of any light field $`\varphi `$ is $`A_{\delta \varphi }H_{}/2\pi `$. Moreover, assuming slow-roll conditions we obtain
$`\delta \vartheta _1(t)`$ $`=`$ $`\delta \vartheta _1(t_{})e^{\eta _\vartheta N},`$ (5.61)
$`\phi _0(t)`$ $`=`$ $`\phi _0(t_{})e^{\eta _\phi N},`$ (5.62)
$`\stackrel{~}{\epsilon }`$ $`=`$ $`\epsilon e^{2\eta _\phi N},`$ (5.63)
where we have used $`N=_t_{}^tH_{\mathrm{inf}}𝑑t`$, $`\eta _\vartheta m_\vartheta ^2/3H_{}^2`$, and Eq. (5.59). A similar expression for the evolution of $`\delta \phi _1`$ is obtained by invoking the constancy of the first-order curvature perturbation $`\zeta _1`$:
$$\delta \phi _1(t)=\delta \phi _1(t_{})e^{\eta _\phi N}.$$
(5.64)
Assuming that $`H_{\mathrm{inf}}`$, $`\eta _\phi `$, and $`\eta _\vartheta `$ are almost constants in time, we end up with
$`\zeta _2^\mathrm{A}(t)\zeta _2^\mathrm{A}(t_{})`$ $`=`$ $`{\displaystyle \frac{1}{\stackrel{~}{\epsilon }H_{}m_P^2}}\{{\displaystyle _t_{}^t}[2^2_i(\delta \dot{\vartheta }_1^i\delta \vartheta _1)^{}+2H_{\mathrm{inf}}(\stackrel{~}{\epsilon }\eta _\phi )\dot{\gamma }_\vartheta `$ (5.65)
$`(\delta \dot{\vartheta }_1)^2\ddot{\gamma }_\vartheta ]dt+[^2_i(\delta \dot{\vartheta }_1^i\delta \vartheta _1)+\dot{\gamma }_\vartheta +\stackrel{~}{\epsilon }H_{\mathrm{inf}}\gamma _\vartheta ]|_t_{}^t\}`$
$`=`$ $`{\displaystyle \frac{1}{\stackrel{~}{\epsilon }H_{}m_P^2}}\{{\displaystyle _t_{}^t}[(\delta \dot{\vartheta }_1)^2+2H_{\mathrm{inf}}\stackrel{~}{\epsilon }\dot{\gamma }_\vartheta ]dt+[^2_i(\delta \dot{\vartheta }_1^i\delta \vartheta _1)`$
$`+H_{\mathrm{inf}}(\stackrel{~}{\epsilon }2\eta _\phi )\gamma _\vartheta ]|_t_{}^t\},`$
where
$$\gamma _\vartheta 3^4_i(_k^k\delta \vartheta _1^i\delta \vartheta _1)+\frac{1}{2}^2(_i\delta \vartheta _1^i\delta \vartheta _1),$$
(5.66)
and we have used the equation of motion $`\delta \ddot{\vartheta }_1+3H_{\mathrm{inf}}\delta \dot{\vartheta }_1+m_\vartheta ^2\delta \vartheta _1=0`$ to go from Eq. (LABEL:ourev) to Eq. (5.65).
The order of magnitude for $`\zeta _2^\mathrm{A}(t)\zeta _2^\mathrm{A}(t_{})`$ is now easily estimated by means of the expressions in Eqs. (5.61) to (5.64), and by neglecting the scale dependence of the non-local terms:
$`\zeta _2^\mathrm{A}(t)\zeta _2^\mathrm{A}(t_{})`$ $``$ $`{\displaystyle \frac{1}{\stackrel{~}{\epsilon }H_{}m_P^2}}\{{\displaystyle _t_{}^t}[\eta _\vartheta ^2H_{\mathrm{inf}}^2|\delta \vartheta _1|^2+\stackrel{~}{\epsilon }\eta _\vartheta H_{\mathrm{inf}}^2|\delta \vartheta _1|^2]dt`$ (5.67)
$`+[\eta _\vartheta H_{\mathrm{inf}}|\delta \vartheta _1|^2+\stackrel{~}{\epsilon }H_{\mathrm{inf}}|\delta \vartheta _1|^2+\eta _\phi H_{\mathrm{inf}}|\delta \vartheta _1|^2]|_t_{}^t\},`$
so that, using Eq. (5.60) to write $`\delta \vartheta _1`$ in terms of $`\zeta _1^\mathrm{A}`$,
$$\zeta _2^\mathrm{A}(t)\zeta _2^\mathrm{A}(t_{})𝒪(\epsilon ,\eta _\phi ,\eta _\vartheta )e^{2N(\eta _\phi \eta _\vartheta )}|\zeta _1^\mathrm{A}|^2.$$
(5.68)
It is unlikely that the exponential factor on the right hand side provides any significant enhancement to $`\zeta _2^\mathrm{A}`$ if $`\phi `$ produces most of the curvature perturbation. Therefore, the overall slow-roll factors give the actual magnitude. We have to remember that, in this case, the right hand side is uncorrelated with the inflaton perturbation $`\delta \varphi `$ which generates $`\zeta _1^\mathrm{A}`$. Thus, Eq. (5.38) as opposed to Eq. (5.33) applies, and the associated $`f_{\mathrm{NL}}`$ would be $`10^9𝒪(\epsilon ^3,\eta _\phi ^3,\eta _\vartheta ^3)`$, which is extremely small. If the observed $`\zeta `$ has a non-gaussian part $`\zeta _2^\mathrm{A}`$ equal to Eq. (5.68) and a gaussian part generated mostly after inflation, one can obtain $`|f_{\mathrm{NL}}|>1`$ by choosing $`\eta _\phi >0.26`$, $`\eta _\vartheta =\eta _\phi /2`$, $`N=70`$, and $`\zeta _2^\mathrm{A}=10^2\zeta `$.
As we will see in the next chapter, where the Enqvist and V$`\ddot{\mathrm{a}}`$ihk$`\ddot{\mathrm{o}}`$nen model is studied by means of the non perturbative $`\delta N`$ formalism (see Subsection 6.5.3), the expression in Eq. (5.68) disagrees with the one calculated using the $`\delta N`$ formalism through the appearance of non-local terms , though the order of magnitude is similar . We point out that the possible source of discrepancy is the use of a set of cosmological perturbation theory equations in Ref. based on those presented in Ref. . A calculation made with another set of cosmological perturbation theory equations that do not involve non-local terms reproduces exactly the result found using the $`\delta N`$ formalism (see Subsection 6.5.3).
#### 5.7.3 Preheating
Now we turn to the possibility that significant non-gaussianity could be generated during preheating. Preheating is a stage of non-perturbative explosive resonant decay of scalar fields which might occur between the end of inflation and reheating , the latter being taken to correspond to the decay of individual particles which leads to more or less complete thermalisation of the Universe. Preheating typically produces marginally-relativistic particles, which decay before reheating.
It was suggested a long time ago that preheating might cause the cosmological curvature perturbation to vary at the level of first-order perturbation theory, perhaps providing its main origin. More recently it has been suggested that preheating might cause the curvature perturbation to vary at second order, providing the main source of its non-gaussianity.
If the separate universe hypothesis is correct, a variation of the curvature perturbation during preheating can occur only in models of preheating which contain a non-inflaton field that is light during inflation. This is not the case for the usual preheating models that were considered in , and accordingly one does not expect that significant non-gaussianity will be generated in those models<sup>15</sup><sup>15</sup>15The preheating model considered in contains a field which may be heavy or light; we refer here to the part of the calculation that considers the former case.. This is not in conflict with the findings of because the curvature perturbation is not actually considered there. Instead the perturbation $`\psi ^{\mathrm{MW}}`$ in the longitudinal gauge is considered, which is only indirectly related to $`\zeta `$ by Eqs. (5.7), (5.11) and (5.12) <sup>16</sup><sup>16</sup>16The slices of the longitudinal gauge are orthogonal to the threads of zero shear, and $`\psi ^{\mathrm{MW}}`$ on them is very different from the curvature perturbation $`\zeta `$.. We conjecture that non-gaussianity for the curvature perturbation on cosmological scales is not generated in the usual preheating models, but that instead the curvature perturbation remains constant on cosmological scales. This should of course be checked, in the same spirit that the constancy of the curvature perturbation was checked at the first-order level .
The situation is different for preheating models which contain a non-inflaton field that is light during inflation. At least three types of models have been proposed with that feature . Except for only the magnitude of the curvature perturbation has been considered, but in all three cases it might be that significant non-gaussianity is also generated.
#### 5.7.4 The curvaton scenario
In the simplest version of the curvaton scenario (see Subsection 2.4.2), the curvaton field $`\sigma `$ is ultra-light during inflation, weakly coupled, and has no significant evolution until it starts to oscillate during some radiation-dominated era. Until this oscillation gets under way, the curvature perturbation is supposed to be negligible (compared with its final observed value). The potential during the oscillation is taken to be quadratic, which will be a good approximation after a few Hubble times even if it fails initially. The curvature perturbation is generated during the oscillation, and is supposed to be conserved after the curvaton decays. Here we give a generally-valid formula for the non-gaussianity in the curvaton scenario, extending somewhat the earlier calculations.
The local energy density $`\rho _\sigma `$ of the curvaton field is given by \[c.f. Eq. (2.53)\]
$$\rho _\sigma (\eta ,𝐱)\frac{1}{2}m_\sigma ^2\sigma _a^2(\eta ,𝐱),$$
(5.69)
where $`\sigma _a(\eta ,𝐱)`$ represents the amplitude of the oscillations and $`m_\sigma `$ is the effective mass. It is proportional to $`a(\eta ,𝐱)^3`$ where $`a`$ is the locally-defined scale factor. This means that the perturbation $`\delta \rho _\sigma /\rho _{\sigma _0}`$ is conserved if the slicing is chosen so that the expansion going from one slicing to the next is uniform . The flat slicing corresponding to $`\psi ^{\mathrm{MW}}=0`$ has this property and accordingly $`\delta \rho _\sigma `$ is defined on that slicing (see Subsection 2.4.2).
Assuming that the fractional perturbation is small (which we shall see is demanded by observation) it is given by
$$\frac{\delta \rho _\sigma }{\rho _{\sigma _0}}2\frac{\delta \sigma _a}{\sigma _{a_0}}+\left(\frac{\delta \sigma _a}{\sigma _{a_0}}\right)^2,$$
(5.70)
where we have extended to second order the Eq. (2.65). We first assume that $`\sigma (𝐱)`$ has no evolution between inflation and the onset of oscillation. Then $`\delta \sigma _a/\sigma _{a_0}`$ will be equal to its value just after horizon exit, which we saw earlier will be practically gaussian.
The total density perturbation is given by
$$\left(\frac{\delta \rho }{\rho _0}\right)_{\mathrm{total}}\mathrm{\Omega }\frac{\delta \rho _\sigma }{\rho _{\sigma _0}},$$
(5.71)
where $`\mathrm{\Omega }\rho _{\sigma _0}/\rho _{\mathrm{total}_0}a`$ is the fraction of energy density contributed by the curvaton. Adopting the sudden-decay approximation, the constant curvature perturbation obtaining after the curvaton decays is given by Eqs. (5.11) and (5.13), evaluated just before curvaton decay and with $`\psi =0`$. In performing that calculation, the exact expression Eq. (5.70) can, without loss of generality, be identified with the first-order part $`\delta \rho _{\sigma _1}/\rho _{\sigma _0}`$, the second- and higher-order parts being set at zero.
Adopting the first-order curvature perturbation in Eq. (5.11), one finds $`\chi ^2`$ non-gaussianity coming from the second term of Eq. (5.70),
$$f_{\mathrm{NL}}=\frac{5}{4r},$$
(5.72)
with
$$r\frac{3\rho _{\sigma _0}}{4\rho _{r_0}+3\rho _{\sigma _0}},$$
(5.73)
evaluated just before decay. Going to the second-order expression one finds additional $`\chi ^2`$ non-gaussianity. The final non-linearity parameter $`f_{\mathrm{NL}}=f_{\mathrm{NL}}^{\mathrm{MW}}+5/3`$ is given by
$$f_{NL}=\frac{5}{3}+\frac{5}{6}r\frac{5}{4r}.$$
(5.74)
If $`\mathrm{\Omega }`$ just before the curvaton decay ($`\mathrm{\Omega }_{\mathrm{dec}}`$) is much less than 1 ($`\mathrm{\Omega }_{\mathrm{dec}}1`$) then $`f_{\mathrm{NL}}`$ is strongly negative and the present bound on it requires $`\mathrm{\Omega }_{\mathrm{dec}}\text{ }\stackrel{>}{}\text{ }0.01`$ (combined with the typical value $`|A_\zeta |5\times 10^5`$, this requires $`\delta \rho _\sigma /\rho _{\sigma _0}1`$ as advertised). If instead $`\mathrm{\Omega }_{\mathrm{dec}}=1`$ to good accuracy, then $`f_{\mathrm{NL}}=+5/4`$. Either of these possibilities may be regarded as generic whereas the intermediate possibility ($`|f_{\mathrm{NL}}|1`$ but $`f_{\mathrm{NL}}5/4`$) requires a special value of $`\mathrm{\Omega }_{\mathrm{dec}}`$ just a bit less than $`1`$.
Finally, we consider the case that $`\sigma `$ evolves between horizon exit and the era when the sinusoidal oscillation begins. If $`\sigma _a`$ (the amplitude of oscillation at the latter era) is some function $`g(\sigma _{})`$ of the value a few Hubble times after horizon exit, then
$$\delta \sigma _a=g^{}\delta \sigma _{}+\frac{1}{2}g^{\prime \prime }(\delta \sigma _{})^2,$$
(5.75)
where the prime means derivative with respect to $`\sigma _{}`$. Repeating the above calculation one finds
$$f_{\mathrm{NL}}=\frac{5}{3}+\frac{5}{6}r\frac{5}{4r}\left(1+\frac{gg^{\prime \prime }}{g^2}\right).$$
(5.76)
The final term is the first-order result (given originally in ), the middle term is the second-order correction found in , and the first term converts from $`f_{\mathrm{NL}}^{\mathrm{MW}}`$ to $`f_{\mathrm{NL}}`$.
#### 5.7.5 The inhomogeneous reheating scenario
The final scenario that has been suggested for the origin of the curvature perturbation is its generation during some spatially inhomogeneous reheating process (see also Refs. ). Before a reheating process the cosmic fluid is dominated by matter (non-relativistic particles, or small scalar field oscillations which are equivalent to particles) which then decay into thermalised radiation. At least one reheating process, presumably, has to occur to give the initial condition for Big Bang Nucleosynthesis, but there might be more than one.
The inhomogeneous reheating scenario in its simplest form supposes that the curvature perturbation is negligible before the relevant reheating process, and constant afterwards. The inhomogeneous reheating corresponds to spatial fluctuations in the decay rate of the inflaton field to ordinary matter, which lead to fluctuations in the reheating temperature. The coupling of the inflaton to normal matter is determined by the vacuum expectation values of scalar fields in the theory. If those fields are light they will fluctuate leading to density perturbations through the described mechanism. Inhomogeneities in the inflaton decay rate lead to a a spatially varying value (a perturbation) of the local Hubble parameter $`H_{\mathrm{reh}}(𝐱)`$ at the decay epoch (or equivalently of the local energy density).
In contrast with the curvaton scenario, where the form $`\rho _\sigma `$ can reasonably be taken as $`\rho _\sigma \sigma ^2`$, the inhomogeneous reheating scenario does not suggest any particular form for $`H_{\mathrm{reh}}(\chi )`$. Depending on the form, the inhomogeneous reheating scenario presumably can produce a wide range of values for $`f_{\mathrm{NL}}`$ .
### 5.8 Conclusions
We have examined a number of scenarios for the production of a non-gaussian primordial curvature perturbation, presenting the results with a unified notation . These are the single-component inflation, multi-component inflation, preheating, curvaton, and inhomogeneous reheating scenarios. Although the trispectrum may give a competitive observational signal , we have focused only on the bispectrum which is characterised by the parameter $`f_{\mathrm{NL}}`$. In all cases our treatment is based on existing ones, though we do not always agree with the original authors.
The preheating and inhomogeneous reheating scenarios cover a range of possibilities, which have not been fully explored but which can presumably allow a wide range for $`f_{\mathrm{NL}}`$. The same is true of multi-component inflation, except that extremely large values comparable with the current bound $`|f_{\mathrm{NL}}|\text{ }\stackrel{<}{}\text{ }10^2`$ seem relatively unlikely. In contrast, the simplest curvaton scenario can produce a strongly negative value (even violating the current bound). However, in the important special case where the curvaton dominates the energy density before it decays, it gives precisely $`f_{\mathrm{NL}}=+5/4`$. Finally, for the single-component inflation case, Maldacena’s calculation combined with current constraints on the spectral tilt show that it has magnitude less than $`10^2`$. These result are summarised in the Table 5.1.
In the near future, results from WMAP or elsewhere may detect a value $`|f_{\mathrm{NL}}|1`$. If that does not happen, then PLANCK or a successor will either detect a value $`|f_{\mathrm{NL}}|1`$, or place a bound $`|f_{\mathrm{NL}}|\text{ }\stackrel{<}{}\text{ }1`$. The precise level at which this will be possible has yet to be determined because it would require a second-order calculation of all relevant observational signatures. The example of the simplest curvaton scenario, where $`f_{\mathrm{NL}}=+5/4`$ is a favoured value, shows that such a calculation and the eventual observations will be well worthwhile.
## Chapter 6 The inflationary prediction for primordial non-gaussianity
### 6.1 Introduction
In this chapter we present for the first time a powerful method to calculate the normalisation $`f_{\mathrm{NL}}`$ of the bispectrum in slow-roll inflation, by means of the knowledge of the evolution of a family of unperturbed universes . The wavevector dependence of $`f_{\mathrm{NL}}`$ will be, in general, negligible compared with the (possibly big) contribution coming from the evolution of the unperturbed universes. This method will be applied to selected examples. In particular we will see how the level of non-gaussianity in the curvaton scenario (see Subsection 5.7.4) and the second order curvature perturbation in the hybrid model of Enqvist and V$`\ddot{\mathrm{a}}`$ihk$`\ddot{\mathrm{o}}`$nen are successfully reproduced.
The primordial curvature perturbation of the Universe, is already present a few Hubble times before cosmological scales start to enter the horizon . Its time-independent value at that stage seems to set the initial condition for the subsequent evolution of all cosmological perturbations. As a result, observation probes the stochastic properties of $`\zeta `$, which is found to be almost gaussian with an almost scale-invariant spectrum.
According to present ideas $`\zeta `$ is supposed to originate from the vacuum fluctuations during inflation of one or more light scalar fields, which on each scale are promoted to classical perturbations around the time of horizon exit . One takes inflation to be almost exponential (quasi de Sitter spacetime) corresponding to a practically constant Hubble parameter $`H_{\mathrm{inf}}`$, and the effective masses of the fields to be much less than $`H_{}`$. This ensures that the fields are almost massless and live in almost unperturbed quasi de Sitter spacetime, making their perturbations indeed almost gaussian and scale invariant (see Subsections 2.4.1 and 2.5.2). This automatically makes $`\zeta `$ almost scale invariant, (see Subsections 2.4.2 and 2.5.3) and can (though not automatically ) make it also almost gaussian.
All of this is of intense interest at the present time, because observation over the next few years will rule out most existing scenarios for the generation of $`\zeta `$, by detecting or bounding the scale dependence and non-gaussianity of $`\zeta `$. We will now describe a general procedure for calculating the level of non-gaussianity, by means of the $`\delta N`$ formalism (see also Refs. ).
### 6.2 Defining the curvature perturbation
Perturbations of the observable Universe are defined with respect to an unperturbed reference universe, which is homogeneous and isotropic (a FRW universe) (see Section 2.2). Its line element may be written as
$$ds^2=dt^2+a^2(t)\delta _{ij}dx^idx^j,$$
(6.1)
defining the unperturbed scale factor $`a(t)`$, time $`t`$, and the Cartesian spatial coordinates $`𝐱`$.
The curvature perturbation is only of interest after the universe has been smoothed on some scale $`\left(k/a\right)^1`$ much bigger than the horizon $`H^1`$. To define it, one takes the fixed-$`t`$ slices of spacetime to have uniform energy density, and the fixed-$`x`$ worldlines to be comoving. The spatial metric is \[c.f. Eq. (5.3)\]
$$g_{ij}=a^2(t)e^{2\zeta (t,𝐱)}\gamma _{ij}(t,𝐱)=\stackrel{~}{a}^2(t,𝐱)\gamma _{ij}(t,𝐱).$$
(6.2)
In this expression, $`\gamma _{ij}(t,𝐱)`$ has unit determinant, so that a volume of the Universe bounded by fixed comoving spatial coordinates is proportional to the locally defined scale factor $`\stackrel{~}{a}^3(t,𝐱)`$. In the inflationary scenario the factor $`\gamma _{ij}`$ just accounts for the tensor perturbation, but its form is irrelevant here (see Subsections 5.2.2 and 5.6.1). According to this definition, $`\zeta `$ is the perturbation in $`\mathrm{ln}\stackrel{~}{a}`$. Only the spatial variation of $`\zeta `$ is significant, and to make contact with observation we can work with its Fourier components in a box a bit bigger than the observable Universe, setting the zero mode equal to zero so that $`\zeta `$ has vanishing spatial average.
One can also consider a slicing whose metric has the form in Eq. (6.2) without the $`\zeta `$ factor, which we call the flat slicing. Starting from any initial flat slice at time $`t_{\mathrm{ini}}`$, let us define the amount of expansion
$$N(t,𝐱)\mathrm{ln}\left[\frac{\stackrel{~}{a}(t)}{a(t_{\mathrm{ini}})}\right],$$
(6.3)
to a final slice of uniform energy density. Then
$$\zeta (t,𝐱)=\delta NN(t,𝐱)N_0(t),$$
(6.4)
where
$$N_0(t)\mathrm{ln}\left[\frac{a(t)}{a(t_{\mathrm{ini}})}\right],$$
(6.5)
is the unperturbed amount of expansion.
To make use of the above formalism we assume that in the superhorizon regime ($`aHk`$), the evolution of the Universe at each position (the local evolution) is well approximated by the evolution of some unperturbed universe . This ‘separate universe’ assumption will presumably be correct on cosmological scales because these scales are so big .
By virtue of the separate universe assumption, $`N(t,𝐱)`$ is the amount of expansion in some unperturbed universe, allowing $`\zeta `$ to be evaluated knowing the evolution of a family of such universes. For a given content of the Universe it can be checked using the gradient expansion method, but we do not wish to assume a specific content.
The separate universe assumption leads also to local energy conservation. Indeed, using the uniform density slicing, and remembering that $`\stackrel{~}{a}`$ determines the expansion,
$$\dot{\rho }(t)=3\stackrel{~}{H}[\rho (t)+P(t,𝐱)]=3\left(H+\dot{\zeta }\right)[\rho (t)+P(t,𝐱)],$$
(6.6)
where $`\stackrel{~}{H}\dot{\stackrel{~}{a}}/\stackrel{~}{a}`$, $`\rho `$ is the energy density, and $`P`$ is the pressure. During any era when $`P`$ is a unique function of $`\rho `$ (the adiabatic condition), $`P`$ is uniform on the chosen slicing; then $`\dot{\zeta }`$ vanishes (because it is uniform and its spatial average vanishes) so that $`\zeta `$ is conserved (see Subsection 2.2.3). This consequence of the separate universe assumption was first recognised in full generality in Refs. (see also Ref. for the case of inflation, Refs. for the case of linear perturbation theory, and Refs. for a coordinate-free treatment).
### 6.3 The inflationary prediction
The evolution of the observable Universe, smoothed on the shortest cosmological scale, is supposed to be determined by the values of one or more light scalar fields when that scale first emerges from the quantum regime a few Hubble times after horizon exit. Defined on a flat slicing, each field $`\varphi _i`$ at this epoch will be of the form $`\varphi _i(𝐱)\varphi _{i_0}+\delta \varphi _i(𝐱)`$.
Because quasi exponential inflation is assumed, and only light fields are considered, it is a good approximation to take the $`\delta \varphi _i`$ to be almost massless fields living in unperturbed quasi de Sitter spacetime . In these circumstances the perturbations $`\delta \varphi _i`$ generated from the vacuum are almost gaussian, with an almost flat spectrum whose amplitude is
$$A_{\delta \varphi _i}\frac{H_{}}{2\pi }.$$
(6.7)
Now we invoke the separate universe assumption, and choose the homogeneous quantities $`\varphi _{i_0}`$ to correspond to the unperturbed universe. Then Eq. (6.4) for $`\zeta `$ becomes
$`\zeta (t,𝐱)=N(\rho (t),\varphi _1(𝐱),\varphi _2(𝐱),\mathrm{})N(\rho (t),\varphi _{1_0},\varphi _{2_0},\mathrm{}).`$ (6.8)
In this expression, the expansion $`N`$ is evaluated in an unperturbed universe, from an epoch when the fields have assigned values to one when the energy density has an assigned value $`\rho `$. This expression allows one to propagate forward the stochastic properties of $`\zeta `$ to the epoch when it becomes observable, given those of the initial field perturbations.
Since the observed curvature perturbation is almost gaussian, it must be given to good accuracy by one or more of the linear terms
$$\zeta (t,𝐱)\underset{i}{}N_{,i}(t)\delta \varphi _i(𝐱),$$
(6.9)
where we use the notation
$`N_{,i}`$ $``$ $`{\displaystyle \frac{N}{\varphi _{i_0}}},`$ (6.10)
$`N_{,ij}`$ $``$ $`{\displaystyle \frac{^2N}{\varphi _{i_0}\varphi _{j_0}}},`$ (6.11)
with the field perturbations being almost gaussian. Here we include for the first time the quadratic terms
$`\zeta (t,𝐱)={\displaystyle \underset{i}{}}N_{,i}(t)\delta \varphi _i+{\displaystyle \frac{1}{2}}{\displaystyle \underset{ij}{}}N_{,ij}(t)\delta \varphi _i\delta \varphi _j.`$ (6.12)
They may be entirely responsible for any observed non-gaussianity if the field perturbations are gaussian to sufficient accuracy<sup>1</sup><sup>1</sup>1Here and elsewhere, we are not displaying a homogeneous term needed to make the spatial average of $`\zeta `$ vanish..
### 6.4 Non-gaussianity
#### 6.4.1 The bispectrum
The stochastic properties of the perturbations are specified through expectation values which, according to the inflationary paradigm, are taken with respect to the time independent (Heisenberg picture) quantum state of the Universe (to be precise, the quantum state of the universe before it somehow collapses to give the observed Universe). Focusing on $`\zeta `$, we consider Fourier components,
$$\zeta _𝐤\frac{d^3k}{(2\pi )^{3/2}}\zeta (t,𝐱)\mathrm{exp}(i𝐤𝐱).$$
(6.13)
The stochastic properties of a gaussian perturbation are specified entirely by the spectrum $`𝒫_\zeta (k)A_\zeta ^2(k/aH_{\mathrm{inf}})^{n_\zeta }`$, defined through
$$\zeta _{𝐤_1}\zeta _{𝐤_2}=P_\zeta (k)\delta ^3(𝐤_1+𝐤_2),$$
(6.14)
and
$$𝒫_\zeta (k)\frac{k^3}{2\pi ^2}P_\zeta (k).$$
(6.15)
From Eqs. (6.7) and (6.9)
$$A_\zeta ^2=\left(\frac{H_{}}{2\pi }\right)^2\underset{i}{}N_{,i}^{}{}_{}{}^{2}.$$
(6.16)
Non-gaussianity is defined through higher correlations. We consider only the three-point correlation<sup>2</sup><sup>2</sup>2The four-point correlation may give a competitive observational signature and can be calculated in a similar fashion .. It defines the bispectrum $`B_\zeta `$ through (see Subsection 5.5.1)
$$\zeta _{𝐤_1}\zeta _{𝐤_2}\zeta _{𝐤_3}=(2\pi )^{3/2}B_\zeta (k_1,k_2,k_3)\delta ^3(𝐤_1+𝐤_2+𝐤_3).$$
(6.17)
Its normalisation is specified by a parameter $`f_{\mathrm{NL}}`$ according to
$$B_\zeta \frac{6}{5}f_{\mathrm{NL}}(k_1,k_2,k_3)[P_\zeta (k_1)P_\zeta (k_2)+\mathrm{cyclic}\mathrm{perturbations}].$$
(6.18)
In first-order cosmological perturbation theory the gauge-invariant gravitational potential $`\mathrm{\Phi }_g`$ during matter domination before horizon entry is $`\mathrm{\Phi }_g=(5/3)\zeta `$, and our definition of $`f_{\mathrm{NL}}`$ coincides with the definition
$$B_{\mathrm{\Phi }_g}2f_{\mathrm{NL}}(k_1,k_2,k_3)[P_{\mathrm{\Phi }_g}(k_1)P_{\mathrm{\Phi }_g}(k_2)+\mathrm{cyclic}\mathrm{perturbations}].$$
(6.19)
At second-order these definitions of $`f_{\mathrm{NL}}`$ differ (see for example Refs. . See also footnotes 8 and 9 in Chapter 5).
We shall take $`𝒫_\zeta (k)`$ and $`f_{\mathrm{NL}}`$ to be evaluated when cosmological scales approach the horizon and $`\zeta `$ becomes observable. Observation gives $`|A_\zeta |5\times 10^5`$ , and $`|f_{\mathrm{NL}}|\text{ }\stackrel{<}{}\text{ }100`$ . Absent of a detection, this will eventually come down to roughly $`|f_{\mathrm{NL}}|\text{ }\stackrel{<}{}\text{ }1`$ .
Ignoring any non-gaussianity of the $`\delta \varphi _i`$, our formula in Eq. (6.12) makes $`f_{\mathrm{NL}}`$ practically independent of the wavenumbers. Indeed, generalising the result found in Ref. , we have calculated (see Appendix A)
$`{\displaystyle \frac{3}{5}}f_{\mathrm{NL}}={\displaystyle \frac{_{ij}N_{,i}N_{,j}N_{,ij}}{2\left[_iN_{,i}^{}{}_{}{}^{2}\right]^2}}+\mathrm{ln}(kL){\displaystyle \frac{A_\zeta ^2}{2}}{\displaystyle \frac{_{ijk}N_{,ij}N_{,jk}N_{,ki}}{\left[_iN_{,i}^{}{}_{}{}^{2}\right]^3}}.`$ (6.20)
In deriving this expression we used the amplitude of the spectrum $`A_{\delta \varphi _i}H_{}/2\pi `$ of the field perturbations, and used Eq. (6.16) to eliminate $`H_{}`$ in favour of $`A_\zeta `$. As discussed in Ref. , the logarithm can be taken to be of order 1, because it involves the size $`k^1`$ of a typical scale under consideration, relative to the size $`L`$ of the region within which the stochastic properties are specified. Except for the logarithm, $`f_{\mathrm{NL}}`$ is scale independent if the field perturbations are gaussian.
If only one $`\delta \varphi _i`$ is relevant, Eq. (6.12) becomes
$$\zeta (t,𝐱)=N_{,i}\delta \varphi _i+\frac{1}{2}N_{,ii}(\delta \varphi _i)^2,$$
(6.21)
and because the first term dominates, Eq. (6.20) becomes
$$\frac{3}{5}f_{\mathrm{NL}}=\frac{1}{2}\frac{N_{,ii}}{N_{,i}^{}{}_{}{}^{2}}.$$
(6.22)
In this case, $`f_{\mathrm{NL}}`$ may equivalently be defined by writing
$$\zeta =\zeta _\mathrm{g}\frac{3}{5}f_{\mathrm{NL}}\zeta _\mathrm{g}^2,$$
(6.23)
where $`\zeta _\mathrm{g}`$ is gaussian.
To include the possible non-gaussianity of the $`\delta \varphi _i`$, we define the bispectra $`B_{ijk}`$ of the dimensionless field perturbations $`(2\pi /H_{})\delta \varphi _i`$ and their normalisation $`f_{ijk}`$, in exactly the same way that we defined $`B_\zeta `$ and $`f_{\mathrm{NL}}`$ \[c.f. Eqs. (5.50) and (5.51)\]. These bispectra add the following contribution to $`f_{\mathrm{NL}}`$ in Eq. (6.20) (see Appendix A):
$$\mathrm{\Delta }f_{\mathrm{NL}}=\frac{_{ijk}N_{,i}N_{,j}N_{,k}f_{ijk}(k_1,k_2,k_3)}{\left(_iN_{,i}^{}{}_{}{}^{2}\right)^{3/2}}|A_\zeta ^1|.$$
(6.24)
The $`f_{ijk}`$, generated directly from the vacuum fluctuation, will depend strongly on the wavenumbers \[c.f. Eq. (5.52)\].
#### 6.4.2 Cosmological perturbation theory
In the superhorizon regime the non-linear theory that we have used is a complete description. The basic expression in Eq. (6.8) is non-perturbative, giving $`\zeta (t,𝐱)`$ in terms of the initial fields and the expansion of a family of unperturbed universes. The second-order expansion in Eq. (6.12) is a matter of convenience. As we shall see it seems to be adequate in practice, but Eq. (6.8) would still be applicable if the expansion converged slowly or not at all.
Cosmological perturbation theory (CPT) is completely different. It is applicable both inside and outside the horizon, being at each instant a power series in the perturbations of the metric and the stress-energy tensor, together with whatever variables are needed to completely specify the latter and close the system of equations. During inflation these variables are the components of the inflaton, while afterwards they may involve oscillating fields and a description of the particle content. First-order CPT is usually adequate and can describe non-gaussianity at the level $`|f_{\mathrm{NL}}|1`$ which has to be generated by the second-order term in Eq. (6.12). Second-order CPT<sup>3</sup><sup>3</sup>3For a full set of CPT equations see e.g. Ref. . is generally needed only to handle non-gaussianity at the level $`|f_{\mathrm{NL}}|1`$.
Quantised CPT is needed to calculate the stochastic properties of the initial field perturbations $`\delta \varphi _i`$, which are the input for our calculation. The slow-roll spectrum in Eq. (6.7) comes from the first-order calculation. The bispectrum is a second-order effect and has, in the context of slow-roll inflation, been calculated in Refs. \[c.f. Eqs. (5.50) - (5.52)\]. It is shown elsewhere that $`\mathrm{\Delta }f_{\mathrm{NL}}`$ is in this case negligible compared to 1, being generically below $`(15/24)f_{SL}\sqrt{r_{T\zeta }\epsilon }\text{ }\stackrel{<}{}\text{ }10^2`$ with $`1/3f_{SL}11/18`$ \[c.f. Eqs. (5.52) and (6.24)\]<sup>4</sup><sup>4</sup>4In Ref. (see also Refs. ) Rigopoulos et. al. calculated the three-point correlator of $`\zeta `$ in general multi-component inflationary models using a stochastic approach. Their result is quite puzzling since the wavevector dependence seems to be significant even well after horizon exit. Reconciliation between the approach followed in Ref. and ours is desirable .. Higher order correlators have not been calculated yet and would give an additional contribution to Eq. (6.20), which presumably is also negligible. A few single field non slow-roll models have been investigated where it is found that $`\mathrm{\Delta }f_{\mathrm{NL}}`$ could be much bigger than 1. From now on we take the $`\delta \varphi _i`$ to be gaussian.
In the regime $`aHk`$, perturbation theory must be compatible with Eq. (6.12). In particular, the non-local terms, present at second order for a generic perturbation, must be absent for $`\zeta `$. Finally, CPT is needed to evolve the perturbations after horizon entry, but that is not our concern here. In the following, we apply our formalism to calculate $`f_{\mathrm{NL}}`$ in various cases and compare it with the CPT result where that is known.
### 6.5 The $`\delta N`$ formalism in some multi-component models
#### 6.5.1 A two-component inflation model
As a first use of Eq. (6.20) we consider the two-component inflation model of Kadota and Stewart , estimating for the first time the non-gaussianity which it predicts. The model works with a complex field $`\mathrm{\Phi }`$, which is supposed to be a modulus with a point of enhanced symmetry at the origin. The scalar potential is given by
$$V=V_hm^2|\mathrm{\Phi }|^2+\frac{1}{3}Am^2[\mathrm{\Phi }^3+\mathrm{\Phi }^3]+\frac{1}{2}\nu (\nu +1)A^2m^2|\mathrm{\Phi }|^4,$$
(6.25)
with $`A`$ being fixed so that the vacuum energy vanishes at the minimum of the potential, and $`\nu =\{1,2,3,\mathrm{}\}`$. Writing
$$\mathrm{\Phi }=|\mathrm{\Phi }|e^{i\theta },$$
(6.26)
the tree level potential has a maximum at $`\mathrm{\Phi }_0=0`$ and depends on both $`|\mathrm{\Phi }|`$ and $`\theta `$. A one-loop correction turns the maximum into a crater and inflation occurs while $`\mathrm{\Phi }`$ is rolling away from the rim of the crater (see Fig. 6.1). The curvature perturbation is supposed to be constant after the end of slow-roll inflation. For $`\theta _0\theta _c`$, with $`\theta _c`$ being a parameter of the model, it is found that
$$N\left|\frac{\theta _c}{\theta _0}\right|.$$
(6.27)
Through the first term of Eq. (6.20)
$$f_{\mathrm{NL}}\left|\frac{\theta _0}{\theta _c}\right|,$$
(6.28)
which is too small ever to be observed.
#### 6.5.2 The curvaton model
In the curvaton model (see also Refs. and Subsection 2.4.2) the curvature perturbation $`\zeta `$ grows, from a negligible value in an initially radiation dominated epoch, due to the oscillations of a weakly coupled light field $`\sigma `$ (the curvaton) around the minimum of its quadratic potential
$$V_\sigma (t,𝐱)=\frac{1}{2}m_\sigma ^2\sigma ^2(t,𝐱),$$
(6.29)
where $`m_\sigma `$ is the curvaton effective mass. Due to the oscillations, the initially negligible curvaton energy density redshifts as
$$\rho _\sigma (t,𝐱)\frac{1}{2}m_\sigma ^2\sigma _a^2(t,𝐱)a^3(t,𝐱),$$
(6.30)
where $`\sigma _a`$ represents the amplitude of the oscillations. Meanwhile the radiation energy density $`\rho _r`$ redshifts as $`a^4`$. Soon after the curvaton decay, the standard Hot Big-Bang is recovered and $`\zeta `$ is assumed to be conserved until horizon reentry.
To calculate $`f_{\mathrm{NL}}`$ using Eq. (6.20) we first realise that $`\sigma _{}`$ (the unperturbed value of $`\sigma `$ a few Hubble times after horizon exit) is the only relevant quantity since the curvature perturbation produced by the inflaton, and imprinted in the radiation fluid during the reheating process, is supposed to be negligible. Thus, Eq. (6.22) applies. Second, we can redefine $`N`$ as the number of e-folds from the beginning of the sinusoidal oscillations to the curvaton decay. This is because the number of e-folds from the end of inflation to the beginning of the oscillations is completely unperturbed as the radiation energy density dominates during that time. Thus, $`N`$ is now a function of three variables
$$N(\rho _{dec},\rho _{osc},\sigma _{})=\frac{1}{3}\mathrm{ln}\left(\frac{\rho _{\sigma _{\mathrm{osc}}}}{\rho _{\sigma _{dec}}}\right)=\frac{1}{3}\mathrm{ln}\left[\frac{\frac{1}{2}m_\sigma ^2[g(\sigma _{})]^2}{\rho _{\sigma _{dec}}}\right],$$
(6.31)
where $`g\sigma _{osc}`$ is the amplitude at the beginning of the sinusoidal oscillations as a function of $`\sigma _{}`$. Here the curvaton energy density just before the curvaton decay $`\rho _{\sigma _{dec}}`$ is expressed in terms of the total energy density $`\rho _{dec}`$ at that time, the total energy density at the beginning of the sinusoidal oscillations $`\rho _{osc}`$, and $`g`$ by
$$\rho _{\sigma _{dec}}=\frac{1}{2}m_\sigma ^2[g(\sigma _{})]^2\left(\frac{\rho _{dec}\rho _{\sigma _{dec}}}{\rho _{osc}}\right)^{3/4}.$$
(6.32)
After evaluating $`/\sigma _{}=g^{}/g`$, at fixed $`\rho _{dec}`$ and $`\rho _{osc}`$, we obtain
$$N_{,\sigma _{}}=\frac{2}{3}r\frac{g^{}}{g},$$
(6.33)
where
$$r\frac{3\rho _{\sigma _{dec}}}{3\rho _{\sigma _{dec}}+4\rho _{r_{dec}}},$$
(6.34)
being $`\rho _{r_{dec}}`$ the radiation energy density just before the curvaton decay, giving
$$A_\zeta =\frac{H_{}}{2\pi }N_{,\sigma _{}}=\frac{H_{}r}{3\pi }\frac{g^{}}{g},$$
(6.35)
in agreement with first-order cosmological perturbation theory in the sudden decay approximation (see Subsection 2.4.2). Differentiating again we find from Eq. (6.22)
$$f_{\mathrm{NL}}=\frac{5}{6}\frac{N_{,\sigma _{}\sigma _{}}}{N_{,\sigma _{}}^2}=\frac{5}{3}+\frac{5}{6}r\frac{5}{4r}\left(1+\frac{gg^{\prime \prime }}{g^2}\right),$$
(6.36)
which nicely agrees with the already calculated $`f_{\mathrm{NL}}`$ using first- and second-order perturbation theory (see Refs. and Subsection 5.7.4).
#### 6.5.3 Another two-component model
Finally we consider the two-component inflation model of Ref. (see also Refs. and Subsection 5.7.2). For at least some number $`N`$ of $`e`$-folds after cosmological scales leave the horizon, the potential in Eq. (5.56) is written as
$$V=V_h\left(1+\frac{1}{2}\eta _\phi \frac{\phi ^2}{m_P^2}+\frac{1}{2}\eta _\vartheta \frac{\vartheta ^2}{m_P^2}\right),$$
(6.37)
with the first term dominating, and $`\eta _\phi `$ and $`\eta _\vartheta `$ being the slow-roll $`\eta `$ parameters. The idea is to use Eq. (6.20) to calculate the non-gaussianity after the $`N`$ e-folds which, barring cancellations, will place a lower limit on the observed non-gaussianity.
The slow-roll equations give the background field values $`\phi _0(N)`$ and $`\vartheta _0(N)`$ after $`N`$ e-folds, in terms of those obtaining just after horizon exit
$`\phi _0(N)`$ $`=`$ $`\phi _0\mathrm{exp}(N\eta _\phi ),`$ (6.38)
$`\vartheta _0(N)`$ $`=`$ $`\vartheta _0\mathrm{exp}(N\eta _\vartheta ).`$ (6.39)
This gives
$$V_0(N,\phi _0,\vartheta _0)=V_h\left(1+\frac{1}{2}\eta _\phi \frac{\phi _0^2}{m_P^2}e^{2N\eta _\phi }+\frac{1}{2}\eta _\vartheta \frac{\vartheta _0^2}{m_P^2}e^{2N\eta _\vartheta }\right),$$
(6.40)
and allows us to calculate the derivatives of $`N`$ with respect to $`\phi _0`$ and $`\vartheta _0`$ at fixed $`V`$. Focusing on the case $`\vartheta _0=0`$ considered in Ref. , we find
$$\zeta =\frac{\delta \phi }{\eta _\phi \phi _0}\frac{\eta _\phi }{2}\left(\frac{\delta \phi }{\eta _\phi \phi _0}\right)^2+\frac{\eta _\vartheta }{2}e^{2N(\eta _\phi \eta _\vartheta )}\left(\frac{\delta \vartheta }{\eta _\phi \phi _0}\right)^2,$$
(6.41)
in agreement with the second-order perturbation calculation of Ref. . If the observed $`\zeta `$ has a non-gaussian part $`\zeta _\sigma `$ equal to the last term of Eq. (6.41) and a gaussian part generated mostly after inflation, one can obtain $`|f_{\mathrm{NL}}|>1`$ by choosing $`\eta _\phi >0.26`$, $`\eta _\vartheta =\eta _\phi /2`$, $`N=70`$, and $`\zeta _\sigma =10^2\zeta `$.
Our calculated expression for the coefficient of $`(\delta \vartheta )^2`$ is in disagreement with the one found in Ref. \[c.f. Eq. (5.68)\] which uses a set of CPT equations based on those presented in Ref. <sup>5</sup><sup>5</sup>5The initial calculations of the structural form of $`\zeta `$ in this model using CPT were in gross conflict with Eq. (6.41). This is because the time evolutions of $`\phi `$ and $`\vartheta `$ outside the horizon were not considered. The sources of discrepancy were recognised in Ref. , showing that the actual order of magnitude is in agreement with Eq. (6.41) except for the presence of non-local terms (see Subsection 5.7.2).. After converting the variable used there to our $`\zeta `$ (see Subsection 5.3.3), these equations give $`\dot{\zeta }`$ in terms of first-order quantities, but they contain non-local terms involving the inverse Laplacian . Comparison with our non-linear expression in Eq. (6.8) shows that such terms must cancel if correctly evaluated.
### 6.6 Conclusions
The $`\delta N`$ formalism is a non-perturbative approach to calculate the curvature perturbation $`\zeta `$ at all orders, in terms only of background quantities that describe the evolution of a family of unperturbed universes. Such a formalism was originally introduced to calculate the spectrum of $`\zeta `$ at first order (see also Ref. ) in multi-component inflationary models. Now, with the increasing interest in the non-gaussian features of $`\zeta `$ in both single- and multi-component inflationary models, the formalism has been extended to calculate the curvature perturbation at second order $`\zeta _2`$ and the normalisation $`f_{\mathrm{NL}}`$ of the bispectrum . The $`\delta N`$ formalism relies on the separate universe assumption, which says that on superhorizon scales the Universe behaves locally as if it were unperturbed , and on the intrinsic gaussianity of the fields $`\varphi _i`$ involved. The quantities $`\zeta _2`$ and $`f_{\mathrm{NL}}`$ are easily given in terms of the first and second derivatives of the unperturbed number of e-folds $`N`$, from an epoch when the fields have assigned values $`\varphi _{i_0}`$ to one when the energy density has an assigned value $`\rho `$, with respect to the unperturbed fields $`\varphi _{i_0}`$. The possible intrinsic non-gaussianity of the fields $`\varphi _i`$ would lead to an additional contribution to $`f_{\mathrm{NL}}`$, highly wavevector dependent but in any case negligible compared to 1 . The $`\delta N`$ formalism reproduces the well known results in single-component inflation , in the curvaton scenario , and in the ‘hybrid’ model of Enqvist and V$`\ddot{\mathrm{a}}`$ihk$`\ddot{\mathrm{o}}`$nen . In addition, the Kadota and Stewart’s modular inflation model has served as an example of the power of this formalism. The $`\delta N`$ formalism is an interesting alternative to cosmological perturbation theory where, order by order, the relevant expressions to calculate $`\zeta _2`$ and $`f_{\mathrm{NL}}`$ tend to be more and more complicated (see for example Refs. ). Nevertheless, it is true that the latter is valid on all the scales, while the former is only valid in the superhorizon regime. Fortunately, the subhorizon effects are negligible making the $`\delta N`$ formalism reliable and completely independent of cosmological perturbation theory.
## Chapter 7 Conclusions
Prior to the standard Hot Big-Bang, a period of accelerated expansion seems to have been crucial . Not only does this period solve the classical problems of the Big-Bang cosmological model, namely the horizon, flatness, and unwanted relics problems, but it also amplifies the fluctuations in the light scalar fields $`\varphi _i`$ living in the Friedmann-Robertson-Walker spacetime . This inflationary process serves also for the scalar field fluctuations to become classical soon after horizon exit , giving birth to the primordial perturbations in the energy density that generate the temperature anisotropies in the cosmic microwave background radiation, and, through gravitational collapse, the large-scale structure observed today. Such primordial perturbations can be defined perturbatively on a homogeneous and isotropic background, but the freedom to choose the perturbed coordinate system makes them gauge dependent. To characterize adequately the primordial perturbations, we introduce the gauge-invariant curvature perturbation $`\zeta `$ which represents the intrinsic spatial curvature on slices of uniform energy density (or slices with zero flow of energy). While the pressure is a unique function of the energy density, $`\zeta `$ is conserved at all orders , which makes this an ideal quantity. In this thesis we have explored some of the theoretical and statistical aspects of the origin of the large-scale structure, such as the two most known scenarios to generate $`\zeta `$ (the inflaton and the curvaton scenario), the required inflationary energy scale in those scenarios, and the non-gaussianity associated to $`\zeta `$.
In general $`\zeta `$ depends on the perturbations in the scalar fields $`\varphi _i`$ during the inflationary period, whose spectra $`𝒫_{\delta \varphi _i}(k)`$ are generically the same for all kinds of quasi exponential expansion either the respective field dominates the energy density or not. The only appreciable difference is in the way the scale dependence is given in terms of the Hubble parameter $`H_{}`$ a few Hubble times after horizon exit and the mass $`m_{\varphi _i}`$ of the respective field. However, the spectrum of $`\zeta `$, $`𝒫_\zeta (k)`$, varies significantly among the different possible scenarios for the origin of the large-scale structure, although in all cases it is almost gaussian and scale invariant. We have studied in Chapter 2 two different scenarios for the origin of $`\zeta `$, the inflaton and the curvaton scenario (see also Refs. ), pointing out their different signatures and connecting them with the amplitude of gravitational waves produced in each scenario. Both scenarios have their advantages and drawbacks. For example, the single-component inflaton scenario presents a consistency relation that relates the amplitude of $`𝒫_\zeta (k)`$ with the amplitude of the gravitational waves spectrum $`𝒫_T(k)`$ . The possible detection of gravitational waves is consistent with this scenario and would serve as a smoking gun if the consistency relation is satisfied. The drawback is that the generation of $`\zeta `$ by the inflaton field $`\phi `$ (the field which drives inflation) imposes severe constraints on the theoretical model building . In contrast, the curvaton scenario circumvents the latter problem since the generation of $`\zeta `$ is assigned to a weakly coupled scalar field $`\sigma `$ (the curvaton) different to the inflaton. The drawbacks are that there is no consistency relation in this case, and that the scenario is inconsistent with a detectable level of gravitational waves .
There exists several well motivated inflationary models that locate the inflaton field within a particle physics framework . Most of these models are however unrealistic due to the strong constraints mentioned above. In particular, the generation of adiabatic perturbations is almost inconsistent with the low inflationary energy scale, given by $`H_{}`$, required to identify the inflaton with one of the many fields present in supersymmetry. The curvaton scenario comes to rescue these models, allowing for a much lower inflationary energy scale. But how low may this energy scale be in the curvaton scenario?. In Chapter 3 we have discussed the lower bound on $`H_{}`$ in the simplest curvaton setup, showing that it is high enough ($`H_{}>10^7`$ GeV) to fail at rescuing the inflationary models whose potentials are generated by some mechanism of gravity-mediated supersymmetry breaking where $`H_{}10^3`$ GeV is required. The general conditions to obtain low scale inflation in the curvaton scenario have been given in that chapter , in terms of three quantities $`ϵ`$, $`f`$, and $`\delta `$, that parameterize respectively the evolution of the curvaton field from the time of horizon exit to the beginning of its oscillations, the effective curvaton mass $`\stackrel{~}{m}_\sigma `$ at the end of a phase transition with respect to the Hubble parameter at the same time, and the time of the phase transition with respect to the time of horizon exit. In Chapters 3 and 4 we have invoked the ‘heavy curvaton’ picture , defined as the setup where $`\sigma `$ suddenly increases its mass at the end of a phase transition much later than inflation . The mass increment is given by the coupling of $`\sigma `$ (parameterized by the constant $`\lambda `$) with another field which acquires a large vacuum expectation value at the end of the same period. Thus, the smallness of the parameters $`f`$ and $`\delta `$ are exploited to allow for an inflationary scale consistent with gravity-mediated susy breaking<sup>1</sup><sup>1</sup>1The parameter $`ϵ`$ is in these cases unmodified because it is assumed that the unperturbed component of $`\sigma `$ is frozen throughout inflation and until oscillations begin.. In the cases presented in these two chapters the phase transition is associated with the end of a second (thermal) inflationary period, which was originally introduced to dilute the abundances of unwanted relics that the first (main) inflationary period is not able to do (for instance the moduli fields) . Thermal inflation is driven by the confinement of a second (flaton) field $`\chi `$ at the origin of the potential due to the thermal effects from the radiation background left by the inflaton decay . The eventual rolling of $`\chi `$, towards the minimum of its potential, ends thermal inflation and triggers an increment in the bare mass $`m_\sigma `$ of $`\sigma `$ through the coupling of the latter with $`\chi `$. Solving the moduli problem while satisfying adequately all the constraints in the first model discussed in Chapter 3, where the curvaton oscillates for some time before decaying, restricts the parameter space to a region where the two important parameters, $`\lambda `$ and $`m_\sigma `$, are required to be very small ($`10^{22}\text{ }\stackrel{<}{}\text{ }\lambda \text{ }\stackrel{<}{}\text{ }10^{10}`$ and $`m_\sigma \text{ }\stackrel{<}{}\text{ }10^1\mathrm{GeV}`$) . It is likely then that the curvaton field is a pseudo Nambu-Goldstone boson. In contrast, as discussed in the second model presented in Chapter 4, if the increment in the mass of the curvaton field is so high that the decay rate overtakes the Hubble parameter, leading to the immediate decay of $`\sigma `$, the parameter space is less constrained, with more natural values for $`\lambda `$ and $`m_\sigma `$ ($`10^{10}\text{ }\stackrel{<}{}\text{ }\lambda \text{ }\stackrel{<}{}\text{ }10^4`$ and $`m_\sigma \text{ }\stackrel{<}{}\text{ }1\mathrm{GeV}`$) .
Although $`\zeta `$ is found to be almost perfectly gaussian in most of the models that give account of its origin, as required by observation, the possibly present small non-gaussianity is being scrutinized by present experiments like the WMAP satellite and will be the focus of future experiments like the PLANCK satellite . The first statistical significant quantity that gives us information about the level of non-gaussianity is the bispectrum, which corresponds to the three-point correlator of $`\zeta `$ . Its normalisation is given by the parameter $`f_{\mathrm{NL}}`$, which has been found to be $`|f_{\mathrm{NL}}|\text{ }\stackrel{<}{}\text{ }10^2`$ but that will eventually go down to $`|f_{\mathrm{NL}}|1`$ in the forthcoming years unless there is an earlier detection . In view of the difficulty at discriminating between models by means of only the spectral index and/or consistency relations, the detection of non-gaussianity and a precise determination of $`f_{\mathrm{NL}}`$ would be useful tools to serve this purpose. For example, in the single-component inflaton scenario the level of non-gaussianity is very small, being $`|f_{\mathrm{NL}}|`$ or the order of the slow-roll parameters . In contrast, in the curvaton scenario $`|f_{\mathrm{NL}}|`$ could be much higher, being $`f_{\mathrm{NL}}`$ negative, or, if $`\sigma `$ has already dominated the energy density before decaying, the precise value for $`f_{\mathrm{NL}}`$ would be $`f_{\mathrm{NL}}=+5/4`$ . In Chapters 5 and 6 we have addressed this subject, following a perturbative approach in the former, and a non-perturbative one in the latter. In Chapter 5 we have presented in an unified way the different definitions of the second-order curvature perturbation $`\zeta _2`$ present in the literature . The translation rules to go from one definition to another have been explicitly given, and would help to avoid the possible confusion when confronting different papers and results which use different definitions for $`\zeta _2`$ . We have examined the predictions for $`f_{\mathrm{NL}}`$ coming from the single-component inflation model, the multi-component one, and the curvaton scenario, discussing also the respective predictions in preheating and the inhomogeneous reheating scenario. Although multi-component inflation, preheating, and the inhomogeneous reheating scenarios do not predict a definite value or set of values for $`f_{\mathrm{NL}}`$, being very dependent on the specific model, the single-component inflaton scenario as well as the curvaton one do give definite predictions for $`f_{\mathrm{NL}}`$ . Multi-component slow-roll inflation is studied in Chapter 6 following the non-perturbative $`\delta N`$ formalism , that allows us to calculate $`\zeta `$ at all orders only by knowing the evolution of a family of unperturbed universes. Immediate applications of this formalism are the calculation of the spectral index , and $`f_{\mathrm{NL}}`$ , in a general multi-component slow-roll inflation model. The formalism relies on the separate universe assumption, which says that the local evolution of the Universe on superhorizon scales is the same as that of an unperturbed universe . The normalisation parameter $`f_{\mathrm{NL}}`$ is easily given in terms of first and second derivatives of the number of e-folds, from an initially flat slice to a final uniform energy density slice, with respect to the fields living in the former. The intrinsic non-gaussianity in the fields $`\varphi _i`$ relevant for the evolution of the family of universes may be also taken into account, but its contribution, which is highly wavevector dependent , is in any case negligible compared to 1 . Comparison of this formalism with cosmological perturbation theory has been made, in the case of the curvaton scenario and the ‘hybrid’ model of Enqvist and V$`\ddot{\mathrm{a}}`$ihk$`\ddot{\mathrm{o}}`$nen . According to Ref. the results for the latter model following the two approaches agree, and moreover they refute the claim about the possible presence of non-local terms in cosmological perturbation theory .
## Appendix A $`f_{\mathrm{NL}}`$ from the $`\delta N`$ formalism
In this appendix we derive explicitly the expression in Eq. (6.20) which gives the normalisation $`f_{\mathrm{NL}}`$ of the bispectrum of $`\zeta `$ defined in Eqs. (6.17) and (6.18).
We begin by writing $`\zeta `$ as given by Eq. (6.12) including the homogeneous term which makes the spatial average of $`\zeta `$ vanish:
$$\zeta =\underset{i}{}N_{,i}\delta \varphi _i+\frac{1}{2}\underset{ij}{}N_{,ij}\delta \varphi _i\delta \varphi _j\frac{1}{2}\underset{ij}{}N_{,ij}\delta \varphi _i\delta \varphi _j.$$
(A.1)
The corresponding mode function is then written as
$$\zeta _𝐤=\underset{i}{}N_{,i}\delta _𝐤\varphi _i+\frac{1}{2}\underset{ij}{}N_{,ij}\frac{d^3p}{(2\pi )^{3/2}}\delta _{𝐤+𝐩}\varphi _i\delta _𝐩\varphi _j^{}\frac{1}{2}(2\pi )^{3/2}\delta ^3(𝐤)\underset{ij}{}N_{,ij}\delta \varphi _i\delta \varphi _j.$$
(A.2)
Making use of the above formula, the product of three mode functions $`\zeta _𝐤`$ is therefore
$`\zeta _{𝐤_1}\zeta _{𝐤_2}\zeta _{𝐤_3}`$ $`=`$ $`{\displaystyle \underset{ijk}{}}N_{,i}N_{,j}N_{,k}\delta _{𝐤_1}\varphi _i\delta _{𝐤_2}\varphi _j\delta _{𝐤_3}\varphi _k`$
$`+`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{ijkl}{}}N_{,i}N_{,j}N_{,kl}[\delta _{𝐤_1}\varphi _i\delta _{𝐤_2}\varphi _j{\displaystyle }{\displaystyle \frac{d^3p}{(2\pi )^{3/2}}}\delta _{𝐤_3+𝐩}\varphi _k\delta _𝐩\varphi _l^{}`$
$`+`$ $`\delta _{𝐤_1}\varphi _i\delta _{𝐤_3}\varphi _j{\displaystyle \frac{d^3p}{(2\pi )^{3/2}}\delta _{𝐤_2+𝐩}\varphi _k\delta _𝐩\varphi _l^{}}`$
$`+`$ $`\delta _{𝐤_2}\varphi _i\delta _{𝐤_3}\varphi _j{\displaystyle }{\displaystyle \frac{d^3p}{(2\pi )^{3/2}}}\delta _{𝐤_1+𝐩}\varphi _k\delta _𝐩\varphi _l^{}]`$
$``$ $`{\displaystyle \frac{1}{2}}(2\pi )^{3/2}{\displaystyle \underset{ijkl}{}}N_{,i}N_{,j}N_{,kl}[\delta _{𝐤_1}\varphi _i\delta _{𝐤_2}\varphi _j\delta ^3(𝐤_3)\delta \varphi _k\delta \varphi _l`$
$`+`$ $`\delta _{𝐤_1}\varphi _i\delta _{𝐤_3}\varphi _j\delta ^3(𝐤_2)\delta \varphi _k\delta \varphi _l`$
$`+`$ $`\delta _{𝐤_2}\varphi _i\delta _{𝐤_3}\varphi _j\delta ^3(𝐤_1)\delta \varphi _k\delta \varphi _l]`$
$`+`$ $`{\displaystyle \frac{1}{4}}{\displaystyle \underset{ijklm}{}}N_{,i}N_{,jk}N_{,lm}[{\displaystyle }{\displaystyle \frac{d^3p}{(2\pi )^{3/2}}}{\displaystyle \frac{d^3p^{}}{(2\pi )^{3/2}}}\delta _{𝐤_1}\varphi _i\delta _{𝐤_2+𝐩}\varphi _j\delta _𝐩\varphi _k^{}\delta _{𝐤_3+𝐩^{}}\varphi _l\delta _𝐩^{}\varphi _m^{}`$
$`+`$ $`{\displaystyle \frac{d^3p}{(2\pi )^{3/2}}\frac{d^3p^{}}{(2\pi )^{3/2}}\delta _{𝐤_2}\varphi _i\delta _{𝐤_1+𝐩}\varphi _j\delta _𝐩\varphi _k^{}\delta _{𝐤_3+𝐩^{}}\varphi _l\delta _𝐩^{}\varphi _m^{}}`$
$`+`$ $`{\displaystyle }{\displaystyle \frac{d^3p}{(2\pi )^{3/2}}}{\displaystyle \frac{d^3p^{}}{(2\pi )^{3/2}}}\delta _{𝐤_3}\varphi _i\delta _{𝐤_1+𝐩}\varphi _j\delta _𝐩\varphi _k^{}\delta _{𝐤_2+𝐩^{}}\varphi _l\delta _𝐩^{}\varphi _m^{}]`$
$``$ $`{\displaystyle \frac{1}{4}}(2\pi )^{3/2}{\displaystyle \underset{ijklm}{}}N_{,i}N_{,jk}N_{,lm}[\delta ^3(𝐤_1)\delta \varphi _j\delta \varphi _k{\displaystyle }{\displaystyle \frac{d^3p}{(2\pi )^{3/2}}}\delta _{𝐤_2}\varphi _i\delta _{𝐤_3+𝐩}\varphi _l\delta _𝐩\varphi _m^{}`$
$`+`$ $`\delta ^3(𝐤_2)\delta \varphi _j\delta \varphi _k{\displaystyle \frac{d^3p}{(2\pi )^{3/2}}\delta _{𝐤_1}\varphi _i\delta _{𝐤_3+𝐩}\varphi _l\delta _𝐩\varphi _m^{}}`$
$`+`$ $`\delta ^3(𝐤_3)\delta \varphi _l\delta \varphi _m{\displaystyle \frac{d^3p}{(2\pi )^{3/2}}\delta _{𝐤_1}\varphi _i\delta _{𝐤_2+𝐩}\varphi _j\delta _𝐩\varphi _k^{}}`$
$`+`$ $`\delta ^3(𝐤_1)\delta \varphi _j\delta \varphi _k{\displaystyle \frac{d^3p}{(2\pi )^{3/2}}\delta _{𝐤_3}\varphi _i\delta _{𝐤_2+𝐩}\varphi _l\delta _𝐩\varphi _m^{}}`$
$`+`$ $`\delta ^3(𝐤_2)\delta \varphi _j\delta \varphi _k{\displaystyle \frac{d^3p}{(2\pi )^{3/2}}\delta _{𝐤_3}\varphi _i\delta _{𝐤_1+𝐩}\varphi _l\delta _𝐩\varphi _m^{}}`$
$`+`$ $`\delta ^3(𝐤_3)\delta \varphi _l\delta \varphi _m{\displaystyle }{\displaystyle \frac{d^3p}{(2\pi )^{3/2}}}\delta _{𝐤_2}\varphi _i\delta _{𝐤_1+𝐩}\varphi _j\delta _𝐩\varphi _k^{}]`$
$`+`$ $`{\displaystyle \frac{1}{4}}(2\pi )^3{\displaystyle \underset{ijklm}{}}N_{,i}N_{,jk}N_{,lm}[\delta ^3(𝐤_1)\delta ^3(𝐤_2)\delta \varphi _j\delta \varphi _k\delta \varphi _l\delta \varphi _m\delta _{𝐤_3}\varphi _i`$
$`+`$ $`\delta ^3(𝐤_1)\delta ^3(𝐤_3)\delta \varphi _j\delta \varphi _k\delta \varphi _l\delta \varphi _m\delta _{𝐤_2}\varphi _i`$
$`+`$ $`\delta ^3(𝐤_2)\delta ^3(𝐤_3)\delta \varphi _j\delta \varphi _k\delta \varphi _l\delta \varphi _m\delta _{𝐤_1}\varphi _i]`$
$`+`$ $`{\displaystyle \frac{1}{8}}{\displaystyle \underset{ijklmn}{}}N_{,ij}N_{,kl}N_{,mn}[`$
$`{\displaystyle }{\displaystyle \frac{d^3p}{(2\pi )^{3/2}}}\delta _{𝐤_1+𝐩}\varphi _i\delta _𝐩\varphi _j^{}{\displaystyle }{\displaystyle \frac{d^3p^{}}{(2\pi )^{3/2}}}\delta _{𝐤_2+𝐩^{}}\varphi _k\delta _𝐩^{}\varphi _l^{}{\displaystyle }{\displaystyle \frac{d^3p^{\prime \prime }}{(2\pi )^{3/2}}}\delta _{𝐤_3+𝐩^{\prime \prime }}\varphi _m\delta _{𝐩^{\prime \prime }}\varphi _n^{}]`$
$``$ $`{\displaystyle \frac{1}{8}}(2\pi )^{3/2}{\displaystyle \underset{ijklmn}{}}N_{,ij}N_{,kl}N_{,mn}[`$
$`\delta ^3(𝐤_1)\delta \varphi _i\delta \varphi _j{\displaystyle \frac{d^3p}{(2\pi )^{3/2}}\delta _{𝐤_2+𝐩}\varphi _k\delta _𝐩\varphi _l^{}\frac{d^3p^{}}{(2\pi )^{3/2}}\delta _{𝐤_3+𝐩^{}}\varphi _m\delta _𝐩^{}\varphi _n^{}}`$
$`+`$ $`\delta ^3(𝐤_2)\delta \varphi _k\delta \varphi _l{\displaystyle \frac{d^3p}{(2\pi )^{3/2}}\delta _{𝐤_1+𝐩}\varphi _i\delta _𝐩\varphi _j^{}\frac{d^3p^{}}{(2\pi )^{3/2}}\delta _{𝐤_3+𝐩^{}}\varphi _m\delta _𝐩^{}\varphi _n^{}}`$
$`+`$ $`\delta ^3(𝐤_3)\delta \varphi _m\delta \varphi _n{\displaystyle }{\displaystyle \frac{d^3p}{(2\pi )^{3/2}}}\delta _{𝐤_1+𝐩}\varphi _i\delta _𝐩\varphi _j^{}{\displaystyle }{\displaystyle \frac{d^3p^{}}{(2\pi )^{3/2}}}\delta _{𝐤_2+𝐩^{}}\varphi _k\delta _𝐩^{}\varphi _l^{}]`$
$`+`$ $`{\displaystyle \frac{1}{8}}(2\pi )^3{\displaystyle \underset{ijklmn}{}}N_{,ij}N_{,kl}N_{,mn}[`$
$`\delta ^3(𝐤_1)\delta ^3(𝐤_2)\delta \varphi _i\delta \varphi _j\delta \varphi _k\delta \varphi _l{\displaystyle \frac{d^3p}{(2\pi )^{3/2}}\delta _{𝐤_3+𝐩}\varphi _m\delta _𝐩\varphi _n^{}}`$
$`+`$ $`\delta ^3(𝐤_1)\delta ^3(𝐤_3)\delta \varphi _i\delta \varphi _j\delta \varphi _m\delta \varphi _n{\displaystyle \frac{d^3p}{(2\pi )^{3/2}}\delta _{𝐤_2+𝐩}\varphi _k\delta _𝐩\varphi _l^{}}`$
$`+`$ $`\delta ^3(𝐤_2)\delta ^3(𝐤_3)\delta \varphi _k\delta \varphi _l\delta \varphi _m\delta \varphi _n{\displaystyle }{\displaystyle \frac{d^3p}{(2\pi )^{3/2}}}\delta _{𝐤_1+𝐩}\varphi _i\delta _𝐩\varphi _j^{}]`$
$``$ $`{\displaystyle \frac{1}{8}}(2\pi )^{9/2}{\displaystyle \underset{ijklmn}{}}N_{,ij}N_{,kl}N_{,mn}\delta ^3(𝐤_1)\delta ^3(𝐤_2)\delta ^3(𝐤_3)\delta \varphi _i\delta \varphi _j\delta \varphi _k\delta \varphi _l\delta \varphi _m\delta \varphi _n.`$
The next step is to take the average of the latter expression. In doing so, we make use of the following decompositions :
$`\delta _{𝐤_1}\varphi _i\delta _{𝐤_2}\varphi _j\delta _{𝐤_3}\varphi _k\delta _{𝐤_4}\varphi _l=`$
$`\delta _{𝐤_1}\varphi _i\delta _{𝐤_2}\varphi _j\delta _{𝐤_3}\varphi _k\delta _{𝐤_4}\varphi _l+\delta _{𝐤_1}\varphi _i\delta _{𝐤_3}\varphi _k\delta _{𝐤_2}\varphi _j\delta _{𝐤_4}\varphi _l+\delta _{𝐤_1}\varphi _i\delta _{𝐤_4}\varphi _l\delta _{𝐤_2}\varphi _j\delta _{𝐤_3}\varphi _k,`$
(A.4)
$`\delta _{𝐤_1}\varphi _i\delta _{𝐤_2}\varphi _j\delta _{𝐤_3}\varphi _k\delta _{𝐤_4}\varphi _l\delta _{𝐤_5}\varphi _m\delta _{𝐤_6}\varphi _n=`$
$`\delta _{𝐤_1}\varphi _i\delta _{𝐤_2}\varphi _j\delta _{𝐤_3}\varphi _k\delta _{𝐤_4}\varphi _l\delta _{𝐤_5}\varphi _m\delta _{𝐤_6}\varphi _n+\delta _{𝐤_1}\varphi _i\delta _{𝐤_2}\varphi _j\delta _{𝐤_3}\varphi _k\delta _{𝐤_5}\varphi _m\delta _{𝐤_4}\varphi _l\delta _{𝐤_6}\varphi _n`$
$`+`$ $`\delta _{𝐤_1}\varphi _i\delta _{𝐤_2}\varphi _j\delta _{𝐤_3}\varphi _k\delta _{𝐤_6}\varphi _n\delta _{𝐤_4}\varphi _l\delta _{𝐤_5}\varphi _m+\delta _{𝐤_1}\varphi _i\delta _{𝐤_3}\varphi _k\delta _{𝐤_2}\varphi _j\delta _{𝐤_4}\varphi _l\delta _{𝐤_5}\varphi _m\delta _{𝐤_6}\varphi _n`$
$`+`$ $`\delta _{𝐤_1}\varphi _i\delta _{𝐤_3}\varphi _k\delta _{𝐤_2}\varphi _j\delta _{𝐤_5}\varphi _m\delta _{𝐤_4}\varphi _l\delta _{𝐤_6}\varphi _n+\delta _{𝐤_1}\varphi _i\delta _{𝐤_3}\varphi _k\delta _{𝐤_2}\varphi _j\delta _{𝐤_6}\varphi _n\delta _{𝐤_4}\varphi _l\delta _{𝐤_5}\varphi _m`$
$`+`$ $`\delta _{𝐤_1}\varphi _i\delta _{𝐤_4}\varphi _l\delta _{𝐤_2}\varphi _j\delta _{𝐤_3}\varphi _k\delta _{𝐤_5}\varphi _m\delta _{𝐤_6}\varphi _n+\delta _{𝐤_1}\varphi _i\delta _{𝐤_4}\varphi _l\delta _{𝐤_2}\varphi _j\delta _{𝐤_5}\varphi _m\delta _{𝐤_3}\varphi _k\delta _{𝐤_6}\varphi _n`$
$`+`$ $`\delta _{𝐤_1}\varphi _i\delta _{𝐤_4}\varphi _l\delta _{𝐤_2}\varphi _j\delta _{𝐤_6}\varphi _n\delta _{𝐤_3}\varphi _k\delta _{𝐤_5}\varphi _m+\delta _{𝐤_1}\varphi _i\delta _{𝐤_5}\varphi _m\delta _{𝐤_2}\varphi _j\delta _{𝐤_3}\varphi _k\delta _{𝐤_4}\varphi _l\delta _{𝐤_6}\varphi _n`$
$`+`$ $`\delta _{𝐤_1}\varphi _i\delta _{𝐤_5}\varphi _m\delta _{𝐤_2}\varphi _j\delta _{𝐤_4}\varphi _l\delta _{𝐤_3}\varphi _k\delta _{𝐤_6}\varphi _n+\delta _{𝐤_1}\varphi _i\delta _{𝐤_5}\varphi _m\delta _{𝐤_2}\varphi _j\delta _{𝐤_6}\varphi _n\delta _{𝐤_3}\varphi _k\delta _{𝐤_4}\varphi _l`$
$`+`$ $`\delta _{𝐤_1}\varphi _i\delta _{𝐤_6}\varphi _n\delta _{𝐤_2}\varphi _j\delta _{𝐤_3}\varphi _k\delta _{𝐤_4}\varphi _l\delta _{𝐤_5}\varphi _m+\delta _{𝐤_1}\varphi _i\delta _{𝐤_6}\varphi _n\delta _{𝐤_2}\varphi _j\delta _{𝐤_4}\varphi _l\delta _{𝐤_3}\varphi _k\delta _{𝐤_5}\varphi _m`$
$`+`$ $`\delta _{𝐤_1}\varphi _i\delta _{𝐤_6}\varphi _n\delta _{𝐤_2}\varphi _j\delta _{𝐤_5}\varphi _m\delta _{𝐤_3}\varphi _k\delta _{𝐤_4}\varphi _l,`$
where we have neglected connected $`n`$-point correlators with $`n>3`$, which nobody has calculated yet but that are presumably very small , and products of $`n`$-point correlators with $`m`$-point correlators, where $`n3`$ and $`m2`$, which we believe give a much smaller contribution than that coming from the three-point correlator .
From Eqs. (LABEL:barethreepoint), (A.4), and (LABEL:decom6), and after doing some algebra, we obtain the three point correlator function of the $`\zeta _𝐤`$ mode functions:
$`\zeta _{𝐤_1}\zeta _{𝐤_2}\zeta _{𝐤_3}={\displaystyle \underset{ijk}{}}N_{,i}N_{,j}N_{,k}\delta _{𝐤_1}\varphi _i\delta _{𝐤_2}\varphi _j\delta _{𝐤_3}\varphi _k`$
$`+{\displaystyle \underset{ij}{}}N_{,i}N_{,j}N_{,ij}(2\pi )^{3/2}\delta ^3(𝐤_1+𝐤_2+𝐤_3)\left({\displaystyle \frac{H_{}}{2\pi }}\right)^4\left[{\displaystyle \frac{4\pi ^4}{k_1^3k_2^3}}+\mathrm{cyclic}\mathrm{permutations}\right]`$
$`+{\displaystyle \frac{1}{2}}{\displaystyle \underset{ijk}{}}N_{,ij}N_{,jk}N_{,ki}(2\pi )^{3/2}\delta ^3(𝐤_1+𝐤_2+𝐤_3)\left({\displaystyle \frac{H_{}}{2\pi }}\right)^6\times `$
$`\times {\displaystyle }d^3p[{\displaystyle \frac{1}{|𝐤_1+𝐩|^3p^3|𝐤_3𝐩|^3}}+{\displaystyle \frac{1}{|𝐤_1+𝐩|^3p^3|𝐤_2𝐩|^3}}].`$ (A.6)
The integrals in the last line of the previous expression are calculated following the arguments presented in Ref. , giving as a result
$$d^3p\left[\frac{1}{|𝐤_1+𝐩|^3p^3|𝐤_3𝐩|^3}+\frac{1}{|𝐤_1+𝐩|^3p^3|𝐤_2𝐩|^3}\right]=8\pi \mathrm{ln}(kL)\left[\frac{1}{k_1^3k_2^3}+\mathrm{cyclic}\right],$$
(A.7)
where $`k^1`$ represents a typical scale under consideration and $`L`$ is the size of the region within which the stochastic properties are specified. That is why $`\mathrm{ln}(kL)`$ can be taken of order 1.
Finally, making use of the $`f_{\mathrm{NL}}`$ definition in Eqs. (6.17) and (6.18):
$$\zeta _{𝐤_1}\zeta _{𝐤_2}\zeta _{𝐤_3}=(2\pi )^{3/2}\delta ^3(𝐤_1+𝐤_2+𝐤_3)\left[\frac{6}{5}f_{\mathrm{NL}}(k_1,k_2,k_3)[P_\zeta (k_1)P_\zeta (k_2)+\mathrm{cyclic}]\right],$$
(A.8)
the results found in Eqs. (A.6) and (A.7), and the expression in Eq. (6.16) for the amplitude of the spectrum of $`\zeta `$ in terms of the first derivatives of $`N`$:
$$A_\zeta ^2=\left(\frac{H_{}}{2\pi }\right)^2\underset{i}{}N_{,i}^2,$$
(A.9)
we arrive to the desired expression
$`{\displaystyle \frac{3}{5}}f_{\mathrm{NL}}={\displaystyle \frac{_{ij}N_{,i}N_{,j}N_{,ij}}{2\left[_iN_{,i}^{}{}_{}{}^{2}\right]^2}}+\mathrm{ln}(kL){\displaystyle \frac{A_\zeta ^2}{2}}{\displaystyle \frac{_{ijk}N_{,ij}N_{,jk}N_{,ki}}{\left[_iN_{,i}^{}{}_{}{}^{2}\right]^3}},`$ (A.10)
with an additional contribution
$$\mathrm{\Delta }f_{\mathrm{NL}}=\frac{_{ijk}N_{,i}N_{,j}N_{,k}f_{ijk}(k_1,k_2,k_3)}{\left(_iN_{,i}^{}{}_{}{}^{2}\right)^{3/2}}|A_\zeta ^1|.$$
(A.11)
The latter expression comes from the first line in Eq. (A.6) containing the three-point correlator $`\delta _{𝐤_1}\varphi _i\delta _{𝐤_2}\varphi _j\delta _{𝐤_3}\varphi _k`$. This three-point correlator is normalised in such a way that the bispectra $`B_{ijk}`$ of the dimensionless perturbations $`(2\pi /H_{})\delta \varphi _i`$ is written in the same way as in Eqs. (6.17) and (6.18) with $`f_{\mathrm{NL}}(k_1,k_2,k_3)`$ being replaced by $`f_{ijk}(k_1,k_2,k_3)`$. |
warning/0507/quant-ph0507261.html | ar5iv | text | # Multiphoton Antiresonance and Quantum Activation in Driven Systems
## 1 Introduction
Much progress has been made recently in experimental studies of periodically modulated vibrational systems. Examples include optically bistable systems, electrons in Penning traps, Josephson junctions, and various nano- and micro-mechanical resonators Drummond80 -Chan05 . All these systems display bistability of forced vibrations. Because of thermal or externally applied noise, there occurs switching between coexisting stable vibrational states. The measured switching probabilities of noise-induced transitions are in a good agreement with the theoretical predictions.
In the present paper we are interested in the dynamics of a quantum oscillator. The generality of the oscillator as a model system and the current interest in quantum computing and coherent phenomena lead to two major questions: (i) does a resonantly driven oscillator display coherent quantum effects that would qualitatively differ from those in two-level systems, and (ii) in the presence of relaxation, what is the probability of switching between coexisting stable states due to quantum fluctuations? These two questions are addressed in the present paper, which is based on the results DS88 -Marthaler05 .
A weakly nonlinear oscillator is a multi-level quantum system with nearly equidistant energy levels $`E_n`$. Therefore a periodic force of frequency $`\omega _F`$ can be nearly resonant for many transitions at a time, i.e., $`\mathrm{}\omega _F`$ can be close to the interlevel distance $`E_{n+1}E_n`$ for many $`n`$. This leads to strong nonlinearity of the response even to comparatively weak resonant fields. A well-known quantum effect of the oscillator nonlinearity is the onset of Rabi oscillations due to resonant multiphoton transitions. Multiphoton Rabi oscillations occur when the spacing between remote energy levels $`n`$ and $`m`$ coincides with the energy of $`nm`$ photons, $`E_nE_m=(nm)\mathrm{}\omega _F`$ Bloembergen76 . The multiphoton transition amplitude is resonantly enhanced, because an $`mn`$ transition occurs via a sequence of virtual field-induced transitions $`kk+1`$ (with $`mkn1`$), all of which are almost resonant. An associated classical effect, in the presence of dissipation, is hysteresis of the amplitude of forced vibrations as function of the field amplitude $`A`$ and $`\omega _F`$.
In this paper (see also Dykman\_Fistul05 ) we show that multiphoton transitions in the oscillator are accompanied by a new effect, antiresonance of the response. When the frequency of the driving field adiabatically passes through a resonant value, the vibration amplitude displays a sharp minimum or maximum, depending on the initially occupied state. We argue that the antiresonance and the multiphoton Rabi oscillations can be observed in Josephson junctions.
If the frequency $`\omega _F`$ is close to twice the oscillator frequency, then $`\mathrm{}\omega _F`$ is close to $`E_{n+2}E_n`$ for many $`n`$ at a time. This leads to parametric resonance in the oscillator, in which it oscillates at frequency $`\omega _F/2`$ in response to the driving. Such oscillations are intrinsically bistable, because their phase can take on two values that differ by $`\pi `$.
We will be most interested in the semiclassical behavior of the oscillator, which, on the one hand, stretches all the way to the classical region, and on the other hand, works well for oscillators even deep in the quantum domain. In the semiclassical picture, resonant multiphoton transitions correspond to tunneling between Floquet states of the oscillator with equal quasienergies. \[The quasienergy $`\epsilon `$ gives the change of the wave function $`\psi (t)`$ when time is incremented by the modulation period $`\tau _F`$, $`\psi (t+\tau _F)=\mathrm{exp}(i\epsilon \tau _F/\mathrm{})\psi (t)`$\]. The occurrence of equal-quasienergy states is related to the bistability of forced vibrations of a classical oscillator.
Tunneling of a driven oscillator is a carefully studiedDyakonov86 example of dynamical tunneling Heller81 . The WKB analysis gives an important insight into the origin of the antiresonance, which goes beyond the perturbation theory in the driving field.
Dynamical tunneling also leads to transitions between coexisting metastable states of forced vibrations, which emerge in the presence of dissipation due to coupling to a thermal reservoir. In terms of quantum mechanics, dissipation is due to interlevel oscillator transitions with energy being transferred to (or from, for nonzero temperature) the reservoir. It turns out that dissipation may also lead to transitions between metastable states of forced vibrations, even for zero temperature DS88 .
For $`T=0`$ there occur only interlevel transitions where the oscillator energy goes to the reservoir (but the energy loss is compensated by the driving field, in the stationary regime). However, the quasi-energy may increase or decrease as a result of a coupling-to-reservoir induced transition, although with different probabilities. Therefore along with drift over quasienergy towards a metastable state, which results from more probable transition, there emerges diffusion away from this state as a sequence of less probable transitions. The diffusion may lead to activated-like escape. Activation in this case has purely quantum nature, and therefore we call it quantum activation.
## 2 The Models
The Hamiltonian of a nonlinear oscillator with mass $`M=1`$ has the form
$$H(t)=\frac{1}{2}p^2+\frac{1}{2}\omega _0^2q^2+\frac{1}{4}\gamma q^4+H_F(t).$$
(1)
We will consider two types of periodic modulation, $`H_F=H_F^{(r,p)}`$, which correspond to resonant and parametric driving,
$`H_F^{(r)}(t)=qA\mathrm{cos}(\omega _Ft),\delta \omega ^{(r)}\omega _F\omega _0\omega _0,`$ (2)
$`H_F^{(p)}={\displaystyle \frac{1}{2}}q^2F\mathrm{cos}(\omega _Ft),\delta \omega ^{(p)}{\displaystyle \frac{1}{2}}\omega _F\omega _0\omega _0`$
(in what follows we set $`\gamma >0`$).
It is convenient to analyze the dynamics in the rotating wave approximation by switching from the fast oscillating operators $`q,p`$ to slowly varying operators $`Q,P`$ using transformations
$$q=\alpha ^{(r)}(Q\mathrm{cos}\omega ^{(r)}t+P\mathrm{sin}\omega ^{(r)}t),p=\alpha ^{(r)}\omega ^{(r)}(Q\mathrm{sin}\omega ^{(r)}tP\mathrm{cos}\omega ^{(r)}t)$$
for resonant driving and
$$q=\alpha ^{(p)}(P\mathrm{cos}\omega ^{(p)}tQ\mathrm{sin}\omega ^{(p)}t),p=\alpha ^{(p)}\omega ^{(p)}(P\mathrm{sin}\omega ^{(p)}t+Q\mathrm{cos}\omega ^{(p)}t)$$
for parametric driving, with
$$\omega ^{(r)}=\omega _F,\alpha ^{(r)}=(8\omega _F\delta \omega ^{(r)}/3\gamma )^{1/2},$$
and
$$\omega ^{(p)}=\omega _F/2,\alpha ^{(p)}=(2F/3\gamma )^{1/2}.$$
The variables $`Q,P`$ are the appropriately scaled coordinate and momentum. The commutation relation for them has a simple form
$$[P,Q]=i\lambda ,\lambda =\mathrm{}\left(\omega ^{(r,p)}\right)^1\left(\alpha ^{(r,p)}\right)^2.$$
(3)
The parameter $`\lambda `$ plays the role of the effective Planck constant. We note that it is proportional to the oscillator nonlinearity $`\gamma `$ scaled by the comparatively small detuning of the field frequency, in the case of nearly resonant driving, or the comparatively small field amplitude, in the case of parametric driving.
The dynamics of $`Q,P`$ in the two cases is described by effective Hamiltonians
$`H^{(r)}=\omega _F\delta \omega ^{(r)}\left(\alpha ^{(r)}\right)^2g^{(r)},`$
$`H^{(p)}=(F/4)\left(\alpha ^{(p)}\right)^2g^{(p)}.`$ (4)
Their eigenvalues are equal to the quasienergies $`\epsilon _n`$ of the oscillator. The functions $`g^{(r,p)}`$ have the forms
$`g^{(r)}(P,Q)={\displaystyle \frac{1}{4}}(Q^2+P^21)^2\beta ^{1/2}Q,\beta ={\displaystyle \frac{3\gamma A^2}{32\omega _F^3\left(\delta \omega ^{(r)}\right)^3}},`$ (5)
$`g^{(p)}={\displaystyle \frac{1}{4}}(Q^2+P^2)^2+{\displaystyle \frac{1}{2}}(1\mu )P^2{\displaystyle \frac{1}{2}}(1+\mu )Q^2,\mu =2\omega _F\delta \omega ^{(p)}/F.`$ (6)
They are shown in Fig. 1. Each of them depends on one parameter. In the region of bistability of period one vibrations, $`0<\beta <4/27`$, the function $`g^{(r)}`$ has a shape of a tilted Mexican hat, with a maximum at the top of the central dome and a minimum at the lowest point of the rim. For a parametrically excited oscillator in the region $`1<\mu <1`$ the function $`g^{(p)}`$ has two symmetrical minima. These extrema of $`g^{(r,p)}`$ correspond to metastable states of the driven oscillator in the presence of weak dissipation. The saddle points of $`g^{(r,p)}`$ correspond to unstable states of forced vibrations.
## 3 Multiphoton antiresonance
We will start with the studies of the coherent response of the oscillator to a nearly resonant field. When the field amplitude $`A0`$, the eigenstates $`|n`$ of the Hamiltonian $`H^{(r)}`$ coincide with the Fock states of the oscillator, and the quasienergies are
$$\epsilon _n=\delta \omega ^{(r)}n+\frac{1}{2}Vn(n+1),V=\frac{3\mathrm{}\gamma }{4\omega _0^2}.$$
(7)
We keep only the lowest-order term in $`V`$, which corresponds to the weak nonlinearity approximation. In this approximation the energy of an $`N`$th oscillator state for $`A=0`$ is $`E_N=\mathrm{}\omega _0N+VN(N+1)/2`$. The $`N`$-photon resonance $`N\mathrm{}\omega _F=E_NE_0`$ occurs, in terms of $`\delta \omega ^{(r)}`$, for
$$\delta \omega ^{(r)}=\delta \omega _N^{(r)}=V(N+1)/2.$$
For the corresponding field frequency $`\omega _F`$ the quasienergies $`\epsilon _0`$ and $`\epsilon _N`$ are equal.
The field leads to mixing of the wave functions of resonating states and to level anticrossing. This anticrossing is clearly seen in the upper right panel of Fig. 2. The minimal splitting of the levels $`\epsilon _0`$ and $`\epsilon _N`$ is given by the multiphoton Rabi frequency $`\mathrm{\Omega }_R`$. For weak field it can be obtained by perturbation theory Bloembergen76 . To the lowest order in the field amplitude $`A`$ in the limit of large $`N`$
$`\mathrm{\Omega }_R=V(A/A_N)^NN^{5/4}(2\pi )^{3/4},A_N=(2\mathrm{}\omega _0)^{1/2}|V|N^{3/2}\mathrm{exp}(3/2)/2.`$ (8)
The Rabi frequency depends on $`N`$ exponentially, $`\mathrm{\Omega }_RA^N`$. This dependence works well numerically in the whole region $`A<A_N`$ Dykman\_Fistul05 .
Coherent response of the oscillator to the driving field is characterized by the expectation value of the coordinate $`q`$. If the oscillator is in an eigenstate $`|n`$ of the Hamiltonian $`H^{(r)}`$, this value has the form
$$q_n=(\mathrm{}/2\omega _0)^{1/2}a_ne^{i\omega _Ft}+\mathrm{c}.\mathrm{c}.,q_n=n|q|n.$$
(9)
To first order in the field, the reduced amplitude of forced vibrations in an $`n`$th state $`a_n`$ is
$$a_n=f\delta \omega ^{(r)}/\left((\delta \omega ^{(r)}Vn)[\delta \omega ^{(r)}V(n+1)]\right),f=(8\mathrm{}\omega _0)^{1/2}A.$$
(10)
Remarkably, for $`\delta \omega ^{(r)}=\delta \omega _N^{(r)}`$ the vibration amplitudes in the resonating states coincide with each other, $`a_{Nn}=a_n`$ for $`0n<N/2`$, see the left lower panel in Fig. 2.
Field-induced multiphoton mixing leads not only to splitting of the quasienergy levels, but also to repulsion of the vibration amplitudes. It can be calculated by diagonalizing the Hamiltonian $`H^{(r)}`$ and is shown in the right lower panel of Fig. 2 as a function of frequency detuning $`\delta \omega \delta \omega ^{(r)}`$. One of the involved resonating states is the ground state of the oscillator $`n=0`$ in the limit $`A0`$. The quantities plotted in Figs. 2(b) are susceptibilities, they are proportional to the ratio of the vibration amplitude to the modulation amplitude $`a_n/A`$.
The antiresonant splitting of the expectation values of the vibration amplitudes is by far the most interesting feature of Fig. 2. It occurs at the adiabatic passage of $`\delta \omega ^{(r)}`$ through resonance, where the system switches between the ground and excited states. In particular, the amplitude displays an antiresonant dip if the oscillator is mostly in the ground state for $`(\delta \omega ^{(r)}\delta \omega _N^{(r)})/V<1`$ or in the state $`N`$ for $`(\delta \omega ^{(r)}\delta \omega _N^{(r)})/V>1`$. The magnitude and sharpness of the dip are determined by $`\mathrm{\Omega }_R/V`$ and depend very strongly on the field and $`N`$. With decreasing $`\mathrm{\Omega }_R/V`$ the dip (and peak) start looking like cusps located at resonant frequency. We note that, in contrast to the case of energy levels, there is no reason for repulsion (anticrossing) of susceptibilities. In fact, as seen from Fig. 2 the susceptibilities do cross, although away from the resonant frequency. The effect of antiresonance is due purely to specific quantum interference Dykman\_Fistul05 .
The dip in the oscillator response has no analogue in two-level systems. There, for nearly resonant driving, the coherent response in the two adiabatic states differs only in sign. It displays a peak when the radiation frequency adiabatically passes through the transition frequency.
### 3.1 The WKB picture of the antiresonance
In the WKB approximation, Rabi oscillations correspond to tunneling between the states with nearly equal quasienergies. Such semiclassical states can be found from the Hamiltonian $`H^{(r)}`$ (2) by using the Bohr-Sommerfeld quantization condition applied to the mechanical action $`P𝑑Q`$ for trajectories $`g(Q,P)=`$ const, with $`\mathrm{}`$ replaced by $`\lambda `$, Eq. (3). It is seen from the left panel of Fig. 1 that, in a certain range of $`g`$, there are two types of trajectories with the same $`g`$, those on the internal dome and those on the external part of the Mexican hat $`g(Q,P)`$. If, as a result of the Bohr-Sommerfeld quantization, the quantized values of $`g`$ on the two parts of the surface $`g(Q,P)`$ coincide with each other, then there may occur resonant tunneling between the corresponding quantum states. The resulting tunneling splitting Dyakonov86 is the Rabi frequency.
Interestingly, one can show that the average value of the coordinate
$$Q(g)=\tau ^1(g)_0^{\tau (g)}Q(t)𝑑t$$
\[$`\tau (g)`$ is the period of oscillations for a given $`g`$\] is the same for the internal and external trajectories with the same $`g`$. This corresponds to the susceptibilities of the resonating quasienergy states being the same, in the neglect of tunneling-induced mixing of the states. We emphasize that the fact that the susceptibilities are equal is not a result of the perturbation theory in the field amplitude, as in the case of Eq. (10), they are equal in all orders of the perturbation theory in $`A`$ as long as tunneling is disregarded. Tunneling-induced state mixing leads to the antiresonance of the response Dykman\_Fistul05 .
## 4 Escape of a driven system: tunneling or quantum activation?
We will now briefly outline the new effects and unanswered questions that emerge when dissipation is taken into account DS88 ; Marthaler05 . We will assume that, even though dissipation is weak, the dissipation rate exceeds the tunneling rate. The problem of fairly general interest that will be addressed is switching between metastable states of forced vibrations of a quantum oscillator. We will consider the most interesting situation where there are many quasienergy states between the extrema of the quasienergy surface. In the case of escape of a particle from a potential well it corresponds to a well with many energy levels.
In systems in thermal equilibrium, the rate of tunneling decay of a metastable state for low temperatures is given by the probability of a tunneling transition from the ground state in a metastable potential well. In the case of a resonantly driven oscillator this corresponds to dynamical tunneling from the top of the dome of the quasienergy surface $`g^{(r)}`$ to the state on the external orbit with the same quasienergy, see the left panel of Fig. 1. The tunneling is shown schematically in the central panel of Fig. 3. For a parametrically driven oscillator the corresponding tunneling occurs between the minima of the surface $`g^{(p)}`$, as shown in the right panel of Fig. 3.
For higher temperatures, again in the case of equilibrium systems, one has to take into account tunneling from excited intrawell states. Escape may occur also via thermal activation over the potential barrier. One of these escape mechanisms dominates, depending on temperature Larkin85 . In the case of a driven oscillator tunneling from excited states corresponds to tunneling with quasienergies that differ from those at the extrema of $`g(Q,P)`$. In addition, there is a probability of activation over the quasienergy barrier. However, since the distribution over quasienergy is not of the Boltzmann form, it is not clear which of the escape mechanisms dominates at a given temperature of the bath.
For weak dissipation the distribution can be described by the balance equation for the occupations $`\rho _n`$ of quasienergy (Floquet) states $`|n`$,
$$\dot{\rho }_n=_mW_{nm}\rho _n+_mW_{mn}\rho _m.$$
(11)
The transitions probabilities $`W_{nm}`$ can be calculated as matrix elements of the operator that describes relaxation of the oscillator. The wave functions $`|n`$ can be found from the Bohr-Sommerfeld approximation disregarding tunneling. A standard WKB calculation allows one to express $`W_{nm}`$ in terms of the Fourier components of the coordinate and momentum $`Q_{mn}(g_n),P_{mn}(g_n)`$ for a given quasienergy $`g_n`$ \[the functions $`Q(t),P(t)`$ are periodic functions of time for a given $`g`$, with period $`\tau (g)`$\]. In the simple case of linear friction, which corresponds to relaxation transitions between nearest energy (not quasienergy) levels of the oscillator shown in the left panel of Fig. 3, $`W_{nm}`$ are simply quadratic in $`Q_{mn}(g_n),P_{mn}(g_n)`$. They exponentially decay with $`|nm|`$.
The probabilities $`W_{nm}`$ are organized so that it is more likely for a system to make a transition toward the value of $`g`$ in the metastable state rather than away from it. This is why the state is metastable. However, in contrast to systems in equilibrium, the probabilities $`W_{nm}`$ do not satisfy the condition $`W_{nm}=W_{mn}\mathrm{exp}[(g_ng_m)/kT]`$. Even for $`T0`$ there is a nonzero probability to make a transition in the direction opposite to the metastable state. This is a consequence of the fact that the Floquet states $`|n`$ are linear combinations of the Fock states of the oscillator. Therefore, even where all transitions between the Fock states go in one direction in energy, as in the left panel of Fig. 3, transitions between the Floquet states go in different directions in quasienergy, although with different probabilities.
Transitions in the “wrong” direction lead to diffusion over quasienergy away from the metastable state. Their immediate consequence is a finite width of the stationary distribution over quasienergy even for $`T0`$. Another closely related consequence is a nonzero probability to reach quasienergy of the saddle of the functions $`g^{(r,p)}`$ starting from a metastable state. The saddle of $`g^{(r,p)}`$ is similar to the top of a potential barrier for a particle in a metastable potential. The logarithm of the probability of reaching the saddle gives the activation energy of escape as a result of diffusion over quasienergy. We call this process quantum activation, since it occurs even for $`T=0`$.
For resonantly and parametrically driven oscillators the transition rates $`W_{nm}`$ can be calculated explicitly using the fact that the classical trajectories in $`(Q,P)`$-variables can be expressed in terms of the Jacobi elliptic functions DS88 ; Marthaler05 . We have compared the activation energies of escape with the tunneling exponents. When the activation energy is smaller than the tunneling exponent in the absolute value, escape occurs via quasienergy diffusion. We found that, both for a resonantly and a parametrically excited oscillator, escape from metastable states occurs via activation, not tunneling. This holds true for all parameter values where a nonlinear oscillator has coexisting stable states. We found an unusual behavior of the distribution for an underdamped parametrically driven oscillator for $`T=0`$. For some quasienergy the distribution displays a sharp decrease, and at the same time the tunneling rate goes to zero.
The physical origin of the fact that escape occurs via activation, not dynamical tunneling, remains not fully understood, the existing arguments are formal DS88 . Apparently, there must be a crossover from activation to tunneling when the system goes to equilibrium, but the models that we have discussed are strongly nonequilibrium, the very presence of coexisting metastable states is due to periodic driving.
In conclusion, we have shown that a simple system, a driven nonlinear oscillator, displays unusual quantum coherent phenomena and unusual switching behavior. The considered effects have no analogue in two-level systems and are qualitatively different from what has been known about switching in thermal equilibrium systems. They are not only of fundamental interest, but are also important for many applications, in particular in sensing and quantum measurements.
This research was supported in part by the Institute for Quantum Sciences at Michigan State University and by the NSF through grant ITR-0085922. |
warning/0507/math0507479.html | ar5iv | text | # Bijective proofs of shifted tableau and alternating sign matrix identities
## 1 Introduction
The expression
$$\underset{1i<jn}{}(x_i+y_j)$$
(1.1)
appears in a number of contexts in symmetric function theory. Given $`𝐲=(y_1,y_2,\mathrm{},y_n)`$ and $`𝐱=(x_1,x_2,\mathrm{},x_n)`$, when $`𝐲=𝐱`$, the expression (1.1) is just the Vandermonde determinant that appears in Weyl’s denominator formula
$$det(x_i^{nj})=\underset{1i<jn}{}(x_ix_j).$$
(1.2)
For $`𝐲=\lambda 𝐱`$, the expression (1.1) becomes the subject of the $`\lambda `$–determinant formula of Robbins and Rumsey \[RR86\]:
$$\underset{1i<jn}{}(x_i+\lambda x_j)=\underset{A𝒜_n}{}\lambda ^{SE(A)}(1+\lambda )^{NS(A)}\underset{i=1}{\overset{n}{}}x_i^{NE_i(A)+SE_i(A)+NS_i(A)},$$
(1.3)
where the exponents are various parameters associated with alternating sign matrices and defined in Section 3. Robbins and Runsey use different notation but do include the square ice concepts, although they use different terminology. Bressoud \[B01\] asked for a combinatorial proof of (1.3). This was provided by Chapman \[C01\] who generalised it to:
$$\underset{1i<jn}{}(x_i+y_j)=\underset{A𝒜_n}{}\underset{i=1}{\overset{n}{}}x_i^{NE_i(A)}y_i^{SE_i(A)}(x_i+y_i)^{NS_i(A)}.$$
(1.4)
For $`𝐲=t𝐱`$, there is also the $`t`$-deformation of a Weyl denominator formula for $`gl(n)`$ due to Tokuyama \[T88\]:
$$\underset{i=1}{\overset{n}{}}x_i\underset{1i<jn}{}(x_i+tx_j)s_\lambda (𝐱)=\underset{ST𝒮𝒯^\mu (n)}{}t^{\mathrm{hgt}(ST)}(1+t)^{\mathrm{str}(ST)n}𝐱^{\mathrm{wgt}(ST)},$$
(1.5)
where the sum is over semistandard shifted tableaux $`ST`$ of shape $`\mu =\lambda +\delta `$ with $`\delta =(n,n1,\mathrm{},1)`$, and where $`\mathrm{hgt},\mathrm{str},`$ and $`\mathrm{wgt}`$ are parameters associated with semistandard shifted tableaux. They are defined in Section 2. Note also that $`s_\lambda (𝐱)`$, the Schur function specified by the partition $`\lambda `$, with a suitable interpretation of the indeterminates $`x_i`$ for $`i=1,2,\mathrm{},n`$, is the character of an irreducible representation of $`gl(n)`$ whose highest weight is specified by the partition $`\lambda `$.
Here we present a general identity that unifies the results (1.2)-(1.5). This identity is our first main result and is expressed in terms of a certain generalisation of Schur $`P`$–functions and also in terms of the corresponding generalisation of Schur $`Q`$–functions. These $`P`$ and $`Q`$ functions are defined combinatorially in Section 2.
###### Proposition 1.1
Let $`\mu =\lambda +\delta `$ be a strict partition of length $`\mathrm{}(\mu )=n`$, with $`\lambda `$ a partition of length $`\mathrm{}(\lambda )n`$ and $`\delta =(n,n1,\mathrm{},1)`$. In addition, let $`𝐱=(x_1,x_2,\mathrm{},x_n)`$ and $`𝐲=(y_1,y_2,\mathrm{},y_n)`$. Then
$$\begin{array}{ccc}\hfill P_\mu (𝐱/𝐲)& =& s_\lambda (𝐱)_{i=1}^nx_i_{1i<jn}(x_i+y_j),\hfill \\ & & \\ \hfill Q_\mu (𝐱/𝐲)& =& s_\lambda (𝐱)_{1ijn}(x_i+y_j),\hfill \end{array}$$
(1.6)
where $`P_\mu (𝐱/𝐲)`$ and $`Q_\mu (𝐱/𝐲)`$ are as defined in Section 2.
A bijective proof of this Proposition, along with a number of corollaries, is provided in Section 3. The case $`𝐱=𝐲`$ is an example of Macdonald \[M95\] (Ex2, p259, 2nd Edition). The case $`𝐲=t𝐱=(tx_1,tx_2,\mathrm{},tx_n)`$ is equivalent to a Weyl denominator deformation theorem due to Tokuyama \[T88\] for the Lie algebra $`gl(n)`$ and given a combinatorial proof by Okada \[O90\]. The case $`\lambda =0`$ is equivalent to an alternating sign matrix (ASM) identity attributed to Robbins and Rumsey \[RR86\] and proved combinatorially by Chapman \[C01\]. The connection with ASMs is provided in Section 5, in which both (1.3) and (1.4) are shown to be simple corollaries of Proposition 1.1.
It should be pointed out that the above Proposition is restricted to the case of a strict partition $`\mu `$ of length $`\mathrm{}(\mu )=n`$. Although a similar result applying to the case $`\mathrm{}(\mu )=n1`$ may be obtained from the above by dividing both sides by $`s_{1^n}(𝐱)=x_1x_2\mathrm{}x_n`$, there is no similar product formula for either $`P_\mu (𝐱/𝐲)`$ or $`Q_\mu (𝐱/𝐲)`$ in the case $`\mathrm{}(\mu )<n1`$.
On the other hand, the above results may all be generalised to the case of certain symplectic tableaux. The analogue of (1.1) in this setting turns out to be
$$\underset{1i<jn}{}(x_i+t^2x_i^1+y_j+t^2y_j^1).$$
(1.7)
When $`𝐲=𝐱`$ and $`t=1`$ the expression (1.7) is a factor of the determinant that appears in Weyl’s denominator formula for $`sp(2n)`$,
$$det(x_i^{nj+1}x_i^{n+j1})=\underset{i=1}{\overset{n}{}}(x_ix_i^1)\underset{1i<jn}{}(x_i+x_i^1x_jx_j^1).$$
(1.8)
More generally, for $`𝐲=t𝐱`$ we have \[HK02\]
$$\begin{array}{c}_{i=1}^n(x_i+tx_i^1)_{1i<jn}(x_i+t^2x_i^1+tx_j+tx_j^1)sp_\lambda (𝐱;t)\hfill \\ \\ =_{ST𝒮𝒯^\mu (n,\overline{n})}t^{\mathrm{var}(ST)+\mathrm{bar}(ST)}(1+t)^{\mathrm{str}(ST)n}𝐱^{\mathrm{wgt}(ST)},\hfill \end{array}$$
(1.9)
where the sum is over semistandard shifted symplectic tableaux of shape $`\mu =\lambda +\delta `$ with $`\delta =(n,n1,\mathrm{},1)`$, and where $`\mathrm{var}`$, $`\mathrm{bar}`$, $`\mathrm{str}`$ and $`\mathrm{wgt}`$ are defined in Section 2. Here $`sp_\lambda (𝐱;t)`$, once again with a suitable interpretation of the indeterminates $`x_i`$ for $`i=1,2,\mathrm{},n`$, is a $`t`$-deformation of the character $`sp_\lambda (𝐱)`$ of the irreducible representation of the Lie algebra $`sp(2n)`$ whose highest weight is specified by the partition $`\lambda `$.
Our second main result then takes the form
###### Proposition 1.2
Let $`\mu =\lambda +\delta `$ be a strict partition of length $`\mathrm{}(\mu )=n`$, with $`\lambda `$ a partition of length $`\mathrm{}(\lambda )n`$ and $`\delta =(n,n1,\mathrm{},1)`$. In addition, let $`𝐱=(x_1,x_2,\mathrm{},x_n)`$, $`𝐲=(y_1,y_2,\mathrm{},y_n)`$, $`\overline{𝐱}=(\overline{x}_1,\overline{x}_2,\mathrm{},\overline{x}_n)`$ and $`\overline{𝐲}=(\overline{y}_1,\overline{y}_2,\mathrm{},\overline{y}_n)`$, with $`\overline{x}_i=x_i^1`$ and $`\overline{y}_i=y_i^1`$ for $`i=1,2,\mathrm{},n`$. Then
$$Q_\mu (𝐱/𝐲;t)=sp_\lambda (𝐱;t)\underset{1ijn}{}(x_i+t^2\overline{x}_i+y_j+t^2\overline{y}_j),$$
(1.10)
where $`Q_\mu (𝐱/𝐲;t)`$ is defined in Section 2.
Here $`Q(𝐱/𝐲;t)`$ is a generalisation of $`Q(𝐱/𝐲)`$ that associates factors of $`t^2`$ with the barred components of $`\overline{𝐱}`$ and $`\overline{𝐲}`$. Although a similar generalisation $`P(𝐱/𝐲;t)`$ of $`P(𝐱/𝐲)`$ exists, as we shall see, there does not exist a corresponding identity for $`P(𝐱/𝐲;t)`$ that is analogous to the identity (1.10) for $`Q(𝐱/𝐲;t)`$.
Our paper is arranged as follows. In Section 2 the necessary background is introduced regarding both the relevant semistandard, shifted and primed tableaux, and the various $`P`$ and $`Q`$ functions and characters of $`gl(n)`$ and $`sp(2n)`$. For the $`gl(n)`$ case, Section 3 opens in Section 4.1 with a formal statement of the combinatorial identity upon which the first main result, Proposition 1.1, is based. A bijective proof of this identity is then provided. A detailed example appears in Section 3.2. In Section 3.3 a number of corollaries are gathered together.
Turning to the $`sp(2n)`$ case, the combinatorial identity necessary to establish the second main result, Proposition 1.2, is stated, bijectively proved and exemplified in Section 4. Once again two corollaries are supplied in Section 4.3, including a proof of Proposition 1.2.
Finally, in Section 5 the connection is made with alternating sign matrices and U-turn alternating sign matrices in the $`gl(n)`$ and $`sp(2n)`$ cases, respectively.
## 2 Background
### 2.1 $`gl(n)`$ tableaux
Let $`\lambda =(\lambda _1,\lambda _2,\mathrm{},\lambda _p)`$ with $`\lambda _1\lambda _2\mathrm{}\lambda _p>0`$ be a partition of weight $`|\lambda |=\lambda _1+\lambda _2+\mathrm{}+\lambda _p`$ and length $`\mathrm{}(\lambda )=p`$, where each $`\lambda _i`$ is a positive integer for all $`i=1,2,\mathrm{},p`$. Then $`\lambda `$ defines a Young diagram $`F^\lambda `$ consisting of $`p`$ rows of boxes of lengths $`\lambda _1,\lambda _2\mathrm{},\lambda _p`$ left-adjusted to a vertical line.
A partition $`\mu =(\mu _1,\mu _2,\mathrm{},\mu _q)`$ of length $`\mathrm{}(\mu )=q`$ is said to be a strict partition if all the parts of $`\mu `$ are distinct; that is, $`\mu _1>\mu _2>\mathrm{}>\mu _q>0`$. A strict partition $`\mu `$ defines a shifted Young diagram $`SF^\mu `$ consisting of $`q`$ rows of boxes of lengths $`\mu _1,\mu _2,\mathrm{},\mu _q`$ left-adjusted this time to a diagonal line.
For any partition $`\lambda `$ of length $`\mathrm{}(\lambda )n`$ let $`𝒯^\lambda (n)`$ be the set of all semistandard tableaux $`T`$ obtained by numbering all the boxes of $`F^\lambda `$ with entries taken from the set $`\{1,2,\mathrm{},n\}`$, subject to the usual total ordering $`1<2<\mathrm{}<n`$. The numbering must be such that the entries are:
| T1 | weakly increasing across each row from left to right; |
| --- | --- |
| T2 | strictly increasing down each column from top to bottom. |
The weight of the tableau $`T`$ is given by $`\mathrm{wgt}(T)=(w_1,w_2,\mathrm{},w_n)`$, where $`w_k`$ is the number of times $`k`$ appears in $`T`$ for $`k=1,2,\mathrm{},n`$. For example in the case $`n=6`$, $`\lambda =(3,3,2,1,1)`$ we have
$$T=\begin{array}{ccccccc}\hfill & 1\hfill & \hfill & 2\hfill & \hfill & 3\hfill & \hfill \\ \hfill & 3\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill \\ \hfill & 4\hfill & \hfill & 6\hfill & \hfill & & \\ \hfill & 5\hfill & \hfill & & \\ \hfill & 6\hfill & \hfill \\ & & \end{array}𝒯^{33211}(6)\text{with}\mathrm{wgt}(T)=(1,1,2,1,3,2).$$
(2.11)
By the same token, for any strict partition $`\mu `$ of length $`\mathrm{}(\mu )n`$, let $`𝒮𝒯^\mu (n)`$ be the set of all semistandard shifted tableaux $`ST`$ obtained by numbering all the boxes of $`SF^\mu `$ with entries taken from the set $`\{1,2,\mathrm{},n\}`$, subject to the total ordering $`1<2<\mathrm{}<n`$. The numbering must be such that the entries are:
| ST1 | weakly increasing across each row from left to right; |
| --- | --- |
| ST2 | weakly increasing down each column from top to bottom; |
| ST3 | strictly increasing down each diagonal from top-left to bottom-right. |
The weight of the tableau $`ST`$ is again given by $`\mathrm{wgt}(ST)=(w_1,w_2,\mathrm{},w_n)`$, where $`w_k`$ is the number of times $`k`$ appears in $`ST`$ for $`k=1,2,\mathrm{},n`$.
The rules ST1-ST3 serve to exclude any $`2\times 2`$ blocks of boxes all containing the same entry, and as a result, each $`ST𝒮𝒯^\mu (n)`$ consists of a sequence of ribbon strips of boxes containing identical entries. Any given ribbon strip may consist of a number of disjoint connected components. Let $`\mathrm{str}(ST)`$ denote the total number of disjoint connected components of all the ribbon strips. Let $`\mathrm{hgt}(ST)`$ be the height of the tableaux, defined $`\mathrm{hgt}(ST)=_{k=1}^n(\mathrm{row}_k(ST)\mathrm{con}_k(ST))`$, where $`\mathrm{row}_k(ST)`$ is the number of rows of $`S`$ containing an entry $`k`$, and $`\mathrm{con}_k(ST)`$ is the number of connected components of the ribbon strip of $`ST`$ consisting of all the entries $`k`$.
By way of illustration, consider the case $`n=6`$, $`\mu =(9,8,6,4,3,1)`$ and the semistandard shifted tableau:
$$ST=\begin{array}{ccccccccccccccccccc}\hfill & 1\hfill & \hfill & 1\hfill & \hfill & 1\hfill & \hfill & 2\hfill & \hfill & 2\hfill & \hfill & 2\hfill & \hfill & 3\hfill & \hfill & 3\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & 2\hfill & \hfill & 2\hfill & \hfill & 3\hfill & \hfill & 3\hfill & \hfill & 4\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill & 6\hfill & \hfill \\ & & & & \hfill & 3\hfill & \hfill & 3\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & 5\hfill & \hfill & 6\hfill & \hfill & & \\ & & & & & & \hfill & 4\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill & & \\ & & & & & & & & \hfill & 5\hfill & \hfill & 6\hfill & \hfill & 6\hfill & \hfill \\ & & & & & & & & & & \hfill & 6\hfill & \hfill & & \end{array}𝒮𝒯^{986431}(6)\text{with}\begin{array}{c}\mathrm{wgt}(ST)=(3,5,6,4,8,5)\hfill \\ \\ \mathrm{str}(ST)=12,\mathrm{hgt}(ST)=6.\hfill \end{array}$$
(2.12)
Refining this construct, for any strict partition $`\mu `$ with $`\mathrm{}(\mu )n`$, let $`𝒫𝒮𝒯^\mu (n)`$ be the set of all primed, or marked, semistandard shifted tableaux $`PST`$ obtained by numbering all the boxes of $`SF^\mu `$ with entries taken from the set $`\{1^{},1,2^{},2,\mathrm{},n^{},n\}`$, subject to the total ordering $`1^{}<1<2^{}<2<\mathrm{}<n^{}<n`$. The numbering must be such that the entries are:
| PST1 | weakly increasing across each row from left to right; |
| --- | --- |
| PST2 | weakly increasing down each column from top to bottom; |
| PST3 | with no two identical unprimed entries in any column; |
| PST4 | with no two identical primed entries in any row; |
| PST5 | with no primed entries on the main diagonal. |
The weight of the tableau $`PST`$ is then defined to be $`\mathrm{wgt}(PST)=(u_1,u_2,\mathrm{},u_n/v_1,v_2,\mathrm{},v_n)`$, where $`u_k`$ and $`v_k`$ are the number of times $`k`$ and $`k^{}`$, respectively, appear in $`PST`$ for $`k=1,2,\mathrm{},n`$.
The passage from $`𝒮𝒯^\mu (n)`$ to $`𝒫𝒮𝒯^\mu (n)`$ is effected merely by adding primes to the entries of each $`ST𝒮𝒯^\mu (n)`$ in all possible ways that are consistent with PST1-5 to give some $`PST𝒫𝒮𝒯^\mu (n)`$. The only entries for which any choice is possible are those in the lower left hand box at the beginning of each connected component of a ribbon strip. Thereafter, in that connected component of the ribbon strip, entries in the boxes of its horizontal portions are unprimed and those in the boxes of its vertical portions are primed. It should be noted that all the boxes on the main diagonal are necessarily at the lower left hand end of a connected component of a ribbon strip, but their entries remain unprimed by virtue of PST5.
To illustrate this let us assign primes to those entries of $`ST`$ in (2.12) for which it is essential (that is, for every entry lying immediately above the same entry) and some of those for which it is optional (those entries off the main diagonal that are at the start of any continuous strip of equal entries). This gives, for example,
$$PST=\begin{array}{ccccccccccccccccccc}\hfill & 1\hfill & \hfill & 1\hfill & \hfill & 1\hfill & \hfill & 2^{}\hfill & \hfill & 2\hfill & \hfill & 2\hfill & \hfill & 3\hfill & \hfill & 3\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & 2\hfill & \hfill & 2\hfill & \hfill & 3^{}\hfill & \hfill & 3\hfill & \hfill & 4^{}\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 6^{}\hfill & \hfill \\ & & & & \hfill & 3\hfill & \hfill & 3\hfill & \hfill & 4^{}\hfill & \hfill & 4\hfill & \hfill & 5^{}\hfill & \hfill & 6\hfill & \hfill & & \\ & & & & & & \hfill & 4\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill & & \\ & & & & & & & & \hfill & 5\hfill & \hfill & 6^{}\hfill & \hfill & 6\hfill & \hfill \\ & & & & & & & & & & \hfill & 6\hfill & \hfill & & \end{array}𝒫𝒮𝒯^{986431}(6)\text{with}\mathrm{wgt}(PST)=(3,4,5,2,5,3/0,1,1,2,3,2).$$
(2.13)
We may replace PST1-4 by identical conditions QST1-4, but discard PST5. This serves to define corresponding primed shifted tableaux $`QST𝒬𝒮𝒯^\mu (n)`$ that now involve both primed and unprimed entries on the main diagonal.
Finally, in this $`gl(n)`$ context, for fixed positive integer $`n`$, let $`\delta =(n,n1,\mathrm{},1)`$ and let $`𝒫𝒟^\delta (n)`$ be the set of all primed shifted tableaux, $`PD`$, of shape $`\delta `$, obtained by numbering the boxes of $`SF^\delta `$ with entries taken from the set $`\{1^{},1,2^{},2,\mathrm{},n^{},n\}`$ in such a way that
| PD1 | each unprimed entry $`k`$ appears only in the $`k`$th row; |
| --- | --- |
| PD2 | each primed entry $`k^{}`$ appears only in the $`k`$th column; |
| PD3 | there are no primed entries on the main diagonal. |
The weight of the tableau $`PD`$ is defined by $`\mathrm{wgt}(PD)=(𝐮/𝐯)=(u_1,u_2,\mathrm{},u_n/v_1,v_2,\mathrm{},v_n)`$, where $`u_k`$ and $`v_k`$ are the numbers of times $`k`$ and $`k^{}`$, respectively, appear in $`PD`$ for $`k=1,2,\mathrm{},n`$. Typically for $`n=6`$ we have
$$PD=\begin{array}{ccccccccccccc}\hfill & 1\hfill & \hfill & 2^{}\hfill & \hfill & 1\hfill & \hfill & 4^{}\hfill & \hfill & 5^{}\hfill & \hfill & 6^{}\hfill & \hfill \\ & & \hfill & 2\hfill & \hfill & 3^{}\hfill & \hfill & 2\hfill & \hfill & 5^{}\hfill & \hfill & 2\hfill & \hfill \\ & & & & \hfill & 3\hfill & \hfill & 4^{}\hfill & \hfill & 3\hfill & \hfill & 3\hfill & \hfill \\ & & & & & & \hfill & 4\hfill & \hfill & 5^{}\hfill & \hfill & 6^{}\hfill & \hfill \\ & & & & & & & & \hfill & 5\hfill & \hfill & 5\hfill & \hfill \\ & & & & & & & & & & \hfill & 6\hfill & \hfill \end{array}𝒫𝒟^{654321}(6)\text{with}\mathrm{wgt}(PD)=(2,3,3,1,2,1/0,1,1,2,3,2).$$
(2.14)
Since the $`i`$th entry on the main diagonal is always $`i`$ and for $`i<j`$ the entry in the $`(i,j)`$th position is either $`i`$ or $`j^{}`$, it is clear that
$$\underset{PD𝒫𝒟^\delta (n)}{}(𝐱/𝐲)^{\mathrm{wgt}(PD)}=\underset{i=1}{\overset{n}{}}x_i\underset{1i<jn}{}(x_i+y_j).$$
(2.15)
By way of a small variation of the above, if we replace PD1-2 by identical conditions QD1-2 and discard the condition PD3, the corresponding set $`𝒬𝒟^\delta (n)`$ of primed shifted tableaux $`QD`$ differs from $`𝒫𝒟^\mu (n)`$ only in allowing primed entries on the main diagonal. It follows that
$$\underset{QD𝒬𝒟^\delta (n)}{}(𝐱/𝐲)^{\mathrm{wgt}(QD)}=\underset{1ijn}{}(x_i+y_j).$$
(2.16)
These formulae (2.15) and (2.16) offer a combinatorial interpretation of factors appearing in the expansions (2.26) of Proposition 1.1. This will be exploited later in Section 3.
### 2.2 $`sp(2n)`$ tableaux
In order to establish a similar approach to Proposition 1.2 it is necessary to extend our already copious list of tableaux to encompass certain tableaux associated with the symplectic algebra $`sp(2n)`$. As before it is helpful to start with definitions of the various types of tableaux, both shifted and unshifted.
For any partition $`\lambda `$ of length $`\mathrm{}(\lambda )n`$, let $`𝒯^\lambda (n,\overline{n})`$ be the set of all semistandard symplectic tableaux $`T`$ obtained by numbering all the boxes of $`F^\lambda `$ with entries from the set $`\{\overline{1},1,\overline{2},2,\mathrm{},\overline{n},n\}`$, subject to the usual total ordering $`\overline{1}<1<\overline{2}<2<\mathrm{}\overline{n}<n`$. The entries are:
| T1 | weakly increasing across each row from left to right; |
| --- | --- |
| T2 | strictly increasing down each column from top to bottom. |
| T$`\overline{3}`$ | $`k`$ or $`\overline{k}`$ may appear no lower than the $`k`$th row. |
The weight of the symplectic tableau $`T`$ is given by $`\mathrm{wgt}(T)=(𝐰)=(w_1,w_2,\mathrm{},w_n)`$, with $`w_k=n_kn_{\overline{k}}`$ where $`n_k`$ and $`n_{\overline{k}}`$ are the number of times $`k`$ and $`\overline{k}`$, respectively, appear in $`T`$ for $`k=1,2,\mathrm{},n`$. The parameter $`bar(T)`$ is equal to the number of barred entries in the tableau. For example in the case $`n=5`$, $`\lambda =(4,3,3)`$ we have
$$T=\begin{array}{ccccccccc}\hfill & \overline{1}\hfill & \hfill & 1\hfill & \hfill & \overline{2}\hfill & \hfill & 4\hfill & \hfill \\ \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & \overline{4}\hfill & \hfill & & \\ \hfill & \overline{4}\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill \end{array}𝒯^{433}(5,5)\text{with}\begin{array}{c}\mathrm{wgt}(T)=(0,1,1,0,0)\hfill \\ \\ \mathrm{bar}(T)=5.\hfill \end{array}$$
(2.17)
For any strict partition $`\mu `$ of length $`\mathrm{}(\mu )n`$, let $`𝒮𝒯^\mu (n,\overline{n})`$ be the set of all semistandard shifted symplectic tableaux $`ST`$ obtained by numbering all the boxes of $`SF^\mu `$ with entries taken from the set $`\{\overline{1},1,\overline{2},2,\mathrm{},\overline{n},n\}`$, subject to the total ordering $`\overline{1}<1<\overline{2}<2<\mathrm{}<\overline{n}<n`$. The numbering must be such that the entries are:
| ST1 | weakly increasing across each row from left to right; |
| --- | --- |
| ST2 | weakly increasing down each column from top to bottom; |
| ST3 | strictly increasing down each diagonal from top-left to bottom-right; |
| ST$`\overline{4}`$ | with $`d_k\{k,\overline{k}\}`$, where $`d_k`$ is the $`k`$th entry on the main diagonal. |
The weight of the shifted symplectic tableau $`ST`$ is given by $`\mathrm{wgt}(ST)=(w_1,w_2,\mathrm{},w_n)`$, with $`w_k=n_kn_{\overline{k}}`$ where $`n_k`$ and $`n_{\overline{k}}`$ are the number of times $`k`$ and $`\overline{k}`$, respectively, appear in $`ST`$ for $`k=1,2,\mathrm{},n`$. Once again it is convenient, following \[HK05\], to introduce $`\mathrm{str}(ST)`$ as the total number of disjoint connected components of all ribbon strips of $`ST`$, and $`\mathrm{var}(ST)=_{k=1}^n(\mathrm{row}_k(ST)\mathrm{con}_k(ST)+\mathrm{col}_{\overline{k}}(ST)\mathrm{con}_{\overline{k}}(ST))`$, where $`\mathrm{row}_k(ST)`$ is the number of rows of $`ST`$ containing an entry $`k`$, $`\mathrm{col}_{\overline{k}}(ST)`$ is the number of columns containing an entry $`\overline{k}`$, while $`\mathrm{con}_k(ST)`$ and $`\mathrm{con}_{\overline{k}}(ST)`$ are the number of connected components of the ribbon strips of $`ST`$ consisting of all the entries $`k`$ and $`\overline{k}`$, respectively, and $`\mathrm{bar}(ST)`$ is equal to the total number of barred entries.
Typically, for $`n=5`$ and $`\mu =(9,7,6,2,1)`$ we have
$$ST=\begin{array}{ccccccccccccccccccc}\hfill & \overline{1}\hfill & \hfill & 1\hfill & \hfill & \overline{2}\hfill & \hfill & 2\hfill & \hfill & \overline{3}\hfill & \hfill & \overline{3}\hfill & \hfill & \overline{4}\hfill & \hfill & 4\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & \overline{2}\hfill & \hfill & \overline{2}\hfill & \hfill & 2\hfill & \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & \overline{4}\hfill & \hfill & 4\hfill & \hfill & & \\ & & & & \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill \\ & & & & & & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & & & & & \\ & & & & & & & & \hfill & \overline{5}\hfill & \hfill \end{array}𝒮𝒯^{97621}(5,\overline{5})\text{with}\begin{array}{c}\mathrm{wgt}(ST)=(0,1,0,4,0)\hfill \\ \\ \mathrm{bar}(ST)=11,\mathrm{str}(ST)=12,\mathrm{var}(ST)=7.\hfill \end{array}$$
(2.18)
Refining this construct, for any strict partition $`\mu `$ with $`\mathrm{}(\mu )n`$, let $`𝒫𝒮𝒯^\mu (n,\overline{n})`$ be the set of all primed semistandard shifted symplectic tableaux $`PST`$ obtained by numbering all the boxes of $`SF^\mu `$ with entries taken from the set $`\{\overline{1}^{},\overline{1},1^{},1,\overline{2}^{},\overline{2},2^{},2,\mathrm{},\overline{n},\overline{n},n^{},n\}`$, subject to the total ordering
$$\overline{1}^{}<\overline{1}<1^{}<1<\overline{2}^{}<\overline{2}<2^{}<2<\mathrm{}<\overline{n}^{}<\overline{n}<n^{}<n.$$
(2.19)
The numbering must be such that the entries are:
| PST1 | weakly increasing across each row from left to right; |
| --- | --- |
| PST2 | weakly increasing down each column from top to bottom; |
| PST3 | with no two identical unprimed entries in any column; |
| PST4 | with no two identical primed entries in any row; |
| PST$`\overline{5}`$ | with $`d_k\{\overline{k},k\}`$, where $`d_k`$ is the $`k`$th entry on the main diagonal. |
The weight of the tableau $`PST`$ is then defined to be $`\mathrm{wgt}(PST)=(𝐮/𝐯)`$ with $`𝐮=(u_1,u_2,\mathrm{},u_n)`$ and $`𝐯=(v_1,v_2,\mathrm{},v_n)`$, where $`u_k=n_kn_{\overline{k}}`$ and $`v_k=n_k^{}n_{\overline{k}^{}}`$, with $`n_k`$, $`n_{\overline{k}}`$, $`n_k^{}`$ and $`n_{\overline{k}^{}}`$ are the number of times $`k`$, $`\overline{k}`$, $`k^{}`$ and $`\overline{k}^{}`$, respectively, appear in $`PST`$ for $`k=1,2,\mathrm{},n`$. In addition, let $`\mathrm{bar}(PST)`$ be the total number of barred entries in $`PST`$.
If we now replace PST1-4 by identical conditions QST1-4 and replace PST$`\overline{5}`$ by:
| QST$`\overline{5}`$ | with $`d_k\{\overline{k}^{},\overline{k},k^{},k\}`$, where $`d_k`$ is the $`k`$th entry on the main diagonal. |
| --- | --- |
Then once again the corresponding primed shifted tableaux $`QST𝒬𝒮𝒯^\mu (n,\overline{n})`$ now have primes allowed on the main diagonal.
Typically, for $`n=5`$ and $`\mu =(9,7,6,2,1)`$ we have
$$QST=\begin{array}{ccccccccccccccccccc}\hfill & \overline{1}\hfill & \hfill & 1\hfill & \hfill & \overline{2}^{}\hfill & \hfill & 2^{}\hfill & \hfill & \overline{3}^{}\hfill & \hfill & \overline{3}\hfill & \hfill & \overline{4}^{}\hfill & \hfill & 4^{}\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & \overline{2}^{}\hfill & \hfill & \overline{2}\hfill & \hfill & 2\hfill & \hfill & 3\hfill & \hfill & \overline{4}^{}\hfill & \hfill & \overline{4}\hfill & \hfill & 4^{}\hfill & \hfill & & \\ & & & & \hfill & 3^{}\hfill & \hfill & \overline{4}^{}\hfill & \hfill & 4^{}\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill \\ & & & & & & \hfill & 4^{}\hfill & \hfill & 4\hfill & \hfill & & & & & & \\ & & & & & & & & \hfill & \overline{5}^{}\hfill & \hfill \end{array}𝒬𝒮𝒯^{97621}(5,\overline{5})\text{with}\begin{array}{c}\mathrm{wgt}(QST)=(0,0,0,3,1/0,1,0,1,1),\hfill \\ \\ \mathrm{bar}(QST)=11.\hfill \end{array}$$
(2.20)
To complete our set of $`sp(2n)`$ tableaux, for fixed positive integer $`n`$, let $`\delta =(n,n1,\mathrm{},1)`$ and let $`𝒫𝒟^\delta (n,\overline{n})`$ be the set of all primed shifted tableaux, $`PD`$, of shape $`\delta `$, obtained by numbering the boxes of $`SF^\delta `$ with entries taken from the set $`\{\overline{1}^{},\overline{1},1^{},1,\overline{2}^{},\overline{2},2^{},2,\mathrm{},\overline{n}^{},\overline{n},n^{},n\}`$ in such a way that
| PD$`\overline{1}`$ | each unprimed entry $`k`$ or $`\overline{k}`$ appears only in the $`k`$th row; |
| --- | --- |
| PD$`\overline{2}`$ | each primed entry $`k^{}`$ or $`\overline{k}^{}`$ appears only in the $`k`$th column; |
| PD3 | there are no primed entries on the main diagonal. |
The weight of the tableau $`PD`$ is defined by $`\mathrm{wgt}(PD)=(𝐮/𝐯)`$ with $`𝐮=(u_1,u_2,\mathrm{},u_n)`$ and $`𝐯=(v_1,v_2,\mathrm{},v_n)`$, where $`u_k=n_kn_{\overline{k}}`$ and $`v_k=n_k^{}n_{\overline{k}^{}}`$, with $`n_k`$, $`n_{\overline{k}}`$, $`n_k^{}`$ and $`n_{\overline{k}^{}}`$ are the number of times $`k`$, $`\overline{k}`$, $`k^{}`$ and $`\overline{k}^{}`$, respectively, appear in $`PD`$ for $`k=1,2,\mathrm{},n`$. In addition let $`\mathrm{bar}(PD)`$ be the total number of barred entries in $`PD`$.
With this notation, since the entry in the $`i`$th position on the main diagonal is either $`i`$ or $`\overline{i}`$ while for $`i<j`$ the entry in the $`(i,j)`$th position is either $`i`$, $`\overline{i}`$, $`j^{}`$ or $`\overline{j}^{}`$, it is clear that
$$\underset{PD𝒫𝒟^\delta (n,\overline{n})}{}t^{2\mathrm{b}\mathrm{a}\mathrm{r}(PD)}(𝐱/𝐲)^{\mathrm{wgt}(PD)}=\underset{i=1}{\overset{n}{}}(x_i+t^2\overline{x}_i)\underset{1i<jn}{}(x_i+t^2\overline{x}_i+y_j+t^2\overline{y}_j).$$
(2.21)
By way of a small variation of the above, if we replace PD1-2 by identical conditions QD1-2 and discard the condition PD3, the corresponding set $`𝒬𝒟^\delta (n)`$ of primed shifted tableaux $`QD`$ differs from $`𝒫𝒟^\mu (n)`$ only in allowing primed entries on the main diagonal.
Typically for $`n=5`$ we have
$$QD=\begin{array}{ccccccccccc}\hfill & \overline{1}\hfill & \hfill & 1\hfill & \hfill & \overline{3}^{}\hfill & \hfill & 4^{}\hfill & \hfill & \overline{1}\hfill & \hfill \\ & & \hfill & 2\hfill & \hfill & 2\hfill & \hfill & 2\hfill & \hfill & \overline{2}\hfill & \hfill \\ & & & & \hfill & \overline{3}^{}\hfill & \hfill & 3\hfill & \hfill & \overline{3}\hfill & \hfill \\ & & & & & & \hfill & 4^{}\hfill & \hfill & 4\hfill & \hfill \\ & & & & & & & & \hfill & \overline{5}^{}\hfill & \hfill \end{array}𝒬𝒟^{54321}(5,\overline{5})\text{with}\begin{array}{c}\mathrm{wgt}(QD)=(1,2,0,1,0/0,0,2,2,1),\hfill \\ \\ \mathrm{bar}(QD)=7.\hfill \end{array}$$
(2.22)
It follows from our definition of $`𝒬𝒟(n,\overline{n})`$ that
$$\underset{QD𝒬𝒟^\delta (n,\overline{n})}{}t^{2\mathrm{b}\mathrm{a}\mathrm{r}(QD)}(𝐱/𝐲)^{\mathrm{wgt}(QD)}=\underset{1ijn}{}(x_i+t^2\overline{x}_i+y_j+t^2\overline{y}_j).$$
(2.23)
These formulae (2.21) and (2.23) have been introduced so as to offer a combinatorial interpretation of factors appearing in the expansions (2.30) of Proposition 1.2. This will be exploited later in Section 3.
### 2.3 Schur’s $`P`$ and $`Q`$ functions and their generalisations
Let $`𝐱=(x_1,x_2,\mathrm{},x_n)`$ be a vector of $`n`$ indeterminates and let $`𝐰=(w_1,w_2,\mathrm{},w_n)`$ be a vector of $`n`$ non-negative integers. Then $`𝐱^𝐰=x_1^{w_1}x_2^{w_2}\mathrm{}x_n^{w_n}`$. With this notation it is well known that each partition $`\lambda `$ of length $`\mathrm{}(\lambda )n`$ specifies a Schur function $`s_\lambda (𝐱)`$ with combinatorial definition:
$$s_\lambda (𝐱)=\underset{T𝒯^\lambda (n)}{}𝐱^{\mathrm{wgt}(T)}$$
(2.24)
Similarly, each strict partition $`\mu `$ of length $`\mathrm{}(\mu )n`$ specifies a Schur $`P`$-function and a Schur $`Q`$-function whose combinatorial definitions take the form:
$$\begin{array}{ccc}\hfill P_\mu (𝐱)& =& _{ST𝒮𝒯^\mu (n)}2^{\mathrm{str}(ST)\mathrm{}(\mu )}𝐱^{\mathrm{wgt}(ST)};\hfill \\ & & \\ \hfill Q_\mu (𝐱)& =& _{ST𝒮𝒯^\mu (n)}2^{\mathrm{str}(ST)}𝐱^{\mathrm{wgt}(ST)}.\hfill \end{array}$$
(2.25)
Now let $`𝐳=(𝐱/𝐲)=(x_1,x_2,\mathrm{},x_n/y_1,y_2,\mathrm{},y_n)`$, where $`𝐱`$ and $`𝐲`$ are two vectors of $`n`$ indeterminates, and let $`𝐰=(𝐮/𝐯)=(u_1,u_2,\mathrm{},u_n/v_1,v_2,\mathrm{},v_n)`$ where $`𝐮`$ and $`𝐯`$ are two vectors of $`n`$ non-negative integers. Then let $`𝐳^𝐰=(𝐱/𝐲)^{(𝐮/𝐯)}=𝐱^𝐮𝐲^𝐯=x_1^{u_1}\mathrm{}x_n^{u_n}y_1^{v_1}\mathrm{}y_n^{v_n}`$. With this notation each strict partition $`\mu `$ of length $`\mathrm{}(\mu )n`$ serves to specify generalised Schur $`P`$ and $`Q`$-functions defined by:
$$\begin{array}{ccc}\hfill P_\mu (𝐱/𝐲)& =& _{PST𝒫𝒮𝒯^\mu (n)}(𝐱/𝐲)^{\mathrm{wgt}(PST)};\hfill \\ & & \\ \hfill Q_\mu (𝐱/𝐲)& =& _{QST𝒬𝒮𝒯^\mu (n)}(𝐱/𝐲)^{\mathrm{wgt}(QST)}.\hfill \end{array}$$
(2.26)
Since the maps back from $`PST𝒫𝒮𝒯^\mu (n)`$ and from $`QST𝒬𝒮𝒯^\mu (n)`$ to some $`ST𝒮𝒯^\mu (n)`$ are effected merely by deleting primes, and there are no primes on the main diagonal in the case of $`PST`$, it follows that
$$Q_\mu (𝐱)=2^{\mathrm{}(\mu )}P_\mu (𝐱)\text{with}P_\mu (𝐱)=P_\mu (𝐱/𝐱)\text{and}Q_\mu (𝐱)=Q_\mu (𝐱/𝐱)$$
(2.27)
It might be noted that $`s_\lambda (𝐱)`$, $`P_\lambda (𝐱)`$ and $`Q_\lambda (𝐱)`$ are nothing other than the specialisations $`P_\lambda (𝐱;1)`$, $`P_\mu (𝐱;1)`$ and $`Q_\mu (𝐱;1)`$, respectively, of the Hall-Littlewood functions $`P_\mu (𝐱;t)`$ and $`Q_\mu (𝐱;t)`$.
Turning to the symplectic case, it is well known that each partition $`\lambda `$ of length $`\mathrm{}(\lambda )n`$ specifies an irreducible representation of $`sp(2n)`$ whose character $`sp_\lambda (𝐱)`$ may be given a combinatorial definition:
$$sp_\lambda (𝐱)=\underset{T𝒯^\lambda (n,\overline{n})}{}𝐱^{\mathrm{wgt}(T)}.$$
(2.28)
This may be $`t`$-deformed to give
$$sp_\lambda (𝐱;t)=\underset{T𝒯^\lambda (n,\overline{n})}{}t^{2\mathrm{b}\mathrm{a}\mathrm{r}(T)}𝐱^{\mathrm{wgt}(T)}.$$
(2.29)
In the case of a strict partition $`\mu `$ of length $`\mathrm{}(\mu )=n`$ the required generalisations of Schur $`P`$ and $`Q`$ functions take the form:
$$\begin{array}{ccc}\hfill P_\mu (𝐱/𝐲;t)& =& _{PST𝒫𝒮𝒯^\mu (n,\overline{n})}t^{2\mathrm{b}\mathrm{a}\mathrm{r}(PST)}(𝐱/𝐲)^{\mathrm{wgt}(PST)};\hfill \\ & & \\ \hfill Q_\mu (𝐱/𝐲;t)& =& _{QST𝒬𝒮𝒯^\mu (n,\overline{n})}t^{2\mathrm{b}\mathrm{a}\mathrm{r}(QST)}(𝐱/𝐲)^{\mathrm{wgt}(QST)}.\hfill \end{array}$$
(2.30)
## 3 The $`gl(n)`$ bijection
### 3.1 Main Result
The generalisations of the combinatorial definitions of $`P_\mu (𝐱)`$ and $`Q_\mu (𝐱)`$ to $`P_\mu (𝐱/𝐲)`$ and $`Q_\mu (𝐱/𝐲)`$, respectively, together with those of $`s_\lambda (𝐱)`$ and the product factors appearing in (2.26), allow us to establish the validity of Proposition 1.1 by first proving the following:
###### Theorem 3.1
Let $`\mu =\lambda +\delta `$ be a strict partition of length $`\mathrm{}(\mu )=n`$, with $`\lambda `$ a partition of length $`\mathrm{}(\lambda )n`$ and $`\delta =(n,n1,\mathrm{},1)`$. There exists a weight preserving, bijective map $`\mathrm{\Theta }`$ from $`𝒫𝒮𝒯^\mu (n)`$ to $`(𝒫𝒟^\delta (n),𝒯^\lambda (n))`$ and from $`𝒬𝒮𝒯^\mu (n)`$ to $`(𝒬𝒟^\delta (n),𝒯^\lambda (n))`$ such that for all $`PST𝒫𝒮𝒯^\mu (n)`$ and for all $`QST𝒬𝒮𝒯^\mu (n)`$
$$\begin{array}{ccc}\hfill \mathrm{\Theta }:PST& & (PD,T)\text{with}\mathrm{wgt}(PST)=\mathrm{wgt}(PD)+\mathrm{wgt}(T).\hfill \\ & & \\ \hfill \mathrm{\Theta }:QST& & (QD,T)\text{with}\mathrm{wgt}(QST)=\mathrm{wgt}(QD)+\mathrm{wgt}(T).\hfill \end{array}$$
(3.31)
with $`PD𝒫𝒟^\delta (n)`$, $`QD𝒬𝒟^\delta (n)`$ and $`T𝒯^\lambda (n)`$.
Proof
We choose to tackle the $`PST`$ case first with the aim of describing a candidate map $`\mathrm{\Theta }`$ and showing that it is both weight preserving and bijective.
The technique is to apply the jeu de taquin (\[S87\], \[W84\], \[SS89\], \[M95\],\[HH92\]) to the primed entries $`k^{}`$ of $`PST`$ taken in turn starting with any $`1^{}`$s (actually there are none), then any $`2^{}`$s (at most one), then any $`3^{}`$s (at most two) and so on. If for fixed $`k`$ there is more than one $`k^{}`$ in $`PST`$ then these are dealt with in turn from top to bottom before a final rearrangement is made of the entries in the $`k`$th column. The map $`\mathrm{\Theta }`$ is thus expressible in the form $`\mathrm{\Theta }=\theta _n^{}\mathrm{}\theta _2^{}\theta _1^{}`$.
We start by describing the map $`\theta _k^{}`$. This involves sliding each $`k^{}`$ in the north-west direction by a sequence of interchanges with either its northern or western neighbour until it reaches a position in the $`k`$th column either in the topmost row, or immediately below another $`k^{}`$, or immediately below some unprimed entry $`i`$ in the $`i`$th row.
This amounts to playing jeu de taquin, treating $`k^{}`$ to be strictly less than all the unprimed entries. At every stage all the unprimed entries must satisfy the semistandardness conditions T1 and T2; that is, they should be weakly increasing across rows and strictly increasing down columns. It is this that ensures that each move made by $`k^{}`$ is uniquely determined. Consider first the situation illustrated by the tableau $`T_0`$ in (3.32) with $`k^{}`$ not yet in the $`k`$th column, nor in the top row. This is to be thought of as the subtableau surrounding a particular $`k^{}`$ awaiting its next move. For the time being we assume that $`b,d`$ are unprimed, while $`a,c,e,f,g,h`$ may be primed, or unprimed, or even absent if $`k^{}`$ is at or near the southern or eastern edge of the complete diagram. However, all the unprimed entries amongst $`a,b,\mathrm{},h`$ must, by hypothesis, satisfy the semistandardness conditions T1 and T2.
Now for the jeu de taquin rules that define the map $`\theta _k^{}`$. If $`db`$ then $`k^{}`$ is to be interchanged with $`b`$ and if $`d>b`$ then $`k^{}`$ is to be interchanged with $`d`$. In the first case $`k^{}`$ moves north and the resulting tableau $`T_N`$ satisfies both $`T1`$ and $`T2`$ since $`dbc<e`$, while in the second, $`k^{}`$ moves west and the resulting tableau $`T_W`$ satisfies both $`T1`$ and $`T2`$ since $`b<d<fg`$.
$$\theta _k^{}:T_0=\begin{array}{ccccccc}\hfill & a\hfill & \hfill & b\hfill & \hfill & c\hfill & \hfill \\ \hfill & d\hfill & \hfill & k^{}\hfill & \hfill & e\hfill & \hfill \\ \hfill & f\hfill & \hfill & g\hfill & \hfill & h\hfill & \hfill \end{array}\{\begin{array}{cc}T_N=\begin{array}{ccccccc}\hfill & a\hfill & \hfill & k^{}\hfill & \hfill & c\hfill & \hfill \\ \hfill & d\hfill & \hfill & b\hfill & \hfill & e\hfill & \hfill \\ \hfill & f\hfill & \hfill & g\hfill & \hfill & h\hfill & \hfill \end{array}& \text{if}bd;\\ & \\ T_W=\begin{array}{ccccccc}\hfill & a\hfill & \hfill & b\hfill & \hfill & c\hfill & \hfill \\ \hfill & k^{}\hfill & \hfill & d\hfill & \hfill & e\hfill & \hfill \\ \hfill & f\hfill & \hfill & g\hfill & \hfill & h\hfill & \hfill \end{array}& \text{if}b<d.\end{array}$$
(3.32)
If $`k^{}`$ is already in the topmost row, so that the row $`abc`$ is absent then $`\theta _k^{}`$ acts on $`T_0`$ as follows:
$$\theta _k^{}:T_0=\begin{array}{ccccccc}\hfill & d\hfill & \hfill & k^{}\hfill & \hfill & e\hfill & \hfill \\ \hfill & f\hfill & \hfill & g\hfill & \hfill & h\hfill & \hfill \end{array}T_W=\begin{array}{ccccccc}\hfill & k^{}\hfill & \hfill & d\hfill & \hfill & e\hfill & \hfill \\ \hfill & f\hfill & \hfill & g\hfill & \hfill & h\hfill & \hfill \end{array}$$
(3.33)
Once again the unprimed entries in $`T_W`$ satisfy both T1 and T2, since we still have $`d<fg`$.
On the other hand, if $`k^{}`$ is already in the $`k`$th column, so that the column $`adf`$ is absent, the map $`\theta _k^{}`$ leaves $`T_0`$ unaltered, that is $`k^{}`$ has reached its final resting place, unless $`k^{}`$ lies in the $`i`$th row with an unprimed entry $`bi`$ immediately above it. In such a case $`\theta _k^{}`$ acts on $`T_0`$ as shown below
$$\theta _k^{}:\begin{array}{c}T_0=\begin{array}{ccccc}\hfill & b\hfill & \hfill & c\hfill & \hfill \\ \hfill & k^{}\hfill & \hfill & e\hfill & \hfill \\ \hfill & g\hfill & \hfill & h\hfill & \hfill \end{array}\\ k^{}\text{ in }i\text{th row and }k\text{th column}\\ bi\end{array}T_N=\begin{array}{ccccc}\hfill & k^{}\hfill & \hfill & c\hfill & \hfill \\ \hfill & b\hfill & \hfill & e\hfill & \hfill \\ \hfill & g\hfill & \hfill & h\hfill & \hfill \end{array}$$
(3.34)
Yet again, the unprimed entries of $`T_N`$ satisfy both T1 and T2, since we still have $`bc<e`$.
Now we return to the two possibilities that we had previously set aside, namely $`b=k^{}`$ or $`d=k^{}`$. The first of these cannot occur in the tableau $`T_0`$ of (3.32), since in such a case the uppermost $`b=k^{}`$ would have been moved either north or west before attempting to move the central $`k^{}`$. In the case of the tableau $`T_0`$ of (3.34), as we have alreay pointed out, if $`b=k^{}`$ then no further move of the $`k^{}`$ below $`b=k^{}`$ is required.
It follows that the only possible impediment to the movement of $`k^{}`$ in a north-westerly direction until it actually reaches the $`k`$th column, is the existence of another $`k^{}`$ to its immediate left, that is in $`T_0`$ of (3.32) or (3.34), we have $`d=k^{}`$. That this cannot occur is a corollary of the fact that the path followed by $`k^{}`$ always remains column by column below (that is strictly south of) the path followed by any preceding $`k^{}`$. To see this consider $`k^{}`$ arriving, as shown below in the diagram on the left of (3.35), at a position due south of an entry $`b`$ which itself lies on the path of the preceding $`k^{}`$.
$$\theta _k^{}:\begin{array}{cccccccccccccccc}\hfill & 𝐩\hfill & \hfill & \mathrm{}\hfill & \hfill & 𝐚\hfill & \hfill & 𝐛\hfill & \hfill & & \mathrm{}\hfill & & & \hfill & 𝐪\hfill & \hfill \\ \hfill & r\hfill & \hfill & \mathrm{}\hfill & \hfill & c\hfill & \hfill & k^{}\hfill & \hfill & & & & & & & \end{array}\begin{array}{c}b<c\\ \end{array}\begin{array}{cccccccccccccccc}\hfill & 𝐩\hfill & \hfill & \mathrm{}\hfill & \hfill & 𝐚\hfill & \hfill & 𝐛\hfill & \hfill & & \mathrm{}\hfill & & & \hfill & 𝐪\hfill & \hfill \\ \hfill & r\hfill & \hfill & \mathrm{}\hfill & \hfill & k^{}\hfill & \hfill & c\hfill & \hfill & & & & & & & \end{array}$$
(3.35)
If this path of the preceding $`k^{}`$ moves north from $`b`$, then there is no problem since the $`k^{}`$ can follow the same path north or move west without violating the strictly south condition. On the other hand the path of the preceding $`k^{}`$ may move west along the indicated boldface track from $`q`$ to $`p`$. In doing so, it must at one stage have displaced $`b`$ from its original position at the site of $`a`$, immediately above $`c`$ and satisfying the T2 condition $`b<c`$. This condition then ensures that the $`k^{}`$ must itself move west as shown in (3.35). It therefore stays south of the path of the preceding $`k^{}`$ that passes through the position of $`a`$. This implies that the path of $`k^{}`$ must always stay strictly south of the path of the preceding $`k^{}`$, thereby excluding the possibility $`d=k^{}`$ in both (3.32) and (3.33). This ensures that each $`k^{}`$ will eventually reach the $`k`$th column by means of a sequence of moves of type (3.32)-(3.34).
Following the action of $`\theta _k^{}`$ the unprimed entry $`k`$ on the main diagonal of $`PST`$ remains fixed, and all $`k^{}`$s are in the $`k`$th column along with distinct unprimed entries $`j`$ with $`1jk`$. If $`k^{}`$ appears in the $`i`$th row, then $`i`$ cannot appear above it, since $`k^{}`$ would then move north as in (3.34), and cannot appear below it, since it would then be to the right of a diagonal entry greater than $`i`$ and thus violate the weakly increasing condition T1. It follows that the unprimed entries $`j`$ in the $`k`$th column do not include the row numbers of $`k^{}`$. Since they are distinct and $`1jk`$, they must include all the other row numbers, and be arranged in strictly increasing order in accordance with T2. This means that each unprimed entry in the $`k`$th column lies in its own row. Since the primed entries in this column are all $`k^{}`$s all the entries in the $`k`$th column satisfy PD1-3.
Iterating this procedure for all $`k=1,2,\mathrm{},n`$ results in all primed entries being moved to the first $`k`$ columns of $`SF^\mu `$ along with some unprimed entries, collectively satisfying PD1-3 in this region of shape $`SF^\delta `$, and leaving only unprimed entries, all satisfying T1-2, in the right hand region of shape $`F^\lambda `$. That is, the result of applying $`\mathrm{\Theta }`$ to $`PST𝒫𝒮𝒯^\mu `$ is a semistandard tableau $`T𝒯^\lambda (n)`$ of shape $`\lambda `$ juxtaposed to a primed tableau $`PD𝒫𝒟^\delta (n)`$. This map is necessarily weight preserving since every individual step is a simple interchange which does not alter the number of $`k`$s or $`k^{}`$s for any $`k`$.
To show that this map $`\mathrm{\Theta }`$ is bijective it should be noted that each step may be reversed. One starts by juxtaposing an arbitrary pair of tableaux $`PD𝒫𝒟^\delta (n)`$ and $`T𝒯^\lambda (n)`$ to create a tableaux of shape $`SF^\mu `$ with $`\mu =\lambda +\delta `$. Then for each $`k`$ taken in turn from $`n`$ to $`n1`$ down to $`1`$ one applies $`\theta _k^{}^1`$ to all the primed entries $`k^{}`$; that is to say one reverses the action of $`\theta _k^{}`$ by playing jeu de taquin in the reverse direction with primed entries $`k^{}`$ treated in turn from bottom to top, moving each one in a south easterly direction with $`k^{}`$ now assumed to be larger than $`i`$ for $`i=1,\mathrm{},k1`$ but less than $`j`$ for $`j=k,k+1,\mathrm{},n`$ with the semistandardness conditions T1 and T2 applying to all unprimed entries at all times. For example, in the following diagram this leads unambiguously from $`T_0`$ to $`T_E`$ if $`e<g`$ and from $`T_0`$ to $`T_S`$ if $`eg`$, with all unprimed entries satisfying the semistandardness conditions T1 and T2 since in addition $`bc<e<g`$ in $`T_E`$ and $`d<fge`$ in $`T_S`$.
θk1:T0=abcdkefgh{TE=abcdekfghife<g;TS=
abcdgefkhifeg.\theta^{-1}_{k^{\prime}}:\ \ T_{0}=\ {\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle a&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle b&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle c&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle d&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle k^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle e&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle f&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle g&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle h&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr}}}\quad\ \ \longrightarrow\ \ \quad\left\{\begin{array}[]{ll}T_{E}=\ {\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle a&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle b&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle c&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle d&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle e&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle k^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle f&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle g&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle h&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr}}}&\quad\hbox{if}\quad e<g;\\
\\
T_{S}=\ {\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle a&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle b&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle c&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle d&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle g&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle e&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle f&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle k^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle h&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr}}}&\quad\hbox{if}\quad e\geq g.\end{array}\right.}}}}}}}}}}}}}}}}}}}}}}}}}}} (3.36)
If $`k^{}`$ is already in the lowest row or rightmost column, the following diagrams illustrate the allowed moves of $`k^{}`$ east and south respectively.
$$\theta _k^{}^1:T_0=\begin{array}{ccccccc}\hfill & a\hfill & \hfill & b\hfill & \hfill & c\hfill & \hfill \\ \hfill & d\hfill & \hfill & k^{}\hfill & \hfill & e\hfill & \hfill \end{array}T_E=\begin{array}{ccccccc}\hfill & a\hfill & \hfill & b\hfill & \hfill & c\hfill & \hfill \\ \hfill & d\hfill & \hfill & e\hfill & \hfill & k^{}\hfill & \hfill \end{array}$$
(3.37)
and
$$\theta _k^{}^1:T_0=\begin{array}{ccccc}\hfill & a\hfill & \hfill & b\hfill & \hfill \\ \hfill & d\hfill & \hfill & k^{}\hfill & \hfill \\ \hfill & f\hfill & \hfill & g\hfill & \hfill \end{array}T_S=\begin{array}{ccccc}\hfill & a\hfill & \hfill & b\hfill & \hfill \\ \hfill & d\hfill & \hfill & g\hfill & \hfill \\ \hfill & f\hfill & \hfill & k^{}\hfill & \hfill \end{array}$$
(3.38)
Once again since the unprimed entries of $`T_0`$ satisfy both T1 and T2, these rules are also satisfied by both $`T_E`$ and $`T_S`$ since we still have $`bc<e`$ and $`d<fg`$.
More importantly, returning to the original jeu de taquin moves illustrated in (3.32), this reversed jeu de taquin is such that if the conditions T1 and T2 are satisfied by the entries in $`T_0`$, then the reverse process leads directly from $`T_W`$ to $`T_0`$ since $`d<f`$ and from $`T_N`$ to $`T_0`$ since $`bc`$. Thus the original steps along each of the $`k^{}`$ paths are retraced precisely. The same is true of the maps (3.33) and (3.34).
The only task remaining is to show that the endpoints of these retraced paths results in an element $`PST`$ of $`𝒫𝒮𝒯^\mu `$. If $`T_0`$ is such that $`k^{}`$ has reached its endpoint then neither $`e`$ nor $`g`$ is $`<k^{}`$. If either $`e`$ or $`g`$ is $`k`$ then this poses no problem. If $`gk^{}`$ there is again no problem since the rules PST1-5 allow two (or more) $`k^{}`$s in the same column. It is only the case $`e=k^{}`$ that produces a violation, in this case of PST4. Fortunately this case is excluded by the following argument analogous to that which led to the strictly south property of the original jeu de taquin. Now we require a strictly north property. The argument goes as follows. The fact that the strictly north property applies to the reverse jeu de taquin follows from a consideration of the following diagram in which a $`k^{}`$ south westerly path meets a preceding west–east path, indicated by means of boldface entries, passing from $`p`$ to $`q`$ through the positions of $`a`$ and $`b`$. The existence of the latter requires that $`a`$ must initially have been immediately south of $`c`$, so that $`c<a`$. This in turn implies that $`k^{}`$ moves eastwards staying strictly north of the preceding path, and ensuring that no two $`k^{}`$s can appear in the same row.
$$\theta _k^1:\begin{array}{ccccccccccccc}& & & & \hfill & k^{}\hfill & \hfill & c\hfill & \hfill & \mathrm{}\hfill & \hfill & s\hfill & \hfill \\ \hfill & 𝐩\hfill & \hfill & \mathrm{}\hfill & \hfill & 𝐚\hfill & \hfill & 𝐛\hfill & \hfill & \mathrm{}\hfill & \hfill & 𝐪\hfill & \hfill \end{array}\begin{array}{c}a>c\\ \end{array}\begin{array}{ccccccccccccc}& & & & \hfill & c\hfill & \hfill & k^{}\hfill & \hfill & \mathrm{}\hfill & \hfill & s\hfill & \hfill \\ \hfill & 𝐩\hfill & \hfill & \mathrm{}\hfill & \hfill & 𝐚\hfill & \hfill & 𝐛\hfill & \hfill & \mathrm{}\hfill & \hfill & 𝐪\hfill & \hfill \end{array}$$
(3.39)
Proceeding in this way, the process terminates when each $`k^{}`$ has moved as far east and south as the jeu de taquin allows. The fact that no two identical $`k^{}`$s may lie in the same row is sufficient, given T1 and T2, to show that the resulting primed shifted tableau satisfies PST1-4. Since we had already noted that the diagonal entries are always unprimed, PST5 is also satisfied.
It follows that $`\mathrm{\Theta }^1`$ is well defined and maps the juxtaposition of any pair of tableaux $`PD𝒫𝒟^\delta (n)`$ and $`T𝒯^\lambda (n)`$ to a unique $`PST𝒫𝒮𝒯^\mu (n)`$. Thus the original map $`\mathrm{\Theta }`$ from $`𝒫𝒮𝒯^\mu (n)`$ to $`(𝒫𝒟^\delta (n),𝒯^\lambda (n))`$ is indeed bijective. Since it is also weight preserving, as argued earlier, this completes the proof of the $`PST`$ case in Theorem 3.1.
The only difference between the $`PST`$ and $`QST`$ cases is the fact that in the latter case primed entries are allowed on the main diagonal. This is reflected in the same distinction between $`PD`$ and $`QD`$ on the right of the above formulae. In fact it is not difficult to see that the map $`\mathrm{\Theta }`$ preserves the entries on the main diagonal in both cases; that is, just as the main diagonal of $`PST`$ coincides with that of $`PD`$, where $`\mathrm{\Theta }:PST(PD,T)`$, so the main diagonal of $`QST`$, complete with any primes, coincides with that of $`QD`$, where $`\mathrm{\Theta }:QST(QD,T)`$. This observation is sufficient to complete the proof of Theorem 3.1.
### 3.2 Example
This bijection is illustrated by the map from $`PST`$ of (2.13) to $`PD`$ of (2.14) and $`T`$ of (2.11); that is,
$$PST=\begin{array}{ccccccccccccccccccc}\hfill & 1\hfill & \hfill & 1\hfill & \hfill & 1\hfill & \hfill & 2^{}\hfill & \hfill & 2\hfill & \hfill & 2\hfill & \hfill & 3\hfill & \hfill & 3\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & 2\hfill & \hfill & 2\hfill & \hfill & 3^{}\hfill & \hfill & 3\hfill & \hfill & 4^{}\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 6^{}\hfill & \hfill \\ & & & & \hfill & 3\hfill & \hfill & 3\hfill & \hfill & 4^{}\hfill & \hfill & 4\hfill & \hfill & 5^{}\hfill & \hfill & 6\hfill & \hfill & & \\ & & & & & & \hfill & 4\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill & & \\ & & & & & & & & \hfill & 5\hfill & \hfill & 6^{}\hfill & \hfill & 6\hfill & \hfill \\ & & & & & & & & & & \hfill & 6\hfill & \hfill & & \end{array}PD=\begin{array}{ccccccccccccc}\hfill & 1\hfill & \hfill & 2^{}\hfill & \hfill & 1\hfill & \hfill & 4^{}\hfill & \hfill & 5^{}\hfill & \hfill & 6^{}\hfill & \hfill \\ & & \hfill & 2\hfill & \hfill & 3^{}\hfill & \hfill & 2\hfill & \hfill & 5^{}\hfill & \hfill & 2\hfill & \hfill \\ & & & & \hfill & 3\hfill & \hfill & 4^{}\hfill & \hfill & 3\hfill & \hfill & 3\hfill & \hfill \\ & & & & & & \hfill & 4\hfill & \hfill & 5^{}\hfill & \hfill & 6^{}\hfill & \hfill \\ & & & & & & & & \hfill & 5\hfill & \hfill & 5\hfill & \hfill \\ & & & & & & & & & & \hfill & 6\hfill & \hfill \end{array},T=\begin{array}{ccccccc}\hfill & 1\hfill & \hfill & 2\hfill & \hfill & 3\hfill & \hfill \\ \hfill & 3\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill \\ \hfill & 4\hfill & \hfill & 6\hfill & \hfill & & \\ \hfill & 5\hfill & \hfill & & \\ \hfill & 6\hfill & \hfill \\ & & \end{array}$$
(3.40)
The paths traced out by the primed entries $`k^{}`$ of $`PST`$ as they move northwest as far as but no further than the $`k`$th column are illustrated by means of boldface entries in the tableaux shown below:
First moving the single $`2^{}`$ under the map $`\theta _2^{}`$ gives:
$$\begin{array}{ccccccccccccccccccc}\hfill & 1\hfill & \hfill & \mathrm{𝟏}\hfill & \hfill & \mathrm{𝟏}\hfill & \hfill & \mathrm{𝟐}^{}\hfill & \hfill & 2\hfill & \hfill & 2\hfill & \hfill & 3\hfill & \hfill & 3\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & 2\hfill & \hfill & 2\hfill & \hfill & 3^{}\hfill & \hfill & 3\hfill & \hfill & 4^{}\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 6^{}\hfill & \hfill \\ & & & & \hfill & 3\hfill & \hfill & 3\hfill & \hfill & 4^{}\hfill & \hfill & 4\hfill & \hfill & 5^{}\hfill & \hfill & 6\hfill & \hfill & & \\ & & & & & & \hfill & 4\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill & & \\ & & & & & & & & \hfill & 5\hfill & \hfill & 6^{}\hfill & \hfill & 6\hfill & \hfill \\ & & & & & & & & & & \hfill & 6\hfill & \hfill & & \end{array}\begin{array}{ccccccccccccccccccc}\hfill & 1\hfill & \hfill & \mathrm{𝟐}^{}\hfill & \hfill & \mathrm{𝟏}\hfill & \hfill & \mathrm{𝟏}\hfill & \hfill & 2\hfill & \hfill & 2\hfill & \hfill & 3\hfill & \hfill & 3\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & 2\hfill & \hfill & 2\hfill & \hfill & 3^{}\hfill & \hfill & 3\hfill & \hfill & 4^{}\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 6^{}\hfill & \hfill \\ & & & & \hfill & 3\hfill & \hfill & 3\hfill & \hfill & 4^{}\hfill & \hfill & 4\hfill & \hfill & 5^{}\hfill & \hfill & 6\hfill & \hfill & & \\ & & & & & & \hfill & 4\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill & & \\ & & & & & & & & \hfill & 5\hfill & \hfill & 6^{}\hfill & \hfill & 6\hfill & \hfill \\ & & & & & & & & & & \hfill & 6\hfill & \hfill & & \end{array}$$
(3.41)
Under $`\theta _3^{}`$ the only $`3^{}`$ moves just one step west where it has, as required, reached the $`3`$rd column. It does not move north because the entry $`1`$ immediately above already lies in its own row:
$$\begin{array}{ccccccccccccccccccc}\hfill & 1\hfill & \hfill & 2^{}\hfill & \hfill & 1\hfill & \hfill & 1\hfill & \hfill & 2\hfill & \hfill & 2\hfill & \hfill & 3\hfill & \hfill & 3\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & 2\hfill & \hfill & \mathrm{𝟐}\hfill & \hfill & \mathrm{𝟑}^{}\hfill & \hfill & 3\hfill & \hfill & 4^{}\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 6^{}\hfill & \hfill \\ & & & & \hfill & 3\hfill & \hfill & 3\hfill & \hfill & 4^{}\hfill & \hfill & 4\hfill & \hfill & 5^{}\hfill & \hfill & 6\hfill & \hfill & & \\ & & & & & & \hfill & 4\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill & & \\ & & & & & & & & \hfill & 5\hfill & \hfill & 6^{}\hfill & \hfill & 6\hfill & \hfill \\ & & & & & & & & & & \hfill & 6\hfill & \hfill & & \end{array}\begin{array}{ccccccccccccccccccc}\hfill & 1\hfill & \hfill & 2^{}\hfill & \hfill & 1\hfill & \hfill & 1\hfill & \hfill & 2\hfill & \hfill & 2\hfill & \hfill & 3\hfill & \hfill & 3\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & 2\hfill & \hfill & \mathrm{𝟑}^{}\hfill & \hfill & \mathrm{𝟐}\hfill & \hfill & 3\hfill & \hfill & 4^{}\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 6^{}\hfill & \hfill \\ & & & & \hfill & 3\hfill & \hfill & 3\hfill & \hfill & 4^{}\hfill & \hfill & 4\hfill & \hfill & 5^{}\hfill & \hfill & 6\hfill & \hfill & & \\ & & & & & & \hfill & 4\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill & & \\ & & & & & & & & \hfill & 5\hfill & \hfill & 6^{}\hfill & \hfill & 6\hfill & \hfill \\ & & & & & & & & & & \hfill & 6\hfill & \hfill & & \end{array}$$
(3.42)
There are two $`4^{}`$s. Under $`\theta _4^{}`$ the upper one must be moved first and then the lower one:
$$\begin{array}{ccccccccccccccccccc}\hfill & 1\hfill & \hfill & 2^{}\hfill & \hfill & 1\hfill & \hfill & \mathrm{𝟏}\hfill & \hfill & \mathrm{𝟐}\hfill & \hfill & 2\hfill & \hfill & 3\hfill & \hfill & 3\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & 2\hfill & \hfill & 3^{}\hfill & \hfill & 2\hfill & \hfill & \mathrm{𝟑}\hfill & \hfill & \mathrm{𝟒}^{}\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 6^{}\hfill & \hfill \\ & & & & \hfill & 3\hfill & \hfill & 3\hfill & \hfill & 4^{}\hfill & \hfill & 4\hfill & \hfill & 5^{}\hfill & \hfill & 6\hfill & \hfill & & \\ & & & & & & \hfill & 4\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill & & \\ & & & & & & & & \hfill & 5\hfill & \hfill & 6^{}\hfill & \hfill & 6\hfill & \hfill \\ & & & & & & & & & & \hfill & 6\hfill & \hfill & & \end{array}\begin{array}{ccccccccccccccccccc}\hfill & 1\hfill & \hfill & 2^{}\hfill & \hfill & 1\hfill & \hfill & \mathrm{𝟒}^{}\hfill & \hfill & \mathrm{𝟏}\hfill & \hfill & 2\hfill & \hfill & 3\hfill & \hfill & 3\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & 2\hfill & \hfill & 3^{}\hfill & \hfill & 2\hfill & \hfill & \mathrm{𝟐}\hfill & \hfill & \mathrm{𝟑}\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 6^{}\hfill & \hfill \\ & & & & \hfill & 3\hfill & \hfill & 3\hfill & \hfill & 4^{}\hfill & \hfill & 4\hfill & \hfill & 5^{}\hfill & \hfill & 6\hfill & \hfill & & \\ & & & & & & \hfill & 4\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill & & \\ & & & & & & & & \hfill & 5\hfill & \hfill & 6^{}\hfill & \hfill & 6\hfill & \hfill \\ & & & & & & & & & & \hfill & 6\hfill & \hfill & & \end{array}=\begin{array}{ccccccccccccccccccc}\hfill & 1\hfill & \hfill & 2^{}\hfill & \hfill & 1\hfill & \hfill & 4^{}\hfill & \hfill & 1\hfill & \hfill & 2\hfill & \hfill & 3\hfill & \hfill & 3\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & 2\hfill & \hfill & 3^{}\hfill & \hfill & 2\hfill & \hfill & 2\hfill & \hfill & 3\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 6^{}\hfill & \hfill \\ & & & & \hfill & 3\hfill & \hfill & \mathrm{𝟑}\hfill & \hfill & \mathrm{𝟒}^{}\hfill & \hfill & 4\hfill & \hfill & 5^{}\hfill & \hfill & 6\hfill & \hfill & & \\ & & & & & & \hfill & 4\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill & & \\ & & & & & & & & \hfill & 5\hfill & \hfill & 6^{}\hfill & \hfill & 6\hfill & \hfill \\ & & & & & & & & & & \hfill & 6\hfill & \hfill & & \end{array}\begin{array}{ccccccccccccccccccc}\hfill & 1\hfill & \hfill & 2^{}\hfill & \hfill & 1\hfill & \hfill & 4^{}\hfill & \hfill & 1\hfill & \hfill & 2\hfill & \hfill & 3\hfill & \hfill & 3\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & 2\hfill & \hfill & 3^{}\hfill & \hfill & 2\hfill & \hfill & 2\hfill & \hfill & 3\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 6^{}\hfill & \hfill \\ & & & & \hfill & 3\hfill & \hfill & \mathrm{𝟒}^{}\hfill & \hfill & \mathrm{𝟑}\hfill & \hfill & 4\hfill & \hfill & 5^{}\hfill & \hfill & 6\hfill & \hfill & & \\ & & & & & & \hfill & 4\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill & & \\ & & & & & & & & \hfill & 5\hfill & \hfill & 6^{}\hfill & \hfill & 6\hfill & \hfill \\ & & & & & & & & & & \hfill & 6\hfill & \hfill & & \end{array}$$
(3.43)
There are three $`5^{}`$s to deal with in turn from top to bottom using $`\theta _5^{}`$, but the last of these is already in the $`3`$rd column and directly below a $`3`$ in the $`3`$rd row, and so does not move:
1214𝟏𝟐𝟑3523223𝟓56343456455556661214𝟓𝟏𝟐3523223𝟑5634345645555666=
1214512352322335634345645555666121451235232𝟓𝟐356343𝟑𝟒645555666formulae-sequence1214𝟏𝟐𝟑35missing-subexpressionmissing-subexpression23223𝟓56missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression343456missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression4555missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression566missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression6missing-subexpressionmissing-subexpression1214𝟓𝟏𝟐35missing-subexpressionmissing-subexpression23223𝟑56missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression343456missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression4555missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression566missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression6missing-subexpressionmissing-subexpression
1214512352322335634345645555666121451235missing-subexpressionmissing-subexpression232𝟓𝟐356missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression343𝟑𝟒6missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression4555missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression566missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression6missing-subexpressionmissing-subexpression{\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf{5^{\prime}}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr}}}\ \ \longrightarrow\ \ {\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf{5^{\prime}}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr}}}\ =\ {\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf{5^{\prime}}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr}}}\ \ \longrightarrow\ \ {\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf{5^{\prime}}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}} (3.44)
Then we deal with the two $`6^{}`$s by applying $`\theta _6^{}`$ to give
12145𝟏𝟐𝟑𝟓2325235𝟔3433464555566612145𝟔𝟏𝟐𝟑2325235𝟓34334645555666=
12145612323252355343346455556661214561232325235534334645𝟔55𝟓6612145𝟏𝟐𝟑𝟓missing-subexpressionmissing-subexpression2325235𝟔missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression343346missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression4555missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression566missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression6missing-subexpressionmissing-subexpression12145𝟔𝟏𝟐𝟑missing-subexpressionmissing-subexpression2325235𝟓missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression343346missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression4555missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression566missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression6missing-subexpressionmissing-subexpression
1214561232325235534334645555666121456123missing-subexpressionmissing-subexpression23252355missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression343346missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression45𝟔5missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression5𝟓6missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression6missing-subexpressionmissing-subexpression{\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf{6^{\prime}}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr}}}\ \ \longrightarrow\ \ {\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf{6^{\prime}}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr}}}\ \ =\ \ {\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf{6^{\prime}}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr}}}\ \longrightarrow\ {\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf{6^{\prime}}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\bf 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}} (3.45)
This results in the juxtaposition of $`PD`$ from (2.14) and $`T`$ from (2.11) as claimed:
12145612323252355343346456555661214562325234334565561233554656
1214561232325235534334645655566121456missing-subexpressionmissing-subexpression23252missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression3433missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression456missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression55missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression6123fragments35546missing-subexpressionmissing-subexpression5missing-subexpressionmissing-subexpression6missing-subexpressionmissing-subexpressionmissing-subexpression{\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr}}}\ \ \equiv\ \ {\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr}}}\quad\cdot\quad{\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&\cr\hrulefill\cr}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}} (3.46)
### 3.3 Corollaries
By associating $`x_k`$ and $`y_k`$ to each entry $`k`$ and $`k^{}`$, respectively, in the various tableaux $`PST`$, $`QST`$, $`PD`$, $`QD`$ and $`T`$ appearing in Theorem 3.1 we immediately have the following corollary.
###### Corollary 3.2
Let $`\mu =\lambda +\delta `$ be a strict partition of length $`\mathrm{}(\mu )=n`$, with $`\lambda `$ a partition of length $`\mathrm{}(\lambda )n`$ and $`\delta =(n,n1,\mathrm{},1)`$.
$$\begin{array}{ccc}\hfill _{PST𝒫𝒮𝒯^\mu (n)}(𝐱/𝐲)^{\mathrm{wgt}(PST)}& =& _{PD𝒫𝒟^\delta (n)}(𝐱/𝐲)^{\mathrm{wgt}(PD)}_{T𝒯^\lambda (n)}𝐱^{\mathrm{wgt}(T)};\hfill \\ & & \\ \hfill _{QST𝒬𝒮𝒯^\mu (n)}(𝐱/𝐲)^{\mathrm{wgt}(QST)}& =& _{QD𝒬𝒟^\delta (n)}(𝐱/𝐲)^{\mathrm{wgt}(QD)}_{T𝒯^\lambda (n)}𝐱^{\mathrm{wgt}(T)}.\hfill \end{array}$$
(3.47)
Thanks to the definitions of $`P(𝐱/𝐲)`$ and $`Q(𝐱/𝐲)`$ given in (2.26), the identities (2.15) and (2.16) and the combinatorial definition of $`s_\lambda (𝐱)`$ given in (2.24), the above result is nothing other than our first main result, Proposition 1.1.
Other corollaries follow as special cases of these results. Setting $`\lambda =0`$ we obtain
$$P_\delta (𝐱/𝐲)=s_{1^n}(𝐱)\underset{1i<jn}{}(x_i+y_j)\text{and}Q_\delta (𝐱/𝐲)=\underset{1ijn}{}(x_i+y_j).$$
(3.48)
Further specialisation to the case $`𝐲=𝐱`$ leads to a result given by Macdonald \[M95\](Sec.III.8,Ex.3 p259):
$$P_\delta (𝐱)=s_\delta (𝐱)\text{and}Q_\delta (𝐱)=2^ns_\delta (𝐱),$$
(3.49)
where use has been made of the fact that
$$\underset{i=1}{\overset{n}{}}x_i\underset{1i<jn}{}(x_i+x_j)=s_{1^n}(𝐱)\underset{1i<jn}{}\frac{(x_i^2x_j^2)}{(x_ix_j)}=s_{1^n}(𝐱)s_{\delta /1^n}(𝐱)=s_\delta (𝐱),$$
(3.50)
where the last step is true when, as here, $`𝐱`$ has $`n`$ components $`x_1,x_2,\mathrm{},x_n`$.
More generally, if $`\mu =\lambda +\delta `$ for any partition $`\lambda `$ of length $`\mathrm{}(\lambda )n`$, but $`𝐲=𝐱`$ we have another result due to Macdonald \[M95\](Sec.III.8,Ex.2 p259):
$$P_{\lambda +\delta }(𝐱)=s_\delta (𝐱)s_\lambda (𝐱).$$
(3.51)
where (3.50) and (2.27) have been applied directly to the $`𝐲=𝐱`$ case of (1.6).
On the other hand the case $`𝐲=t𝐱=(tx_1,tx_2,\mathrm{},tx_n)`$ of (1.6) is equivalent to (1.5), the $`t`$-deformation of Weyl’s denominator formula for the Lie algebra $`gl(n)`$ due to Tokuyama \[T88\]:
###### Corollary 3.3
$$\underset{i=1}{\overset{n}{}}x_i\underset{1i<jn}{}(x_i+tx_j)s_\lambda (𝐱)=\underset{ST𝒮𝒯^\mu (n)}{}t^{\mathrm{hgt}(ST)}(1+t)^{\mathrm{str}(ST)n}𝐱^{\mathrm{wgt}(ST)},$$
(3.52)
Proof While there is a combinatorial proof of this result due to Okada \[O90\], it follows immediately from Theorem 3.1 by setting $`y_k=tx_k`$ for all $`k=1,2,\mathrm{},n`$, noting that deleting primes from the entries $`k^{}`$ in each $`PST𝒫𝒮𝒯^\mu (n)`$ gives a shifted tableaux $`ST𝒮𝒯^\mu (n)`$ with a factor of $`t`$ arising from each primed entry of $`PST`$, and observing that these must occur in precisely those boxes contributing to $`\mathrm{hgt}(ST)`$ and are optional, thereby giving rise to a factor of $`(1+t)`$ in those $`\mathrm{str}(ST)n`$ boxes at the lower left hand end of all continuous strips of identical entries other than those starting on the main diagonal.
The remaining corollaries mentioned in the Introduction are the formulae (1.3) of Robbins and Rumsey \[RR86\] and (1.4) of Chapman \[C01\]. These require for their elucidation a link with alternating sign matrices. This is provided in Section 5.
## 4 The $`sp(2n)`$ bijection
### 4.1 Main Result
The analogue in the symplectic case of Theorem 3.1 is the following:
###### Theorem 4.1
Let $`\mu =\lambda +\delta `$ be a strict partition of length $`\mathrm{}(\mu )=n`$, with $`\lambda `$ a partition of length $`\mathrm{}(\lambda )n`$ and $`\delta =(n,n1,\mathrm{},1)`$. There exists a weight and barred weight preserving, bijective map $`\mathrm{\Phi }`$ from $`𝒬𝒮𝒯^\mu (n,\overline{n})`$ to $`(𝒬𝒟^\delta (n,\overline{n}),𝒯^\lambda (n,\overline{n}))`$ such that for all $`QST𝒬𝒮𝒯^\mu (n,\overline{n})`$
$$\mathrm{\Phi }:QST(QD,T)\text{with}\{\begin{array}{c}\mathrm{wgt}(QST)=\mathrm{wgt}(QD)+\mathrm{wgt}(T)\hfill \\ \\ \mathrm{bar}(QST)=\mathrm{bar}(QD)+\mathrm{bar}(T)\hfill \end{array}$$
(4.53)
and $`QD𝒬𝒟^\delta (n,\overline{n})`$ and $`T𝒯^\lambda (n,\overline{n})`$.
Proof
The Theorem is proved by the identification of a suitable map $`\mathrm{\Phi }`$ that it is both weight preserving and bijective. The underlying procedure is the same as before, in that the jeu de taquin is applied successively to all primed entries of $`QST`$ dealing in sequence with all entries $`\overline{k}^{}`$ and then $`k^{}`$ for $`k=1,2\mathrm{},n`$. In the case of the $`\overline{k}^{}`$s there is no impediment to moving all these entries to the $`k`$th column by means of the jeu de taquin, but something slightly more subtle is required in the case of the $`k^{}`$s. As well, the transformations applied to entries in the $`k`$th column now include not only permutations but an additional weight preserving transformation.
The structure of $`\mathrm{\Phi }`$ is such that $`\mathrm{\Phi }=\varphi _n^{}\varphi _{\overline{n}^{}}\mathrm{}\varphi _2^{}\varphi _{\overline{2}^{}}\varphi _1^{}\varphi _{\overline{1}^{}}`$. Here $`\varphi _{\overline{k}^{}}`$ differs from $`\theta _k^{}`$ only in that the jeu de taquin is played with $`\overline{k}^{}`$s rather than the $`k^{}`$s, while $`\varphi _k^{}=\chi _k^{}\psi _k^{}`$ where $`\psi _k^{}`$ differs from $`\theta _k^{}`$ only if the final step of the path of $`k^{}`$ into the $`i`$th row of the $`k`$th column is blocked by an entry $`\overline{k}^{}`$. In such a situation the horizontal pair of entries $`\overline{k}^{}k^{}`$ in the $`i`$th row is replaced by the horizontal pair $`i\overline{i}`$. Having moved all the $`k^{}`$s into the $`k`$th column or annihilated them as above, there may remain in the $`k`$th column vertical pairs $`\overline{i}i`$. It is then necessary to invoke $`\chi _k^{}`$. This replaces the lowest such pair, for which $`i`$ is necessarily in the $`i`$th row, by a vertical pair $`k^{}\overline{k}^{}`$, moves the resulting $`k^{}`$ and $`\overline{k}^{}`$ north as far as possible whilst still satisfyng T$`\overline{3}`$, and then acts in the same way on the next lowest vertical pair $`\overline{j}j`$, replacing them by another vertical pair $`k^{}\overline{k}^{}`$, and so on. Having removed all unprimed vertical pairs in this way any remaining unbarred entries $`i`$ or $`\overline{i}`$, but not both, lie in their own $`i`$th row, as required for consistency with QD$`\overline{1}`$.
To demonstrate that these various maps are well defined, we exhibit the relevant individual steps as below, first for the $`\overline{k}^{}`$ case. The starting point is the tableau $`T_0`$ in which the entries $`a,b,\mathrm{},h`$ satisfy the semistandardness conditions T1 and T2, and the symplectic constraint T$`\overline{3}`$. This tableau $`T_0`$ is to be thought of as the subtableau surrounding a particular $`\overline{k}^{}`$ after the jeu de taquin has been applied to all $`\overline{k}^{}`$s appearing initially above the $`\overline{k}^{}`$ in question, moving them into the $`k`$th column. Some further steps of the jeu de taquin may have already been applied to the central $`\overline{k}^{}`$ and the diagram is intended to indicate under what conditions its next move is north or west, given that it currently lies in the $`i`$th row.
If $`db`$ then $`\overline{k}^{}`$ is to be interchanged with $`b`$ and if $`d>b`$ then $`\overline{k}^{}`$ is to be interchanged with $`d`$. In the first case $`\overline{k}^{}`$ moves north and the resulting tableau $`T_N`$ satisfies both $`T1`$ and $`T2`$ since $`dbc<e`$, while in the second $`\overline{k}^{}`$ moves west and the resulting tableau $`T_W`$ satisfies both $`T1`$ and $`T2`$ since $`b<d<fg`$. What is particularly important to note here is that by virtue of T$`\overline{3}`$ applied to $`T_0`$ it is known that $`d\overline{i}`$. It follows that if $`db`$ and the map is from $`T_0`$ to $`T_N`$ then $`b\overline{i}`$ so that $`T_N`$ also satisfies T$`\overline{3}`$. Conversely if $`b<\overline{i}`$ then we must have $`b<d`$ and the map must be from $`T_0`$ to $`T_W`$ in which case T$`\overline{3}`$ is still satisfied.
$$\varphi _{\overline{k}^{}}:\begin{array}{c}T_0=\begin{array}{ccccccc}\hfill & a\hfill & \hfill & b\hfill & \hfill & c\hfill & \hfill \\ \hfill & d\hfill & \hfill & \overline{k}^{}\hfill & \hfill & e\hfill & \hfill \\ \hfill & f\hfill & \hfill & g\hfill & \hfill & h\hfill & \hfill \end{array}\\ \overline{k}^{}\text{ in }i\text{th row}\\ \overline{i}d\end{array}\{\begin{array}{cc}T_N=\begin{array}{ccccccc}\hfill & a\hfill & \hfill & \overline{k}^{}\hfill & \hfill & c\hfill & \hfill \\ \hfill & d\hfill & \hfill & b\hfill & \hfill & e\hfill & \hfill \\ \hfill & f\hfill & \hfill & g\hfill & \hfill & h\hfill & \hfill \end{array}\hfill & \text{if}\overline{i}db\hfill \\ & \\ T_W=\begin{array}{ccccccc}\hfill & a\hfill & \hfill & b\hfill & \hfill & c\hfill & \hfill \\ \hfill & \overline{k}^{}\hfill & \hfill & d\hfill & \hfill & e\hfill & \hfill \\ \hfill & f\hfill & \hfill & g\hfill & \hfill & h\hfill & \hfill \end{array}\hfill & \text{if}b<\overline{i}d\hfill \end{array}$$
(4.54)
If $`\overline{k}^{}`$ is already in the topmost row or leftmost column the following diagrams illustrate the allowed moves of $`\overline{k}^{}`$ west and north respectively:
$$\varphi _{\overline{k}^{}}:T_0=\begin{array}{ccccccc}\hfill & d\hfill & \hfill & \overline{k}^{}\hfill & \hfill & e\hfill & \hfill \\ \hfill & f\hfill & \hfill & g\hfill & \hfill & h\hfill & \hfill \end{array}T_W=\begin{array}{ccccccc}\hfill & \overline{k}^{}\hfill & \hfill & d\hfill & \hfill & e\hfill & \hfill \\ \hfill & f\hfill & \hfill & g\hfill & \hfill & h\hfill & \hfill \end{array}$$
(4.55)
and
$$\varphi _{\overline{k}^{}}:\begin{array}{c}T_0=\begin{array}{ccccc}\hfill & b\hfill & \hfill & c\hfill & \hfill \\ \hfill & \overline{k}^{}\hfill & \hfill & e\hfill & \hfill \\ \hfill & g\hfill & \hfill & h\hfill & \hfill \end{array}\\ \overline{k}^{}\text{ in }i\text{th row}\end{array}T_N=\begin{array}{ccccc}\hfill & \overline{k}^{}\hfill & \hfill & c\hfill & \hfill \\ \hfill & b\hfill & \hfill & e\hfill & \hfill \\ \hfill & g\hfill & \hfill & h\hfill & \hfill \end{array}\text{if}\overline{i}b.$$
(4.56)
Notice that in order to satisfy T$`\overline{3}`$ the map from $`T_0`$ to $`T_N`$ only takes place if $`b\overline{i}`$, otherwise $`T_0`$ is unaltered and $`\overline{k}^{}`$ occupies the site in the $`i`$th row of the $`k`$th column, until disturbed by any incoming $`k^{}`$s, as we shall see later. In the meantime it is to be noted that $`T_W`$ satisfies T$`\overline{3}`$, since by hypothesis $`T_0`$ does so. By the same token, since the unprimed entries of $`T_0`$ satisfy both T1 and T2, these conditions are also satisfied by both $`T_W`$ and $`T_N`$ by virtue of the fact that we still have $`d<fg`$ and $`bc<e`$.
It still remains to be shown that by means of the above moves of $`\overline{k}^{}`$ in a north-westerly direction it actually reaches the $`k`$th column. This time the only possible impediments to this is the existence of another $`\overline{k}^{}`$ to its immediate left. However, by the same argument as that used in the $`\theta _k^{}`$ case, this cannot occur because the path followed by $`\overline{k}^{}`$ always remains strictly south of the path followed by all preceding $`\overline{k}^{}`$s.
Having completed the jeu de taquin moves for all $`\overline{k}^{}`$s and moved them as far north as possible in the $`k`$th column, it remains to deal with any $`k^{}`$s in $`QST`$ using $`\varphi _k^{}=\chi _k^{}\psi _k^{}`$. The action of $`\psi _k^{}`$ is carried out in the same way as before with the diagrams of (4.54), (4.55) and (4.56) just altered by changing $`\overline{k}^{}`$ to $`k^{}`$, provided that $`d`$ is unprimed. If $`d`$ is primed then $`dk^{}`$ since $`d=k^{}`$ would give a horizontal pair $`k^{}k^{}`$, and this cannot occur since the path of the second $`k^{}`$ must stay strictly south of the path of the first $`k^{}`$. To deal with the case $`d=\overline{k}^{}`$ the following map is required if $`a`$ is unprimed
ϕk:T0=abck¯kefghk in ith rowa<i¯{TN=
akc¯kbefghifa<i¯bTW=abcii¯efghifab<i¯\phi_{{k}^{\prime}}:\ \ \begin{array}[]{c}T_{0}=\ {\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle a&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle b&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle c&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{k}^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle k^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle e&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle f&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle g&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle h&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr}}}\\
\hbox{$k^{\prime}$ in $i$th row}\\
a<\overline{i}\\
\end{array}\ \ \longrightarrow\ \ \left\{\begin{array}[]{ll}T_{N}=\ {\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle a&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle k^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle c&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{k}^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle b&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle e&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle f&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle g&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle h&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr}}}&\quad\hbox{if}\quad a<\overline{i}\leq b\\
\\
T_{W}=\ {\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle a&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle b&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle c&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle i&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{i}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle e&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle f&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle g&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle h&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr}}}&\quad\hbox{if}\quad a\leq b<\overline{i}\\
\end{array}\right.}}}}}}}}}}}}}}}}}}}}}}}}}}} (4.57)
where advantage has been taken of the fact that $`a<\overline{i}<i`$, since if this were not the case the action of $`\varphi _{\overline{k}^{}}`$ would have required that $`a`$ and $`\overline{k}^{}`$ be interchanged in $`T_0`$. It follows that $`T_W`$ satisfies T$`\overline{3}`$. The fact that the pair $`i\overline{i}`$ violates $`T1`$ is not a problem because the $`i`$ lies in the $`k`$th column of what will become $`QD`$ and the $`\overline{i}`$ lies either in the $`(k+1)`$th column of $`QD`$ or the first column of $`T`$. In neither case does the condition T$`1`$ apply to both $`i`$ and $`\overline{i}`$. If they are both in $`QD`$ they automatically satisfy the condition QD1=PD1, and if one is in $`QD`$ and the other in $`T`$ then $`i`$ and $`\overline{i}`$ automatically satisfy QST$`\overline{5}`$ and the combination of ST1 and ST$`\overline{4}`$, respectively. In the case that $`\varphi _k^{}`$ maps $`T_0`$ to $`T_N`$, then one reverts to the use of $`\psi _k^{}=\theta _k^{}`$ for the next step (or equivalently $`\varphi _{\overline{k}^{}}`$ with $`\overline{k}^{}`$ replaced by $`k^{}`$ in (4.54)-(4.56)).
Finally, if $`d=\overline{k}^{}`$ and $`a`$ is primed then for $`a=\overline{k}^{}`$ the action of $`\psi _k^{}`$ gives
$$\psi _k^{}:\begin{array}{c}T_0=\begin{array}{ccccccc}\hfill & \overline{k}^{}\hfill & \hfill & b\hfill & \hfill & c\hfill & \hfill \\ \hfill & \overline{k}^{}\hfill & \hfill & k^{}\hfill & \hfill & e\hfill & \hfill \\ \hfill & f\hfill & \hfill & g\hfill & \hfill & h\hfill & \hfill \end{array}\\ k^{}\text{ in }i\text{th row}\end{array}\{\begin{array}{cc}T_N=\begin{array}{ccccccc}\hfill & \overline{k}^{}\hfill & \hfill & k^{}\hfill & \hfill & c\hfill & \hfill \\ \hfill & \overline{k}^{}\hfill & \hfill & b\hfill & \hfill & e\hfill & \hfill \\ \hfill & f\hfill & \hfill & g\hfill & \hfill & h\hfill & \hfill \end{array}\hfill & \text{if}\overline{i}b\hfill \\ & \\ T_W=\begin{array}{ccccccc}\hfill & \overline{k}^{}\hfill & \hfill & b\hfill & \hfill & c\hfill & \hfill \\ \hfill & i\hfill & \hfill & \overline{i}\hfill & \hfill & e\hfill & \hfill \\ \hfill & f\hfill & \hfill & g\hfill & \hfill & h\hfill & \hfill \end{array}\hfill & \text{if}b<\overline{i}\hfill \end{array}$$
(4.58)
while for $`a=k^{}`$ the action is
$$\psi _k^{}:\begin{array}{c}T_0=\begin{array}{ccccccc}\hfill & k^{}\hfill & \hfill & b\hfill & \hfill & c\hfill & \hfill \\ \hfill & \overline{k}^{}\hfill & \hfill & k^{}\hfill & \hfill & e\hfill & \hfill \\ \hfill & f\hfill & \hfill & g\hfill & \hfill & h\hfill & \hfill \end{array}\\ \overline{k}^{}\text{ in }i\text{th row}\end{array}\begin{array}{c}b<\overline{i}\\ \end{array}T_W=\begin{array}{ccccccc}\hfill & k^{}\hfill & \hfill & b\hfill & \hfill & c\hfill & \hfill \\ \hfill & i\hfill & \hfill & \overline{i}\hfill & \hfill & e\hfill & \hfill \\ \hfill & f\hfill & \hfill & g\hfill & \hfill & h\hfill & \hfill \end{array}$$
(4.59)
since the uppermost $`k^{}`$ must have arrived at its present position by displacing $`b`$, and $`b`$ could only have been in that position above $`\overline{k}^{}`$ in the $`i`$th row if it satisfied $`b<\overline{i}`$. This prevents the lower $`k^{}`$ from moving north by interchanging with $`b`$, since the latter move would violate T$`\overline{3}`$.
If $`k^{}`$ is already in the topmost row or leftmost column the following diagrams (4.60)and (4.61) illustrate the allowed moves of $`k^{}`$ west and north respectively:
$$\psi _k^{}:T_0=\begin{array}{ccccccc}\hfill & d\hfill & \hfill & k^{}\hfill & \hfill & e\hfill & \hfill \\ \hfill & f\hfill & \hfill & g\hfill & \hfill & h\hfill & \hfill \end{array}T_W=\begin{array}{ccccccc}\hfill & k^{}\hfill & \hfill & d\hfill & \hfill & e\hfill & \hfill \\ \hfill & f\hfill & \hfill & g\hfill & \hfill & h\hfill & \hfill \end{array}$$
(4.60)
and
$$\psi _k^{}:\begin{array}{c}T_0=\begin{array}{ccccc}\hfill & b\hfill & \hfill & c\hfill & \hfill \\ \hfill & \overline{k}^{}\hfill & \hfill & e\hfill & \hfill \\ \hfill & g\hfill & \hfill & h\hfill & \hfill \end{array}\\ k^{}\text{ in }i\text{th row}\end{array}\begin{array}{c}\overline{i}b\\ \end{array}T_N=\begin{array}{ccccc}\hfill & k^{}\hfill & \hfill & c\hfill & \hfill \\ \hfill & b\hfill & \hfill & e\hfill & \hfill \\ \hfill & g\hfill & \hfill & h\hfill & \hfill \end{array}$$
(4.61)
Following all these moves of $`k^{}`$ and thereby completing the action of $`\psi _k^{}`$ one may still have one or more unprimed pairs $`i`$ and $`\overline{i}`$ in the same column. If so then $`\chi _k^{}`$ acts on the lowest such pair. This pair must be such that $`\overline{i}`$ and $`i`$ lie in the $`(i1)`$th and $`i`$th rows, respectively. To see this let them lie in the $`j1`$th and $`j`$th rows with $`ji`$ by virtue of T$`\overline{3}`$. Then following the action of $`\psi _k^{}`$ all unprimed entries below $`i`$ are unpaired and must lie in their own row, whether they are barred or unbarred, by virtue of the argument already given in the $`\theta _k^{}`$ case. Thus the entry immediately below $`i`$ is either $`j+1`$, $`\overline{j+1}`$, $`k^{}`$ or $`\overline{k}^{}`$. In all four cases the condition T2 and the rule for shifting $`k^{}`$ and $`\overline{k}^{}`$ as far north as is consistent with T$`\overline{3}`$ imply that $`ij`$. Since $`ji`$ it follows that $`j=i`$, as claimed. The action of $`\chi _k^{}`$ is then to map this vertical pair $`\overline{i}i`$ with $`i`$ in the $`i`$th row to $`k^{}\overline{k}^{}`$.
$$\chi _k^{}:\begin{array}{c}T_0=\begin{array}{ccccc}\hfill & \overline{i}\hfill & \hfill & b\hfill & \hfill \\ \hfill & i\hfill & \hfill & d\hfill & \hfill \end{array}\\ i\text{ in }i\text{th row}\\ \overline{i}b<d\end{array}T=\begin{array}{ccccc}\hfill & k^{}\hfill & \hfill & b\hfill & \hfill \\ \hfill & \overline{k}^{}\hfill & \hfill & d\hfill & \hfill \end{array}$$
(4.62)
Then moving the new $`k^{}`$ and $`\overline{k}^{}`$ as far north as permitted by QD1=PD1, the process is repeated for any other vertical pair $`\overline{j}j`$ in the $`k`$th column with $`j`$ in the $`j`$th row, until all such pairs are eliminated. The result of this is that every unprimed entry $`i`$ or $`\overline{i}`$ in the $`k`$th column lies in the $`i`$th row, with all other entries in this column equal to $`k^{}`$ or $`\overline{k}^{}`$.
Repeating this procedure for all $`k=1,2,\mathrm{},n`$ results, as required, in the juxtaposition of a primed tableau $`QD𝒫𝒟^\delta (n,\overline{n})`$ and an umprimed tableau $`T𝒯^\lambda (n,\overline{n})`$. It should be noted that each indvidual step of the maps constituting $`\mathrm{\Phi }`$ is either a simple interchange or a map from a horizontal pair $`\overline{k}^{}k^{}`$ to a horizontal pair $`i\overline{i}`$, or a map from a vertical pair $`\overline{i}i`$ to a vertical pair $`k^{}\overline{k}^{}`$. In each case the weight of each pair is zero, and the number of barred entries is one. It follows that such steps are all weight and barred weight preserving, so that (4.53) is always satisfied.
To show that the map $`\mathrm{\Phi }`$ from $`QST𝒬𝒮𝒯^\mu (n,\overline{n})`$ to $`(QD,T)`$ with $`QD𝒫𝒟^\delta (n,\overline{n})`$ and $`T𝒯^\lambda (n,\overline{n})`$ is bijective it is sufficient to note that each step of the map $`\mathrm{\Phi }`$ may be reversed and that if $`\mathrm{\Phi }^1`$ is defined by such a reversal of each step, then the action of $`\mathrm{\Phi }^1`$ on the juxtaposition of any $`QD𝒫𝒟^\delta (n,\overline{n})`$ and any $`T𝒯^\lambda (n,\overline{n})`$ always leads to some $`QST𝒬𝒮𝒯^\mu (n,\overline{n})`$.
To see this note that since $`\mathrm{\Phi }=\varphi _n^{}\varphi _{\overline{n}^{}}\mathrm{}\varphi _2^{}\varphi _{\overline{2}^{}}\varphi _1^{}\varphi _{\overline{1}^{}}`$ with $`\varphi _k^{}=\chi _k^{}\psi _k^{}`$, then $`\mathrm{\Phi }^1=\varphi _{\overline{1}^{}}^1\varphi _1^{}^1\varphi _{\overline{2}^{}}^1\varphi _2^{}^1\mathrm{}\varphi _{\overline{n}^{}}^1\varphi _n^{}^1`$ with $`\varphi _k^{}^1=\psi _k^{}^1\chi _k^{}^1`$. The action of the $`\varphi _{\overline{k}^{}}^1`$ is defined by the action of $`\theta _k^{}^1`$ illustrated in (3.36) with $`k^{}`$ replaced by $`\overline{k}^{}`$, while that of $`\psi _k^{}^1`$ coincides precisely with that of $`\theta _k^{}^1`$ together with additional transformations of the type
ψk1:T0=abcii¯efghi in the ith rowb<i¯TE=
abc¯kkefgh\psi^{-1}_{k^{\prime}}:\ \ \begin{array}[]{c}T_{0}=\ {\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle a&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle b&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle c&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle i&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{i}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle e&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle f&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle g&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle h&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr}}}\\
\hbox{$i$ in the $i$th row}\\
\end{array}\quad\begin{array}[]{c}\ b<\overline{i}\\
\ \ \longrightarrow\\
\end{array}T_{E}=\ {\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle a&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle b&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle c&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{k}^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle k^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle e&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle f&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle g&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle h&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr}}}}}}}}}}}}}}}}}}}}} (4.63)
where, as indicated, the pair $`i\overline{i}`$ lies in the $`i`$th row of the tableau. The fact that we necessarily have $`b<\overline{i}`$ ensures that this is the exact inverse of the map $`\varphi _k^{}`$ that takes $`T_0`$ to $`T_W`$ in (4.57). If $`a=\overline{k}^{}`$ then the above gives the inverse of the map $`\varphi _k^{}`$ that takes $`T_0`$ to $`T_W`$ in (4.55), while if $`a=k^{}`$ then the inverse of (4.59) takes the form
ψk1:T0=
kbci¯iefghi in the ith rowb<i¯TE=kbck¯kefgh\psi^{-1}_{k^{\prime}}:\ \ \begin{array}[]{c}T_{0}=\ {\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle k^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle b&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle c&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle i&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{i}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle e&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle f&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle g&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle h&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr}}}\\
\hbox{$i$ in the $i$th row}\\
\end{array}\quad\begin{array}[]{c}\ \ b<\overline{i}\\
\ \ \longrightarrow\\
\end{array}\quad T_{E}=\ {\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle k^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle b&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle c&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{k}^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle k^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle e&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle f&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle g&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle h&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr}}}}}}}}}}}}}}}}}}}}} (4.64)
where it is important to note that when applying $`\psi _k^{}^1`$ the map of any $`i\overline{i}`$ to $`\overline{k}^{}k^{}`$ in the $`i`$th row must be carried out before moving any $`k^{}`$s that may appear in the tableau higher than the $`i`$th row.
Similarly, the basic action of the inverse of $`\chi _k^{}`$, which must be applied before $`\psi _k^{}^1`$ comes into play, lowers $`k^{}`$s in the $`k`$th column in accordance with
$$\chi _k^{}^1:T_0=\begin{array}{ccccc}\hfill & k^{}\hfill & \hfill & b\hfill & \hfill \\ \hfill & a\hfill & \hfill & d\hfill & \hfill \end{array}T=\begin{array}{ccccc}\hfill & a\hfill & \hfill & b\hfill & \hfill \\ \hfill & k^{}\hfill & \hfill & d\hfill & \hfill \end{array}$$
(4.65)
if $`a`$ is unprimed and $`ab`$. If $`a=k^{}`$ then the action of $`\chi _k^{}^1`$ produces no change, while if $`a=\overline{k}^{}`$ then
$$\chi _k^{}^1:\begin{array}{c}T_0=\begin{array}{ccccc}\hfill & k^{}\hfill & \hfill & b\hfill & \hfill \\ \hfill & \overline{k}^{}\hfill & \hfill & d\hfill & \hfill \end{array}\\ \overline{k}^{}\text{ in the }i\text{th row}\\ \overline{i}b<d\end{array}T=\begin{array}{ccccc}\hfill & \overline{i}\hfill & \hfill & b\hfill & \hfill \\ \hfill & i\hfill & \hfill & d\hfill & \hfill \end{array}$$
(4.66)
where the constraint $`\overline{i}b<d`$ is required so as to ensure that $`T_0`$ satisfies the semistandardness conditions T1 and T2. Of course if $`b<\overline{i}`$ then $`\chi _k^{}^1`$ cannot map as above, so it simply leaves $`T_0`$ unchanged but then subject to $`\varphi _k^{}^1`$ which acts as follows
$$\varphi _k^{}^1:T=\begin{array}{ccccc}\hfill & k^{}\hfill & \hfill & b\hfill & \hfill \\ \hfill & \overline{k}^{}\hfill & \hfill & d\hfill & \hfill \end{array}T_E=\begin{array}{ccccc}\hfill & b\hfill & \hfill & k^{}\hfill & \hfill \\ \hfill & \overline{k}^{}\hfill & \hfill & d\hfill & \hfill \end{array}$$
(4.67)
in accordance with the inverse of $`\theta _k^{}`$ applied in the case $`b<\overline{i}`$. In applying $`\chi _k^{}^1`$ as in (4.65) and (4.66) one starts the action with the most northerly $`k^{}`$ and the most northerly vertical pairs $`k^{}\overline{k}^{}`$, respectively, working down the $`k`$th column. This is in contrast to the subsequent action of $`\psi _k^{}^1`$ which acts on the most southerly $`k^{}`$ first.
Finally, it is necessary to show that all $`k^{}`$s and $`\overline{k}^{}`$s are mapped by $`\varphi _k^{}^1`$ and $`\varphi _{\overline{k}^{}}^1`$ to endpoints consistent with the conditions QST1-4 and QST$`\overline{5}`$ on all primed tableaux $`QST𝒬𝒮𝒯^\mu (n,\overline{n})`$. First it should be noted that QST3 and QST$`\overline{5}`$ are satisfied throughout the application of $`\mathrm{\Phi }^1`$. Furthermore, a violation of QST1 or QST2 by any $`k^{}`$ or $`\overline{k}^{}`$ simply means that application of $`\varphi _k^{}^1`$ or $`\varphi _{\overline{k}^{}}^1`$, respectively, has not been completed. The argument used in Section 3 regarding reverse paths staying strictly north of one another in any given column, is sufficient to ensure that no two identical primed entries may appear in the same row, thereby ensuring that the final condition QST3 is always satisfied. Thus the image of $`\mathrm{\Phi }^1`$ of any pair $`(QD,T)`$ with $`QD𝒫𝒟^\delta (n,\overline{n})`$ and $`T𝒯^\lambda (n,\overline{n})`$ is some $`QST𝒬𝒮𝒯^\mu (n,\overline{n})`$.
This implies that the original map $`\mathrm{\Phi }`$ is bijective. Since it is also both weight and barred weight preserving, this completes the proof of Theorem 4.1.
It should be pointed out that, unlike the $`gl(n)`$ case, a corresponding result does not apply to $`PST𝒫𝒮𝒯^\mu (n,\overline{n})`$ in the $`sp(2n)`$ case because of the necessity of using (4.62). If used, as will sometimes be necessary, in the case $`i=k`$ it results in a primed entry $`\overline{k}^{}`$ appearing on the main diagonal, in direct violation of PD3.
### 4.2 Example
The bijection $`\mathrm{\Phi }`$ is illustrated by the following map for $`n=5`$ and $`\mu =(9,7,6,2,1)`$:
PST=1¯12¯23¯34¯4¯52¯2¯3¯3¯4443¯34¯555445PD=
11¯34¯1¯2¯245¯3¯4¯3445,T=1¯4¯4¯53444¯55PST=\ {\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{1}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{2}^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{3}^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{4}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{4}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{2}^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{2}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{3}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{3}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{3}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{4}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr}}}\ \ \longrightarrow\ \ PD\ =\ \ {\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{3}^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{1}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{2}^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{2}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{3}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{4}^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{3}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr}}}\ \ ,\ \ T\ =\ \ {\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{1}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{4}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{4}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle\overline{4}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}} (4.68)
Once again we indicate by means of boldface entries both the paths traced out by elements $`\overline{k}^{}`$ and $`k^{}`$ as they move to the $`k`$th column under the action of $`\varphi _{\overline{k}^{}}`$ and $`\psi _k^{}`$, respectively, as well as annihilations and creation of $`\overline{k}^{}k^{}`$ pairs under $`\psi _k^{}`$ and $`\chi _k^{}`$, respectively.
There are no $`\overline{1}^{}`$s, so we first move the single $`1^{}`$ under the action of $`\psi _1^{}`$ as shown:
$$\begin{array}{ccccccccccccccccccc}\hfill & \overline{\mathrm{𝟏}}\hfill & \hfill & \mathrm{𝟏}^{}\hfill & \hfill & \overline{2}^{}\hfill & \hfill & 2^{}\hfill & \hfill & \overline{3}^{}\hfill & \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & \overline{4}\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & \overline{2}^{}\hfill & \hfill & \overline{2}\hfill & \hfill & \overline{3}\hfill & \hfill & \overline{3}\hfill & \hfill & 4^{}\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & \\ & & & & \hfill & \overline{3}\hfill & \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill \\ & & & & & & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & & & & & \\ & & & & & & & & \hfill & 5^{}\hfill & \hfill \end{array}\begin{array}{ccccccccccccccccccc}\hfill & \mathrm{𝟏}^{}\hfill & \hfill & \overline{\mathrm{𝟏}}\hfill & \hfill & \overline{2}^{}\hfill & \hfill & 2^{}\hfill & \hfill & \overline{3}^{}\hfill & \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & \overline{4}\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & \overline{2}^{}\hfill & \hfill & \overline{2}\hfill & \hfill & \overline{3}\hfill & \hfill & \overline{3}\hfill & \hfill & 4^{}\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & \\ & & & & \hfill & \overline{3}\hfill & \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill \\ & & & & & & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & & & & & \\ & & & & & & & & \hfill & 5^{}\hfill & \hfill \end{array}$$
(4.69)
The application of $`\varphi _{\overline{2}}^{}`$ then gives
$$\begin{array}{ccccccccccccccccccc}\hfill & 1^{}\hfill & \hfill & \overline{\mathrm{𝟏}}\hfill & \hfill & \overline{\mathrm{𝟐}}^{}\hfill & \hfill & 2^{}\hfill & \hfill & \overline{3}^{}\hfill & \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & \overline{4}\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & \overline{2}^{}\hfill & \hfill & \overline{2}\hfill & \hfill & \overline{3}\hfill & \hfill & \overline{3}\hfill & \hfill & 4^{}\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & \\ & & & & \hfill & \overline{3}\hfill & \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill \\ & & & & & & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & & & & & \\ & & & & & & & & \hfill & 5^{}\hfill & \hfill \end{array}\begin{array}{ccccccccccccccccccc}\hfill & 1^{}\hfill & \hfill & \overline{\mathrm{𝟐}}^{}\hfill & \hfill & \overline{\mathrm{𝟏}}\hfill & \hfill & 2^{}\hfill & \hfill & \overline{3}^{}\hfill & \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & \overline{4}\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & \overline{2}^{}\hfill & \hfill & \overline{2}\hfill & \hfill & \overline{3}\hfill & \hfill & \overline{3}\hfill & \hfill & 4^{}\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & \\ & & & & \hfill & \overline{3}\hfill & \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill \\ & & & & & & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & & & & & \\ & & & & & & & & \hfill & 5^{}\hfill & \hfill \end{array}$$
(4.70)
where there is no possibility of moving the lower $`\overline{2}^{}`$. Then the application of $`\psi _2^{}`$ on the only $`2^{}`$ involves first a transposition and then the replacement of the resulting horizontal pair $`\overline{2}^{}\mathrm{\hspace{0.17em}2}^{}`$ in the first row by $`1\overline{1}`$:
$$\begin{array}{ccccccccccccccccccc}\hfill & 1^{}\hfill & \hfill & \overline{2}^{}\hfill & \hfill & \overline{\mathrm{𝟏}}\hfill & \hfill & \mathrm{𝟐}^{}\hfill & \hfill & \overline{3}^{}\hfill & \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & \overline{4}\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & \overline{2}^{}\hfill & \hfill & \overline{2}\hfill & \hfill & \overline{3}\hfill & \hfill & \overline{3}\hfill & \hfill & 4^{}\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & \\ & & & & \hfill & \overline{3}\hfill & \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill \\ & & & & & & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & & & & & \\ & & & & & & & & \hfill & 5^{}\hfill & \hfill \end{array}\begin{array}{ccccccccccccccccccc}\hfill & 1^{}\hfill & \hfill & \overline{2}^{}\hfill & \hfill & \mathrm{𝟐}^{}\hfill & \hfill & \overline{\mathrm{𝟏}}\hfill & \hfill & \overline{3}^{}\hfill & \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & \overline{4}\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & \overline{2}^{}\hfill & \hfill & \overline{2}\hfill & \hfill & \overline{3}\hfill & \hfill & \overline{3}\hfill & \hfill & 4^{}\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & \\ & & & & \hfill & \overline{3}\hfill & \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill \\ & & & & & & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & & & & & \\ & & & & & & & & \hfill & 5^{}\hfill & \hfill \end{array}=\begin{array}{ccccccccccccccccccc}\hfill & 1^{}\hfill & \hfill & \overline{\mathrm{𝟐}}^{}\hfill & \hfill & \mathrm{𝟐}^{}\hfill & \hfill & \overline{1}\hfill & \hfill & \overline{3}^{}\hfill & \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & \overline{4}\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & \overline{2}^{}\hfill & \hfill & \overline{2}\hfill & \hfill & \overline{3}\hfill & \hfill & \overline{3}\hfill & \hfill & 4^{}\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & \\ & & & & \hfill & \overline{3}\hfill & \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill \\ & & & & & & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & & & & & \\ & & & & & & & & \hfill & 5^{}\hfill & \hfill \end{array}\begin{array}{ccccccccccccccccccc}\hfill & 1^{}\hfill & \hfill & \mathrm{𝟏}\hfill & \hfill & \overline{\mathrm{𝟏}}\hfill & \hfill & \overline{1}\hfill & \hfill & \overline{3}^{}\hfill & \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & \overline{4}\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & \overline{2}^{}\hfill & \hfill & \overline{2}\hfill & \hfill & \overline{3}\hfill & \hfill & \overline{3}\hfill & \hfill & 4^{}\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & \\ & & & & \hfill & \overline{3}\hfill & \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill \\ & & & & & & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & & & & & \\ & & & & & & & & \hfill & 5^{}\hfill & \hfill \end{array}$$
(4.71)
The single $`\overline{3}^{}`$ is moved as shown to the $`3`$rd column under the action of $`\psi _3^{}`$:
$$\begin{array}{ccccccccccccccccccc}\hfill & 1^{}\hfill & \hfill & 1\hfill & \hfill & \overline{\mathrm{𝟏}}\hfill & \hfill & \overline{\mathrm{𝟏}}\hfill & \hfill & \overline{\mathrm{𝟑}}^{}\hfill & \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & \overline{4}\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & \overline{2}^{}\hfill & \hfill & \overline{2}\hfill & \hfill & \overline{3}\hfill & \hfill & \overline{3}\hfill & \hfill & 4^{}\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & \\ & & & & \hfill & \overline{3}\hfill & \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill \\ & & & & & & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & & & & & \\ & & & & & & & & \hfill & 5^{}\hfill & \hfill \end{array}\begin{array}{ccccccccccccccccccc}\hfill & 1^{}\hfill & \hfill & 1\hfill & \hfill & \overline{\mathrm{𝟑}}^{}\hfill & \hfill & \overline{\mathrm{𝟏}}\hfill & \hfill & \overline{\mathrm{𝟏}}\hfill & \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & \overline{4}\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & \overline{2}^{}\hfill & \hfill & \overline{2}\hfill & \hfill & \overline{3}\hfill & \hfill & \overline{3}\hfill & \hfill & 4^{}\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & \\ & & & & \hfill & \overline{3}\hfill & \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill \\ & & & & & & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & & & & & \\ & & & & & & & & \hfill & 5^{}\hfill & \hfill \end{array}$$
(4.72)
Next, the single $`4^{}`$ is moved under $`\varphi _4^{}`$ as follows:
$$\begin{array}{ccccccccccccccccccc}\hfill & 1^{}\hfill & \hfill & 1\hfill & \hfill & \overline{3}^{}\hfill & \hfill & \overline{\mathrm{𝟏}}\hfill & \hfill & \overline{\mathrm{𝟏}}\hfill & \hfill & \mathrm{𝟑}\hfill & \hfill & \overline{4}\hfill & \hfill & \overline{4}\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & \overline{2}^{}\hfill & \hfill & \overline{2}\hfill & \hfill & \overline{3}\hfill & \hfill & \overline{3}\hfill & \hfill & \mathrm{𝟒}^{}\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & \\ & & & & \hfill & \overline{3}\hfill & \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill \\ & & & & & & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & & & & & \\ & & & & & & & & \hfill & 5^{}\hfill & \hfill \end{array}\begin{array}{ccccccccccccccccccc}\hfill & 1^{}\hfill & \hfill & 1\hfill & \hfill & \overline{3}^{}\hfill & \hfill & \mathrm{𝟒}^{}\hfill & \hfill & \overline{\mathrm{𝟏}}\hfill & \hfill & \overline{\mathrm{𝟏}}\hfill & \hfill & \overline{4}\hfill & \hfill & \overline{4}\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & \overline{2}^{}\hfill & \hfill & \overline{2}\hfill & \hfill & \overline{3}\hfill & \hfill & \overline{3}\hfill & \hfill & \mathrm{𝟑}\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & \\ & & & & \hfill & \overline{3}\hfill & \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill \\ & & & & & & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & & & & & \\ & & & & & & & & \hfill & 5^{}\hfill & \hfill \end{array}$$
(4.73)
There are no $`\overline{4}^{}`$s. However, the $`4`$th column contains the pair $`\overline{3}\mathrm{\hspace{0.17em}3}`$ which must be replaced by $`4^{}\overline{4}^{}`$ under the action of $`\chi _4^{}`$:
$$\begin{array}{ccccccccccccccccccc}\hfill & 1^{}\hfill & \hfill & 1\hfill & \hfill & \overline{3}^{}\hfill & \hfill & 4^{}\hfill & \hfill & \overline{1}\hfill & \hfill & \overline{1}\hfill & \hfill & \overline{4}\hfill & \hfill & \overline{4}\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & \overline{2}^{}\hfill & \hfill & \overline{2}\hfill & \hfill & \overline{\mathrm{𝟑}}\hfill & \hfill & \overline{3}\hfill & \hfill & 3\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & \\ & & & & \hfill & \overline{3}\hfill & \hfill & \mathrm{𝟑}\hfill & \hfill & \overline{4}\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill \\ & & & & & & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & & & & & \\ & & & & & & & & \hfill & 5^{}\hfill & \hfill \end{array}\begin{array}{ccccccccccccccccccc}\hfill & 1^{}\hfill & \hfill & 1\hfill & \hfill & \overline{3}^{}\hfill & \hfill & 4^{}\hfill & \hfill & \overline{1}\hfill & \hfill & \overline{1}\hfill & \hfill & \overline{4}\hfill & \hfill & \overline{4}\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & \overline{2}^{}\hfill & \hfill & \overline{2}\hfill & \hfill & \mathrm{𝟒}^{}\hfill & \hfill & \overline{3}\hfill & \hfill & 3\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & \\ & & & & \hfill & \overline{3}\hfill & \hfill & \overline{\mathrm{𝟒}}^{}\hfill & \hfill & \overline{4}\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill \\ & & & & & & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & & & & & \\ & & & & & & & & \hfill & 5^{}\hfill & \hfill \end{array}$$
(4.74)
There are no $`\overline{5}^{}`$s and it is then important to notice that one does not replace the vertical pair $`\overline{4}\mathrm{\hspace{0.17em}4}`$ in the $`5`$th column by $`5^{}\overline{5}^{}`$ under the action of $`\chi _5^{}`$ because one must first apply $`\psi _5^{}`$ to the two $`5^{}`$s. In any case the premature action of $`\chi _5^{}`$ would lead to two $`5^{}`$s in the same row, which is forbidden. Instead, the algorithm dictates that one first acts on the two $`5^{}`$s with $`\psi _5^{}`$. The uppermost $`5^{}`$ is moved into the $`5`$th column and up that column until it is just below the entry $`\overline{1}`$ which cannot be moved into the second row. This leaves both the $`\overline{3}`$ and $`4`$ in their own rows in the $`5`$th column, and no $`\overline{4}\mathrm{\hspace{0.17em}4}`$ pair.
$$\begin{array}{ccccccccccccccccccc}\hfill & 1^{}\hfill & \hfill & 1\hfill & \hfill & \overline{3}^{}\hfill & \hfill & 4^{}\hfill & \hfill & \overline{1}\hfill & \hfill & \overline{1}\hfill & \hfill & \overline{4}\hfill & \hfill & \overline{4}\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & \overline{2}^{}\hfill & \hfill & \overline{2}\hfill & \hfill & 4^{}\hfill & \hfill & \overline{\mathrm{𝟑}}\hfill & \hfill & 3\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & \\ & & & & \hfill & \overline{3}\hfill & \hfill & \overline{4}^{}\hfill & \hfill & \overline{\mathrm{𝟒}}\hfill & \hfill & \mathrm{𝟓}^{}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill \\ & & & & & & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & & & & & \\ & & & & & & & & \hfill & 5^{}\hfill & \hfill \end{array}\begin{array}{ccccccccccccccccccc}\hfill & 1^{}\hfill & \hfill & 1\hfill & \hfill & \overline{3}^{}\hfill & \hfill & 4^{}\hfill & \hfill & \overline{1}\hfill & \hfill & \overline{1}\hfill & \hfill & \overline{4}\hfill & \hfill & \overline{4}\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & \overline{2}^{}\hfill & \hfill & \overline{2}\hfill & \hfill & 4^{}\hfill & \hfill & \mathrm{𝟓}^{}\hfill & \hfill & 3\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & \\ & & & & \hfill & \overline{3}\hfill & \hfill & \overline{4}^{}\hfill & \hfill & \overline{\mathrm{𝟑}}\hfill & \hfill & \overline{\mathrm{𝟒}}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill \\ & & & & & & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & & & & & \\ & & & & & & & & \hfill & 5^{}\hfill & \hfill \end{array}$$
(4.75)
The second $`5^{}`$ does not move under the action of $`\psi _5^{}`$ since the $`4`$ immediately above it cannot move into the $`5`$th row. The final result can then be seen to be, as claimed, the juxtaposition of a primed tableaux $`PD𝒫𝒟^{54321}(5,\overline{5})`$ and an unprimed tableaux $`T𝒯^{433}(5,\overline{5})`$:
$$\begin{array}{ccccccccccccccccccc}\hfill & 1^{}\hfill & \hfill & 1\hfill & \hfill & \overline{3}^{}\hfill & \hfill & 4^{}\hfill & \hfill & \overline{1}\hfill & \hfill & \overline{1}\hfill & \hfill & \overline{4}\hfill & \hfill & \overline{4}\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & \overline{2}^{}\hfill & \hfill & \overline{2}\hfill & \hfill & 4^{}\hfill & \hfill & 5^{}\hfill & \hfill & 3\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & \\ & & & & \hfill & \overline{3}\hfill & \hfill & \overline{4}^{}\hfill & \hfill & \overline{3}\hfill & \hfill & \overline{4}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill \\ & & & & & & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & & & & & \\ & & & & & & & & \hfill & 5^{}\hfill & \hfill \end{array}\begin{array}{ccccccccccc}\hfill & 1^{}\hfill & \hfill & 1\hfill & \hfill & \overline{3}^{}\hfill & \hfill & 4^{}\hfill & \hfill & \overline{1}\hfill & \hfill \\ & & \hfill & \overline{2}^{}\hfill & \hfill & \overline{2}\hfill & \hfill & 4^{}\hfill & \hfill & 5^{}\hfill & \hfill \\ & & & & \hfill & \overline{3}\hfill & \hfill & \overline{4}^{}\hfill & \hfill & \overline{3}\hfill & \hfill \\ & & & & & & \hfill & 4\hfill & \hfill & 4\hfill & \hfill \\ & & & & & & & & \hfill & 5^{}\hfill & \hfill \end{array}\begin{array}{ccccccccc}\hfill & \overline{1}\hfill & \hfill & \overline{4}\hfill & \hfill & \overline{4}\hfill & \hfill & 5\hfill & \hfill \\ \hfill & 3\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & \\ \hfill & \overline{4}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill \end{array}$$
(4.76)
### 4.3 Corollaries
By associating $`x_k`$, $`t^2\overline{x}_k`$, $`y_k`$ and $`t^2\overline{y}_k`$ to each entry $`k`$, $`\overline{k}`$, $`k^{}`$ and $`\overline{k}^{}`$, respectively, in the various tableaux $`QST`$, $`QD`$ and $`T`$ appearing in Theorem 4.1 we immediately have the following corollary.
###### Corollary 4.2
Let $`\mu =\lambda +\delta `$ be a strict partition of length $`\mathrm{}(\mu )=n`$, with $`\lambda `$ a partition of length $`\mathrm{}(\lambda )n`$ and $`\delta =(n,n1,\mathrm{},1)`$.
$$\underset{QST𝒬𝒮𝒯^\mu (n)}{}t^{2\mathrm{b}\mathrm{a}\mathrm{r}(QST)}(𝐱/𝐲)^{\mathrm{wgt}(QST)}=\underset{QD𝒬𝒟^\delta (n)}{}t^{2\mathrm{b}\mathrm{a}\mathrm{r}(QD)}(𝐱/𝐲)^{\mathrm{wgt}(QD)}\underset{T𝒯^\lambda (n)}{}t^{2\mathrm{b}\mathrm{a}\mathrm{r}(T)}𝐱^{\mathrm{wgt}(T)}.$$
(4.77)
Thanks to the definition of $`Q(𝐱/𝐲;t)`$ given in (2.25), the identity (2.23) and the combinatorial definition of the $`t`$-deformation $`sp_\lambda (𝐱;t)`$ of symplectic characters given in (2.29), the above result is precisely our second main result Proposition 1.2.
Other corollaries follow as special cases of these results. Setting $`\lambda =0`$ we obtain
$$Q_\delta (𝐱/𝐲;t)=\underset{1ijn}{}(x_i+t^2\overline{x}_i+y_j+t^2\overline{y}_j).$$
(4.78)
On the other hand the case $`𝐲=t𝐱=(tx_1,tx_2,\mathrm{},tx_n)`$ of (1.10) is equivalent to the $`t`$-deformation of Weyl’s denominator formula (1.9) for the Lie algebra $`sp(2n)`$ derived elsewhere \[HK02, HK05\] by much more circuitous means.
###### Corollary 4.3
$$\begin{array}{c}_{i=1}^n(x_i+t\overline{x}_i)_{1i<jn}(x_i+t^2\overline{x}_i+tx_j+t\overline{x}_j)sp_\lambda (𝐱;t)\hfill \\ \\ =_{ST𝒮𝒯^\mu (n\overline{n})}t^{\mathrm{var}(ST)+\mathrm{bar}(ST)}(1+t)^{\mathrm{str}(ST)n}𝐱^{\mathrm{wgt}(ST)}.\hfill \end{array}$$
(4.79)
Proof First it should be noted that
$$\underset{1ijn}{}(x_i+t^2\overline{x}_i+tx_j+t\overline{x}_j)=\underset{i=1}{\overset{n}{}}(1+t)(x_i+t\overline{x}_i)\underset{1i<jn}{}(1+t\overline{x}_i\overline{x}_j)(1+t\overline{x}_i\overline{x}_j),$$
(4.80)
which includes a factor $`(1+t)^n`$.
Then it suffices to recognise that deleting primes from the entries $`k^{}`$ and $`\overline{k}^{}`$ in $`QST𝒬𝒮𝒯^\mu (n,\overline{n})`$ gives a symplectic shifted tableau $`ST𝒮𝒯^\mu (n,\overline{n})`$ with a factor of $`t`$ arising from each primed entry since $`y_k=tx_k`$ and $`t^2y_k^1=tx_k^1`$. Additional factors of $`t^2`$ arise from each barred entry $`k`$ since these are associated with $`t^2x_k^1`$. Compulsorily primed entries in each primed shifted tableau $`QST`$ corresponding to a fixed shifted tableau $`ST`$ appear once and only once in each row of each connected strip of identical entries, whether barred or unbarrred, while the leftmost lowest entry of each connected strip may each be primed or unprimed. To summarise, each connected component of a $`k`$-strip gives rise to a factor $`(1+t)t^{\mathrm{row}_k1}`$, where $`\mathrm{row}_k`$ is the number of rows of the $`k`$-strip component, and each component of a $`\overline{k}`$-strip gives rise to a factor of $`(1+t^1)t^{2\mathrm{b}\mathrm{a}\mathrm{r}_{\overline{k}}\mathrm{row}_{\overline{k}}+1}=(1+t)t^{\mathrm{bar}_{\overline{k}}+\mathrm{col}_{\overline{k}}1}`$ where $`\mathrm{bar}_{\overline{k}}`$ is the length of the component of the $`\overline{k}`$-strip, while $`\mathrm{row}_{\overline{k}}`$ and $`\mathrm{col}_{\overline{k}}`$ are the numbers of rows and columns, respectively, that it occupies, with $`\mathrm{bar}_{\overline{k}}=\mathrm{row}_{\overline{k}}+\mathrm{col}_{\overline{k}}1`$. Combining all these factors for $`k=1,2,\mathrm{},n`$ gives $`(1+t)^{\mathrm{str}(ST)}t^{\mathrm{var}(ST)+\mathrm{bar}(ST)}`$, as required, since as defined earlier $`\mathrm{var}(ST)=(\mathrm{row}_k+\mathrm{col}_{\overline{k}}1)`$ where the sum is over all connected components of all $`k`$ and $`\overline{k}`$ strips, for all $`k`$, and $`\mathrm{bar}(ST)=\mathrm{bar}_{\overline{k}}`$ is the total number of barred entries in $`ST`$.
Another significant corollary involves a link with U-turn alternating sign matrices. This is provided in Section 5.
## 5 Connection to Alternating Sign Matrices
### 5.1 $`gl(n)`$ case
In this section we show how to map from shifted tableaux, $`ST`$, to alternating sign matrices. Using the analogous relationship for primed shifted tableaux, $`PST`$, a result of Chapman \[C01\] is a straightforward consequence of Theorem 3.1.
An alternating sign matrix (ASM) is an $`n\times n`$ matrix filled with $`0`$’s, $`1`$’s, and $`1`$’s such that the first and last nonzero entries of each row and column are $`1`$’s and the nonzero entries within a row or column alternate in sign. There is a famous formula, conjectured by Mills, Robbins, and Rumsey \[MRR83\] and proved by Zeilberger \[Z96\], that counts the number of ASM of size $`n`$ as $`_{j=0}^{n1}(3j+1)!/(n+j)!`$. See also Bressoud \[B99\].
We work with a generalisation of ASM called $`\mu `$–ASM introduced by Okada \[O93\] that can be associated with semistandard shifted tableaux. The new feature here is that the alternating sign matrix is no longer square. Its row sums are all $`1`$ but its column sums are $`1`$ or $`0`$ according as the column label is or is not a part of some partition $`\mu `$. To be more precise, given a partition $`\mu `$ with distinct parts and such that $`\mathrm{}(\mu )=n`$ and $`\mu _1m`$ for some $`mn`$, the set $`𝒜^\mu (n)`$ of $`\mu `$–alternating sign matrices is the set of $`n\times m`$ matrices $`A=(a_{ij})_{1i,jn}`$ that satisfy the following conditions:
| ASM1 | $`a_{iq}\{1,0,1\}`$ for $`1in,1qm`$; |
| --- | --- |
| ASM2 | $`_{q=p}^ma_{iq}\{0,1\}`$ for $`1in,1pm`$; |
| ASM3 | $`_{i=j}^na_{iq}\{0,1\}`$ for $`1jn,1qm`$ |
| ASM4 | $`_{q=1}^ma_{iq}=1`$ for $`1in`$; |
| ASM5 | $`_{i=1}^na_{iq}=1`$ if $`q=\mu _j`$ for some $`j`$; or $`_{i=1}^na_{iq}=0`$ otherwise; for $`1qm`$. |
In what follows we also require U-turn alternating sign matrices, UASMs, and their generalisation $`\mu `$–UASMs that are associated with semistandard shifted symplectic tableaux \[HK02, HK05\]. In fact the bijection between semistandard shifted tableaux $`ST𝒮𝒯^\mu (n)`$ and $`\mu `$–ASMs, $`A𝒜^\mu (n)`$, is a special case of a bijection between semistandard shifted symplectic tableaux $`ST𝒮𝒯^\mu (n,\overline{n})`$ and $`\mu `$–UASMs $`A𝒜^\mu (n,\overline{n})`$ \[HK05\].
Briefly, in the $`\mu `$–ASM case, we associate to each semistandard shifted tableaux $`ST𝒮𝒯^\mu (n)`$ of shape $`\mu `$ with $`\mathrm{}(\mu )=n`$ and $`\mu _1=m`$ an $`n\times m`$ matrix $`M(ST)`$ filled with the entries from $`ST`$ together with zeros such that if there is an $`i`$ on diagonal $`j`$ of $`ST`$ (where the main diagonal is diagonal 1 and the last box in the first row is in diagonal $`\mu _1=m`$), then there is an $`i`$ in row $`i`$, column $`j`$ of the matrix. All other entries are zero.
For example, in the case $`n=6`$ and $`\mu =(9,8,6,4,3,1)`$ a given semistandard shifted tableau $`ST`$ of shape $`\mu `$ yields a $`6\times 9`$ matrix, $`M(ST)`$, as shown:
$$ST=\begin{array}{ccccccccccccccccccc}\hfill & 1\hfill & \hfill & 1\hfill & \hfill & 1\hfill & \hfill & 2\hfill & \hfill & 3\hfill & \hfill & 3\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill \\ & & \hfill & 2\hfill & \hfill & 2\hfill & \hfill & 2\hfill & \hfill & 3\hfill & \hfill & 4\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill \\ & & & & \hfill & 3\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & 5\hfill & \hfill & 6\hfill & \hfill & & \\ & & & & & & \hfill & 4\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill & 6\hfill & \hfill & & \\ & & & & & & & & \hfill & 5\hfill & \hfill & 6\hfill & \hfill & 6\hfill & \hfill \\ & & & & & & & & & & \hfill & 6\hfill & \hfill & & \end{array}M(ST)=\left[\begin{array}{ccccccccc}1\hfill & 1\hfill & 1\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 2\hfill & 2\hfill & 2\hfill & 2\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 3\hfill & 0\hfill & 0\hfill & 3\hfill & 3\hfill & 3\hfill & 0\hfill & 0\hfill & 0\hfill \\ 4\hfill & 4\hfill & 4\hfill & 4\hfill & 4\hfill & 0\hfill & 4\hfill & 4\hfill & 4\hfill \\ 5\hfill & 5\hfill & 5\hfill & 0\hfill & 5\hfill & 5\hfill & 5\hfill & 5\hfill & 0\hfill \\ 6\hfill & 6\hfill & 6\hfill & 6\hfill & 0\hfill & 6\hfill & 0\hfill & 0\hfill & 0\hfill \end{array}\right]$$
(5.81)
A primed semistandard shifted tableau $`PST𝒫𝒮𝒯^\mu (N)`$ yields a similar matrix $`M(PST)`$ in the same way:
$$PST=\begin{array}{ccccccccccccccccccc}\hfill & 1\hfill & \hfill & 1\hfill & \hfill & 1\hfill & \hfill & 2^{}\hfill & \hfill & 3^{}\hfill & \hfill & 3\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill \\ & & \hfill & 2\hfill & \hfill & 2\hfill & \hfill & 2\hfill & \hfill & 3^{}\hfill & \hfill & 4^{}\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 5\hfill & \hfill \\ & & & & \hfill & 3\hfill & \hfill & 4^{}\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & 5\hfill & \hfill & 6\hfill & \hfill & & \\ & & & & & & \hfill & 4\hfill & \hfill & 5^{}\hfill & \hfill & 5\hfill & \hfill & 6^{}\hfill & \hfill & & \\ & & & & & & & & \hfill & 5\hfill & \hfill & 6^{}\hfill & \hfill & 6\hfill & \hfill \\ & & & & & & & & & & \hfill & 6\hfill & \hfill & & \end{array}M(PST)=\left[\begin{array}{ccccccccc}1\hfill & 1\hfill & 1\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 2\hfill & 2\hfill & 2\hfill & 2^{}\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 3\hfill & 0\hfill & 0\hfill & 3^{}\hfill & 3^{}\hfill & 3\hfill & 0\hfill & 0\hfill & 0\hfill \\ 4\hfill & 4^{}\hfill & 4\hfill & 4\hfill & 4^{}\hfill & 0\hfill & 4\hfill & 4\hfill & 4\hfill \\ 5\hfill & 5^{}\hfill & 5\hfill & 0\hfill & 5\hfill & 5^{}\hfill & 5\hfill & 5\hfill & 0\hfill \\ 6\hfill & 6^{}\hfill & 6\hfill & 6^{}\hfill & 0\hfill & 6\hfill & 0\hfill & 0\hfill & 0\hfill \end{array}\right]$$
(5.82)
where as we shall see it is possible to distinguish various types of entry $`0`$ as characterised by their set of nearest non-vanishing neighbours.
Each of these matrices can be converted into a $`\mu `$–alternating sign matrix by replacing the rightmost entry of each continuous sequence of nonzero entries by a $`1`$ and each zero immediately to the left of a nonzero entry by $`1`$, leaving all other entries $`0`$. In the case of the above example we obtain in this way
$$A=\left[\begin{array}{ccccccccc}\hfill 0& \hfill 0& \hfill 1& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0\\ \hfill 0& \hfill 0& \hfill 0& \hfill 1& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0\\ \hfill 1& \hfill 0& \hfill 1& \hfill 0& \hfill 0& \hfill 1& \hfill 0& \hfill 0& \hfill 0\\ \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 1& \hfill 1& \hfill 0& \hfill 0& \hfill 1\\ \hfill 0& \hfill 0& \hfill 1& \hfill 1& \hfill 0& \hfill 0& \hfill 0& \hfill 1& \hfill 0\\ \hfill 0& \hfill 0& \hfill 0& \hfill 1& \hfill 1& \hfill 1& \hfill 0& \hfill 0& \hfill 0\end{array}\right]𝒜^{986431}(6)$$
(5.83)
Square ice provides a further refinement of the relationship between shifted tableaux and $`\mu `$–ASM. Square ice is a directed graph that models the orientation of oxygen and hydrogen molecules in frozen water. The vertices are laid out in an $`n\times n`$ grid and each vertex has two incoming and two outgoing edges in a north, south, east, west orientation. The square ice graph corresponding to (5.83) appears in Figure 1.
At each vertex there are six possible orientations of the four directed edges. These orientations may be specified by the pairs of compass points giving the directions of the incoming edges. In this way the above square ice graph is specified by a corresponding “compass points” matrix:
$$CM=\left[\begin{array}{ccccccccc}\hfill NE& \hfill NE& \hfill WE& \hfill NW& \hfill NW& \hfill NW& \hfill NW& \hfill NW& \hfill NW\\ \hfill NE& \hfill NE& \hfill SE& \hfill WE& \hfill NW& \hfill NW& \hfill NW& \hfill NW& \hfill NW\\ \hfill WE& \hfill NW& \hfill NS& \hfill SE& \hfill NE& \hfill WE& \hfill NW& \hfill NW& \hfill NW\\ \hfill SE& \hfill NE& \hfill NE& \hfill SE& \hfill WE& \hfill NS& \hfill NE& \hfill NE& \hfill WE\\ \hfill SE& \hfill NE& \hfill WE& \hfill NS& \hfill SE& \hfill NE& \hfill NE& \hfill WE& \hfill SW\\ \hfill SE& \hfill NE& \hfill SE& \hfill WE& \hfill NS& \hfill WE& \hfill NW& \hfill SW& \hfill SW\end{array}\right]$$
(5.84)
The bijection between compass point matrices, square ice graphs and $`\mu `$-ASMs is provided by the following correspondences:
The horizontal orientation (with both horizontal edges directed in), $`WE`$, corresponds to each entry $`+1`$ in $`A`$, and the vertical orientation (with both vertical edges directed in), $`NS`$, corresponds to each entry $`1`$ in $`A`$; the other four orientations, $`NE`$, $`SW`$, $`NW`$ and $`SE`$ correspond to the entries $`0`$ in $`A`$. Accordingly there are northwest zeros (with edges pointing in the north and west directions), southwest zeros, northeast zeros, and southeast zeros. Northwest zeros are those whose nearest nonzero neighbour to the right, if it has one, is $`1`$, and whose nearest nonzero neighbour below is $`1`$. Southwest zeros are those whose nearest nonzero neighbour to the right, if it has one, is $`1`$, and whose nearest nonzero neighbour below, if it has one, is $`1`$. Northeast zeros are those whose nearest nonzero neighbour to the right is $`1`$, and whose nearest nonzero neighbour below is $`1`$. Southeast zeros are those whose nearest nonzero neighbour to the right is $`1`$, and whose nearest nonzero neighbour below, if it has one, is $`1`$.
The compass points matrices $`CM`$ can then be associated to the set of all primed shifted tableaux $`PST`$ that may be obtained by adding primes to the entries of the unprimed tableau $`ST`$. For example, the entries $`NE`$ in the $`k`$th row are associated with an entry $`k`$ in $`PST`$ and correspondingly to a weight factor $`x_k`$. The entries $`SE`$ in the $`k`$th row are associated with an entry $`k^{}`$ in $`PST`$ and correspondingly to a weight factor $`y_k`$. The entries $`NS`$ in the $`k`$th row are associated with the two possible labels $`k`$ and $`k^{}`$ of the first box of each connected component of $`\mathrm{str}_k(PST)`$ other than the one starting on the main diagonal. Correspondingly each $`NS`$ in row $`k`$ is associated with a weight factor $`(x_k+y_k)`$. It should be pointed out that the main diagonal is not included at all in the compass points matrix so that the first column corresponds to the second diagonal and indeed in general, column $`k`$ of $`CM`$ corresponds to diagonal $`k+1`$ of $`PST`$. This implies that the above weighting excludes the weight $`x_1x_2\mathrm{}x_n`$ arising from the entries $`1,2,\mathrm{},n`$ on the main diagonal of each $`PST`$.
Combining the weight factors we have a total weight associated with each $`A𝒜^\mu (n)`$ given by
$$\underset{k=1}{\overset{n}{}}x_k^{NE_k(A)}y_k^{SE_k(A)}(x_k+y_k)^{NS_k(A)}$$
(5.85)
where $`NE_k(A)`$, $`SE_k(A)`$ and $`NS_k(A)`$ are the numbers of entries $`NE`$, $`SE`$ and $`NS`$ in the $`k`$th row of the compass matrix $`CM(A)`$ corresponding to $`A`$.
Thanks to the connection already made between $`PST`$s and weighted $`ST`$s, the following is then an immediate corollary of Proposition 1.1:
###### Corollary 5.1
Let $`\mu =\lambda +\delta `$ be a strict partition of length $`\mathrm{}(\mu )=n`$, with $`\lambda `$ a partition of length $`\mathrm{}(\lambda )<n`$ and $`\delta =(n,n1,\mathrm{},1)`$. Then for all $`𝐱=(x_1,x_2,\mathrm{},x_n)`$ and $`𝐲=(y_1,y_2,\mathrm{},y_n)`$ we have
$$\underset{1i<jn}{}(x_i+y_j)s_\lambda (𝐱)=\underset{A𝒜^\mu (n)}{}\underset{k=1}{\overset{n}{}}x_k^{NE_k(A)}y_k^{SE_k(A)}(x_k+y_k)^{NS_k(A)}.$$
(5.86)
This generalises a result of Chapman \[C01\]. In his original paper he weights by column instead of row so the parameters in his paper correspond to the transpose matrix. Now setting $`\lambda =0`$ so that $`\mu =\delta `$, and noting that $`𝒜^\delta (n)=𝒜(n)`$, the set of all $`n\times n`$ ASMs, we have
###### Corollary 5.2 (Chapman \[C01\])
$$\underset{1i<jn}{}(x_i+y_j)=\underset{A𝒜(n)}{}\underset{k=1}{\overset{n}{}}x_k^{NE_k(A)}y_k^{SE_k(A)}(x_k+y_k)^{NS_k(A)}.$$
(5.87)
Corollary 5.86 has a further consequence:
###### Corollary 5.3
Let $`\mu =\lambda +\delta `$ be a strict partition of length $`\mathrm{}(\mu )=n`$, with $`\lambda `$ a partition of length $`\mathrm{}(\lambda )n`$ and $`\delta =(n,n1,\mathrm{},1)`$. For any $`m`$ for which $`m>n`$ and $`\mu _1m`$, let $`𝒜(n,m,\mu )𝒜(m)`$ be the subset consisting of those ASMs, $`C`$, whose top $`n`$ rows constitute an ASM, $`A`$, in $`𝒜^\mu (n)`$. Then for all $`(x_1,\mathrm{},x_n,x_{n+1},\mathrm{},x_m)`$ and $`(y_1,\mathrm{},y_n,y_{n+1},\mathrm{},y_m)`$ we have
$`{\displaystyle \underset{1i<jn}{}}(x_i+y_j)s_\lambda (x_1,\mathrm{},x_n){\displaystyle \underset{n+1i<jm}{}}(x_i+y_j)s_\kappa (y_{n+1},\mathrm{},y_m)`$ (5.88)
$`={\displaystyle \underset{C𝒜(n,m,\mu )}{}}{\displaystyle \underset{k=1}{\overset{m}{}}}x_k^{NE_k(C)}y_k^{SE_k(C)}(x_k+y_k)^{NS_k(C)},`$ (5.89)
where $`\kappa `$ is the conjugate of the complement of $`\lambda `$ with respect to the rectangular partition $`((mn)^n)`$, that is $`\kappa =((mn)^n/\lambda )^{}`$.
Proof Let the top $`n`$ rows of $`C`$ and the bottom $`(mn)`$ rows of $`C`$, reversed in order, form the matrices $`A`$ and $`B`$, respectively. Then the application of Corollary 5.86 to $`A`$ gives the contribution made by the top $`n`$ rows of each $`C`$ on the right hand side of (5.89) in the form of the first two factors on the left hand side. Similarly, the remaining two factors on the left hand side arise from the contribution of the bottom $`(mn)`$ rows of each $`C`$. To see this one applies Corollary 5.86 to $`B`$, but this time with $`\mu =\lambda +\delta `$ replaced by $`\nu =\kappa +ϵ`$, where $`ϵ=(mn,mn1,\mathrm{},1)`$, and with $`(x_1,\mathrm{},x_n)`$ and $`(y_1,\mathrm{},y_n)`$ replaced by $`(y_m,,\mathrm{},y_{n+1})`$ and $`(x_m,,\mathrm{},x_{n+1})`$, respectively. It only remains to relate $`\lambda `$ and $`\kappa `$. Since the parts of $`\mu `$ and $`\nu `$ specify those columns of $`A`$ and $`B`$, respectively, whose column sums are $`1`$, and $`A`$ and $`B`$ are constructed from an ASM $`C`$, all these parts must be distinct and together constitute $`(m,m1,\mathrm{},1)`$. It follows that the union of $`\{\lambda _i+ni+1|i=1,\mathrm{},n\}`$ and $`\{\kappa _j+(mn)j+1|j=1,\mathrm{},mn\}`$ must be $`\{1,\mathrm{},m\}`$. However, it is well known \[M95\]p3 that the complement of $`\{\lambda _i+ni+1|i=1,\mathrm{},n\}`$ in $`\{1,\mathrm{},m\}`$ is $`\{n+k\lambda _k^{}|k=1,\mathrm{},mn\}`$. By setting $`k=mnj+1`$ it can then be seen that $`\kappa _j=n\lambda _{mnj+1}^{}`$ for $`j=1,\mathrm{},mn`$, so that $`\kappa =(n^{mn}/\lambda ^{}=((mn)^n/\lambda )^{}`$, as claimed.
Alternatively, Corollary 5.3, may be proved bijectively by taking each primed shifted tableau $`PST`$ specified by some $`C𝒜(n,m,\mu )`$ and using the jeu de taquin, first as described above, to move all entries $`k^{}`$ with $`1kn`$ north-west to the $`k`$th column, and then in an analogous manner, to move all entries $`k`$ with $`n+1km`$ south-east to the $`k`$th row. The result is a pair of triangular subtableaux, both of type $`PD`$ but with all entries $`k`$ and $`k^{}`$ such that $`kn`$ in one case and $`k>n`$ in the other, together with a pair of semistandard tableaux, one of shape $`\lambda `$ with unprimed entries subject to the order relation $`1<2<\mathrm{}<n`$ and the other of shape $`\kappa `$ with primed entries subject to the order relation $`m^{}<(m1)^{}<\mathrm{}<(n+1)^{}`$.
By way of an example, let $`m=6`$ and consider the following $`6\times 6`$ ASM
$$C=\left[\begin{array}{cccccc}\hfill 0& \hfill 0& \hfill 1& \hfill 0& \hfill 0& \hfill 0\\ \hfill 0& \hfill 1& \hfill 1& \hfill 1& \hfill 0& \hfill 0\\ \hfill 1& \hfill 1& \hfill 1& \hfill 0& \hfill 0& \hfill 0\\ \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 1& \hfill 0\\ \hfill 0& \hfill 1& \hfill 0& \hfill 0& \hfill 1& \hfill 1\\ \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 1& \hfill 0\end{array}\right]𝒜(6).$$
(5.90)
Taking $`n=2`$, the top two rows of $`C`$ constitute $`A`$, and the bottom four rows of $`C`$, reversed in order, constitute $`B`$, where:
$$A=\left[\begin{array}{cccccc}\hfill 0& \hfill 0& \hfill 1& \hfill 0& \hfill 0& \hfill 0\\ \hfill 0& \hfill 1& \hfill 1& \hfill 1& \hfill 0& \hfill 0\end{array}\right]𝒜^{4,2}(2)\text{and}B=\left[\begin{array}{cccccc}\hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 1& \hfill 0\\ \hfill 0& \hfill 1& \hfill 0& \hfill 0& \hfill 1& \hfill 1\\ \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 1& \hfill 0\\ \hfill 1& \hfill 1& \hfill 1& \hfill 0& \hfill 0& \hfill 0\end{array}\right]𝒜^{6,5,3,1}(4).$$
(5.91)
The superscripts on $`𝒜^{4,2}(2)`$ and $`𝒜^{6,5,3,1}(4)`$ specify those columns of $`A`$ and $`B`$, respectively, having column sums $`1`$. They indicate that $`A𝒜(2,4,\mu )`$ with $`\mu =(4,2)`$ so that $`\lambda =(2,1)`$, while $`\nu =(6,5,3,1,)`$ so that $`\kappa =(2,2,1)`$. This is in accordance with the formula $`\kappa =(4^2/\lambda )^{}=(3,2)^{}=(2,2,1)`$.
The compass point matrix corresponding to $`C`$ takes the form:
$$CM=\left[\begin{array}{ccccccccc}\hfill NE& \hfill NE& \hfill WE& \hfill NW& \hfill NW& \hfill NW& & & \\ \hfill NE& \hfill WE& \hfill NS& \hfill WE& \hfill NW& \hfill NW& & & \\ \hfill WE& \hfill NS& \hfill WE& \hfill SW& \hfill NW& \hfill NW& & & \\ \hfill SE& \hfill NE& \hfill SE& \hfill SE& \hfill WE& \hfill NW& & & \\ \hfill SE& \hfill WE& \hfill SW& \hfill SW& \hfill NS& \hfill WE& & & \\ \hfill SE& \hfill SE& \hfill SE& \hfill SE& \hfill WE& \hfill SW& & & \end{array}\right]$$
(5.92)
so that the contribution of $`C`$ to the right hand side of (5.89) is
$$x_1^2x_2(x_2+y_2)(x_3+y_3)x_4y_4^3y_5(x_5+y_5)y_6^4.$$
(5.93)
The three factors $`(x_k+y_k)`$ arise from three entries $`NS`$ in $`CM`$, that themselves arise from the three entries $`1`$ in $`C`$. There must be therefore be precisely $`2^3`$ primed shifted tableaux $`PST`$ corresponding to $`C`$. Choosing just one of these for illustrative purposes, the use of the jeu de taquin to move all $`k^{}`$s with $`k2`$ north-west and all $`k`$s with $`k>2`$ south-east, gives the following bijective map:
$$PST=\begin{array}{ccccccccccccc}\hfill & 1\hfill & \hfill & 1\hfill & \hfill & 1\hfill & \hfill & 2^{}\hfill & \hfill & 4^{}\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & 2\hfill & \hfill & 2\hfill & \hfill & 3\hfill & \hfill & 4^{}\hfill & \hfill & 6^{}\hfill & \hfill \\ & & & & \hfill & 3\hfill & \hfill & 4^{}\hfill & \hfill & 4\hfill & \hfill & 6^{}\hfill & \hfill \\ & & & & & & \hfill & 4\hfill & \hfill & 5^{}\hfill & \hfill & 6^{}\hfill & \hfill \\ & & & & & & & & \hfill & 5\hfill & \hfill & 6^{}\hfill & \hfill \\ & & & & & & & & & & \hfill & 6\hfill & \hfill \end{array}\begin{array}{ccccccccccccc}\hfill & 1\hfill & \hfill & 2^{}\hfill & \hfill & 1\hfill & \hfill & 1\hfill & \hfill & 4^{}\hfill & \hfill & 6^{}\hfill & \hfill \\ & & \hfill & 2\hfill & \hfill & 2\hfill & \hfill & 4^{}\hfill & \hfill & 5^{}\hfill & \hfill & 6^{}\hfill & \hfill \\ & & & & \hfill & 3\hfill & \hfill & 4^{}\hfill & \hfill & 3\hfill & \hfill & 6^{}\hfill & \hfill \\ & & & & & & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & 6^{}\hfill & \hfill \\ & & & & & & & & \hfill & 5\hfill & \hfill & 5\hfill & \hfill \\ & & & & & & & & & & \hfill & 6\hfill & \hfill \end{array}$$
(5.94)
The corresponding contribution to the left hand side of (5.89) is then given by
$$x_1^2x_2y_2x_3x_4y_4^3x_5y_5y_6^4=y_2x_1^2x_2x_3x_4y_4x_5y_6^2y_4^2y_5y_6^2,$$
(5.95)
where the arrangement of the terms on the right exhibits the contributions to each of the four factors constituting the left hand side of (5.89). Both tableaux in (5.94) may be displayed, as shown below, in terms of suitably re-oriented subtableaux involving entries $`k`$ and $`k^{}`$, with all $`k2`$ in one case, and all $`k>2`$ on the other.
1112226666655544444331221126566543443
665441112missing-subexpressionmissing-subexpression22missing-subexpressionmissing-subexpression666665missing-subexpressionmissing-subexpression55444missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression443missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression3missing-subexpressionmissing-subexpression12missing-subexpressionmissing-subexpression2112missing-subexpressionmissing-subexpression6566missing-subexpressionmissing-subexpression543missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression44missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpression3
66544{\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr}}}\ \ \cdot\ \ {\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr}}}\ \ \longleftrightarrow\ \ {\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr}}}\ \ \cdot\ \ {\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 1&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 2&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr}}}\ \ \cdot\ \ {\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&&\hbox to8.9pt{\hss$\scriptstyle&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 3&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\leaders\hrule height=0.0pt\hfill&\hrulefill\cr}}}\ \ \cdot\ \ {\vbox{\offinterlineskip\halign{&\mystrut\vrule#&\mybox{\hss$\scriptstyle#$\hss}\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 6^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 5^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule&\hbox to8.9pt{\hss$\scriptstyle 4^{\prime}&\hbox{\vrule height=10.6pt,depth=2.0pt,width=0.0pt}\vrule\cr\hrulefill\cr}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}} (5.96)
This illustrates the outcome of applying the jeu de taquin to primed shifted tableaux corresponding to the submatrices $`A`$ and $`B`$ of the ASM $`C`$. The resulting contribution of the four final tableaux to the left hand side of (5.89), is then confirmed to be as given on the right hand side of (5.95).
### 5.2 $`sp(2n)`$ case
The symplectic case involves a modified alternating sign matrix called a U–turn $`\mu `$–ASM or $`\mu `$–UASM. Informally, the U–turn condition means that two consecutive rows and the U–turn between them must follow the $`\mu `$–ASM summation rules, ASM2–5; that is, the cumulative sum must be zero or one, and the total sum must be one. These $`\mu `$–UASM were first defined in Hamel and King \[HK02\] where they were called sp($`2n`$)–generalised alternating sign matrics. They are discussed at length in Hamel and King \[HK05\]. A formal definition is as follows:
Let $`\mu `$ be a partition of length $`\mathrm{}(\mu )=n`$, all of whose parts are distinct, and for which $`\mu _1m`$. Then the matrix $`UA=(a_{iq})_{1i2n,1qm}`$ is said to belong to the set $`𝒰𝒜^\mu (2n)`$ of $`\mu `$–alternating sign matrices with a U–turn boundary if it is a $`2n\times m`$ matrix whose elements $`a_{iq}`$ satisfy the conditions:
| UA1 | $`a_{iq}\{1,0,1\}`$ | for $`1i2n`$, $`1qm`$; |
| --- | --- | --- |
| UA2 | $`_{q=p}^ma_{iq}\{0,1\}`$ | for $`1i2n`$, $`1pm`$; |
| UA3 | $`_{i=j}^{2n}a_{iq}\{0,1\}`$ | for $`1j2n`$, $`1qm`$. |
| UA4 | $`_{q=1}^m(a_{2i1,q}+a_{2i,q})=1`$ | for $`1in`$; |
| UA5 | $`_{i=1}^{2n}a_{iq}=\{\begin{array}{cc}1\hfill & \text{if }q=\mu _k\text{ for some }k\hfill \\ 0\hfill & \text{otherwise}\hfill \end{array}`$ | for $`1qm`$, $`1kn`$. |
In the case $`\mu =\delta =(n,n1,\mathrm{},1)`$ and $`m=n`$, for which UA5 becomes $`_{i=1}^{2n}a_{iq}=1`$ for $`1qn`$, this definition is such that the set of $`\mu `$–UASM coincides with the set of U–turn alternating sign matrices, UASMs, defined by Kuperberg \[K02\].
As noted above, Hamel and King \[HK05\] established a bijection between $`\mu `$–UASM and semistandard shifted symplectic tableaux. An example of this association is illustrated below in the case $`n=5`$ and $`\mu =(9,7,6,2,1)`$:
$$\begin{array}{ccccccccccccccccccc}\hfill & \overline{1}\hfill & \hfill & 1\hfill & \hfill & \overline{2}\hfill & \hfill & 2\hfill & \hfill & \overline{3}\hfill & \hfill & \overline{3}\hfill & \hfill & \overline{4}\hfill & \hfill & 4\hfill & \hfill & 5\hfill & \hfill \\ & & \hfill & \overline{2}\hfill & \hfill & \overline{2}\hfill & \hfill & 2\hfill & \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & \overline{4}\hfill & \hfill & 4\hfill & \hfill & & \\ & & & & \hfill & 3\hfill & \hfill & \overline{4}\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & 4\hfill & \hfill \\ & & & & & & \hfill & 4\hfill & \hfill & 4\hfill & \hfill & & & & & & \\ & & & & & & & & \hfill & \overline{5}\hfill & \hfill \end{array}\left[\begin{array}{ccccccccc}\hfill \overline{1}& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0\\ \hfill 0& \hfill 1& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0\\ \hfill \overline{2}& \hfill \overline{2}& \hfill \overline{2}& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0\\ \hfill 0& \hfill 0& \hfill 2& \hfill 2& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0\\ \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill \overline{3}& \hfill \overline{3}& \hfill 0& \hfill 0& \hfill 0\\ \hfill 3& \hfill 0& \hfill 0& \hfill 3& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0\\ \hfill 0& \hfill \overline{4}& \hfill 0& \hfill 0& \hfill \overline{4}& \hfill \overline{4}& \hfill \overline{4}& \hfill 0& \hfill 0\\ \hfill 4& \hfill 4& \hfill 4& \hfill 4& \hfill 4& \hfill 4& \hfill 4& \hfill 4& \hfill 0\\ \hfill \overline{5}& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0\\ \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 5\end{array}\right]\left[\begin{array}{ccccccccc}\hfill 1& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0\\ \hfill 1& \hfill 1& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0\\ \hfill 0& \hfill 0& \hfill 1& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0\\ \hfill 0& \hfill 1& \hfill 0& \hfill 1& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0\\ \hfill 0& \hfill 0& \hfill 0& \hfill 1& \hfill 0& \hfill 1& \hfill 0& \hfill 0& \hfill 0\\ \hfill 1& \hfill 0& \hfill 1& \hfill 1& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0\\ \hfill 1& \hfill 1& \hfill 0& \hfill 1& \hfill 0& \hfill 0& \hfill 1& \hfill 0& \hfill 0\\ \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 1& \hfill 0\\ \hfill 1& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0\\ \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 0& \hfill 1& \hfill 1\end{array}\right]$$
(5.97)
where the columns are labeled from left to right $`1,2,\mathrm{},9=m=\mu _1`$, and the rows from top to bottom $`\overline{1},1,\overline{2},2,\mathrm{},\overline{5},5=n`$.
The translation to square ice is also natural and just requires a modification of the left boundary by the insertion of a U–turn. The square ice graph in Figure 3 corresponds to the above $`\mu `$-UASM matrix. The same example appeared in Hamel and King \[HK05\].
In this symplectic case the bijection from the U–turn $`\mu `$-ASMs to U–turn square ice graphs is precisely as before, with entries $`+1`$ and $`1`$ mapped to $`WE`$ and $`NS`$ vertex orientations, and $`NE`$, $`SW`$, $`NW`$ and $`SE`$ entries $`0`$ distinguished by their nearest non-zero neighbouring entries. This map is encoded in the corresponding compass points matrix. For the above example, this takes the form:
$$CM=\left[\begin{array}{ccccccccc}WE& NW& NW& NW& NW& NW& NW& NW& NW\\ NS& WE& NW& NW& NW& NW& NW& NW& NW\\ NE& SE& WE& NW& NW& NW& NW& NW& NW\\ NW& NS& SE& WE& NW& NW& NW& NW& NW\\ NW& NW& SW& NS& NE& WE& NW& NW& NW\\ WE& NW& NS& WE& NW& SW& NW& NW& NW\\ NS& WE& NW& NS& NE& SE& WE& NW& NW\\ NE& SE& NE& NE& NE& SE& SE& WE& NW\\ WE& SW& NW& NW& NW& SW& SW& SW& NW\\ SW& SW& NW& NW& NW& SW& SW& NS& WE\end{array}\right]$$
(5.98)
Then we can generate a weighting in the same manner as for the $`gl(n)`$ case, with the $`k`$th column of $`CM`$ corresponding to the $`(k+1)`$th diagonal of $`QST`$. In this case we have unbarred entries corresponding to odd rows and barred entries corresponding to even rows. An entry $`NE`$ in the $`k`$th row is associated to an entry $`k`$ in $`QST`$ and correspondingly to a weight factor $`x_k`$. An entry $`SE`$ in row $`k`$ is associated to an entry $`k^{}`$ in $`QST`$ and correspondingly to a weight factor $`y_k`$. An entry $`NE`$ in row $`\overline{k}`$ is associated to an entry $`\overline{k}`$ in $`QST`$ and correspondingly to a weight factor $`t^2x_k^1`$. An entry $`SE`$ in row $`\overline{k}`$ is associated to an entry $`\overline{K}^{}`$ in $`QST`$ and correspondingly to $`t^2y_k^1`$. An entry $`NS`$ in the $`k`$th row is associated with the four possible labels $`k,k^{},\overline{k},\overline{k}^{}`$ of the first box of each connected component of $`str_k(QST)`$ (other than the one starting on the main diagonal) and correspondingly to a weight $`(x_i+y_i+t^2\overline{x}_i+t^2\overline{y}_i)`$. Combining the weight factors we have
$$\underset{k=1}{\overset{n}{}}x_k^{NE_k(A)}(t^2\overline{x}_k)^{NE_{\overline{k}}(A)}y_k^{SE_k(A)}(t^2\overline{y}_k)^{SE_{\overline{k}}(A)}(x_k+y_k)^{NS_k(A)}(t^2\overline{x}_k+t^2\overline{y}_k)^{NS_{\overline{k}}(A)}$$
where $`SE_k(A),NE_k(A),NS_k(A)`$ (resp. $`SE_{\overline{k}}(A),NE_{\overline{k}}(A),NS_{\overline{k}}(A)`$) are the numbers of entries $`SE,NE,NS`$ in row $`k`$ (resp. $`\overline{k}`$) of the compass matrix $`CM(A)`$.
We then have the following immediate corollary of Proposition 1.2:
###### Corollary 5.4
$`{\displaystyle \underset{1i<jn}{}}(x_i+t^2\overline{x}_i+y_j+t^2\overline{y}_j)sp_\lambda (x;t)`$ $`=`$ $`{\displaystyle \underset{A}{}}{\displaystyle \underset{k=1}{\overset{n}{}}}x_k^{NE_k(A)}(t^2\overline{x}_k)^{NE_{\overline{k}}(A)}y_k^{SE_k(A)}(t^2\overline{y}_k)^{SE_{\overline{k}}(A)}`$ (5.100)
$`\times (x_k+y_k)^{NS_k(A)}(t^2\overline{x}_k+t^2\overline{y}_k)^{NS_{\overline{k}}(A)}`$
This Corollary is a generalisation of Theorem 6.4 of Hamel and King \[HK05\]. This Theorem 6.4 may be recovered from Corollary 5.4 by setting $`𝐲=t𝐱`$, exploiting the bijection between compass point matrices $`CM(A)`$ and the U–turn $`\mu `$-ASM’s $`A`$, and noting that the number of entries $`NS`$ and $`WE`$ in any row of $`CM`$ are either the same or differ by one according to the nature, barred or unbarred, of the corresponding entry on the main diagonal of the associated semistandard shifted symplectic tableau $`ST`$. Note also that Theorem 6.4 includes the weighting for the main diagonal on each side of the equation, whereas Corollary 5.4 does not.
Acknowledgements
The existence of Corollary 5.3 was pointed out to us by an anonymous referee of an earlier version of this paper. The first author (AMH) acknowledges the support of a Discovery Grant from the Natural Sciences and Engineering Research Council of Canada, and the second (RCK) the support of a Leverhulme Emeritus Fellowship. |
warning/0507/hep-ph0507307.html | ar5iv | text | # A Unified Picture with Neutrino As a Central Feature Invited talk presented at the XI International Workshop on “Neutrino Telescopes” held at Venice, February 21–25, 2005, to appear in the proceedings.
## 1 Introduction
Since the discoveries (confirmations) of the atmospheric and solar neutrino oscillations , the neutrinos have emerged as being among the most effective probes into the nature of higher unification. Although almost the feeblest of all the entities of nature, simply by virtue of their tiny masses, they seem to possess a subtle clue to some of the deepest laws of nature pertaining to the unification-scale and (even more important) to the nature of the unification-symmetry. In this sense the neutrinos provide us with a rare window to view physics at truly short distances. As we will see, these turn out to be as short as about $`10^{30}`$ cm. In addition,it appears most likely that the origin of their tiny masses may be at the root of the origin of matter-antimatter asymmetry in the early universe. In short, the neutrinos may be crucial to shedding light not only on unification but also on our own origin!
The main purpose of this talk would be two-fold. First I discuss in the next section the issue of the choice of the effective symmetry in 4D. Here, I explain why (a) observed neutrino oscillations, (b) the likely need for leptogenesis as the means for baryogenesis , and (c) the success of certain fermion mass relations, together, seem to select out the route to higher unification based on the symmetry SU(4)-color . The effective symmetry near the GUT/string scale in 4D should thus be either SO(10) , or minimally $`G\left(224\right)=SU\left(2\right)_L\times SU\left(2\right)_R\times SU\left(4\right)^c`$ , as opposed to other alternatives. The second part of my talk is based on recent works on fermion masses and neutrino oscillations , and CP and flavor violations , all treated within a promising SO(10)/G(224) framework. The purpose of this second part is to present a unified description of a set of diverse phenomena, including:
* Fermion masses and mixings
* Neutrino oscillations
* CP non-conservation
* Flavor violations (in quark and lepton sectors),
* Baryogenesis via leptogenesis, and
* Proton Decay.
As it turns out, the neutrino plays a central role in arriving at this unified picture. My goal here will be to exhibit that the first five phenomena hang together neatly, in accord with observations, within a single predictive framework, based on an effective symmetry in 4D which is either SO(10) or G(224).
As we will see, the predictions of the framework not only account for many of the features of the five phenomena listed above (including the smallness of $`V_{cb}`$, the near maximality of $`\mathrm{\Theta }_{23}^\nu `$, $`m_b\left(m_b\right)`$, $`\mathrm{\Delta }m^2\left(\nu _2\nu _3\right)`$, $`ϵ_K`$, $`S\left(B_dJ/\psi K_S\right)`$, baryon asymmetry $`Y_B`$, and more), but also include features involving CP and flavor violations (such as edm, the asymmetry parameter $`S\left(B_d\varphi K_S\right)`$ and $`\mu e\gamma `$) which can clearly test the framework on many fronts.
To set the background for this discussion I first remark in the next section on the choice of the effective symmetry in 4D and the need for SU(4)-color. In this connection, I also clarify the historical origin of some of the concepts that are common to both G(224) and SO(10) and are now crucial to an understanding of neutrino masses and implementing baryogenesis via leptogenesis. In the following section, I briefly review the SO(10)/G(224)-framework proposed in Ref. for considerations of fermion masses and neutrino oscillations, and in the subsequent sections discuss the issues of CP and flavor violations as well as baryogenesis via leptogenesis , within the same framework. Expectations for proton decay are noted at the end.
## 2 On the choice of the Effective Symmetry in 4D: The need for SU(4)-color
The idea of grand unification was motivated by the desire to explain (a) the observed quantum numbers of the members of a family, and (b) quantization of electric charge on the one hand, and simultaneously to achieve (c) unification of quarks and leptons and (d) a unity of the basic forces on the other hand. While these four, together with the observed gauge coupling unification , still provide the strongest support – on aesthetic and empirical grounds– in favor of grand unification, they leave open the question of the choice of the effective symmetry G in 4D near the GUT scale which achieves these four goals.
For instance, should the symmetry group G be of rank 4, that is SU(5) , which is devoid of SU(4)-color? Or, should G possess SU(4)-color and thus minimally be SO(10) of rank 5, or even $`E_6`$ of rank 6? Or, should G be a string-derived semi-simple group G(224) $``$ SO(10), still of rank 5? Or, should G be \[SU(3)\]$`{}_{}{}^{3}E_6`$, of rank 6, but devoid of SU(4)-color?
An answer to these questions that helps select out the effective symmetry G in 4D is provided, however, if together with the four features (a)–(d) listed above, one folds in the following three:
$`\begin{array}{c}\text{(e) Neutrino oscillations}\hfill \\ \text{(f) The likely need for leptogenesis as the means for baryogenesis, and}\hfill \\ \text{(g) The success of certain fermion mass relations noted below}\hfill \end{array}`$
One can argue that the last three features, together with the first four listed above, clearly suggest that the standard model symmetry very likely emerges, near the GUT-scale $`M_U2\times 10^{16}`$ GeV, from the spontaneous breaking of a higher gauge symmetry G that should possess the symmetry SU(4)-color . The relevant symmetry in 4D could then maximally be SO(10) (possibly even $`E_6`$ ) or minimally the symmetry G(224); either one of these symmetries may be viewed to have emerged in 4D from a string/M theory near the string scale $`M_{st}\stackrel{>}{_{}}M_{GUT}`$ <sup>1</sup><sup>1</sup>1The relative advantage of an effective string-derived SO(10) over a G(224)-solution and vice versa have been discussed in detail in . Briefly speaking, for the case of a string derived G(224)-solution, coupling unification being valid near the string scale, one needs to assume that the string scale is not far above the GUT scale ($`M_{st}(23)M_{GUT}`$, say) to explain observed gauge coupling unification. While such a possibility can well arise in the string theory context , for an SO(10)-solution, coupling unification at the GUT-scale is ensured regardless of the gap between string and GUT-scales. The advantage of a G(224)-solution over an SO(10) solution is, however, that doublet-triplet splitting (DTS) can emerge naturally for the former in 4D through the process of string compactification (see Ref. ), while for an SO(10)-solution this feature is yet to be realized. As we will see, SO(10) and G(224) share many common advantages, aesthetic and practical, in particular as regards an understanding of fermion masses, neutrino oscillations and baryogenesis via leptogenesis; but they can be distinguished empirically through phenomena involving CP and flavor violations as well as proton decay.. The theory thus described should of course possess weak scale supersymmetry so as to avoid unnatural fine tuning in Higgs mass and to ensure gauge coupling unification.
To see the need for having SU(4)-color as a component of the higher gauge symmetry, it is useful to recall the family-multiplet structure of G(224), which is retained by SO(10) as well. The symmetry G(224), subject to left-right discrete symmetry which is natural to G(224), organizes members of a family into a single left-right self-conjugate multiplet ($`\mathrm{F}_\mathrm{L}^\mathrm{e}\mathrm{F}_\mathrm{R}^\mathrm{e}`$) given by :
$`\begin{array}{c}\mathrm{F}_{\mathrm{L},\mathrm{R}}^\mathrm{e}=\left[\begin{array}{cccc}\mathrm{u}_\mathrm{r}& \mathrm{u}_\mathrm{y}& \mathrm{u}_\mathrm{b}& \nu _𝐞\\ \mathrm{d}_\mathrm{r}& \mathrm{d}_\mathrm{y}& \mathrm{d}_\mathrm{b}& \mathrm{e}^{}\end{array}\right]_{\mathrm{L},\mathrm{R}}\end{array}`$ (4)
The multiplets $`\mathrm{F}_\mathrm{L}^\mathrm{e}`$ and $`\mathrm{F}_\mathrm{R}^\mathrm{e}`$ are left-right conjugates of each other transforming respectively as (2, 1, 4) and (1, 2, 4) of G(224); likewise for the muon and the tau families. Note that each family of G(224), subject to left-right symmetry, must contain sixteen two-component objects as opposed to fifteen for SU(5) or the standard model. While the symmetries $`SU\left(2\right)_{L,R}G\left(224\right)`$ treat each column of $`\mathrm{F}_{\mathrm{L},\mathrm{R}}^\mathrm{e}`$ as doublets, the symmetry SU(4)-color unifies quarks and leptons by treating each row of $`\mathrm{F}_\mathrm{L}^\mathrm{e}`$ and $`\mathrm{F}_\mathrm{R}^\mathrm{e}`$ as a quartet. Thus SU(4)-color treats the left and right-handed neutrinos ($`\nu _L^e`$ and $`\nu _R^e`$) as the fourth color-partners of the left and right-handed up quarks (u<sub>L</sub> and u<sub>R</sub>) respectively. Here in lies the distinctive feature of SU(4)-color. It necessitates the existence of the RH neutrino ($`\nu _R^e`$) on par with that of the RH up quark (u<sub>R</sub>) by relating them through a gauge symmetry transformation; and likewise for the mu and the tau families. As we will see, this in turn leads to some very desirable fermion mass relations for the third family that help distinguish it from alternative symmetries. An accompanying characteristic of SU(4)-color is that it also introduces $`BL`$ as a local symmetry . This in turn plays a crucial role in protecting the Majorana masses of the right-handed neutrinos from acquiring Planck-scale values.
In anticipation of sections. 3, 4 and 7 where some of the statements made below will become clear, I may now state the following. The need for SU(4)-color (mentioned above) arises because it provides the following desirable features:
$`\begin{array}{cc}\left(1\right)\mathrm{RH}\mathrm{neutrino}\left(\nu _\mathrm{R}^\mathrm{i}\right)\mathrm{as}\mathrm{an}\mathrm{essential}\hfill & \mathrm{Needed}\mathrm{to}\mathrm{implement}\mathrm{the}\mathrm{seesaw}\mathrm{mechanism}\hfill \\ \mathrm{member}\mathrm{of}\mathrm{each}\mathrm{family}\hfill & \mathrm{and}\mathrm{leptogenesis}\left(\mathrm{see}\mathrm{Secs}\mathrm{.\; 3}\mathrm{and}7\right).\hfill \\ \left(2\right)\mathrm{B}\mathrm{L}\mathrm{as}\mathrm{a}\mathrm{local}\mathrm{symmetry}\hfill & \mathrm{Needed}\mathrm{to}\mathrm{protect}\nu _\mathrm{R}^{}\mathrm{s}\mathrm{from}\mathrm{acquiring}\mathrm{Planck}\hfill \\ & \mathrm{scale}\mathrm{masses}\mathrm{and}\mathrm{to}\mathrm{set}\mathrm{M}\left(\nu _\mathrm{R}^\mathrm{i}\right)\mathrm{M}_{\mathrm{B}\mathrm{L}}\mathrm{M}_{\mathrm{GUT}}.\hfill \end{array}`$
(3) Two simple mass relations for
the 3rd family:
$`\begin{array}{cc}\left(\mathrm{a}\right)\mathrm{m}\left(\nu _{\mathrm{Dirac}}^\tau \right)\mathrm{m}_{\mathrm{top}}\left(\mathrm{M}_{\mathrm{GUT}}\right)\hfill & \mathrm{Needed}\mathrm{for}\mathrm{success}\mathrm{of}\mathrm{seesaw}\left(\mathrm{see}\mathrm{section}3\right).\hfill \\ \left(\mathrm{b}\right)\mathrm{m}_\mathrm{b}\left(\mathrm{M}_{\mathrm{GUT}}\right)\mathrm{m}_\tau \hfill & \mathrm{Empirically}\mathrm{successful}.\hfill \end{array}`$
These three ingredients ((1), (2) and (3a)), together with the SUSY unification-scale, are indeed crucial (see sections 3 and 4) to an understanding of the neutrino masses via the seesaw mechanism . The first two ingredients are important also for implementing baryogenesis via leptogenesis (see section 7). Hence the need for having SU(4)-color as a component of the unification symmetry which provides all four ingredients.
By contrast SU(5), devoid of SU(4)-color, does not provide the ingredients of (1), (2) and (3a) (though it does provide (3b)); hence it does not have a natural setting for understanding neutrino masses and implementing baryogenesis via leptogenesis (see discussion in section 4 and especially footnote 2). Symmetries like $`G\left(2213\right)=SU\left(2\right)_L\times SU\left(2\right)_R\times U\left(1\right)_{BL}\times SU\left(3\right)^c`$ and $`\left[SU\left(3\right)\right]^3`$ provide (1) and (2) but neither (3a) nor (3b), while flipped $`SU\left(5\right)^{}\times U\left(1\right)^{}`$ provides (1), (2) and (3a) but not (3b). In summary, the need for the combination of the four ingredients (1), (2), (3a) and (3b) seems to select out the route to higher unification based on SU(4)-color, and thereby as mentioned above an effective symmetry like G(224) or SO(10) being operative in 4D near the string scale.
At this point, an intimate link between $`SU\left(4\right)`$-color and the left-right symmetric gauge structure $`SU\left(2\right)_L\times SU\left(2\right)_R`$ is worth noting. Assuming that $`SU\left(4\right)^c`$ is gauged and demanding an explanation of quantization of electric charge lead one to gauge minimally the left-right symmetric flavor symmetry $`SU\left(2\right)_L\times SU\left(2\right)_R`$ (rather than $`SU\left(2\right)_L\times U\left(1\right)_{I_{3R}}`$). The resulting minimal gauge symmetry that contains $`SU\left(4\right)`$-color and explains quantization of electric charge is then $`G\left(224\right)=SU\left(2\right)_L\times SU\left(2\right)_R\times SU\left(4\right)^c`$ . With SU(4)-color being vectorial, such a symmetry structure (as also G(2213) which is a subgroup of G(224)) in turn naturally suggests the attractive idea that L–R discrete symmetry and thus parity (i.e. F$`{}_{\mathrm{L}}{}^{}`$ F<sub>R</sub>, W$`{}_{\mathrm{L}}{}^{}`$ W<sub>R</sub> with g$`{}_{\mathrm{L}}{}^{\left(0\right)}=`$g$`{}_{}{}^{\left(0\right)}{}_{\mathrm{R}}{}^{}`$) is preserved at a basic level and is broken only spontaneously . In other words, observed parity violation is only a low-energy phenomenon which should disappear at sufficiently high energies. We thus see that the concepts of SU(4)-color and left-right symmetry are intimately inter-twined, through the requirement of quantization of electric charge.
A Historical Note: Advantages of G(224)
As a historical note, it is worth noting that the symmetry $`SU\left(4\right)`$-color, and thereby the three desirable features listed above, were introduced into the literature, as a step towards higher unification, through the minimal symmetry G(224) , rather than through SO(10) . The symmetry G(224) (supplemented by L–R discrete symmetry which is natural to G(224)) in turn brought a host of desirable features. Including those mentioned above they are:
(i) Unification of all sixteen members of a family within one left-right self-conjugate multiplet, with a neat explanation of their quantum numbers;
(ii) Quantization of electric charge;
(iii) Quark-lepton unification through $`SU\left(4\right)`$-color;
(iv) Conservation of parity at a fundamental level;
(v) RH neutrino as a compelling member of each family;
(vi) $`BL`$ as a local symmetry; and
(vii) The rationale for the now successful mass-relations (3a) and (3b).
These seven features constitute the hallmark of G(224). Any simple or semi-simple group that contains G(224) as a subgroup would of course naturally possess these features. So does therefore SO(10) which is the smallest simple group containing G(224). Thus, as alluded to above, all the attractive features of SO(10), which distinguish it from SU(5) and are now needed to understand neutrino masses and baryogenesis via leptogenesis, were in fact introduced through the symmetry G(224) , long before the SO(10) papers appeared . These in particular include the features (i) as well as (iii)–(vii). SO(10) of course preserved these features for reasons stated above; it even preserved the family multiplet structure of G(224) without needing additional fermions (unlike $`E_6`$) in that the L–R conjugate 16-plet ( = $`F_LF_R`$) of G(224) precisely corresponds to the spinorial 16 (= $`F_L\left(F_R\right)^c`$) of SO(10). Furthermore, with SU(4)-color being vectorial, G(224) is anomaly-free; so also is SO(10).
SO(10) brought of course one added and desirable feature relative to G(224)– that is manifest coupling unification. Again, as a historical note, it is worth mentioning that the idea of coupling unification was initiated in and was first manifested explicitly within a minimal model through the suggestion of SU(5) in .
As mentioned before, believing in string unification, either G(224) or SO(10) may be viewed to have its origin in a still higher gauge symmetry (like $`E_8`$) in 10D. To realize the existence of the right-handed neutrinos, $`BL`$ as a local symmetry and the fermion mass-relations (3a), which are needed for understanding neutrino masses and implementing baryogenesis via leptogenesis, I have argued that one needs $`SU\left(4\right)^c`$ as a component of the effective symmetry in 4D, and therefore minimally G(224) (or even G(214)) or maximally perhaps SO(10) in 4D near the string scale. The relative advantages of G(224) over SO(10) and vice versa as 4D symmetries in addressing the issues of doublet-triplet splitting on the one hand and gauge coupling unification on the other hand have been discussed in Ref. and briefly noted in footnote 1.
In the following sections I discuss how either one of these symmetries G(224) or SO(10) link together fermion masses, neutrino oscillations, CP and flavor violations and leptogenesis. As we will see, while G(224) and SO(10) lead to essentially identical results for fermion masses and neutrino oscillations, which are discussed in the next two sections, they can be distinguished by processes involving CP and/or flavor violations, which are discussed in sections 5 and 6, and proton decay, discussed in section 8.
## 3 Seesaw and SUSY Unification with SU(4)-color
The idea of the seesaw mechanism is simply this. In a theory with RH neutrinos as an essential member of each family, and with spontaneous breaking of $`BL`$ and $`I_{3R}`$ at a high scale ($`M_{BL}`$), both already inherent in , the RH neutrinos can and generically will acquire a superheavy Majorana mass ($`M\left(\nu _R\right)M_{BL}`$) that violates lepton number and $`BL`$ by two units. Combining this with the Dirac mass of the neutrino ($`m\left(\nu _{\mathrm{Dirac}}\right)`$), which arises through electroweak symmetry breaking, one would then obtain a mass for the LH neutrino given by
$$m\left(\nu _L\right)m\left(\nu _{\mathrm{Dirac}}\right)^2/M\left(\nu _R\right)$$
(5)
which would be naturally super-light because $`M\left(\nu _R\right)`$ is naturally superheavy. This then provided a *simple but compelling reason* for the lightness of the known neutrinos. In turn it took away the major burden that faced the ideas of $`SU\left(4\right)`$-color and left-right symmetry from the beginning. In this sense, the seesaw mechanism was indeed the missing piece that was needed to be found for consistency of the ideas of $`SU\left(4\right)`$-color and left-right symmetry.
In turn, of course, the seesaw mechanism needs the ideas of $`SU\left(4\right)`$-color and SUSY grand unification so that it may be quantitatively useful. Because the former provides (a) the RH neutrino as a compelling feature (crucial to seesaw), and (b) the Dirac mass for the tau neutrino accurately in terms of the top quark mass (cf. feature (3a)), while the latter provides the superheavy Majorana mass of the $`\nu _R^\tau `$ in terms of the SUSY unification scale (see Sec. 4). Both these masses enter crucially into the seesaw formula and end up giving the *right mass-scale* for the atmospheric neutrino oscillation as observed. To be specific, $`SU\left(4\right)`$-color yields: $`m\left(\nu _{\mathrm{Dirac}}^\tau \right)m_{\mathrm{top}}\left(M_U\right)120\mathrm{GeV}`$; and the SUSY unification scale, together with the protection provided by $`BL`$ that forbids Planck-scale contributions to the Majorana mass of $`\nu _R^\tau `$, naturally yields: $`M\left(\nu _R^\tau \right)M_{GUT}^2/M4\times 10^{14}\mathrm{GeV}\left(1/2\text{}2\right)`$, where $`M10^{18}`$ GeV $`\left(1/2\text{}2\right)`$ \[cf. Sec. 4\]. The seesaw formula (without 2-3 family mixing) then yields:
$`m\left(\nu _L^3\right)\left(120\mathrm{GeV}\right)^2/(4\times 10^{14}\mathrm{GeV}(1/2\text{}2))(1/28\mathrm{eV})(1/2\text{}2))`$ (6)
With hierarchical pattern for fermion mass-matrices (see Sec. 4), one necessarily obtains $`m\left(\nu _L^2\right)m\left(\nu _L^3\right)`$ (see section 4), and thus $`\sqrt{\mathrm{\Delta }m_{23}^2}m\left(\nu _L^3\right)1/28\mathrm{eV}\left(1/2\text{}2\right)`$. This is just the right magnitude to go with the mass scale observed at SuperK !
*Without an underlying reason as above for at least the approximate values of these two vastly differing mass-scales — $`m(\nu _{\mathrm{Dirac}}^\tau )`$ and $`M(\nu _R^\tau )`$ — the seesaw mechanism by itself would have no clue, quantitatively, to the mass of the LH neutrino.* In fact it would yield a rather arbitrary value for $`m\left(\nu _L^\tau \right)`$, which could vary quite easily by more than 10 orders of magnitude either way around the observed mass scale. This would in fact be true if one introduces the RH neutrinos as a singlet of the SM or of SU(5).<sup>2</sup><sup>2</sup>2To see this, consider for simplicity just the third family. Without $`SU(4)`$-color, even if a RH two-component fermion $`N`$ (the analogue of $`\nu _R`$) is introduced by hand as a *singlet* of the gauge symmetry of the SM or $`SU(5)`$, *such an $`N`$ by no means should be regarded as a member of the third family, because it is not linked by a gauge transformation to the other fermions in the third family*. Thus its Dirac mass term given by $`m(\nu _{\mathrm{Dirac}}^\tau )[\overline{\nu }_L^\tau N+h.c.]`$ is completely arbitrary, except for being bounded from above by the electroweak scale $`200\text{ GeV}`$. In fact a priori (within the SM or $`SU(5)`$) it can well vary from say 1 GeV (or even 1 MeV) to 100 GeV. Using Eq. (5), this would give a variation in $`m_(\nu _L)`$ by at least four orders of magnitude if the Majorana mass $`M(N)`$ of $`N`$ is held fixed. Furthermore, $`N`$ being a singlet of the SM as well as of $`SU(5)`$, the Majorana mass $`M(N)`$, unprotected by $`BL`$, could well be as high as the Planck or the string scale ($`10^{18}\text{}10^{17}\text{ GeV}`$), and as low as say 1 TeV; this would introduce a further arbitrariness by fourteen orders of magnitude in $`m(\nu _L)`$. Such arbitrariness both in the Dirac and in the Majorana masses, is drastically reduced, however, once $`\nu _R`$ is related to the other fermions in the family by an $`SU(4)`$-color gauge transformation and a SUSY unification is assumed.
In short, the seesaw mechanism needs the ideas of SUSY unification and $`SU\left(4\right)`$-color, and of course vice-versa; *together* they provide an understanding of neutrino masses as observed. Schematically, one thus finds:
$$\begin{array}{ccc}\hfill \overline{)\begin{array}{c}\text{SUSY UNIFICATION}\\ \text{WITH }SU\left(4\right)\text{-COLOR}\end{array}}& & \text{SEESAW}\hfill \\ & & \\ \hfill m\left(\nu _L^3\right)& & 1/10\text{ eV}.\hfill \end{array}$$
(7)
In summary, as noted in section 2, the agreement of the expected $`\sqrt{\mathrm{\Delta }m_{23}^2}`$ with the observed SuperK value clearly seems to favor the idea of the seesaw and select out the route to higher unification based on supersymmetry and SU(4)-color, as opposed to other alternatives.
I will return to a more quantitative discussion of the mass scale and the angle associated with the atmospheric neutrino oscillations in Sec. 4.
## 4 Fermion Masses and Neutrino Oscillations in G(224)/SO(10): A Review of the BPW framework
Following Ref. , I now present a simple and predictive pattern for fermion mass-matrices based on $`SO\left(10\right)`$ or the $`G\left(224\right)`$-symmetry.<sup>3</sup><sup>3</sup>3I will present the Higgs system for $`SO(10)`$. The discussion would remain essentially unaltered if one uses the corresponding $`G(224)`$-submultiplets instead. One can obtain such a mass mass-matrix for the fermions by utilizing only the minimal Higgs system that is used also to break the gauge symmetry $`SO\left(10\right)`$ to $`SU\left(3\right)^c\times U\left(1\right)_{em}`$. It consists of the set:
$$H_{\mathrm{minimal}}=\{\mathrm{𝟒𝟓}_𝐇,\mathrm{𝟏𝟔}_𝐇,\overline{\mathrm{𝟏𝟔}}_𝐇,\mathrm{𝟏𝟎}_𝐇\}$$
(8)
Of these, the VEV of $`\mathrm{𝟒𝟓}_𝐇M_U`$ breaks $`SO\left(10\right)`$ in the B-L direction to $`G\left(2213\right)=SU\left(2\right)_L\times SU\left(2\right)_R\times U\left(1\right)_{BL}\times SU\left(3\right)^c`$, and those of $`\mathrm{𝟏𝟔}_𝐇=\overline{\mathrm{𝟏𝟔}}_𝐇M_U`$ along $`\stackrel{~}{\overline{\nu }}_{RH}`$ and $`\stackrel{~}{\nu }_{RH}`$ break $`G\left(2213\right)`$ into the SM symmetry $`G\left(213\right)`$ at the unification-scale $`M_U`$. Now $`G\left(213\right)`$ breaks at the electroweak scale by the VEVs of $`\mathrm{𝟏𝟎}_𝐇`$ and of the EW doublet in $`\mathrm{𝟏𝟔}_𝐇`$ to $`SU\left(3\right)^c\times U\left(1\right)_{em}`$.
Before discussing fermion masses and mixings, I should comment briefly on the use of the minimal Higgs system noted above as opposed to large-dimensional tensorial multiplets of SO(10) including ($`126_H,\overline{126_H}`$), $`210_H`$ and possibly $`120_H`$, which have been used widely in the literature to break SO(10) to the SM symmetry and give masses to the fermions. We have preferred to use the low-dimensional Higgs multiplets ($`45_H,16_H`$ and $`\overline{16_H}`$) rather than the large dimensional ones like ($`126_H,\overline{126_H}`$, $`210_H`$ and possibly $`120_H`$) in part because these latter tend to give too large GUT-scale threshold corrections to $`\alpha _3\left(m_Z\right)`$ from split sub-multiplets (typically exceeding 15–20% with either sign), which would render observed gauge coupling unification fortuitous (see Appendix D of Ref. for a discussion on this point). By contrast, with the low-dimensional multiplets ($`45_H,16_H`$ and $`\overline{16_H}`$), the threshold corrections to $`\alpha _3\left(m_Z\right)`$ are tamed and are found, for a large range of the relevant parameters, to have the right sign and magnitude (nearly -5 to -8%) so as to account naturally for the observed gauge coupling unification.
Another disadvantage of $`126_H`$, which contributes to EW symmetry breaking through its ($`2,2,15`$) component of G(224), is that it gives $`BL`$ dependent contribution to family-diagonal fermion masses. Such a contribution, barring adjustment of parameters against the contribution of $`10_H`$, could in general make the success of the relation $`m_b\left(GUT\right)m_\tau `$ fortuitous. By contrast, the latter relation emerges as a robust prediction of the minimal Higgs system ($`45_H,16_H,\overline{16}_H\text{ and }10_H`$), subject to a hierarchical pattern, because the only $`\left(BL\right)`$ dependent contribution in this case can come effectively through $`10_H`$$`45_H`$/$`M`$ which is family-antisymmetric and cannot contribute to diagonal entries (see below) .
One other issue involves the question of achieving doublet-triplet splitting by a natural mechanism as opposed to that of fine tuning, and incorporating the associated GUT-scale threshold correction to $`\alpha _3\left(m_Z\right)`$. For the case of ($`45_H,16_H`$, $`\overline{16_H}`$ and $`10_H`$), there exists a simple mechanism which achieves the desired splitting naturally with the introduction of an extra $`10_H^{}`$ , and the effect of this splitting on GUT-scale threshold correction to $`\alpha _3\left(m_Z\right)`$ has been evaluated in to conform with natural coupling unification on the one hand and the limit on proton lifetime on the other hand. To the best of my knowledge, an analogous study for the system involving ($`126_H,\overline{126_H}`$) has not been carried out as yet.
Balancing against these advantages of the minimal Higgs system, the large-dimensional system ($`126_H,\overline{126}_H,210_H\text{ and possibly }120_H`$) has an advantage over the minimal system, because $`126`$ and $`\overline{126}`$ break $`BL`$ by two units and thus automatically preserve the familiar R-parity = $`\left(1\right)^{3\left(BL\right)+2S}`$. By contrast, $`16`$ and $`\overline{16}`$ break $`BL`$ by one unit and thereby break the familiar R-parity. This difference is, however, not really significant, because for the minimal system one can still define consistently a matter-parity (i.e. $`16_i16_i,16_H16_H,\overline{16_H}\overline{16_H},45_H45_H,10_H10_H`$), which serves the desired purpose by allowing all the desired interactions but forbidding the dangerous $`d=4`$ proton decay operators and yielding stable LSP to serve as CDM. Given the net advantages of the minimal Higgs system $`H_{\mathrm{minimal}}`$, as noted above, I will proceed to present the results of which uses this system.<sup>4</sup><sup>4</sup>4Personally I feel, however, that it would be important to explore thoroughly the theoretical and phenomenological consequences of both the minimal and the large-dimensional Higgs systems involving issues such as doublet-triplet splitting, GUT-scale threshold corrections to gauge couplings, CP and flavor violations and proton decay. The aim would be to look for avenues by which the two systems can be distinguished experimentally.
The $`3\times 3`$ Dirac mass matrices for the four sectors $`(u,d,l,\nu )`$ proposed in Ref. were motivated in part by the notion that flavor symmetries are responsible for the hierarchy among the elements of these matrices (i.e., for “33”$``$“23”$``$“32”$``$“22”$``$“12”$``$“11”, etc.), and in part by the group theory of SO(10)/G(224), relevant to a minimal Higgs system. Up to minor variants , they are as follows <sup>5</sup><sup>5</sup>5A somewhat analogous scheme based on low dimensional $`SO(10)`$ Higgs multiplets, has been proposed by C. Albright and S. Barr \[AB\] , who, however use two pairs of ($`16_H,\overline{16_H}`$), while BPW use only one. One major difference between the work of AB and that of BPW (stemming from the use of two pairs of ($`16_H,\overline{16_H}`$) by AB compared to one by BPW) is that the AB model introduces the so-called “lop-sided” pattern in which some of the “23” and “32” elements are even greater than the “33” element; in the BPW model on the other hand, the pattern is consistently hierarchical with individual “23” and “32” elements (like $`\eta `$, $`ϵ`$ and $`\sigma `$) being much smaller in magnitude than the “33” element of 1. It turns out that this difference leads to a characteristically different explanation for the large (maximal) $`\nu _\mu \nu _\tau `$ oscillation angle in the two models, and in particular to a much more enhanced rate for $`\mu e\gamma `$ in the AB model compared to that in the BPW model (see Sec. 7).<sup>,</sup><sup>6</sup><sup>6</sup>6An alternate SO(10)-based pattern differing from BPW and AB models is proposed in .:
$`\begin{array}{cc}M_u=\left[\begin{array}{ccc}0& ϵ^{}& 0\\ ϵ^{}& \zeta _{22}^u& \sigma +ϵ\\ 0& \sigma ϵ& 1\end{array}\right]_u^0;& M_d=\left[\begin{array}{ccc}0& \eta ^{}+ϵ^{}& 0\\ \eta ^{}ϵ^{}& \zeta _{22}^d& \eta +ϵ\\ 0& \eta ϵ& 1\end{array}\right]_d^0\\ & \\ M_\nu ^D=\left[\begin{array}{ccc}0& 3ϵ^{}& 0\\ 3ϵ^{}& \zeta _{22}^u& \sigma 3ϵ\\ 0& \sigma +3ϵ& 1\end{array}\right]_u^0;& M_l=\left[\begin{array}{ccc}0& \eta ^{}3ϵ^{}& 0\\ \eta ^{}+3ϵ^{}& \zeta _{22}^d& \eta 3ϵ\\ 0& \eta +3ϵ& 1\end{array}\right]_d^0\end{array}`$ (24)
These matrices are defined in the gauge basis and are multiplied by $`\overline{\mathrm{\Psi }}_L`$ on left and $`\mathrm{\Psi }_R`$ on right. For instance, the row and column indices of $`M_u`$ are given by $`(\overline{u}_L,\overline{c}_L,\overline{t}_L)`$ and $`(u_R,c_R,t_R)`$ respectively. Note the group-theoretic up-down and quark-lepton correlations: the same $`\sigma `$ occurs in $`M_u`$ and $`M_\nu ^D`$, and the same $`\eta `$ occurs in $`M_d`$ and $`M_l`$. It will become clear that the $`ϵ`$ and $`ϵ^{}`$ entries are proportional to $`BL`$ and are antisymmetric in the family space (as shown above). Thus, the same $`ϵ`$ and $`ϵ^{}`$ occur in both ($`M_u`$ and $`M_d`$) and also in ($`M_\nu ^D`$ and $`M_l`$), but $`ϵ3ϵ`$ and $`ϵ^{}3ϵ^{}`$ as $`ql`$. Such correlations result in enormous reduction of parameters and thus in increased predictiveness. Such a pattern for the mass-matrices can be obtained, using a minimal Higgs system $`\mathrm{𝟒𝟓}_H,\mathrm{𝟏𝟔}_H,\overline{\mathrm{𝟏𝟔}}_H\text{ and }\mathrm{𝟏𝟎}_H`$ and a singlet $`S`$ of SO(10), through effective couplings as follows (see Ref. and for details):
$`_{\mathrm{Yuk}}`$ $`=`$ $`h_{33}\mathrm{𝟏𝟔}_3\mathrm{𝟏𝟔}_3\mathrm{𝟏𝟎}_H+[h_{23}\mathrm{𝟏𝟔}_2\mathrm{𝟏𝟔}_3\mathrm{𝟏𝟎}_H(S/M)`$ (25)
$`+`$ $`a_{23}\mathrm{𝟏𝟔}_2\mathrm{𝟏𝟔}_3\mathrm{𝟏𝟎}_H(\mathrm{𝟒𝟓}_H/M^{})(S/M)^p+g_{23}\mathrm{𝟏𝟔}_2\mathrm{𝟏𝟔}_3\mathrm{𝟏𝟔}_H^d(\mathrm{𝟏𝟔}_H/M^{\prime \prime })(S/M)^q]`$
$`+`$ $`\left[h_{22}\mathrm{𝟏𝟔}_2\mathrm{𝟏𝟔}_2\mathrm{𝟏𝟎}_H\left(S/M\right)^2+g_{22}\mathrm{𝟏𝟔}_2\mathrm{𝟏𝟔}_2\mathrm{𝟏𝟔}_H^d\left(\mathrm{𝟏𝟔}_H/M^{\prime \prime }\right)\left(S/M\right)^{q+1}\right]`$
$`+`$ $`\left[g_{12}\mathrm{𝟏𝟔}_1\mathrm{𝟏𝟔}_2\mathrm{𝟏𝟔}_H^d\left(\mathrm{𝟏𝟔}_H/M^{\prime \prime }\right)\left(S/M\right)^{q+2}+a_{12}\mathrm{𝟏𝟔}_1\mathrm{𝟏𝟔}_2\mathrm{𝟏𝟎}_H\left(\mathrm{𝟒𝟓}_H/M^{}\right)\left(S/M\right)^{p+2}\right].`$
Typically we expect $`M^{}`$, $`M^{\prime \prime }`$ and $`M`$ to be of order $`M_{\mathrm{string}}`$ or (possibly) of order $`M_{GUT}`$. The VEV’s of $`\mathrm{𝟒𝟓}_H`$ (along $`BL`$), $`\mathrm{𝟏𝟔}_H=\overline{\mathrm{𝟏𝟔}}_H`$ (along standard model singlet sneutrino-like component) and of the SO(10)-singlet $`S`$ are of the GUT-scale, while those of $`\mathrm{𝟏𝟎}_H`$ and of the down type SU(2)<sub>L</sub>-doublet component in $`\mathrm{𝟏𝟔}_H`$ (denoted by $`\mathrm{𝟏𝟔}_H^d`$) are of the electroweak scale . Depending upon whether $`M^{}\left(M^{\prime \prime }\right)M_{\mathrm{GUT}}`$ or $`M_{\mathrm{string}}`$ (see ), the exponent $`p\left(q\right)`$ is either one or zero .
The entries 1 and $`\sigma `$ arise respectively from $`h_{33}`$ and $`h_{23}`$ couplings, while $`\widehat{\eta }\eta \sigma `$ and $`\eta ^{}`$ arise respectively from $`g_{23}`$ and $`g_{12}`$-couplings. The $`\left(BL\right)`$-dependent antisymmetric entries $`ϵ`$ and $`ϵ^{}`$ arise respectively from the $`a_{23}`$ and $`a_{12}`$ couplings. \[Effectively, with $`\mathrm{𝟒𝟓}_H`$ $`BL`$, the product $`\mathrm{𝟏𝟎}_H\times \mathrm{𝟒𝟓}_H`$ contributes as a 120, whose coupling is family-antisymmetric.\] The relatively small entry $`\zeta _{22}^u`$ arises from the $`h_{22}`$-coupling, while $`\zeta _{22}^d`$ arises from the joint contributions of $`h_{22}`$ and $`g_{22}`$-couplings.
Such a hierarchical form of the mass-matrices, with $`h_{33}`$-term being dominant, is attributed in part to a U(1)-flavor gauge symmetry that distinguishes between the three families and introduces powers of $`S/M1/10`$, and in part to higher dimensional operators involving for example $`\mathrm{𝟒𝟓}_H/M^{}`$ or $`\mathrm{𝟏𝟔}_H/M^{\prime \prime }`$, which are suppressed by $`M_{\mathrm{GUT}}/M_{\mathrm{string}}1/10`$, if $`M^{}`$ and/or $`M^{\prime \prime }M_{\mathrm{string}}`$.
The right-handed neutrino masses arise from the effective couplings of the form :
$$_{\mathrm{Maj}}=f_{ij}\mathrm{𝟏𝟔}_i\mathrm{𝟏𝟔}_j\overline{\mathrm{𝟏𝟔}}_H\overline{\mathrm{𝟏𝟔}}_H/M$$
(26)
where the $`f_{ij}`$’s include appropriate powers of $`S/M`$. The hierarchical form of the Majorana mass-matrix for the RH neutrinos is :
$`M_R^\nu =\left[\begin{array}{ccc}x& 0& z\\ 0& 0& y\\ z& y& 1\end{array}\right]M_R`$ (30)
Following flavor charge assignments (see ), we have $`1yzx`$. The magnitude of M<sub>R</sub> is estimated by putting $`f_{33}1`$ and $`\overline{\mathrm{𝟏𝟔}}_HM_{GUT}2\times 10^{16}`$ GeV. We expect that the effective scale M of Eq. (26) should lie between $`M_{string}4\times 10^{17}`$ GeV and $`\left(M_{Pl}\right)_{reduced}2\times 10^{18}`$ GeV. Thus we take $`M10^{18}`$ GeV (1/2–2) . We then get the Majorana mass of the heaviest RH neutrino to be given by $`M_3M_R=f_{33}\overline{\mathrm{𝟏𝟔}}_H^2/M\left(4\times 10^{14}\text{ GeV}\right)\left(1/2\text{}2\right)`$.
Ignoring possible phases in the parameters and thus the source of CP violation for a moment, and also setting $`\zeta _{22}^d=\zeta _{22}^u=0`$, as was done in Ref. , the parameters $`(\sigma ,\eta ,ϵ,ϵ^{},\eta ^{},_u^0,\mathrm{and}_d^0)`$ can be determined by using, for example, $`m_t^{\mathrm{phys}}=174`$ GeV, $`m_c\left(m_c\right)=1.37`$ GeV, $`m_s\left(1\text{ GeV}\right)=110116`$ MeV, $`m_u\left(1\text{ GeV}\right)=6`$ MeV, and the observed masses of $`e`$, $`\mu `$, and $`\tau `$ as inputs. One is thus led, for this CP conserving case, to the following fit for the parameters, and the associated predictions :
$`\begin{array}{c}\sigma 0.110,\eta 0.151,ϵ0.095,\left|\eta ^{}\right|4.4\times 10^3,\hfill \\ \begin{array}{c}ϵ^{}2\times 10^4,_u^0m_t\left(M_X\right)100\text{ GeV},_d^0m_\tau \left(M_X\right)1.1\text{ GeV}.\hfill \end{array}\hfill \end{array}`$ (34)
These output parameters remain stable to within 10% corresponding to small variations ($`\stackrel{<}{_{}}10`$%) in the input parameters of $`m_t`$, $`m_c`$, $`m_s`$, and $`m_u`$. These in turn lead to the following predictions for the quarks and light neutrinos , :
$`\begin{array}{c}m_b\left(m_b\right)\left(4.7\text{}4.9\right)\text{ GeV},\hfill \\ \sqrt{\mathrm{\Delta }m_{23}^2}m\left(\nu _3\right)\text{(1/24 eV)(1/2–2)},\hfill \\ \begin{array}{ccc}V_{cb}\hfill & & \left|\sqrt{\frac{m_s}{m_b}\left|\frac{\eta +ϵ}{\eta ϵ}\right|}\sqrt{\frac{m_c}{m_t}\left|\frac{\sigma +ϵ}{\sigma ϵ}\right|}\right|\hfill \\ & & 0.044,\hfill \end{array}\hfill \\ \{\begin{array}{ccc}\theta _{\nu _\mu \nu _\tau }^{\mathrm{osc}}\hfill & & \left|\sqrt{\frac{m_\mu }{m_\tau }}\left|\frac{\eta 3ϵ}{\eta +3ϵ}\right|^{1/2}+\sqrt{\frac{m_{\nu _2}}{m_{\nu _3}}}\right|\hfill \\ & & \left|0.437+\left(0.378\pm 0.03\right)\right|\text{ (for }\frac{m\left(\nu _2\right)}{m\left(\nu _3\right)}1/6\text{),}\hfill \\ \multicolumn{3}{c}{\text{Thus, }\mathrm{sin}^22\theta _{\nu _\mu \nu _\tau }^{\mathrm{osc}}0.993,}\end{array}\hfill \\ V_{us}\left|\sqrt{\frac{m_d}{m_s}}\sqrt{\frac{m_u}{m_c}}\right|0.20,\hfill \\ \left|\frac{V_{ub}}{V_{cb}}\right|\sqrt{\frac{m_u}{m_c}}0.07,\hfill \\ m_d\left(\text{1 GeV}\right)\text{8 MeV}.\hfill \end{array}`$ (47)
It has been noted that small non-seesaw contribution to $`\nu _L^e\nu _L^\mu `$ mass term ($``$ few $`\times 10^3`$ eV) which can arise through higher dimensional operators in accord with flavor symmetry, but which have been ignored in the analysis given above, can lead quite plausibly to large $`\nu _e\nu _\mu `$ oscillation angle in accord with the LMA MSW solution for the solar neutrino problem. Including the seesaw contribution obtained by combining $`M_\nu ^D`$ (Eq. (24)) and $`M_R^\nu `$ (Eq. (30)) and with an input value of $`y1/17`$ (Note that by flavor symmetry , we a priori expect $`\left|y\right|1/10`$) we get:
$`\begin{array}{c}m\left(\nu _2\right)\left(67\right)\times 10^3\mathrm{eV}\left(\mathrm{from}\mathrm{seesaw}\right)\left(a\right)\hfill \\ \begin{array}{c}m\left(\nu _1\right)\left(1few\right)\times 10^3;\mathrm{thus}\mathrm{\Delta }\mathrm{m}_{12}^2\left(35\right)\times 10^5\mathrm{eV}^2\left(b\right)\hfill \\ \begin{array}{c}\mathrm{sin}^22\theta _{\nu _e\nu _\mu }^{osc}\left(0.50.7\right)\left(\mathrm{from}\mathrm{non}\mathrm{seesaw}\right)\left(c\right)\hfill \\ \begin{array}{c}\theta _{13}\stackrel{<}{_{}}\left(25\right)\times 10^2\left(d\right)\hfill \end{array}\hfill \end{array}\hfill \end{array}\hfill \end{array}`$ (55)
While the results in Eq. (47) are compelling predictions of the model, the LMA-compatible solution for $`\theta _{\nu _e\nu _\mu }^{osc}`$ listed in (55)(c) should be regarded as a plausible and consistent possibility rather than as a compelling prediction of the framework.
The Majorana masses of the RH neutrinos ($`N_{iR}N_i`$) are given by :
$`M_3`$ $``$ $`M_R4\times 10^{14}\text{ GeV (1/2-2)},`$
$`M_2`$ $``$ $`\left|y^2\right|M_310^{12}\text{ GeV(1/2-2)},`$ (56)
$`M_1`$ $``$ $`\left|xz^2\right|M_3\left(1/4\text{-}2\right)10^4M_3`$
$`4\times 10^{10}\text{ GeV}\left(1/84\right).`$
where $`y1/17`$ and $`xz^210^4\left(1/22\right)`$ have been used, in accord with flavor-symmetry . Note that we necessarily have a hierarchical pattern for the light as well as the heavy neutrinos with normal hierarchy $`m_1\stackrel{<}{_{}}m_2m_3`$ and $`M_1M_2M_3`$.
Leaving aside therefore the question of the $`\nu _e\nu _\mu `$ oscillation angle, it seems quite remarkable that all seven predictions in Eq.(47) agree with observations to within 10%. Particularly intriguing is the $`(BL`$)-dependent group-theoretic correlation between $`V_{cb}`$ and $`\theta _{\nu _\mu \nu _\tau }^{osc}`$, which explains simultaneously why one is small ($`V_{cb}`$) and the other is so large ($`\theta _{\nu _\mu \nu _\tau }^{osc}`$) .
Why $`V_{cb}`$ is small while $`\theta _{\nu _\mu \nu _\tau }^{osc}`$ is large?
A Comment is in order about this last feature. Often it has been remarked by several authors that while the “observed” near equality of $`m_b`$ and $`m_\tau `$ at the GUT-scale supports quark-lepton unification, the sharp difference between $`V_{cb}`$ versus $`\theta _{\nu _\mu \nu _\tau }^{osc}`$ disfavors such a unification. I believe that the truth is quite the opposite. This becomes apparent if one notices a simple group-theoretic property of the minimal Higgs system $`(\mathrm{𝟒𝟓}_𝐇,\mathrm{𝟏𝟔}_𝐇,\overline{\mathrm{𝟏𝟔}}_𝐇,\mathrm{𝟏𝟎}_𝐇)`$. While such a system makes SU(4)-color preserving family-symmetric contributions to fermion masses through $`\mathrm{𝟏𝟎}_𝐇`$ (which yields $`m_b^{}=m_\tau ^{}`$), it can make SU(4)-color breaking ($`BL`$)- dependent contribution denoted by “$`ϵ`$” (see Eq. (24)) only through the combination $`\mathrm{𝟏𝟎}_𝐇.\mathrm{𝟒𝟓}_𝐇/M`$, which, however, is family-antisymmetric. As a result, the $`\left(BL\right)`$-dependent contribution enters into the “23” and the “32” entries but not into the “33”-entry (see Eq. (24)).
With “$`ϵ`$” being hierarchical (of order 1/10), following diagonalization, this in turn means that the SU(4)-color breaking effect for the masses of the third family-fermions are small (of order $`ϵ^2`$) as desired to preserve the near equality $`m_b^{}m_\tau ^{}`$; but such breaking effects are necessarily large for the masses of the second family fermions (likewise for the first family), again just as desired to account for $`m_\mu ^{}m_s^{}`$. The SU(4)-color breaking effects are also large for the mixings between second and the third family fermions (arising from the “23” and “32” entries), which precisely explain why $`V_{cb}\theta _{\nu _\mu \nu _\tau }^{osc}`$ and yet $`m_b^{}m_\tau ^{}`$.
To be specific, it may be noted from the expressions for $`V_{cb}`$ and $`\theta _{\nu _\mu \nu _\tau }^{osc}`$ in Eq. (47), that while the family asymmetric and ($`BL`$)- dependent square root factors like $`\left(\eta +ϵ/\eta ϵ\right)^{1/2}`$ suppress $`V_{cb}`$, if $`ϵ`$ is relatively negative compared to $`\eta `$, the analogous factor $`\left(\eta 3ϵ/\eta +3ϵ\right)^{1/2}`$, necessarily enhances $`\theta _{\nu _\mu \nu _\tau }^{osc}`$ in a predictable manner for the same sign of $`ϵ`$ relative to $`\eta `$ (the magnitudes of $`\eta ,\sigma `$ and $`ϵ`$ are of course fixed by quark-lepton masses ). In other words, this correlation between the suppression of $`V_{cb}`$ and the enhancement of $`\theta _{\nu _\mu \nu _\tau }^{osc}`$ has come about due to the group theoretic property of $`\mathrm{𝟏𝟎}_𝐇.\mathrm{𝟒𝟓}_𝐇/M`$ which is proportional to $`BL`$, but family-antisymmetric. Note this correlation would be absent if $`126_H`$ were used to introduce $`\left(BL\right)`$dependence because its contributions would be family-symmetric, and the corresponding square root factors would reduce to unity.
Another interesting point of the hierarchical BPW model is that with $`\left|y\right|`$ being hierarchical (of order 1/10 as opposed to being of order 1) and $`m\left(\nu _2\right)/m\left(\nu _3\right)`$ being of order 1/5–1/10, it as shown in Ref. that the mixing angle from the neutrino sector $`\sqrt{m(\nu _2)/m(\nu _3)}`$ necessarily add (rather than subtract) to the contribution from the charged lepton sector (see Eq. (47)). As a result, in the BPW model, both charged lepton and neutrino-sectors give medium-large contribution ($`0.4`$) which add to naturally yield a maximal $`\theta _{\nu _\mu \nu _\tau }^{osc}`$. This thus becomes a simple and compelling prediction of the model, based essentially on the group theory of the minimal Higgs system in the context of SO(10) or G(224) and the hierarchical nature of the mass-matrices.<sup>7</sup><sup>7</sup>7The explanation of the largeness of $`\theta _{\nu _\mu \nu _\tau }^{osc}`$ together with the smallness of $`V_{cb}`$ outlined above, based on medium-large contributions from the charged lepton and neutrino sectors, is quite distinct from alternative explanations. In paricular, in the lop-sided Albright-Barr model , the largeness of $`\theta _{\nu _\mu \nu _\tau }^{osc}`$ arises almost entirely from the lop-sidedness of the charged lepton mass matrix. This distinction between the BPW and the AB models leads to markedly different predictions for the rate of $`\mu e\gamma `$ decay in the two models (see remarks later)..
The success of the model as regards the seven predictions listed above provides some confidence in the $`\mathrm{𝑔𝑟𝑜𝑠𝑠}\mathrm{𝑝𝑎𝑡𝑡𝑒𝑟𝑛}`$ of the Dirac mass matrices presented above and motivates the study of CP and flavor violations and baryogenesis within the same framework. This is what I do in the next sections.
## 5 CP and Flavor Violations in the SUSY SO(10)/G(224) Framework
### 5.1 Some Experimental Facts
On the experimental side there are now four well measured quantities reflecting CP and/or $`\mathrm{\Delta }F=2`$ flavor violations. They are:<sup>8</sup><sup>8</sup>8$`ϵ_K^{}`$ reflecting direct $`\mathrm{\Delta }F=1`$ CP violation is well measured, but its theoretical implications are at present unclear due to uncertainties in the matrix element. We discuss this later.
$`\mathrm{\Delta }m_K,ϵ_K,\mathrm{\Delta }m_{B_d}\mathrm{and}S\left(B_dJ/\mathrm{\Psi }K_S\right)`$ (57)
where $`S\left(B_dJ/\mathrm{\Psi }K_S\right)`$ denotes the asymmetry parameter in ($`B_d`$ versus $`\overline{B_d})J/\mathrm{\Psi }K_S`$ decays. It is indeed remarkable that the observed values including the signs of all four quantities as well as the empirical lower limit on $`\mathrm{\Delta }m_{B_s}`$ can consistently be realized within the standard CKM model for a single choice of the Wolfenstein parameters :
$`\overline{\rho }_W=0.178\pm 0.046;\overline{\eta }_W=0.341\pm 0.028.`$ (58)
This fit is obtained using the observed values of $`ϵ_K`$ = 2.27$`\times 10^3`$, $`V_{us}`$ = 0.2240 $`\pm `$ 0.0036, $`\left|V_{ub}\right|`$ = (3.30 $`\pm `$ 0.24)$`\times 10^3`$, $`\left|V_{cb}\right|`$ = (4.14 $`\pm `$ 0.07)$`\times 10^2`$ , $`\left|\mathrm{\Delta }m_{B_d}\right|`$ = (3.3 $`\pm `$ 0.06) $`\times 10^{13}`$ GeV and $`\mathrm{\Delta }m_{B_d}/\mathrm{\Delta }m_{B_s}`$ $`>`$ 0.035, and allowing for uncertainties in the hadronic matrix elements of up to 15%. One can then predict the asymmetry parameter $`S\left(B_dJ/\mathrm{\Psi }K_S\right)`$ in the SM to be $`0.685\pm 0.052`$. This agrees remarkably well with the observed value $`S\left(B_dJ/\mathrm{\Psi }K_S\right)_{expt.}`$ = $`0.734\pm 0.054`$, representing an average of the BABAR and BELLE results . This agreement of the SM prediction with the data in turn poses a challenge for physics beyond the SM, especially for supersymmetric grand unified (SUSY GUT) models, as these generically possess new sources of CP and flavor violations beyond those of the SM.
### 5.2 Origin of CKM CP Violation in SO(10)/G(224)
At the outset I need to say a few words about the origin of CP violation within the G(224)/SO(10)-framework presented above. Following Ref. , the discussion so far has ignored, for the sake of simplicity, possible CP violating phases in the parameters ($`\sigma `$, $`\eta `$, $`ϵ`$, $`\eta ^{}`$, $`ϵ^{}`$, $`\zeta _{22}^{u,d}`$, $`y`$, $`z`$, and $`x`$) of the Dirac and Majorana mass matrices \[Eqs. (24, and (30)\]. In general, however, these parameters can and generically will have phases . Some combinations of these phases enter into the CKM matrix and define the Wolfenstein parameters $`\rho _W`$ and $`\eta _W`$, which in turn induce CP violation by utilizing the standard model interactions.It should be stressed, however, that the values of $`(\overline{\rho }_W,\overline{\eta }_W)`$ obtained this way from a given pattern of mass matrices based on SO(10) (as in Eq. (24)) need not agree (even nearly) with the SM-based phenomenological values shown in Eq. (58), for any choice of phases of the parameters of the mass-matrices. That in turn would pose a challenge for the SO(10)-model in question as to whether it can adequately describe observed CP and flavor violations (see discussion below).
We choose to diagonalize the quark mass matrices $`M_u`$ and $`M_d`$ at the GUT scale $`2\times 10^{16}`$ GeV, by bi-unitary transformations - i.e.
$`M_d^{diag}=X_L^dM_dX_R^d\mathrm{and}M_u^{diag}=X_L^uM_uX_R^u`$ (59)
with phases of $`q_{L,R}^i`$ chosen such that the eigenvalues are real and positive and that the CKM matrix $`V_{CKM}`$ (defined below) has the Wolfenstein form ). Approximate analytic expressions for $`X_{L,R}`$ are given in Ref. .
The CKM elements in the Wolfenstein basis are given by the matrix $`V_{CKM}=e^{i\alpha }\left(X_L^uX_L^d\right)`$, where $`\alpha =\left(\varphi _{\sigma ϵ}\varphi _{\eta ϵ}\right)\left(\varphi _ϵ^{}\varphi _{\eta ^{}+ϵ^{}}\right)`$.
### 5.3 SUSY CP and Flavor Violations
SUSY Breaking
As is well known, since the model is supersymmetric, non-standard CP and flavor violations would generically arise in the model through sfermion/gaugino quantum loops involving scalar $`\left(mass\right)^2`$ transitions . The latter can either preserve chirality (as in $`\stackrel{~}{q}_{L,R}^i\stackrel{~}{q}_{L,R}^j`$) or flip chirality (as in $`\stackrel{~}{q}_{L,R}^i\stackrel{~}{q}_{R,L}^j`$). Subject to our assumption on SUSY breaking (specified below), it would turn out that these scalar $`\left(mass\right)^2`$ parameters get completely determined within our model by the fermion mass-matrices, and the few parameters of SUSY breaking.
We assume that flavor-universal soft SUSY-breaking is transmitted to the SM-sector at a messenger scale M, where M$`{}_{GUT}{}^{}<`$ M$`{}_{}{}^{}`$ M<sub>string</sub>. This may naturally be realized e.g. in models of mSUGRA , or gaugino-mediation . With the assumption of extreme universality as in CMSSM, supersymmetry introduces five parameters at the scale M:
$`m_o,m_{1/2},A_o,\mathrm{tan}\beta \mathrm{and}sgn\left(\mu \right).`$
For most purposes, we will adopt this restricted version of SUSY breaking with the added restriction that $`A_o`$ = 0 at M . However, we will not insist on strict Higgs-squark-slepton mass universality. Even though we have flavor preservation at M, flavor violating scalar (mass)<sup>2</sup>–transitions and A-terms arise in the model through RG running from M to $`M_{GUT}`$ and from $`M_{GUT}`$ to the EW scale. As described below, we thereby have three sources of flavor violation.
(i) RG Running of Scalar Masses from M to M<sub>GUT</sub>.
With family universality at the scale M, all sfermions have the mass m<sub>o</sub> at this scale and the scalar (mass)<sup>2</sup> matrices are diagonal. Due to flavor dependent Yukawa couplings, with $`h_t=h_b=h_\tau \left(=h_{33}\right)`$ being the largest, RG running from M to M<sub>GUT</sub> renders the third family lighter than the first two (see e.g. ) by the amount:
$`\mathrm{\Delta }\widehat{m}_{\stackrel{~}{b}_L}^2=\mathrm{\Delta }\widehat{m}_{\stackrel{~}{b}_R}^2=\mathrm{\Delta }\widehat{m}_{\stackrel{~}{\tau }_L}^2=\mathrm{\Delta }\widehat{m}_{\stackrel{~}{\tau }_R}^2\mathrm{\Delta }\left({\displaystyle \frac{30m_o^2}{16\pi ^2}}\right)h_t^2ln\left(M^{}/M_{GUT}\right).`$ (60)
Note the large coefficient “30”, which is a consequence of SO(10). The factor 30$``$12 for the case of G(224). The squark and slepton (mass)<sup>2</sup> matrices thus have the form $`\stackrel{~}{\mathrm{M}}^{\left(o\right)}`$ = diag(m$`{}_{}{}^{2}{}_{o}{}^{}`$, m$`{}_{}{}^{2}{}_{o}{}^{}`$, m$`{}_{o}{}^{2}\mathrm{\Delta }`$). Transforming $`\stackrel{~}{\mathrm{M}}^{\left(o\right)}`$ by $`X_{L,R}^f`$, which diagonalize fermion mass-matrices, i.e. evaluating X$`{}_{}{}^{f}{}_{L}{}^{}`$($`\stackrel{~}{\mathrm{M}}^{\left(o\right)}`$)<sub>LL</sub> X$`{}_{}{}^{f}{}_{L}{}^{}`$ and similarly for L$``$R, where f = u, d, l, introduces off-diagonal elements in the so-called SUSY basis (at the GUT-scale) given by:
$`\left(\widehat{\delta }_{LL,RR}^f\right)_{ij}=\left(X_{L,R}^f\left(\stackrel{~}{\mathrm{M}}^{\left(o\right)}\right)X_{L,R}^f\right)_{ij}/m_{\stackrel{~}{f}}^2`$ (61)
These induce flavor and CP violating transitions $`\stackrel{~}{q}_{L,R}^i\stackrel{~}{q}_{L,R}^j`$ and $`\stackrel{~}{l}_{L,R}^i\stackrel{~}{l}_{L,R}^j`$. Note that these transitions depend upon the matrices $`X_{L,R}^f`$, which are of course determined by the entries (including phases) in the fermion mass matrices (Eq. (24)). Here m$`_{\stackrel{~}{f}}`$ denotes an average squark or slepton mass (as appropriate) and the hat signifies GUT-scale values.
(ii) RG Running of the $`A`$parameters from M to M<sub>GUT</sub>.
Even if $`A_o`$ = 0 at the scale M (as we assume for concreteness, see also ). RG running from M to M<sub>GUT</sub> induces $`A`$parameters at M<sub>GUT</sub>, involving the SO(10)/G(224) gauginos and yukawa couplings ; these yield chirality flipping transitions $`\stackrel{~}{q}_{L,R}^i\stackrel{~}{q}_{R,L}^j`$ and ($`\stackrel{~}{l}_{L,R}^i\stackrel{~}{l}_{R,L}^j`$). Because of large SO(10) Casimirs, these induced A-terms arising from post-GUT physics are large even if $`\mathrm{ln}\left(M^{}/M_{GUT}\right)1`$. The chirality flipping transition angles are given by:
$`\left(\delta _{LR}^f\right)_{ij}\left(A_{LR}^f\right)_{ij}\left({\displaystyle \frac{v_f}{m_{\stackrel{~}{f}}^2}}\right).`$ (62)
Here f = u, d, l. The matrices $`A_{LR}^{ij}`$ are given explicitly in Refs. and . Note that these induced A-terms are also completely determined by the fermion mass matrices, for any given choice of the universal SUSY parameters ($`m_o,m_{1/2},\mathrm{tan}\beta `$ and $`M^{}`$).
$`\left(\mathrm{𝐢𝐢𝐢}\right)\mathrm{𝐅𝐥𝐚𝐯𝐨𝐫}\mathrm{𝐕𝐢𝐨𝐥𝐚𝐭𝐢𝐨𝐧}\mathrm{𝐓𝐡𝐫𝐨𝐮𝐠𝐡}\mathrm{𝐑𝐆}\mathrm{𝐑𝐮𝐧𝐧𝐢𝐧𝐠}\mathrm{𝐅𝐫𝐨𝐦}𝐌_{\mathrm{𝐆𝐔𝐓}}\mathrm{𝐭𝐨}𝐦_𝐖\mathrm{𝐢𝐧}\mathrm{𝐌𝐒𝐒𝐌}:`$ It is well known that, even with universal masses at the GUT scale, RG running from $`M_{GUT}`$ to $`m_W`$ in MSSM, involving contribution from the top Yukawa coupling, gives a significant correction to the mass of $`\stackrel{~}{b}_L^{}=V_{td}\stackrel{~}{d}_L+V_{ts}\stackrel{~}{s}_L+V_{tb}\stackrel{~}{b}_L`$, which is not shared by the mass-shifts of $`\stackrel{~}{b}_R,\stackrel{~}{d}_{L,R}`$ and $`\stackrel{~}{s}_{L,R}`$. This in turn induces flavor violation. Here, $`\stackrel{~}{d}_L,\stackrel{~}{s}_L`$ and $`\stackrel{~}{b}_L`$ are the SUSY partners of the physical $`d_L,s_L`$ and $`b_L`$ respectively. The differential mass shift of $`\stackrel{~}{b}_L^{}`$ arising as above, may be expressed by an effective Lagrangian : $`\mathrm{\Delta }=\left(\mathrm{\Delta }m_L^{}_{}{}^{}2\right)\stackrel{~}{b}_L^{{}_{}{}^{}}\stackrel{~}{b}_L^{}`$, where
$`\mathrm{\Delta }m_L^{}_{}{}^{}2=3/2m_o^2\eta _t+2.3A_om_{1/2}\eta _t\left(1\eta _t\right)\left(A_o^2/2\right)\eta _t\left(1\eta _t\right)+m_{1/2}^2\left(3\eta _t^27\eta _t\right).`$ (63)
Here $`\eta _t=\left(h_t/h_f\right)\left(m_t/v\mathrm{sin}\beta \right)^2\left(1/1.21\right)0.836`$ for tan$`\beta `$ = 3. Expressing $`\stackrel{~}{b}_L^{}`$ in terms of down-flavor squarks in the SUSY basis as above, Eq. (63) yields new contributions to off diagonal squark mixing. Normalizing to $`m_{sq}^2`$, they are given by
$`\delta _{LL}^{{}_{}{}^{}(12,13,23)}=\left({\displaystyle \frac{\mathrm{\Delta }m_L^{}_{}{}^{}2}{m_{sq}^2}}\right)(V_{td}^{}V_{ts},V_{td}^{}V_{tb},V_{ts}^{}V_{tb}).`$ (64)
The net chirality preserving squark $`\left(mass\right)^2`$ off-diagonal elements at $`m_W`$ are then obtained by adding the respective GUT-scale contributions from Eqs. (61) to that from Eq. (64). They are:
$`\delta _{LL}^{ij}=\widehat{\delta }_{LL}^{ij}+\delta _{LL}^{{}_{}{}^{}ij};\delta _{RR}^{ij}=\widehat{\delta }_{RR}^{ij}`$ (65)
### 5.4 The Challenge for SUSY SO(10)/G(224)
The interesting point is that the net values including phases of the off-diagonal squark-mixings, arising from the three sources listed above, and thereby the flavor and CP violations induced by them, are entirely determined within our approach by the entries in the quark mass-matrices and the choice of the universal SUSY parameters ($`m_0`$, $`m_{1/2}`$, $`M^{}`$, $`\mathrm{tan}\beta `$ and sgn$`\left(\mu \right)`$). Within the $`G\left(224\right)/SO\left(10\right)`$ framework presented in Sec. 4, the quark mass-matrices are however tightly constrained by our considerations of fermion masses and neutrino-oscillations.
The question thus arises: Can observed CP and/or flavor-violations in the quark and lepton sectors (including the empirical limits on some of these) emerge consistently within the $`G(224)/SO(10)`$-framework, for *any* choice of phases in the fermion mass-matrices of Eq. (24), while preserving all its successes with respect to fermion masses and neutrino oscillations?
This is indeed a *non-trivial challenge* to meet within the $`SO\left(10\right)`$ or $`G\left(224\right)`$-framework, since the constraints from both CP and flavor violations on the one hand and fermion masses and neutrino oscillations on the other hand are severe.
To be specific, the fact that all four entities ($`\mathrm{\Delta }m_K,ϵ_K,\mathrm{\Delta }m_{B_d}`$ and $`S\left(B_dJ/\psi K_S\right)`$) can be realized consistently in accord with experiments within the standard CKM model for a single choice of the Wolfenstein parameters $`\overline{\rho }_W`$ and $`\overline{\eta }_W`$ (Eq. (58)) strongly suggests that even for the SUSY SO(10)/G(224)-model, the corresponding SM-contributions, at least to these four entities, should be the dominant ones, with SUSY contributions being sub-dominant or small.<sup>9</sup><sup>9</sup>9The alternative of SUSY-contributions being relatively important compared to the SO(10)-based SM contributions and correcting for its pitfalls in just the right way for each of these four entities appear to be rather contrived and may require arbitrary adjustment of the many MSSM parameters. Such a scenario would at the very least mean that the good agreement between the SM-predictions and experiments is fortuitous. This in turn means that there should exist a choice of the parameters of the SO(10)-based mass matrices (like $`\sigma ,\eta ,ϵ,ϵ^{}`$ etc.), viewed in general as complex, for which not only (a) the fermion masses and (b) the CKM mixings $`\left|V_{ij}\right|`$ should be described correctly (as in Eq.(47)), but also (c) the Wolfenstein parameters $`\overline{\rho }_W^{}`$ and $`\overline{\eta }_W^{}`$ derived from the SO(10)-based mass-matrices should be close to the phenomenological SM values (Eq. (58)). A priori, a given SO(10)-model, with a specified pattern for fermion mass matrices, may not in fact be able to satisfy all three constraints (a), (b) and (c) simultaneously.<sup>10</sup><sup>10</sup>10For a discussion of the difficulties in this regard within a recently proposed SO(10)-model see e.g. Ref. .
### 5.5 The Results
Without further elaboration, I will now briefly summarize the main results of Refs. and .
(1) Allowing for phases ($`1/10`$ to $`1/2`$) in the parameters $`\eta `$, $`\sigma `$, $`ϵ^{}`$ and $`\zeta _{22}^d`$ of the $`G\left(224\right)/SO\left(10\right)`$-framework (see Eq. (24)) we found that there do exist solutions which yield masses and mixings of quarks and leptons including neutrinos, all in good accord with observations (to within 10 %), and at the same time yield the following values for the Wolfenstein parameters (see Ref. for details):
$$\overline{\rho }_W^{}0.15,\overline{\eta }_W^{}0.37.(\mathrm{SO}\left(10\right)/\mathrm{G}\left(224\right)\mathrm{model})$$
(66)
The prime here signifies that these are the values of $`\overline{\rho }_W`$ and $`\overline{\eta }_W`$ which are derived (for a suitable choice of phases in the parameters of the fermion mass matrices) from within the structure of the SO(10)-based mass-matrices (Eq. (24)). The corresponding phenomenological values are listed in Eq. (58). Note, as desired, the $`G\left(224\right)/SO\left(10\right)`$-framework presented here has turned out to be capable of yielding $`\overline{\rho }_W^{}`$ and $`\overline{\eta }_W^{}`$ close to the SM-values of $`\overline{\rho }_W`$ and $`\overline{\eta }_W`$ while preserving the successes with respect to fermion masses and neutrino oscillations as in Sec. 4. As mentioned above, this is indeed a non-trivial but most desirable feature.
(2) Including both the SM-contribution (with $`\overline{\rho }_W^{}`$ and $`\overline{\eta }_W^{}`$ as above) and the SUSY-contribution (with a plausible choice of the spectrum-e.g. $`m_{sq}\left(0.81\right)`$ TeV and $`x=\left(m_{\stackrel{~}{g}}^2/m_{sq}^2\right)0.60.8`$), we obtain :
$`\left(\mathrm{\Delta }m_K\right)_{shortdist}3\times 10^{15}\text{ GeV};`$
$`ϵ_K\left(2\text{to}\mathrm{\hspace{0.17em}2.5}\right)\times 10^3;`$
$`\mathrm{\Delta }m_{B_d}\left(3.5\text{to}\mathrm{\hspace{0.17em}3.6}\right)\times 10^{13}\text{ GeV};`$
$`S\left(B_dJ/\mathrm{\Psi }K_s\right)0.680.74.`$ (67)
We have used $`\widehat{B}_K=0.86`$ and $`f_{Bd}\sqrt{\widehat{B}_{Bd}}=215`$ MeV (see ). Now all four on which there is reliable data are in good agreement with observations (within 10%). The spectrum of ($`m_{sq}`$, $`m_{\stackrel{~}{g}}`$) considered above can be realized, for example for a choice of $`(m_0,m_{1/2})(600,220)`$ GeV. For a more complete presentation of the results involving other choices of $`(m_o,m_{1/2})`$, and a discussion on the issue of consistency with WMAP results on the LSP as cold dark matter, see Refs. and .
In all these cases, the SUSY-contribution turns out to be rather small ($`\stackrel{<}{_{}}5\%`$ in amplitude), except however for $`ϵ_K`$, for which it is sizable ($`2030\%`$) and has opposite sign, compared to the SM-contribution. Had the SUSY contribution to $`ϵ_K`$ been positive relative to the SM-contribution, $`ϵ_K`$(total) would have been too large ($`\left(3.13.5\right)\times 10^3`$), in strong disagreement with the observed value of $`2.27\times 10^3`$, despite the uncertainty in $`\widehat{B}_K`$. In short, the SUSY contribution of the model to $`ϵ_K`$ has just the right sign and nearly the right magnitude to play the desired role. This seems to be an intriguing feature of the model.
*We thus see that the SUSY $`G(224)`$ or $`SO(10)`$-framework (remarkably enough) has met all the challenges so far in being able to reproduce the observed features of both CP and quark-flavor violations as well as fermion masses and neutrino-oscillations!*
Other Predictions
Other predictions of the model which incorporate contributions from $`\delta _{LL}^{23},\delta _{RR}^{23},\delta _{LR}^{23}`$ and $`\delta _{RL}^{23}`$, include (see Ref. for details):
$`S\left(B_d\varphi K_S\right)\left(TotSM\right)0.650.73`$ (68)
$`\mathrm{\Delta }m_{B_s}\left(TotSM\right)17.3ps^1\left({\displaystyle \frac{f_{B_s}\sqrt{\widehat{B}_{B_s}}}{245MeV}}\right)^2.`$ (69)
$`A\left(bs\gamma \right)_{SUSY}\left(15\right)\%\mathrm{of}A\left(bs\gamma \right)_{SM}`$ (70)
$`Re\left(ϵ^{}/ϵ\right)_{SUSY}+\left(8.8\times 10^4\right)\left(B_G/4\right)\left(5/\mathrm{tan}\beta \right).`$ (71)
Particularly interesting is the prediction of the model that the asymmetry parameter $`S\left(B_d\varphi K_S\right)`$ should be close to the SM value of $`0.70\pm 0.10`$. At present, there is conflicting data: $`S\left(B_d\varphi K_S\right)=\left(+0.50\pm 0.25_{0.04}^{+0.07}\right)_{BaBar};\left(+0.06\pm 0.33\pm 0.09\right)_{BELLE}`$ <sup>11</sup><sup>11</sup>11At the time of completing this manuscript, the BELLE group reported a new value of $`S(B_d\varphi K_S)=+0.44\pm 0.27^\pm 0.05`$ at the 2005 Lepton-Photon Symposium . This value is close to that reported by BaBar and enhances the possibilty of the true value being close to the SM value.. It will thus be extremely interesting to see both from the point of view of the present model and the SM whether the true value of $`S\left(B_d\varphi K_S\right)`$ will turn out to be close to the SM prediction or not.
EDM’s
For a representative choice of ($`m_o,m_{1/2}`$) = (600, 300) GeV (i.e. $`m_{sq}`$ = 1 TeV, $`m_{\stackrel{~}{g}}`$ = 900 GeV, $`m_{\stackrel{~}{l}}`$ = 636 GeV and $`m_{\stackrel{~}{B}}`$ = 120 GeV), the induced A-terms (see Eq. (62)) lead to :
$`\left(d_n\right)_{A_{ind}}=(1.6,1.08)\times 10^{26}ecm\mathrm{for}\mathrm{tan}\beta =(5,10).`$ (72)
$`\left(d_e\right)_{A_{ind}}={\displaystyle \frac{1.1\times 10^{28}}{\mathrm{tan}\beta }}ecm.`$ (73)
Given the experimental limits $`d_n<6.3\times 10^{26}`$ e cm and $`d_e<4.3\times 10^{27}`$ e cm , we see that the predictions of the model (arising only from the induced $`A`$-term contributions) especially for the EDM of the neutron is in an extremely interesting range suggesting that it should be discovered with an improvement of the current limit by a factor of about 10.
## 6 Lepton Flavor Violation in SUSY SO(10)/G(224)
It has been recognized for sometime that lepton flavor violating processes (such as $`\mu e\gamma ,\tau \mu \gamma ,\mu NeNetc.`$), can provide sensitive probes into new physics beyond the SM, especially that arising in SUSY grand-unification , and that too with heavy right-handed neutrinos . In our case these get contributions from three sources:
(i) The slepton (mass)<sup>2</sup> elements $`\left(\delta m^2\right)_{LL}^{ij}`$ arising from RG-running of scalar masses from $`M^{}M_{GUT}`$ in the context of SO(10)/G(224) (see Eq. (61)),
(ii) The chirality flipping slepton (mass)<sup>2</sup> elements $`\left(\delta m^2\right)_{LR}^{ij}`$ arising from A-terms induced through RG-running from $`M^{}M_{GUT}`$ in the context of SO(10) or G(224) (see Sec. 5.3), and
(iii) $`\left(\delta m^2\right)_{LL}^{ij}`$ arising from RG-running from $`M_{GUT}`$ to the RH neutrino mass-scales $`M_{R_i}`$ involving $`\nu _R^i`$ Dirac Yukawa couplings corresponding to Eq.(24) which (in the leading log approximation) yield:
$`\left(\delta _{LL}^l\right)_{ij}^{RHN}={\displaystyle \frac{\left(3m_o^2+A_o^2\right)}{8\pi ^2}}{\displaystyle \underset{k=1}{\overset{3}{}}}\left(Y_N\right)_{ik}\left(Y_N^{}\right)_{jk}ln\left({\displaystyle \frac{M_{GUT}}{M_{R_k}}}\right).`$ (74)
Note that the masses M$`_{R_i}`$ of RH neutrinos are fairly well determined within the model (see Eq. (4)).
There is a vast literature on the subject of lepton flavor violation (LFV). (For earlier works see Ref. ; and for a partial list of references including recent works see Ref. ). Most of the works in the literature have focused only on the contribution from the third source, involving the Yukawa couplings of the RH neutrinos, which is proportional to $`\mathrm{tan}\beta `$ in the amplitude. It turns out, however, that the contributions from the first two sources arising from post-GUT physics (i.e. $`SO\left(10\right)`$-running from $`M^{}`$ to $`M_{GUT}`$) are in fact the dominant ones for $`\mathrm{tan}\beta \stackrel{<}{_{}}10`$, as long as $`\mathrm{ln}\left(M^{}/M_{GUT}\right)\stackrel{>}{_{}}1`$. We consider the contribution from all three sources by summing the corresponding amplitudes, and by varying ($`m_0`$, $`m_{1/2}`$, $`\mathrm{tan}\beta `$ and $`sgn\left(\mu \right)`$)
Here I present the predictions of the model for five different choices of $`(m_o,m_{1/2})`$ with $`\mathrm{tan}\beta =10`$ or 20 and $`\mathrm{ln}\left(M^{}/M_{GUT}\right)=1`$, to indicate the nature of the predictions. (Results for a wider choice of parameters and a more detailed discussion may be found in Ref. ). We have set $`(M_{R_1},M_{R_2},M_{R_3})=(10^{10},10^{12},5\times 10^{14})`$ GeV (see Eq. (4)) and $`A_o\left(M^{}\right)=0`$. The predicted rates for G(224) are smaller than those for SO(10) approximately by a factor of 4 to 6 (see comments in Sec. 5). The results for SO(10) are presented in Table 1.
The following points regarding these results are worth noting:
(1) We find that the contribution due to the presence of the RH neutrinos<sup>12</sup><sup>12</sup>12In the context of contributions due to the RH neutrinos alone, there exists an important distinction (partially observed by Barr, see Ref. ) between the hierarchical BPW form and the lop-sided Albright-Barr (AB) form of the mass-matrices. The amplitude for $`\mu e\gamma `$ from this source turns out to be proportional to the difference between the (23)-elements of the Dirac mass-matrices of the charged leptons and the neutrinos, with (33)-element being 1. This difference is (see Eq. (24)) is $`\eta \sigma 0.041`$, which is naturally small for the hierarchical BPW model (incidentally it is also $`V_{cb}`$), while it is order one for the lop-sided AB model. This means that the rate for $`\mu e\gamma `$ due to RH neutrinos would be about $`600`$ times larger in the AB model than the BPW model (for the same input SUSY parameters). For a comparative study of the BPW and the AB models using processes such as $`\mu e\gamma `$ and edm’s, see forthcoming paper by P. Rastogi . is about an order of magnitude smaller, in the amplitude, than those of the others arising from post-GUT physics (proportional to $`\widehat{\delta }_{LL}^{ij},\delta _{LR}^{ij}`$ and $`\delta _{RL}^{ij}`$). The latter arise from RG running of the scalar masses and the $`A`$parameters in the context of SO(10) or G(224) from $`M^{}`$ to M<sub>GUT</sub>. It seems to us that the latter, which have commonly been omitted in the literature, should exist in any SUSY GUT model for which the messenger scale for SUSY-breaking is high ($`M^{}>M_{GUT}`$), as in a mSUGRA model. The inclusion of these new contributions to LFV processes arising from post-GUT physics, that too in the context of a predictive and realistic framework, is the distinguishing feature of the study carried out in Ref. .<sup>13</sup><sup>13</sup>13For the sake of comparison, should one drop the post-GUT contribution by setting $`M^{}=M_{GUT}`$, however, the predicted $`Br\left(\mu e\gamma \right)`$, based on RHN contributions only, would be reduced significantly in our model to e.g. ($`4.2,2.9,`$ and $`8.6`$)$`\times 10^{15}`$ for cases I, II and IV respectively.
(2) Owing to the general prominence of the new contributions from post-GUT physics, we see from table 1 that case V, (with low $`m_o`$ and high $`m_{1/2}`$) is clearly excluded by the empirical limit on $`\mu e\gamma `$-rate (see Sec. 1). Case III is also excluded, for the case of SO(10), yielding a rate that exceeds the limit by a factor of about 2 (for $`\kappa =\mathrm{ln}\left(M^{}/M_{GUT}\right)\stackrel{>}{_{}}1`$), though we note that for the case of G(224), Case III is still perfectly compatible with the observed limit (see remark below table 1). All the other cases (I, II, IV, VI, and VII), with medium or moderately heavy ($`\stackrel{>}{_{}}`$ 500 GeV) sleptons , are compatible with the empirical limit, even for the case of SO(10). The interesting point about these predictions of our model, however, is that $`\mu e\gamma `$ should be discovered, even with moderately heavy sleptons ($`8001000`$ GeV), both for SO(10) and G(224), with improvement in the current limit by a factor of 10–100. Such an improvement is being planned at the forthcoming MEG experiment at PSI.
(3) We see from table 1 that $`\tau \mu \gamma `$ (leaving aside case V, which is excluded by the limit on $`\mu e\gamma `$), is expected to have a branching ratio in the range of $`2\times 10^8`$ (Case VII) to about $`\left(1\mathrm{or}2\right)\times 10^9`$ (Case VI or II). The former may be probed at BABAR and BELLE, while the latter can be reached at the LHC or a super B factory. The process $`\tau e\gamma `$ would, however, be inaccessible in the foreseeable future (in the context of our model).
(4) The WMAP-Constraint: Of the cases exhibited in table 1, Case V ($`m_o=100`$ GeV, $`m_{1/2}=440`$ GeV) would be compatible with the WMAP-constraint on relic dark matter density, in the context of CMSSM, assuming that the lightest neutralino is the LSP and represents cold dark matter (CDM), accompanying co-annihilation mechanism. (See e.g. ). As mentioned above (see table 1), a spectrum like Case V, with low $`m_o`$ and higher $`m_{1/2}`$, is however excluded in our model by the empirical limit on $`\mu e\gamma `$. Thus we infer that in the context of our model CDM cannot be associated with the co-annihilation mechanism.
Several authors (see e.g. Refs. and ), have, however considered the possibility that Higgs-squark-slepton mass universality need not hold even if family universality does. In the context of such non-universal Higgs mass (NUHM) models, the authors of Ref. show that agreement with the WMAP data can be obtained over a wide range of mSUGRA parameters. In particular, such agreement is obtained for ($`m_\varphi /m_o`$) of order unity (with either sign) for almost all the cases (I, II, III, IV, VI and VII)<sup>14</sup><sup>14</sup>14We thank A. Mustafayev and H. Baer for private communications in this regard., with the LSP (neutralino) representing CDM.<sup>15</sup><sup>15</sup>15We mention in passing that there may also be other posibilities for the CDM if we allow for either non-universal gaugino masses, or axino or gravitino as the LSP, or R-parity violation (with e.g. axion as the CDM). (Here $`m_\varphi sign\left(m_{H_{u,d}}^2\right)\sqrt{\left|m_{H_{u,d}}^2\right|}`$, see ). All these cases (including Case III for G(224)) are of course compatible with the limit on $`\mu e\gamma `$.
(5) Coherent $`\mu e`$ conversion in nuclei: In our framework, $`\mu e`$ conversion (i.e. $`\mu ^{}+Ne^{}+N`$) will occur when the photon emitted in the virtual decay $`\mu e\gamma ^{}`$ is absorbed by the nucleus (see e.g. ). In such situations, there is a rather simple relation connecting the $`\mu e`$ conversion rate with $`B\left(\mu e\gamma \right)`$: $`B\left(\mu e\gamma \right)/\left(\omega _{conversion}/\omega _{capture}\right)=R\left(230400\right)`$, depending on the nucleus. For example, $`R`$ has been calculated to be $`R389`$ for $`{}_{}{}^{27}Al`$, 238 for $`{}_{}{}^{48}Ti`$ and 342 for $`{}_{}{}^{208}Pb`$ in this type of models. (These numbers were computed in for the specific model of , but they should approximately hold for our model as well.) With the branching ratios listed in Table 1 ($`10^{11}`$ to $`10^{13}`$) for our model, $`\omega _{conversion}/\omega _{capture}`$ (40–1) $`\times 10^{15}`$. The MECO experiment at Brookhaven is expected to have a sensitivity of $`10^{16}`$ for this process, and thus will test our model.
In summary, lepton flavor violation is studied in within a predictive SO(10)/G(224)-framework, possessing supersymmetry, that was proposed in Refs. . The framework seems most realistic in that it successfully describes five phenomena: (i) fermion masses and mixings, (ii) neutrino oscillations, (iii) CP violation, (iv) quark flavor-violations, as well as (v) baryogenesis via leptogenesis (see below) . LFV emerges as an important prediction of this framework bringing no new parameters, barring the few flavor-preserving SUSY parameters.
Our results show that – (i) The decay $`\mu e\gamma `$ should be seen with improvement in the current limit by a factor of 10 – 100, even if sleptons are moderately heavy ($`800`$ GeV, say); (ii) for the same reason, $`\mu e`$ conversion ($`\mu NeN`$) should show in the planned MECO experiment, and (iii) $`\tau \mu \gamma `$ may be accessible at the LHC and a super B-factory.
## 7 Baryogenesis Via Leptogenesis Within the $`G(224)/SO(10)`$-Framework
The observed matter-antimatter asymmetry provides an important clue to physics at truly short distances. Given the existence of the RH neutrinos, as required by the symmetry $`SU\left(4\right)`$-color or $`SU\left(2\right)_R`$, possessing superheavy Majorana masses which violate B-L by two units, baryogenesis via leptogenesis has emerged as perhaps the most viable and natural mechanism for generating the baryon asymmetry of the universe. The most interesting aspect of this mechanism is that it directly relates our understanding of the light neutrino masses to our own origin. The question of whether this mechanism can quantitatively explain the magnitude of the observed baryon-asymmetry depends however crucially on the Dirac as well as the Majorana mass-matrices of the neutrinos, including the phases and the eigenvalues of the latter-i.e. $`M_1`$, $`M_2`$ and $`M_3`$ (see Eq. (4)).
This question has been considered in a recent work in the context of a realistic and predictive framework for fermion masses and neutrino oscillations, based on the symmetry $`G\left(224\right)`$ or $`SO\left(10\right)`$ , as discussed in Sec. 4, with CP violation treated as in Sec. 5. It has also been discussed in a recent review . Here I will primarily quote the results and refer the reader to Ref. for more details especially including the discussion on inflation and relevant references.
The basic picture is this. Following inflation, the lightest RH neutrinos ($`N_1`$’s) with a mass $`10^{10}`$ GeV ($`1/33`$) are produced either from the thermal bath following reheating ($`T_{RH}\text{ few}\times 10^9`$ GeV), or non-thermally directly from the decay of the inflaton <sup>16</sup><sup>16</sup>16In this case the inflaton can naturally be composed of the Higgs-like objects having the quantum numbers of the RH sneutrinos ($`\stackrel{~}{\nu }_{RH}`$ and $`\stackrel{~}{\overline{\nu }}_{RH}`$) lying in $`(1,2,4)_H`$ and $`(1,2,\overline{4})_H`$ for $`G(224)`$ (or $`16_H`$ and $`\overline{16}_H`$ for $`SO(10)`$), whose VEV’s break B-L and give Majorana masses to the RH neutrinos via the coupling shown in Eq. (26). (with $`T_{RH}`$ in this case being about $`10^710^8`$ GeV). In either case, the RH neutrinos having Majorana masses decay by utilizing their Dirac Yukawa couplings into both $`l+H`$ and $`\overline{l}+\overline{H}`$ (and corresponding SUSY modes), thus violating B-L. In the presence of CP violating phases, these decays produce a net lepton-asymmetry $`Y_L=\left(n_Ln_{\overline{L}}\right)/s`$ which is converted to a baryon-asymmetry $`Y_B=\left(n_Bn_{\overline{B}}\right)/s=CY_L`$ ($`C1/3`$ for MSSM) by the EW sphaleron effects. Using the Dirac and the Majorana mass-matrices of Sec. 4, with the introduction of CP-violating phases in them as discussed in Sec. 5, the lepton-asymmetry produced per $`N_1`$ (or ($`\stackrel{~}{N}_1+\overline{\stackrel{~}{N}}_1)`$-pair) decay is found to be :
$`ϵ_1`$ $``$ $`{\displaystyle \frac{1}{8\pi }}\left({\displaystyle \frac{_u^0}{v}}\right)^2\left|\left(\sigma +3ϵ\right)y\right|^2\mathrm{sin}\left(2\varphi _{21}\right)\times \left(3\right)\left({\displaystyle \frac{M_1}{M_2}}\right)`$ (75)
$``$ $`\left(2.0\times 10^6\right)\mathrm{sin}\left(2\varphi _{21}\right)\times \left[{\displaystyle \frac{\left(M_1/M_2\right)}{5\times 10^3}}\right]`$
*Here $`\varphi _{21}`$ denotes an effective phase depending upon phases in the Dirac as well as Majorana mass-matrices (see Ref. ).* Note that the parameters $`\sigma `$, $`ϵ`$, $`y`$ and $`\left(_u^0/v\right)`$ are already determined within our framework (to within 10 %) from considerations of fermion masses and neutrino oscillations (see Sec. 4 and 5). Furthermore, from Eq. (4) we see that $`M_1\left(1/33\right)\times 10^{10}`$ GeV, and $`M_22\times 10^{12}`$ GeV, thus $`M_1/M_2\left(5\times 10^3\right)\left(1/33\right)`$. In short, leaving aside the phase factor, the RHS of Eq. (75) is pretty well determined within our framework (to within about a factor of 5), as opposed to being uncertain by orders of magnitude either way. *This is the advantage of our obtaining the lepton-asymmetry in conjunction with a predictive framework for fermion masses and neutrino oscillations.* Now the phase angle $`\varphi _{21}`$ is uncertain because we do not have any constraint yet on the phases in the Majorana sector $`\left(M_R^\nu \right)`$. At the same time, since the phases in the Dirac sector are relatively large (see Sec. 5 and Ref. ), barring unnatural cancellation between the Dirac and Majorana phases, we would naturally expect $`\mathrm{sin}\left(2\varphi _{21}\right)`$ to be sizable-i.e. of order $`1/10`$ to $`1`$ (say).
The lepton-asymmetry is given by $`Y_L=\kappa \left(ϵ_1/g^{}\right)`$, where $`\kappa `$ denotes an efficiency factor representing wash-out effects and $`g^{}`$ denotes the light degrees of freedom ($`g^{}228`$ for MSSM). For our model, using recent discussions on $`\kappa `$ from Ref. , we obtain: $`\kappa \left(1/181/60\right)`$, for the thermal case, depending upon the $`{}_{}{}^{\prime \prime }31_{}^{\prime \prime }`$ entries in the neutrino-Dirac and Majorana mass-matrices (see Ref. ). Thus, for the thermal case, we obtain:
$$\left(Y_B\right)_{thermal}/\mathrm{sin}\left(2\varphi _{21}\right)\left(1030\right)\times 10^{11}$$
(76)
where, for concreteness, we have chosen $`M_14\times 10^9`$ GeV and $`M_21\times 10^{12}`$ GeV, in accord with Eq. (4). In this case, the reheat temperature would have to be about few $`\times 10^9`$ GeV so that $`N_1`$’s can be produced thermally. We see that the derived values of $`Y_B`$ can in fact account for the recently observed value $`\left(Y_B\right)_{WMAP}\left(8.7\pm 0.4\right)\times 10^{11}`$ , for a natural value of the phase angle $`\mathrm{sin}\left(2\varphi _{21}\right)\left(1/31\right)`$. As discussed below, the case of non-thermal leptogenesis can allow even lower values of the phase angle. It also typically yields a significantly lower reheat temperature ($`10^710^8`$ GeV) which may be in better accord with the gravitino-constraint.
For the non-thermal case, to be specific one may assume an effective superpotential : $`W_{eff}^{infl}=\lambda S\left(\overline{\mathrm{\Phi }}\mathrm{\Phi }M^2\right)+`$ (non-ren. terms) so as to implement hybrid inflation; here $`S`$ is a singlet field and $`\mathrm{\Phi }`$ and $`\overline{\mathrm{\Phi }}`$ are Higgs fields transforming as $`(1,2,4)`$ and $`(1,2,\overline{4})`$ of $`G\left(224\right)`$ which break B-L at the GUT scale and give Majorana masses to the RH neutrinos. Following the discussion in , , one obtains: $`m_{infl}=\sqrt{2}\lambda M`$, where $`M=<(1,2,4)_H>2\times 10^{16}`$ GeV; $`T_{RH}\left(1/7\right)\left(\mathrm{\Gamma }_{infl}M_{Pl}\right)^{1/2}\left(1/7\right)\left(M_1/M\right)\left(m_{infl}M_{Pl}/8\pi \right)^{1/2}`$ and $`Y_B\left(1/2\right)\left(T_{RH}/m_{infl}\right)\epsilon _1`$. Taking the coupling $`\lambda `$ in a plausible range $`\left(10^510^6\right)`$ (which lead to the desired reheat temperature, see below) and the asymmetry-parameter $`\epsilon _1`$ for the $`G\left(224\right)/SO\left(10\right)`$-framework as given in Eq. (75), the baryon-asymmetry $`Y_B`$ can then be derived. The values for $`Y_B`$ thus obtained are listed in Table 2.
The variation in the entries correspond to taking $`M_1=\left(2\times 10^{10}\text{ GeV}\right)\left(11/3\right)`$ with $`M_2=\left(2\times 10^{12}\right)`$ GeV in accord with Eq. (4). We see that for this case of non-thermal leptogenesis, one quite plausibly obtains
$`\left(Y_B\right)_{NonThermal}\left(89\right)\times 10^{11}`$ (77)
in full accord with the WMAP data, for natural values of the phase angle $`\mathrm{sin}\left(2\varphi _{21}\right)\left(1/31/10\right)`$, and with $`T_{RH}`$ being as low as $`10^7`$ GeV $`\left(21/2\right)`$. Such low values of the reheat temperature are fully consistent with the gravitino-constraint for $`m_{3/2}400`$ GeV $`1`$ TeV (say), even if one allows for possible hadronic decays of the gravitinos for example via $`\gamma \stackrel{~}{\gamma }`$-modes .
In summary, I have presented two alternative scenarios (thermal as well as non-thermal) for inflation and leptogenesis. We see that the $`G\left(224\right)/SO\left(10\right)`$-framework provides a simple and unified description of not only fermion masses, neutrino oscillations (consistent with maximal atmospheric and large solar oscillation angles) *and* CP violation, but also of baryogenesis via leptogenesis, in either scenario. Each of the following features - (a) the existence of the RH neutrinos, (b) B-L local symmetry, (c) $`SU\left(4\right)`$-color, (d) the SUSY unification scale, (e) the seesaw mechanism, and (f) the pattern of $`G\left(224\right)/SO\left(10\right)`$ mass-matrices allowed in the minimal Higgs system (see Sec. 4)-have played crucial roles in realizing this *unified and successful description*.
## 8 Proton Decay
Perhaps the most dramatic prediction of grand unification is proton decay. I have discussed proton decay in the context of the SUSY $`SO\left(10\right)/G\left(224\right)`$-framework presented here in some detail in recent reviews which are updates of the results obtained in . Here, I will present only the salient features and the updated results. In SUSY unification there are in general three distinct mechanisms for proton decay.
1. The familiar d=6 operators mediated by $`X`$ and $`Y`$ gauge bosons of $`SU\left(5\right)`$ and $`SO\left(10\right)`$ As is well known, these lead to $`e^+\pi ^0`$ as the dominant mode with a lifetime $`10^{35.3\pm 1}`$ yrs.
2. The “standard” $`d=5`$ operators which arise through the exchange of the color-triplet Higgsinos which are in the $`5_H+\overline{5}_H`$ of $`SU\left(5\right)`$ or $`10_H`$ of $`SO\left(10\right)`$. These operators require (for consistency with proton lifetime limits) that the color-triplets be made superheavy while the EW-doublets are kept light by a suitable doublet-triplet splitting mechanism (for $`SO\left(10\right)`$, see Ref. . They lead to dominant $`\overline{\nu }K^+`$ and comparable $`\overline{\nu }\pi ^+`$ modes with lifetimes varying from about $`10^{29}`$ to $`10^{34}`$ years, depending upon a few factors, which include the nature of the SUSY-spectrum and the matrix elements (see below). In the present context, see . Some of the original references on contributions of standard $`d=5`$ operators to proton decay may be found in
3. The so called “new” $`d=5`$ operators which can generically arise through the exchange of color-triplet Higgsinos in the Higgs multiplets like $`\left(16_H+\overline{16}_H\right)`$ of $`SO\left(10\right)`$. Such exchanges are possible by utilizing the joint effects of (a) the couplings given in Eq. (26) which assign superheavy Majorana masses to the RH neutrinos through the VEV of $`\overline{16}_H`$, and (b) the coupling of the form $`g_{ij}16_i16_j16_H16_H/M`$ (see Eq. (25)) which are needed, at least for the minimal Higgs-system, to generate CKM-mixings. These operators also lead to $`\overline{\nu }K^+`$ and $`\overline{\nu }\pi ^+`$ as the dominant modes, and they can quite plausibly lead to lifetimes in the range of $`10^{32}10^{34}`$ yrs \[see below\]. These operators, though most natural in a theory with Majorana masses for the RH neutrinos, have been invariably omitted in the literature.
One distinguishing feature of the new $`d=5`$ operator is that they directly link proton decay to neutrino masses via the Majorana masses of the RH neutrinos. *The other, and perhaps most important, is that these new $`d=5`$ operators can induce proton decay even when the $`d=6`$ and standard $`d=5`$ operators mentioned above are absent.* This is what could happen if the string theory or a higher dimensional GUT-theory leads to an effective $`G\left(224\right)`$-symmetry in $`4D`$, which would be devoid of both $`X`$ and $`Y`$ gauge bosons and the dangerous color-triplets in the $`10_H`$ of $`SO\left(10\right)`$. *By the same token, for an effective $`G(224)`$-theory, these new $`d=5`$ operators can become the sole and viable source of proton decay leading to lifetimes in an interesting range (see below).*
Our study of proton decay carried out in Ref. and updated in and has a few distinctive features: (i) It is based on a *realistic framework for fermion masses and neutrino oscillations*, as discussed in Sec. 4;(ii) It includes the *new $`d=5`$ operators* in addition to the standard $`d=5`$ and $`d=6`$ operators; (iii) It restricts *GUT-scale threshold-corrections* to $`\alpha _3\left(m_Z\right)`$ so as to be in accord with the demand of “natural” coupling unification and thereby restricts $`M_{eff}`$ that controls the strength of the standard $`d=5`$ operators; and (iv) It allows for the ESSM extension of MSSM motivated on several grounds (see e.g. and ), which introduces two vectorlike families in $`16+\overline{16}`$ of $`SO\left(10\right)`$ with masses of order $`1`$ TeV, in addition to the three chiral families.
Guided by recent calculation based on quenched lattice QCD in the continuum limit and renormalization factors $`A_L`$ and $`A_s`$ for d = 5 as in , we take (see Ref. for details): $`\left|\beta _H\right|\left|\alpha _H\right|\left(0.009\text{ GeV}^3\right)\left(1/\sqrt{2}\sqrt{2}\right)`$; $`m_{\stackrel{~}{q}}m_{\stackrel{~}{l}}1.2`$ TeV $`\left(1/22\right)`$; $`\left(m_{\stackrel{~}{W}}/m_{\stackrel{~}{q}}\right)=1/6\left(1/22\right)`$; $`M_{H_C}\left(\text{min}SU\left(5\right)\right)10^{16}`$ GeV, $`A_L0.32`$, $`A_S0.93`$, $`\mathrm{tan}\beta 3`$; $`M_XM_Y10^{16}`$ GeV $`\left(1\pm 25\%\right)`$, and $`A_R\left(d=6,e^+\pi ^0\right)3.4`$.
The theoretical predictions for proton decay for the cases of minimal SUSY $`SU\left(5\right)`$, SUSY $`SO\left(10\right)`$ and $`G\left(224\right)`$-models developed in Secs. 3 and 4, are summarized in Table 3. They are obtained by following the procedure as in and using the parameters as mentioned above.<sup>17</sup><sup>17</sup>17The chiral Lagrangian parameter ($`D+F`$) and the renormalization factor $`A_R`$ entering into the amplitude for $`pe^+\pi ^0`$ decay are taken to be 1.25 and 3.4 respectively.
It should be stressed that the upper limits on proton lifetimes given in Table 3 are quite conservative in that they are obtained (especially for the top two cases) by stretching the uncertainties in the matrix element and the SUSY spectra to their extremes so as to prolong proton lifetimes. In reality, the lifetimes should be shorter than the upper limits quoted above.
Now the experimental limits set by SuperK studies are as follows :
$`\mathrm{\Gamma }^1\left(pe^+\pi ^0\right)_{\mathrm{expt}}`$ $``$ $`6\times 10^{33}\text{ yrs}`$
$`\mathrm{\Gamma }^1\left(p\overline{\nu }K^+\right)_{\mathrm{expt}}`$ $``$ $`1.9\times 10^{33}\text{ yrs}`$ (116)
The following comments are in order.
1. By comparing the upper limit given in Eq. (83) with the experimental lower limit, we see that the *minimal* SUSY $`SU\left(5\right)`$ with the conventional MSSM spectrum is clearly excluded by a large margin by proton decay searches. This is in full agreement with the conclusion reached by other authors (see e.g. Ref. ).<sup>18</sup><sup>18</sup>18See, however, Refs. and , where attempts are made to save minimal SUSY SU(5) by a set of scenarios. These include a judicious choice of sfermion mixings, higher dimensional operators and squarks of first two families having masses of order 10 TeV.
2. By comparing Eq. (89) with the empirical lower limit, we see that the case of MSSM embedded in $`SO\left(10\right)`$ is already tightly constrained to the point of being disfavored by the limit on proton lifetime. The constraint is of course augmented by our requirement of *natural coupling unification*, which prohibits accidental large cancelation between different threshold corrections (see ).
3. In contrast to the case of MSSM, that of ESSM embedded in $`SO\left(10\right)`$, which has been motivated on several grounds<sup>19</sup><sup>19</sup>19The case of ESSM, which introduces two vector like families, i.e. $`16+\overline{16}`$ of SO(10), with a mass of order 1 TeV, has been motivated by a number of considerations independently of proton decay . These include: (a) dilaton stabilization through a semi-perturbative unification, (b) coupling unification with a better prediction for $`\alpha _3(m_Z)`$ compared to that for MSSM, (c) a simple understanding of the inter-family mass hierarchy, and (d) a possible explanation of a 2.7 $`\sigma `$ anomaly in $`(g2)_\mu `$. The vector like families with mass of order 1 TeV can of course be searched for at the LHC., is fully compatible with the SuperK limit (see Eq. (99)). In this case, $`\mathrm{\Gamma }_{\mathrm{Med}}^1\left(p\overline{\nu }K^+\right)10^{33}10^{34}\text{ yrs}`$, given in Eq. (99), corresponds to the parameters involving the SUSY spectrum and the matrix element $`\beta _H`$ being in the *median range*, close to their central values.
4. We see from Eq. (109) that the contribution of the new operators related to the Majorana masses of the RH neutrinos (which is the same for MSSM and ESSM and is independent of $`\mathrm{tan}\beta `$) is fully compatible with the SuperK limit. These operators can quite naturally lead to proton lifetimes in the range of $`10^{33}10^{34}`$ yrs with an upper limit of about $`2\times 10^{34}`$ yrs.
In summary for this section, within the $`SO\left(10\right)/G\left(224\right)`$ framework and with the inclusion of the standard as well as the new $`d=5`$ operators, one obtains (see Eqs. (89)–(8)) a conservative upper limit on proton lifetime given by:
$$\begin{array}{cc}\tau _{\mathrm{proton}}\stackrel{<}{_{}}\left(1/32\right)\times 10^{34}\text{ yrs}\hfill & \hfill \left(\begin{array}{c}\text{SUSY}\\ SO\left(10\right)/G\left(224\right)\end{array}\right)\end{array}$$
(117)
with $`\overline{\nu }K^+`$ and $`\overline{\nu }\pi ^+`$ being the dominant modes and quite possibly $`\mu ^+K^0`$ being prominent.
The $`e^+\pi ^0`$-mode induced by gauge boson-exchanges should have an inverse decay rate in the range of $`10^{34}10^{36}`$ years (see Eq. (35)). The implication of these predictions for a next-generation detector is noted in the next section.
## 9 Concluding Remarks
The neutrinos seem to be as elusive as revealing. Simply by virtue of their tiny masses, they provide crucial information on the unification-scale, and even more important on the nature of the unification-symmetry. In particular, as argued in Secs. 4 and 6, (a) the magnitude of the superK-value of $`\sqrt{\delta m_{23}^2}\left(1/20\text{ eV}\right)`$, (b) the $`b/\tau `$ mass-ratio, and (c) the need for baryogenesis via leptogenesis, together, provide clear support for: (i) the existence of the $`SU\left(4\right)`$-color symmetry in 4D above the GUT-scale which provides not only the RH neutrinos but also B-L as a local symmetry and a value for $`m\left(\nu _{\text{Dirac}}^\tau \right)`$; (ii) the familiar SUSY unification-scale which provides the scale of $`M_R`$; and (iii) the seesaw mechanism. *In turn this chain of arguments selects out the effective symmetry in $`4D`$ being either a string-derived $`G(224)`$ or $`SO(10)`$-symmetry, as opposed to the other alternatives like $`SU(5)`$ or flipped $`SU(5)^{}\times U(1)`$*.
It is furthermore remarkable that the tiny neutrino-masses also seem to hold the key to the origin of baryon excess and thus to our own origin!
In this talk, I have tried to highlight that the $`G\left(224\right)/SO\left(10\right)`$-framework as described here is capable of providing a *unified description* of a set of phenomena including: fermion masses, neutrino oscillations, CP and flavor violations as well as of baryogenesis via leptogenesis. This seems non-trivial.
The neutrinos have clearly played a central role in arriving at this unified description, first (a) by providing a clue to the nature of the unification-symmetry (as noted above), second (b) by confirming certain group-theoretic correlations between the quark and lepton sectors as regards their masses and mixings (cf. $`m\left(\nu _\tau \right)_{\mathrm{Dirac}}`$ versus $`m_{top}`$ and $`\theta _{\nu _\mu \nu _\tau }^{osc}`$ versus $`V_{cb}`$), and (c) by yielding naturally the desired magnitude for the baryon excess. Hence the title of the paper.
The framework is also highly predictive and can be further tested by studies of CP and flavor violations in processes such as (a) $`B_d\varphi K_S`$-decay, (b) $`(B_S,\overline{B}_S)`$-decays, (c) edm of neutron, and (d) leptonic flavor violations as in $`\mu e\gamma `$ and $`\tau \mu \gamma `$-decays, and in $`\mu NeN`$.
To conclude, the evidence in favor of supersymmetric grand unification, based on a string-derived $`G\left(224\right)`$-symmetry in 4D (as described in Sec. 3) or $`SO\left(10\right)`$-symmetry, appears to be strong. It includes:
* Quantum numbers of all members in a family,
* Quantization of electric charge,
* Gauge coupling unification,
* $`m_b^0m_\tau ^0`$
* $`\sqrt{\delta m^2\left(\nu _2\nu _3\right)}1/20`$ eV,
* A maximal $`\mathrm{\Theta }_{23}^\nu \pi /4`$ with a minimal $`V_{cb}0.04`$, and
* Baryon Excess $`Y_B10^{10}`$.
All of these features and more including (even) CP and flavor violations hang together neatly within a single unified framework based on a presumed string-derived four-dimensional $`G\left(224\right)`$ or $`SO\left(10\right)`$-symmetry, with supersymmetry. It is hard to believe that this neat fitting of all these pieces emerging as predictions of one and the same framework can be a mere coincidence. It thus seems pressing that dedicated searches be made for the two missing pieces of this picture-that is supersymmetry and proton decay. The search for supersymmetry at the LHC and a possible future NLC is eagerly awaited. That for proton decay will need a next-generation megaton-size underground detector.
Acknowledgments: I would like to thank Milla Baldo-Ceolini for her kind hospitality. I have benefited from many collaborative discussions with Kaladi S. Babu, Parul Rastogi and Frank Wilczek on topics covered in this lecture. Conversations with Milla Baldo-Ceolini, Manfred Lindner, Antonio Masiero and Qaisar Shafi during the Venice conference leading to bets on some interesting issues in particle physics, which will be settled soon by experiments, were enlightening. The research presented here is supported in part by DOE grant No. DE-FG02-96ER-41015. |
warning/0507/hep-th0507025.html | ar5iv | text | # Effects of a mixed vector-scalar screened Coulomb potential for spinless particles
## 1 Introduction
In a two-dimensional space-time the screened Coulomb potential ($`e^{|x|/\lambda }`$) has been analyzed and its analytical solutions have been found for the Dirac equation with vector , scalar and pseudoscalar couplings and for the Klein-Gordon (KG) equation with vector and scalar couplings. As has been emphasized in Refs. and , the solution of the KG equation with this sort of potential may find applications in the study of pionic atoms, doped Mott insulators, doped semiconductors, interaction between ions, quantum dots surrounded by a dielectric or a conducting medium, protein structures, etc.
In the present work the problem of a spinless particle in the background of a screened Coulomb potential is considered with a general mixing of vector and scalar Lorentz structures. This sort of mixing beyond its potential physical applications, shows to be a powerful tool to obtain a deeper insight about the nature of the KG equation and its solutions. The problem is mapped into an exactly solvable Sturm-Liouville problem of a Schrödinger-like equation with an effective symmetric Morse-like potential, or an effective screened Coulomb potential in particular circumstances. The cases of pure vector and scalar potentials, already analyzed in -, are obtained as particular cases.
In the presence of vector and scalar potentials the 1+1 dimensional time-independent KG equation for a spinless particle of rest mass $`m`$ reads
$$\mathrm{}^2c^2\frac{d^2\psi }{dx^2}+\left(mc^2+V_s\right)^2\psi =\left(EV_v\right)^2\psi $$
(1)
where $`E`$ is the energy of the particle, $`c`$ is the velocity of light and $`\mathrm{}`$ is the Planck constant. The vector and scalar potentials are given by $`V_v`$ and $`V_s`$, respectively. The subscripts for the terms of potential denote their properties under a Lorentz transformation: $`v`$ for the time component of the 2-vector potential and $`s`$ for the scalar term. It is worth to note that the KG equation is covariant under $`xx`$ if $`V_v(x)`$ and $`V_s(x)`$ remain the same. Also note that $`\psi `$ remains invariant under the simultaneous transformations $`EE`$ and $`V_vV_v`$. Furthermore, for $`V_v=0`$, the case of a pure scalar potential, the negative- and positive-energy levels are disposed symmetrically about $`E=0`$.
The KG equation can also be written as
$$H_{eff}\psi =\frac{\mathrm{}^2}{2m}\psi ^{\prime \prime }+V_{eff}\psi =E_{eff}\psi $$
(2)
where
$$E_{eff}=\frac{E^2m^2c^4}{2mc^2},V_{eff}=\frac{V_s^2V_v^2}{2mc^2}+V_s+\frac{E}{mc^2}V_v$$
(3)
From this one can see that for potentials which tend to $`\pm \mathrm{}`$ as $`|x|\mathrm{}`$ it follows that $`V_{eff}\left(V_s^2V_v^2\right)/\left(2mc^2\right)`$, so that the KG equation furnishes a purely discrete (continuum) spectrum for $`|V_s|>|V_v|`$ ($`|V_s|<|V_v|`$). On the other hand, if the potentials vanish as $`|x|\mathrm{}`$ the continuum spectrum is omnipresent but the necessary conditions for the existence of a discrete spectrum is not an easy task for general functional forms. The boundary conditions on the eigenfunctions come into existence by demanding that the effective Hamiltonian given (2) is Hermitian, viz.
$$_a^b𝑑x\psi _n^{}\left(H_{eff}\psi _n^{^{}}\right)=_a^b𝑑x\left(H_{eff}\psi _n\right)^{}\psi _n^{^{}}$$
(4)
where $`\psi _n`$ is an eigenfunction corresponding to an effective eigenvalue $`\left(E_{eff}\right)_n`$ and $`(a,b)`$ is the interval under consideration. In passing, note that a necessary consequence of Eq. (4) is that the eigenfunctions corresponding to distinct effective eigenvalues are orthogonal. It can be shown that (4) is equivalent to
$$\left[\psi _n^{}\frac{d\psi _n^{^{}}}{dx}\frac{d\psi _n^{}}{dx}\psi _n^{^{}}\right]_{x=a}^{x=b}=0$$
(5)
In the nonrelativistic approximation (potential energies small compared to $`mc^2`$ and $`Emc^2`$) Eq. (1) becomes
$$\left(\frac{\mathrm{}^2}{2m}\frac{d^2}{dx^2}+V_v+V_s\right)\psi =\left(Emc^2\right)\psi $$
(6)
so that $`\psi `$ obeys the Schrödinger equation with binding energy equal to $`Emc^2`$ without distinguishing the contributions of vector and scalar potentials.
It is remarkable that the KG equation with a scalar potential, or a vector potential contaminated with some scalar coupling, is not invariant under $`VV+const.`$, this is so because only the vector potential couples to the positive-energies in the same way it couples to the negative-ones, whereas the scalar potential couples to the mass of the particle. Therefore, if there is any scalar coupling the absolute values of the energy will have physical significance and the freedom to choose a zero-energy will be lost. It is well known that a confining potential in the nonrelativistic approach is not confining in the relativistic approach when it is considered as a Lorentz vector. It is surprising that relativistic confining potentials may result in nonconfinement in the nonrelativistic approach. This last phenomenon is a consequence of the fact that vector and scalar potentials couple differently in the KG equation whereas there is no such distinction among them in the Schrödinger equation. This observation permit us to conclude that even a “repulsive” potential can be a confining potential. The case $`V_v=V_s`$ presents bounded solutions in the relativistic approach, although it reduces to the free-particle problem in the nonrelativistic limit. The attractive vector potential for a particle is, of course, repulsive for its corresponding antiparticle, and vice versa. However, the attractive (repulsive) scalar potential for particles is also attractive (repulsive) for antiparticles. For $`V_v=V_s`$ and an attractive vector potential for particles, the scalar potential is counterbalanced by the vector potential for antiparticles as long as the scalar potential is attractive and the vector potential is repulsive. As a consequence there is no bounded solution for antiparticles. For $`V_v=0`$ and a pure scalar attractive potential, one finds energy levels for particles and antiparticles arranged symmetrically about $`E=0`$. For $`V_v=V_s`$ and a repulsive vector potential for particles, the scalar and the vector potentials are attractive for antiparticles but their effects are counterbalanced for particles. Thus, recurring to this simple standpoint one can anticipate in the mind that there is no bound-state solution for particles in this last case of mixing.
## 2 The mixed vector-scalar screened Coulomb potential
Now let us focus our attention on scalar and vector potentials in the form
$$V_s=\frac{g_s}{2\lambda }\mathrm{exp}\left(\frac{|x|}{\lambda }\right),V_v=\frac{g_v}{2\lambda }\mathrm{exp}\left(\frac{|x|}{\lambda }\right)$$
(7)
where the coupling constants, $`g_s`$ and $`g_v`$, are dimensionless real parameters and $`\lambda `$, related to the range of the interaction, is a positive parameter. In this case the second equation of (3) transmutes into
$$V_{eff}=V_1\mathrm{exp}\left(\frac{|x|}{\lambda }\right)+V_2\mathrm{exp}\left(2\frac{|x|}{\lambda }\right)$$
(8)
where
$$V_1=\frac{1}{2\lambda }\left(g_s+\frac{E}{mc^2}g_v\right),V_2=\frac{g_s^2g_v^2}{8\lambda ^2mc^2}$$
(9)
Therefore, one has to search for bounded solutions in an effective symmetric Morse-like potential for $`g_s^2g_v^2`$, or screened Coulomb potential for $`g_s^2=g_v^2`$. The KG eigenvalues are obtained by inserting the effective eigenvalues into the first equation of (3). Since the effective potential is even under $`xx`$, the KG eigenfunction can be expressed as a function of definite parity. Thus, we can concentrate our attention on the positive half-line and impose boundary conditions on $`\psi `$ at $`x=0`$ and $`x=+\mathrm{}`$. From (5) one can see that in addition to $`\psi \left(\mathrm{}\right)=0`$, the boundary conditions can be met in two distinct ways: the odd function obeys the Neumann condition at the origin ($`d\psi /dx|_{x=0}=0`$) whereas the even function obeys the Dirichlet condition ($`\psi \left(0\right)=0`$) .
Note carefully that the potentials $`V_s`$ and $`V_v`$ vanish as $`|x|\mathrm{}`$ and the KG equation can furnish a discrete spectrum when $`V_1<0`$ and $`V_20`$, or $`V_1<|V_2|`$ and $`V_2<0`$. Only in those circumstances the effective potentials present potential-well structures permitting bounded solutions in the range $`|E|<mc^2`$. The eigenenergies in the range $`|E|>mc^2`$ correspond to the continuum.
Now we move to consider a quantitative treatment of our problem by considering the two distinct classes of effective potentials.
### 2.1 The effective screened Coulomb potential ($`g_s^2=g_v^2`$)
For this class of effective potential, the discrete spectrum arises when $`V_1<0`$ and $`V_2=0`$, corresponding to $`g_s\left[1+\mathrm{sgn}\left(g_v\right)E/\left(mc^2\right)\right]>0`$ and $`g_s=|g_v|`$. Defining the dimensionless quantities
$`y`$ $`=`$ $`y_0\mathrm{exp}\left({\displaystyle \frac{|x|}{2\lambda }}\right),y_0={\displaystyle \frac{2}{\mathrm{}}}\sqrt{\lambda mg_s\left[1+{\displaystyle \frac{E}{mc^2}}\mathrm{sgn}\left(g_v\right)\right]}`$
$`\mu `$ $`=`$ $`{\displaystyle \frac{2\lambda mc}{\mathrm{}}}\sqrt{1{\displaystyle \frac{E^2}{m^2c^4}}}`$
and using (2)-(3) and (8)-(9) one obtains the differential Bessel equation
$$y^2\psi ^{\prime \prime }+y\psi ^{}+\left(y^2\mu ^2\right)\psi =0$$
(11)
where the prime denotes differentiation with respect to $`y`$. The solution finite at $`y=0`$ ($`|x|=\mathrm{}`$) is given by the Bessel function of the first kind and order $`\mu `$ :
$$\psi (y)=N_\mu J_\mu (y)$$
(12)
where $`N_\mu `$ is a normalization constant. In fact, the normalizability of $`\psi `$ demands that the integral $`_0^{y_0}y^1|J_\mu (y)|^2𝑑y`$ must be convergent. Since $`J_\mu (y)`$ behaves as $`y^\mu `$ at the lower limit, one can see that $`\mu 1/2`$ so that square-integrable KG eigenfuntions are allowed only if $`\lambda \lambda _c/4`$, where $`\lambda _c=\mathrm{}/(mc)`$ is the Compton wavelength. The boundary conditions at $`x=0`$ ($`y=y_0`$) imply that
$$\begin{array}{cc}\frac{dJ_\mu (y)}{dy}|_{y=y_0}=0,\hfill & \text{for even states}\hfill \\ & \\ J_\mu (y_0)=0,\hfill & \text{for odd states}\hfill \end{array}$$
(13)
Since the KG eigenenergies are dependent on $`\mu `$ and $`y_0`$, it follows that Eq. (13) is a quantization condition. The allowed values for the parameters $`\mu `$ and $`y_0`$, and $`E`$ as an immediate consequence, are determined by solving Eq. (13). The oscillatory character of the Bessel function and the finite range for $`y`$ ($`0<yy_0`$) imply that there is a finite number of discrete KG eigenenergies. Of course, the number of bound-state solutions increases as $`y_0`$ is increased. As we let $`\lambda \lambda _c/4`$, it can now be realized that the energy levels approach $`E=0`$ and tend to disappear one after another. It happens that an isolated zero-energy solution survives when $`\lambda =\lambda _c/4`$, regardless the relative values of the coupling constants (recall that $`g_s>0`$).
The roots of $`J_\mu (y)`$ and $`J_\mu ^{}(y)`$ are listed in tables of Bessel functions only for a few special values of $`\mu `$. A bit of time and effort can be saved in the numerical calculation of the roots of $`J_\mu ^{}(y)`$ if one uses the recurrence relation $`J_{\mu 1}J_{\mu +1}=2J_\mu ^{}`$, in such a manner that the quantization condition for even states translates into $`J_{\mu +1}(y_0)=J_{\mu 1}(y_0)`$.
When $`g_s=g_v`$ ($`g_s=g_v`$) the single-well potential is deeper (shallower) for positive-energy levels than that one for negative-energy levels. Thus, the capacity to hold bound states depends on the sign of the eigenenergy and one might expect that the number of positive (negative) energy levels is greater than the number of negative (positive) energy levels. By the way, the positive (negative) energy solutions are not to be promptly identified with the solutions for particles (antiparticles). Rather, whether it is positive or negative, an eigenenergy can be unambiguously identified with a bounded solution for a particle (antiparticle) only by observing if the energy level emerges from the upper (lower) continuum.
The KG eigenenergies are plotted in Fig. 1 for the four lowest bound states as a function of $`g_s`$ for $`g_v=g_s`$ and $`\lambda =2\lambda _c`$. The eigenenergies for $`g_v=g_s`$ can be obtained by changing $`E`$ by $`E`$, as mentioned before. Note that in the case illustrated in Fig. 1 the eigenenergies correspond to bounded solutions for particles. There are no energy levels for antiparticles. Also, note that the energy level corresponding to the ground-state solution ($`\psi `$ even) always makes its appearance and that the number of energy levels grows with $`g_s`$. The spectrum consists of a finite set of energy levels of alternate parities. The nonrelativistic limit is only viable for $`g_s=g_v`$ whereas the case $`g_s=g_v`$ is an essentially relativistic problem. Furthermore, one has $`Emc^2`$ as long as $`g_s1`$.
### 2.2 The effective Morse-like potential ($`g_s^2g_v^2`$)
For this class, the existence of bound-state solutions permits us to distinguish two subclasses: 1) $`V_1<0`$ and $`V_2>0`$, corresponding to $`g_s+g_vE/\left(mc^2\right)>0`$ and $`g_s>|g_v|`$; 2) $`V_1<|V_2|`$ and $`V_2<0`$, corresponding to $`g_s+g_vE/\left(mc^2\right)>\left(g_v^2g_s^2\right)/\left(4mc^2\right)`$ and $`|g_s|<|g_v|`$. Included into the first (second) subclass is the case of a pure scalar (vector) coupling. The first subclass, as well as the second one on the condition that $`g_v>|g_s|`$, contain the nonrelativistic theory as a limiting case. On the contrary, i.e. $`g_s<|g_v|`$ and $`g_v<0`$, the theory is essentially relativistic. Let us define
$`z`$ $`=`$ $`z_0\mathrm{exp}\left({\displaystyle \frac{|x|}{\lambda }}\right),z_0={\displaystyle \frac{\sqrt{g_s^2g_v^2}}{\mathrm{}c}}`$
$`\rho `$ $`=`$ $`{\displaystyle \frac{\lambda m}{\mathrm{}^2z_0}}\left(g_s+{\displaystyle \frac{E}{mc^2}}g_v\right),\nu ={\displaystyle \frac{\lambda mc}{\mathrm{}}}\sqrt{1{\displaystyle \frac{E^2}{m^2c^4}}}`$
so that
$$z\psi ^{\prime \prime }+\psi ^{}+\left(\frac{z}{4}\frac{\nu ^2}{z}+\rho \right)\psi =0$$
(15)
Note that $`\psi `$ is a function of complex variable, $`z`$, if $`g_s^2<g_v^2`$. Following the steps of Refs. and , we make the transformation $`\psi =z^{1/2}\varphi `$ to obtain the Whittaker equation :
$$\varphi ^{\prime \prime }+\left(\frac{1}{4}+\frac{\rho }{z}+\frac{1/4\nu ^2}{z^2}\right)\varphi =0$$
(16)
whose solution vanishing at the infinity is written as $`\varphi =Nz^{\nu +1/2}e^{z/2}M(a,b,z)`$, where $`N`$ is a normalization constant and $`M`$ is the regular confluent hypergeometric function with
$$a=\nu +\frac{1}{2}\rho ,b=2\nu +1$$
(17)
Thus,
$$\psi =Nz^\nu e^{z/2}M(a,b,z)$$
(18)
For this class of effective potential there is no restriction on the size of $`\lambda `$ in order to make the existence of a bounded solution possible as there is for the previous class. The KG eigenfunction is normalizable for any $`\nu `$ as easy inspection shows. Therefore, one can think of a very short-ranged potential in the sense of $`\lambda 0`$, i.e. a potential approaching the $`\delta `$-function. Indeed, this sort of limit has already been realized by Domínguez-Adame and Rodríguez for the case of a pure vector potential. From Eq. (18) one can see now that the boundary conditions at $`x=0`$ ($`z=z_0`$) imply into the quantization conditions
$$\begin{array}{cc}\frac{M(a+1,b+1,z_0)}{M(a,b,z_0)}=\frac{z_02\nu }{2\frac{a}{b}z_0},\hfill & \text{for even states}\hfill \\ & \\ M(a,b,z_0)=0,\hfill & \text{for odd states}\hfill \end{array}$$
(19)
If $`g_s`$ happens to vanish, the spectrum will only consist of positive (negative) energy levels for $`g_v>0`$ ($`g_v<0`$). If $`g_s0`$, though, the spectrum may acquiesce both signs of eigenenergies. The presence of both signs of eigenenergies depends, of course, on the relative strength between the vector and scalar potentials. When $`g_v=0`$ the negative- and positive-energy levels are disposed symmetrically about $`E=0`$, as commented before, so that there are as many positive-energy levels as negative ones. In the case $`g_s>|g_v|`$ it is reasonable to expect a two-fold degeneracy as $`g_s/|g_v|\mathrm{}`$ due to the double-well structure with an infinitely high barrier potential between the wells. That degeneracy in an one-dimensional quantum-mechanical problem is due to the fact that even eigenfunctions tend to vanish at the origin as $`g_s/|g_v|\mathrm{}`$.
The Fig. 2 illustrates the four lowest states of the spectrum for this class of effective potential as a function of $`g_s/|g_v|`$ with $`g_v>0`$. The energy level corresponding to the ground-state solution ($`\psi `$ even) always makes its appearance. As before, the eigenenergies for $`g_v<0`$ can be obtained by replacing $`E`$ by $`E`$ (recall that $`g_s>|g_v|`$). Note that only bounded solutions for particles are present for $`g_s/|g_v|<1`$ (even if $`E<0`$) and that a new branch of solutions corresponding to antiparticles emerges from the lower continuum as $`g_s/|g_v|`$ increases starting from $`g_s/|g_v|=1`$. In that last case as $`g_s/|g_v|\mathrm{}`$ the even and odd parities solutions tend to be degenerate and the spectrum tends to exhibit a symmetry about $`E=0`$.
## 3 Conclusions
Using the same method used in prior works, we have succeed in the proposal of searching the solution for a more general screened Coulomb potential with the KG equation. An opportunity was given by that generalization to analyze some aspects of the KG equation which would not be feasibly only with the special cases already approached in the literature. Thus, the use of the mixing of vector and scalar Lorentz structures for other kinds of potentials may lead to a better understanding of the KG equation and its solutions. Free from doubt, this sort of mixing also deserves to be explored with the Dirac equation.
It is worthwhile to mention that the solutions of the KG equation with a screened Coulomb potential present a continuous transition as the ratio $`g_s/g_v`$ varies. However, a phase transition occurs when $`|g_s/g_v|=1`$. Although the phase transition does not always show its face for the KG eigenenergies (observe carefully the continuity of the KG eigenenergies for the particle energy levels in Fig. 2), it clearly shows it for the KG eigenfunctions (note, for instance, that the behaviour of the KG eigenfunction for both classes of effective potentials differ at the neighborhood of the origin).
Finally, we draw attention to the fact that no matter how strong the potentials are, as far as $`g_s|g_v|`$, the energy levels for particles (antiparticles) never dive into the lower (upper) continuum. Thus there is no room for the production of particle-antiparticle pairs. This all means that Klein´s paradox never comes to the scenario.
Acknowledgments
This work was supported in part by means of funds provided by CNPq and FAPESP. |
warning/0507/nucl-ex0507022.html | ar5iv | text | # Jefferson Lab’s results on the 𝑄²-evolution of moments of spin structure functions
## 1 Moments of Spin Structure Functions
Polarized DIS has provided a testing ground for the study of the strong force. Moments of spin structure functions (SSF), among them the Bjorken sum, has played an important rôle in this study. The n-th (Cornwall-Norton) moment of SSF is the integral of the $`x^ng_{1,2}(x,Q^2)`$ SSF over $`x`$. Moments are specially useful because sum rules relate them to other quantities. Such sum rules for $`\mathrm{\Gamma }_1`$, the first moment of $`g_1`$, are the Ellis-Jaffe EJSR and the Bjorken sum rules Bjorken , derived at large $`Q^2`$, and the related Gerasimov-Drell-Hearn (GDH) sum rule GDH at $`Q^2=0`$. The first moment of $`g_2`$, $`\mathrm{\Gamma }_2`$, is given by the Burkhardt-Cottingham (BC) sum rule BC . Rules can be also derived for higher moments, e.g., spin polarizability or $`d_2`$ sum rules.
These relations are useful in many ways: checks the theory on which the rule is based (e.g. QCD and the Bjorken sum rule); checks hypotheses used in the sum rule derivation (e.g. the Ellis-Jaffe sum rules); or checks calculations such as chiral perturbation theory ($`\chi pt`$), lattice QCD or Operator Product Expansion (OPE). If a sum rule rests on solid grounds or is well tested, it can be used to extract quantities otherwise hard to measure (e.g. generalized spin polarizabilities). Because $`\mathrm{\Gamma }_{1,2}`$ are calculable at any $`Q^2`$ using either $`\chi pt`$, lattice QCD or OPE, they are particularly suited to study the transition between the hadronic to partonic descriptions of the strong force. Measurements in the transition region (intermediate $`Q^2`$) have recently been made at Jefferson Lab (JLab).
## 2 Measurements at Jefferson Lab
At moderate $`Q^2`$, resonances saturate moments. JLab’s accelerator delivers CW electron beam with a maximum energy up to 6 GeV. This makes JLab the suited place to measure moments up to $`Q^2`$ of a few GeV<sup>2</sup>. The beam current can reach 200 $`\mu `$A with a polarization now reaching 85% although at the time of the experiments reported here, it was typically 70%. The beam is sent simultaneously to three halls (A, B and C), all of them equipped with polarized targets. In this talk, we report on results from halls A and B.
Hall A HallA nim contains a polarized <sup>3</sup>He gaseous target and two high resolution spectrometers (HRS) with 6 mSr acceptance. The target can be polarized longitudinally or transversally at typically 40% polarization with 10-15 $`\mu `$A of beam. The target’s $`10`$ atm. of <sup>3</sup>He gives a luminosity greater than $`10^{36}`$cm<sup>-2</sup>s<sup>-1</sup>. Hall B HallB nim luminosity is typically 5$`\times 10^{33}`$cm<sup>-2</sup>s<sup>-1</sup> but is compensated by the large acceptance (about 2.5$`\pi `$) of the CLAS spectrometer. Cryogenic polarized targets (NH<sub>3</sub> and ND<sub>3</sub>) are well suited for the low beam currents ($``$nA) utilized in Hall B. The target is longitudinally polarized with average 75% (NH<sub>3</sub>) and 40% (ND<sub>3</sub>) polarizations. Both halls can cover the large region of $`Q^2`$ and $`x`$ needed to extract moments at various $`Q^2`$, either because of the large CLAS acceptance (Hall B) or because of large luminosity allowing to quickly gather data at various beam energies and HRS settings (Hall A).
I report here on the Hall A E94010 E94010 and Hall B EG1 experiments. EG1 was split in two runs: EG1a (1998) which results are published eg1a , and EG1b (2000) that is still being analyzed. SSF are extracted differently in halls A and B. In Hall A, *absolute* cross sections asymmetries $`\mathrm{\Delta }\sigma ^{()}`$ were measured for longitudinal (transverse) target spin orientations. $`g_1`$ and $`g_2`$ are linear combinations of these $`\mathrm{\Delta }\sigma `$ and are extracted without external input. Furthermore, unpolarized contributions, e.g. target cell windows or the (mostly) unpolarized protons in the <sup>3</sup>He nucleus, cancel out. The *relative* longitudinal asymmetry $`A_{}`$ is measured in Hall B. Models for $`F_1`$, $`g_2`$ and $`R=\sigma _L/\sigma _T`$ are then used to extract $`g_1`$. $`F_1`$ and $`R`$ are constrained at low $`Q^2`$ by recent Hall C data E94110 . $`g_2`$ is estimated using models (resonance region) or its leading twist part $`g_2^{ww}`$ (DIS domain). The unmeasured low-$`x`$ part of the moment is estimated using a parametrization developed by the EG1 collaboration, while the E94010 group used a Regge-type fit of DIS data Bianchi .
Results on $`\mathrm{\Gamma }_1^p`$, $`\mathrm{\Gamma }_1^n`$ and $`\mathrm{\Gamma }_1^d`$ are shown in Fig. 1, together with $`\chi pt`$ calculations meissner chipt ; ji chipt models Burkert and Ioffe ; Soffer and leading twist OPE prediction. HERMES HERMES and SLAC e143 results are also shown. The halls A and B data, reanalyzed at matched $`Q^2`$ points and with a consistent low-$`x`$ estimate Bianchi were used to form the Bjorken sum $`\mathrm{\Gamma }_1^{pn}`$ deur . Preliminary $`\mathrm{\Gamma }_1^{pn}`$ from EG1b is also shown. $`\mathrm{\Gamma }_1^{pn}`$ is a unique quantity to study parton-hadron transition because its non-singlet structure makes it an easier quantity to handle for $`\chi pt`$, lattice QCD and OPE. These data form, for both nucleons, an accurate mapping at intermediate $`Q^2`$ that connects to SLAC, HERMES and CERN DIS data. At low $`Q^2`$, $`\chi pt`$ disagrees with the data above $`Q^2=0.2`$ GeV<sup>2</sup>, while models based on different physics reproduce equally well the data. Twist 2 description also works well down to low $`Q^2`$, indicating an overall suppressed higher twist rôle. Indeed, in OPE analysis results deur ; osipenko ; ZEM , twist 4 and 6 coefficients are either small or canceling each others at $`Q^2`$=1 GeV<sup>2</sup>.
The availability of transverse data in Hall A allows us to form $`\mathrm{\Gamma }_2^n`$ and thereby check the BC sum rule $`(\mathrm{\Gamma }_2=0)`$ on the neutron (fig. 2). The sum rule is based on dispersion relations and is $`Q^2`$-invariant. A striking feature is the almost perfect cancellation between elastic and resonance contributions leading to the verification of the sum rule.
Other sum rules link SSF moments to the generalized spin polarizabilities $`\gamma _0`$ and $`\delta _{LT}`$:
$`\gamma _0={\displaystyle \frac{4e^2M^2}{\pi Q^6}}{\displaystyle _0^1^{}}x^2(g_1{\displaystyle \frac{4M^2}{Q^2}}x^2g_2)𝑑x;\delta _{LT}={\displaystyle \frac{4e^2M^2}{\pi Q^6}}{\displaystyle _0^1^{}}x^2(g_1+g_2)𝑑x`$
Results from Hall A can be seen in fig. 2 E94010-3 . $`\delta _{LT}`$ is interesting because the $`\mathrm{\Delta }_{1232}`$ rôle is suppressed. Hence $`\delta _{LT}`$ is easier to access by $`\chi pt`$. However, calculations and data disagree for both $`\gamma _0`$ and $`\delta _{LT}`$. The MAID model MAID , however, well reproduces the data. Another higher moment that can be formed is $`d_2^n`$, the integral of $`x^2(g_2g_2^{ww})`$ where $`g_2^{ww}`$ is the leading twist part of $`g_2`$. Thus $`d_2`$ is sensitive to twist 3 and higher. The measured $`\overline{d}_2^n`$ (the bar indicates the exclusion of $`x=1`$) trends toward the lattice QCD results, although larger $`Q^2`$ data are necessary to establish a possible agreement.
## 3 Summary and Perspectives
The hadron-parton transition region is covered by data of the SSF moments from JLab. These can be calculated at any $`Q^2`$, thus providing a ground for studying the link between hadronic and partonic descriptions of the strong force. An OPE analysis reveals that in this domain, high twist effects are small. The BC sum rule was shown on the neutron and found to be valid. Data and sum rules were used to extract neutron generalized spin polarizabilities. Those disagree with the present $`\chi pt`$ calculations. Further data from Hall A E01-012 e01012 , Hall B EG1b, and Hall C RSS RSS will be available shortly in the resonance region. New data at very low $`Q^2`$ have been taken on the neutron in Hall A e97110 and will be gathered early 2006 for the proton in Hall B e03006 . The 12 GeV upgrade of JLab will allow us to access both larger-$`x`$ and lower-$`x`$. This will allow for more precise measurements of the moments, in particular by addressing the low$`x`$ issue.
This work is supported by the U.S. Department of Energy (DOE) and the U.S. National Science Foundation. The Southeastern Universities Research Association operates the Thomas Jefferson National Accelerator Facility for the DOE under contract DE-AC05-84ER40150. |
warning/0507/nucl-ex0507026.html | ar5iv | text | # Optical calibration hardware for the Sudbury Neutrino Observatory
## 1 Introduction
The Sudbury Neutrino Observatory (SNO) detector is designed to detect Cherenkov light produced by solar neutrino interactions in heavy water (D<sub>2</sub>O). The accuracy of the solar neutrino measurements depends on a detailed knowledge of the detector operating conditions, of which the optical properties play a dominant role.
The SNO detector distinguishes itself from other large-volume water Cherenkov detectors in its use of 1 kt of D<sub>2</sub>O contained in a 12 m diameter acrylic vessel (AV) . The AV is suspended in H<sub>2</sub>O which provides shielding from natural radioactivity in the detector components and the surrounding rock cavity. To provide shielding from cosmic rays, the detector is located 2 km underground in the INCO Creighton mine near Sudbury, Canada. Light emitted in the detector is captured by 9456 photomultiplier tubes (PMTs) arranged on a 17 m diameter geodesic support structure (PSUP). The determination of light attenuation in the heavy and light water and the PMT response as a function of incident angle are part of the optical calibration (OCA) . The determination of the relative timing offsets of the PMT electronic channels are part of the PMT calibration (PCA). These calibrations determine crucial parameters for the precise reconstruction of the energy, position and direction of neutrino candidate events used in the analyses of solar neutrinos , as well as for the searches for nucleon decay via ’invisible’ modes , electron antineutrinos , supernova neutrinos and analyses of atmospheric neutrinos and muons .
The optical calibration hardware described in this paper consists of a light diffusing sphere (“laserball”), an underwater optical fibre umbilical cable and a pulsed dye-laser light source. These components are shown in Figure 1, along with the source manipulator system that is used to manoeuvre different calibration sources, including the laserball, inside the AV. An overview of the SNO detector design, including a description of the manipulator system, is given in Ref. . The overall system requirements are briefly described in the next section. The following three sections contain a detailed description of the design and construction of each of the optical calibration subsystems, followed by a brief summary of the laserball performance in SNO.
## 2 System requirements
The OCA hardware must provide good coverage of the detectable Cherenkov spectrum, which extends from 300 to 700 nm. The pulse intensity must be adjustable for single photoelectron level, at $`5\%`$ of PMTs triggered per laser pulse, equivalent to a ratio of multi- to single-photon hits of $`2.5\%`$, in order to maximize the timing measurement precision.
While the OCA uses data taken at many laserball positions, the PCA is done with the laserball only at the centre of the detector and at 505 nm only. The choice of this wavelength for the PCA is done for several reasons:
* minimum effects from scattering (Rayleigh scattering coefficient is about $`1.8\times 10^5`$ cm<sup>-1</sup> at 505 nm);
* reduced time-delayed fluorescence in the optical fibres (low level below 400 nm, reaching $`13\pm 4\%`$ at 337 nm , below detection above 400 nm);
* high PMT quantum efficiency (85% of the peak value, which is $`21.5\%`$ at 440 nm);
* good transmission in all the materials: acrylic ($`>90\%`$ above 370 nm), H<sub>2</sub>O ($`>90\%`$ between 280 and 540 nm) and D<sub>2</sub>O ($`>90\%`$ above 340 nm).
Good timing of the detected photons is required by both the OCA and the PCA to allow the discrimination between light coming directly from the source and late light which has been scattered or reflected from other detector elements . Since the PMT hit times are used in the event reconstruction, the PCA requires the optical pulse timing width – determined essentially by the laser pulse width and by the dispersion in the fibres and the laserball – to be small relative to the intrinsic 1.5 ns resolution of the PMTs .
## 3 Laserball
The laserball was designed and built to calibrate the optical characteristics of the SNO detector by emitting a quasi-isotropic light distribution in the far-field (i.e., at distances large compared to the laserball’s dimensions) . The final design of the laserball, described in this paper and shown schematically in Figure 2, was achieved after a series of iterations.
The laserball consists of a 10.9 cm diameter quartz flask filled with 0.5 kg of silicone gel , in which 2 g of 50 $`\mu `$m diameter air-filled hollow glass spheres are suspended. In the assembly of the laserball, uniform distribution of the scattering spheres is achieved by continuous agitation of the flask during the 15 minutes cure time of the silicone gel. It was verified that bubbles formed in the gel (so it had to be refilled) if the laserball was brought to the surface from the underground SNO laboratory.
The mounting hardware is made of stainless steel except for the acrylic window (for inspection of the fibre loop during assembly). The quartz flask has a straight neck with a restraining ring, and is completely filled with silicone gel.
The polished tip of the fibre bundle coming from the umbilical cable (see Section 4) is coupled to a single rigid fibre, the bottom tip of which is placed slightly above the centre of the laserball such that the light emitted by the fibres is redirected by refraction and total internal reflection at the glass-air interface inside the hollow glass spheres. This results in a smoothly varying, quasi-isotropic far-field intensity distribution. This distribution is only weakly dependent on the wavelength even for low concentrations of scatterers. The mean free path of light between the hollow glass spheres is $`1`$ cm, which is a compromise between guaranteeing far-field uniformity and reducing time dispersion and light absorption. Monte Carlo simulations of the laserball indicate that the light appears to come from a diffuse region with an effective diameter of 4 cm.
The far-field light distribution of the laserball can be adjusted by moving the fibre tip: a displacement of 1 mm results in a change of approximately 10% in the intensity pattern at 385 nm. Only a coarse adjustment is required, since the PCA depends only on the timing, and for the OCA the laserball light distribution is fit along with the optical parameters using an optical model of the detector (as it will be described in Section 6).
In addition to the small residual anisotropy in the light distribution of the laserball due to the placement of the optical fibres, there is a larger asymmetry due to shadowing by the mounting hardware. The light within approximately 60 of the vertical is progressively shadowed, reaching a reduction of $`50\%`$ directly above the laserball. To minimize reflections from the mounting hardware and to prevent the detection of light emitted from the neck of the flask, a cylindrical polished stainless steel light shield extends to just above the ball surface.
## 4 Umbilical cables
The multi-purpose underwater “umbilical” cables were designed to be flexible and robust for use with the source manipulator system, and to provide services for the operation of all the SNO calibration sources . The optical fibre umbilical contains a bundle of optical fibres for transmitting light from the laser down to the laserball. This section describes the requirements, design and fabrication process of the umbilical cables.
### 4.1 Requirements and design
Design requirements for the umbilicals included:
* Radioactivity cleanliness: suitable for deployment in SNO for periods of up to several weeks at a time without introducing a significant <sup>222</sup>Rn contamination (emanation lower than 10 mBq, which correspond to 10% of the level in the D<sub>2</sub>O);
* Impermeability to water to protect cable components and the source itself;
* High flexibility with a small bending radius, for use with the manipulator;
* Length of 30 m, to reach the farthest available positions inside SNO;
* Neutral buoyancy in D<sub>2</sub>O with constant linear density over its length.
Cables with these specifications were not available commercially, so the expertise to construct them was developed within the SNO Collaboration. A short length of umbilical cable is shown in Figure 3, with its cross-sectional view depicted in Figure 4.
An umbilical consists of a 30.5 m long silicone tube with an outer diameter (OD) of 12.7 mm containing a smaller 6.35 mm OD polyethylene tube helically wrapped with four thin Teflon-insulated hook-up wires and a thin coaxial cable . The four wires and the coaxial cable centre the polyethylene tube inside the umbilical, and the helical wrapping permits longitudinal tensile and compressive forces on the wires to be relieved locally as the umbilical goes over the pulleys of the manipulator system. Both of these effects are important in reducing the minimum bending radius for use with the manipulator system with its 15.2 cm and 20.3 cm diameter pulleys. Additionally, the central polyethylene tube provides mechanical support to the fibres, both radially (against crushing) as well as longitudinally (against stretching).
For the laserball umbilical, 20 optical fibres were used to allow for redundancy in the case of breakage, and to simplify the injection of light from the laser while preserving adequate light intensity at the laserball. The high-OH (hydroxyl) core fibres provide low attenuation and low time dispersion over the wavelength range 337–619 nm probed in the optical calibration.
The optical fibres are put in a small Teflon tube , which protects them during assembly and deployment. The fibres are deliberately kept loose inside the Teflon tube so that they do not take any load and can move relative to the umbilical. Having the fibres on the neutral bending axis of the umbilical also greatly reduces mechanical stresses during deployment.
Variations on the basic umbilical design include replacing the fibre bundle with a gas capillary for two short-lived radioactive gas sources (<sup>16</sup>N and <sup>8</sup>Li ), replacing the 6.35 mm polyethylene tube with a high-voltage cable for a small proton-tritium accelerator source , or a fast coaxial cable for a thoriated anode proportional counter source . Thus one basic umbilical design has been adapted to provide services to all the SNO calibration sources.
### 4.2 Umbilical Fabrication
The umbilicals are made by first wrapping the four small hook-up wires and the thin coaxial wire around the polyethylene tube, and this assembly is then drawn into the silicone tube. The volume between the silicone tube and the polyethylene tube is filled with a clear, transparent, two-component liquid silicone which forms a robust, impermeable layer between the silicone tube and the inner components. The filling of the umbilicals with silicone is a critical step in their manufacturing. At room temperature, the highly viscous ($`4000`$ Pa$``$s) RTV615 silicone remains fluid for four to six hours after its components are mixed. During this time, the viscosity increases steadily until the fluid stops flowing freely. Therefore, a high-pressure injection system was developed to force the RTV615 into the umbilical within the first hour after mixing.
To hold the silicone tubing in place, it is inserted into eight 3.65 m and one 1.3 m sections (total length: 30.5 m) of thin-walled 15.875 mm OD copper piping. The inner diameter of the copper piping is only slightly larger than the OD of the silicone tube. The silicone tube is wrapped back around the ends of the copper pipe to form a seal and is “inflated” by evacuating the volume between the copper pipes and the outside of the silicone tubing. This anchors the silicone tube to the copper pipe by friction and allows the helically-wrapped polyethylene tube to be drawn through the silicone tube without stretching the latter.
The umbilical in its copper pipe housing is connected at one end to an injection piston and at the other to an evacuated drum, which eliminates trapped air between cables or at the front edge of the flow of liquid silicone. At the injection end of the umbilical, a special “T”-fitting is used to ensure the silicone is forced into the annular region between the polyethylene and the silicone tubes. The injection piston was designed to fulfill two goals: degassing the silicone immediately after mixing at partial vacuum pressure ($`3.3`$ kPa), and then injecting the silicone into the umbilical at relatively high pressure ($`2.1`$ MPa). Because of the limited time available for the umbilical filling, the degassing step was kept short ($`20`$ minutes). In consequence, a long horizontal acrylic piston (183 cm $`\times `$ 7.62 cm OD, wall thickness 6.35 mm) was used, exposing a large surface area of fluid to the vacuum and permitting visual monitoring during the degassing and injection steps.
Under the injection pressure, the umbilical is expanded against the constraining copper pipes by the silicone. The remaining time until the silicone stops flowing freely allows this pressure to be relieved (the silicone is permitted to escape from both ends). The finished umbilical can be easily removed from the copper pipes once the RTV615 is fully cured two days later.
## 5 Laser system
The laser system consists of a short pulse-length nitrogen laser ($`\lambda =337.1`$ nm) which can be made to pump a series of up to five laser dyes at longer wavelengths ($`\lambda `$ = 369, 385, 420, 505 and 619 nm). Figure 5 shows the layout of the laser system, consisting of the following components:
1. Commercial N<sub>2</sub> TEA (Transversely Excited Atmosphere) thyratron triggered ultraviolet pump laser (Class IIIb ).
The laser head is a channel 10 cm long with a 4 mm electrode gap and a feedback mirror at one end of a high voltage (15 kV) parallel plate capacitor. The output is super-radiant and therefore lacks the coherence of a cavity mode laser. The laser is enclosed in a copper radio-frequency (RF) shield, which attenuates the high levels of RF radiation emitted during the rapid high voltage discharge across the gas cell. This is critical because the RF noise is synchronous with the laser light.
2. Optical table where four dye laser resonator units and associated beam optics are mounted. The dye laser cell is selected by moving a mirror mounted on a computer controlled lead screw carriage. The fifth dye cell is swapped manually.
The nitrogen laser beam is focused through a cylindrical lens to a line just inside the dye cell cuvette wall. The cavity mirror and feedback coupler generate a well defined beam along the focused edge of the cuvette (mounted at an angle so that feedback from internal reflections are not amplified). The resulting beam is approximately 2 mm in diameter with a half-angle divergence of 3 mrad and rectangular diffraction fringes. The cuvette holders incorporate a motor that magnetically drives a stirring agitator in the bottom of the cuvette.
The beam optics are set so that the beams from the four dye laser units converge on a single axis using four semi-reflective mirrors. By putting the longest wavelength in the fourth dye cell and the shortest in the first, these steering mirrors partially compensate for the higher attenuation of shorter wavelengths in the optical fibres.
3. Two computer controlled attenuator wheels with 8 positions each: open, beam stop, 6 coarse adjustment neutral density (ND) filters in the first wheel, and 6 finely spaced ND filters in the second wheel. The combined neutral density is adjusted to produce pulses with an intensity equivalent on average to single photoelectron illumination of all the tubes (as mentioned before, this allows about 5% of all the PMTs to be illuminated by each pulse).
4. An intensity homogenizer (1 mm diameter fibre, 1 m long) to remove pulse-to-pulse beam pattern instabilities.
5. Remote and local system power control and trigger control of pump laser and trigger lockout.
6. Pump beam and dye beam laser energy monitors, amplifiers, and a fast beam activated event trigger. The latter is generated by a fast reverse biased MRD500 PIN diode, that produces a negative pulse of 0.8 ns full width. This provides superior timing compared to command triggering of the laser, which would include about 5 ns timing jitter due to the logic circuits and the thyratron switch.
7. Heavy aluminum box consisting of a $`1.27\times 61\times 186`$ cm base plate, an angle frame and 1.5 mm thick sheet aluminum covers on all sides. This box provides mechanical stability for the optical components and prevents dust from getting in, and RF or laser radiation from getting out.
The dye wavelengths used are listed with other key parameters of the laser system in Table 1. Figure 6 shows the stimulated emission spectra of each of the dyes. The spectra were measured by using the dyes in a commercial N<sub>2</sub>/dye laser unit coupled to a spectrophotometer that has a 2 nm resolution. The measured peak positions and widths are compatible with the manufacturer’s nominal values. Differences in peak wavelength of about 1% or lower are observed when comparing the spectra of dyes used for 3 years in SNO with those of newly mixed dyes. Such differences can cause a negligible systematic effect of about 0.1% in the refractive index, so the impact of dye degradation in the analysis of SNO data is not significant.
Table 2 shows the relevant contributions to timing dispersion in the laser system coupled through optical fibres (total length 45 m) to the laserball. The fibre dispersion measurements were made by using a setup in which the fibre is looped over a large drum and a small PMT detects photons that scatter out of the fibre for each pass of a light pulse produced by a fast N<sub>2</sub> pulser lamp. The PMT timing spectrum contains multiple peaks offset by the loop transit time and the width of those gives the dispersion as a function of the distance.
All the source manipulation operations can be done remotely, except for the actual source insertion (and removal) within the AV. A fifth resonator unit was recently added to the optical table, so all wavelengths can be selected with no manual intervention. Therefore, with the exception of the occasional replacement of the laser N<sub>2</sub> supply, the optical calibration system can be operated remotely, which makes it easier to conduct around-the-clock, multiple-day calibrations since access to the SNO underground laboratory is limited.
## 6 Performance of the optical source
The laserball (LB) distribution is implemented in the OCA model as a combination of a a 2-dimensional $`12\times 36`$ matrix of the polar angle cosine $`\mathrm{cos}\theta _{\mathrm{LB}}`$ versus the azimuthal angle $`\varphi _{\mathrm{LB}}`$ and a sixth-order polynomial function on $`\mathrm{cos}\theta _{\mathrm{LB}}`$ only. The polynomial function allows the model to accomodate a rapidly varying distribution without increasing the size of the matrix.
Figure 7 shows the far-field laserball intensity distribution, determined by the OCA for four different calibration scans at 505 nm, as a function of the polar angle cosine. A strong variation of the polar light distribution is expected at $`\mathrm{cos}\theta _{\mathrm{LB}}`$ close to 1 due to shadowing by the source support hardware. Away from the top region, the polar variation of the light intensity can be up to $`\pm 40`$% and is likely determined at assembly by the exact configuration of gaps between the fibre bundle and the rigid fibre (since this can influence the angular distribution of light arriving to the gel). The laserball used in the September 2000 scan is different from the one used in the later scans. The variations between the September 2001, May 2002 and August 2003 are likely due to variations in the laserball-umbilical coupling and degradation of the gel.
Figure 8 shows the laserball intensity distribution determined by the OCA for each wavelength , after taking into account the polar angle dependence shown in Figure 7. The grey scale indicates the relative light intensity as a function of the cosine of the polar angle, $`\mathrm{cos}\theta _{\mathrm{LB}}`$, and azimuthal angle, $`\varphi _{\mathrm{LB}}`$. The light intensity varies by $`\pm 15`$% at 619 nm, increasing to $`\pm 25`$% at 337.1 nm The azimuthal light distribution is dominated by a dipole distribution, caused by misalignment of the fibre tip inside the laserball. The larger asymmetry at shorter wavelengths is likely the effect of increased absorption in the laserball gel and glass spheres.
The OCA must determine average optical properties of the whole detector. So it is important for the variation in azimuthal distribution to be smooth and relatively small, as shown in Figure 8, to guarantee a good sampling of all the PMTs. Fluctuations in the laserball intensity that occur in a short timescale (less than a week) can affect the accuracy of the OCA during a multi-day scan. Hence, runs are repeated at specific positions throughout the scan to ensure the stability of the light intensity.
## 7 Conclusions
An accurate calibration of the optical properties of SNO is a vital step in understanding the performance of the detector. Measurements with the laserball system at six wavelengths from 337.1 nm to 619 nm allow the optical properties to be sampled over the detectable Cherenkov light spectrum. These optical data for the OCA are taken approximately twice per year, each time for one week around the clock. The system is fully remotely controlled so that continuous underground access to the SNO laboratory is not necessary.
The short pulse length and good uniformity characteristics of the light distribution permit the simultaneous relative timing and collected charge calibrations(PCA) of all 9456 PMT channels to be accomplished efficiently. The timing calibration is performed at 505 nm because the light intensity varies by only $`\pm `$15%, and absorption and scattering are low in all three media (D<sub>2</sub>O, H<sub>2</sub>O and acrylic). These calibrations are done monthly.
The laserball system has proven very robust over more than 2500 hours of operation. Limited maintenance has been required, such as new charge transfer boards for the laser head and realignment of the mirrors. The umbilical cables, made using a novel manufacturing technique, are flexible, clean, robust and versatile: the basic umbilical design was easily modified to provide gas, electrical connections or simply act as a support line for each of the eight other SNO calibration devices.
The optical calibration hardware enables the detector optics to be probed *in situ* and provides the essential timing reference for all PMT channels. These calibrations are crucial to the energy and position reconstruction of events, and are essential to the precision of SNO’s solar neutrino data analysis.
## Acknowledgements
This research was supported in part by the Natural Sciences and Engineering Research Council of Canada, and the Fonds pour les Chercheurs et l’Aide à la Recherche of the province of Québec, Canada. |
warning/0507/physics0507194.html | ar5iv | text | # Contents
## 1 Introduction
### 1.1 Why classical mechanics?
All hail the rise of modern physics! Between 1890 and 1930, the quantum and relativity revolutions and the consolidation of statistical physics through the discovery of atoms, utterly transformed our understanding of nature; and had an enormous influence on philosophy; (e.g. Kragh 1999; Ryckman 2005). Accordingly, this Handbook concentrates on those three pillars of modern physics—quantum theories, spacetime theories and thermal physics. So some initial explanation of the inclusion of a Chapter on classical mechanics, indeed the classical mechanics of finite-dimensional systems, is in order.
The first point to make is that the various fields of classical physics, such as mechanics and optics, are wonderfully rich and deep, not only in their technicalities, but also in their implications for the philosophy and foundations of physics. From Newton’s time onwards, classical mechanics and optics have engendered an enormous amount of philosophical reflection. As regards mechanics, the central philosophical topics are usually taken (and have traditionally been taken) to be space, time, determinism and the action-at-a-distance nature of Newtonian gravity. Despite their importance, I will not discuss these topics; but some other Chapters will do so (at least in part, and sometimes in connection with theories other than classical mechanics). I will instead focus on the theory of symplectic reduction, which develops the well-known connection between continuous symmetries and conserved quantities, summed up in Noether’s “first theorem”. I choose this focus partly by way of preparation for parts of some other Chapters; and partly because, as we will see in a moment, symplectic reduction plays a central role in the current renaissance of classical mechanics, and in its relation to quantum physics.
I said that classical physics engendered a lot of philosophical reflection. It is worth stressing two, mutually related, reasons for this: reasons which today’s philosophical emphasis on the quantum and relativity revolutions tends to make us forget.
First: in the two centuries following Newton, these fields of classical physics were transformed out of all recognition, so that the framework for philosophical reflection about them also changed. Think of how in the nineteenth century, classical mechanics and optics gave rise to classical field theories, especially electromagnetism. And within this Chapter’s specific field, the classical mechanics of finite-dimensional systems, think of how even its central theoretical principles were successively recast, in fundamental ways, by figures such Euler, Lagrange, Hamilton and Jacobi.
Second, various difficult problems beset the attempt to rigorously formulate classical mechanics and optics; some of which have considerable philosophical aspects. It is not true that once we set aside the familiar war-horse topics—space, time, determinism and action-at-a-distance—the world-picture of classical mechanics is straightforward: just “matter in motion”. On the contrary. Even if we consider only finite-dimensional systems, we can ask, for example:
(i) For point-particles (material points): can they have different masses, and if so how? What happens when they collide? Indeed, for point-particles interacting only by Newtonian gravity, a collision involves infinite kinetic energy.
(ii) For extended bodies treated as finite-dimensional because rigid: what happens when they collide? Rigidity implies that forces, and displacements, are transmitted “infinitely fast” through the body. Surely that should not be taken literally? But if so, what justifies this idealization; and what are its scope and limits?
As to infinite-dimensional systems (elastic solids, fluids and fields), many parts of their theories remain active research areas, especially as regards rigorous formulations and results. For contemporary work on elastic solids, for example, cf. Marsden and Hughes (1982). As to fluids, the existence and uniqueness of rigorous solutions of the main governing equations, the Navier-Stokes equations, is still an open problem. This problem not only has an obvious bearing on determinism; it is regarded as scientifically significant enough that its solution would secure a million-dollar Clay Millennium prize.
These two reasons—the successive reformulations of classical mechanics, and its philosophical problems—are of course related. The monumental figures of classical mechanics recognized and debated the problems, and much of their technical work was aimed at solving them. As a result, there was a rich debate about the foundations of classical physics, in particular mechanics, for the two centuries after Newton’s Principia (1687). A well-known example is Duhem’s instrumentalist philosophy of science, which arose in large measure from his realization how hard it was to secure rigorous foundations at the microscopic level for classical mechanics. A similar example is Hilbert’s being prompted by his contemporaries’ continuing controversies about the foundations of mechanics, to choose as the sixth of his famous list of outstanding mathematical problems, the axiomatization of mechanics and probability; (but for some history of this list, cf. Grattan-Guinness (2000)). A third example, spanning both centuries, concerns variational principles: the various principles of least action formulated first by Maupertuis, then by Euler and later figures—first for finite classical mechanical systems, then for infinite ones—prompted much discussion of teleology. Indeed, this discussion ensnared the logical empiricists (Stöltzner 2003); it also bears on contemporary philosophy of modality (Butterfield 2004).
In the first half of the twentieth century, the quantum and relativity revolutions tended to distract physicists, and thereby philosophers, from these and similar problems. The excitement of developing the new theories, and of debating their implications for natural philosophy, made it understandable, even inevitable, that the foundational problems of classical mechanics were ignored.
Besides, this tendency was strengthened by the demands of pedagogy: the necessity of including the new theories in physics undergraduate degrees. By mid-century, the constraints of time on the physics curriculum had led many physics undergraduates’ education in classical mechanics to finish with the elementary parts of analytical mechanics, especially of finite-dimensional systems: for example, with the material in Goldstein’s well-known textbook (1950). Such a restriction is understandable, not least because: (i) the elementary theory of Lagrange’s and Hamilton’s equations requires knowledge of ordinary differential equations, and (ii) elementary Hamiltonian mechanics forms a springboard to learning elementary canonical quantization (as does Hamilton-Jacobi theory, from another perspective). Besides, as I mentioned: even this restricted body of theory provides plenty of material for philosophical analysis—witness my examples above, and the discussions of the great figures such Euler, Lagrange, Hamilton and Jacobi.
However, the second half of the twentieth century saw a renaissance in research in classical mechanics: hence my first motto. There are four obvious reasons for this: the first two “academic”, and the second two “practical”.
(i): Thanks partly to developments in mathematics in the decades after Hilbert’s list of problems, the foundational questions were addressed afresh, as much by mathematicians and mathematically-minded engineers as by physicists. The most relevant developments lay in such fields as topology, differential geometry, measure theory and functional analysis. In this revival, the contributions of the Soviet school, always strong in mechanics and probability, were second to none. And relatedly:—
(ii): The quest to deepen the formulation of quantum theory, especially quantum field theory, prompted investigation of (a) the structure of classical mechanics and (b) quantization. For both (a) and (b), special interest attaches to the generally much harder case of infinite systems.
(iii): The coming of spaceflight, which spurred the development of celestial mechanics. And relatedly:—
(iv): The study of non-linear dynamics (“chaos theory”), which was spurred by the invention of computers.
With these diverse causes and aspects, this renaissance continues to flourish—and accordingly, I shall duck out of trying to further adumbrate it! I shall even duck out of trying to survey the philosophical questions that arise from the various formulations of mechanics from Newton to Jacobi and Poincaré. Suffice it to say here that to the various topics mentioned above, one could add, for example, the following two: the first broadly ontological, the second broadly epistemological.
(a): The analysis of notions such as mass and force (including how they change over time). For this topic, older books include Jammer (1957, 1961) and McMullin (1978); recent books include Boudri (2002), Jammer (2000), Lutzen (2005) and Slovik (2002); and Grattan-Guinness (2006) is a fine recent synopsis of the history, with many references.
(b): The analysis of what it is to have an explicit solution of a mechanical problem (including how the notion of explicit solution was gradually generalized). This topic is multi-faceted. It not only relates to the gradual generalization of the notion of function (a grand theme in the history of mathematics—well surveyed by Lutzen 2003), and to modern non-linear dynamics (cf. (iv) above). It also relates to the simplification of problems by exploiting a symmetry so as to reduce the number of variables one needs—and this is the core idea of symplectic reduction. I turn to introducing it.
### 1.2 Prospectus
The strategy of simplifying a mechanical problem by exploiting a symmetry so as to reduce the number of variables is one of classical mechanics’ grand themes. It is theoretically deep, practically important, and recurrent in the history of the subject. The best-known general theorem about the strategy is undoubtedly Noether’s theorem, which describes a correspondence between continuous symmetries and conserved quantities. There is both a Lagrangian and a Hamiltonian version of this theorem, though for historical reasons the name ‘Noether’s theorem’ is more strongly attached to the Lagrangian version. However, we shall only need the Hamiltonian version of the theorem: it will be the “springboard” for our exposition of symplectic reduction.<sup>2</sup><sup>2</sup>2For discussion of the Lagrangian version, cf. e.g. Brading and Castellani (this vol., ch. 13) or (restricted to finite-dimensional systems) Butterfield (2004a: Section 4.7). For an exposition of both versions that is complementary to this paper (and restricted to finite-dimensional systems), cf. Butterfield (2006). Brading and Castellani also bring out that, even apart from Noether’s theorems in other branches of mathematics, there are other ‘Noether’s theorems’ about symmetries in classical dynamics; so the present theorem is sometimes called Noether’s “first theorem”. Note also (though I shall not develop this point) that symplectic structure can be seen in the classical solution space of Lagrange’s equations, so that symplectic reduction can be developed in the Lagrangian framework; cf. e.g. Marsden and Ratiu (1999: p. 10, Sections 7.2-7.5, and 13.5).
So I shall begin by briefly reviewing the Hamiltonian version in Section 2.1. For the moment, suffice it to make four comments (in ascending order of importance for what follows):
(i): Both versions are underpinned by the theorems in elementary Lagrangian and Hamiltonian mechanics about cyclic (ignorable) coordinates and their corresponding conserved momenta.<sup>3</sup><sup>3</sup>3Here we glimpse the long history of our subject: these theorems were of course clear to these subjects’ founders. Indeed the strategy of exploiting a symmetry to reduce the number of variables occurs already in 1687, in Newton’s solution of the Kepler problem; (or more generally, the problem of two bodies exerting equal and opposite forces along the line between them). The symmetries are translations and rotations, and the corresponding conserved quantities are the linear and angular momenta. In what follows, these symmetries and quantities will provide us with several examples.
(ii): In fact, the Hamiltonian version of the theorem is stronger. This reflects the fact that the canonical transformations form a “larger” group than the point transformations. A bit more precisely: though the point transformations $`qq^{}`$ on the configuration space $`Q`$ induce canonical transformations on the phase space $`\mathrm{\Gamma }`$ of the $`q`$s and $`p`$s, $`qq^{},pp^{}`$ , there are yet other canonical transformations which “mix” the $`q`$s and $`p`$s in ways that transformations induced by point transformations do not.
(iii): I shall limit our discussion to (a) time-independent Hamiltonians and (b) time-independent transformations. Agreed, analytical mechanics can be developed, in both Lagrangian and Hamiltonian frameworks, while allowing time-dependent dynamics and transformations. For example, in the Lagrangian framework, allowing velocity-dependent potentials and-or time-dependent constraints would prompt one to use what is often called the ‘extended configuration space’ $`Q\times \mathrm{I}\mathrm{R}`$. And in the Hamiltonian framework, time-dependence prompts one to use an ‘extended phase space’ $`\mathrm{\Gamma }\times \mathrm{I}\mathrm{R}`$. Besides, from a philosophical viewpoint, it is important to consider time-dependent transformations: for they include boosts, which are central to the philosophical discussion of spacetime symmetry groups, and especially of relativity principles. But beware: rough-and-ready statements about symmetry, e.g. that the Hamiltonian must be invariant under a symmetry transformation, are liable to stumble on these transformations. To give the simplest example: the Hamiltonian of a free particle is just its kinetic energy, which can be made zero by transforming to the particle’s rest frame; i.e. it is not invariant under boosts.
So a full treatment of symmetry in Hamiltonian mechanics, and thereby of symplectic reduction, needs to treat time-dependent transformations—and to beware! But I will set aside all these complications. Here it must suffice to assert, without any details, that the modern theory of symplectic reduction does cope with boosts; and more generally, with time-dependent dynamics and transformations.
(iv): As we shall see in detail, there are three main ways in which the theory of symplectic reduction generalizes Noether’s theorem. As one might expect, these three ways are intimately related to one another.
(a): Noether’s theorem is “one-dimensional” in the sense that for each symmetry (a vector field of a special kind on the phase space), it provides a conserved quantity, i.e. a real-valued function on the phase space, whose value stays constant over time. So in particular, different components of a conserved vector quantity, such as total linear momentum, are treated separately; (in this example, the corresponding vector fields generate translations in three different spatial directions). But in symplectic reduction, the notion of a momentum map provides a “unified” description of these different components.
(b): Given a symmetry, Noether’s theorem enables us to confine our attention to the level surface of the conserved quantity, i.e. the sub-manifold of phase space on which the quantity takes its initial value: for the system’s time-evolution is confined to that surface. In that sense, the number of variables we need to consider is reduced. But in symplectic reduction, we go further and form a quotient space from the phase space. That is, in the jargon of logic: we define on phase space an equivalence relation (not in general so simple as having a common value for a conserved quantity) and form the set of equivalence classes. In the jargon of group actions: we form the set of orbits. Passage to this quotient space can have various good technical, and even philosophical, motivations. And under good conditions, this set is itself a manifold with lower dimension.
(c): Hamiltonian mechanics, and so Noether’s theorem, is usually formulated in terms of symplectic manifolds, in particular the cotangent bundle $`T^{}Q`$ of the configuration space $`Q`$. (Section 2.1 will give details.) But in symplectic reduction, we often need a (mild) generalization of the idea of a symplectic manifold, called a Poisson manifold, in which a bracket, with some of the properties of the Poisson bracket, is taken as the primitive notion. Besides, this is related to (b) in that we are often led to a Poisson manifold, and dynamics on it, by taking the quotient of a symplectic manifold (i.e. a phase space of the usual kind) by the action of a symmetry group.
As comment (iv) hints, symplectic reduction is a large subject. So there are several motivations for expounding it. As regards physics, many of the ideas and results can be developed for finite-dimensional classical systems (to which I will confine myself), but then generalized to infinite-dimensional systems. And in either the original or the generalized form, they underpin developments in quantum theories. So these ideas and results have formed part of the contemporary renaissance in classical mechanics; cf. (i) and (ii) at the end of Section 1.1.
As regards philosophy, symmetry is both a long-established focus for philosophical discussion, and a currently active one: cf. Brading and Castellani (2003). But philosophical discussion of symplectic reduction seems to have begun only recently, especially in some papers of Belot and Earman. This delay is presumably because the technical material is more sophisticated: indeed, the theory of symplectic reduction was cast in its current general form only in the 1970s. But as Belot and Earman emphasise, the philosophical benefits are worth the price of learning the technicalities. The most obvious issue is that symplectic reduction’s device of quotienting a state space casts light on philosophical issues about whether two apparently distinct but utterly indiscernible possibilities should be ruled to be one and the same. In Section 2, I will follow Belot in illustrating this issue with “relationist” mechanics. Indeed, I have selected the topics for my exposition with an eye to giving philosophical readers the background needed for some of Belot’s discussions. His papers (which I will cite in Section 2) make many judicious philosophical points, without burdening the reader with an exposition of technicalities: excellent stuff—but to fully appreciate the issues, one of course has to slog through the details.
Finally, in the context of this volume, symplectic reduction provides some background for the Chapters on the representation of time in mechanics (Belot, this vol., ch. 2), and on the relations between classical and quantum physics (Landsman, this vol., ch. 5, especially Sections 4.3-4.5 and 6.5; Dickson, this vol., ch. 4).
The plan of the Chapter is as follows. I first review Noether’s theorem in Hamiltonian mechanics as usually formulated, in Section 2.1. Then I introduce the themes mentioned in (b) and (c) above, of quotienting a phase space, and Poisson manifolds (Section 2.2); and illustrate these themes with “relationist” mechanics (Section 2.3).
Thereafter, I expound the basics of symplectic reduction: (confining myself to finite-dimensional Hamiltonian mechanics). Section by Section, the plan will be as follows. Sections 3 and 4 review the modern geometry that will be needed. Section 3 is mostly about Frobenius’ theorem, Lie algebras and Lie groups.<sup>4</sup><sup>4</sup>4Its first two Subsections also provide some pre-requisites for Malament (this vol.). Section 4 expounds Lie group actions. It ends with the central idea of the co-adjoint representation of a Lie group $`G`$ on the dual $`𝔤^{}`$ of its Lie algebra. This review enables us to better understand the motivations for Poisson manifolds (5.1); and then to exhibit examples, and prove some main properties (Section 5.2 onwards). Section 6 applies this material to symmetry and conservation in mechanical systems. In particular, it expresses conserved quantities as momentum maps, and proves Noether’s theorem for Hamiltonian mechanics on Poisson manifolds. Finally, in Section 7, we prove one of the several main theorems about symplectic reduction. It concerns the case where the natural configuration space for a system is itself a Lie group $`G`$: this occurs both for the rigid body and ideal fluids. In this case, quotienting the natural phase space (the cotangent bundle on $`G`$) gives a Poisson manifold that “is” the dual $`𝔤^{}`$ of $`G`$’s Lie algebra.<sup>5</sup><sup>5</sup>5In this endeavour, my sources are four books by masters of the subject: Abraham and Marsden (1978), Arnold (1989), Marsden and Ratiu (1999) and Olver (2000). But again, be warned: my selection is severe, as anyone acquainted with these or similar books will recognize.
To sum up:— The overall effect of this exposition is, I hope, to illustrate this Chapter’s mottoes: that classical mechanics is alive and kicking, not least through deepening our understanding of time-honoured systems such as the rigid body—whose analysis in traditional textbooks can be all too confusing!
## 2 Symplectic reduction: an overview
We begin by briefly reviewing Hamiltonian mechanics and Noether’s theorem, in Section 2.1.<sup>6</sup><sup>6</sup>6For more details about differential geometry, cf. Sections 3.1 and 3.2. For more details about the geometric formulation of mechanics, cf. Arnold (1989) or Marsden and Ratiu (1999); or Singer (2001) (more elementary than this exposition) or Abraham and Marsden (1978) (more advanced); or Butterfield (2006) (at the same level). Of many good textbooks of mechanics, I admire especially Desloge (1982) and Johns (2005). This prepares us for the idea of symplectic reduction, Section 2.2: which we then illustrate using “relationist” mechanics, Section 2.3.
### 2.1 Hamiltonian mechanics and Noether’s theorem: a review
#### 2.1.1 Symplectic manifolds; the cotangent bundle as a symplectic manifold
A symplectic structure or symplectic form on a manifold $`M`$ is defined to be a differential 2-form $`\omega `$ on $`M`$ that is closed (i.e. its exterior derivative $`𝐝\omega `$ vanishes) and is non-degenerate. That is: for any $`xM`$, and any two tangent vectors at $`x`$, $`\sigma ,\tau T_x`$:
$$𝐝\omega =0\text{ and }\tau 0,\sigma :\omega (\tau ,\sigma )0.$$
(2.1)
Such a pair $`(M,\omega )`$ is called a symplectic manifold. There is a rich theory of symplectic manifolds; but we shall only need a small fragment of it. (In particular, the fact that we mostly avoid the theory of canonical transformations means we will not need the theory of Lagrangian sub-manifolds.)
First, it follows from the non-degeneracy of $`\omega `$ that $`M`$ is even-dimensional. The reason lies in a theorem of linear algebra, which one then applies to the tangent space at each point of $`M`$. Namely, for any bilinear form $`\omega :V\times V\mathrm{I}\mathrm{R}`$: if $`\omega `$ is antisymmetric of rank $`rm\mathrm{dim}(V)`$, then $`r`$ is even. That is: $`r=2n`$ for some integer $`n`$, and there is a basis $`e_1,\mathrm{},e_i,\mathrm{},e_m`$ of $`V`$ for which $`\omega `$ has a simple expansion as wedge-products
$$\omega =\mathrm{\Sigma }_{i=1}^ne^ie^{i+n};$$
(2.2)
equivalently, $`\omega `$ has the $`m\times m`$ matrix
$$\omega =\left(\begin{array}{ccc}\mathrm{𝟎}& \mathrm{𝟏}& \mathrm{𝟎}\\ \mathrm{𝟏}& \mathrm{𝟎}& \mathrm{𝟎}\\ \mathrm{𝟎}& \mathrm{𝟎}& \mathrm{𝟎}\end{array}\right).$$
(2.3)
where $`\mathrm{𝟏}`$ is the $`n\times n`$ identity matrix, and similarly for the zero matrices of various sizes. This normal form of antisymmetric bilinear forms is an analogue of the Gram-Schmidt theorem that an inner product space has an orthonormal basis, and is proved by an analogous argument.
So if an antisymmetric bilinear form is non-degenerate, then $`r2n=m`$. That is: eq. 2.3 loses its bottom row and right column consisting of zero matrices, and reduces to the $`2n\times 2n`$ symplectic matrix $`\omega `$ given by
$$\omega :=\left(\begin{array}{cc}\mathrm{𝟎}& \mathrm{𝟏}\\ \mathrm{𝟏}& \mathrm{𝟎}\end{array}\right).$$
(2.4)
Second, the non-degeneracy of $`\omega `$ implies that at any $`xM`$, there is a basis-independent isomorphism $`\omega ^{\mathrm{}}`$ from the tangent space $`T_x`$ to its dual $`T_x^{}`$. Namely: for any $`xM`$ and $`\tau T_x`$, the value of the 1-form $`\omega ^{\mathrm{}}(\tau )T_x^{}`$ is defined by
$$\omega ^{\mathrm{}}(\tau )(\sigma ):=\omega (\sigma ,\tau )\sigma T_x.$$
(2.5)
This also means that a symplectic structure enables a covector field, i.e. a differential one-form, to determine a vector field. Thus for any function $`H:M\mathrm{I}\mathrm{R}`$, so that $`dH`$ is a differential 1-form on $`M`$, the inverse of $`\omega ^{\mathrm{}}`$ (which we might write as $`\omega ^{\mathrm{}}`$), carries $`dH`$ to a vector field on $`M`$, written $`X_H`$. This is the key idea whereby in Hamiltonian mechanics, a scalar function $`H`$ determines a dynamics; cf. Section 2.1.2.
So far, we have noted some implications of $`\omega `$ being non-degenerate. The other part of the definition of a symplectic form (for a manifold), viz. $`\omega `$ being closed, $`𝐝\omega =0`$, is also important. We shall see in Section 2.1.3 that it implies that a vector field $`X`$ on a symplectic manifold $`M`$ preserves the symplectic form $`\omega `$ (i.e. in more physical jargon: generates (a one-parameter family of) canonical transformations) iff $`X`$ is Hamiltonian in the sense that there is a scalar function $`f`$ such that $`X=X_f\omega ^{\mathrm{}}(df)`$. Or in terms of the Poisson bracket, with $``$ representing the argument place for a scalar function: $`X()=X_f()\{,f\}`$.
So much by way of introducing symplectic manifolds. I turn to showing that any cotangent bundle $`T^{}Q`$ is such a manifold. That is: it has, independently of a choice of coordinates or bases, a symplectic structure.
Given a manifold $`Q`$ (dim($`Q`$)=$`n`$) which we think of as the system’s configuration space, choose any local coordinate system $`q`$ on $`Q`$ , and the natural local coordinates $`q,p`$ thereby induced on $`T^{}Q`$. We define the 2-form
$$dpdq:=dp_idq^i:=\mathrm{\Sigma }_{i=1}^ndp_idq^i.$$
(2.6)
In fact, eq. 2.6 defines the same 2-form, whatever choice we make of the chart $`q`$ on $`Q`$. For $`dpdq`$ is the exterior derivative of a 1-form on $`T^{}Q`$ which is defined naturally (i.e. independently of coordinates or bases) from the derivative (also known as: tangent) map of the projection
$$\pi :(q,p)T^{}QqQ.$$
(2.7)
Thus consider a tangent vector $`\tau `$ (not to $`Q`$, but) to the cotangent bundle $`T^{}Q`$ at a point $`\eta =(q,p)T^{}Q,`$ i.e. $`qQ`$ and $`pT_q^{}`$. Let us write this as: $`\tau T_\eta (T^{}Q)T_{(q,p)}(T^{}Q)`$. The derivative map, $`D\pi `$ say, of the natural projection $`\pi `$ applies to $`\tau `$:
$$D\pi :\tau T_{(q,p)}(T^{}Q)(D\pi (\tau ))T_q.$$
(2.8)
Now define a 1-form $`\theta _H`$ on $`T^{}Q`$ by
$$\theta _H:\tau T_{(q,p)}(T^{}Q)p(D\pi (\tau ))\mathrm{I}\mathrm{R};$$
(2.9)
where in this definition of $`\theta _H`$, $`p`$ is defined to be the second component of $`\tau `$’s base-point $`(q,p)T^{}Q`$; i.e. $`\tau T_{(q,p)}(T^{}Q)`$ and $`pT_q^{}`$.
This 1-form is called the canonical 1-form on $`T^{}Q`$. One now checks that in any natural local coordinates $`q,p`$, $`\theta _H`$ is given by
$$\theta _H=p_idq^i.$$
(2.10)
Finally, we define a 2-form by taking the exterior derivative of $`\theta _H`$:
$$𝐝(\theta _H):=𝐝(p_idq^i)dp_idq^i.$$
(2.11)
One checks that this 2-form is closed (since $`𝐝^2=0`$) and non-degenerate. So $`(T^{}Q,𝐝(\theta _H))`$ is a symplectic manifold. Accordingly, $`𝐝(\theta _H)`$, or its negative $`𝐝(\theta _H)`$, is called the canonical symplectic form, or canonical 2-form.
There is a theorem (Darboux’s theorem) to the effect that locally, any symplectic manifold “looks like” a cotangent bundle: or in other words, a cotangent bundle is locally a “universal” example of symplectic structure. We will not go into details; but in Section 5.3.4, we will discuss the generalization of this theorem for Poisson manifolds. But first we review, in the next two Subsections, Hamilton’s equations, and Noether’s theorem.
#### 2.1.2 Geometric formulations of Hamilton’s equations
As we already emphasised, the main geometric idea behind Hamilton’s equations is that a gradient, i.e. covector, field $`dH`$ determines a vector field $`X_H`$. So to give a geometric formulation of Hamilton’s equations at a point $`x=(q,p)`$ in a cotangent bundle $`T^{}Q`$, let us write $`\omega ^{\mathrm{}}`$ for the (basis-independent) isomorphism from the cotangent space to the tangent space, $`T_x^{}T_x`$, induced by $`\omega :=𝐝(\theta _H)=dq^idp_i`$ (cf. eq. 2.5). Then Hamilton’s equations may be written as:
$$\dot{x}=X_H(x)=\omega ^{\mathrm{}}(𝐝H(x))=\omega ^{\mathrm{}}(dH(x)).$$
(2.12)
There are various other formulations. Applying $`\omega ^{\mathrm{}}`$, the inverse isomorphism $`T_xT_x^{}`$, to both sides, we get
$$\omega ^{\mathrm{}}X_H(x)=dH(x).$$
(2.13)
In terms of the symplectic form $`\omega `$ at $`x`$, this is: for all vectors $`\tau T_x`$
$$\omega (X_H(x),\tau )=dH(x)\tau ;$$
(2.14)
or in terms of the contraction (also known as: interior product) $`𝐢_X\alpha `$ of a differential form $`\alpha `$ with a vector field $`X`$, with $``$ marking the argument place of $`\tau T_x`$:
$$𝐢_{X_H}\omega :=\omega (X_H(x),)=dH(x)().$$
(2.15)
More briefly, and now written for any function $`f`$, it is:
$$𝐢_{X_f}\omega =df.$$
(2.16)
Finally, recall the relation between the Poisson bracket and the directional derivative (or the Lie derivative $``$) of a function: viz.
$$_{X_f}g=dg(X_f)=X_f(g)=\{g,f\}.$$
(2.17)
Combining this with eq. 2.16, we can state the relation between the symplectic form and Poisson bracket in the form:
$$\{g,f\}=dg(X_f)=𝐢_{X_f}dg=𝐢_{X_f}(𝐢_{X_g}\omega )=\omega (X_g,X_f).$$
(2.18)
#### 2.1.3 Noether’s theorem
The core idea of Noether’s theorem, in both the Lagrangian and Hamiltonian frameworks, is that to every continuous symmetry of the system there corresponds a conserved quantity (a first integral, a constant of the motion). The idea of a continuous symmetry is made precise along the following lines: a symmetry is a vector field on the state-space that (i) preserves the Lagrangian (respectively, Hamiltonian) and (ii) “respects” the structure of the state-space.
In the Hamiltonian framework, the heart of the proof is a “one-liner” based on the fact that the Poisson bracket is antisymmetric. Thus for any scalar functions $`f`$ and $`H`$ on a symplectic manifold $`(M,\omega )`$ (and so with a Poisson bracket given by eq. 2.18), we have that at any point $`xM`$
$$X_f(H)(x)\{H,f\}(x)=0\text{ iff }\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}0}=\{f,H\}(x)X_H(f)(x).$$
(2.19)
In words: around $`x`$, $`H`$ is constant under the flow of the vector field $`X_f`$ (i.e. under what the evolution would be if $`f`$ was the Hamiltonian) iff $`f`$ is constant under the flow $`X_H`$. Thinking of $`H`$ as the physical Hamiltonian, so that $`X_H`$ represents the real time-evolution (sometimes called: the dynamical flow), this means: around $`x`$, $`X_f`$ preserves the Hamiltonian iff $`f`$ is constant under time-evolution, i.e. $`f`$ is a conserved quantity (a constant of the motion).
But we need to be careful about clause (ii) above: the idea that a vector field respects” the structure of the state-space. In the Hamiltonian framework, this is made precise as preserving the symplectic form. Thus we define a vector field $`X`$ on a symplectic manifold $`(M,\omega )`$ to be symplectic (also known as: canonical) iff the Lie-derivative along $`X`$ of the symplectic form vanishes, i.e. $`_X\omega =0`$. (This definition is equivalent to $`X`$’s generating (active) canonical transformations, and to its preserving the Poisson bracket. But I will not go into details about these equivalences: for they belong to the theory of canonical transformations, which, as mentioned, I will not need to develop.)
We also define a Hamilton system to be a triple $`(M,\omega ,H)`$ where $`(M,\omega )`$ is a symplectic manifold and $`H:M\mathrm{I}\mathrm{R}`$, i.e. $`M(M)`$. And then we define a (continuous) symmetry of a Hamiltonian system to be a vector field $`X`$ on $`M`$ that:
(i) preserves the Hamiltonian function, $`_XH=0`$; and
(ii) preserves the symplectic form, $`_X\omega =0`$.
These definitions mean that to prove Noether’s theorem from eq. 2.19, it will suffice to prove that a vector field $`X`$ is symplectic iff it is locally of the form $`X_f`$. Such a vector field is called locally Hamiltonian. (And a vector field is called Hamiltonian if there is a global scalar $`f:M\mathrm{I}\mathrm{R}`$ such that $`X=X_f`$.) In fact, two results from the theory of differential forms, the Poincaré Lemma and Cartan’s magic formula, make it easy to prove this; (for a vector field on any symplectic manifold $`(M,\omega )`$, i.e. $`(M,\omega )`$ does not need to be a cotangent bundle).
Again writing $`𝐝`$ for the exterior derivative, we recall that a $`k`$-form $`\alpha `$ is called:
(i): exact if there is a $`(k1)`$-form $`\beta `$ such that $`\alpha =𝐝\beta `$; (cf. the elementary definition of an exact differential);
(ii): closed if $`𝐝\alpha =0`$.
The Poincaré Lemma states that every closed form is locally exact. To be precise: for any open set $`U`$ of $`M`$, we define the vector space $`\mathrm{\Omega }^k(U)`$ of $`k`$-form fields on $`U`$. Then the Poincaré Lemma states that if $`\alpha \mathrm{\Omega }^k(M)`$ is closed, then at every $`xM`$ there is a neighbourhood $`U`$ such that $`\alpha _U\mathrm{\Omega }^k(U)`$ is exact.
Cartan’s magic formula is a useful formula (proved by straightforward calculation) relating the Lie derivative, contraction and the exterior derivative. It says that if $`X`$ is a vector field and $`\alpha `$ a $`k`$-form on a manifold $`M`$, then the Lie derivative of $`\alpha `$ with respect to $`X`$ (i.e. along the flow of $`X`$) is
$$_X\alpha =\mathrm{𝐝𝐢}_X\alpha +𝐢_X𝐝\alpha .$$
(2.20)
We now argue as follows. Since $`\omega `$ is closed, i.e. $`𝐝\omega =0`$, Cartan’s magic formula, eq. 2.20, applied to $`\omega `$ becomes
$$_X\omega \mathrm{𝐝𝐢}_X\omega +𝐢_X𝐝\omega =\mathrm{𝐝𝐢}_X\omega .$$
(2.21)
So for $`X`$ to be symplectic is for $`𝐢_X\omega `$ to be closed. But by the Poincaré Lemma, if $`𝐢_X\omega `$ closed, it is locally exact. That is: there locally exists a scalar function $`f:M\mathrm{I}\mathrm{R}`$ such that
$$𝐢_X\omega =df\mathrm{i}.\mathrm{e}.X=X_f.$$
(2.22)
So for $`X`$ to be symplectic is equivalent to $`X`$ being locally Hamiltonian.
Thus we have
> Noether’s theorem for a Hamilton system If $`X`$ is a symmetry of a Hamiltonian system $`(M,\omega ,H)`$, then locally $`X=X_f`$; so by the anti-symmetry of the Poisson bracket, eq. 2.19, $`f`$ is a constant of the motion. And conversely: if $`f:M\mathrm{I}\mathrm{R}`$ is a constant of the motion, then $`X_f`$ is a symmetry.
We will see in Section 6.2 that most of this approach to Noether’s theorem, in particular the “one-liner” appeal to the anti-symmetry of the Poisson bracket, eq. 2.19, carries over to the more general framework of Poisson manifolds. For the moment, we mention an example (which we will also return to).
For most Hamiltonian systems in euclidean space $`\mathrm{I}\mathrm{R}^3`$, spatial translations and rotations are (continuous) symmetries. Let us consider in particular a system we will discuss in more detail in Section 2.3: $`N`$ point-particles interacting by Newtonian gravity. The Hamiltonian is a sum of two terms, which are each individually invariant under translations and rotations:
(i) a kinetic energy term $`K`$; though I will not go into details, it is in fact defined by the euclidean metric of $`\mathrm{I}\mathrm{R}^3`$, and is thereby invariant; and
(ii) a potential energy term $`V`$; it depends only on the particles’ relative distances, and is thereby invariant.
The corresponding conserved quantities are the total linear and angular momentum.<sup>7</sup><sup>7</sup>7By the way, this Hamiltonian is not invariant under boosts. But as I said in (iii) of Section 1.2, I restrict myself to time-independent transformations; the treatment of symmetries that “represent the relativity of motion” needs separate discussion.
### 2.2 The road ahead
In this Subsection, four comments will expand on the introductory comment (iv) of Section 1.2, and also give some information about the history of symplectic reduction and about some crucial examples.
(1): Generalizing from Noether’s theorem; Poisson manifolds:—
Noether’s theorem tells us that a continuous symmetry, i.e. a one-parameter group of symmetries, determines a first integral (i.e. a constant of the motion). So a larger group of symmetries, i.e. a group with several parameters, implies several first integrals. The phase flow is therefore confined to the intersection of the level surfaces of these integrals: an intersection which is in general a manifold. In other words: the simultaneous level manifold of these integrals is an invariant manifold of the phase flow.
It turns out that, in many useful cases, this manifold is also invariant under an appropriately chosen subgroup of the group of symmetries; and that the quotient space, i.e. the set of orbits under the action of this subgroup, is a manifold with a natural structure induced by the original Hamiltonian system that is sufficient to do mechanics in Hamiltonian style. The quotient space is therefore called the ‘reduced phase space’.
But in some cases, this natural structure is not a symplectic form, but a (mild) generalization in which the the form is allowed to be degenerate; i.e. like eq. 2.3 rather than eq. 2.4. A manifold equipped with such a structure need not be a quotient manifold. It can instead be defined in terms of a generalization of the usual Poisson bracket, as defined in terms of the symplectic form by eq. 2.18.
The key idea is to postulate a bracket, acting on the scalar functions $`F:M\mathrm{I}\mathrm{R}`$ on any manifold $`M`$, and possessing four properties enjoyed by the usual Poisson bracket. One of the properties is anti-symmetry, emphasised in Section 2.1.3’s proof of Noether’s theorem. The other three are that the postulated bracket, again written $`\{,\}`$, is: to be bilinear; to obey the Jacobi identity for any real functions $`F,G,H`$ on $`M`$, i.e.
$$\{\{F,H\},G\}+\{\{G,F\},H\}+\{\{H,G\},F\}=0;$$
(2.23)
and to obey Leibniz’ rule for products, i.e.
$$\{F,HG\}=\{F,H\}G+H\{F,G\}.$$
(2.24)
We will see in Section 5 that such a bracket, again called ‘Poisson bracket’, provides a sufficient framework for mechanics in Hamiltonian style. In particular, it induces an anti-symmetric bilinear form that may be degenerate, as in eq. 2.3. A manifold $`M`$ equipped with such a bracket is called a Poisson manifold.
The allowance of degeneracy means that a Poisson manifold can have odd dimension; while we saw in Section 2.1.1 that any symplectic manifold is even-dimensional. On the other hand, this generalized Hamiltonian mechanics will have clear connections with the usual formulation of Section 2.1. The main connection will be the result that any Poisson manifold $`M`$ is a disjoint union of even-dimensional manifolds, on which $`M`$’s degenerate antisymmetric bilinear form restricts to be non-degenerate.<sup>8</sup><sup>8</sup>8Because of these clear connections, it is natural to still call the more general framework ‘Hamiltonian’; as is usually done. But of course this is just a verbal matter.
(2): Historical roots:—
The theory of symplectic reduction has deep historical roots in the work of classical mechanics’ monumental figures. In part, this is no surprise. As mentioned in (i) of Section 1.2, cyclic coordinates underpin the role of symmetry in mechanics, and in particular Noether’s theorem. And Newton’s solution of the Kepler problem provides an example: witness textbooks’ expositions of the transition to centre-of-mass coordinates, and of polar coordinates with the angle being cyclic (yielding angular momentum as the conserved quantity). So it is unsurprising that various results and ideas of symplectic reduction can be seen in the work of such masters as Euler, Lagrange, Hamilton, Jacobi, Lie and Poincaré; for example (as we will see), in Euler’s theory of the rigid body.
But the history also holds a surprise. It turns out that Lie’s epoch-making work on Lie groups already contained a detailed development of much of the general, modern theory.<sup>9</sup><sup>9</sup>9The main source is his (1890). Besides, Arnold (1989: 456) reports that the prototype example of a Poisson manifold, viz. the dual of a finite-dimensional Lie algebra, was already understood by Jacobi. The sad irony is that most of Lie’s insights were not taken up—and were then repeatedly re-discovered. So this is yet another example (of which the history of mathematics has so many!) of the saying that he who does not learn from history is doomed to repeat it. The consolation is of course that it is often easier, and more fun, to re-discover something than to learn it from another…
Thus it was only from the mid-1960s that the theory, in essentially the form Lie had, was recovered and cast in the geometric language adopted by modern mechanics; namely, by contemporary masters such as Arnold, Kostant, Marsden, Meyer, Smale, Souriau and Weinstein; (cf. this Chapter’s first motto). Happily, several of these modern authors are scholars of the history, and even their textbooks give some historical details: cf. Marsden and Ratiu (1999, pp. 336-8, 369-370, 430-432), and the notes to each Chapter of Olver (2000: especially p.427-428). (Hawkins (2000) is a full history of Lie groups from 1869 to 1926; for Lie, cf. especially its Sections 1.3, 2.5 and Chapter 3, especially 3.2.)
In any case, setting history aside: symplectic reduction has continued since the 1970s to be an active research area in contemporary mechanics, and allied fields such as symplectic geometry. So it has now taken its rightful place as a major part of the contemporary renaissance of classical mechanics: as shown by …
(3): Two examples: the rigid body and the ideal fluid:—
Two examples illustrate vividly how symplectic reduction can give new physical understanding, even of time-honoured examples: the rigid body and the ideal fluid—as attested by this Chapter’s mottoes. (Section 2.3 will develop a third example, more closely related to philosophy.)
As to the rigid body: we will see (especially in Section 5) that symplectic reduction considerably clarifies the elementary theory of the rigid body (Euler’s equations, Euler angles etc.): which, notoriously, can be all too confusing! For simplicity, I shall take the rigid body to be pivoted, so as to set aside translational motion. This will mean that the group of symmetries defining the quotienting procedure will be the rotation group. It will also mean that the rigid body’s configuration space is given by the rotation group, since any configuration can be labelled by the rotation that obtains it from some reference-configuration. So in this application of symplectic reduction, the symmetry group (viz. the rotation group) will act on itself as the configuration space. This example will also give us our prototype example of a Poisson manifold.
As to the ideal fluid, i.e. a fluid that is incompressible and inviscid (with zero viscosity): this is of course an infinite-dimensional system, and so (as I announced in Section 1.2) outside the scope of this Chapter. So I will not go into any details, but just report the main idea.
The equations of motion of an ideal fluid, Euler’s equations, are usually derived either by applying Newton’s second law $`𝐅=m𝐚`$ to a small fluid element; or by a heuristic use of the Lagrangian or Hamiltonian approach (as in heuristic classical field theories). But in the mid-1960s, Arnold showed how the latter derivations could be understood in terms of a striking, even beautiful, analogy with the above treatment of the rigid body. Namely, the analogy shows that the configuration space of the fluid is an infinite-dimensional group; as follows. The configuration of an ideal fluid confined to some container occupying a volume $`V\mathrm{I}\mathrm{R}^3`$ is an assignment to each spatial position $`xV`$ of an infinitesimal fluid element. Given such an assignment as a reference-configuration, any other configuration can be labelled by the volume-preserving diffeomorphism $`d`$ from $`V`$ to $`V`$ that carries the reference-configuration to the given one, by dragging each fluid element along by $`d`$. So given a choice of reference-configuration, the fluid’s configuration space is given by the infinite-dimensional group $`𝒟`$ of diffeomorphisms $`d:VV`$: just as the rotation group is the configuration space of a (pivoted) rigid body. $`𝒟`$ then forms the basis for rigorous Lagrangian and Hamiltonian theories of an ideal fluid.
These theories turn out to have considerable analogies with the Lagrangian and Hamiltonian theories of the rigid body, thanks to the fact that in both cases the symmetry group forms the configuration space. In particular, Euler’s equations for ideal fluids are the analogues of Euler’s equations for a rigid body. Besides, these rigorous theories of fluids (and symplectic reduction applied to them) are scientifically important: they have yielded various general theorems, and solved previously intractable problems. (For more details, cf. Abraham and Marsden (1978: Sections 4.4 and 4.6 for the rigid body, and 5.5.8 for the ideal fluid), Arnold (1989: Appendix 2:C to 2:F for the rigid body, and 2:G to 2:L for the ideal fluid), and Marsden and Ratiu (1999: Chapters 1.4 and 15 for the rigid body, and 1.5, p. 266, for the ideal fluid).)
(4): Philosophical importance:—
Symplectic reduction is also, I submit, philosophically important; in at least two ways. The first way is specific: it illustrates some methodological morals about how classical mechanics analyses problems. I develop this theme in (Butterfield 2005). The second way is more general: the theory, or rather various applications of it, is directly relevant to disputes in the philosophy of space and time, and of mechanics. This relevance is recognized in contemporary philosophy of physics. So far as I know, the authors who develop these connections in most detail are Belot and Earman. They discuss symplectic reduction in connection with such topics as:
(i) the treatment of symmetries, including gauge symmetries;
(ii) the dispute between absolute and relationist conceptions of space and time; and
(iii) the interpretation of classical general relativity (a topic which connects (i) and (ii), and bears on heuristics for quantum gravity).
Thus Belot (1999, 2000, 2001, 2003, 2003a) and Earman (2003) discuss mainly (i) and-or (ii); Belot and Earman (2001) discusses (iii). For (i) and (ii), I also recommend Wallace (2003).
But these papers have a demanding pre-requisite: they invoke, but do not expound, the theory of symplectic reduction. They also discuss infinite-dimensional systems (especially classical electromagnetism and general relativity), without developing finite-dimensional examples like the rigid body. Indeed, there is, so far as I know, no article-length exposition of the theory which is not unduly forbidding for philosophers. So I aim to give such an exposition, to help readers of papers such as those cited.<sup>10</sup><sup>10</sup>10As I said in Section 1.2, my material is drawn from the books by Abraham and Marsden, Arnold, Marsden and Ratiu, and Olver. More precisely, I will mostly draw on: Abraham and Marsden (1978: Sections 3.1-3.3, 4.1-4.3), Arnold (1989: Appendices 2, 5 and 14), Marsden and Ratiu (1999: Chapters 9-13) and Olver (2000: Chapter 6). And much of what follows—in spirit, and even in letter—is already in Lie (1890)! As a (non-philosophical) introduction to symplectic reduction, I also recommend Singer (2001). It is at a yet more elementary level than what follows; e.g. it omits Poisson manifolds and co-adjoint representations.
As an appetizer for this exposition, I will first (in Section 2.3) follow Belot in presenting the general features of a finite-dimensional symplectic reduction which has vivid philosophical connections, viz. to the absolute vs. relationist debate. This example concerns a system of point-particles in Euclidean space, either moving freely or interacting by a force such as Newtonian gravity. (The symmetries defining the quotienting procedure are given by the Euclidean group of translations and rotations.) For philosophers, this will be a good appetizer for symplectic reduction, since it sheds considerable light on relationism about space of the sort advocated by Leibniz and Mach.
### 2.3 Appetizer: Belot on relationist mechanics
#### 2.3.1 Comparing two quotienting procedures
In several papers, Belot discusses how symplectic reduction bears on the absolute-vs.-relational debate about space. I shall pick out one main theme of his discussions: the comparison of a relational classical mechanical theory with what one gets by quotienting the orthodox absolutist (also called a ‘substantivalist’) classical mechanics, by an appropriate symmetry group. His main contention—which I endorse—is that this comparison sheds considerable light on relationism: on both its motivation, and its advantages and disadvantages.<sup>11</sup><sup>11</sup>11The main references are Belot (1999, 2001, 2003: Sections 3.5, 5). Cf. also his (2000: Sections 4 to 5.3), (2003a: Section 6). Though I recommend all these papers, the closest template for what follows is (2001: Section VI et seq.).
Belot’s overall idea is as follows. Where the relationist admits one possible configuration, as (roughly) a specification of all the distances (and thereby angles) between all the parts of matter, the absolutist (or substantivalist) sees an infinity of possibilities: one for each way the relationist’s configuration (a relative configuration) can be embedded in the absolute space. This makes it natural to take the relationist to be envisaging a mechanics which is some sort of “quotient” of the absolutist’s mechanics.
In particular, on the traditional conception of space as Euclidean (modelled by $`\mathrm{I}\mathrm{R}^3`$), each of the relationist’s relative configurations corresponds to an equivalence class of absolutist configurations (i.e. embeddings of arrangements of matter into $`\mathrm{I}\mathrm{R}^3`$), with the members of the class related by spatial translations and rotations, i.e. elements of the Euclidean group. In the jargon of group actions, to be developed in Section 4: the Euclidean group acts on the set of all absolutist configurations, and a relative configuration corresponds to an orbit of this action. So it is natural to take the relationist to be envisaging a mechanics which quotients this action of the Euclidean group, to get a relative configuration space. A relationist mechanics, of Lagrangian or Hamiltonian type, is then to be built up on this space of relative configurations.
But as Belot emphasises, one can instead consider quotienting the absolutist’s state-space—i.e. in a Hamiltonian framework, the phase space—rather than their configuration space. Indeed, this is exactly what one does in symplectic reduction. In particular, the Euclidean group’s action on the absolutist’s configuration space, $`Q`$ say, can be lifted to give an action on the cotangent bundle $`T^{}Q`$; which is accordingly called the ‘cotangent lift’. One can then take the quotient, i.e. consider the orbits into which $`T^{}Q`$ is partitioned by the cotangent lift.
We thus have two kinds of theories to compare: (i) the relationist theories, built up from the relative configuration space; which for the sake of comparison with symplectic reduction we take to be Hamiltonian, rather than Lagrangian; (ii) theories obtained by quotienting “later”, i.e. quotienting the absolutist’s cotangent bundle.
I will now spell out this comparison. But I will not try to summarize Belot’s more detailed conclusions, about what such a comparison reveals about the advantages and disadvantages of relationism. They are admirably subtle, and so defy summary: they can mainly be found at his (2000: p. 573-574, 582; 2001: Sections VIII to X). (Rovelli (this volume) also discusses relationism.)
As befits an appetizer, I will also (like Belot) concentrate on as simple a case as possible: a mechanics of $`N`$ point-particles, which is to assume a Euclidean spatial geometry. Of course, the absolutist make this assumption by postulating a Euclidean space; but for the relationist, the assumption is encoded in constraints relating the various inter-particle distances. The main current example of a relationist mechanics of such a system is due to Barbour and Bertotti (1982), though they develop it in the Lagrangian framework; (to be precise, in terms of Jacobi’s principle). Belot also discusses other relational theories, including field theories, i.e. theories of infinite systems; some of them also due to Barbour, and in a Lagrangian framework. But in this Section I only consider $`N`$ point-particles.
Also, I will also not discuss boosts, though of course the relationist traditionally proposes to identify any two absolutist states of motion related by a boost. In terms of group actions, this means I will consider quotienting by an action of the euclidean group, but not the Galilei group. (Cf. how I set aside time-dependent transformations already in (iii) of Section 1.2.) I will also postpone to later Sections technical details, even when our previous discussion makes them accessible.
Finally, a warning to avoid later disappointment! The later Sections will not include a full analysis of the euclidean group’s actions on configuration space and phase space, and their quotients. That would involve technicalities going beyond an appetizer. Instead (as mentioned at the end of Section 1.2), the material in later Sections is chosen so as to lead up to Section 7’s theorem, the Lie-Poisson reduction theorem, about quotienting the phase space of a system whose configuration space is a Lie group. Further reasons for presenting the material for this theorem will be given in Section 5.1.
#### 2.3.2 The spaces and group actions introduced
Let us begin by formulating the orthodox absolutist mechanics of $`N`$ point-particles interacting by Newtonian gravity, together with the action of the Euclidean group.
Each point-particle occupies a point of $`\mathrm{I}\mathrm{R}^3`$, so that the configuration space $`Q`$ is $`\mathrm{I}\mathrm{R}^{3N}`$: dim($`Q`$) = 3$`N`$. So the phase space for Hamiltonian mechanics will be the cotangent bundle $`T^{}Q(q,p)`$: dim($`T^{}Q`$) = 6$`N`$.
The Hamiltonian is a sum of kinetic and potential terms, $`K`$ and $`V`$. $`K`$ depends only on the $`p`$s, and $`V`$ only on the $`q`$s. In cartesian coordinates, with $`i`$ now labelling particles $`i=1,\mathrm{},N`$ rather than degrees of freedom, we have the familiar expressions:
$$H(q,p)=K(p)+V(q)\mathrm{with}K=\mathrm{\Sigma }_i\frac{𝐩_i^2}{2m_i},V(q)=G\mathrm{\Sigma }_{i<j}\frac{m_im_j}{𝐪_i𝐪_j}$$
(2.25)
where $`m_i`$ are the masses and $`G`$ is the gravitational constant.<sup>12</sup><sup>12</sup>12From the broader philosophical perspective, the most significant feature of eq. 2.25 is no doubt the fact that the potential is a sum of all the two-body potential energies for the configuration $`qQ`$: there are no many-body interactions.<sup>,</sup><sup>13</sup><sup>13</sup>13Incidental remark. In fact, the kinetic energy can be represented by a metric $`g`$ on the configuration space. For Hamiltonian mechanics, this means that the kinetic energy scalar $`K`$ on the cotangent bundle $`T^{}Q`$ can be defined by applying $`Q`$’s metric $`g`$ to the projections of the momenta $`p`$, where at each point $`(q,p)T^{}Q`$ the projection is made with the preferred isomorphism $`\omega ^{\mathrm{}}:T_q^{}T_q`$; (cf. eq. 2.12). That is:—
$$K:(q,p)T^{}Qg_q(\omega ^{\mathrm{}}(p),\omega ^{\mathrm{}}(p)).$$
(2.26)
The euclidean group $`E`$ (aka: $`E(3)`$) is the group (under composition) of translations, rotations and reflections on $`\mathrm{I}\mathrm{R}^3`$. But since we will be interested in continuous symmetries, we will ignore reflections, and so consider the subgroup of orientation-preserving translations and rotations; i.e. the component of the group connected to the identity transformation (which I will also write as $`E`$). This is a Lie group, i.e. a group which is also a manifold, with the group operations smooth with respect to the manifold structure. Section 3 will give formal details. Here we just note that we need three real numbers to specify a translation ($`𝐱=(x,y,z)`$), and three to specify a rotation (two for an axis, and one for the angle through which to rotate); and accordingly, it is unsurprising that as a manifold, the dimension of $`E`$ is 6: dim($`E`$) = 6.
$`E`$ acts in the obvious sense on $`\mathrm{I}\mathrm{R}^3`$. For example, if $`gE`$ is translation by $`𝐱\mathrm{I}\mathrm{R}^3`$, $`g`$ induces the map $`𝐪\mathrm{I}\mathrm{R}^3𝐪+𝐱`$. Similarly for a rotation induces: again, Section 3 will give a formal definition.
Now let $`E`$ act in this way on each of the $`N`$ factor spaces $`\mathrm{I}\mathrm{R}^3`$ of our system’s configuration manifold $`Q=\mathrm{I}\mathrm{R}^{3N}`$. This defines an action $`\mathrm{\Phi }`$ on $`Q`$: i.e. for all $`gE`$, there is a map $`\mathrm{\Phi }_g:QQ`$. For example, for $`g=`$ a translation by $`𝐱\mathrm{I}\mathrm{R}^3`$, we have
$$\mathrm{\Phi }_g:(𝐪_j)=(𝐪_1,\mathrm{},𝐪_N)Q(𝐪_1+𝐱,\mathrm{},𝐪_N+𝐱)Q;$$
(2.27)
and similarly for rotations. Since the potential function $`V:Q\mathrm{I}\mathrm{R}`$ of eq. 2.25 depends only on inter-particle distances, each map $`\mathrm{\Phi }_g:QQ`$ is a symmetry of the potential; i.e. we have $`V(\mathrm{\Phi }_g(q))=V(q)`$.
The action $`\mathrm{\Phi }`$ (i.e. the assignment $`gE\mathrm{\Phi }_g`$) induces an action of $`E`$ on $`T^{}Q=T^{}\mathrm{I}\mathrm{R}^{3N}`$, called the cotangent lift of $`\mathrm{\Phi }`$ to $`T^{}Q`$, and usually written as $`\mathrm{\Phi }^{}`$; so that we have for each $`gE`$ a lifted map $`\mathrm{\Phi }_g^{}:T^{}QT^{}Q`$. Again, the details can wait till later (Section 4). But the idea is that each map $`\mathrm{\Phi }_g`$ on $`Q`$ is smooth, and so maps curves to curves, and so vectors to vectors, and so covectors to covectors, and so on.
Unsurprisingly, each of the lifted maps $`\mathrm{\Phi }_g^{}:T^{}QT^{}Q`$ leaves the potential $`V`$, now considered as a scalar on $`T^{}Q`$, invariant: i.e. we have $`V(\mathrm{\Phi }_g^{}(q,p))=V(q,p)V(q)`$. But furthermore, each of the lifted maps $`\mathrm{\Phi }_g^{}`$ is a symmetry of the Hamilton system, in our previous sense (Section 2.1.3). That is: $`\mathrm{\Phi }_g^{}`$ preserves the Hamiltonian (indeed the kinetic and potential terms are separately invariant); and it preserves the symplectic structure. This means the dynamics is invariant under the action of all $`gG`$: the dynamical histories of the system through $`(q,p)`$ and through $`\mathrm{\Phi }_g^{}(q,p)`$ match exactly at each time. They are qualitatively indistinguishable: in contemporary metaphysical jargon, they are duplicates.
At this point, of course, we meet the absolute-vs.-relational debate about space. The absolutist asserts, and the relationist denies, that there being two such indistinguishable possibilities makes sense.<sup>14</sup><sup>14</sup>14The locus classicus for this debate is of course the Leibniz-Clarke correspondence, though the protagonists’ argumentation is of course sometimes theological. Clarke the absolutist maintains that there are many possible arrangements of bits of matter in space consistent with a specification of all relative distances, saying ‘if \[the mere will of God\] could in no case act without a pre-determining cause … this would tend to take away all power of choosing, and to introduce fatality.’ Leibniz claims there is only one such arrangement: ‘those two states … would not at all differ from one another. Their difference therefore is only to be found in our chimerical supposition of the reality of space in itself.’ So the relationist, presented with the theory above, says we should cut down the space of possibilities. As I said in Section 2.3.1, it is natural to make this precise in terms of quotienting the action of the euclidean group: a set of absolutist possibilities related one to another by elements of the euclidean group form an equivalence class (an orbit) which is to represent one relationist possibility.
But here we need to distinguish two different quotienting procedures. I will call them Relationism and Reductionism (with capital R’s), since the former is close to both traditional and contemporary relationist proposals, and the latter is an example of the orthodox idea of symplectic reduction. As I said in Section 2.3.1, the main difference will be that:
(i): Relationism performs the quotient on $`E`$’s action on the configuration space $`Q`$; the set of orbits form a relative configuration space, on which the relationist proposes to build a dynamics, whether Lagrangian or Hamiltonian—yielding in the latter case, a relative phase space; whereas
(ii): Reductionism performs the quotient on $`E`$’s action on the usual phase space $`T^{}Q`$, the set of orbits forming a reduced phase space.
Since our discussion adopts the Hamiltonian framework, it will not matter for what follows, that Relationism, as defined, can adopt the Lagrangian framework, while Reductionism is committed to the Hamiltonian one. What will matter is that (i) and (ii) make for phase spaces of different dimensions; the reduced phase space has six more dimensions than the relative phase space. The “dimension gap” is six.
We will see that four of the six variables that describe these dimensions are constants of the motion; the other two vary with time. And for certain choices of values of the constants of the motion (roughly: no rotation), the time-varying variables drop out, and the dynamics according to the Reductionist theory simplifies so as to coincide with that of the Relationist theory. In other words: if we impose no rotation, then the heterodox Relationist dynamics matches the conventional Reduced dynamics.
#### 2.3.3 The Relationist procedure
The Relationist seeks a mechanics based on the relative configuration space (RCS). An element of the RCS is to be a pattern of inter-particle distances and angles that is geometrically possible, i.e. compatible with the $`N`$ particles being embedded in $`\mathrm{I}\mathrm{R}^3`$. So, roughly speaking, an element of the RCS is a euclidean configuration, modulo isometries; and the RCS will be the set of orbits $`\mathrm{I}\mathrm{R}^{3N}/E`$.
Even before giving a more precise statement, we can state the “punchline” about dimensions, as follows. Since dim$`(E)=6`$, quotienting by $`E`$ subtracts six dimensions: that is, the dimension of the RCS will be 3$`N`$-6.
But we need to be more precise about the RCS. For the orbits and quotient spaces to be manifolds, and for dimensions to add or subtract in this simple way, we need to excise two classes of “special” points from $`\mathrm{I}\mathrm{R}^{3N}`$, before we quotient. (But I postpone till Section 4 the technical rationale for these excisions.)
Let $`\delta _Q\mathrm{I}\mathrm{R}^{3N}`$ be the set of configurations which are symmetric: i.e. each is fixed by some element of $`E`$ (other than the identity element!). Any configuration in which all the point-particles are collinear provides an example: the configuration is fixed by any rotation about the line as axis. Let $`\mathrm{\Delta }_Q`$ be the set of collision configurations; i.e. configurations in which two or more particles are coincident in the usual configuration space $`\mathrm{I}\mathrm{R}^{3N}`$. (The $`Q`$ subscripts will later serve as a reminder that these sets are sets of configurations.) $`\delta _Q`$ and $`\mathrm{\Delta }_Q`$ are both of measure zero in $`\mathrm{I}\mathrm{R}^{3N}`$. Excise both of them, and call the resulting space, which is again of dimension $`3N`$: $`Q:=\mathrm{I}\mathrm{R}^{3N}(\delta _Q\mathrm{\Delta }_Q)`$.
$`\delta _Q`$ and $`\mathrm{\Delta }_Q`$ are each closed under the action of $`E`$. That is, each is a union of orbits: a euclidean transformation of a symmetric (collision) configuration is also symmetric (collision). So $`E`$ acts on $`Q`$. Now quotient $`Q`$ by $`E`$. $`Q/E`$ is the Relationist’s RCS. Since dim$`(E)=6`$, we have: dim($`Q/E`$) = 3$`N`$-6.
These $`3N6`$ variables encode all of a (relative) configuration’s particle-pair relative distances, $`r_{ij}\mathrm{I}\mathrm{R}`$ (with $`i,j`$ labelling particles). Note that there are $`N(N1)/2`$ such relative distances; and for $`N>4`$, this is greater than $`3N6`$: (for $`N>>4`$, it is much greater). So the relative distances, though physically intuitive, give an over-complete set of coordinates on $`Q/E`$. (So they cannot be freely chosen: there are constraints between them.)
So the Relationist seeks a mechanics that uses this RCS. Newton’s second law being second-order in time means that she will also need quantities like velocities (in a Lagrangian framework) or like momenta (in a Hamiltonian framework). For the former, she will naturally consider the $`N(N1)/2`$ relative velocities $`\dot{r}_{ij}:=\frac{d}{dt}r_{ij}`$; and for the latter, the corresponding momenta $`p_{ij}:=\frac{L}{\dot{r}_{ij}}`$. Again, she must beware of constraints. The tangent and cotangent bundles built on her RCS $`Q/E`$ will each have dimension $`2(3N6)=6N12`$. So again, for $`N>4`$, the number $`N(N1)/2`$ of relative velocities $`\dot{r}_{ij}`$, or of relative momenta $`p_{ij}`$, is greater than the number of degrees of freedom concerned; and for $`N>>4`$, it is much greater. So again, the relative velocities or relative momenta are over-complete: there are constraints.
On the other hand, if the Relationist uses only these relative quantities, $`r_{ij}`$ and either $`\dot{r}_{ij}`$ or $`p_{ij}`$ (or “equivalent” coordinates on $`T(Q/E)`$ or $`T^{}(Q/E)`$ that are not over-complete), she faces a traditional problem—whatever the other details of her theory. At least, she faces a problem if she hopes for a deterministic theory which is empirically equivalent to the orthodox absolutist theory. I will follow tradition and state the problem in terms of relative velocities rather than momenta.
The problem concerns rotation; (and herein lies the strength of Newton’s and Clarke’s position in the debate against Leibniz). For according to the absolutist theory two systems of point-particles could match with respect to all relative distances and relative velocities, and yet have different future evolutions; so that a theory allowing the same possibilities as the absolutist one, yet using only these relative quantities (or “equivalent” variables) would have to be indeterministic.
The simplest example is an analogue for point-particles of Newton’s two globes thought-experiment. Thus the systems could each consist of just two point-particles with zero relative velocity. One system could be non-rotating, so that the point-particles fall towards each other under gravity; while the other system could be rotating about an axis normal to the line between the particles, and rotating at just such a rate as to balance the attractive force of gravity.
The Relationist has traditionally replied that they do not hope for a theory empirically equivalent to the absolutist one. Rather, they envisage a mechanics in which, of the two systems mentioned, only the non-rotating evolution is possible: more generally, a mechanics in which the universe as a whole must have zero angular momentum. Originally, in authors like Leibniz and Mach, this reply was a promissory note. But modern Relationist theories such as Barbour and Bertotti’s (1982) have made good the promise; and they have been extended well beyond point-particles interacting by Newtonian gravity. Besides, since the universe seems in fact to be non-rotating, these theories can even claim to be empirically adequate, at least as regards this principal difference from absolutist theories.<sup>15</sup><sup>15</sup>15An advocate of the absolutist theory might say that it is odd to make what seems a contingent feature of the universe, non-rotation, a principle of mechanics; and the Relationist might reply that their view has the merit of predicting that the universe does not rotate! I fear there are no clear criteria for settling this methodological dispute; anyway, I will not pursue it.
But it is not my brief to go into these theories’ details, except by way of comparison with a quotiented version of the absolutist theory: cf. Section 2.3.4.
#### 2.3.4 The Reductionist procedure
The Reductionist’s main idea is to quotient only after passing to the orthodox phase space for $`N`$ point-particles, i.e. the cotangent bundle $`T^{}\mathrm{I}\mathrm{R}^{3N}`$ of $`\mathrm{I}\mathrm{R}^{3N}`$. So the idea is to consider $`(T^{}\mathrm{I}\mathrm{R}^{3N})/E`$, i.e. the quotient of $`T^{}\mathrm{I}\mathrm{R}^{3N}`$ by the cotangent-lifted action $`\mathrm{\Phi }^{}`$ of the euclidean group $`E`$.
More precisely, we again proceed by first excising special points that would give technical trouble. But now the points to be excised are in the cotangent bundle $`T^{}\mathrm{I}\mathrm{R}^{3N}`$, not in $`\mathrm{I}\mathrm{R}^{3N}`$. So let $`\delta T^{}\mathrm{I}\mathrm{R}^{3N}`$ be the set of phase space states whose configurations are symmetric (in the sense of Section 2.3.3’s $`\delta _Q`$). Let $`\mathrm{\Delta }T^{}\mathrm{I}\mathrm{R}^{3N}`$ be the set of collision points; i.e. states in which two or more particles are coincident in the configuration space $`\mathrm{I}\mathrm{R}^{3N}`$. Both $`\delta `$ and $`\mathrm{\Delta }`$ are of measure zero. Excise both of them, and call the resulting phase space, which is again of dimension $`6N`$: $`M:=T^{}\mathrm{I}\mathrm{R}^{3N}(\delta \mathrm{\Delta })`$.
$`\delta `$ and $`\mathrm{\Delta }`$ are each closed under the cotangent-lifted action of $`E`$ on $`T^{}\mathrm{I}\mathrm{R}^{3N}`$. That is, each is a union of orbits: the cotangent lift of a euclidean transformation acting on a phase space state with a symmetric (collision) configuration yields a state which also has a symmetric (collision) configuration. So $`E`$ acts on $`M`$. Now quotient $`M`$ by $`E`$, getting $`\overline{M}:=M/E`$. This is called reduced phase space. We have: dim($`\overline{M}`$) = dim($`M`$) - dim($`E`$) = $`6N6`$.
As emphasised at the end of Section 2.3.2, $`\overline{M}`$ has six more dimensions than the corresponding Relationist phase space (whether the velocity phase space (tangent bundle) or the momentum phase space (cotangent bundle)). The dimension of those phase spaces is $`2(3N6)=6N12`$. Indeed, we can better understand both the reduced phase space $`\overline{M}`$ and Relationist phase spaces by considering this “dimension gap”. There are two extended comments to make.
(1): Obtaining the Relationist phase space:—
We can obtain the Relationist momentum phase space from our original phase space $`M`$. Thus let $`M_0`$ be the subspace of $`M`$ in which the system has total linear momentum and total angular momentum both equal to zero. Since these are constants of the motion, $`M_0`$ is dynamically closed and so supports a Hamiltonian dynamics given just by restriction of the original dynamics. With linear and angular momentum each contributing three real numbers, dim($`M_0`$) = dim($`M`$) - 6 = $`6N6`$. Furthermore, $`M_0`$ is closed under (is a union of orbits under) the cotangent-lifted action of $`E`$. So let us quotient $`M_0`$ by this action of $`E`$, and write $`\overline{M}_0:=M_0/E`$. Then dim($`\overline{M}_0`$) = $`6N66=6N12`$.
Now recall that this is the dimension of the phase space of the envisaged Relationist theory built on the RCS $`Q/E`$. And indeed, as one would hope: $`\overline{M}_0`$ is the Hamiltonian version of Barbour and Bertotti’s 1982 Relational theory; (recall that they work in a Lagrangian framework).
That is: $`\overline{M}_0`$ is a symplectic manifold, and points in $`\overline{M}_0`$ are parametrized by all the particle-to-particle relative distances and relative velocities. There is a deterministic dynamics which matches that of the original absolutist theory, once the original dynamics is projected down to Section 2.3.3’s relative configuration space $`Q/E`$.
In short: the vanishing total linear and angular momenta mean that an initial state comprising only relative quantities is sufficient to determine all future relative quantities.
(2): Decomposing the Reductionist reduced phase space:—
Let us return to the reduced phase space $`\overline{M}`$. The first point to make is that since the Hamiltonian $`H`$ on $`M`$, or indeed on $`T^{}\mathrm{I}\mathrm{R}^{3N}`$, is invariant under the cotangent-lifted action of $`E`$, the usual dynamics on $`M`$ projects down to $`\overline{M}=M/E`$. That is: the reduced phase space dynamics captures all the $`E`$-invariant features of the usual dynamics.
In fact, $`\overline{M}`$ is a Poisson manifold. So it is our first example of the more general framework for Hamiltonian mechanics announced in (1) of Section 2.2. Again, I postpone technical detail till later (especially Sections 5.1 and 5.2.4). But the idea is that a Poisson manifold has a degenerate antisymmetric bilinear map, which implies that the manifold is a disjoint union of symplectic manifolds. Each symplectic manifold is called a leaf of the Poisson manifold. The leaves’ symplectic structures “mesh” with one another; and within each leaf there is a conventional Hamiltonian dynamics.
Even without a precise definition of a Poisson manifold, we can describe how $`M`$ is decomposed into symplectic manifolds, each with a Hamiltonian dynamics. Recall that we have: dim($`\overline{M}`$) = dim($`M`$) - dim($`E`$) = $`6N6`$. This breaks down as:
$$6N6=(6N12)+3+3=2(3N6)+3+3=:\alpha +\beta +\gamma $$
(2.28)
where the right hand side defines $`\alpha ,\beta ,\gamma `$ respectively as $`2(3N6),3`$ and $`3`$. In terms of $`\overline{M}`$, this means the following.
(i): $`\alpha `$ corresponds to (1)’s $`\overline{M}_0`$, i.e. to $`T^{}(Q/E)`$. As discussed, $`3N6`$ variables encode all the particle-pair relative distances; and the other $`3N6`$ variables encode all the particle-pair relative momenta.
The six extra variables additional to these 6$`N`$-12 relative quantities consist of: four constants of the motion, and two other variables which are dynamical, i.e. change in time.
(ii): $`\beta `$ stands for three of the four constants of the motion: viz. the three variables that encode the total linear momentum of the system, i.e. the momentum of the centre of mass. These constants of the motion are “just parameters” in the sense that: (a) not only does specifying a value for all three of them fix a surface, i.e. a $`(6N9)`$-dimensional hypersurface in $`\overline{M}`$, on which there is a Hamiltonian dynamics; also (b) this Hamiltonian and symplectic structure is independent of the values we specify.<sup>16</sup><sup>16</sup>16As mentioned at the end of Section 2.3.1, the relationist traditionally proposes to identify absolutist states of motion that differ just by the value of the total momentum. And indeed, the proposal can be implemented by considering an action of the Galilean group on the absolutist phase space $`M`$, and identifying points related by Galilean boosts. For discussion and references, cf. Belot (2000: Section 5.3).
(iii): $`\gamma `$ stands for the three variables that encode the total angular momentum of the system. One of these is a fourth constant of the motion, viz. the magnitude $`L`$ of the total angular momentum. The other two time-varying quantities fix a point on a sphere (2-sphere) of radius $`L`$, encoding the direction of the angular momentum of the system in a frame rotating with it. The situation is as in the elementary theory of the rigid body: though the total angular momentum relative to coordinates fixed in space is a constant of the motion (three constant real numbers), the total angular momentum relative to the body is constant only in magnitude (one real number $`L`$), not in direction. This will be clearer in Section 5 onwards, when we describe the Poisson manifold structure in the theory of the rigid body. For the moment, there are two main comments to make about the $`N`$ particle system:—
(a): If we specify $`L`$, in addition to the momentum of the centre of mass of the system, we get a $`(6N10)`$-dimensional hypersurface in $`\overline{M}`$, on which (as in (ii)) there is a Hamiltonian dynamics. So we can think of $`\overline{M}`$ as consisting of the four real-parameter family of these hypersurfaces, with each point of each hypersurface being equipped with a sphere of radius $`L`$; (subject to a qualification in (b) below).
Note that here ‘each point being equipped’ does not mean that the sphere gives the extra dimensions that would constitute $`\overline{M}`$ as a fibre bundle; (there would be two dimensions lacking). Rather: in the point’s representation by $`6N10`$ real numbers, two of the numbers can be taken to represent a point on a sphere.
(b): But unlike the situation for $`\beta `$ in (ii) above, the Hamiltonian dynamics on such a hypersurface depends on the value of $`L`$. In particular, if $`L=0`$ the sphere representing the body angular momentum is degenerate: it is of radius zero, and the other two time-varying quantities drop out. A point in the hypersurface is represented by $`6N12`$ real numbers; i.e. the hypersurface is $`6N12`$-dimensional.
Now recall from Section 2.3.3 or (1) above that $`6N12`$ is the dimension of the phase space of the envisaged Relationist theory built on the RCS $`Q/E`$. And indeed, just as one would hope: the hypersurface with $`L=0`$ and also with vanishing linear momentum, with its dynamics, is the symplectic manifold and dynamics that is the Hamiltonian version of Barbour and Bertotti’s 1982 Relational theory of $`N`$ point-particles. In terms of (1)’s notation, this hypersurface is $`\overline{M}_0`$.
We can sum up this comparison as follows. On this hypersurface $`\overline{M}_0`$, the dynamics in the reduced phase space coincides with the dynamics one obtains for the relative variables, if one arbitrarily embeds their initial values in the usual absolutist phase space $`T^{}\mathrm{I}\mathrm{R}^{3N}`$, subject to the constraint that the total angular and linear momenta vanish, and then reads off (just by projection) their evolution from the usual evolution in $`T^{}\mathrm{I}\mathrm{R}^{3N}`$.
#### 2.3.5 Comparing the Relationist and Reductionist procedures
In comparing the Relationist and Reductionist procedures, I shall just make just two extended comments, and refer to Belot for further discussion. The gist of both comments is that Reductionism suffices: Relationism is not needed. The first is a commonplace point; the second is due to Belot.
##### 2.3.5.A Reductionism allows for rotation
The first comment reiterates the Reductionist’s ability, and the Relationist’s inability, to endorse Newton’s globes (or bucket) thought-experiment. The Reductionist can work in either
(i) the $`(6N6)`$-dimensional phase space $`\overline{M}=M/E`$; or
(ii) the $`(6N9)`$-dimensional hypersurface got from (i) by specifying the centre of mass’ linear momentum; or
(iii) the $`(6N10)`$-dimensional hypersurface got from (ii) by also specifying a non-zero value of $`L`$.
In all three cases, the Reductionist can describe rotation in a way that the Relationist with their $`(6N12)`$-dimensional space cannot. For she has to hand the three extra non-relative variables ($`L`$ and two others) that describe the rotation of the system as a whole. (Incidentally: that they describe the system as a whole is suggested by there being just three of them, whatever the value of $`N`$.) In particular, she can distinguish states of rotation and non-rotation ($`L=0`$), in the sense of endorsing the distinctions advocated by the globes and bucket thought-experiments.
The Reductionist can also satisfy a traditional motivation for relationism, which concerns general philosophy, rather than the theory of motion. It is especially associated with Leibniz: namely, our theory (or our metaphysics) should not admit distinct but utterly indiscernible possibilities. One might well ask why we should endorse this “principle of the identity of indiscernibles” for possibilities rather than objects. For Leibniz himself, the answer lies (as Belot’s (2001) brings out) in his principle of sufficient reason, and ultimately in theology.
But in any case the Reductionist can satisfy the requirement. Agreed, the usual absolutist theory, cast in $`T^{}\mathrm{I}\mathrm{R}^{3N}`$ (or if you prefer, $`M=T^{}\mathrm{I}\mathrm{R}^{3N}(\delta \mathrm{\Delta })`$) has nine variables that describe (i) the position of the centre of mass, (ii) the orientation of the system about its centre of mass, and (iii) the system’s total linear momentum: i.e. three variables, a vector in $`\mathrm{I}\mathrm{R}^3`$, for each of (i)-(iii). So the usual absolutist theory has a nine-dimensional “profligacy” of distinct but indiscernible possibilities. But as we have seen, the Reductionist quotients by the action of the euclidean group $`E`$, and so works in $`\overline{M}=M/E`$: which removes the profigacy about (i) and (ii).
As to (iii), I agree that for all I have said, a job remains to be done. The foliation of $`\overline{M}`$ by a three real-parameter family of $`(6N9)`$-dimensional hypersurfaces, labelled by the system’s total linear momentum, codifies the profligacy—but does not eliminate it. But as I mentioned above (cf. footnote 16), the Reductionist can in fact quotient further, by considering the action of Galilean boosts and identifying phase space points that differ by a boost; i.e. defining orbits transverse to these hypersurfaces.
##### 2.3.5.B Analogous reductions in other theories
I close my philosophers’ appetizer for symplectic reduction by summarizing some general remarks of Belot’s (2001: Sections VIII-IX); cf. also his (2003a, Sections 12, 13). They are about how our discussion of relational mechanics is typical of many cases; and how symplectic reduction can be physically important. I label them (1)-(3).
(1): A general contrast: when to quotient:—
The example of $`N`$ point-particles interacting by Newtonian gravity is typical of a large class of cases (infinite-dimensional, as well as finite-dimensional). There is a configuration space $`Q`$, acted on by a continuous group $`G`$ of symmetries, which lifts to the cotangent bundle $`T^{}Q`$, with the cotangent lift leaving invariant the Hamiltonian, and so the dynamics. So we can quotient $`T^{}Q`$ by $`G`$ to give a reduced theory. (There is a Lagrangian analogue; but as above, we set it aside.) But there is also some motivation for quotienting $`G`$’s action on $`Q`$, irrespective of how we then go on the construct dynamics. Let us adopt ‘relationism’ as a mnemonic label for whatever motivates quotienting the configuration space. Then with suitable technical conditions assumed (recall our excision of $`\delta `$ and $`\mathrm{\Delta }`$), we will have:
(i): for the reduced Hamiltonian theory: dim($`(T^{}Q)/G`$) = 2 dim$`Q`$ \- dim$`G`$;
(ii): for the relationist theory, in a Lagrangian or Hamiltonian framework:
dim$`(T(Q/G))`$ = dim$`(T^{}(Q/G))`$ = 2(dim$`Q`$ \- dim$`G`$)
So we have in the reduced theory, dim $`G`$ variables that do not occur in the relationist theory: let us call them ‘non-relational variables’.
(2): The non-relational variables:—
Typically, these non-relational variables represent global, i.e. collective, properties of the system. That is unsurprising since the number, dim $`G`$, of these variables is independent of the number of degrees of freedom of the system (dim $`Q`$, or 2dim $`Q`$ if you count rate of change degrees of freedom separately).
Some of these variables are conserved quantities, which arise (by Noether’s theorem) from the symmetries. Furthermore, there can be specific values of the conserved quantities, like the vanishing angular momentum of Section 2.3.4, for which the reduced theory collapses into the relationist theory. That is, not only are the relevant state spaces of equal dimension; but also their dynamics agree.
(3): The reduced theory:—
Typically, the topology and geometry of the reduced phase space $`(T^{}Q)/G`$, and the Hamiltonian function on it, $`\overline{H}:(T^{}Q)/G\mathrm{I}\mathrm{R}`$ say, are more complex than the corresponding features of the unreduced theory on $`T^{}Q`$. In particular, the reduced Hamiltonian $`\overline{H}`$ typically has potential energy terms corresponding to forces that are absent from the unreduced theory. But this should not be taken as necessarily a defect, for two reasons.
First, there are famous cases in which the reduced theory has a distinctive motivation. One example is Hertz’ programme in mechanics, viz. to “explain away” the apparent forces of our macroscopic experience (e.g. gravity) as arising from reduction of a theory that has suitable symmetries. (The programme envisaged cyclic variables for microscopic degrees of freedom that were unknown to us; cf. Lutzen (1995, 2005).) Another famous example is the Kaluza-Klein treatment of the force exerted on a charged particle by the electromagnetic field. That is: the familiar Lorentz force-law describing a charged particle’s motion in four spacetime dimensions can be shown to arise by symplectic reduction from a theory postulating a spacetime with a fifth (tiny and closed) spatial dimension, in which the particle undergoes straight-line motion. Remarkably, the relevant conserved quantity, viz. momentum along the fifth dimension, can be identified with electric charge; so that the theory can claim to explain the conservation of electric charge. (This example generalizes to other fields: for details and references, cf. Marsden and Ratiu (1999, Section 7.6).)
Second, the reduced theory need not be so complicated as to be impossible to work with. Indeed, these two examples prove this point, since in them the reduced theory is entirely tractable: for it is the familiar theory—that one might resist abandoning for the sake of the postulated unreduced theory.<sup>17</sup><sup>17</sup>17And here one should resist being prejudiced because of familiarity. Why not have Newtonian gravity arise from a microscopic cyclic degree of freedom? Why not have the Lorentz force law arise from geodesic motion in a five-dimensional spacetime with the fifth dimension wrapped up, so that conservation of charge is explained, in Noether’s theorem fashion, by a symmetry? Besides, Belot describes how, even when the reduced theory seems complicated (and not just because it is unfamiliar!), the general theory of symplectic reduction, as developed over the last forty years, has shown that one can often “do physics” in the reduced phase space: and that, as in the Kaluza-Klein example, the physics in the reduced phase space can be heuristically, as well as interpretatively, valuable.
## 3 Some geometric tools
So much by way of an appetizer. The rest of the Chapter, comprising this Section and the next four, is the five-course banquet! This Section expounds some modern differential geometry, especially about Lie algebras and Lie groups. Section 4 takes up actions by Lie groups. Then Section 5 describes Poisson manifolds as a generalized framework for Hamiltonian mechanics. As I mentioned in (2) of Section 2.2, Lie himself developed this framework; so in effect, he knew everything in these two Sections—so it is a true (though painful!) pun to say that these three Sections give us the “Lie of the land”. In any case, these two Sections will prepare us for Section 6’s description of symmetry and conservation in terms of momentum maps. Finally, Section 7 will present one of the main theorems about symplectic reduction. It concerns the case where the natural configuration space for a system is itself a Lie group $`G`$; (cf. (3) of Section 2.2). Quotienting the natural phase space (the cotangent bundle on $`G`$) will give a Poisson manifold that “is” the dual of $`G`$’s Lie algebra.
In this Section, I first sketch some notions of differential geometry, and fix notation (Section 3.1). Then I introduce Lie algebras and Lie brackets of vector fields (Section 3.2). Though most of this Section (indeed this Chapter!) is about differential rather than integral notions, I will later need Frobenius’ theorem, which I present in Section 3.3. Then I give some basic information about Lie groups and their Lie algebras (Section 3.4).
### 3.1 Vector fields on manifolds
#### 3.1.1 Manifolds, vectors, curves and derivatives
By way of fixing ideas and notation, I begin by giving details about some ideas in differential geometry (some already used in Section 2.1), and introducing some new notation for them.
A manifold $`M`$ will be finite-dimensional, except for obvious and explicit exceptions such as the infinite-dimensional group of diffeomorphisms of a (as usual: finite-dimensional!) manifold. I will not be concerned about the degree of differentiability in the definition of a manifold, or of any associated geometric objects: ‘smooth’ can be taken throughout what follows to mean $`C^{\mathrm{}}`$. I will often not be concerned with global, as against local, structures and results; (though the reduction results we are driving towards are global in nature). For example, I will not be concerned about whether curves are inextendible, or flows are complete.
I shall in general write a vector at a point $`xM`$ as $`X`$; or in terms of local coordinates $`x^i`$, as $`X=X^i\frac{}{x^i}`$ (summation convention). From now on, I shall write the tangent space at a point $`xM`$ as $`T_xM`$ (rather than just $`T_x`$), thus explicitly indicating the manifold $`M`$ to which it is tangent. As before, I write the tangent bundle, consisting of the “meshing collection” of these tangent spaces, as $`TM`$. Similarly, I write a 1-form (covector) at a point $`xM`$ as $`\alpha `$; and so the cotangent space at $`xM`$ as $`T_x^{}M`$; and as before, the cotangent bundle as $`T^{}M`$.
A smooth map $`f:MN`$ between manifolds $`M`$ and $`N`$ (maybe $`N=M`$) maps smooth curves to smooth curves, and so tangent vectors to tangent vectors; and so on for 1-forms and higher tensors. It is convenient to write $`Tf`$, called the derivative or tangent of $`f`$ (also written as $`f_{}`$ or $`df`$ or $`Df`$), for the induced map on the tangent bundle.
In more detail: let us take a curve $`c`$ in $`M`$ to be a smooth map from an interval $`I\mathrm{I}\mathrm{R}`$ to $`M`$, and a tangent vector at $`xM`$, $`XT_xM`$, to be an equivalence class $`[c]_x`$ of curves through $`x`$. (The equivalence relation is that the curves be tangent at $`x`$, with respect to every local chart at $`x`$; but I omit the details of this.) Then we define $`Tf:TMTN`$ (also written $`f_{}:TMTN`$) by
$$f_{}([c]_x)Tf([c]_x):=[fc]_{f(x)},\mathrm{for}\mathrm{all}xM.$$
(3.29)
We sometimes write $`T_xf`$ for the restriction of $`Tf`$ to just the tangent space $`T_xM`$ at $`x`$; i.e.
$$T_xf:[c]_xT_xM[fc]_{f(x)}T_{f(x)}N.$$
(3.30)
In Section 3.1.2.B, we will discuss how one can instead define tangent vectors to be differential operators on the set of all scalar functions defined in some neighbourhood of the point in question, rather than equivalence classes of curves. One can then define the tangent map $`f_{}Tf`$ in a way provably equivalent to that above.
#### 3.1.2 Vector fields, integral curves and flows
We will be especially concerned with vector fields defined on $`M`$, i.e. $`X:xMX(x)T_xM`$, or on a subset $`UM`$. So suppose that $`X`$ is vector field on $`M`$ and $`f:MN`$ is a smooth map, so that $`T_xf:T_xMT_{f(x)}N`$.
##### 3.1.2.A Push-forwards and pullbacks
It is important to note $`(T_xf)(X(x))`$ does not in general define a vector field on $`N`$. For $`f(M)`$ may not be all of $`N`$, so that for $`y(N\mathrm{ran}(f))`$ $`(T_xf)(X(x))`$ assigns no element of $`T_yN`$. And $`f`$ may not be injective, so that we could have $`x,x^{}M`$ and $`f(x)=f(x^{})`$ with $`(T_xf)(X(x))(T_x^{}f)(X(x^{}))`$. Thus we say that vector fields do not push forward.
On the other hand, suppose that $`f:MN`$ is a diffeomorphism onto $`N`$: that is, the smooth map $`f`$ is a bijection, and its inverse $`f^1`$ is also smooth. Then for any vector field $`X`$ on $`M`$, $`Tf(X)`$ is a vector field on $`N`$. So in this case, the vector field does push forward. Accordingly, $`Tf(X)`$ is called the push-forward of $`X`$; it is often written as $`f_{}(X)`$. So for any $`xM`$, the pushed forward vector field at the image point $`f(x)`$ is given by
$$(f_{}(X))(f(x)):=T_xfX(x).$$
(3.31)
(Note the previous use of the asterisk-subscript for the derivative of $`f`$, in eq. 3.29.)
This prompts three more general comments.
(1): More generally: we say that two vector fields, $`X`$ on $`M`$ and $`Y`$ on $`N`$, are $`f`$-related on $`M`$ (respectively: on $`SM`$) if $`(Tf)(X)=Y`$ at all $`xM`$ (respectively: $`xS`$).
(2): We can generalize the idea that a diffeomorphism implies that a vector field can be pushed forward, in two ways. First, the diffeomorphism need only be defined locally, on some neighbourhood of the point $`xM`$ of interest. Second, a diffeomorphism establishes a one-one correspondence, not just between vector fields defined on its domain and codomain, but also between all differential geometric objects defined on its domain and codomain: in particular, 1-form fields, and higher rank tensors.
(3): (This continues comment (2).) Though vector fields do not in general push forward, 1-form fields do in general pull back. This is written with an asterisk-superscript. That is: for any smooth $`f:MN`$, not necessarily a diffeomorphism (even locally), and any 1-form field (differential 1-form) $`\alpha `$ on $`N`$, we define the pullback $`f^{}(\alpha )`$ to be the 1-form on $`M`$ whose action, for each $`xM`$, and each $`XT_xM`$, is given by:
$$(f^{}(\alpha ))(X):=\alpha _{f(x)}(Tf(X)).$$
(3.32)
Similarly, of course if the map $`f`$ is defined only locally on a subset of $`M`$: a 1-form defined on the range of $`f`$ pulls back to a 1-form on the domain of $`f`$.
##### 3.1.2.B The correspondence between vector fields and flows
The leading idea about vector fields is that, for any manifold, the theorems on the local existence, uniqueness and differentiability of solutions of systems of ordinary differential equations (e.g. Arnold (1973: 48-49, 77-78, 249-250), Olver (2000: Prop 1.29)) secure a one-one correspondence between four notions:
(i): Vector fields $`X`$ on a subset $`UM`$, on which they are non-zero; $`X:xUX(x)T_xM,X(x)0`$;
(ii): Non-zero directional derivatives at each point $`xU`$, in the direction of the vector $`X(x)`$. In terms of coordinates $`𝐱=x^1,\mathrm{},x^n`$, these are first-order linear differential operators $`X^1(𝐱)\frac{}{x^1}+\mathrm{}+X^n(𝐱)\frac{}{x^n}`$, with $`X^i(𝐱)`$ the $`i`$-component in this coordinate system of the vector $`X(x)`$. Such an operator is often introduced abstractly as a derivation: a map on the set of smooth real-valued functions defined on a neighbourhood of $`x`$, that is linear and obeys the Leibniz rule.
(iii): Integral curves (aka: solution curves) of the fields $`X`$ in $`U`$; i.e. smooth maps $`\varphi :IM`$ from a real open interval $`I\mathrm{I}\mathrm{R}`$ to $`U`$, with $`0I`$, $`\varphi (0)=xU`$, and whose tangent vector at each $`\varphi (\tau ),\tau I`$ is $`X(\varphi (\tau )).`$
(iv): Flows $`X^\tau `$ mapping, for each field $`X`$ and each $`xU`$, some appropriate subset of $`U`$ to another: $`X^\tau :UM`$. This flow is guaranteed to exist only in some neighbourhood of a given point $`x`$, and for $`\tau `$ in some neighbourhood of $`0\mathrm{I}\mathrm{R}`$; but this will be enough for us. Such a flow is a one-parameter subgroup of the “infinite-dimensional group” of all local diffeomorphisms.
I spell out this correspondence in a bit more detail:— In local coordinates $`x^1,\mathrm{},x^n`$, any smooth curve $`\varphi :IM`$ is given by $`n`$ smooth functions $`\varphi (\tau )=(\varphi ^1(\tau ),\mathrm{},\varphi ^n(\tau ))`$, and the tangent vector to $`\varphi `$ at $`\varphi (\tau )M`$ is
$$\dot{\varphi }(\tau )=\dot{\varphi }^1(\tau )\frac{}{x^1}+\mathrm{}+\dot{\varphi }^n(\tau )\frac{}{x^n}.$$
(3.33)
So for $`\varphi `$ to be an integral curve of $`X`$ requires that for all $`i=1,\mathrm{},n`$ and all $`\tau I`$
$$\dot{\varphi }^i(\tau )=X^i(\tau ).$$
(3.34)
The local existence and uniqueness, for a given vector field $`X`$ and $`xM`$, of the integral curve $`\varphi _{X,x}`$ through $`x`$ (with $`\varphi (0)=x`$) then ensures that the flow, written either as $`X^\tau `$ or as $`\varphi _X(\tau )`$
$$X^\tau :xMX^\tau (x)\varphi _{X,x}(\tau )M,$$
(3.35)
is (at least locally) well-defined. The flow is a one-parameter group of transformations of $`M`$, and $`X`$ is called its infinitesimal generator.
The exponential notation
$$\mathrm{exp}(\tau X)(x):=X^\tau (x)\varphi _{X,x}(\tau )$$
(3.36)
is suggestive. For example, the group operation in the flow, i.e.
$$X^{\tau +\sigma }(x)=X^\tau (X^\sigma (x)),$$
(3.37)
is written in the suggestive notation
$$\mathrm{exp}((\tau +\sigma )X)(x)=\mathrm{exp}(\tau X)(\mathrm{exp}(\sigma X)(x)).$$
(3.38)
So computing the flow for a given $`X`$ (i.e. solving a system of $`n`$ first-order differential equations!) is called exponentiation of the vector field $`X`$.
Remark:— The above correspondence can be related to our discussion of diffeomorphisms and pushing forward vector fields. In particular: if two vector fields, $`X`$ on $`M`$ and $`Y`$ on $`N`$, are $`f`$-related by $`f:MN`$, so that $`(Tf)(X(x))=Y(f(x))`$, then $`f`$ induces a map from integral curves of $`X`$ to integral curves of $`Y`$. We can express this in terms of exponentiation of $`X`$ and $`Y=(Tf)(X)`$:
$$f(\mathrm{exp}(\tau X)x)=\mathrm{exp}(\tau (Tf)(X))(f(x)).$$
(3.39)
Remark:— I emphasise that the above correspondence between (i), (ii), (iii) and (iv) is not true at a single point. More precisely:
(a): On the one hand: the correspondence between (i) and (ii) holds at a point; and also holds for zero vectors. That is: a single vector $`XT_xM`$ corresponds to a directional derivative operator (derivation) at $`x`$; and $`X=0`$ corresponds to the zero derivative operator mapping all local scalars to 0. (Indeed, as I mentioned: vectors are often defined as such operators/derivations). But:
(b): On the other hand: the correspondence between (i) and (iii), or between (i) and (iv), requires a neighbourhood. For a single vector $`XT_xM`$ corresponds to a whole class of curves (and so: of flows) through $`x`$, not to a single curve. Namely, it corresponds to all the curves (flows) with $`X`$ as their tangent vector.
However, we shall see (starting in Section 3.4) that for a manifold with suitable extra structure, a single vector does determine a curve. (And we will again talk of exponentiation.)
We need to generalize one aspect of the above correspondence (i)-(iv), namely the (i)-(ii) correspondence between vectors and directional derivatives. This generalization is the Lie derivative.
#### 3.1.3 The Lie derivative
Some previous Sections have briefly used the Lie derivative. Since we will use it a lot in the sequel, we now introduce it more thoroughly.
We have seen that given a vector field $`X`$ on a manifold $`M`$, a point $`xM`$, and any scalar function $`f`$ defined on a neighbourhood of $`x`$, there is a naturally defined rate of change of $`f`$ along $`X`$ at $`x`$: the directional derivative $`X(x)(f)`$.
Now we will define the Lie derivative along $`X`$ as an operator $`_X`$ that defines a rate of change along $`X`$: not only for locally defined functions (for which the definition will agree with our previous notion, i.e. we will have $`_X(f)=X(f)`$); but also for vector fields and differential 1-forms.<sup>18</sup><sup>18</sup>18Indeed, the definition can be extended to all higher rank tensors. But I will not develop those details, since—apart from Section 2.1.3’s mention of the Lie derivative of the symplectic form $`_X\omega `$ (viz. the requirement that if $`X`$ is a symmetry, $`_X\omega =0`$)—we shall not need them. We proceed in three stages.
(1): We first define the Lie derivative as an operator on scalar functions, in terms of the vector field $`X`$ on $`M`$. We define the Lie derivative along the field $`X`$ (aka: the derivative in the direction of $`X`$), $`_X`$, as the operator on scalar functions $`f:M\mathrm{I}\mathrm{R}`$ defined by:
$$_X:f_Xf:M\mathrm{I}\mathrm{R}\mathrm{with}xM:(_Xf)(x):=\frac{d}{d\tau }_{\tau =0}f(X^\tau (x))X(x)(f).$$
(3.40)
Though this definition assumes that both $`X`$ and $`f`$ are defined globally, i.e. on all of $`M`$, it can of course be restricted to a neighbourhood. Thus defined, $`_X`$ is linear and obeys the Leibniz rule, i.e.
$$_X(fg)=f_X(g)+g_X(f);$$
(3.41)
In coordinates $`𝐱=x^1,\mathrm{},x^n`$, $`_Xf`$ is given by
$$_Xf=X^1(𝐱)\frac{f}{x^1}+\mathrm{}+X^n(𝐱)\frac{f}{x^n},$$
(3.42)
with $`X^i(𝐱)`$ the $`i`$-component of the vector $`X(x)`$. Eq. 3.42 means that despite eq. 3.40’s mention of the flow $`X^\tau `$, the Lie derivative of a scalar agrees with our previous notion of directional derivative: that is, for all $`f`$, $`_X(f)=X(f)`$.
(2): In (1), the vector field $`X`$ determined the operator $`_X`$: in terms of Section 3.1.2.B’s correspondence, we moved from (i) to (ii). But we can conversely define a vector field in terms of its Lie derivative; and in Section 3.2.2’s discussion of the Lie bracket, we shall do exactly this.
In a bit more detail:— We note that the set $`(M)`$ of all scalar fields on $`M`$, $`f:M\mathrm{I}\mathrm{R}`$ forms an (infinite-dimensional) real vector space under pointwise addition. So also does the set $`𝒳(M)`$ of all vector fields on $`M`$, $`X:xMX(x)T_xM`$. Furthermore, $`𝒳(M)`$ is isomorphic as a real vector space, and as an module over the scalar fields, to the collection of operators $`_X`$. The isomorphism is given by the map $`\theta :X_X`$ defined in (1).
(3): We now extend the definition of $`_X`$ so as to define it on vector fields $`Y`$ and 1-forms $`\alpha `$. We can temporarily use $`\theta `$ as notation for either a vector field $`Y`$ or a differential 1-form $`\alpha `$. Given a vector field $`X`$ and flow $`X^\tau \varphi _X(\tau )`$, we need to compare $`\theta `$ at the point $`xM`$ with $`\theta `$ at the nearby point $`X^\tau (x)\varphi _{X,x}(\tau )`$, in the limit as $`\tau `$ tends to zero. But the value of $`\theta `$ at $`X^\tau (x)`$ is in the tangent space, or cotangent space, at $`X^\tau (x)`$: $`T_{X^\tau (x)}M`$ or $`T_{X^\tau (x)}^{}M`$. So to make the comparison, we need to somehow transport back this value to $`T_xM`$ or $`T_x^{}M`$.
Fortunately, the vector field $`X`$ provides a natural way to define such a transport. For the vector field $`Y`$, we use the differential (i.e. push-forward) of the inverse flow, to “get back” from $`X^\tau (x)`$ to $`x`$. Using $`\varphi ^{}(\tau )`$ for this “pullback” of $`\varphi _{X,x}(\tau )`$, we define
$$\varphi ^{}(\tau ):=T(\mathrm{exp}(\tau X))d\mathrm{exp}(\tau X):T_{X^\tau (x)}MT_{\mathrm{exp}(\tau X)(x)}MT_xM.$$
(3.43)
For the 1-form $`\alpha `$, we define the transport by the pullback, already defined by eq. 3.32:
$$\varphi ^{}(\tau ):=(\mathrm{exp}(\tau X))^{}:T_{X^\tau (x)}^{}MT_{\mathrm{exp}(\tau X)(x)}^{}MT_x^{}M.$$
(3.44)
With these definitions of $`\varphi ^{}(\tau )`$, we now define the Lie derivative $`_X\theta `$, where $`\theta `$ is a vector field $`Y`$ or a differential 1-form $`\alpha `$, as the vector field or differential 1-form respectively, with value at $`x`$ given by
$$\underset{\tau 0}{lim}\frac{\varphi ^{}(\tau )(\theta _{X^\tau (x)})\theta _x}{\tau }=\frac{d}{d\tau }_{\tau =0}\varphi ^{}(\tau )(\theta _{X^\tau (x)}).$$
(3.45)
Finally, an incidental result to illustrate this Chapter’s “story so far”. It connects Noether’s theorem, from Section 2.1.3, to this Section’s details about the Lie derivative, and to the theorem stating the local existence and uniqueness of solutions of ordinary differential equations (cf. the start of Section 3.1.2.B). This latter theorem implies that on any manifold any vector field $`X`$ can be “straightened out”, in the sense that around any point at which $`X`$ is non-zero, there is a local coordinate system in which $`X`$ has all but one component vanish and the last component equal to 1. Using this theorem, it is straightforward to show that on any even-dimensional manifold any vector field $`X`$ is locally Hamiltonian, with respect to some symplectic form, around a point where $`X`$ is non-zero. One just defines the symplectic form by Lie-dragging from a surface transverse to $`X`$’s integral curves.
### 3.2 Lie algebras and brackets
I now introduce Lie algebras and the Lie bracket of two vector fields.
#### 3.2.1 Lie algebras
A Lie algebra is a vector space $`V`$ equipped with a bilinear anti-symmetric operation, usually denoted by square brackets (and called ‘bracket’ or ‘commutator’), $`[,]:V\times VV`$, that satisfies the Jacobi identity, i.e.
$$[[X,Y],Z]+[[Y,Z],X]+[[Z,X],Y]=0.$$
(3.46)
##### 3.2.1.A Examples; rotations introduced
Here are three examples.
(i): $`n\times n`$ matrices equipped with the usual commutator, i.e. $`[X,Y]:=XYYX`$. (So the matrix multiplication “contributes” to the bracket, but not to the underlying vector space structure.)
(ii): $`3\times 3`$ anti-symmetric matrices, equipped with the usual commutator.
(iii): $`\mathrm{I}\mathrm{R}^3`$ equipped with vector multiplication. In fact, example (iii) is essentially the same as example (ii); and this example will recur in what follows, in connection with rotations and the rigid body. (We will also see that example (ii) is in a sense more fundamental.)
To explain this, we first recall that every anti-symmetric operator $`A`$ on a three-dimensional oriented euclidean space is the operator of vector multiplication by a fixed vector, $`\omega `$ say. That is: for all $`𝐪,A𝐪=[\omega ,𝐪]\omega 𝐪`$. (Proof: the anti-symmetric operators on $`\mathrm{I}\mathrm{R}^3`$ for a 3-dimensional vector space, since an anti-symmetric $`3\times 3`$ matrix has three independent components. Vector multiplication by a vector $`\omega `$ is a linear and anti-symmetric operator; varying $`\omega `$ we get a subspace of the space of all anti-symmetric operators on $`\mathrm{I}\mathrm{R}^3`$; but this subspace has dimension 3; so it coincides with the space of all anti-symmetric operators.)
With this result in hand, the following three points are all readily verified.
(1): The matrix representation of $`A`$ in cartesian coordinates is then
$$A=\left(\begin{array}{ccc}0& \omega _3& \omega _2\\ \omega _3& 0& \omega _1\\ \omega _2& \omega _1& 0\end{array}\right).$$
(3.47)
We can write
$$A\omega \mathrm{or}A_{ij}=ϵ_{ijk}\omega _k\mathrm{or}\omega _i=\frac{1}{2}ϵ_{ijk}A_{jk}.$$
(3.48)
(2): The plane $`\mathrm{\Pi }`$ of vectors perpendicular to $`\omega `$ is an invariant subspace for $`A`$, i.e. $`A(\mathrm{\Pi })=\mathrm{\Pi }`$. And $`\omega `$ is an eigenvector for $`A`$ with eigenvalue 0. This suggest a familiar elementary interpretation, which will be confirmed later (Section 3.4): viz. that any $`3\times 3`$ anti-symmetric matrix $`A`$ represents a infinitesimal rotation, and $`\omega `$ represents instantaneous angular velocity. That is, we will have, for all $`𝐪\mathrm{I}\mathrm{R}^3`$: $`\dot{𝐪}=A𝐪=[\omega ,𝐪]`$.
(3): The commutator of any two $`3\times 3`$ anti-symmetric matrices $`A,B`$, i.e. $`[A,B]:=ABBA`$, corresponds by eq. 3.48 to vector multiplication of the axes of rotation. That is: writing eq. 3.48’s bijection from vectors to matrices as $`\mathrm{\Theta }:\omega A=:\mathrm{\Theta }(\omega )`$, we have for vectors $`𝐪,𝐫,𝐬`$
$`(\mathrm{\Theta }(𝐪)\mathrm{\Theta }(𝐫)\mathrm{\Theta }(𝐫)\mathrm{\Theta }(𝐪))𝐬=\mathrm{\Theta }(𝐪)[𝐫,𝐬]\mathrm{\Theta }(𝐫)[𝐪,𝐬]`$ (3.49)
$`=[𝐪,[𝐫,𝐬]][𝐫,[𝐪,𝐬]]`$ (3.50)
$`=[[𝐪,𝐫],𝐬]=\mathrm{\Theta }([𝐪,𝐫])𝐬.`$ (3.51)
where the \[,\] represents vector multiplication, i.e. $`[𝐪,𝐫]𝐪𝐫`$.
Eq. 3.51 means that $`\mathrm{\Theta }`$ gives a Lie algebra isomorphism; and so our example (iii) is essentially the same as example (ii).
Besides, we can already glimpse why example (ii) is in a sense more fundamental. For this correspondence between anti-symmetric operators (or matrices) and vectors, eq. 3.48, is specific to three dimensions. In $`n`$ dimensions, the number of independent components of an anti-symmetric matrix is $`n(n1)/2`$: only for $`n=3`$ is this equal to $`n`$. Yet we will see later (Section 3.4.4) that rotations on euclidean space $`\mathrm{I}\mathrm{R}^n`$ of any dimension $`n`$ are generated, in a precise sense, by the Lie algebra of $`n\times n`$ anti-symmetric matrices. So only for $`n=3`$ is there a corresponding representation of rotations by vectors in $`\mathrm{I}\mathrm{R}^n`$.
In the next two Subsections, we shall see other examples of Lie algebras: whose vectors are vector fields (Section 3.2.2), or tangent vectors at the identity element of a Lie group (Section 3.4). The first example will be an infinite-dimensional Lie algebra; the second finite-dimensional (since we will only consider finite-dimensional Lie groups). Besides, the above examples (i) and (ii) (equivalently: (i) and (iii)) will recur: each will be the vector space of tangent vectors at the identity element of a Lie group.
##### 3.2.1.B Structure constants
A finite-dimensional Lie algebra is characterized, relative to a basis, by a set of numbers, called structure constants that specify the bracket operation. Thus if $`\{v_1,\mathrm{},v_n\}`$ is a basis of a Lie algebra $`V`$, we define the structure constants $`c_{ij}^k,(i,j,k=1,\mathrm{},n)`$ by expanding, in terms of this basis, the bracket of any two basis elements
$$[v_i,v_j]=\mathrm{\Sigma }_kc_{ij}^kv_k.$$
(3.52)
The bilinearity of the bracket implies that eq. 3.52 determines the bracket of all pairs of vectors $`v,wV`$. And the bracket’s obeying anti-symmetry and the Jacobi identity implies that, for any basis, the structure constants obey
$$c_{ij}^k=c_{ji}^k;\mathrm{\Sigma }_k(c_{ij}^kc_{kl}^m+c_{li}^kc_{kj}^m+c_{jl}^kc_{ki}^m)=0$$
(3.53)
Conversely, any set of constants $`c_{ij}^k`$ obeying eq. 3.53 are the structure constants of an $`n`$-dimensional Lie algebra.
#### 3.2.2 The Lie bracket of two vector fields
Given two vector fields $`X,Y`$ on a manifold $`M`$, the corresponding flows do not in general commute: $`X^tY^sY^sX^t`$. The non-commutativity is measured by the commutator of the Lie derivatives of $`X`$ and of $`Y`$, i.e. $`_X_Y_Y_X`$. (Cf. eq. 3.40 and 3.45 for a definition of the Lie derivative.) Here, ‘measured’ can be made precise by considering Taylor expansions; but I shall not go into detail about this.
What matters for us is that this commutator, which is at first glance seems to be a second-order operator, is in fact a first-order operator. This is verified by calculating in a coordinate system, and seeing that the second derivatives occur twice with opposite signs:
$`(_X_Y_Y_X)f=\mathrm{\Sigma }_iX^i{\displaystyle \frac{}{x^i}}\left(\mathrm{\Sigma }_jY^j{\displaystyle \frac{f}{x^j}}\right)\mathrm{\Sigma }_jY^j{\displaystyle \frac{}{x^j}}\left(\mathrm{\Sigma }_iX^i{\displaystyle \frac{f}{x^i}}\right)`$ (3.54)
$`=\mathrm{}=\mathrm{\Sigma }_{i,j}\left(X^i{\displaystyle \frac{Y^j}{x^i}}Y^i{\displaystyle \frac{X^j}{x^i}}\right){\displaystyle \frac{f}{x^j}}.`$ (3.55)
So $`_X_Y_Y_X`$ corresponds to a vector field: (recall (2) of Section 3.1.3, about defining a vector field from its Lie derivative). We call this field $`Z`$ the Lie bracket (also known as: Poisson bracket, commutator, and Jacobi-Lie bracket!) of the fields $`X`$ and $`Y`$, and write it as $`[X,Y]`$. It is also written as $`_XY`$ and called the Lie derivative of $`Y`$ with respect to $`X`$. (Beware: some books use an opposite sign convention.)
Thus $`Z[X,Y]_XY`$ is defined to be the vector field such that
$$_Z_{[X,Y]}=_X_Y_Y_X.$$
(3.56)
It follows that $`Z[X,Y]`$’s components in a coordinate system are given by eq. 3.55. This formula can be remembered by writing it (with summation convention, i.e. omitting the $`\mathrm{\Sigma }`$) as
$$[X^i\frac{}{x^i},Y^j\frac{}{x^J}]=X^i\frac{Y^j}{x^i}\frac{}{x^j}Y^j\frac{X^i}{x^j}\frac{}{x^i}$$
(3.57)
Another way to write eq. 3.55 is as:
$$[X,Y]^j=(X)Y^j(Y)X^j;$$
(3.58)
or without coordinates, writing $`𝐃`$ for the derivative map given by the Jacobian matrix, as
$$[X,Y]=𝐃YX𝐃XY.$$
(3.59)
Again, the vector field $`Z[X,Y]`$ measures the non-commutation of the flows $`X^t`$ and $`Y^s`$: in particular, these flows commute iff $`[X,Y]=0`$.
We will need three results about the Lie bracket. They concern, respectively, the relation to Lie algebras, to Poisson brackets, and to Frobenius’ theorem.
(1): The Lie bracket is obviously a bilinear and anti-symmetric operation on the (infinite-dimensional) vector space $`𝒳(M)`$ of all vector fields on $`M`$: $`[,]:𝒳(M)\times 𝒳(M)𝒳(M)`$. One readily checks that it satisfied the Jacobi identity. (Expand $`_{[[X,Y],Z]}=_{[X,Y]}_Z_Z_{[X,Y]}`$ etc.) So: $`𝒳(M)`$ is an (infinite-dimensional) Lie algebra.
(2): Returning to Hamiltonian mechanics (Section 2.1): there is a simple and fundamental relation between the Lie bracket and the Poisson bracket, via the notion of Hamiltonian vector fields (Section 2.1.3).
Namely: the Hamiltonian vector field of the Poisson bracket of two scalar functions $`f,g`$ on the symplectic manifold $`M`$ is, upto a sign, the Lie bracket of the Hamiltonian vector fields, $`X_f`$ and $`X_g`$, of $`f`$ and $`g`$:
$$X_{\{f,g\}}=[X_f,X_g]=[X_g,X_f].$$
(3.60)
Proof: apply the rhs to an arbitrary scalar $`h:M\mathrm{I}\mathrm{R}`$. One easily obtains $`X_{\{f,g\}}(h)`$, by using:
(i) the definition of a Hamiltonian vector field;
(ii) the Lie derivative of a function equals its elementary directional derivative eq. 3.40; and
(iii) the Poisson bracket is antisymmetric and obeys the Jacobi identity.
This result means that the Hamiltonian vector fields on a symplectic manifold $`M`$, equipped with the Poisson bracket, form an (infinite-dimensional) Lie subalgebra of the Lie algebra $`𝒳(M)`$ of all vector fields on the symplectic manifold $`M`$. Later, it will be important that this result extends from symplectic manifolds to Poisson manifolds; (details in Section 5.2.2).
(3): For Frobenius’ theorem (Section 3.3), we need to relate the Lie bracket to Section 3.1.2’s idea of vector fields being $`f`$-related by a map $`f:MN`$ between manifolds $`M`$ and $`N`$. In short: if two pairs of vector fields are $`f`$-related, so is their Lie bracket. More explicitly: if $`X,Y`$ are vector fields on $`M`$, and $`f:MN`$ is a map such that $`(Tf)(X),(Tf)(Y)`$ are well-defined vector fields on $`N`$, then $`Tf`$ commutes with the Lie bracket:
$$(Tf)[X,Y]=[(Tf)X,(Tf)Y].$$
(3.61)
### 3.3 Submanifolds and Frobenius’ theorem
This Subsection differs from the preceding ones in three ways. First, it emphasises integral, rather than differential, notions.
Second: Section 3.1.2.B have emphasised that the integral curves of a vector field correspond to integrating a system of ordinary differential equations. Since such curves are one-dimensional submanifolds of the given manifold, our present topic, viz. higher-dimensional submanifolds, naturally suggests partial differential equations. For their integration involves finding, given an assignment to each point $`x`$ of a manifold $`M`$ of a subspace $`S_x`$ (with dimension greater than one) of the tangent space $`T_xM`$, an integral surface, i.e. a submanifold $`S`$ of $`M`$ whose tangent space at each of its points is $`S_x`$.<sup>19</sup><sup>19</sup>19Beware: there is no analogue for partial differential equations of the local existence and uniqueness theorem for ordinary differential equations. Even a field of two-dimensional planes in three-dimensional space is in general not integrable, e.g. the field of planes given by the equation $`dz=ydx`$. So integrable fields of planes, or other tangent subspaces on a manifold, are an exception; and accordingly, the integration theory for partial differential equations is less unified, and more complicated, than that for ordinary differential equations.
However, we will not be concerned with partial differential equations. For us, submanifolds of dimension higher than one arise when the span $`S_x`$ of the tangent vectors at $`x`$ to a set of vector fields fit together to form a submanifold. Thus Frobenius’ theorem states, roughly speaking, that a finite set of vector fields is integrable in this sense iff the vector fields are in involution. That is: iff their pairwise Lie brackets are expandable in terms of the fields; i.e. the vector fields form a Lie subalgebra of the entire Lie algebra of vector fields. We will not need to prove this theorem. But we need to state it and use it—in particular, for the foliation of Poisson manifolds.
Third: a warning is in order. The intuitive idea of a subset $`SM`$ that is a smooth manifold “in its own right” can be made precise in different ways. So there are subtleties about the definition of ‘submanifold’, and terminology varies between expositions—in a way it does not for the material in previous Sections. I will adopt what seems to be a widespread, if not majority, terminology.<sup>20</sup><sup>20</sup>20My treatment is based on Marsden and Ratiu (1999, p. 124-127, 140) for Section 3.3.1, and Olver (2000, p. 38-40) for Section 3.3.2. As to varying terminology: Olver (2000, p. 9) defines ‘submanifold’ to be what we will call an immersed submanifold; (which latter, for us, does not have to be a submanifold, since the immersion need not be an embedding). Bishop and Goldberg (1980, p. 40-41) provide a similar example. For a detailed introduction to the different notions of submanifold, cf. Darling (1994, Chapters 3 and 5). Note that I will also omit some details, in particular about Frobenius’ theorem providing regular immersions.
#### 3.3.1 Submanifolds
The fundamental definition is:
Given a manifold $`M`$ (dim($`M`$)=$`n`$), a submanifold of $`M`$ of dimension $`k`$ is a subset $`NM`$ such that for every $`yN`$ there is an admissible local chart (i.e. a chart in $`M`$’s maximal atlas) $`(U,\varphi )`$ with $`yU`$ and with the submanifold property, viz.
$$(\mathrm{SM}).\varphi :U\mathrm{I}\mathrm{R}^k\times \mathrm{I}\mathrm{R}^{nk}\mathrm{and}\varphi (UN)=\varphi (U)(\mathrm{I}\mathrm{R}^k\times \{\mathrm{𝟎}\}).$$
(3.62)
The set $`N`$ becomes a manifold, generated by the atlas of all charts of the form $`(UN,\varphi (UN))`$, where $`(U,\varphi )`$ is a chart of $`M`$ having the submanifold property. (This makes the topology of $`N`$ the relative topology.)
We need to take note of two ways in which submanifolds can be specified in terms of smooth functions between manifolds.
(1): A submanifold can be specified as the set on which a smooth function $`f:MP`$ between manifolds takes a certain value. In effect, this will be a generalization of eq. 3.62’s requirement that $`nk`$ coordinate-components of a chart $`\varphi `$ take the value zero. This will involve the idea that the tangent map $`Tf`$ is surjective, in which case $`f`$ will be called a submersion. We will need this approach for quotients of actions of Lie groups.
(2): A submanifold can be specified parametrically, as the set of values of a local parametrization: i.e. as the range of a smooth function $`f`$ with $`M`$ as codomain. This will involve the idea that the tangent map $`Tf`$ is injective, in which case $`f`$ will be called an immersion. We will need this approach for Frobenius’ theorem.
(1): Submersions:—
If $`f:MP`$ is a smooth map between manifolds, a point $`xM`$ is called a regular point if the tangent map $`T_xf`$ is surjective; otherwise $`x`$ is a critical point of $`f`$. If $`CM`$ is the set of critical points of $`M`$, we say $`f(C)`$ is the set of critical values of $`f`$, and $`Pf(C)`$ is the set of regular values of $`f`$. So if $`pP`$ is a regular value of $`f`$, then at every $`xM`$ with $`f(x)=p`$, $`T_xf`$ is surjective.
The submersion theorem states that if $`pP`$ is a regular value of $`f`$, then:
(i): $`f^1(p)`$ is a submanifold of $`M`$ of dimension dim($`M`$) - dim($`P`$); and
(ii): the tangent space of this submanifold at any point $`xf^1(p)`$ is the kernel of $`f`$’s tangent map:
$$T_x(f^1(p))=\mathrm{ker}T_xf.$$
(3.63)
If $`T_xf`$ is surjective for every $`xM`$, $`f`$ is called a submersion.
(2): Immersions:—
A smooth map between manifolds $`f:MP`$ is called an immersion if $`T_xf`$ is injective at every $`xM`$. The immersion theorem states that $`T_xf`$ is injective iff there is a neighbourhood $`U`$ of $`x`$ in $`M`$ such that $`f(U)`$ is a submanifold of $`P`$ and $`f_U:Uf(U)`$ is a diffeomorphism.
NB: This does not say that $`f(M)`$ is a submanifold of $`P`$. For $`f`$ may not be injective (so that $`f(M)`$ has self-intersections). And even if $`f`$ is injective, $`f`$ can fail to be a homeomorphism between $`M`$ and $`f(M)`$, equipped with the relative topology induced from $`P`$. A standard simple example is an injection of an open interval of $`\mathrm{I}\mathrm{R}`$ into an “almost-closed” figure-of-eight in $`\mathrm{I}\mathrm{R}^2`$.
Nevertheless, when $`f:MP`$ is an immersion, and is also injective, we call $`f(M)`$ an injectively immersed submanifold (or shorter: an immersed submanifold): though $`f(M)`$ might not be a submanifold.
We also define an embedding to be an immersion that is also a homeomorphism (and so injective) between $`M`$ and $`f(M)`$ (where the latter has the relative topology induced from $`P`$). If $`f`$ is an embedding, $`f(M)`$ is a submanifold of $`N`$ and $`f`$ is a diffeomorphism $`f:Mf(M)`$.
In fact, Frobenius’ theorem will provide injectively immersed submanifolds that need not be embedded, and so need not be submanifolds. (They must also obey another condition, called ‘regularity’, that I will not go into.)
#### 3.3.2 The theorem
We saw at the end of Section 3.2.2 that if two pairs of vector fields are $`f`$-related, so is their Lie bracket: cf. eq. 3.61. This result immediately yields a necessary condition for two vector fields to be tangent to an embedded submanifold: namely
If $`X_1,X_2`$ are vector fields on $`M`$ that are tangent to an embedded submanifold $`S`$ (i.e. at each $`xS`$, $`X_i(x)T_xS<T_xM`$), then their Lie bracket $`[X_1,X_2]`$ is also tangent to $`S`$.
This follows by considering the diffeomorphism $`f:\stackrel{~}{S}S`$ that gives an embedding of $`S`$ in $`M`$. One then uses the fact that $`Tf`$ commutes with the Lie bracket, eq. 3.61. That is: the Lie bracket of the $`f`$-related vector fields $`\stackrel{~}{X}_1,\stackrel{~}{X}_2`$ on $`\stackrel{~}{S}`$, which is of course tangent to $`\stackrel{~}{S}`$, is carried by $`Tf`$ to the Lie bracket $`[X_1,X_2]`$ of $`X_1`$ and $`X_2`$. So $`[X_1,X_2]`$ is tangent to $`S`$.
The idea of Frobenius’ theorem will be that this necessary condition of two vector fields being tangent to a submanifold is also sufficient. To be more precise, we need the following definitions.
A distribution $`D`$ on a manifold $`M`$ is a subset of the tangent bundle $`TM`$ such that at each $`xM`$, $`D_x:=DT_xM`$ is a vector space. The dimension of $`D_x`$ is the rank of $`D`$ at $`x`$. If the rank of $`D`$ is constant on $`M`$, we say the distribution is regular.
A distribution is smooth if for every $`xM`$, and every $`X_0D_x`$, there is a neighbourhood $`UM`$ of $`x`$, and a smooth vector field $`X`$ on $`U`$ such that (i) $`X(x)=X_0`$, (ii) for all $`yU`$, $`X(y)D_y`$. Such a vector field $`X`$ is called a local section of $`D`$. Example: a set of $`r`$ vector fields, $`X_1,\mathrm{},X_r`$ each defined on $`M`$, together define a smooth distribution of rank at most $`r`$.
A distribution is involutive if for any pair $`X_1,X_2`$ of local sections, the Lie bracket $`[X_1,X_2](y)D_y`$ in the two sections’ common domain of definition.
We similarly say that a set of $`r`$ smooth vector fields, $`X_1,\mathrm{},X_r`$, on a manifold $`M`$ is in involution if everywhere in $`M`$ they span their Lie brackets. That is: there are smooth real functions $`h_{ij}^k:MR,i,j,k=1,\mathrm{},r`$ such that at each $`xM`$
$$[X_i,X_j](x)=\mathrm{\Sigma }_kh_{ij}^k(x)X_k(x).$$
(3.64)
(Beware: involution is used in a different sense in connection with Liouville’s theorem, viz. a set of real functions on phase space is said to be in involution when all their pairwise Poisson brackets vanish.)
A distribution $`D`$ on $`M`$ is integrable if for each $`xM`$ there is a local submanifold $`N(x)`$ of $`M`$ whose tangent bundle equals the restriction of $`D`$ to $`N(x)`$. If $`D`$ is integrable, the various $`N(x)`$ can be extended to get, through each $`xM`$, a unique maximal connected set whose tangent space at each of its elements $`y`$ is $`D_y`$. Such a set is called a (maximal) integral manifold.
NB: In general, each integral manifold is injectively immersed in $`M`$, but not embedded in it; and so, by the discussion in (2) of Section 3.3.1, an integral manifold might not be a submanifold of $`M`$. But (like most treatments), I shall ignore this point, and talk of them as submanifolds, integral submanifolds.
If the rank of $`D`$ is constant on $`M`$, all the integral submanifolds have a common dimension: the rank of $`D`$. But in general the rank of $`D`$ varies across $`M`$, and so does the dimension of the integral submanifolds.
We similarly say that a set of $`r`$ vector fields, $`X_1,\mathrm{},X_r`$, is integrable; viz. if through every $`xM`$ there passes a local submanifold $`N(x)`$ of $`M`$ whose tangent space at each of its points is spanned by $`X_1,\mathrm{},X_r`$. (Again: we allow that at some $`x`$, $`X_1(x),\mathrm{},X_r(x)`$ may be linearly dependent, so that the dimension of the submanifolds varies.)
We say (both for distributions and sets of vector fields) that the collection of integral manifolds is a foliation of $`M`$, and its elements are leaves. Again: if the dimension of the leaves is constant on $`M`$, we say the foliation is regular.
With these definitions in hand, we can now state Frobenius’ theorem: both in its usual form, which concerns the case of constant rank, i.e. regular distributions and vector fields that are everywhere linearly independent; and in a generalized form. The usual form is:
> Frobenius’ theorem (usual form) A smooth regular distribution is integrable iff it is involutive.
> Or in terms of vector fields: a set of $`r`$ smooth vector fields, $`X_1,\mathrm{},X_r`$, on a manifold $`M`$, that are everywhere linearly independent, is integrable iff it is in involution.
The generalization comes in two stages. The first stage concerns varying rank, but assumes a finite set of vector fields. It is straightforward: this very same statement holds. That is: a set of $`r`$ smooth vector fields, $`X_1,\mathrm{},X_r`$, on a manifold $`M`$ (perhaps not everywhere linearly independent) is integrable iff it is in involution.
But for the foliation of Poisson manifolds (Section 5.3.3), we need to consider an infinite set of vector fields, perhaps with varying rank; and for such a set, this statement fails. Fortunately, there is a useful generalization; as follows.
Let $`𝒳`$ be a set of vector fields on a manifold $`M`$, that forms a vector space. So in the above discussion of $`r`$ vector fields, $`𝒳`$ can be taken as all the linear combinations $`\mathrm{\Sigma }_{i=1}^rf_i(x)X_i(x),xM`$, where the $`f_i`$ are arbitrary smooth functions $`f:M\mathrm{I}\mathrm{R}`$. Such an $`𝒳`$ is called finitely generated.
For any $`𝒳`$ forming a vector space, we say (as before) that $`𝒳`$ is in involution if $`[X,Y]𝒳`$ whenever $`X,Y𝒳`$. Let $`𝒳_x`$ be the subspace of $`T_xM`$ spanned by the $`X(x)`$ for all $`X𝒳`$. As before, we define: an integral manifold of $`𝒳`$ is a submanifold $`NM`$ such that for all $`yN`$, $`T_yN=𝒳_y`$; and $`𝒳`$ is called integrable iff through each $`xM`$ there passes an integral manifold.
As before: if $`𝒳`$ is integrable, it is in involution. But the converse fails. A further condition is needed, as follows.
We say that $`𝒳`$ is rank-invariant if for any vector field $`X𝒳`$, the dimension of the subspace $`𝒳_{\mathrm{exp}(\tau X)(x)}`$ along the flow generated by $`X`$ is a constant, independent of $`\tau `$. (But it can depend on the point $`x`$.)
Since the integral curve $`\mathrm{exp}(\tau X)(x)`$ through $`x`$ should be contained in any integral submanifold, rank-invariance is certainly a necessary condition of integrability. (It also follows from $`𝒳`$ being finitely generated.) In fact we have:
> Frobenius’ theorem (generalized form) A system $`𝒳`$ of vector fields on $`M`$ is integrable iff it is rank-invariant and in involution.
The idea of the proof is to directly construct the integral submanifolds. The submanifold through $`x`$ is obtained as
$$N=\{\mathrm{exp}(X_1)\mathrm{exp}(X_2)\mathrm{}.\mathrm{exp}(X_p)(x):p1,X_i𝒳\}.$$
(3.65)
The rank-invariance secures that for any $`yN`$, $`𝒳_y`$ has dimension dim($`N`$).
### 3.4 Lie groups, and their Lie algebras
I introduce Lie groups and their Lie algebras. By the last two Subsections (Sections 3.4.3 and 3.4.4), we will have enough theory to compute efficiently the Lie algebra of a fundamentally important Lie group, the rotation group.
#### 3.4.1 Lie groups and matrix Lie groups
A Lie group is a group $`G`$ which is also a manifold, and for which the product and inverse operations $`G\times GG`$ and $`GG`$ are smooth.
Examples:—-
(i): $`\mathrm{I}\mathrm{R}^n`$ under addition.
(ii): The group of linear isomorphisms of $`\mathrm{I}\mathrm{R}^n`$ to $`\mathrm{I}\mathrm{R}^n`$, denoted $`GL(n,\mathrm{I}\mathrm{R})`$ and called the general linear group; represented by the real invertible $`n\times n`$ matrices. This is an open subset of $`\mathrm{I}\mathrm{R}^{n^2}`$, and so a manifold of dimension $`n^2`$; and the formulas for the product and inverse of matrices are smooth in the matrix components.
(iii) The group of rotations about the origin of $`\mathrm{I}\mathrm{R}^3`$, represented by $`3\times 3`$ orthogonal matrices of determinant 1; denoted $`SO(3)`$, where $`S`$ stands for ‘special’ (i.e. determinant 1), and $`O`$ for ‘orthogonal’.
In fact, all three examples can be regarded as Lie groups of matrices, with matrix multiplication as the operation. In example (i), consider the isomorphism $`\theta `$ between $`\mathrm{I}\mathrm{R}^n`$ under addition and $`(n+1)\times (n+1)`$ matrices with diagonal entries all equal to 1, other rightmost column entries equal to the given vector in $`\mathrm{I}\mathrm{R}^n`$, and all other entries zero. Thus consider, for the case $`n=3`$:
$$\theta :\left(\begin{array}{c}x\\ y\\ z\end{array}\right)\left(\begin{array}{cccc}1& 0& 0& x\\ 0& 1& 0& y\\ 0& 0& 1& z\\ 0& 0& 0& 1\end{array}\right).$$
(3.66)
This suggests that we define a matrix Lie group to be any set of invertible real matrices, under matrix multiplication, that is closed under multiplication, inversion and taking of limits. That a matrix Lie group is a Lie group will then follow from $`GL(n,\mathrm{I}\mathrm{R})`$ being a Lie group, and the theorem below (in Section 3.4.3) that any closed subgroup of a Lie group is itself a Lie group.
For matrix Lie groups, some of the theory below simplifies. For example, the definition of exponentiation of an element of the group’s Lie algebra reduces to exponentiation of a matrix. But we will develop some of the general theory, since (as always!) it is enlightening and powerful.
#### 3.4.2 The Lie algebra of a Lie group
The main result in this Subsection is that for any Lie group $`G`$, the tangent space $`T_eG`$ at the identity $`eG`$ has a natural Lie algebra structure that is induced by certain natural vector fields on $`G`$; as follows.
##### 3.4.2.A Left-invariant vector fields define the Lie algebra
:
Let $`G`$ be a Lie group. Each $`gG`$ defines a diffeomorphism of $`G`$ onto itself by left translation, and similarly by right translation:
$$L_g:hGghG;R_g:hGhgG.$$
(3.67)
Remark: In Section 4 we will describe this in the language of group actions, saying that in eq. 3.67 $`G`$ acts on itself by left and right translation.
Now consider the induced maps on the tangent spaces, i.e. the tangent (aka: derivative) maps; cf. eq.s 3.29, 3.30. They are $`(L_g)_{}=:L_g,(R_g)_{}=:R_g`$ where for each $`hG`$:
$$L_g:T_hGT_{gh}G\text{ and}R_g:T_hGT_{hg}G.$$
(3.68)
In particular: the derivative $`(R_g)_{}`$ at $`eG`$ maps $`T_eG`$ to $`T_gG`$. This implies that every vector $`\xi T_eG`$ defines a vector field on $`G`$: its value at any $`gG`$ is the image $`(R_g)_{}\xi `$ of $`\xi `$ under $`(R_g)_{}`$. Such a vector field is called a right-invariant vector field: it is uniquely defined by (applying the derivative of right translation to) its value at the identity $`eG`$.
In more detail, and now defining left-invariant vector fields:—
A vector field $`X`$ on $`G`$ is called left-invariant if for every $`gG`$, $`(L_g)_{}X=X`$. More explicitly, let us write $`T_hL_g`$ for the tangent or derivative of $`L_g`$ at $`h`$, i.e. for $`L_g:T_hGT_{gh}G`$. Then left-invariance requires that
$$(T_hL_g)X(h)=X(gh)\mathrm{for}\mathrm{every}g\mathrm{and}hG.$$
(3.69)
Thus every vector $`\xi T_eG`$ defines a left-invariant vector field, written $`X_\xi `$, on $`G`$: $`X_\xi `$’s value at any $`gG`$ is the image $`(L_g)_{}\xi `$ of $`\xi `$ under $`(L_g)_{}`$. In other words: $`X_\xi (g):=(T_eL_g)\xi `$.
Not only is a left-invariant vector field uniquely defined by its value at the identity $`eG`$. Also, the set $`𝒳_L(G)`$ of left-invariant vector fields on $`G`$ is isomorphic as a vector space to the tangent space $`T_eG`$ at the identity $`e`$. For the linear maps $`\alpha ,\beta `$ defined by
$$\alpha :X𝒳_L(G)X(e)T_eG;\mathrm{and}\beta :\xi T_eG\{gX_\xi (g):=(T_eL_g)\xi \}𝒳_L(G)$$
(3.70)
compose to give the identity maps:
$$\beta \alpha =id_{𝒳_L(G)};\alpha \beta =id_{T_eG}.$$
(3.71)
$`𝒳_L(G)`$ is a Lie subalgebra of the Lie algebra of all vector fields on $`G`$, because it is closed under the Lie bracket. That is: the Lie bracket of left-invariant vector fields $`X`$ and $`Y`$ is itself left-invariant, since one can check that for every $`gG`$ we have (with $`L`$ meaning ‘left’ not ‘Lie’!)
$$L_g[X,Y]=[L_gX,L_gY]=[X,Y].$$
(3.72)
If we now define a bracket on $`T_eG`$ by
$$[\xi ,\eta ]:=[X_\xi ,X_\eta ](e)$$
(3.73)
then $`T_eG`$ becomes a Lie algebra. It is called the Lie algebra of $`G`$, written $`𝔤`$ (or, to avoid ambiguity about which Lie group is in question: $`𝔤(G)`$). It follows from eq. 3.72 that
$$[X_\xi ,X_\eta ]=X_{[\xi ,\eta ]};$$
(3.74)
that is to say, the maps $`\alpha ,\beta `$ are Lie algebra isomorphisms.
This result, that $`T_eG`$ has a natural Lie algebra structure, is very important. For, as we shall see in the rest of Section 3.4: the structure of a Lie group is very largely determined by the structure of this Lie algebra. Accordingly, as we shall see in Sections 4 and 5 et seq.: this Lie algebra underpins most of the constructions made with the Lie group, e.g. in Lie group actions. Thus Olver writes that this result ‘is the cornerstone of Lie group theory … almost the entire range of applications of Lie groups to differential equations ultimately rests on this one construction!’ (Olver 2000: 42).
Before turning in the next Subsection to examples, and the topic of subgroups and subalgebras, I end with four results, (1)-(4), which will be needed later; and a remark.
##### 3.4.2.B Four results
:
(1): Lie group structure determines Lie algebra structure in the following sense. If $`G,H`$ are Lie groups, and $`f:GH`$ is a smooth homomorphism, then the derivative of $`f`$ at the identity $`T_ef:𝔤(G)𝔤(H)`$ is a Lie algebra homomorphism. In particular, for all $`\xi ,\eta 𝔤(G)`$, $`(T_ef)[\xi ,\eta ]=[T_ef(\xi ),T_ef(\eta )]`$. (Cf. eq. 3.61.)
(2): Exponentiation again; a correspondence between left-invariant vector fields and one-dimensional subgroups:
Recall from Section 3.1, especially eq. 3.36, that each vector field $`X`$ on the manifold $`G`$ determines an integral curve $`\varphi _X`$ in $`G`$ passing through the identity $`e`$ (with $`\varphi _X(0)=e`$). We now write the points in (the image of) this curve as $`g_\tau `$ ($`X`$ and $`e`$ being understood):
$$\mathrm{exp}(\tau X)(e)X^\tau (e)\varphi _{X,e}(\tau )=:g_\tau .$$
(3.75)
It is straightforward to show that if $`X`$ is left-invariant, this (image of a) curve is a one-parameter subgroup of $`G`$: i.e. not just as eq. 3.35 et seq., a one-parameter subgroup of the group of diffeomorphisms of the manifold $`G`$. In fact:
$$g_{\tau +\sigma }=g_\tau g_\sigma g_0=eg_\tau ^1=g_\tau .$$
(3.76)
Besides, the group is defined for all $`\tau \mathrm{I}\mathrm{R}`$; and is isomorphic to either $`\mathrm{I}\mathrm{R}`$ or the circle group $`SO(2)`$. Conversely, any connected one-parameter subgroup of $`G`$ is generated by a left-invariant vector field in this way.
Accordingly, we define exponentiation of elements $`\xi `$ of $`𝔤`$ by reference to the isomorphisms eq. 3.70 and 3.71. It is also convenient to define this as a map taking values in $`G`$. Thus for $`\xi 𝔤`$ and its corresponding left-invariant vector field $`X_\xi `$ that takes as value at $`gG`$, $`X_\xi (g):=(T_eL_g)(\xi )`$, we write the integral curve of $`X_\xi `$ that passes through $`e`$ (with value $`e`$ for argument $`\tau =0`$) as
$$\varphi _\xi :\tau \mathrm{I}\mathrm{R}\mathrm{exp}(\tau X_\xi )(e)G.$$
(3.77)
Then we define the exponential map of $`𝔤`$ into $`G`$ to be the map
$$\mathrm{exp}:\xi 𝔤\varphi _\xi (1)G.$$
(3.78)
Using the linearity of $`\beta `$ as defined by eq. 3.70, these two equations, eq. 3.77 and 3.78, are related very simply:
$$\mathrm{exp}(\tau \xi ):=\varphi _{\tau \xi }(1):=\mathrm{exp}(1.X_{\tau \xi })(e)=\mathrm{exp}(\tau X_\xi ).$$
(3.79)
We write $`\mathrm{exp}_G`$ rather than $`\mathrm{exp}`$ when the context could suggest a Lie group other than $`G`$.
The map $`\mathrm{exp}`$ is a local diffeomorphism of a neighbourhood of $`0𝔤`$ to a neighbourhood of $`eG`$; but not in general a global diffeomorphism onto $`G`$. In modern terms, this result follows by applying the inverse function theorem to the discussion above. (It also represents an interesting example of the history of subject; cf. Hawkins (2000: 82-83) for Lie’s version of this result, without explicit mention of its local nature.)
The map $`\mathrm{exp}`$ also has the basic property, adding to result (1) above, that …
(3): Homomorphisms respect exponentiation:
If $`f:GH`$ is a smooth homomorphism of Lie groups, then for all $`\xi 𝔤`$,
$$f(\mathrm{exp}_G\xi )=\mathrm{exp}_H((T_ef)(\xi )).$$
(3.80)
(4): Right-invariant vector fields as an alternative approach:
We have followed the usual practice of defining $`𝔤`$ in terms of left-invariant vector fields. One can instead use right-invariant vector fields. This produces some changes in signs, and in whether certain defined operations respect or reverse the order of two elements used in their definition. I will not go into many details about this. But some will be needed when we consider:
(i): Lie group actions, and especially their infinitesimal generators (Section 4.4 and 4.5);
(ii): reduction on the cotangent bundle of a Lie group—as occurs in the theory of the rigid body (Section 6.5 and 7.3.3).
For the moment we just note two basic results, (A) and (B); postponing others to Section 4.4 et seq..
(A): Corresponding to the vector space isomorphism between $`𝔤`$ and the left-invariant vector fields, as in eq. 3.70. viz.
$$\xi T_eGX_\xi 𝒳_L(G)\mathrm{with}X_\xi (g):=(T_eL_g)\xi ,$$
(3.81)
there is a vector space isomorphism to the set of right-invariant vector fields
$$\xi T_eGY_\xi 𝒳_R(G)\mathrm{with}Y_\xi (g):=(T_eR_g)\xi .$$
(3.82)
Besides, the Lie bracket of right-invariant vector fields is itself right-invariant. So corresponding to our previous definition, eq. 3.73, of a Lie bracket on $`T_eG`$, and its corollary eq. 3.74, i.e. $`[X_\xi ,X_\eta ]=X_{[\xi ,\eta ]}`$, that makes $`T_eG𝒳_L(G)`$ a Lie algebra isomorphism: we can also define a Lie bracket on $`T_eG`$ by
$$[\xi ,\eta ]_R:=[Y_\xi ,Y_\eta ](e),$$
(3.83)
and get a Lie algebra isomorphism $`T_eG𝒳_R(G)`$.
(B): But the two Lie brackets, eq. 3.73 and 3.83, on $`T_eG`$ are different. In fact one can show that:
(i): $`X_\xi `$ and $`Y_\xi `$ are related by
$$I_{}X_\xi =Y_\xi $$
(3.84)
where $`I:GG`$ is the inversion map $`I(g):=g^1`$, and $`I_{}`$ is the push-forward on vector fields induced by $`I`$, cf. eq. 3.31, i.e.
$$(I_{}X_\xi )(g):=(TIX_\xi I^1)(g).$$
(3.85)
Besides, since $`I`$ is a diffeomorphism, eq. 3.84 makes $`I_{}`$ a vector space isomorphism.
(ii): It follows from eq. 3.84 that
$$[X_\xi ,X_\eta ](e)=[Y_\xi ,Y_\eta ](e);\mathrm{so}[\xi ,\eta ]=[\xi ,\eta ]_R.$$
(3.86)
Finally, a remark about physics. In applications to physics, $`G`$ is usually the group of symmetries of a physical system, and so a vector field on $`G`$ is the infinitesimal generator of a one-parameter group of symmetries. For mechanics, we saw this repeatedly in Section 2, especially as regards the group of translations and rotations about the origin, in physical space $`\mathrm{I}\mathrm{R}^3`$. This Subsection’s isomorphism between the Lie algebra $`𝔤`$ and left-invariant vector fields on $`G`$ means that we can think of $`𝔤`$ also as consisting of infinitesimal symmetries of the system. (The $`\xi 𝔤`$ are also called generators of the group $`G`$.)
#### 3.4.3 Examples, subgroups and subalgebras
I begin with the first two of Section 3.4.1’s three examples. That will prompt a little more theory, which will enable us to deal efficiently in the next Subsection with the third example, viz. the rotation group.
(1): Examples:—
(i): $`G:=\mathrm{I}\mathrm{R}^n`$ under addition. $`G`$ is abelian so that left and right translation coincide. The invariant vector fields are just the constant vector fields, so that $`𝒳_L(G)𝒳_R(G)\mathrm{I}\mathrm{R}^n`$. So the tangent space at the identity $`T_eG`$, i.e. the Lie algebra $`𝔤`$, is itself $`\mathrm{I}\mathrm{R}^n`$. The bracket structure is wholly degenerate: for all invariant vector fields $`X,Y`$, $`[X,Y]=0`$; and for all $`\xi ,\eta 𝔤`$, $`[\xi ,\eta ]=0`$.
(ii): $`G:=GL(n,\mathrm{I}\mathrm{R})`$, the general linear group. Since $`G`$ is open in $`End(\mathrm{I}\mathrm{R}^n,\mathrm{I}\mathrm{R}^n)`$, the vector space of all linear maps on $`\mathrm{I}\mathrm{R}^n`$ (‘$`End`$’ for ‘endomorphism’), $`G`$’s Lie algebra, as a vector space, is $`End(\mathrm{I}\mathrm{R}^n,\mathrm{I}\mathrm{R}^n)`$; (cf. example (i)). To compute what the Lie bracket is, we first note that any $`\xi End(\mathrm{I}\mathrm{R}^n,\mathrm{I}\mathrm{R}^n)`$ defines a corresponding vector field on $`GL(n,\mathrm{I}\mathrm{R})`$ by
$$X_\xi :AGL(n,\mathrm{I}\mathrm{R})A\xi End(\mathrm{I}\mathrm{R}^n,\mathrm{I}\mathrm{R}^n).$$
(3.87)
Besides, $`X_\xi `$ is left-invariant, since for every $`BGL(n,\mathrm{I}\mathrm{R})`$, the left translation
$$L_B:AGL(n,\mathrm{I}\mathrm{R})BAGL(n,\mathrm{I}\mathrm{R})$$
(3.88)
is linear, and so
$$X_\xi (L_BA)=BA\xi =T_AL_BX_\xi (A).$$
(3.89)
Applying now eq. 3.59 at the identity $`IGL(n,\mathrm{I}\mathrm{R})`$ to the definition of the bracket in the Lie algebra, eq. 3.73, we have:
$$[\xi ,\eta ]:=[X_\xi ,X_\eta ](I)=𝐃X_\eta (I)X_\xi (I)𝐃X_\xi (I)X_\eta (I).$$
(3.90)
But $`X_\eta A=A\eta `$ is linear in $`A`$, so $`𝐃X_\eta (I)B=B\eta `$. This means that
$$𝐃X_\eta (I)X_\xi (I)=\xi \eta ;$$
(3.91)
and similarly
$$𝐃X_\xi (I)X_\eta (I)=\eta \xi .$$
(3.92)
So the Lie algebra $`End(\mathrm{I}\mathrm{R}^n,\mathrm{I}\mathrm{R}^n)`$ has the usual matrix commutator as its bracket: $`[\xi ,\eta ]=\xi \eta \eta \xi `$. This Lie algebra is often written $`𝔤𝔩(n,\mathrm{I}\mathrm{R})`$.
Let us apply to this example, result (2) from Section 3.4.2.B. In short, the result said that left-invariant vector fields correspond (by exponentiation through $`eG`$) to connected one-parameter subgroups of $`G`$. To find the one-parameter subgroup $`\mathrm{exp}(\tau X_\xi )(e)`$ of $`GL(n,\mathrm{I}\mathrm{R})`$, we take the matrix entries $`x_{ij},(i,j=1,\mathrm{},n)`$ as the $`n^2`$ coordinates on $`GL(n,\mathrm{I}\mathrm{R})`$, so that the tangent space at the identity matrix $`I`$ is the set of vectors
$$\mathrm{\Sigma }_{ij}\xi _{ij}\frac{}{x_{ij}}_I$$
(3.93)
with $`\xi =(\xi _{ij})`$ an arbitrary matrix. For given $`\xi `$, $`\mathrm{exp}(\tau X_\xi )e`$ is found by integrating the $`n^2`$ ordinary differential equations
$$\frac{dx_{ij}}{d\tau }=\mathrm{\Sigma }_k\xi _{ik}x_{kj};x_{ij}(0)=\delta _{ij}.$$
(3.94)
The solution is just the matrix exponential:
$$X(\tau )=\mathrm{exp}(\tau \xi ).$$
(3.95)
More generally, let us return to Section 3.4.1’s idea of a matrix Lie group. For a matrix Lie group $`G`$, the definition of its Lie algebra can be given as:
$$𝔤=\{\mathrm{the}\mathrm{set}\mathrm{of}\mathrm{matrices}\xi =\varphi ^{}(0):\varphi \mathrm{a}\mathrm{differentiable}\mathrm{map}:\mathrm{I}\mathrm{R}G,\varphi (0)=e_G\}.$$
(3.96)
The deduction of the structure of the Lie algebra then proceeds straightforwardly. In particular, we get the result that the one-parameter subgroup generated by $`\xi 𝔤`$ is given by matrix exponentials, as in eq. 3.95: the group is $`\{\mathrm{exp}(\tau \xi ):\tau \mathrm{I}\mathrm{R}\}`$.
This result will help us compute our third example: finding the Lie algebra of the rotation group. But for that example, it is worth first developing a little the result (2) from Section 3.4.2.B: i.e. the correspondence between left-invariant vector fields and connected one-parameter subgroups of $`G`$.
(2): More theory:—
First, a warning remark. We will later need to take notice of the fact that a subgroup, even a one-parameter subgroup, of a Lie group $`G`$ need not be a submanifold of $`G`$. Here we recall Section 3.3.1’s definitions of immersion and embedding. Accordingly, we now define a subgroup $`H`$ of a Lie group $`G`$ to be a Lie subgroup of $`G`$ if the inclusion map $`i:HG`$ is an injective immersion.
Just as we saw in Section 3.3.1 that not every injective immersion is an embedding, so also there are examples of Lie subgroups that are not submanifolds. Example: the torus $`𝖳^2`$ can be made into a Lie group in a natural way (exercise: do this!); the one-parameter subgroups on the torus $`𝖳^2`$ that wind densely on the torus are Lie subgroups that are not submanifolds. (For more details about this example, cf. Arnold (1973: 160-167) or Arnold (1989: 72-74) or Butterfield (2004a: Section 2.1.3.B).)
But it turns out that being closed is a sufficient, and necessary, further condition. That is:
> If $`H`$ is a closed subgroup of a Lie group $`G`$, then $`H`$ is a submanifold of $`G`$ and in particular a Lie subgroup. And conversely, if $`H`$ is a Lie subgroup that is also a submanifold, then $`H`$ is closed.
Result (2) from Section 3.4.2.B, i.e. the correspondence between one-dimensional subgroups of $`G`$ and one-dimensional subspaces (and so subalgebras) of $`𝔤`$, generalizes to higher-dimensional subgroups and subalgebras. That is to say:
> If $`HG`$ is a Lie subgroup of $`G`$, then its Lie algebra $`𝔥:=𝔤(H)`$ is a subalgebra of $`𝔤𝔤(G)`$. In fact
>
> $$𝔥=\{\xi 𝔤:\mathrm{exp}(\tau X_\xi )(e)H,\mathrm{for}\mathrm{all}\tau \mathrm{I}\mathrm{R}\}.$$
> (3.97)
> And conversely, if $`𝔥`$ is any $`m`$-dimensional subalgebra of $`𝔤`$, then there is a unique connected $`m`$-dimensional Lie subgroup $`H`$ of $`G`$ with Lie algebra $`𝔥`$.
The proof of the first two statements uses result (1) of Section 3.4.2.B. For the third, i.e. converse, statement, the main idea is that $`𝔥`$ defines $`m`$ vector fields on $`G`$ that are linearly independent and in involution, so that one can apply Frobenius’ theorem to infer an integral submanifold. One then has to prove that $`H`$ is a Lie subgroup: Olver (2000: Theorem 1.51) and Marsden and Ratiu (1999: 279-280) give details and references. (Historical note: to see that this result, sometimes called Lie’s ‘third fundamental theorem’, is close to what Lie himself called the main theorem of his theory of groups, cf. Hawkins (2000: 83).)
This general correspondence between Lie subgroups and Lie subalgebras prompts the question whether every finite-dimensional Lie algebra $`𝔤`$ is the Lie algebra of a Lie group. The answer is Yes. Besides, the question reduces to the case of a matrix Lie group (i.e. a Lie subgroup of $`GL(n,\mathrm{I}\mathrm{R})`$), in the sense that: every finite-dimensional Lie algebra $`𝔤`$ is isomorphic to a subalgebra of $`𝔤𝔩(n,\mathrm{I}\mathrm{R})`$, for some $`n`$. But be warned: this does not imply (and it is not true) that every Lie group is realizable as a matrix Lie group, i.e. that every Lie group is isomorphic to a Lie subgroup of $`GL(n,\mathrm{I}\mathrm{R})`$.
This general correspondence also simplifies greatly the computation of the Lie algebras of Lie groups, for example $`H:=SO(3)`$, that are Lie subgroups of $`GL(n,\mathrm{I}\mathrm{R})`$. We only need to combine it with example (ii) above, that $`𝔤𝔩(n,\mathrm{I}\mathrm{R})`$ is $`End(\mathrm{I}\mathrm{R}^n,\mathrm{I}\mathrm{R}^n)`$ with the usual matrix commutator as its bracket: $`[\xi ,\eta ]=\xi \eta \eta \xi `$.
Thus we infer that the Lie algebra of $`SO(3)`$, written $`𝔰𝔬(3)`$, is a subalgebra of $`End(\mathrm{I}\mathrm{R}^n,\mathrm{I}\mathrm{R}^n)`$ with the matrix commutator as bracket. Besides, we can identify $`𝔰𝔬(3)`$ by looking at all the one-dimensional subgroups of $`G`$ contained in it. Combining eq. 3.95 and 3.97, we have
$$𝔰𝔬(3)=\{\xi 𝔤𝔩(n,\mathrm{I}\mathrm{R}):\mathrm{the}\mathrm{matrix}\mathrm{exponential}\mathrm{exp}(\tau \xi )SO(3),\tau \mathrm{I}\mathrm{R}\}.$$
(3.98)
With this result in hand, we can now compute $`𝔰𝔬(3)`$.
#### 3.4.4 The Lie algebra of the rotation group
Our first aim is to calculate the Lie algebra $`𝔰𝔬(3)`$ (also written: $`so(3)`$) of $`H:=SO(3)`$, the rotation group. This will lead us back to Section 3.2.1.A’s correspondence between anti-symmetric matrices and vectors in $`\mathrm{I}\mathrm{R}^3`$.
$`SO(3)`$ is represented by $`3\times 3`$ orthogonal matrices of determinant 1. So the requirement in eq. 3.98 becomes, now writing $`e`$, not $`\mathrm{exp}`$:
$$(e^{\tau \xi })(e^{\tau \xi })^T=I\mathrm{and}\mathrm{det}(e^{\tau \xi })=1.$$
(3.99)
Differentiating the first equation with respect to $`\tau `$ and setting $`\tau =0`$ yields
$$\xi +\xi ^T=0.$$
(3.100)
So $`\xi `$ must be anti-symmetric, i.e. represented by an anti-symmetric matrix. Conversely, for any such anti-symmetric matrix $`\xi `$, we can show that det$`(e^{\tau \xi })=1`$. So, indeed:
$$𝔰𝔬(3)=\{3\times 3\mathrm{antisymmetric}\mathrm{matrices}\}.$$
(3.101)
Notice that the argument is independent of choosing $`n=3`$. It similarly computes $`𝔰𝔬(n)`$ for any integer $`n`$:
$$𝔰𝔬(n)=\{n\times n\mathrm{antisymmetric}\mathrm{matrices}\}.$$
(3.102)
Thus the rotations on euclidean space $`\mathrm{I}\mathrm{R}^n`$ of any dimension $`n`$ are generated by the Lie algebra of $`n\times n`$ anti-symmetric matrices.
This justifies our assertion at the end of Section 3.2.1.A that the rotation group in three dimensions is special in being representable by vectors in the space on which it acts, i.e. $`\mathrm{I}\mathrm{R}^3`$. For as we have just seen, in general the infinitesimal generators of rotations are anti-symmetric matrices, which in $`n`$ dimensions have $`n(n1)/2`$ independent components. But only for $`n=3`$ does this equal $`n`$.
Remark: An informal computation of $`𝔰𝔬(3)`$, based on the idea that higher-order terms in $`e^{\tau \xi }`$ can be neglected (cf. the physical idea that $`\xi `$ represents an infinitesimal rotation), goes as follows.
For $`(I+\tau \xi )`$ to be a rotation requires that
$$(I+\tau \xi )(I+\tau \xi )^T=I\mathrm{and}\mathrm{det}(I+\xi \tau )=1.$$
(3.103)
Dropping higher-order terms, the first equation yields
$$I+\tau (\xi +\xi ^T)=I\mathrm{i}.\mathrm{e}.\xi +\xi ^T=0.$$
(3.104)
Besides, the second equation in eq. 3.103 yields no further constraint, since for any anti-symmetric matrix $`\xi `$ written as (cf. eq. 3.47)
$$\xi =\left(\begin{array}{ccc}0& \xi _3& \xi _2\\ \xi _3& 0& \xi _1\\ \xi _2& \xi _1& 0\end{array}\right),$$
(3.105)
we immediately compute that det$`(I+\xi \tau )=1+\tau ^2(\xi _1^2+\xi _2^2+\xi _3^2)`$. So, dropping higher-order terms, det$`(I+\xi \tau )=1`$. In short, we again conclude that
$$𝔰𝔬(3)=\{3\times 3\mathrm{antisymmetric}\mathrm{matrices}\}.$$
(3.106)
For later use (e.g. Sections 4.4 and 4.5.1), we note that the three matrices
$$A^x=\left(\begin{array}{ccc}0& 0& 0\\ 0& 0& 1\\ 0& 1& 0\end{array}\right),A^y=\left(\begin{array}{ccc}0& 0& 1\\ 0& 0& 0\\ 1& 0& 0\end{array}\right),A^z=\left(\begin{array}{ccc}0& 1& 0\\ 1& 0& 0\\ 0& 0& 0\end{array}\right)$$
(3.107)
span $`𝔰𝔬(3)`$, and generate the one-parameter subgroups
$$R_\theta ^x=\left(\begin{array}{ccc}1& 0& 0\\ 0& \mathrm{cos}\theta & \mathrm{sin}\theta \\ 0& \mathrm{sin}\theta & \mathrm{cos}\theta \end{array}\right),R_\theta ^y=\left(\begin{array}{ccc}\mathrm{cos}\theta & 0& \mathrm{sin}\theta \\ 0& 1& 0\\ \mathrm{sin}\theta & 0& \mathrm{cos}\theta \end{array}\right),R_\theta ^z=\left(\begin{array}{ccc}\mathrm{cos}\theta & \mathrm{sin}\theta & 0\\ \mathrm{sin}\theta & \mathrm{cos}\theta & 0\\ 0& 0& 1\end{array}\right)$$
(3.108)
representing anticlockwise rotation around the respective coordinate axes in the physical space $`\mathrm{I}\mathrm{R}^3`$.
Having computed $`𝔰𝔬(3)`$ to consist of antisymmetric matrices, we can use Section 3.2.1.A’s correspondence between these and vectors in $`\mathrm{I}\mathrm{R}^3`$ so as to realize $`𝔰𝔬(3)`$ as vectors with the Lie bracket as vector multiplication. With these realizations in hand, we can readily obtain several further results about rotations. We will not need any. But a good example, which uses eq. 3.48’s isomorphism $`\mathrm{\Theta }`$ from vectors $`\omega \mathrm{I}\mathrm{R}^3`$ to matrices $`A𝔰𝔬(3)`$, is as follows:—
$`\mathrm{exp}(\tau \mathrm{\Theta }(\omega ))`$ is a rotation about the axis $`\omega `$ by the angle $`\tau \omega `$.
We can now begin to see the point of this Chapter’s second motto (from Arnold), that the elementary theory of the rigid body confuses six conceptually different three-dimensional spaces. For our discussion has already distinguished three of the six spaces which Arnold lists (in a different notation). Namely, we have just distinguished:
(i) $`\mathrm{I}\mathrm{R}^3`$, especially when taken as physical space; from (ii) $`𝔰𝔬(3)T_e(SO(3))`$, the generators of rotations; though they are isomorphic as Lie algebras, by eq. 3.48’s bijection $`\mathrm{\Theta }`$ from vectors $`\omega \mathrm{I}\mathrm{R}^3`$ to matrices $`A𝔰𝔬(3)`$;
(ii) $`𝔰𝔬(3)T_e(SO(3))`$ from its isomorphic copy under the derivative of left translation by $`g`$ (i.e. under $`(L_g)_{}`$), viz. $`T_g(SO(3))`$: cf. eq. 3.69. (In the motto, Arnold writes $`g`$ for $`𝔰𝔬(3)`$ and $`G`$ for $`SO(3)`$.)
In Section 5.2.4 we will grasp (even without developing the theory of the rigid body!) the rest of the motto. That is, we will see why Arnold also mentions the three corresponding dual spaces, $`\mathrm{I}\mathrm{R}^3,𝔰𝔬(3)^{}`$ and $`T_g^{}(SO(3))`$. But we can already say more about the two tangent spaces $`𝔰𝔬(3)T_e(SO(3))`$ and $`T_g(SO(3))`$, in connection with the idea that for a pivoted rigid body, the configuration space can be taken as $`SO(3)`$; (cf. (3) of Section 2.2). We will show that there are two isomorphisms from $`T_g(SO(3))`$ to $`T_e(SO(3))`$ that are natural, not only in the mathematical sense of being basis-independent but also in the sense of having a physical interpretation. Namely, they represent the computation of the angular velocity from the Lagrangian generalized velocity, i.e. $`\dot{q}`$. In effect, one isomorphism computes the angular velocity’s components with respect to an orthonormal frame fixed in space (called spatial coordinates); and the other computes it with respect to a frame fixed in the rigid body (body coordinates). In fact, these isomorphisms are the derivatives of right and left translation, respectively; (cf. eq. 3.67 and 3.68).
So suppose a pivoted rigid body has a right-handed orthonormal frame $`\{a,b,c\}`$ fixed in it. We can think of the three unit vectors $`a,b,c`$ as column vectors in $`\mathrm{I}\mathrm{R}^3`$. Arranging them in a $`3\times 3`$ matrix $`g:=(abc)GL(3,\mathrm{I}\mathrm{R})`$, we get a matrix that maps the unit $`x`$-vector $`e_1`$ to $`a`$, the unit $`y`$-vector $`e_2`$ to $`b`$, etc. That is: $`g`$ maps the standard frame $`e_1,e_2,e_3`$ to $`a,b,c`$, and $`g`$ is an orthogonal matrix: $`gS0(3)=\{gGL(3,\mathrm{I}\mathrm{R})\stackrel{~}{g}g=I\}`$. Thus $`g`$ represents the configuration of the body, and the configuration space is $`SO(3)`$.
By differentiating the condition $`\stackrel{~}{g}g=I`$, we deduce that the tangent space at a specific $`g`$ $`T_g(SO(3))`$, i.e. the space of velocities $`\dot{g}`$, is the 3-dimensional vector subspace of $`GL(3,\mathrm{I}\mathrm{R})`$:
$$T_g(SO(3))=\{\dot{g}GL(3,\mathrm{I}\mathrm{R})\dot{\stackrel{~}{g}}g+\stackrel{~}{g}\dot{g}=0\}$$
(3.109)
Now recall examples (ii) and (iii) of Section 3.2.1.A. We saw there that though the angular velocity of the body is usually taken to be the vector $`\omega `$ such that, with our “body-vectors” $`a,b,c`$,
$$\dot{a}=\omega a,\dot{b}=\omega b,\dot{c}=\omega c:$$
(3.110)
we can instead encode the angular velocity by the antisymmetric matrix $`A:=\mathrm{\Theta }(\omega )𝔤T_e(SO(3))`$. As we saw, eq. 3.110 then becomes
$$\dot{a}=\mathrm{\Theta }(\omega )a,\dot{b}=\mathrm{\Theta }(\omega )b,\dot{c}=\mathrm{\Theta }(\omega )c:$$
(3.111)
or equivalently the matrix equation for the configuration $`g=(abc)`$,
$$\dot{g}(\dot{a}\dot{b}\dot{c})=\mathrm{\Theta }(\omega )g;\mathrm{i}.\mathrm{e}.\mathrm{\Theta }(\omega )=\dot{g}g^1.$$
(3.112)
Thus we see that the map from $`T_g(SO(3))`$ to $`𝔤=T_e(SO(3))`$
$$\dot{g}T_g(SO(3))\dot{g}g^1\dot{g}\stackrel{~}{g}𝔤$$
(3.113)
maps the generalized velocity $`\dot{g}`$ to the angular velocity $`\mathrm{\Theta }(\omega )`$. This is the angular velocity represented in the usual elementary way, with respect to coordinates fixed in space. One immediately checks that it is an isomorphism (exercise!).
On the other hand, let us consider $`\mathrm{\Theta }(\omega )`$ as a linear transformation $`\mathrm{\Theta }(\omega ):\mathrm{I}\mathrm{R}^3\mathrm{I}\mathrm{R}^3`$, and express it in the body coordinates $`a,b,c`$. This gives $`g^1\mathrm{\Theta }(\omega )gg^1\dot{g}`$. Thus the map
$$\dot{g}T_g(SO(3))g^1\dot{g}\stackrel{~}{g}\dot{g}𝔤$$
(3.114)
maps the generalized velocity $`\dot{g}`$ to the angular velocity expressed in body coordinates. It also is clearly an isomorphism.
Summing up: we have two natural isomorphisms that compute the angular velocity, in spatial and body coordinates respectively, from the generalized velocity $`\dot{g}`$.
Incidentally, one can verify directly that the images $`\dot{g}\stackrel{~}{g}`$ and $`\stackrel{~}{g}\dot{g}`$ of the isomorphisms eq. 3.113 and 3.114 lie in $`𝔤`$, i.e. are antisymmetric matrices. Thus with $``$ for the elementary dot-product, we have:
$$g^1\dot{g}\stackrel{~}{g}\dot{g}=\left(\begin{array}{c}\stackrel{~}{a}\\ \stackrel{~}{b}\\ \stackrel{~}{c}\end{array}\right)(\dot{a}\dot{b}\dot{c})=\left(\begin{array}{ccc}0& a\dot{b}& a\dot{c}\\ b\dot{a}& 0& b\dot{c}\\ c\dot{a}& c\dot{b}& 0\end{array}\right).$$
(3.115)
This is an antisymmetric matrix, since differentiating $`ab=bc=ac=0`$ with respect to time gives $`a\dot{b}+\dot{a}b=0`$ etc. Finally, we deduce that $`\dot{g}\stackrel{~}{g}`$ is antisymmetric from the facts that $`\dot{g}\stackrel{~}{g}=g(g^1\dot{g})g^1`$ and antisymmetry is preserved by conjugation by $`g`$.
We end this Subsection with two incidental remarks; (they will not be used in what follows).
(1): In Section 2.1.1, we could have specialized the discussion from a symplectic manifold to a symplectic vector space, i.e. a (real, finite-dimensional) vector space equipped with a non-degenerate anti-symmetric bilinear form $`\omega :Z\times Z\mathrm{I}\mathrm{R}`$. It follows that $`Z`$ is of even dimension. The question then arises which linear maps $`A:ZZ`$ preserve the normal form of $`\omega `$ given by eq. 2.4. It is straightforward to show that this is equivalent to $`A`$ preserving the form of Hamilton’s equations (for any Hamiltonian); so that these maps $`A`$ are called symplectic (or canonical, or Poisson). The set of all such maps form a Lie group, the symplectic group, written Sp($`Z,\omega `$). But since this Chapter will not need the theory of canonical transformations, I leave the study of Sp($`Z,\omega `$)’s structure as an exercise! (For details, cf. e.g. Abraham and Marsden (1978: 167-174), Marsden and Ratiu (1999: 69-72, 293-299).)
(2): Finally, a glimpse of the infinite-dimensional manifolds that this Chapter has foresworn. Consider the infinite-dimensional Lie group $`Diff(M)`$ of all diffeomorphisms on $`M`$. An element of its Lie algebra, i.e. a vector $`AT_e(Diff(M))`$, is a vector field, or equivalently a flow, on $`M`$. Besides, the Lie bracket in this Lie algebra $`T_e(Diff(M))`$, as defined by eq. 3.73 turns out to be the usual Lie bracket of the vector fields on $`M`$, as defined in Section 3.2.2.
## 4 Actions of Lie groups
We turn to actions of Lie groups on manifolds. The notions, results and examples in this Section will be crucial from Section 5.4 onwards. Fortunately, the foregoing provides several examples of the notions and results we need. Section 4.1 will give basic material, including the crucial notion of cotangent lifts. Sections 4.2 and 4.3 describe conditions for orbits and quotient spaces to be manifolds. Section 4.4 describes actions infinitesimally, i.e. in terms of their infinitesimal generators. Section 4.5 presents two important representations of a Lie group, its adjoint and co-adjoint representations, on its Lie algebra $`𝔤`$ and on the dual $`𝔤^{}`$ respectively. Finally, Section 4.6 gathers some threads concerning our central, recurring example, viz. the rotation group.
### 4.1 Basic definitions and examples
A left action of a Lie group $`G`$ on a manifold $`M`$ is a smooth map $`\mathrm{\Phi }:G\times MM`$ such that:
(i): $`\mathrm{\Phi }(e,x)=x`$ for all $`xM`$
(ii): $`\mathrm{\Phi }(g,\mathrm{\Phi }(h,x))=\mathrm{\Phi }(gh,x)`$ for all $`g,hG`$ and all $`xM`$.
We sometimes write $`gx`$ for $`\mathrm{\Phi }(g,x)`$.
Similarly, a right action of a Lie group $`G`$ on a manifold $`M`$ is a smooth map $`\mathrm{\Psi }:M\times GM`$ satisfying (i) $`\mathrm{\Psi }(x,e)=x`$ and (ii) $`\mathrm{\Psi }(\mathrm{\Psi }(x,g),h)=\mathrm{\Psi }(x,gh)`$. We sometimes write $`xg`$ for $`\mathrm{\Psi }(x,g)`$.
It is convenient to also have a subscript notation. For every $`gG`$, we define
$$\mathrm{\Phi }_g:MM:x\mathrm{\Phi }(g,x).$$
(4.116)
In this notation, (i) becomes $`\mathrm{\Phi }_e=id_M`$ and (ii) becomes $`\mathrm{\Phi }_{gh}=\mathrm{\Phi }_g\mathrm{\Phi }_h`$. For right actions, (ii) becomes $`\mathrm{\Psi }_{gh}=\mathrm{\Psi }_h\mathrm{\Psi }_g`$.
One immediately verifies that any left action $`\mathrm{\Phi }`$ of $`G`$ on a manifold $`M`$, $`g\mathrm{\Phi }_g:MM`$, defines a right action $`\mathrm{\Psi }`$ by
$$g\mathrm{\Psi }_g:=\mathrm{\Phi }_{g^1}:MM;\mathrm{i}.\mathrm{e}.\mathrm{\Psi }:(x,g)M\times G\mathrm{\Phi }(g^1,x)M.$$
(4.117)
(Use the fact that in $`G`$, $`(gh)^1=h^1g^1`$.) Similarly, a right action defines a left action, by taking the inverse in $`G`$. We will occasionally make use of this left-right “flip”.
The definition of left action is equivalent to saying that the map $`g\mathrm{\Phi }_g`$ is a homomorphism of $`G`$ into Diff($`M`$), the group of diffeomorphisms of $`M`$. In the special case where $`M`$ is a Banach space $`V`$ and each $`\mathrm{\Phi }_g:VV`$ is a continuous linear transformation, the action of $`G`$ on $`V`$ is called a representation of $`G`$ on $`V`$.
The orbit of $`xM`$ (under the action $`\mathrm{\Phi }`$) is the set
$$\mathrm{Orb}(x)=\{\mathrm{\Phi }_g(x):gG\}M.$$
(4.118)
The action is called transitive if there is just one orbit, i.e. for all $`x,yM`$, there is a $`gG`$ such that $`gx=y`$. It is called effective (or faithful) if $`\mathrm{\Phi }_g=\mathrm{id}_M`$ implies $`g=e`$, i.e. if $`g\mathrm{\Phi }_g`$ is one-to-one. It is called free if it has no fixed points for any $`ge`$: that is, $`\mathrm{\Phi }_g(x)=x`$ implies $`g=e`$. In other words, it is free if for each $`xM`$, $`g\mathrm{\Phi }_g(x)`$ is one-to-one. (So: every free action is faithful.)
##### 4.1.A Examples; cotangent lifts
We begin with geometric examples; and then return to mechanics, giving first some general theory, followed by some examples.
(1): Geometric examples:—
(i): $`SO(3)`$ acts on $`\mathrm{I}\mathrm{R}^3`$ by $`(A,x)Ax`$. The action is faithful. But it is neither free (each rotation fixes the points on its axis) nor transitive (the orbits are the spheres centred at the origin).
(ii): $`GL(n,\mathrm{I}\mathrm{R})`$ acts on $`\mathrm{I}\mathrm{R}^n`$ by $`(A,x)Ax`$. The action is faithful, not free, and “almost transitive”: the zero subspace $`\{\mathrm{𝟎}\}`$ is an orbit, and so is $`\mathrm{I}\mathrm{R}^n\{\mathrm{𝟎}\}`$.
(iii): Suppose $`X`$ is a vector field on $`M`$ which is complete in the sense that the flow $`\varphi _X(\tau )`$ of eq. 3.35 is defined for all $`\tau \mathrm{I}\mathrm{R}`$. Then this flow defines an action of $`\mathrm{I}\mathrm{R}`$ on $`M`$.
We turn to two examples which will be central, and recurring, in our discussion of symplectic reduction.
(iv): Left translation by each $`gG`$, $`L_g:hGghG`$ (cf. eq. 3.67), defines a left action of $`G`$ on itself. Since $`G`$ is a group, it is transitive and free (and so faithful). Similarly, right translation, $`gR_g`$ with $`R_g:hGhgG`$, defines a right action. And $`gR_{g^1}`$ defines a left action; cf. eq. 4.117.
One readily proves that left translation lifts to the tangent bundle $`TG`$ as a left action. That is: one verifies by the chain rule that
$$\mathrm{\Phi }_g:TGTG:vv_hT_hG(T_hL_g)(v)T_{gh}G$$
(4.119)
defines a left action on $`TG`$. Similarly, right translation lifts to a right action on $`TG`$. But our interest in Hamiltonian mechanics of course makes us more interested in cotangent lifts. See (2) below for the general definitions, and example (viii) in (3) below for the cotangent lift of left translation.
(v): $`G`$ acts on itself by conjugation (inner automorphism): $`gK_g:=R_{g^1}L_g`$. That is: $`K_g:hGghg^1G`$. Each $`K_g`$ is an isomorphism of $`G`$. The orbits are conjugacy classes. Section 4.5 will introduce two “differentiated versions” of action by conjugation, viz. the adjoint and co-adjoint actions, which will be important in symplectic reduction.
(2): Hamiltonian symmetries and cotangent lifts:—
We turn to Hamiltonian mechanics. Following the discussion in Section 2.1.3, we say: given a Hamilton system $`(M,\omega ,H)`$ with $`(M,\omega )`$ a symplectic manifold and $`H:M\mathrm{I}\mathrm{R}`$, a Hamiltonian group of symmetries is a Lie group $`G`$ acting on $`M`$ such that each $`\mathrm{\Phi }_g:MM`$ preserves both $`\omega `$ and $`H`$. Then the simplest possible examples are spatial translations and-or rotations acting on the free particle. The details of these examples, (vi) and (vii) below, will be clearer if we first develop some general theory.
This theory will illustrate the interaction between the left-right contrast for actions, and the tangent-cotangent contrast for bundles. Besides, both the general theory and the examples’ details will carry over straightforwardly, i.e. component by component, to the case of $`N`$ particles interacting by Newtonian gravity, discussed in Section 2.3.2: the action defined on a single particle is just repeated for each of the $`N`$ particles.
So we will take $`M:=(\mathrm{I}\mathrm{R}^3)\times (\mathrm{I}\mathrm{R}^3)^{},\omega :=dq^idp^i,H:=p^2/2m.`$ In the first place, both translations (by $`𝐱\mathrm{I}\mathrm{R}^3`$) and rotations (by $`ASO(3)`$) act on the configuration space $`Q=\mathrm{I}\mathrm{R}^3`$. We have actions of $`\mathrm{I}\mathrm{R}^3`$ and $`SO(3)`$ on $`\mathrm{I}\mathrm{R}^3`$ by
$$\mathrm{\Phi }_𝐱(𝐪)=𝐪+𝐱;\mathrm{\Phi }_A(𝐪)=A𝐪.$$
(4.120)
But these actions lift to the cotangent bundle $`T^{}Q=(\mathrm{I}\mathrm{R}^3)\times (\mathrm{I}\mathrm{R}^3)^{}\mathrm{I}\mathrm{R}^6`$; (as mentioned in Section 2.3.2). The lift of these actions is defined using a result that does not use the notion of an action. Namely:
> Any diffeomorphism $`f:Q_1Q_2`$ induces a cotangent lift $`T^{}f:T^{}Q_2T^{}Q_1`$ (i.e. in the opposite direction) which is symplectic, i.e. maps the canonical one-form, and so symplectic form, on $`T^{}Q_2`$ to that of $`T^{}Q_1`$.
To define the lift of an action, it is worth going into detail about the definition of $`T^{}f`$. (But I will not prove the result just stated; for details, cf. Marsden and Ratiu (1999: Section 6.3).)
The idea is that $`T^{}f`$ is to be the “pointwise adjoint” of the tangent map $`Tf:TQ_1TQ_2`$ (eq. 3.29). That is: we define $`T^{}f`$ in terms of the contraction of its value, for an arbitrary argument $`\alpha T_{q_2}^{}Q_2`$, with an arbitrary tangent vector $`vT_{f^1(q_2)}Q_1`$. (Here it will be harmless to (follow many presentations and) conflate a point in $`T^{}Q_2`$, i.e. strictly speaking a pair $`(q_2,\alpha ),q_2Q_2,\alpha T_{q_2}^{}Q_2`$, with its form $`\alpha `$. And similarly it will be harmless to conflate a point $`(q_1,v)`$ in $`TQ_1`$ with its vector $`vT_{q_1}Q_1`$.)
We recall that any finite-dimensional vector space is naturally, i.e. basis-independently, isomorphic to its double dual: $`(V^{})^{}V`$; and we will use angle brackets $`<;>`$ for the natural pairing between $`V`$ and $`V^{}`$. So we define $`T^{}f;T^{}Q_2T^{}Q_1`$ by requiring:
$$<(T^{}f)(\alpha );v>:=<\alpha ;(Tf)(v)>,\alpha T_{q_2}^{}Q_2,vT_{f^1(q_2)}Q_1.$$
(4.121)
NB: Because $`T^{}f`$ “goes in the opposite direction”, the composition of lift with function-composition involves a reversal of the order. That is: if $`Q_1=Q_2Q`$ and $`f,g`$ are two diffeomorphisms of $`Q`$, then
$$T^{}(fg)=T^{}gT^{}f.$$
(4.122)
With this definition of $`T^{}f`$, a left action $`\mathrm{\Phi }`$ of $`G`$ on the manifold $`Q`$ induces for each $`gG`$ the cotangent lift of $`\mathrm{\Phi }_g:QQ`$. That is: we have the map
$$T^{}\mathrm{\Phi }_gT^{}(\mathrm{\Phi }_g):T^{}QT^{}Q,\mathrm{with}\alpha T_q^{}Q(T^{}\mathrm{\Phi }_g)(\alpha )T_{g^1q}^{}Q.$$
(4.123)
Now consider the map assigning to each $`gG`$, $`T^{}\mathrm{\Phi }_g`$:
$$gGT^{}\mathrm{\Phi }_g:T^{}QT^{}Q.$$
(4.124)
To check that this is indeed an action of $`G`$ on $`T^{}Q`$, we first check that since $`\mathrm{\Phi }_e=id_Q`$, $`T\mathrm{\Phi }_e:TQTQ`$ is $`id_{TQ}`$ and $`T^{}(\mathrm{\Phi }_e)`$ is $`id_{T^{}Q}`$. But beware: eq. 4.122 yields
$$T^{}\mathrm{\Phi }_{gh}=T^{}(\mathrm{\Phi }_g\mathrm{\Phi }_h)=T^{}\mathrm{\Phi }_hT^{}\mathrm{\Phi }_g,$$
(4.125)
so that eq. 4.124 defines a right action.
But here we recall that any left action defines a right action by using the inverse; cf. eq. 4.117. Combining this with the idea of the cotangent lift of an action on $`Q`$, we get:
The left action $`\mathrm{\Phi }`$ on $`Q`$ defines, not only the right action eq. 4.124 on $`T^{}Q`$, but also a left action on $`T^{}Q`$, viz. by
$$gG\mathrm{\Psi }_g:=T^{}(\mathrm{\Phi }_{g^1}):T^{}QT^{}Q.$$
(4.126)
For since $`(gh)^1=h^1g^1`$,
$$\mathrm{\Psi }_{gh}T^{}(\mathrm{\Phi }_{(gh)^1})=T^{}(\mathrm{\Phi }_{h^1g^1})=T^{}(\mathrm{\Phi }_{h^1}\mathrm{\Phi }_{g^1})=T^{}\mathrm{\Phi }_{g^1}T^{}\mathrm{\Phi }_{h^1}\mathrm{\Psi }_g\mathrm{\Psi }_h.$$
(4.127)
In short, the two reversals of order cancel out. This sort of left-right flip will recur in some important contexts in the following, in particular in Sections 6.5 and 7.
(3): Mechanical examples:—
So much by way of generalities. Now we apply them to translations and rotations of a free particle, to rotations of a pivoted rigid body, and to $`N`$ point-particles.
(vi): Let the translation group $`G=(\mathrm{I}\mathrm{R}^3,+)`$ act on the free particle’s configuration space $`Q=\mathrm{I}\mathrm{R}^3`$ by
$$\mathrm{\Phi }_𝐱(𝐪)=𝐪+𝐱.$$
(4.128)
Since $`G`$ is abelian, the distinction between left and right actions of $`G`$ collapses. (And if we identify $`G`$ with $`Q`$, this is left=right translation by $`\mathrm{I}\mathrm{R}^3`$ on itself, i.e. example (iv) again: and so transitive and free.) But of course the lifted actions we have defined, “with $`g`$” and “with $`g^1`$”, eq. 4.124 and 4.126 respectively, remain distinct actions.
Then, writing $`\alpha =(𝐪,𝐩)T_𝐪^{}Q`$, and using the fact that $`T\mathrm{\Phi }_𝐱(𝐪𝐱,\dot{𝐪})=(𝐪,\dot{𝐪})`$, we see that eq. 4.121 implies that: first,
$$T^{}(\mathrm{\Phi }_𝐱)(𝐪,𝐩)T_{𝐪𝐱}^{}Q;$$
(4.129)
and second, that for all $`\dot{𝐪}T_{𝐪𝐱}Q`$,
$$<T^{}(\mathrm{\Phi }_𝐱)(𝐪,𝐩);(𝐪𝐱,\dot{𝐪})>=<(𝐪,𝐩);(𝐪,\dot{𝐪})>𝐩(\dot{𝐪}).$$
(4.130)
For eq. 4.130 to hold for all $`\dot{𝐪}T_{𝐪𝐱}Q`$ requires that $`T^{}(\mathrm{\Phi }_𝐱)(𝐪,𝐩)`$ does not affect $`𝐩`$, i.e.
$$T^{}(\mathrm{\Phi }_𝐱)(𝐪,𝐩)=(𝐪𝐱,𝐩).$$
(4.131)
So this is the lifted action “with $`g`$”, corresponding to eq. 4.124. Similarly, the lifted action “with $`g^1`$”, corresponding to eq. 4.126, is: $`\mathrm{\Psi }_𝐱(𝐪,𝐩):=T^{}(\mathrm{\Phi }_𝐱)(𝐪,𝐩)=(𝐪+𝐱,𝐩)`$.
One readily checks that these lifted actions preserve both $`\omega =dq^idp^i`$ (an exercise in manipulating the exterior derivative) and $`H:=p^2/2m.`$ So we have a Hamiltonian symmetry group. The action is not transitive: the orbits are labelled by their values of $`𝐩(\mathrm{I}\mathrm{R}^3)^{}`$. But it is free.
(vii): Let $`SO(3)`$ act on the left on $`Q=\mathrm{I}\mathrm{R}^3`$ by
$$\mathrm{\Phi }_A(𝐪)=A𝐪.$$
(4.132)
(This is example (i) again.) Let us lift this action “with $`g`$”, i.e. eq. 4.124, so as to get a right action on $`T^{}Q`$.
As in example (vi), we write $`\alpha =(𝐪,𝐩)T_𝐪^{}Q`$. Using the fact that $`T\mathrm{\Phi }_A(𝐪,\dot{𝐪})=(A𝐪,A\dot{𝐪})`$, eq. 4.121 then implies that: first,
$$T^{}(\mathrm{\Phi }_A)(𝐪,𝐩)T_{A^1𝐪}^{}Q;$$
(4.133)
and second, that for all $`\dot{𝐪}T_{A^1𝐪}Q`$,
$$<T^{}(\mathrm{\Phi }_A)(𝐪,𝐩);(A^1𝐪,\dot{𝐪})>=<(𝐪,𝐩);(𝐪,A\dot{𝐪})>𝐩(A\dot{𝐪})p_iA_j^i\dot{q}^j.$$
(4.134)
For eq. 4.134 to hold for all $`\dot{𝐪}T_{A^1𝐪}Q`$ requires that
$$T^{}(\mathrm{\Phi }_A)(𝐪,𝐩)=(A^1𝐪,𝐩A),$$
(4.135)
where $`𝐩A`$ is a row-vector. Or if one thinks of the $`𝐩`$ components as a column vector, it requires:
$$T^{}(\mathrm{\Phi }_A)(𝐪,𝐩)=(A^1𝐪,\stackrel{~}{A}𝐩)=(A^1𝐪,A^1𝐩),$$
(4.136)
where $`\stackrel{~}{}`$ represents the transpose of a matrix, and the last equation holds because $`A`$ is an orthogonal matrix.
So this is the lifted action “with $`g`$”, corresponding to eq. 4.124. Similarly, the lifted action “with $`g^1`$”, corresponding to eq. 4.126, is: $`\mathrm{\Psi }_A(𝐪,𝐩):=T^{}(\mathrm{\Phi }_{A^1})(𝐪,𝐩)=(A𝐪,A𝐩)`$.
Again, one readily checks that these lifted actions preserve both $`\omega =dq^idp^i`$ (another exercise in manipulating the exterior derivative!) and $`H:=p^2/2m.`$ So $`SO(3)`$ is a Hamiltonian symmetry group.
Like the original action of $`SO(3)`$ on $`Q`$, these actions are faithful. But they are not transitive: the orbits are labelled by the radii of two spheres centred at the origins of $`\mathrm{I}\mathrm{R}^3`$ and $`(\mathrm{I}\mathrm{R}^3)^{}`$. And they are not free: suppose $`𝐪`$ and $`𝐩`$ are parallel and on the axis of rotation of $`A`$.
(viii): Now we consider the pivoted rigid body. But unlike examples (vi) and (vii), we will consider only kinematics, not dynamics: even for a free body. That is, we will say nothing about the definitions of, and invariance of, $`\omega `$ and $`H`$; for details of these, cf. e.g. Abraham and Marsden (1978: Sections 4.4 and 4.6) and the other references given in (3) of Section 2.2. We will in any case consider the dynamics of this example in more general terms (using momentum maps) in Sections 6.5.3 and 7.
We recall from the discussion at the end of Section 3.4.4 that the configuration space of the pivoted rigid body is $`SO(3)=:G`$. We also saw there that the space and body representations of the angular velocity $`v=\dot{g}T_gG`$ are given by right and left translation. Thus eq. 3.113 and 3.114 give:
$$v^S\dot{g}^S:=T_gR_{g^1}(\dot{g})\mathrm{and}v^B\dot{g}^B:=T_gL_{g^1}(\dot{g}).$$
(4.137)
But we are now concerned with the cotangent lift of left (or right) translation. So let $`SO(3)`$ act on itself by left translation: $`\mathrm{\Phi }_ghL_gh=gh`$. Let us lift this action “with $`g`$”, i.e. eq. 4.124, to get a right action on $`T^{}G`$. So let $`\alpha T_h^{}G`$ and $`(TL_g)(h,\dot{h})=(gh,g\dot{h})`$. Then eq. 4.121 implies that: first
$$(T^{}L_g)(\alpha )T_{g^1h}^{}G,$$
(4.138)
and second that for all $`vT_{g^1h}G`$
$$<T^{}(L_g)(\alpha );v>=<\alpha ;gv>.$$
(4.139)
In other words, on analogy with eq. 4.131 and 4.135: for eq. 4.139 to hold for all $`vT_{g^1h}G`$ requires that with $`gvT_hG`$:
$$T^{}(L_g)(\alpha ):vT_{g^1h}G\alpha (gv).$$
(4.140)
Similarly, the lifted action “with $`g^1`$” corresponding to eq. 4.126, i.e. the left action on $`T^{}G`$, is
$$<T^{}(L_{g^1})(\alpha );v>=<\alpha ;g^1v>,\alpha T_h^{}G,vT_{gh}G$$
(4.141)
We will continue this example in Section 4.6, after developing more of the theory of Lie group actions.
Finally, let us sketch another mechanical example: the case of $`N`$ particles with configuration space $`Q:=\mathrm{I}\mathrm{R}^{3N}`$ interacting by Newtonian gravity—discussed in Section 2.3.2. This will combine and generalize examples (vi) and (vii); and lead on to the next Sections’ discussions of orbits and quotients.
(ix): As I mentioned above (before eq. 4.120), the cotangent-lifted actions of translations and rotations on a single particle carry over straightforwardly to the case of $`N`$ particles: the action defined on a single particle is just repeated, component by component, for each of the $`N`$ particles to give an action on $`T^{}Q\mathrm{I}\mathrm{R}^{3N}\times (\mathrm{I}\mathrm{R}^{3N})^{}`$.
Furthermore, the groups of translations and rotations are subgroups of a single group, the Euclidean group $`E`$. I shall not define $`E`$ exactly. Here, let it suffice to say that:
(a): $`E`$’s component-wise action on the configuration space $`Q:=\mathrm{I}\mathrm{R}^{3N}`$ has a cotangent lift, which is of course also component by component.
(b): $`E`$’s cotangent-lifted action is not transitive, nor free; but it is faithful.
(c): If we take as the Hamiltonian function the $`H`$ of eq. 2.25, describing the particles as interacting by Newtonian gravity, then $`E`$ is a Hamiltonian symmetry group. In fact, the kinetic and potential energies are separately invariant, essentially because the particles’ interaction depends only on the inter-particle distances, not on their positions or orientations; cf. the discussion in Section 2.3.2.
A final comment about example (ix), which points towards the following Sections:—
Recall that in Sections 2.3.3 and 2.3.4, we used this example as a springboard to discussing Relationist and Reductionist procedures, which quotiented the configuration space or phase space. But in order for the quotient spaces (and orbits) to be manifolds, and in particular for dimensions to add or subtract in a simple way, we needed to excise two classes of “special” points, before quotienting. These were: the class of symmetric configurations or states (i.e. those fixed by some element of $`E`$), and the class of collision configurations or states. For the quotienting of phase space advocated by Reductionism, the classes of states were $`\delta T^{}\mathrm{I}\mathrm{R}^{3N}`$ and $`\mathrm{\Delta }T^{}\mathrm{I}\mathrm{R}^{3N}`$; (cf. Section 2.3.4 for definitions.)
With examples (vi) to (ix) in hand, we can now see that:
(a): $`\delta `$ and $`\mathrm{\Delta }`$ are each closed under the cotangent-lifted action of $`E`$ on $`T^{}\mathrm{I}\mathrm{R}^{3N}`$; i.e., each is a union of orbits. So $`E`$ acts on $`M:=T^{}\mathrm{I}\mathrm{R}^{3N}(\delta \mathrm{\Delta })`$.
(b): $`E`$ acts freely on $`M`$.
We will see in the sequel (especially in Sections 4.3.B and 5.5) that an action being free is one half (one conjunct) of an important sufficient condition for orbits and quotient spaces to be manifolds. The other conjunct will be the notion of an action being proper: which we will define in Section 4.3.
### 4.2 Quotient structures from group actions
In finite dimensions, any orbit $`\mathrm{Orb}(x)`$ is an immersed submanifold of $`M`$. This can be proved directly (Abraham and Marsden (1978: Ex. 1.6F(b), p. 51, and 4.1.22 p. 265)). But for our purposes, this is best seen as a corollary of some conditions under which quotient structures are manifolds; as follows.
The relation, $`xy`$ if there is a $`gG`$ such that $`gx=y`$, is an equivalence relation, with the orbits as equivalence classes. We denote the quotient space, i.e. the set of orbits, by $`M/G`$ (sometimes called the orbit space). We write the canonical projection as
$$\pi :MM/G,x\mathrm{Orb}(x);$$
(4.142)
and we give $`M/G`$ the quotient topology by defining $`UM/G`$ to be open iff $`\pi ^1(U)`$ is open in $`M`$.
Simple examples (e.g. (ii) of Section 4.1.A) show that this quotient topology need not be Hausdorff. However, it is easy to show that if the set
$$R:=\{(x,\mathrm{\Phi }_gx)M\times M:(g,x)G\times M\}$$
(4.143)
is a closed subset of $`M\times M`$, then the quotient topology on $`M/G`$ is Hausdorff.
But to ensure that $`M/G`$ has a manifold structure, further conditions are required. The main one (and a much harder theorem) is:
> $`R`$ is a closed submanifold of $`M\times M`$ iff $`M/G`$ is a manifold with $`\pi :MM/G`$ a submersion.
This theorem has two Corollaries which are important for us.
(1): A map $`h:M/GN`$, from the manifold $`M/G`$, for which $`\pi :MM/G`$ is a submersion, to the manifold $`N`$, is smooth iff $`h\pi :MN`$ is smooth.
This corollary has a useful implication, called passage to the quotients, about the notion of equivariance—which will be important in symplectic reduction.
A smooth map $`f:MN`$ is called equivariant if it respects the action of a Lie group $`G`$ on the manifolds. That is: Let $`G`$ act on $`M`$ and $`N`$ by $`\mathrm{\Phi }_g:MM`$ and $`\mathrm{\Psi }_g:NN`$ respectively. $`f:MN`$ is called equivariant with respect to these actions if for all $`gG`$
$$f\mathrm{\Phi }_g=\mathrm{\Psi }_gf.$$
(4.144)
That is, $`f`$ is equivariant iff for all $`g`$, the following diagram commutes:
$$\begin{array}{c}M\\ \mathrm{\Phi }_g\\ M\end{array}\begin{array}{c}\stackrel{f}{}\\ \\ \stackrel{f}{}\end{array}\begin{array}{c}N\\ \mathrm{\Psi }_g\\ N\end{array}$$
(4.145)
Equivariance immediately implies that $`f`$ naturally induces a map, $`\widehat{f}`$ say, on the quotients. That is: the map
$$\widehat{f}:\mathrm{Orb}(x)M/G\mathrm{Orb}(f(x))N/G$$
(4.146)
is well-defined, i.e. independent of the chosen representative $`x`$ for the orbit.
Applying the corollary we have: If $`f:MN`$ is equivariant, and the quotients $`M/G`$ and $`M/N`$ are manifolds with the canonical projections both submersions, then $`f`$ being smooth implies that $`\widehat{f}`$ is smooth. This is called passage to the quotients.
(2): Let $`H`$ be a closed subgroup of the Lie group $`G`$. (By (2) of Section 3.4.3, this is equivalent to $`H`$ being a subgroup that is a submanifold of $`G`$.) Let $`H`$ act on $`G`$ by left translation: $`(h,g)H\times GhgG`$, so that the orbits are the right cosets $`Hg`$. Then $`G/H`$ is a manifold and $`\pi :GG/H`$ is a submersion.
### 4.3 Proper actions
By adding to the Section 4.2’s main theorem (i.e., $`R`$ is a closed submanifold of $`M\times M`$ iff $`M/G`$ is a manifold with $`\pi :MM/G`$ a submersion), the notion of a proper action we can give useful sufficient conditions for:
(A): orbits to be submanifolds;
(B): $`M/G`$ to be a manifold.
An action $`\mathrm{\Phi }:G\times MM`$ is called proper if the map
$$\stackrel{~}{\mathrm{\Phi }}:(g,x)G\times M(x,\mathrm{\Phi }(g,x))M\times M$$
(4.147)
is proper. By this we mean that if $`\{x_n\}`$ is a convergent sequence in $`M`$, and $`\{\mathrm{\Phi }_{g_n}(x_n)\}`$ is a convergent sequence in $`M`$, then $`\{g_n\}`$ has a convergent subsequence in $`G`$. In finite dimensions, this means that compact sets have compact inverse images; i.e. if $`KM\times M`$ is compact, then $`\stackrel{~}{\mathrm{\Phi }}^1(K)`$ is compact.
If $`G`$ is compact, this condition is automatically satisfied. Also, the action of a group on itself by left (or by right) translation (Example (iv) of Section 4.1.A) is always proper. Furthermore, the cotangent lift of left (or right) translation ((2) and Example (viii) of Section 4.1.A) is always proper. We shall not prove this, but it will be important in the sequel.
##### 4.3.A Isotropy groups; orbits as manifolds
For $`xM`$ the isotropy (or stabilizer or symmetry) group of $`\mathrm{\Phi }`$ at $`x`$ is
$$G_x:=\{gG:\mathrm{\Phi }_g(x)\mathrm{\Phi }(g,x)=x\}G.$$
(4.148)
(So an action is free iff for all $`xM`$, $`G_x=\{e\}`$.)
So if we define
$$\mathrm{\Phi }^x:GM:\mathrm{\Phi }^x(g):=\mathrm{\Phi }(g,x)$$
(4.149)
we have: $`G_x=(\mathrm{\Phi }^x)^1(x)`$. (The notation $`\mathrm{\Phi }^x`$ is a “cousin” of the notation $`\mathrm{\Phi }_g`$ defined in eq. 4.116.)
So since $`\mathrm{\Phi }^x`$ is continuous, $`G_x`$ is a closed subgroup of $`G`$. So, by the result in (2) of Section 3.4.3 (i.e. the result before eq. 3.97), $`G_x`$ is a submanifold (as well as Lie subgroup) of $`G`$. And if the action is proper, $`G_x`$ is compact.
Furthermore, the fact that for all $`hG_x`$ we have $`\mathrm{\Phi }^x(gh)=\mathrm{\Phi }_g\mathrm{\Phi }_h(x)=\mathrm{\Phi }_g(x)`$, implies that $`\mathrm{\Phi }^x`$ naturally induces a map
$$\stackrel{~}{\mathrm{\Phi }}^x:[g]=gG_xG/G_x\mathrm{\Phi }_gx\mathrm{Orb}(x)M.$$
(4.150)
That is, this map is well-defined. $`\stackrel{~}{\mathrm{\Phi }}^x`$ is injective because if $`\mathrm{\Phi }_gx=\mathrm{\Phi }_hx`$ then $`g^1hG_x`$, so that $`gG_x=hG_x`$.
It follows from Section 4.2’s main theorem (i.e., $`R`$ is a closed submanifold of $`M\times M`$ iff $`M/G`$ is a manifold with $`\pi :MM/G`$ a submersion) that:
(a): If $`\mathrm{\Phi }:G\times MM`$ is an action and $`xM`$, then $`\stackrel{~}{\mathrm{\Phi }}^x`$ defined by eq. 4.150 is an injective immersion.
Here we recall from Section 3.3.1 that injective immersions need not be embeddings. But:—
(b): If also $`\mathrm{\Phi }`$ is proper, the orbit $`\mathrm{Orb}(x)`$ is a closed submanifold of $`M`$ and $`\stackrel{~}{\mathrm{\Phi }}^x`$ is a diffeomorphism. In other words: the manifold structure of $`\mathrm{Orb}(x)`$ is given by the bijective map $`[g]G/G_xgx\mathrm{Orb}(x)`$ being a diffeomorphism.
Examples:—
(We use the numbering of corresponding examples in Section 4.1.A):—
(i): $`G=SO(3)`$ acts on $`M=\mathrm{I}\mathrm{R}^3`$ by $`(A,x)Ax`$. Since $`\mathrm{Orb}(x)`$ is a sphere centred at the origin of radius $`x`$, $`M/G\mathrm{I}\mathrm{R}^+`$: which is not a manifold. But results (a) and (b) are illustrated: the isotropy group $`G_x`$ at $`x`$ is the group of rotations with $`x`$ on the axis; the action is proper (for $`G`$ is compact); the orbit $`\mathrm{Orb}(x)`$ is a closed manifold of $`M`$; and the isotropy group’s cosets $`[g]G/G_x`$ are mapped diffeomorphically by $`\stackrel{~}{\mathrm{\Phi }}^x`$ to points on the sphere $`\mathrm{Orb}(x)`$.
(iii): Let $`X`$ be the constant vector field $`_x`$ on $`M=\mathrm{I}\mathrm{R}^3`$. $`X`$ is complete. The action of $`\mathrm{I}\mathrm{R}`$ on $`M`$ has as orbit through the point $`𝐱=(x,y,z)\mathrm{I}\mathrm{R}^3`$, the line $`y=`$ constant, $`z=`$ constant. The action is free, and therefore faithful and the isotropy groups are trivial. So $`G/G_xG`$. The action is proper. Again results (a) and (b) are illustrated: the orbits $`\mathrm{Orb}(𝐱)`$ are closed submanifolds of $`M`$, viz. copies of the real line $`\mathrm{I}\mathrm{R}=GG/G_x`$ that are diffeomorphic to $`\mathrm{I}\mathrm{R}`$ by $`\stackrel{~}{\mathrm{\Phi }}^𝐱`$.
##### 4.3.B A sufficient condition for the orbit space $`M/G`$ to be a manifold
With result (b) from the end of Section 4.3.A,, we can prove that:
> If $`\mathrm{\Phi }:G\times MM`$ is a proper free action, then the orbit space $`M/G`$ is a manifold with $`\pi :MM/G`$ a submersion.
Examples: (again using the numbering in Section 4.1.A):—
(i): $`G=SO(3)`$ acts on $`M=\mathrm{I}\mathrm{R}^3`$ by $`(A,x)Ax`$. Since $`\mathrm{Orb}(x)`$ is a sphere centred at the origin of radius $`x`$, $`M/G\mathrm{I}\mathrm{R}^+`$: which is not a manifold, and indeed the action is not free.
(iii): Let $`X`$ be the constant vector field $`_x`$ on $`M=\mathrm{I}\mathrm{R}^3`$. $`X`$ is complete, and the action of $`\mathrm{I}\mathrm{R}`$ on $`M`$ has as orbits the lines $`y=`$ constant, $`z=`$ constant. The action is faithful, free and proper, so that the orbit space $`M/G`$ is a manifold: $`M/G\mathrm{I}\mathrm{R}^2`$.
(iv): Left (or right) translation is obviously a free action of a group $`G`$ on itself, and we noted above that it is proper. But since it is transitive, the orbit space $`G/G`$ is the trivial 0-dimensional manifold (the singleton set of $`G`$).
(viii): The cotangent lift of left (or right) translation by $`SO(3)`$, or more generally, by a Lie group $`G`$. This action is proper (noted after eq. 4.147), and obviously free.
(ix): The Euclidean group $`E`$ acts freely on $`M:=T^{}\mathrm{I}\mathrm{R}^{3N}(\delta \mathrm{\Delta })`$. This action is also proper: a (harder!) exercise for the reader.
### 4.4 Infinitesimal generators of actions
We now connect this Subsection’s topic, group actions, with the Lie algebra of the Lie group concerned, i.e. with the topic of Section 3.4, especially 3.4.2.
Let $`\mathrm{\Phi }:G\times MM`$ be a (left) action by the Lie group $`G`$ on a manifold $`M`$. Then each $`\xi 𝔤`$ defines an action of $`\mathrm{I}\mathrm{R}`$ on $`M`$, which we write as $`\mathrm{\Phi }^\xi `$, in the following way.
We can think either in terms of exponentiation of $`\xi `$’s corresponding left-invariant vector field $`X_\xi `$ (cf. eq. 3.36 and 3.75); or in terms of of exponentiating $`\xi `$ itself (cf. eq. 3.78 and 3.79):
$$\mathrm{\Phi }^\xi :\mathrm{I}\mathrm{R}\times MM:\mathrm{\Phi }^\xi (\tau ,x):=\mathrm{\Phi }(\mathrm{exp}(\tau X_\xi ),x)\mathrm{\Phi }(\mathrm{exp}(\tau \xi ),x).$$
(4.151)
That is, in terms of our subscript notation for the original action $`\mathrm{\Phi }`$ (cf. eq. 4.116): $`\mathrm{\Phi }_{\mathrm{exp}(\tau X_\xi )}\mathrm{\Phi }_{\mathrm{exp}(\tau \xi )}:MM`$ is a flow on $`M`$.
That the flow is complete, i.e. that an action of all of $`\mathrm{I}\mathrm{R}`$ is defined, follows from (2) Exponentiation again of Section 3.4.2, especially after eq. 3.76. Cf. also example (iii) of Section 4.1.
We say that the corresponding vector field on $`M`$, written $`\xi _M`$, i.e. the vector field defined at $`xM`$ by
$$\xi _M(x):=\frac{d}{d\tau }_{\tau =0}\mathrm{\Phi }_{\mathrm{exp}(\tau X_\xi )}(x)\frac{d}{d\tau }_{\tau =0}\mathrm{\Phi }_{\mathrm{exp}(\tau \xi )}(x)$$
(4.152)
is the infinitesimal generator of the action corresponding to $`\xi `$.
In terms of the map $`\mathrm{\Phi }^x`$ defined in eq. 4.149, we have that for all $`\xi 𝔤`$
$$\xi _M(x)=(T_e\mathrm{\Phi }^x)(\xi ).$$
(4.153)
So NB: the words ‘infinitesimal generator’ are used in different, though related, ways. In Remark (2) at the end of Section 3.4.2, a vector field on the group $`G`$, or an element $`\xi 𝔤`$, was called an ‘infinitesimal generator’. Here the infinitesimal generator is a vector field on the action-space $`M`$. Similarly, beware the notation: $`\xi _M`$ is a vector field on $`M`$, while $`X_\xi `$ is a vector field on $`G`$.
As an example, we again take the rotation group $`SO(3)`$ acting on $`\mathrm{I}\mathrm{R}^3`$: $`(A,𝐱)SO(3)\times \mathrm{I}\mathrm{R}A𝐱`$. One readily checks that with $`\omega \mathrm{I}\mathrm{R}^3`$, so that $`\mathrm{\Theta }(\omega )𝔰𝔬(3)`$, the infinitesimal generator of the action corresponding to $`\xi \mathrm{\Theta }(\omega )`$ is the vector field on $`\mathrm{I}\mathrm{R}^3`$
$$\xi _{\mathrm{I}\mathrm{R}^3}(𝐱)(\mathrm{\Theta }(\omega ))_{\mathrm{I}\mathrm{R}^3}(𝐱)=\omega 𝐱.$$
(4.154)
In particular, the vector field on $`\mathrm{I}\mathrm{R}^3`$ representing infinitesimal anti-clockwise rotation about the $`x`$-axis is $`e_1:=y_zz_y`$ (cf. eq. 3.107). Similarly, the infinitesimal generators of the action of rotating about the $`y`$ axis and about the $`z`$-axis are, respectively: $`e_2:=z_xx_z`$ and $`e_3:=x_yy_x`$. The Lie brackets are given by:
$$[e_1,e_2]=e_3[e_3,e_1]=e_2[e_2,e_3]=e_1.$$
(4.155)
The minus signs here are a general feature of the transition $`\xi 𝔤\xi _M𝒳(M)`$; cf. result (4) below.
As another example, we take the infinitesimal generator of left and right translation on the group $`G`$. (We will need this example for our theorems about symplectic reduction; cf. Sections 6.5.3, 7.2 and 7.3.3.) NB: There will be a “left-right flip” here, which continues the discussion in (4) of Section 3.4.2.B, comparing using left-invariant vs. right-invariant vector fields to define the Lie algebra of a Lie group.
For left translation $`\mathrm{\Phi }(g,h)L_gh:=gh`$, we have for all $`\xi 𝔤`$:
$$\mathrm{\Phi }^\xi (\tau ,h)=(\mathrm{exp}\tau \xi )h=R_h(\mathrm{exp}\tau \xi );$$
(4.156)
so that the infinitesimal generator is
$$\xi _G(g)=(T_eR_g)\xi .$$
(4.157)
So $`\xi _G`$ is a right-invariant vector field; and unless $`G`$ is abelian, it is not equal to the left-invariant vector field $`gX_\xi (g):=(T_eL_g)\xi `$; cf. eq. 3.68 and 3.70.
Similarly, for right translation (which is a right action, cf. (1) (iv) in Section 4.1.A), the infinitesimal generator is the left-invariant vector field
$$gX_\xi (g):=(T_eL_g)\xi .$$
(4.158)
Three straightforward results connect the notion of an infinitesimal generator with previous ideas. I will not give proofs, but will present them in the order of the previous ideas.
(1): Recall the correspondence between Lie subgroups and Lie subalgebras, at the end of Section 3.4.3; eq. 3.97. This implies that the Lie algebra of the isotropy group $`G_x,xM`$ (called the isotropy algebra), is
$$𝔤_x=\{\xi 𝔤:\xi _M(x)=0\}.$$
(4.159)
(2): Infinitesimal generators $`\xi _M`$ give a differential version of the notion of equivariance, discussed in (1) of Section 4.2: a version called infinitesimal equivariance.
In eq. 4.144, we set $`g=\mathrm{exp}(\tau \xi )`$ and differentiate with respect to $`\tau `$ at $`\tau =0`$. This gives $`Tf\xi _M=\xi _Nf`$. That is: $`\xi _M`$ and $`\xi _N`$ are $`f`$-related. In terms of the pullback $`f^{}`$ of $`f`$, we have: $`f^{}\xi _N=\xi _M`$.
(3): Suppose the action $`\mathrm{\Phi }`$ is proper, so that by result (b) at the end of Section 4.3.A: the orbit Orb($`x`$) of any point $`xM`$ is a (closed) submanifold of $`M`$. Then the tangent space to Orb($`x`$) at a point $`y`$ in Orb($`x`$) is
$$T\mathrm{Orb}(x)_y=\{\xi _M(y):\xi 𝔤\}.$$
(4.160)
Finally, there is a fourth result relating infinitesimal generators $`\xi _M`$ to previous ideas; as follows. (But it is less straightforward than the previous (1)-(3): its proof requires the notion of the adjoint representation, described in the next Section.)
(4): The infinitesimal generator map $`\xi \xi _M`$ establishes a Lie algebra anti-homomorphism between $`𝔤`$ and the Lie algebra $`𝒳_M`$ of all vector fields on $`M`$. (Contrast the Lie algebra isomorphism between $`𝔤`$ and the set $`𝒳_L(G)`$ of left-invariant vector fields on the group $`G`$; Section 3.4.2 especially eq. 3.70.) That is:
$$(a\xi +b\eta )_M=a\xi _M+b\eta _M;[\xi _M,\eta _M]=[\xi ,\eta ]_M\xi ,\eta 𝔤,\mathrm{and}a,b\mathrm{I}\mathrm{R}.$$
(4.161)
Incidentally, returning to (4) of Section 3.4.2.B, which considered defining the Lie algebra of a Lie group in terms of right-invariant vector fields, instead of left-invariant vector fields: had we done so, the corresponding map $`\xi \xi _M`$ would have been a Lie algebra homomorphism.
### 4.5 The adjoint and co-adjoint representations
A leading idea of later Sections (especially Sections 5.4, 6.4 and 7) will be that there is a natural symplectic structure in the orbits of a certain natural representation of any Lie group: namely a representation of the group on the dual of its own Lie algebra, called the co-adjoint representation. Here we introduce this representation. But we lead up to it by first describing the adjoint representation of a Lie group on its own Lie algebra. Even apart from symplectic structure (and so applications in mechanics), both representations illustrate the ideas of previous Subsections. I will again use $`SO(3)`$ and $`𝔰𝔬(3)`$ as examples.
#### 4.5.1 The adjoint representation
We proceed in four stages. We first define the representation, then discuss infinitesimal generators, then discuss matrix Lie groups, and finally discuss the rotation group.
(1): The representation defined:—
Let $`G`$ be a Lie group and $`𝔤`$ its Lie algebra, i.e. the tangent space to the group at the identity $`eG`$, equipped with the commutator bracket operation $`[,]`$.
Recall (e.g. from the beginning of Section 3.4.2) that $`G`$ acts on itself by left and right translation: each $`gG`$ defines diffeomorphisms of $`G`$ onto itself by
$$L_g:hGghG;R_g:hGhgG.$$
(4.162)
The induced maps of the tangent spaces are, for each $`hG`$:
$$L_g:TG_hTG_{gh}\text{ and}R_g:TG_hTG_{hg}.$$
(4.163)
The diffeomorphism $`K_g:=R_{g^1}L_g`$ (i.e. conjugation by $`g,K_g:hghg^1`$) is an inner automorphism of $`G`$. (Cf. example (v) at the end of Section 4.1.) Its derivative at the identity $`eG`$ is a linear map from the Lie algebra $`𝔤`$ to itself, which is denoted:
$$Ad_g:=(R_{g^1}L_g)_e:𝔤𝔤.$$
(4.164)
So letting $`g`$ vary through $`G`$, the map $`Ad:gAd_g`$ assigns to each $`g`$ a member of End($`𝔤`$), the space of linear maps on (endomorphisms of) $`𝔤`$. The chain rule implies that $`Ad_{gh}=Ad_gAd_h`$. So
$$Ad:gAd_g$$
(4.165)
is a left action, a representation, of $`G`$ on $`𝔤`$: $`G\times 𝔤𝔤`$. It is called the adjoint representation.
Three useful results about $`Ad`$ follow from our results (1) and (3) in Section 3.4.2.B (cf. eq. 3.80: Homomorphisms respect exponentiation):
: If $`\xi 𝔤`$ generates the one-parameter subgroup $`H=\{\mathrm{exp}(\tau X_\xi ):\tau \mathrm{I}\mathrm{R}\}`$, then $`Ad_g(\xi )`$ generates the conjugate subgroup $`K_g(H)=gHg^1`$.
$$\mathrm{exp}(Ad_g(\xi ))=K_g(\mathrm{exp}\xi ):=g(\mathrm{exp}\xi )g^1.$$
(4.166)
Incidentally, eq. 4.166 has a many-parameter generalization. Let $`H`$ and $`H^{}`$ be two connected $`r`$-dimensional Lie subgroups of the Lie group $`G`$, with corresponding Lie subalgebras $`𝔥`$ and $`𝔥^{}`$ of the Lie algebra $`𝔤=𝔤(G)`$. Then $`H`$ and $`H^{}`$ are conjugate subgroups, $`H^{}=gHg^1`$, iff $`𝔥`$ and $`𝔥^{}`$ are corresponding conjugate subalgebras, i.e. $`𝔥^{}=Ad_g(𝔥)`$.
: Eq. 4.166 also implies another result which will be needed for a crucial result about symplectic reduction, in Section 6.5.2. (The many-parameter generalization just mentioned will not be needed.) It relates $`Ad`$ to the pullback of an arbitrary action $`\mathrm{\Phi }`$.
Thus let $`\mathrm{\Phi }`$ be a left action of $`G`$ on $`M`$. Then for every $`gG`$ and $`\xi 𝔤`$
$$(Ad_g\xi )_M=\mathrm{\Phi }_{g^1}^{}\xi _M,$$
(4.167)
where $`\mathrm{\Phi }^{}`$ indicates pullback of the vector field. For we have:
$`(Ad_g\xi )_M(x):={\displaystyle \frac{d}{d\tau }}_{\tau =0}\mathrm{\Phi }(\mathrm{exp}(\tau Ad_g\xi ),x)`$ (4.168)
$`={\displaystyle \frac{d}{d\tau }}_{\tau =0}\mathrm{\Phi }(g(\mathrm{exp}\tau \xi )g^1,x)\mathrm{by}\mathrm{eq}.\text{4.166}`$ (4.169)
$`={\displaystyle \frac{d}{d\tau }}_{\tau =0}(\mathrm{\Phi }_g\mathrm{\Phi }_{\mathrm{exp}\tau \xi }\mathrm{\Phi }_{g^1}(x))`$ (4.170)
$`=T_{\mathrm{\Phi }_{g^1}(x)}\mathrm{\Phi }_g(\xi _M(\mathrm{\Phi }_{g^1}(x)))\mathrm{by}\mathrm{the}\mathrm{chain}\mathrm{rule}\mathrm{and}\mathrm{eq}.\text{4.152}`$ (4.171)
$`=\left(\mathrm{\Phi }_{g^1}^{}\xi _M\right)(x)\mathrm{by}\mathrm{the}\mathrm{definition}\mathrm{of}\mathrm{pullback}.`$ (4.172)
Not only is this result needed later. Also, incidentally: it is the main part of the proof of result (4) at the end of Section 4.4, that $`\xi \xi _M`$ is a Lie algebra anti-homomorphism.
: $`Ad_g`$ is an algebra homomorphism, i.e.
$$Ad_g[\xi ,\eta ]=[Ad_g\xi ,Ad_g\eta ],\xi ,\eta 𝔤.$$
(4.173)
(2): Infinitesimal generators: the map $`ad`$:—
The map $`Ad`$ is differentiable. Its derivative at $`eG`$ is a linear map from the Lie algebra $`𝔤`$ to the space of linear maps on $`𝔤`$. This map is called $`ad`$, and its value for argument $`\xi 𝔤`$ is written $`ad_\xi `$. That is:
$$ad:=Ad_e:𝔤\text{End}𝔤;ad_\xi =\frac{d}{d\tau }_{\tau =0}Ad_{\mathrm{exp}(\tau \xi )}$$
(4.174)
where $`\mathrm{exp}(\tau \xi )`$ is the one-parameter subgroup with tangent vector $`\xi `$ at the identity. But if we apply the definition eq. 4.152 of the infinitesimal generator of an action, to the adjoint action $`Ad`$, we get that for each $`\xi 𝔤`$, the generator $`\xi _𝔤`$, i.e. a vector field on $`𝔤`$, is
$$\xi _𝔤:\eta 𝔤\xi _𝔤(\eta )𝔤\mathrm{with}\xi _𝔤(\eta ):=\frac{d}{d\tau }_{\tau =0}Ad_{\mathrm{exp}(\tau \xi )}(\eta ).$$
(4.175)
Comparing eq. 4.174, we see that $`ad_\xi `$ is just the infinitesimal generator $`\xi _𝔤`$ of the adjoint action corresponding to $`\xi `$:
$$ad_\xi =\xi _𝔤.$$
(4.176)
We now compute the infinitesimal generators of the adjoint action. It will be crucial to later developments (especially Section 5.4) that these are given by the Lie bracket in $`𝔤`$.
We begin by considering the function $`Ad_{\mathrm{exp}(\tau \xi )}(\eta )`$ to be differentiated. By eq. 4.164, we have
$`Ad_{\mathrm{exp}(\tau \xi )}(\eta )=T_e(R_{\mathrm{exp}(\tau \xi )}L_{\mathrm{exp}(\tau \xi )})(\eta )`$ (4.177)
$`=(T_{\mathrm{exp}(\tau \xi )}(R_{\mathrm{exp}(\tau \xi )})T_eL_{\mathrm{exp}(\tau \xi )})(\eta )`$
$`=(T_{\mathrm{exp}(\tau \xi )}(R_{\mathrm{exp}(\tau \xi )})X_\eta (\mathrm{exp}(\tau \xi ))`$
where the second line follows by the chain rule, and the third by definition of left-invariant vector field. Writing the flow of $`X_\xi `$ as $`\varphi _\tau (g)=g\mathrm{exp}\tau \xi =R_{\mathrm{exp}(\tau \xi )}g`$, and applying the definition of the Lie derivative (eq. 3.45), we then have
$`\xi _𝔤(\eta ):={\displaystyle \frac{d}{d\tau }}_{\tau =0}Ad_{\mathrm{exp}(\tau \xi )}(\eta )={\displaystyle \frac{d}{d\tau }}\left[T_{\varphi _\tau (e)}\varphi _\tau ^1X_\eta (\varphi _\tau (e))\right]_{\tau =0}`$ (4.178)
$`=[X_\xi ,X_\eta ](e)=[\xi ,\eta ].`$
where the final equation is the definition eq. 3.73 of the Lie bracket in the Lie algebra.
So for the adjoint action, the infinitesimal generator corresponding to $`\xi `$ is taking the Lie bracket: $`\eta [\xi ,\eta ]`$. To sum up: eq. 4.174 and 4.175 now become
$$ad=Ad_e:𝔤\text{End}𝔤;ad_\xi =\frac{d}{d\tau }_{\tau =0}Ad_{\mathrm{exp}(\tau \xi )}=\xi _𝔤:\eta 𝔤[\xi ,\eta ]𝔤.$$
(4.179)
(3): Example: matrix Lie groups:—
In the case where $`GGL(n,\mathrm{I}\mathrm{R})`$ is a matrix Lie group with Lie algebra $`𝔤𝔤𝔩(n)`$, these results are easy to verify. Writing $`n\times n`$ matrices as $`A,BG`$, conjugation is $`K_A(B)=ABA^1`$, and the adjoint map $`Ad`$ is also given by conjugation
$$Ad_A(X)=AXA^1,AG,X𝔤.$$
(4.180)
So with $`A(\tau )=\mathrm{exp}(\tau X)`$, so that $`A(0)=I`$ and $`A^{}(0)=X`$, we have with $`Y𝔤`$
$`{\displaystyle \frac{d}{d\tau }}_{\tau =0}Ad_{\mathrm{exp}\tau X}Y={\displaystyle \frac{d}{d\tau }}_{\tau =0}\left[A(\tau )YA(\tau )^1\right]`$ (4.181)
$`=A^{}(0)YA^1(0)+A(0)YA^1^{}(0).`$
But differentiating $`A(\tau )A^1(\tau )=I`$ yields
$$\frac{d}{d\tau }(A^1(\tau ))=A^1(\tau )A^{}(\tau )A^1(\tau ),\mathrm{and}\mathrm{so}A^1^{}(0)=A^{}(0)=X$$
(4.182)
so that indeed we have
$$\frac{d}{d\tau }_{\tau =0}Ad_{\mathrm{exp}\tau X}Y=XYYX=[X,Y].$$
(4.183)
(4): Example: the rotation group:—
It is worth giving details for the case of $`G=SO(3)`$, $`𝔤=𝔰𝔬(3)`$. We saw in Section 3.4.4 (eq. 3.107) that the three matrices
$$A^x=\left(\begin{array}{ccc}0& 0& 0\\ 0& 0& 1\\ 0& 1& 0\end{array}\right),A^y=\left(\begin{array}{ccc}0& 0& 1\\ 0& 0& 0\\ 1& 0& 0\end{array}\right),A^z=\left(\begin{array}{ccc}0& 1& 0\\ 1& 0& 0\\ 0& 0& 0\end{array}\right)$$
(4.184)
span $`𝔰𝔬(3)`$, and generate the one-parameter subgroups
$$R_\theta ^x=\left(\begin{array}{ccc}1& 0& 0\\ 0& \mathrm{cos}\theta & \mathrm{sin}\theta \\ 0& \mathrm{sin}\theta & \mathrm{cos}\theta \end{array}\right),R_\theta ^y=\left(\begin{array}{ccc}\mathrm{cos}\theta & 0& \mathrm{sin}\theta \\ 0& 1& 0\\ \mathrm{sin}\theta & 0& \mathrm{cos}\theta \end{array}\right),R_\theta ^z=\left(\begin{array}{ccc}\mathrm{cos}\theta & \mathrm{sin}\theta & 0\\ \mathrm{sin}\theta & \mathrm{cos}\theta & 0\\ 0& 0& 1\end{array}\right)$$
(4.185)
representing anticlockwise rotation around the respective coordinate axes in the physical space $`\mathrm{I}\mathrm{R}^3`$. To calculate the adjoint action of $`R_\theta ^x`$ on the generator $`A^y`$, we differentiate the product $`R_\theta ^xR_\tau ^yR_\theta ^x`$ with respect to $`\tau `$ and set $`\tau =0`$. That is, we find
$$Ad_{R_\theta ^x}(A^y)=R_\theta ^x(A^y)R_\theta ^x=\left(\begin{array}{ccc}0& \mathrm{sin}\theta & \mathrm{cos}\theta \\ \mathrm{sin}\theta & 0& 0\\ \mathrm{cos}\theta & 0& 0\end{array}\right)=\mathrm{cos}\theta A^y+\mathrm{sin}\theta A^z.$$
(4.186)
We similarly find
$$Ad_{R_\theta ^x}(A^x)=A^x,Ad_{R_\theta ^x}(A^z)=\mathrm{sin}\theta A^y+\mathrm{cos}\theta A^z.$$
(4.187)
So the adjoint action of the subgroup $`R_\theta ^x`$ representing rotations around the $`x`$-axis of physical space is given by rotations around the $`A^x`$-axis in the Lie algebra space $`𝔰𝔬(3)`$. Similarly for the other subgroups representing rotations around the $`y`$ or $`z`$-axis. And so for any rotation matrix $`RSO(3)`$, relative to given axes $`x,y,z`$ for $`\mathrm{I}\mathrm{R}^3`$, its adjoint map $`Ad_R`$ acting on $`𝔰𝔬(3)\mathrm{I}\mathrm{R}^3`$ has the same matrix representation relative to the induced basis $`\{A^x,A^y,A^z\}`$ of $`𝔰𝔬(3)`$. (NB: This agreement between $`SO(3)`$’s adjoint representation and its natural physical interpretation is special to $`SO(3)`$: it does not hold for other matrix Lie groups.)
Finally, the infinitesimal generators of the adjoint action are given by differentiation. For example, using eq. 4.186, we find that
$$ad_{A^x}(A^y):=\frac{d}{d\theta }_{\theta =0}Ad_{R_\theta ^x}A^y=A^z;$$
(4.188)
which agrees with the commutator: $`A^z=[A^x,A^y]`$.
#### 4.5.2 The co-adjoint representation
Again we proceed in stages. We first define the representation, then discuss infinitesimal generators, and then take the rotation group as an example.
(1): The representation defined:—
We recall that a linear map $`A:VW`$ induces (basis-independently) a transpose (dual) map, written $`A^{}`$ (or $`\stackrel{~}{A}`$ or $`A^T`$), $`A^{}:W^{}V^{}`$ on the dual spaces, $`V^{}:=\{\alpha :V\mathrm{I}\mathrm{R}\alpha \mathrm{linear}\}`$ and similarly for $`W^{}`$; by
$$\alpha W^{},vV:A^{}(\alpha )(v)<A^{}(\alpha );v>:=\alpha (A(v))(\alpha A)(v).$$
(4.189)
So any representation, $``$ say, of a group $`G`$ on a vector space $`V`$, $`:G\mathrm{End}(V)`$, induces a representation $`^{}`$ of $`G`$ on the dual space $`V^{}`$, by taking the transpose. We shall call $`^{}`$ the dual or transpose of $``$; it is also sometimes called a ‘contragredient representation’. That is: for $`(g):VV`$, we define $`^{}(g):V^{}V^{}`$ by
$$^{}(g):\alpha V^{}^{}(g)(\alpha ):=\alpha ((g))V^{}.$$
(4.190)
Thus the adjoint representation of $`G`$ on $`𝔤`$ induces a co-adjoint representation of $`G`$ on the dual $`𝔤^{}`$ of its Lie algebra $`𝔤`$, i.e. on the cotangent space to the group $`G`$ at the identity, $`𝔤^{}=T_e^{}G`$. The co-adjoint representation will play a central role in symplectic reduction (starting in Section 5.4).
So let $`Ad_g^{}:𝔤^{}𝔤^{}`$ be the dual (aka: transpose) of $`Ad_g`$, defined by
$$\alpha 𝔤^{},\xi 𝔤:<Ad_g^{}\alpha ;\xi >:=<\alpha ;Ad_g\xi >.$$
(4.191)
Since $`Ad:gAd_g`$ is a left action ($`Ad_{gh}=Ad_gAd_h`$), the assignment $`gAd_g^{}`$ is a right action. So to define a left action, we use the inverse $`g^1`$; cf. eq. 4.117 and 4.126. Namely, we define the left action
$$(g,\alpha )G\times 𝔤^{}Ad_{g^1}^{}\alpha 𝔤^{};$$
(4.192)
called the co-adjoint action of $`G`$ on $`𝔤^{}`$. And the corresponding co-adjoint representation of $`G`$ on $`𝔤^{}`$ is denoted by
$$Ad^{}:G\mathrm{End}(𝔤^{}),Ad_{g^1}^{}=(T_e(R_gL_{g^1}))^{}.$$
(4.193)
(2): The map $`ad^{}`$; infinitesimal generators:—
The map $`Ad^{}`$ is differentiable. Its derivative at $`eG`$ is a linear map from the Lie algebra $`𝔤`$ to the space of linear maps on $`𝔤^{}`$. This map is called $`ad^{}`$, and its value for argument $`\xi 𝔤`$ is written $`ad_\xi ^{}`$. Thus $`ad_\xi ^{}`$ is an endomorphism of $`𝔤^{}`$, and we have
$$ad^{}=Ad_e^{}:\xi 𝔤ad_\xi ^{}\text{End}𝔤^{}.$$
(4.194)
Now recall our deduction from eq. 4.174 and 4.175 that $`ad_\xi =\xi _𝔤`$, i.e. eq. 4.176. In the same way we here deduce an equality to the infinitesimal generator of the co-adjoint action:
$$ad_\xi ^{}=\xi _𝔤^{}.$$
(4.195)
In fact, $`ad_\xi ^{}`$ is, modulo a minus sign, the adjoint of $`ad_\xi `$, in the usual sense of the natural pairing of a vector space with its dual: as we now show. (So the notation $`ad^{}`$ is justified, modulo a minus sign.)
Let us compute for this action, the value of the infinitesimal generator $`\xi _𝔤^{}`$ (a vector field on $`𝔤^{}`$, induced by $`\xi 𝔤`$) at the point $`\alpha 𝔤^{}`$. That is, we will compute the value $`\xi _𝔤^{}(\alpha )`$. As usual, we identify the tangent space $`(T𝔤^{})_\alpha `$ in which this value lives, with $`𝔤^{}`$ itself; and similarly for $`𝔤`$. So, with $`\xi _𝔤^{}`$ acting on $`\eta 𝔤`$, we compute:
$`<ad_\xi ^{}(\alpha );\eta ><\xi _𝔤^{}(\alpha );\eta >={\displaystyle \frac{d}{d\tau }}_{\tau =0}Ad_{\mathrm{exp}\tau \xi }^{}(\alpha );\eta `$ (4.196)
$`={\displaystyle \frac{d}{d\tau }}_{\tau =0}Ad_{\mathrm{exp}\tau \xi }^{}(\alpha );\eta ={\displaystyle \frac{d}{d\tau }}_{\tau =0}\alpha ;Ad_{\mathrm{exp}\tau \xi }\eta `$ (4.197)
$`=\alpha ;{\displaystyle \frac{d}{d\tau }}_{\tau =0}Ad_{\mathrm{exp}\tau \xi }\eta =<\alpha ;[\xi ,\eta ]>=<\alpha ;ad_\xi (\eta )>.`$ (4.198)
So $`ad_\xi ^{}`$, defined as the derivative of $`Ad^{}`$ is, up to a sign, the adjoint of $`ad_\xi `$.
(3): Example: the rotation group:—
Let us now write the elementary vector product in $`\mathrm{I}\mathrm{R}^3`$ as $``$, and identify $`𝔰𝔬(3)(\mathrm{I}\mathrm{R}^3,)`$ and $`𝔰𝔬(3)^{}\mathrm{I}\mathrm{R}^3^{}`$. And let us have the natural pairing given by the elementary euclidean inner product $``$. Then the result just obtained (now with $``$ marking the argument-place)
$$<\xi _𝔤^{}(\alpha );>=<\alpha ;[\xi ,]>$$
(4.199)
becomes for $`\alpha 𝔰𝔬(3)^{}`$ and $`\xi so(3)`$
$$\xi _{𝔰𝔬(3)^{}}(\alpha )=\alpha (\xi ).$$
(4.200)
So for $`\eta 𝔰𝔬(3)`$, we have
$$<\xi _{𝔰𝔬(3)^{}}(\alpha );\eta >=\xi _{𝔰𝔬(3)^{}}(\alpha )\eta =\alpha (\xi \eta )=(\alpha \xi )\eta =<\alpha \xi ;\eta >.$$
(4.201)
In short:
$$\xi _{𝔰𝔬(3)^{}}(\alpha )=\alpha \xi =\xi \alpha .$$
(4.202)
Now since $`SO(3)`$ is compact, we know that the co-adjoint action is proper; so $`\mathrm{Orb}(\alpha )`$ is a closed submanifold of $`𝔰𝔬(3)^{}`$, and eq. 4.160 of Section 4.4 applies. So if we fix $`\alpha `$, and let $`\xi `$ vary through $`𝔰𝔬(3)\mathrm{I}\mathrm{R}^3`$, we get all of the tangent space $`T_\alpha \mathrm{Orb}(\alpha )`$ to the orbit passing through $`\alpha `$. Then eq. 4.202 implies that the tangent space is the plane normal to $`\alpha `$, and passing through $`\alpha `$’s end-point. Letting $`\alpha `$ vary through $`𝔰𝔬(3)^{}`$, we conclude that the co-adjoint orbits are the spheres centred on the origin.
In the following Sections, we will see that the orbits of the co-adjoint representation of any Lie group $`G`$ have a natural symplectic structure. So the orbits are always even-dimensional; and by considering all Lie groups and all possible orbits, we can get a series of examples of symplectic manifolds.
Besides, this fact will play a central role in our generalized formulation of Hamiltonian mechanics, and in symplectic reduction. And we will (mercifully!) get a good understanding of that role, already in Section 5.1. To prepare for that, it is worth gathering some threads about our recurrent example, $`SO(3)`$; and generalizing them to other Lie groups …
### 4.6 Kinematics on Lie groups
To summarize some aspects of this Section, and to make our later discussion of reduction clearer, it is worth collecting and generalizing some of our results about $`SO(3)`$ and the description it provides of the rigid body. More precisely, we will now combine:
(i): the description of space and body coordinates in terms of left and right translation, at the end of Section 3.4.4;
(ii): the cotangent lift of translation (example (viii) of Section 4.1.A);
(iii): the adjoint and co-adjoint representations of $`SO(3)`$ (as in (4) of Section 4.5.1, and (3) of Section 4.5.2.
We will also generalize: namely, we will consider (i) to (iii) for an arbitrary Lie group $`G`$, not just for $`SO(3)`$. (The point of doing so will become clear in (3) of Section 5.1.) This will occur already in Section 4.6.1. Then in Section 4.6.2, we will show how this material yields natural diffeomorphisms
$$TGG\times 𝔤\mathrm{and}T^{}GG\times 𝔤^{};$$
(4.203)
(so if $`\mathrm{dim}G=n`$, then all four manifolds are $`2n`$-dimensional). We will also see that by applying Section 4.2’s notion of equivariance, we can “pass to the quotients”, and get from eq. 4.203, the natural diffeomorphisms
$$TG/G𝔤\mathrm{and}T^{}G/G𝔤^{};$$
(4.204)
where the quotients on the left hand sides (the domains) is by the action of left translation; (to be precise: by the action of its derivative for $`TG`$, and its cotangent lift for $`T^{}G`$).
#### 4.6.1 Space and body coordinates generalized to $`G`$
So let a (finite-dimensional) Lie group $`G`$ act on itself by left and right translation, $`L_g`$ and $`R_g`$. For any $`gG`$, we define
$$\lambda _g:T_gG𝔤\mathrm{by}vT_gG(T_eL_g)^1(v)(T_gL_{g^1})(v)𝔤.$$
(4.205)
We similarly define
$$\rho _g:vT_gG(T_eR_g)^1(v)(T_gR_{g^1})(v)𝔤.$$
(4.206)
On analogy with the case of the pivoted rigid body (cf. eq. 3.113 and 3.114, or eq. 4.137), we say that $`\lambda _g`$ represents $`vT_gG`$ in body coordinates, and $`\rho _g`$ represents $`v`$ in space coordinates. We also speak of body and space representations. The transition from body to space coordinates is then an isomorphism of $`𝔤`$; viz. by eq. 4.164
$$\xi 𝔤,(\rho _g\lambda _g^1)(\xi )=\rho _g(T_eL_g(\xi ))Ad_g\xi .$$
(4.207)
So we can combine the $`S`$ and $`B`$ superscript notation of eq. 4.137 with Section 4.5.1’s notion of the adjoint representation, and write
$$v^S=Ad_gv^B.$$
(4.208)
In a similar way, the cotangent lifts of left and right translation provide isomorphisms between the dual spaces $`T_g^{}G,gG`$ and $`𝔤^{}`$. Thus for any $`gG`$, we define
$$\overline{}\lambda _g:T_g^{}G𝔤^{}\mathrm{by}\alpha T_g^{}G\alpha T_eL_g(T_eL_g)^{}(\alpha )(T_e^{}L_g)(\alpha )𝔤^{};$$
(4.209)
and similarly
$$\overline{\rho }_g:\alpha T_g^{}G\alpha T_eR_g(T_e^{}R_g)(\alpha )𝔤^{}.$$
(4.210)
And we again use the $`S`$ and $`B`$ superscript notation of eq. 4.137, and define for $`\alpha T_g^{}G`$
$$\alpha ^S:=(T_e^{}R_g)(\alpha )\overline{\rho }_g(\alpha )\mathrm{and}\alpha ^B:=(T_e^{}L_g)(\alpha )\overline{}\lambda _g(\alpha ),$$
(4.211)
which are called the space (or ‘spatial’) and body representations, respectively, of $`\alpha `$. The transition from body to space representations is now an isomorphism of $`𝔤^{}`$; viz.
$$\alpha 𝔤^{},(\overline{\rho }_g\overline{}\lambda _g^1)(\alpha )=Ad_{g^1}^{}(\alpha ),\mathrm{i}.\mathrm{e}.\alpha ^S=Ad_{g^1}^{}(\alpha ^B).$$
(4.212)
#### 4.6.2 Passage to the quotients
For later purposes, we need to develop the details of how the element $`gG`$ “carries along throughout” in eq. 4.205 to 4.212. More precisely, we have two isomorphisms:
$$TGG\times 𝔤\mathrm{and}T^{}GG\times 𝔤^{}.$$
(4.213)
These are isomorphisms of vector bundles; but we shall not develop the language of fibre bundles. What matters for us is that once we exhibit these isomorphisms, we will see that we have equivariant maps relating two group actions, in the sense of eq. 4.144 and 4.145. And this will mean that we can pass to the quotients to infer that $`TG/G`$ is diffeomorphic to $`𝔤`$, and correspondingly that $`T^{}G/G`$ is diffeomorphic to $`𝔤^{}`$.
This last diffeomorphism will form the first part of Section 7’s main theorem, the Lie-Poisson reduction theorem, which says that $`T^{}G/G`$ and $`𝔤^{}`$ are isomorphic as Poisson manifolds. In Section 5 onwards, we will develop the notion of a Poisson manifold, and the significance of this isomorphism for the reduction of mechanical problems.
I should note here that there is a parallel story about the first diffeomorphism, i.e. about $`TG/G`$ being diffeomorphic to $`𝔤`$. It forms the first part of another reduction theorem, which is the Lagrangian analogue of Section 7’s Lie-Poisson theorem. But since this Chapter has adopted the Hamiltonian approach, I will not go into details. They can be found in Marsden and Ratiu (1999: Sections 1.2, 13.5, 13.6), under the title ‘Euler-Poincaré reduction’.
Thus corresponding to eq. 4.205, we define the isomorphism
$$\lambda :TGG\times 𝔤\mathrm{by}\lambda (v):=(g,(T_eL_g)^1(v))(g,(T_gL_{g^1})(v))$$
(4.214)
with $`vT_gG`$, i.e. $`g=\pi _G(v)`$ and $`\pi _G:TGG`$ the canonical projection. (As mentioned concerning eq. 4.121, it is harmless to (follow many presentations and) conflate a point in $`TG`$, i.e. strictly speaking a pair $`(g,v),gG,vT_gG`$, with its vector $`v`$.) And corresponding to eq. 4.206, we define the isomorphism
$$\rho :TGG\times 𝔤\mathrm{by}\rho (v):=(g,(T_eR_g)^1(v))(g,(T_gR_{g^1})(v)).$$
(4.215)
The transition from body to space representations given by eq. 4.207 now implies
$$(\rho \lambda ^1)(g,\xi )=\rho (g,T_eL_g(\xi ))=(g,(T_eR_g)^1T_eL_g(\xi ))=(g,Ad_g\xi ).$$
(4.216)
In a similar way, the cotangent bundle $`T^{}G`$ is isomorphic in two ways to $`G\times 𝔤^{}`$: namely by
$$\overline{}\lambda (\alpha ):=(g,\alpha T_eL_g)(g,(T_e^{}L_g)\alpha )G\times 𝔤^{},$$
(4.217)
and by
$$\overline{\rho }(\alpha ):=(g,\alpha T_eR_g)(g,(T_e^{}R_g)\alpha )G\times 𝔤^{}$$
(4.218)
where $`\alpha T_g^{}G`$, i.e. $`g=\pi _G^{}(\alpha )`$ with $`\pi _G^{}:T^{}GG`$ the canonical projection. (Again, we harmlessly conflate a point $`(g,\alpha )`$ in $`T^{}G`$ with its form $`\alpha T_g^{}G`$.)
Let us now compute in the body representation, the actions of: (i) the (derivative of the) left translation map, $`TL_g`$, and (ii) the corresponding cotangent lift $`T^{}L_g`$. This will show that $`\lambda `$ and $`\overline{}\lambda `$ are equivariant maps for certain group actions.
(i): We compute:
$`(\lambda TL_g\lambda ^1)(h,\xi )=(\lambda TL_g)(h,TL_h(\xi ))=\lambda (gh,(TL_gTL_h)(\xi ))`$ (4.219)
$`=(gh,((TL_{gh})^1TL_{gh})(\xi ))=(gh,\xi ).`$ (4.220)
So in the body representation, left translation does not act on the vector component. (That is intuitive, in that the vector $`\xi `$ is “attached to the body” and so should not vary relative to coordinates fixed in it.) Eq. 4.220 means that $`\lambda `$ is an equivariant map relating left translation $`TL_g`$ on $`TG`$ to the $`G`$-action on $`G\times 𝔤`$ given just by left translation on the first component:
$$\mathrm{\Phi }_g((h,\xi ))g(h,\xi ):=(gh,\xi ).$$
(4.221)
Equivariance means that $`\lambda `$ induces a map $`\widehat{}\lambda `$ on the quotients. That is: as in eq. 4.146, the map
$$\widehat{}\lambda :TG/G(G\times 𝔤)/G$$
(4.222)
defined as mapping, for any $`g`$, the orbit of any $`vT_gG`$ to the orbit of $`\lambda (v)`$, i.e.
$`\widehat{}\lambda :\mathrm{Orb}(v)\{uTGT_gL_h(v)=u,\mathrm{some}hG\}\mathrm{Orb}(\lambda (v))`$ (4.223)
$`\{(hg,(T_eL_g)^1(v))\mathrm{some}hG\}`$ (4.224)
is well-defined, i.e. independent of the chosen representative $`v`$ of the orbit.
Besides, since the canonical projections, $`vTG\mathrm{Orb}(v)TG/G`$ and $`(g,\xi )\mathrm{Orb}((g,\xi ))(G\times 𝔤)/G`$, are submersions, we can apply result (1) of Section 4.2 and conclude that $`\widehat{}\lambda `$ is smooth.
Finally, we notice that since the action of left translation is transitive, we can identify each orbit of the $`\mathrm{\Phi }`$ of eq. 4.221 with its right component $`\xi 𝔤`$; and so we can identify the set of orbits $`(G\times 𝔤)/G`$ with $`𝔤`$.
To sum up: we have shown that $`TG/G`$ and $`(G\times 𝔤)/G`$, i.e. in effect $`𝔤`$, are diffeomorphic:
$$\widehat{}\lambda :TG/G(G\times 𝔤)/G𝔤.$$
(4.225)
(ii): The results for the cotangent bundle are similar to those in (i). On analogy with eq. 4.220, the action of the cotangent lift of left translation $`T^{}L_g`$ is given in body representation by applying eq. 4.217 to get
$$(\overline{}\lambda (T^{}L_g)\overline{}\lambda ^1)(h,\alpha )=(g^1h,\alpha );$$
(4.226)
or equivalently, now taking the cotangent lift of left translation to define a left action (cf. eq. 4.126),
$$(\overline{}\lambda (T^{}L_{g^1})\overline{}\lambda ^1)(h,\alpha )=(gh,\alpha ).$$
(4.227)
So in body representation, left translation does not act on the covector component; (again, an intuitive result in so far as $`\alpha `$ is “attached to the body”). So eq. 4.227 means that $`\overline{}\lambda `$ is an equivariant map relating the cotangent lifted left action of left translation on $`T^{}G`$ to the $`G`$-action on $`G\times 𝔤^{}`$ given just by left translation on the first component:
$$\mathrm{\Phi }_g((h,\alpha ))g(h,\alpha ):=(gh,\alpha ).$$
(4.228)
So, on analogy with eq. 4.222 and 4.224, we can pass to the quotients, defining a map
$$\widehat{\overline{}\lambda }:T^{}G/G(G\times 𝔤^{})/G$$
(4.229)
by requiring that for $`\alpha T_g^{}G`$, so that $`T^{}L_{h^1}\alpha T_{hg}^{}G`$:
$`\widehat{\overline{}\lambda }:\mathrm{Orb}(\alpha )\{\beta T^{}G\beta =T^{}L_{h^1}(\alpha ),\mathrm{some}hG\}`$ (4.230)
$`\mathrm{Orb}(\overline{}\lambda (\alpha ))\{(hg,(T_e^{}L_g)(\alpha ))\mathrm{some}hG\}\{(h,(T_e^{}L_g)\alpha )\mathrm{some}hG\}.`$ (4.231)
And finally, we identify the set of orbits $`(G\times 𝔤^{})/G`$ with $`𝔤^{}`$, so that we conclude that $`T^{}G/G`$ and $`𝔤^{}`$ are diffeomorphic. That is, we think of the diffeomorphism $`\widehat{\overline{}\lambda }`$ as mapping $`T^{}G/G`$ to $`𝔤^{}`$:
$$\widehat{\overline{}\lambda }:\mathrm{Orb}(\alpha )\{\beta T^{}G\beta =T^{}L_{h^1}(\alpha ),\mathrm{some}hG\}T^{}G/G(T_e^{}L_g)(\alpha )𝔤^{}.$$
(4.232)
As I said above, this diffeomorphism is the crucial first part of Section 7’s main reduction theorem. But we will see its role there, already in (3) of Section 5.1.
Finally, a result which will not be needed later. To calculate the derivatives and cotangent lifts of left translation in space representation, we replace $`\lambda `$ and $`\overline{}\lambda `$ by $`\rho `$ and $`\overline{\rho }`$ as defined by eq. 4.215 and 4.218. We get as the analogues of eq. 4.220 and 4.226 respectively:
$$(\rho TL_g\rho ^1)(h,\xi )=(gh,Ad_g(\xi )),$$
(4.233)
and
$$(\overline{\rho }T^{}L_g\overline{\rho }^1)(h,\alpha )=(g^1h,Ad_g^{}(\alpha )).$$
(4.234)
Though these results are not needed later, they are also analogues of some later results, eq. 6.403 and 6.404, which we will need. (Note that, in accordance with the discussion between eq. 4.191 and 4.192, eq. 4.234 involves right actions.)
## 5 Poisson manifolds
### 5.1 Preamble: three reasons for Poisson manifolds
Now that we are equipped with Sections 3 and 4’s toolbox of modern geometry, we can develop, in this Section and the two to follow, the theory of symplectic reduction. This Section develops the general theory of Poisson manifolds, as a framework for a generalized Hamiltonian mechanics. Its main results concern the foliation, and quotienting, of Poisson manifolds. Then Section 6 returns us to symmetries and conserved quantities: topics which are familiar from Section 2.1.3, but which Section 6 will discuss in the generalized framework using the notion of a momentum map. Finally, in Section 7 all the pieces of our jigsaw puzzle will come together, in our symplectic reduction theorem.
We already glimpsed in (1) of Section 2.2 the idea of a Poisson manifold as a generalization of a symplectic manifold, that provides the appropriate framework for a generalized Hamiltonian mechanics. It is a manifold equipped with a bracket, called a ‘Poisson bracket’, that has essentially the same formal defining properties as in symplectic geometry except that it can be “degenerate”. In particular, the dimension $`m`$ of a Poisson manifold $`M`$ can be even or odd. As we will see, Hamiltonian mechanics can be set up on Poisson manifolds, in a natural generalization of the usual formalism: there are $`m`$ first-order ordinary differential equations for the time evolution of local coordinates $`x^1,\mathrm{},x^m`$, and the time-derivative of any dynamical variable (scalar function on the Poisson manifold $`M`$) is given by its Poisson bracket with the Hamiltonian. Besides, this generalization reduces to the usual formalism in the following sense. Any Poisson manifold $`M`$ is foliated into symplectic manifolds, and any Hamiltonian mechanics of our generalized kind defined on $`M`$ restricts on each symplectic leaf to a conventional Hamiltonian mechanics using the induced symplectic form.
This last point, the invariance of the symplectic leaves under the dynamics, prompts the question ‘why bother with the Poisson manifold, since the dynamics can be written down on each leaf?’. There are three reasons. I will just mention the first; the rest of Section 5 will develop the second; and the two subsequent Sections will develop the third.
(1): Parameters and stability:—
The first two reasons concern the fact that for many problems in Hamiltonian mechanics, it is natural to consider an odd-dimensional state-space. One principal way this happens is if the system is characterized by some odd number, say $`s`$ (maybe $`s=1`$), of parameters that are constant in time. Then even though for a fixed value of the parameter(s), there is a Hamiltonian mechanics on a symplectic manifold, of dimension $`2n`$ say, it is useful to envisage the $`2n+s`$ dimensional space in order to keep track of how the behaviour of systems depends on the parameters. For example, this is very useful for analysing stability, especially if one can somehow control the value of the parameters. Stability theory (and related fields such as bifurcation theory) are crucially important, and vast, topics—which I will not go into.<sup>21</sup><sup>21</sup>21Except to note a broad philosophical point. These parameters illustrate the modal or counterfactual involvements of mechanics. The $`s`$ dimensions of the state-space, and the mathematical constructions built on them, show how rich and structured these involvement are. For a detailed discussion of the modal involvements of mechanics, cf. Butterfield (2004).
(2): Odd-dimensional spaces: the rigid body again:—
Secondly, even in the absence of such controllable parameters, there are mechanical systems whose description leads naturally to an odd-dimensional state-space. The paradigm elementary example is the rigid body pivoted at a point (mentioned in (3) of Section 2.2). An elementary analysis, repeated in every textbook, leads to a description of the body by the three components of the angular momentum (relative to body coordinates, i.e. coordinates fixed in the body): these components evolve according to the three first-order Euler equations.
This situation prompts two foundational questions; (which of course most textbooks ignore!). First, we note that a configuration of the body is given by three real numbers: viz. to specify the rotation required to rotate the body into the given configuration, from a fiducial configuration. So a conventional Hamiltonian description of the rigid body would use six first-order equations. (Indeed, similarly for a Lagrangian description, if we treat the three $`\dot{q}`$s as variables.) So how is the description by Euler’s equations related to a six-dimensional Hamiltonian (or indeed Lagrangian) description?
Second, can the description by the Euler equations be somehow regarded as itself Hamiltonian, or Lagrangian?
This Chapter will not pursue these questions about the rigid body; for details, cf. the references at the end of (3) of Section 2.2. For us, the important point is that the theory of symplectic reduction shows that the answer to the second question is Yes. Indeed, a “resounding Yes”. For we will see very soon (in Section 5.2.4.A) that the three-dimensional space of the components, in body coordinates, of the angular momentum is our prototype example of a Poisson manifold; and the evolution by Euler’s equations is the Hamiltonian mechanics on each symplectic leaf of this manifold. In short: in our generalized framework, Euler’s equations are already in Hamiltonian form.
Furthermore, this Poisson manifold is already familiar: it is $`𝔰𝔬(3)^{}`$, the dual of the Lie algebra of the rotation group. Here we connect with several previous discussions (and this Chapter’s second motto).
First: we connect with the discussion of rotation in Relationist and Reductionist mechanics (Sections 2.3.3 to 2.3.5). In particular, cf. comment (iii) about $`\gamma `$, the three variables encoding the total angular momentum of the system, at the end of Section 2.3.4. (So as regards (1)’s idea of labelling the symplectic leaves by parameters constant in time: in this example, it is the magnitude $`L`$ of the total body angular momentum which is the parameter.)
Second: we connect with Section 3.4.4’s discussion of $`𝔰𝔬(3)`$, with Section 4.5.2’s discussion of the co-adjoint representation on $`𝔰𝔬(3)^{}`$, and with Section 4.6’s discussion of kinematics on an arbitrary Lie group. As regards the rigid body, the main physical idea is that the action of $`SO(3)`$ on itself by left translation is interpreted in terms of the coordinate transformation, i.e. rotation, between the space and body coordinate systems.
But setting aside the rigid body: recall that in Section 4.5.2 we saw that for $`𝔰𝔬(3)^{}`$, the co-adjoint orbits are the spheres centred on the origin. I also announced that they have a natural symplectic structure—and that this was true for the orbits of the co-adjoint representation of any Lie group. Now that we have the notion of a Poisson manifold, we can say a bit more, though of course the proofs are yet to come:—
> For any Lie group $`G`$, the dual of its Lie algebra $`𝔤^{}`$ is a Poisson manifold; and $`G`$ has on $`𝔤^{}`$ a co-adjoint representation, whose orbits are the symplectic leaves of $`𝔤^{}`$ as a Poisson manifold.
In particular, we remark that the theory of the rigid body just sketched is independent of the dimension of physical space being three: it carries over to $`𝔰𝔬(n)^{}`$ for any $`n`$. So we can readily do the Hamiltonian mechanics of the rigid body in arbitrary dimensions. That sounds somewhat academic! But it leads to a more general point, which is obviously of vast practical importance.
In engineering we often need to analyse or design bodies consisting of two or more rigid bodies jointed together, e.g. at a universal joint. Often the configuration space of such a jointed body can be given by a sequence of rotations (in particular about the joints) and-or translations from a fiducial configuration; so that we can take an appropriate Lie group $`G`$ as the body’s configuration space. If so, we can try to mimic our strategy for the rigid body, i.e. to apply the result just announced. And indeed, for such bodies, the action of left translation, and so the adjoint and co-adjoint representations of $`G`$ on $`𝔤`$ and $`𝔤^{}`$, can often be physically significant.
But leaving engineering aside, let us sum up this second reason for Poisson manifolds as follows. For some mechanical systems the natural state-space for a Hamiltonian mechanics is a Poisson manifold. And in the paradigm case of the rigid body, there is a striking interpretation of the Poisson manifold’s leaves as the orbits of the co-adjoint representation of the rotation group $`SO(3)`$.
(3): Reduction:—
My first two reasons have not mentioned reduction. But unsurprisingly, they have several connections with the notion. Here I shall state just one main connection, which links Section 4.6’s kinematics on Lie groups to our main reduction theorem: this will be my third motivation for studying Poisson manifolds.
In short, the connection is that:—
(i): For various systems, the configuration space is naturally taken to be a Lie group $`G`$; (as we have just illustrated with the rigid body).
(ii): So it is natural to set up an orthodox Hamiltonian mechanics of the system on the cotangent bundle $`T^{}G`$. But (as in the Reductionist procedure of Section 2.3.4) it is also natural to quotient by the lift to the cotangent bundle of $`G`$’s action on itself by left translation.
(iii): When we do this, the resulting reduced phase space $`T^{}G/G`$ is a Poisson manifold. Indeed it is an isomorphic copy of $`𝔤^{}`$. That is, we have an isomorphism of Poisson manifolds: $`𝔤^{}T^{}G/G`$. This is the Lie-Poisson reduction theorem.
I shall give a bit more detail about each of (i)-(iii).
(i): For various systems, any configuration can be obtained by acting with an element of the Lie group $`G`$ on some reference configuration which can itself be labelled by an element of $`G`$, say the identity $`eG`$. So we take the Lie group $`G`$ to be the configuration space. As mentioned in (3) of Section 2.2, there is even an infinite-dimensional example of this: the ideal fluid.
(ii): So $`T^{}G`$ is the conventional Hamiltonian phase space of the system. But $`G`$ acts on itself by left translation. We can then consider the quotient of $`T^{}G`$ by the cotangent lift of left translation. Intuitively, this is a matter of “rubbing out” the way that $`T^{}G`$ encodes (i)’s choice of reference configuration. By passing to the quotients as in Section 4.6, we infer that $`T^{}G/G`$ is a manifold. But of course it is in general not even-dimensional. For its dimension is $`\frac{1}{2}\mathrm{dim}(T^{}G)\mathrm{dim}(G)`$. So consider any odd-dimensional $`G`$: for example, our old friend, the three-dimensional rotation group $`SO(3)`$.
(iii): But $`T^{}G/G`$ is always a Poisson manifold. And it is always isomorphic as a Poisson manifold to $`𝔤^{}`$, with its symplectic leaves being the co-adjoint orbits of $`𝔤^{}`$: $`𝔤^{}T^{}G/G`$.
I end this third reason for studying Poisson manifolds with two remarks about examples.
The first remark echoes the end of Section 4.5.2, where I said that by considering all possible Lie groups and all the orbits of their co-adjoint representations, we get a series of examples of symplectic manifolds. We can now put this together with the notion of a Poisson manifold, and with the comment at the end of Section 3.4.3, that every (finite-dimensional) Lie algebra is the Lie algebra of a Lie group. In short: we get a series of examples of Poisson manifolds, in either of two equivalent ways: from the dual $`𝔤^{}`$ of any (finite-dimensional) Lie algebra $`𝔤`$; or from the quotient $`T^{}G/G`$ of the cotangent lift of left translation. In either case, the example is the co-adjoint representation.
The second remark is that there are yet other examples of Poisson manifolds and reductions. Indeed, we noted one in Section 2.3.4: viz. the Reductionist’s reduced phase space $`\overline{M}:=M/E`$, obtained by quotienting the phase space $`M:=T^{}\mathrm{I}\mathrm{R}^{3N}(\delta \mathrm{\Delta })`$ by the (cotangent lift) of the action of the euclidean group $`E`$ on $`\mathrm{I}\mathrm{R}^{3N}`$. But I shall not go into further details about this example; (for which cf. the Belot papers listed in Section 2.3.1, and references therein). Here it suffices to note that this example is not of the above form: $`\mathrm{I}\mathrm{R}^{3N}`$ is not $`E`$, and the action of $`E`$ on $`\mathrm{I}\mathrm{R}^{3N}`$ is not left translation. This of course echoes my remarks at the end of Section 1.2 that the theory of symplectic reduction is too large and intricate for this Chapter to be more than an “appetizer”.
So much by way of motivating Poisson manifolds. The rest of this Section will cover reasons (1) and (2); but reason (3), about reduction, is postponed to Sections 6 and 7. We give some basics about Poisson manifolds, largely in coordinate-dependent language, in Section 5.2. In Section 5.3, we move to a more coordinate-independent language and show that Poisson manifolds are foliated into symplectic manifolds. In Section 5.4, we show that the leaves of the foliation of a finite-dimensional Lie algebra $`𝔤^{}`$ are the orbits of the co-adjoint representation of $`G`$ on $`𝔤^{}`$. Finally in Section 5.5, we prove a general theorem about quotienting a Poisson manifold by the action of Lie group, which will be important for Section 7’s main theorem.
### 5.2 Basics
In Sections 5.2.1 to 5.2.3, we develop some basic definitions and results about Poisson manifolds. This leads up to Section 5.2.4, where we see that the dual of any finite-dimensional Lie algebra has a natural (i.e. basis-independent) Poisson manifold structure. Throughout, there will be some obvious echoes of previous discussions of anti-symmetric forms, Poisson brackets, Hamiltonian vector fields and Lie brackets (Sections 2.1 and 3.2). But I will for the most part not articulate these echoes.
#### 5.2.1 Poisson brackets
A manifold $`M`$ is called a Poisson manifold if it is equipped with a Poisson bracket (also known as: Poisson structure). A Poisson bracket is an assignment to each pair of smooth real-valued functions $`F,H:M\mathrm{I}\mathrm{R}`$, of another such function, denoted by $`\{F,H\}`$, subject to the following four conditions:—
(a) Bilinearity:
$$\{aF+bG,H\}=a\{F,H\}+b\{G,H\};\{F,aG+bH\}=a\{F,G\}+b\{F,H\}a,b\mathrm{I}\mathrm{R}.$$
(5.235)
(b) Anti-symmetry:
$$\{F,H\}=\{H,F\}.$$
(5.236)
(c) Jacobi identity:
$$\{\{F,H\},G\}+\{\{G,F\},H\}+\{\{H,G\},F\}=0.$$
(5.237)
(d) Leibniz’ rule:
$$\{F,HG\}=\{F,H\}G+H\{F,G\}.$$
(5.238)
In other words: $`M`$ is a Poisson manifold iff both: (i) the set $`(M)`$ of smooth scalar functions on $`M`$, equipped with the bracket $`\{,\}`$, is a Lie algebra; and (ii) the bracket $`\{,\}`$ is a derivation in each factor.
Any symplectic manifold is a Poisson manifold. The Poisson bracket is defined by the manifold’s symplectic form; cf. eq. 2.18.
“Canonical” Example:—
Let $`M=\mathrm{I}\mathrm{R}^m,m=2n+l`$, with standard coordinates $`(q,p,z)=(q^1,\mathrm{},q^n,p^1,\mathrm{},p^n,z^1,\mathrm{},z^l)`$. Define the Poisson bracket of any two functions $`F(q,p,z)`$, $`H(q,p,z)`$ by
$$\{F,H\}:=\mathrm{\Sigma }_i^n\left(\frac{F}{q^i}\frac{H}{p^i}\frac{F}{p^i}\frac{H}{q^i}\right).$$
(5.239)
Thus this bracket ignores the $`z`$ coordinates; and if $`l`$ were equal to zero, it would be the standard Poisson bracket for $`\mathrm{I}\mathrm{R}^{2n}`$ as a symplectic manifold. We can immediately deduce the Poisson brackets for the coordinate functions. Those for the $`q`$s and $`p`$s are as for the usual symplectic case:
$$\{q^i,q^j\}=0\{p^i,p^j\}=0\{q^i,p^j\}=\delta _{ij}.$$
(5.240)
On the other hand, all those involving the $`z`$s vanish:
$$\{q^i,z^j\}=\{p^i,z^j\}=\{z^i,z^j\}0.$$
(5.241)
Besides, any function $`F`$ depending only on the $`z`$’s, $`FF(z)`$ will have vanishing Poisson brackets with all functions $`H:\{F,H\}=0.`$
This example seems special in that $`M`$ is foliated into $`2n`$-dimensional symplectic manifolds, each labelled by $`l`$ constant values of the $`z`$s. But Section 5.3.4 will give a generalization for Poisson manifolds of Darboux’s theorem (mentioned at the end of Section 2.1.1): a generalization saying, roughly speaking, that every Poisson manifold “looks locally like this”.
For any Poisson manifold, we say that a function $`F:M\mathrm{I}\mathrm{R}`$ is distinguished or Casimir if its Poisson bracket with all smooth functions $`H:M\mathrm{I}\mathrm{R}`$ vanishes identically: $`\{F,H\}=0.`$
#### 5.2.2 Hamiltonian vector fields
Given a smooth function $`H:M\mathrm{I}\mathrm{R}`$, consider the map on smooth functions: $`F\{F,H\}`$. The fact that the Poisson bracket is bilinear and obeys Leibniz’s rule implies that this map $`F\{F,H\}`$ is a derivation on the space of smooth functions, and so determines a vector field on $`M`$; (cf. (ii) of Section 3.1.2.B). We call this vector field the Hamiltonian vector field associated with (also known as: generated by) $`H`$, and denote it by $`X_H`$.
But independently of the Poisson structure, the action of any vector field $`X_H`$ on a smooth function $`F`$, $`X_H(F)`$, also equals $`L_{X_H}(F)dF(X_H)`$; (cf. eq. 3.40). So we have for all smooth $`F`$
$$L_{X_H}(F)dF(X_H)X_H(F)=\{F,H\}.$$
(5.242)
The equations describing the flow of $`X_H`$ are called Hamilton’s equations, for the choice of $`H`$ as “Hamiltonian”.
In the previous example with $`M=\mathrm{I}\mathrm{R}^{2n+l}`$, we have
$$X_H=\mathrm{\Sigma }_i^n\left(\frac{H}{p^i}\frac{}{q^i}\frac{H}{q^i}\frac{}{p^i}\right),$$
(5.243)
and the flow is given by the ordinary differential equations
$$\frac{dq^i}{dt}=\frac{H}{p^i}\frac{dp^i}{dt}=\frac{H}{q^i}\frac{dz^j}{dt}=0.i=1,\mathrm{},n;j=1,\mathrm{},l.$$
(5.244)
Again, the $`z`$s, and any function $`F(z)`$ solely of them, are distinguished and have a vanishing Hamiltonian vector field. On the other hand, the coordinate functions $`q^i`$ and $`p^i`$ generate the Hamiltonian vector fields $`\frac{}{p^i}`$ and $`\frac{}{q^i}`$ respectively.
Two further remarks about eq. 5.242:—
(1): It follows that a function $`H`$ is distinguished (i.e. has vanishing Poisson brackets with all functions) iff its Hamiltonian vector field $`X_H`$ vanishes everywhere. And since the Poisson bracket is antisymmetric, this is so iff $`H`$ is constant along the flow of all Hamiltonian vector fields.
(2): This equation is the beginning of the theory of constants of the motion (first integrals), and of Noether’s theorem, for Poisson manifolds; just as the corresponding equation was the beginning for the symplectic case. This will be developed in Section 6.
Poisson brackets and Lie brackets:—
With the definition eq. 5.242 in hand, we can readily establish our first important connection between Poisson manifolds and Section 3’s Lie structures. Namely: result (2) at the end of Section 3.2.2, eq. 3.60, is also valid for Poisson manifolds.
That is: the Hamiltonian vector field of the Poisson bracket of scalars $`F,H`$ on a Poisson manifold $`M`$ is, upto a sign, the Lie bracket of the Hamiltonian vector fields, $`X_F`$ and $`X_H`$, of $`F`$ and $`H`$:
$$X_{\{F,H\}}=[X_F,X_H]=[X_H,X_F].$$
(5.245)
The proof is exactly as for eq. 3.60.
So the Hamiltonian vector fields, with the Poisson bracket, form a Lie subalgebra of the Lie algebra $`𝒳_M`$ of all vector fields on the Poisson manifold $`M`$. This result will be important in Section 5.3.3’s proof that every Poisson manifold is a disjoint union of symplectic manifolds.
#### 5.2.3 Structure functions
We show that to compute the Poisson bracket of any two functions given in some local coordinates $`𝐱=x^1,\mathrm{},x^m`$, it suffices to know the Poisson brackets of the coordinates. For any function $`H:M\mathrm{I}\mathrm{R}`$, let the components of its Hamiltonian vector field in the coordinate system $`𝐱`$ be written as $`h^i(x)`$. So $`X_H=\mathrm{\Sigma }_i^mh^i(x)\frac{}{x^i}`$. Then for any other function $`F`$, we have
$$\{F,H\}=X_H(F)=\mathrm{\Sigma }_i^mh^i(x)\frac{F}{x^i}.$$
(5.246)
Taking $`x^i`$ as the function $`F`$, we get: $`\{x^i,H\}=X_H(x^i)=h^i(x)`$. So eq. 5.246 becomes
$$\{F,H\}=\mathrm{\Sigma }_i^m\{x^i,H\}\frac{F}{x^i}.$$
(5.247)
If we now put $`x^i`$ for $`H`$ and $`H`$ for $`F`$ in eq. 5.247, we get
$$\{x^i,H\}=\{H,x^i\}=X_{x^i}(H)=\mathrm{\Sigma }_j^m\{x^j,x^i\}\frac{H}{x^j}..$$
(5.248)
Combining eq.s 5.247 and 5.248, we get the basic formula for the Poisson bracket of any two functions in terms of the Poisson bracket of local coordinates:
$$\{F,H\}=\mathrm{\Sigma }_i^m\mathrm{\Sigma }_j^m\{x^i,x^j\}\frac{F}{x^i}\frac{H}{x^j}.$$
(5.249)
We assemble these basic brackets, which we call the structure functions of the Poisson manifold,
$$J^{ij}(x):=\{x^i,x^j\}i,j=1,\mathrm{},m$$
(5.250)
into a $`m\times m`$ anti-symmetric matrix of functions, $`J(x)`$, called the structure matrix of $`M`$. More precisely, it is the structure matrix for $`M`$ relative to our coordinate system $`𝐱`$. Of course, the transformation of $`J`$ under a coordinate change $`x^i:=x^i(x^1,\mathrm{},x^m)`$ is determined by setting $`F:=x^i,H:=x^j`$ in the basic formula eq. 5.249.
Then, writing $`H`$ for the (column) gradient vector of $`H`$, eq. 5.249 becomes
$$\{F,H\}=FJH.$$
(5.251)
For example, the canonical bracket on $`\mathrm{I}\mathrm{R}^{2n+l}`$, eq.5.239, written in the $`(q,p,z)`$ coordinates, has the simple form
$$J=\left(\begin{array}{ccc}0& I& 0\\ I& 0& 0\\ 0& 0& 0\end{array}\right).$$
(5.252)
where $`I`$ is the $`n\times n`$ identity matrix.
We can write the Hamiltonian vector field, and the Hamilton’s equations, associated with the function $`H`$ in terms of $`J`$. Since
$$\{x^i,H\}=\mathrm{\Sigma }_j^m\{x^i,x^j\}\frac{H}{x^j}$$
(5.253)
we get:
$$X_H=\mathrm{\Sigma }_i^m\left(\mathrm{\Sigma }_j^mJ^{ij}(x)\frac{H}{x^j}\frac{}{x^i}\right),$$
(5.254)
or in matrix notation: $`X_H=(JH)_x`$. Similarly, Hamilton’s equations
$$\frac{dx^i}{dt}=\{x^i,H\}$$
(5.255)
get the matrix form
$$\frac{dx}{dt}=J(x)H(x);\mathrm{i}.\mathrm{e}.\frac{dx^i}{dt}=\mathrm{\Sigma }_j^mJ^{ij}(x)\frac{H}{x^j}.$$
(5.256)
To summarize how we have generalized from the usual form of Hamilton’s equations: compare eq. 5.256, 5.251 and 5.252 respectively with eq. 2.12, 2.18 and 2.3.
Note that not every $`m\times m`$ anti-symmetric matrix of functions on an $`m`$-dimensional manifold (or even: on an open subset of $`\mathrm{I}\mathrm{R}^m`$) is the structure matrix of a Poisson manifold: for the Jacobi identity constrains the functions. In fact it is readily shown that the Jacobi identity corresponds to the following $`m^3`$ partial differential equations governing the $`J^{ij}(x)`$, which are in general non-linear. Writing as usual $`_l`$ for $`/x^l`$:
$$\mathrm{\Sigma }_{l=1}^m(J^{il}_lJ^{jk}+J^{kl}_lJ^{ij}+J^{jl}_lJ^{ki})=0i,j,k,=1,\mathrm{},m;xM.$$
(5.257)
In particular, any constant anti-symmetric matrix $`J`$ defines a Poisson structure.
#### 5.2.4 The Poisson structure on $`𝔤^{}`$
We can now show that any $`m`$-dimensional Lie algebra $`𝔤`$ defines a Poisson structure, often called the Lie-Poisson bracket, on any $`m`$-dimensional vector space $`V`$. We proceed in two stages.
(1): We first present the definition in a way that seems to depend on a choice of bases, both in $`𝔤`$ (where the definition makes a choice of structure constants) and in the space $`V`$.
(2): Then we will see that choosing $`V`$ to be $`𝔤^{}`$, the definition is in fact basis-independent.
This Poisson structure on $`𝔤^{}`$ will be of central importance from now on. As Marsden and Ratiu write: ‘Besides the Poisson structure on a symplectic manifold, the Lie-Poisson bracket on $`𝔤^{}`$, the dual of a Lie algebra, is perhaps the most fundamental example of a Poisson structure’ (1999: 415). Here we return to our motivating discussion of Poisson manifolds, especially reasons (2) and (3) of Section 5.1: which concerned the rigid body and reduction, respectively. Indeed, we will see already in the Example at the end of this Subsection (Section 5.2.4.A) how the Lie-Poisson bracket on the special case $`𝔤^{}:=𝔰𝔬(3)^{}`$ clarifies the theory of the rigid body. And we will see in Sections 7.2 and 7.3.3 how for any $`𝔤`$, the Lie-Poisson bracket on $`𝔤^{}`$ is induced by reduction, from the canonical Poisson (viz. symplectic) structure on the cotangent bundle $`T^{}G`$. This will be our reduction theorem, that $`T^{}G/G𝔤^{}`$.
After (2), we will see that the Lie-Poisson bracket on $`𝔤^{}`$ implies that Hamilton’s equations on $`𝔤^{}`$ can be expressed using $`ad^{}`$: a form that will be needed later. This will be (3) below. Then we will turn in Section 5.2.4.A to the example $`𝔤^{}:=𝔰𝔬(3)^{}`$.
(1): A Poisson bracket on any vector space $`V`$:—
Take a basis, say $`e_1,\mathrm{},e_m`$, in $`𝔤`$, and so structure constants $`c_{ij}^k`$ (cf. eq. 3.52). Consider the space $`V`$ as a manifold, and coordinatize it by taking a basis, $`ϵ_1,\mathrm{},ϵ_m`$ say, determining coordinates $`x^1,\mathrm{},x^m`$. We now define the Poisson bracket (in this case, often called the Lie-Poisson bracket) between two smooth functions $`F,H:V\mathrm{I}\mathrm{R}`$ to be
$$\{F,H\}:=\mathrm{\Sigma }_{i,j,k=1}^mc_{ij}^kx^k\frac{F}{x^i}\frac{H}{x^j}.$$
(5.258)
This takes the form of eq. 5.249, with linear structure functions $`J^{ij}(x)=\mathrm{\Sigma }_k^mc_{ij}^kx^k`$. One easily checks that anti-symmetry, and the Jacobi identity, for the structure constants, eq. 3.53, implies that these $`J^{ij}`$ are anti-symmetric and obey their Jacobi identity eq. 5.257. So eq. 5.258 defines a Poisson bracket on $`V`$.
In particular, the associated Hamiltonian equations, eq.s 5.255 and 5.256, take the form
$$\frac{dx^i}{dt}=\mathrm{\Sigma }_{j,k=1}^mc_{ij}^kx^k\frac{H}{x^j}.$$
(5.259)
(2): The Lie-Poisson bracket on $`𝔤^{}`$:—
To give a basis-independent characterization of the Lie-Poisson bracket, we first recall that:
(i): the gradient $`F(x)`$ of $`F:V\mathrm{I}\mathrm{R}`$ at any point $`xV`$ is in the dual space $`V^{}`$ of (continuous) linear functionals on $`V`$;:
(ii): any finite-dimensional vector space is canonically, i.e. basis-independently, isomorphic to its double dual: $`(V^{})^{}V`$.
Then writing $`<;>`$ for the natural pairing between $`V`$ and $`V^{}`$, we have, for any $`yV`$
$$<F(x);y>:=\mathrm{lim}_{\tau 0}\frac{F(x+\tau y)F(x)}{\tau }.$$
(5.260)
Now let us take $`V`$ in our definition of the Lie-Poisson bracket to be $`𝔤^{}`$. So we will show that $`𝔤`$ makes $`𝔤^{}`$ a Poisson manifold, in a basis-independent way. And let the basis $`ϵ_1,\mathrm{},ϵ_m`$ be dual to the basis $`e_1,\mathrm{},e_m`$ of $`𝔤`$. If $`F:𝔤^{}\mathrm{I}\mathrm{R}`$ is any smooth function, its gradient $`F(x)`$ at any point $`x𝔤^{}`$ is an element of $`(𝔤^{})^{}𝔤`$. One now checks that the Lie-Poisson bracket defined by eq. 5.258 has the basis-independent expression
$$\{F,H\}(x)=<x;[F(x),H(x)]>,x𝔤^{}$$
(5.261)
where $`[,]`$ is the ordinary Lie bracket on the Lie algebra $`𝔤`$ itself.
(3): Hamilton’s equations on $`𝔤^{}`$:—
We can also give a basis-independent expression of the Hamilton’s equations eq. 5.259: viz. by expressing the Lie bracket in eq. 5.261 in terms of $`ad`$, as indicated by eq. 4.179.
Thus let $`F(𝔤^{})`$ be an arbitrary smooth scalar function on $`𝔤^{}`$. By the chain rule
$$\frac{dF}{dt}=𝐃F(x)\dot{x}=<\dot{x};F(x)>.$$
(5.262)
But applying eq.s 4.179 and 4.198 to eq. 5.261 implies:
$$\{F,H\}(x)=<x;[F(x),H(x)]>=<x;ad_{H(x)}(F(x))>=<ad_{H(x)}^{}(x);F(x)>.$$
(5.263)
Since $`F`$ is arbitrary and the pairing is non-degenerate, we deduce that Hamilton’s equations take the form
$$\frac{dx}{dt}=ad_{H(x)}^{}(x).$$
(5.264)
##### 7.2.4.A Example: $`𝔰𝔬(3)`$ and $`𝔰𝔬(3)^{}`$
As an example of the dual of a Lie algebra as a Poisson manifold, let us consider again our standard example $`𝔰𝔬(3)^{}`$. We will thereby make good our promise in (2) of Section 5.1, to show that Euler’s equations for a rigid body are already in Hamiltonian form—in our generalized sense. We will also see why in the Chapter’s second motto, Arnold mentions the three dual spaces, $`\mathrm{I}\mathrm{R}^3,𝔰𝔬(3)^{}`$ and $`T^{}(SO(3))_g`$; (cf. the discussion at the end of Section 3.4.4).
The Lie algebra $`𝔰𝔬(3)`$ of $`SO(3)`$ has a basis $`e_1,e_2,e_3`$ representing infinitesimal rotations around the $`x`$-, $`y`$\- and $`z`$-axes of $`\mathrm{I}\mathrm{R}^3`$. As we have seen, we can think of these basis elements: as vectors in $`\mathrm{I}\mathrm{R}^3`$ with $`[,]`$ as elementary vector multiplication; or as anti-symmetric matrices with $`[,]`$ as the matrix commutator; or as left-invariant vector fields on $`SO(3)`$ with $`[,]`$ as the vector field commutator (i.e. Lie bracket).
Let $`ϵ_1,ϵ_2,ϵ_3`$ be a dual basis for $`𝔰𝔬(3)^{}`$, with $`x=x^1ϵ_1+x^2ϵ_2+x^3ϵ_3`$ a typical point therein. If $`F:𝔰𝔬(3)^{}\mathrm{I}\mathrm{R}`$, its gradient at $`x`$ is the vector
$$F=\frac{F}{x^1}e_1+\frac{F}{x^2}e_2+\frac{F}{x^3}e_3𝔰𝔬(3).$$
(5.265)
Then eq. 5.261 tells us that, if we write $`𝔰𝔬(3)`$ as $`\mathrm{I}\mathrm{R}^3`$ with $`\times `$ for elementary vector multiplication, the Lie-Poisson bracket on $`𝔰𝔬(3)^{}`$ is
$`\{F,H\}(x)=x^1\left({\displaystyle \frac{F}{x^3}}{\displaystyle \frac{H}{x^2}}{\displaystyle \frac{F}{x^2}}{\displaystyle \frac{H}{x^3}}\right)+\mathrm{}+x^3\left({\displaystyle \frac{F}{x^2}}{\displaystyle \frac{H}{x^1}}{\displaystyle \frac{F}{x^1}}{\displaystyle \frac{H}{x^2}}\right)`$ (5.266)
$`=x(F\times H).`$ (5.267)
So the structure matrix $`J(x)`$ is
$$J(x)=\left(\begin{array}{ccc}0& x_3& x_2\\ x_3& 0& x_1\\ x_2& x_1& 0\end{array}\right),x𝔰𝔬(3)^{}.$$
(5.268)
Hamilton’s equations corresponding to the Hamiltonian function $`H(x)`$ are therefore
$$\frac{dx}{dt}=x\times H(x).$$
(5.269)
Now consider the Hamiltonian representing the kinetic energy of a free pivoted rigid body
$$H(x)=\frac{1}{2}\left(\frac{(x^1)^2}{I_1}+\frac{(x^2)^2}{I_2}+\frac{(x^3)^2}{I_2}\right),$$
(5.270)
in which the $`I_i`$ are the moments of inertia about the three coordinate axes, and the $`x^i`$ are the corresponding components of the body angular momentum. For this Hamiltonian, Hamilton’s equations eq. 5.269 become
$$\frac{dx^1}{dt}=\frac{I_2I_3}{I_2I_3}x^2x^3,\frac{dx^2}{dt}=\frac{I_3I_1}{I_3I_1}x^3x^1,\frac{dx^3}{dt}=\frac{I_1I_2}{I_1I_2}x^1x^2,$$
(5.271)
Indeed, these are the Euler equations for a free pivoted rigid body. I shall not go into details about the rigid body. I only note that:
(i): In the elementary theory of such a body, the magnitude $`L`$ of the angular momentum is conserved, and eq. 5.271 describes the motion of the $`x^i`$ on a sphere of radius $`L`$ centred at the origin.
(ii): In Section 5.4, we will return to seeing these spheres as the orbits of the co-adjoint representation of $`SO(3)`$ on $`𝔰𝔬(3)^{}`$ (cf. Section 4.5.2).
(iii): Let us sum up this theme by saying, with Marsden and Ratiu (1999, p.11) that here we see: ‘a simple and beautiful Hamiltonian structure for the rigid body equations’.
### 5.3 The symplectic foliation of Poisson manifolds
We first reformulate some ideas of Section 5.2 in more coordinate-independent language, starting with Section 5.2.3’s idea of the structure matrix $`J(x)`$ (Section 5.3.1). Then we discuss canonical transformations on a Poisson manifold (Section 5.3.2). This will lead up to showing that any Poisson manifold is foliated by symplectic leaves (Section 5.3.3). Finally, we state a generalization of Darboux’s theorem; and again take $`𝔰𝔬(3)`$ as an example (Section 5.3.4).
#### 5.3.1 The Poisson structure and its rank
We now pass from the structure matrix $`J`$, eq. 5.250, to a coordinate-independent object, the Poisson structure (also known as: co-symplectic structure), written $`𝖡`$. Whereas $`J`$ multiplied naive gradient vectors, as in eq. 5.251 and 5.256, $`𝖡`$ is to map the 1-form $`dH`$ into its Hamiltonian vector field; as follows.
At each point $`x`$ in a Poisson manifold $`M`$, there is a unique linear map $`𝖡_x`$, which we will also write as $`𝖡`$
$$𝖡𝖡_x:T_x^{}MT_xM$$
(5.272)
such that
$$𝖡_x(dH(x))=X_H(x).$$
(5.273)
For the requirement eq. 5.273 implies, by eq. 5.254, that for each $`j=1,\mathrm{},m`$
$$𝖡_x(dx^j)=\mathrm{\Sigma }_iJ^{ij}(x)\frac{}{x^i}_x$$
(5.274)
Since the differentials $`dx^i`$ span $`T_x^{}M`$, this fixes $`𝖡_x`$, by linearity. $`𝖡_x`$’s action on any one-form $`\alpha =\mathrm{\Sigma }a_jdx^j`$ is:
$$𝖡_x(\alpha )=\mathrm{\Sigma }_{i,j}J^{ij}(x)a_j\frac{}{x^i}_x$$
(5.275)
so that $`𝖡_x`$ is essentially matrix multiplication by $`J(x)`$. Here, compare again eq. 5.255 and 5.256.
Here we recall that any linear map between (real finite-dimensional) vector spaces, $`B:VW^{}`$, has an associated bilinear form $`B^{\mathrm{}}`$ on $`V\times W^{}V\times W`$ given by
$$B^{\mathrm{}}(v,w):=<B(v);w>.$$
(5.276)
Accordingly, some authors introduce the Poisson structure as a bilinear form $`𝖡_x^{\mathrm{}}:T_x^{}M\times T_x^{}M\mathrm{I}\mathrm{R}`$, often called the Poisson tensor. Thus eq. 5.276 gives, for $`\alpha ,\beta T_x^{}M`$
$$𝖡_x^{\mathrm{}}(\alpha ,\beta ):=<𝖡(\alpha ),\beta >.$$
(5.277)
$`𝖡_x^{\mathrm{}}`$ is antisymmetric, since the matrix $`J(x)`$ is. So, if we now let $`x`$ vary over $`M`$, we can sum up in the traditional terminology of tensor analysis: $`𝖡^{\mathrm{}}`$ is an antisymmetric contravariant two-tensor field.
Example:— Consider our first example, $`M=\mathrm{I}\mathrm{R}^{2n+l}`$ with the “usual bracket” eq. 5.239, from the start of Section 5.2.1. For any one-form
$$\alpha =\mathrm{\Sigma }_{i=1}^n(a_idq^i+b_idp^i)+\mathrm{\Sigma }_{j=1}^lc_jdz^j$$
(5.278)
we have
$$𝖡(\alpha )=\mathrm{\Sigma }_{i=1}^n\left(b_i\frac{}{q^i}a_i\frac{}{p^i}\right).$$
(5.279)
In this example the form of $`𝖡`$ is the same from point to point. In particular, the kernel of $`𝖡`$ has everywhere the same dimension, viz. $`l`$, the number of distinguished coordinates.
We now define the rank at $`x`$ of a Poisson manifold $`M`$ to be the rank of its Poisson structure $`𝖡`$ at $`x`$, i.e. the dimension of the range of $`𝖡_x`$. This range is also the span of all the Hamiltonian vector fields on $`M`$ at $`x`$:
$$\mathrm{ran}(𝖡_x):=\{XT_xM:X=𝖡_x(\alpha ),\mathrm{some}\alpha T_x^{}M\}=\{X_H(x):H:M\mathrm{I}\mathrm{R}\mathrm{smooth}\}.$$
(5.280)
So the rank of $`M`$ at $`x`$ is also equal to the dimension of $`𝖡_x`$’s domain, i.e. dim($`T_x^{}M`$)=dim($`M`$), minus the dimension of the kernel, $`\mathrm{dim}(𝖡_x)`$.
Since in local coordinates, $`𝖡_x`$ is given by multiplication by the structure matrix $`J(x)`$, the rank of $`M`$ at $`x`$ is the rank (the same in any coordinates) of the matrix $`J(x)`$. That $`J(x)`$ is anti-symmetric implies that the rank of $`M`$ is even: cf. again the normal form of antisymmetric bilinear forms, eq. 2.2 and 2.3.
The manifold $`M`$ being symplectic corresponds, of course, to the rank of $`𝖡`$ being everywhere maximal, i.e. equal to dim($`M`$).
In this case, the kernel of $`𝖡`$ is trivial, and any distinguished function $`H`$ is constant on $`M`$. For $`H`$ is distinguished iff $`X_H=0`$; and if the rank is maximal, then $`dH=0`$, so that $`H`$ is constant.
Besides, each of the Poisson structure and symplectic form on $`M`$ determine the other. In particular, the Poisson tensor $`𝖡^{\mathrm{}}`$ of eq. 5.277 is, up to a sign, the “contravariant cousin” of $`M`$’s symplectic form $`\omega `$. For recall: (i) the relation between a symplectic manifold’s Poisson bracket and its form, eq. 2.18, viz.
$$\{F,H\}=dF(X_H)=\omega (X_F,X_H);$$
(5.281)
and (ii) eq. 5.242 for Hamiltonian vector fields on a Poisson manifold, viz.
$$X_H(F)=\{F,H\}.$$
(5.282)
Applying these equations yields, if we start from eq. 5.277 and eq. 5.273:
$$𝖡^{\mathrm{}}(dH,dF):=<B(dH),dF>=dF(X_H)=X_H(F)=\{F,H\}=\omega (X_F,X_H).$$
(5.283)
We have also seen examples where the Poisson structure $`𝖡`$ is of non-maximal rank:
(i): In our opening “canonical” example, the Poisson bracket eq. 5.239 on $`M=\mathrm{I}\mathrm{R}^{2n+l}`$ has rank $`2n`$ everywhere.
(ii): In the Lie-Poisson structure on $`𝔰𝔬(3)^{}`$, the rank varies across the manifold: it is 2 everywhere, except at the origin $`x=0`$ where it is 0. (Cf. the rank of the matrix $`J`$ in eq. 5.268.)
#### 5.3.2 Poisson maps
Already at the beginning of our development of Poisson manifolds, we saw that a scalar function $`H:M\mathrm{I}\mathrm{R}`$ defines equations of motion, with $`H`$ as “Hamiltonian”, for all other functions $`F:M\mathrm{I}\mathrm{R}`$, of the familiar Poisson bracket type:
$$\dot{F}=\{F,H\}.$$
(5.284)
(Cf. Section 5.2.2, especially the remarks around eq. 5.242.) We now develop the generalization for Poisson manifolds of some related notions and results.
We say that a smooth map $`f:M_1M_2`$ between Poisson manifolds $`(M_1,\{,\}_1)`$ and $`(M_2,\{,\}_2)`$ is Poisson or canonical iff it preserves the Poisson bracket. To be precise: we first need the idea of the pullback of a function; cf. Section 3.1.2.A. In this context, the pullback $`f^{}`$ of a function $`F:M_2\mathrm{I}\mathrm{R}`$ is given by
$$f^{}F:=Ff;\mathrm{i}.\mathrm{e}.f^{}F:xM_1F(f(x))\mathrm{I}\mathrm{R}.$$
(5.285)
Then we say that $`f:M_1M_2`$ is Poisson iff for all smooth functions $`F,G:M_2\mathrm{I}\mathrm{R}`$ ($`F,G(M_2)`$)
$$f^{}\{F,G\}_2=\{f^{}F,f^{}G\}_1;$$
(5.286)
where by the definition eq. 5.285, the lhs $`\{F,G\}_2f`$, and the rhs $`\{Ff,Gg\}_1`$.
We note the special case where $`M_1=M_2=:M`$ and $`M`$ is symplectic; i.e. the Poisson bracket is of maximal rank, and so defines a symplectic form on $`M`$, as in eq. 5.283. In this case, we return to the equivalence in Section LABEL:Hammechs’s usual formulation of Hamiltonian mechanics, between preserving the Poisson bracket and preserving the symplectic form. That is: a map $`f:MM`$ on a symplectic manifold $`M`$ is Poisson iff it is symplectic.
Besides, we already have for symplectic manifolds an infinitesimal version of the idea of a Poisson or symplectic map: viz. the idea of a locally Hamiltonian vector field; cf. Section 2.1.3. Similarly for Poisson manifolds, we will need the corresponding infinitesimal version of a Poisson map; but not till Section 6.1.1.
One can show (using in particular the Jacobi identity) that the flows of a Hamiltonian vector field are Poisson. (Here of course, $`(M_1,\{,\}_1)=(M_2,\{,\}_2)`$.) That is: if $`\varphi _\tau `$ is the flow of $`X_H`$ (i.e. $`\varphi _\tau =\mathrm{exp}(\tau X_H)`$), then
$$\varphi _\tau ^{}\{F,G\}=\{\varphi _\tau ^{}F,\varphi _\tau ^{}G\}\mathrm{i}.\mathrm{e}.\{F,G\}\varphi _\tau =\{F\varphi _\tau ,G\varphi _\tau \}.$$
(5.287)
Similarly, one can readily show the equivalent proposition, that along the flow of a Hamiltonian vector field the Lie derivative of the Poisson tensor $`𝖡^{\mathrm{}}`$ vanishes. That is: for any smooth function $`H:M\mathrm{I}\mathrm{R}`$, we have:
$$_{X_H}𝖡^{\mathrm{}}=0.$$
(5.288)
Since preserving the Poisson bracket implies in particular preserving its rank, it follows from eq. 5.287 (or from eq. 5.288) that:
If $`X_H`$ is a Hamiltonian vector field on a Poisson manifold $`M`$, then for any $`\tau \mathrm{I}\mathrm{R}`$ and $`xM`$, the rank of $`M`$ at $`\mathrm{exp}(\tau X_H)(x)`$ is the same as the rank at $`x`$. In other words: Hamiltonian vector fields are rank-invariant in the sense used in the general form of Frobenius’ theorem (Section 3.3.2).
This result will be important for the foliation theorem for Poisson manifolds.
We will also need the result (also readily shown) that Poisson maps push Hamiltonian flows forward to Hamiltonian flows. More precisely: let $`f:M_1M_2`$ be a Poisson map; so that at each $`xM_1`$, we have the derivative map on the tangent space, $`Tf:(TM_1)_x(TM_2)_{f(x)}`$. And let $`H:M_2\mathrm{I}\mathrm{R}`$ be a smooth function. If $`\varphi _\tau `$ is the flow of $`X_H`$ and $`\psi _\tau `$ is the flow (on $`M_1`$) of $`X_{Hf}`$, then:
$$\varphi _\tau f=f\psi _\tau \mathrm{and}TfX_{Hf}=X_Hf.$$
(5.289)
In particular, this square commutes:
$$\begin{array}{c}M_1\\ \psi _\tau \\ M_1\end{array}\begin{array}{c}\stackrel{f}{}\\ \\ \stackrel{f}{}\end{array}\begin{array}{c}M_2\\ \varphi _\tau \\ M_2\end{array}$$
(5.290)
#### 5.3.3 Poisson submanifolds: the foliation theorem
To state the foliation theorem for Poisson manifolds, we need the idea of a Poisson immersion, which leads to the closely related idea of a Poisson submanifold. In effect, these ideas combine the idea of a Poisson map with the ideas about injective immersions in (2) of Section 3.3.1. We recall from that discussion that for an injective immersion, $`f:NM`$, the range $`f(N)`$ is not necessarily a submanifold of $`M`$: but $`f(N)`$ is nevertheless called an ‘injectively immersed submanifold’ of $`M`$. (But as mentioned in Section 3.3.2, many treatments ignore this point: they in effect assume that an injective immersion $`f`$ is also an embedding, i.e. a homeomorphism between $`N`$ and $`f(N)`$, so that $`f(N)`$ is indeed a submanifold of $`M`$ and $`f`$ is a diffeomorphism.)
An injective immersion $`f:NM`$, with $`M`$ a Poisson manifold, is called a Poisson immersion if any Hamiltonian vector field defined on an open subset of $`M`$ containing $`f(N)`$ is in the range of the derivative map of $`f`$ at $`yN`$, i.e. ran($`T_yf`$), at all points $`f(y)`$ for $`yN`$.
Being a Poisson immersion is equivalent to the following rather technical condition.
> Characterization of Poisson immersions An injective immersion $`f:NM`$, with $`M`$ a Poisson manifold, is a Poisson immersion iff:
> if $`F,G:VN\mathrm{I}\mathrm{R}`$, where $`V`$ is open in $`N`$, and if $`\overline{F},\overline{G}:U\mathrm{I}\mathrm{R}`$ are extensions of $`Ff^1,Gf^1:f(V)\mathrm{I}\mathrm{R}`$ to an open neighbourhood $`U`$ of $`f(V)`$ in $`M`$, then $`\{\overline{F},\overline{G}\}_{f(V)}`$ is well-defined and independent of the extensions.
The main point of this equivalence is that it ensures that if $`f:NM`$ is a Poisson immersion, then $`N`$ has a Poisson structure, and $`f:NM`$ is a Poisson map. It is worth seeing how this comes about—by proving the equivalence.
Proof: Let $`f:NM`$ be a Poisson immersion, and let $`F,G:VN\mathrm{I}\mathrm{R}`$ and let $`\overline{F},\overline{G}:Uf(V)\mathrm{I}\mathrm{R}`$ be extensions of $`Ff^1,Gf^1:f(V)\mathrm{I}\mathrm{R}`$. Then for $`yV`$, there is a unique vector $`vTN_y`$ such that
$$X_{\overline{G}}(f(y))=(T_yf)(v).$$
(5.291)
So evaluating the Poisson bracket of $`\overline{F}`$ and $`\overline{G}`$ at $`f(y)`$ yields, by eq. 5.242,
$$\{\overline{F},\overline{G}\}(f(y))=d\overline{F}(f(y))X_{\overline{G}}(f(y))=d\overline{F}(f(y))(T_yf)(v)=d(\overline{F}f)(y)vdF(y)v.$$
(5.292)
So $`\{\overline{F},\overline{G}\}(f(y))`$ is independent of the extension $`\overline{F}`$ of $`Ff^1`$. Since the Poisson bracket is antisymmetric, it is also independent of the extension $`\overline{G}`$ of $`Gf^1`$. So we can define a Poisson structure on $`N`$ by defining for any $`y`$ in an open $`VN`$
$$\{F,G\}_N(y):=\{\overline{F},\overline{G}\}_M(f(y)).$$
(5.293)
This makes $`f:NM`$ a Poisson map, since for any $`\overline{F},\overline{G}`$ on $`M`$ and any $`yN`$, we have that
$$[f^{}\{\overline{F},\overline{G}\}_M](y)[\{\overline{F},\overline{G}\}_Mf](y)=\{F,G\}_N(y)\{f^{}\overline{F},f^{}\overline{G}\}_N(y);$$
(5.294)
where the middle equality uses eq. 5.293.
For the converse implication, assume that eq. 5.292 holds, and let $`H:U\mathrm{I}\mathrm{R}`$ be a Hamiltonian defined on an open subset $`U`$ of $`M`$ that intersects $`f(N)`$. Then as we have just seen, $`N`$ is a Poisson manifold and $`f:NM`$ is a Poisson map. Because $`f`$ is Poisson, it pushes $`X_{Hf}`$ to $`X_H`$. That is: eq. 5.289 implies that if $`yN`$ is such that $`f(y)U`$, then
$$X_H(f(y))=(T_yf)(X_{Hf}(y)).$$
(5.295)
So $`X_H(f(y))`$ is in the range of $`T_yf`$; so $`f:NM`$ is a Poisson immersion. QED.
Now suppose that the inclusion $`id:NM`$ is a Poisson immersion. Then we call $`N`$ a Poisson submanifold of $`M`$. We emphasise, in line with the warning we recalled from (2) of Section 3.3.1, that $`N`$ need not be a submanifold of $`M`$; but it is nevertheless called an ‘injectively immersed submanifold’ of $`M`$.
From the definition of a Poisson immersion, it follows that any Hamiltonian vector field must be tangent to a Poisson submanifold. In other words: writing $`𝒳`$ for the system of Hamiltonian vector fields on $`M`$, and $`𝒳_x`$ for their values at $`xM`$, we have: if $`N`$ is a Poisson submanifold of $`M`$, and $`xN`$, $`𝒳_xTN_x`$.
For the special case where $`M`$ is a symplectic manifold, we have $`𝒳_x=T_xM`$, and the only Poisson submanifolds of $`M`$ are its open sets.
Finally, we define the following equivalence relation on a Poisson manifold $`M`$. Two points $`x_1,x_2M`$ are on the same symplectic leaf if there is a piecewise smooth curve in $`M`$ joining them, each segment of which is an integral curve of a locally defined Hamiltonian vector field. An equivalence class of this equivalence relation is a symplectic leaf.
We can now state and prove that Poisson manifolds are foliated.
##### 7.3.3.A Foliation theorem for Poisson manifolds
The result is:—
> A Poisson manifold $`M`$ is the disjoint union of its symplectic leaves. Each symplectic leaf is an injectively immersed Poisson submanifold, and the induced Poisson structure on the leaf is symplectic. The leaf through the point $`x`$, $`N_x`$ say, has dimension equal to the rank of the Poisson structure at $`x`$; and the tangent space to the leaf at $`x`$ equals
>
> $`TN_x=\mathrm{ran}(𝖡_x):=\{XT_xM:X=𝖡_x(\alpha ),\mathrm{some}\alpha T_x^{}M\}`$ (5.296)
> $`=\{X_H(x):H(U),U\mathrm{open}\mathrm{in}M\}`$ (5.297)
Proof: We apply the general form of Frobenius’ theorem (Section 3.3.2) to the system $`𝒳`$ of Hamiltonian vector fields on $`M`$. We know from eq. 5.245 (Section 5.2.2) that $`𝒳`$ is involutive, and from eq. 5.287 above that it is rank-invariant. So by Frobenius’ theorem, $`𝒳`$ is integrable. The integral submanifolds are by definition given by the rhs of eq. 5.297. QED.
One also readily shows that:
(i): One can evaluate the Poisson bracket of $`F,G:M\mathrm{I}\mathrm{R}`$ at $`xM`$ by restricting $`F`$ and $`G`$ to the symplectic leaf $`N_x`$ through $`x`$, and evaluating the Poisson bracket that is defined by the symplectic form on the leaf $`N_x`$; (i.e. the Poisson bracket defined in eq. 2.18).
(ii): A distinguished function is constant on any symplectic leaf $`N_x`$ of $`M`$.
We end with two remarks. The first is a mathematical warning; the second concerns physical interpretation.
(1): Recall our warning that symplectic leaves need not be submanifolds. This also means that all the distinguished functions being constants does not imply that the Poisson structure is non-degenerate. Indeed, one can readily construct an example in which the symplectic leaves are not manifolds, all distinguished functions are constants, and the Poisson structure is degenerate. Namely, one adapts an example mentioned before, in Section 3.4.3: the flows on the torus $`𝖳^2`$ that wind densely around it. (For more details about this example, cf. Arnold (1973: 160-167) or Arnold (1989: 72-74) or Butterfield (2004a: Section 2.1.3.B); for how to adapt it, cf. Marsden and Ratiu (1999: 347).
(2): As we have seen, any integral curve of any Hamiltonian vector field $`X_H`$ is confined to one of the symplectic leaves. So if we are interested only in the behaviour of a single solution through a point $`xM`$, we can restrict our attention to the symplectic leaf $`N_x`$ through $`x`$: for the solution will always remain in $`N_x`$. But as stressed in Section 5.1, there are at least three good reasons not to ignore the more general Poisson structure!
#### 5.3.4 Darboux’s theorem
At the end of Section 2.1.1, we mentioned Darboux’s theorem: it said that any symplectic manifold “looks locally like” a cotangent bundle. The generalization for Poisson manifolds says that any Poisson manifold “looks locally like” our canonical example on $`\mathrm{I}\mathrm{R}^m,m=2n+l`$, given at the start of Section 5.2.1. More precisely, we have:
> Let $`M`$ be an $`m`$-dimensional Poisson manifold, and let $`xM`$ be a point with an open neighbourhood $`UM`$ throughout which the rank is a constant $`2nm`$. Then defining $`l:=m2n`$, there is a possibly smaller neighbourhood $`U^{}U`$ of $`x`$, on which there exist local coordinates $`(q,p,z)=(q^1,\mathrm{},q^n,p^1,\mathrm{},p^n,z^1,\mathrm{},z^l)`$, for which the Poisson bracket takes the form
>
> $$\{F,H\}:=\mathrm{\Sigma }_i^n\left(\frac{F}{q^i}\frac{H}{p^i}\frac{F}{p^i}\frac{H}{q^i}\right).$$
> (5.298)
> (So the Poisson brackets for the coordinate functions take the now-familiar form given by eq. 5.240 and 5.241.) The symplectic leaves of $`M`$ intersect the coordinate chart in the slices $`\{z^1=c_1,\mathrm{},z^l=c_l\}`$ given by constant values of the distinguished coordinates $`z`$.
We shall not give the proof. Suffice it to say that:
(i): Like Darboux’s theorem for symplectic manifolds: it proceeds by induction on the “half-rank” $`n`$; and it begins by taking any function $`F`$ as the “momentum” $`p^1`$ and constructing the canonically conjugate coordinate $`q^1`$ such that $`\{q^1,p^1\}=1`$.
(ii): The induction step invokes a version of Frobenius’ theorem in which the fact that the rank $`2n`$ is constant throughout $`U`$ secures a coordinate system in which the $`2n`$-dimensional integral manifolds are given by slices defined by constant values of the remaining $`l`$ coordinates. The Poisson structure then secures that these remaining coordinates are distinguished.
##### 7.3.4.A Example: $`𝔰𝔬(3)^{}`$ yet again
We illustrate (1) the foliation theorem and (2) Darboux’s theorem, with $`𝔰𝔬(3)^{}`$; whose Lie-Poisson structure we described in Section 5.2.4.A.
(1): At $`x𝔰𝔬(3)^{}`$, the subspace $`𝒳_x:=\{X_H(x):H(U),U\mathrm{open}\mathrm{in}M\}`$ of values of locally Hamiltonian vector fields is spanned by $`e_1:=y_zz_y`$ representing infinitesimal rotation about the $`x`$-axis (cf. eq. LABEL:infmlrotnx); $`e_2:=z_xx_z`$ for rotation about the $`y`$-axis; and $`e_3:=x_yy_x`$ for rotation about the $`z`$-axis. If $`x0`$, these vectors span a two-dimensional subspace of $`T𝔰𝔬(3)_x^{}`$: viz. the tangent plane to the sphere $`S_x`$ of radius $`x`$ centred at the origin. So the foliation theorem implies that $`𝔰𝔬(3)^{}`$’s symplectic leaves are these spheres; and the origin.
We can compute the Poisson bracket of $`F,G:S_x\mathrm{I}\mathrm{R}`$ by extending $`F`$ and $`G`$ to a neighbourhood of $`S_x`$; cf. eq. 5.293. That is: we can consider extensions $`\overline{F},\overline{G}:US_x\mathrm{I}\mathrm{R}`$, and calculate the Poisson bracket in $`𝔰𝔬(3)^{}`$, whose Poisson structure we already computed in eq. 5.267.
Adopting spherical polar coordinates with $`r=x`$, i.e. $`x^1=r\mathrm{cos}\theta \mathrm{sin}\varphi ,x^2=r\mathrm{sin}\theta \mathrm{sin}\varphi ,x^3=r\mathrm{cos}\varphi `$, we can define $`\overline{F},\overline{G}`$ merely by $`\overline{F}(r,\theta ,\varphi ):=F(\theta ,\varphi ),\overline{G}(r,\theta ,\varphi ):=G(\theta ,\varphi )`$; so that the partial derivatives with respect to the spherical angles $`\theta ,\varphi `$ are equal, i.e. $`\overline{F}_\theta =F_\theta ,\overline{F}_\varphi =F_\varphi ,\overline{G}_\theta =G_\theta ,\overline{G}_\varphi =G_\varphi `$.
Besides, eq. 5.249 implies that we need only calculate the Poisson bracket in $`𝔰𝔬(3)^{}`$ of the spherical angles $`\theta `$ and $`\varphi `$. So eq. 5.267 gives
$$\{\theta ,\varphi \}=x(\theta \times \varphi )=\frac{1}{r\mathrm{sin}\varphi };$$
(5.299)
and eq. 5.293 and 5.249 give
$$\{F,G\}=\{\overline{F},\overline{G}\}=\frac{1}{r\mathrm{sin}\varphi }(F_\theta G_\varphi F_\varphi G_\theta ).$$
(5.300)
(2): $`z:=x^3`$ defines the Hamiltonian vector field $`X_z=x^2_{x^1}x^1_{x^2}`$ that generates clockwise rotation about the $`zx^3`$-axis. So away from the origin the polar angle $`\theta :=\mathrm{arctan}(x^2/x^1)`$ has a Poisson bracket with $`z`$ equal to: $`\{\theta ,z\}=X_z(\theta )=1`$. Exprssing $`F,H:𝔰𝔬(3)^{}\mathrm{I}\mathrm{R}`$ in terms of the coordinates $`z,\theta `$ and $`r:=x`$, we find that the Lie-Poisson bracket is: $`\{F,H\}=F_zH_\theta F_\theta H_z`$. So $`(z,\theta ,r)`$ are canonical coordinates.
### 5.4 The symplectic structure of the co-adjoint representation
Section 5.2.4 described how the dual $`𝔤^{}`$ of a finite-dimensional Lie algebra of a Lie group $`G`$ has the structure of a Poisson manifold. In this case, the foliation established in the previous Subsection has an especially neat interpretation. Namely: the leaves are the orbits of the co-adjoint representation of $`G`$ on $`𝔤^{}`$.
This symplectic structure in the co-adjoint representation sums up themes from Sections 4.5 (especially 4.5.2), and 5.2.4 and 5.3. In particular, it connects two properties of the Lie bracket in $`𝔤`$, which we have already seen: viz.
(i): The Lie bracket in $`𝔤`$ gives the infinitesimal generators of the adjoint action; cf. eq. 4.179.
(ii): The Lie bracket in $`𝔤`$ defines (in a basis-independent way) a Lie-Poisson bracket on $`𝔤^{}`$, thus making $`𝔤^{}`$ a Poisson manifold. (Cf. the definition in eq. 5.258, shown to be basis-independent by eq. 5.261.)
In fact, there is a wealth of instructive results and examples about the structure of the co-adjoint representation: we will only scratch the surface—as in other Sections! We will give a proof, under a simplifying assumption, of one main result; and then make a few remarks about other results.
The result is:
> The orbits of the co-adjoint representation are $`𝔤^{}`$’s leaves
> Let $`G`$ be a Lie group, with its co-adjoint representation $`Ad^{}`$ on $`𝔤^{}`$. That is, recalling eq. 4.193, we have:
>
> $$Ad^{}:G\mathrm{End}(𝔤^{}),Ad_{g^1}^{}=(T_e(R_gL_{g^1}))^{}.$$
> (5.301)
> The orbits of this representation are the symplectic leaves of $`𝔤^{}`$, taken as equipped with its natural Poisson structure, i.e. the Lie-Poisson bracket eq. 5.261.
Proof:— We shall prove this under the simplifying assumption that the co-adjoint action of $`G`$ on $`𝔤^{}`$ is proper. (We recall from the definition of proper actions, eq. 4.147, that for any compact Lie group, such as $`SO(3)`$, this condition is automatically satisfied.) Then we know from result (3) and eq. 4.160, at the end of Section 4.4, that this implies that the co-adjoint orbit $`\mathrm{Orb}(\alpha )`$ of any $`\alpha 𝔤^{}`$ is a closed submanifold of $`𝔤^{}`$, and that the tangent space to $`\mathrm{Orb}(\alpha )`$ at a point $`\beta \mathrm{Orb}(\alpha )`$ is
$$T\mathrm{Orb}(\alpha )_\beta =\{\xi _𝔤^{}(\beta ):\xi 𝔤\}.$$
(5.302)
We will see shortly how this assumption implies that $`𝔤^{}`$’s symplectic leaves are submanifolds.<sup>22</sup><sup>22</sup>22To verify that our condition is indeed simplifying—i.e. that in general the co-adjoint orbits in $`𝔤^{}`$ are not submanifolds—consider the example in Marsden and Ratiu (1999: 14.1.(f), p. 449); taken from Kirillov (1976: 293).
We now argue as follows. For $`\xi 𝔤`$, consider the scalar function on $`𝔤^{}`$, $`K_\xi :\alpha 𝔤^{}K_\xi (\alpha ):=<\alpha ;\xi >\mathrm{I}\mathrm{R}`$; and its Hamiltonian vector field $`X_{K_\xi }`$. At each $`\alpha 𝔤^{}`$, the gradient $`K_\xi (\alpha )dK_\xi (\alpha )`$, considered as an element of $`(T^{}𝔤^{})_\alpha 𝔤`$, is just $`\xi `$ itself. Now we will compute $`X_{K_\xi }(F)(\alpha )`$ for any $`F:𝔤^{}\mathrm{I}\mathrm{R}`$ and any $`\alpha 𝔤^{}`$, using in order:
(i): the intrinsic definition of the Lie-Poisson bracket on $`𝔤^{}`$, eq. 5.261;
(ii): the fact that the infinitesimal generator of the adjoint action is the Lie bracket in $`𝔤`$, eq. 4.179;
(iii): the fact that the derivative $`ad^{}`$ of the co-adjoint action $`Ad^{}`$ is, up to a sign, the adjoint of $`ad_\xi `$; eq. 4.198.
Thus we get, for all $`F:𝔤^{}\mathrm{I}\mathrm{R}`$ and $`\alpha 𝔤^{}`$:
$`X_{K_\xi }(F)(\alpha )\{F,K_\xi \}(\alpha )=<\alpha ;[F(\alpha ),K_\xi (\alpha )]>`$ (5.303)
$`=<\alpha ;[F(\alpha ),\xi ]>=<\alpha ;[\xi ,F(\alpha )]>`$ (5.304)
$`=<\alpha ;ad_\xi (F(\alpha ))>`$ (5.305)
$`=<ad_\xi ^{}(\alpha );F(\alpha )>.`$ (5.306)
But on the other hand, the vector field $`X_{K_\xi }`$ is uniquely determined by its action on all such functions $`F`$ at all $`\alpha 𝔤^{}`$:
$$X_{K_\xi }(F)(\alpha )<X_{K_\xi }(\alpha );F(\alpha )>.$$
(5.307)
So we conclude that at each $`\alpha 𝔤^{}`$:
$$X_{K_\xi }=ad_\xi ^{}.$$
(5.308)
But the subspace $`𝒳_\alpha `$ of values at $`\alpha `$ of Hamiltonian vector fields is spanned by the $`X_{K_\xi }(\alpha )`$, with $`\xi `$ varying through $`𝔤`$. And as $`\xi `$ varies through $`𝔤`$, $`ad_\xi ^{}(\alpha )`$ is the tangent space $`T\mathrm{Orb}(\alpha )_\alpha `$ to the co-adjoint orbit $`\mathrm{Orb}(\alpha )`$ of $`G`$ through $`\alpha `$. So
$$𝒳_\alpha =T\mathrm{Orb}(\alpha )_\alpha .$$
(5.309)
So the integral submanifolds of the system $`𝒳`$ of Hamiltonian vector fields, which are the symplectic leaves of $`𝔤^{}`$ by Section 5.3.3.A’s foliation theorem, are the co-adjoint orbits. QED.
For the illustration of this theorem by our standard example, $`𝔰𝔬(3)^{}`$, cf. our previous discussions of it: in Section 4.5.2 for its co-adjoint structure; in Section 5.2.4.A for its Lie-Poisson structure; and in Section 5.3.4.A for its symplectic leaf structure.
We end this Subsection by stating two other results. They are not needed later, but they are enticing hints of how rich is the theory of co-adjoint orbits.
(1): For each $`gG`$, the co-adjoint map $`Ad_g^{}:𝔤^{}𝔤^{}`$ is a Poisson map that preserves the symplectic leaves of $`𝔤^{}`$.
(2): A close cousin of the theorem just proven is that the Lie bracket on $`𝔤`$ defines (via its definition of the Lie-Poisson bracket on $`𝔤^{}`$, eq. 5.261) a symplectic form, i.e. a non-degenerate closed two-form, on each co-adjoint orbit, by:
$$\omega (\alpha )(ad_\xi ^{}(\alpha ),ad_\eta ^{}(\alpha )):=<\alpha ;[\xi ,\eta ]_𝔤>,\alpha 𝔤^{},\xi ,\eta 𝔤.$$
(5.310)
This theorem is proven in detail (without our simplifying assumption that $`G`$’s action is proper) by Marsden and Ratiu (1999: Thm 14.3.1, pp. 453-456); and much more briefly by Arnold (1989: 321, 376-377, 457); and rather differently (even without using the notion of a Poisson manifold!) in Abraham and Marsden (1978: 302-303).
### 5.5 Quotients of Poisson manifolds
We now end Section 5 with the simplest general theorem about quotienting a Lie group action on a Poisson manifold, so as to get a quotient space (set of orbits) that is itself a Poisson manifold. So this theorem combines themes from Sections 4—in particular, the idea from Section 4.3.B that for a free and proper group action, the orbits and quotient space are manifolds—with material about Poisson manifolds from Section 5.2. (The material in Sections 5.3 and 5.4 will not be needed.) This theorem will be important in Section 7. We call this result the
> Poisson reduction theorem: Suppose the Lie group $`G`$ acts on Poisson manifold $`M`$ is such a way that each $`\mathrm{\Phi }_g:MM`$ is a Poisson map. Suppose also that the quotient space $`M/G`$ is a manifold and the projection $`\pi :MM/G`$ is a smooth submersion (say because $`G`$’s action on $`M`$ is free and proper, cf. Section 4.3.B). Then there is a unique Poisson structure on $`M/G`$ such that $`\pi `$ is a Poisson map. The Poisson bracket on $`M/G`$ is called the reduced Poisson bracket.
Proof: Let us first assume that $`M/G`$ is a Poisson manifold and that $`\pi `$ is a Poisson map; and show uniqueness. We first note that for any $`f:M/G\mathrm{I}\mathrm{R}`$, the function $`\overline{f}:=f\pi :M\mathrm{I}\mathrm{R}`$ is obviously the unique $`G`$-invariant function on $`M`$ that projects by $`\pi `$ to $`f`$. That is: if $`[x]\mathrm{Orb}(x)Gx`$ is the orbit of $`xM`$, then $`\overline{f}`$ assigns the same value $`f([x])`$ to all elements of the orbit $`[x]`$. Besides, in terms of pullbacks (eq. 5.285), $`\overline{f}=\pi ^{}f`$.
Then the condition that $`\pi `$ be Poisson, eq. 5.286, is that for any two smooth scalars $`f,h:M/G\mathrm{I}\mathrm{R}`$, we have an equation of smooth scalars on $`M`$:
$$\{f,h\}_{M/G}\pi =\{f\pi ,h\pi \}_M=\{\overline{f},\overline{h}\}_M$$
(5.311)
where the subscripts indicate on which space the Poisson bracket is defined. Since $`\pi `$ is surjective, eq. 5.311 determines the value $`\{f,h\}_{M/G}`$ uniquely.
But eq. 5.311 also defines $`\{f,h\}_{M/G}`$ as a Poisson bracket; in two stages. (1): The facts that $`\mathrm{\Phi }_g`$ is Poisson, and $`\overline{f}`$ and $`\overline{h}`$ are constant on orbits imply that
$$\{\overline{f},\overline{h}\}(gx)=(\{\overline{f},\overline{h}\}\mathrm{\Phi }_g)(x)=\{\overline{f}\mathrm{\Phi }_g,\overline{h}\mathrm{\Phi }_g\}(x)=\{\overline{f},\overline{h}\}(x).$$
(5.312)
That is: $`\{\overline{f},\overline{h}\}`$ is also constant on orbits, and so defines $`\{f,h\}`$ uniquely.
(2): We show that $`\{f,h\}`$, as thus defined, is a Poisson structure on $`M/G`$, by checking that the required properties, such as the Jacobi identity, follow from the Poisson structure $`\{,\}_M`$ on $`M`$. QED.
This theorem is a “prototype” for material to come. We spell this out in two brief remarks, which look forward to the following two Sections.
(1): Other theorems:— This theorem is one of many that yield new Poisson manifolds and symplectic manifolds from old ones by quotienting. In particular, as we will see in detail in Section 7, this theorem is exemplified by the case where $`M=T^{}G`$ (so here $`M`$ is symplectic, since it is a cotangent bundle), and $`G`$ acts on itself by left translations, and so acts on $`T^{}G`$ by a cotangent lift. In this case, we will have $`M/G𝔤^{}`$; and the reduced Poisson bracket just defined, by eq. 5.311, will be the Lie-Poisson bracket we have already met in Section 5.2.4.
(2): Reduction of dynamics:— Using this theorem, we can already fill out a little what is involved in reduced dynamics; which we only glimpsed in our introductory discussions, in Section 2.3 and 5.1. We can make two basic points, as follows.
(A): If $`H`$ is a $`G`$-invariant Hamiltonian function on $`M`$, it defines a corresponding function $`h`$ on $`M/G`$ by $`H=h\pi `$. The fact that Poisson maps push Hamiltonian flows forward to Hamiltonian flows (eq. 5.289) implies, since $`\pi `$ is Poisson, that $`\pi `$ transforms $`X_H`$ on $`M`$ to $`X_h`$ on $`M/G`$. That is:
$$T\pi X_H=X_h\pi ;$$
(5.313)
i.e. $`X_H`$ and $`X_h`$ are $`\pi `$-related. Accordingly, we say that the Hamiltonian system $`X_H`$ on $`M`$ reduces to that on $`M/G`$.
(B): We shall see in Section 6.2 that $`G`$-invariance of $`H`$ is associated with a family of conserved quantities (constants of the motion, first integrals), viz. a constant of the motion $`J(\xi ):M\mathrm{I}\mathrm{R}`$ for each $`\xi 𝔤`$. Here, $`J`$ being conserved means $`\{J,H\}=0`$; just as in our discussion of Noether’s theorem in ordinary Hamiltonian mechanics (Section 2.1.3). Besides, if $`J`$ is also $`G`$-invariant, then the corresponding function $`j`$ on $`M/G`$ is conserved by $`X_h`$ since
$$\{j,h\}\pi =\{J,H\}=0\mathrm{implies}\{j,h\}=0.$$
(5.314)
## 6 Symmetry and conservation revisited: momentum maps
We now develop the topics of symmetry and conserved quantities (and so Noether’s theorem) in the context of Poisson manifolds. At the centre of these topics lies the idea of a momentum map of a Lie group action on a Poisson manifold; which we introduce in Section 6.1. This is the modern geometric generalization of a conserved quantity, such as linear or angular momentum for the Euclidean group—hence the name. Formally, it will be a map $`𝐉`$ from the Poisson manifold $`M`$ to the dual $`𝔤^{}`$ of the Lie algebra of the symmetry group $`G`$. Since its values lie in a vector space, it has components. So our description of conserved quantities will no longer be “one-dimensional”, i.e. focussed on a single vector field in the state space, as it was in Sections LABEL:NoetherLag and LABEL:Hammechs. The map $`𝐉`$ will be associated with a linear map $`J`$ from $`𝔤`$ to $`(M)`$, the scalar functions on the manifold $`M`$. That is: for each $`\xi 𝔤`$, $`J(\xi )`$ will be a conserved quantity if the Hamiltonian $`H`$ is invariant under the infinitesimal generator $`\xi _M`$, i.e. if $`\xi _M(H)=0`$.
The conservation of momentum maps will be expressed by the Poisson manifold version of Noether’s theorem (Section 6.2), and illustrated by the familiar examples of linear and angular momentum (Section 6.3). Then we discuss the equivariance of momentum maps, with respect to the co-adjoint representation of $`G`$ on $`𝔤^{}`$; Section 6.4. Finally in Section 6.5, we discuss the crucial special case of momentum maps on cotangent bundles, again with examples.
### 6.1 Canonical actions and momentum maps
We first apply the definition of Poisson maps (from Section 5.3.2) to group actions (Section 6.1.1). This will lead to the idea of the momentum map (Section 6.1.2).
#### 6.1.1 Canonical actions and infinitesimal generators
Let $`G`$ be a Lie group acting on a Poisson manifold $`M`$ by a smooth left action $`\mathrm{\Phi }:G\times MM`$; so that as usual we write $`\mathrm{\Phi }_g:xM\mathrm{\Phi }_g(x):=gxM`$. As in the definition of a Poisson map (eq. 5.286), we say the action is canonical if
$$\mathrm{\Phi }_g^{}\{F_1,F_2\}=\{\mathrm{\Phi }_g^{}F_1,\mathrm{\Phi }_g^{}F_2\}$$
(6.315)
for any $`F_1,F_2(M)`$ and any $`gG`$. If $`M`$ is symplectic with symplectic form $`\omega `$, then the action is canonical iff it is symplectic, i.e. $`\mathrm{\Phi }_g^{}\omega =\omega `$ for all $`gG`$.
We will be especially interested in the infinitesimal version of this notion; and so with infinitesimal generators of actions. We recall from eq. 4.152 that the infinitesimal generator of the action corresponding to a Lie algebra element $`\xi 𝔤`$ is the vector field $`\xi _M`$ on $`M`$ obtained by differentiating the action with respect to $`g`$ at the identity in the direction $`\xi `$:
$$\xi _M(x)=\frac{d}{d\tau }[\mathrm{exp}(\tau \xi )x]_{\tau =0}.$$
(6.316)
So we differentiate eq. 6.315 with respect to $`g`$ in the direction $`\xi `$, to give:
$$\xi _M(\{F_1,F_2\})=\{\xi _M(F_1),F_2\}+\{F_1,\xi _M(F_2)\}.$$
(6.317)
Such a vector field $`\xi _M`$ is called an infinitesimal Poisson automorphism.
Side-remark:— We will shortly see that it is the universal quantification over $`gG`$ in eq. 6.315, and correspondingly in eq. 6.317 and 6.319 below, that means our description of conserved quantities is no longer focussed on a single vector field; and in particular, that a momentum map representing a conserved quantity has components.
In the symplectic case, differentiating $`\mathrm{\Phi }_g^{}\omega =\omega `$ implies that the Lie derivative $`_{\xi _M}\omega `$ of $`\omega `$ with respect to $`\xi `$ vanishes: $`_{\xi _M}\omega =0`$. We saw in Section 2.1.3 that this is equivalent to $`\xi _M`$ being locally Hamiltonian, i.e. there being a local scalar $`J:UM\mathrm{I}\mathrm{R}`$ such that $`\xi _M=X_J`$. This was how Section 2.1.3 vindicated eq. 2.19’s “one-liner” approach to Noether’s theorem: because the vector field $`X_f`$ is locally Hamiltonian, it preserves the symplectic structure, i.e. Lie-derives the symplectic form $`_{X_f}\omega =0`$—as a symmetry should.
We also saw in result (2) at the end of Section 3.2.2 that the “meshing”, up to a sign, of the Poisson bracket on scalars with the Lie bracket on vector fields implied that the locally Hamiltonian vector fields form a Lie subalgebra of the Lie algebra $`𝒳(M)`$ of all vector fields.
Turning to the context of Poisson manifolds, we need to note two points. The first is a similarity with the symplectic case; the second is a contrast.
(1): One readily checks, just by applying eq. 6.317, that the infinitesimal Poisson automorphisms are closed under the Lie bracket. So we write the Lie algebra of these vector fields as $`𝒫(M)`$: $`𝒫(M)𝒳(M)`$.
(2): On the other hand, Section 2.1.3’s equivalence between a vector field being locally Hamiltonian and preserving the geometric structure of the state-space breaks down.
Agreed, the first implies the second: a locally Hamiltonian vector field preserves the Poisson bracket. We noted this already in Section 5.3.2. The differential statement was that such a field $`X_H`$ Lie-derives the Poisson tensor: $`_{X_H}𝖡^{\mathrm{}}=0`$ (eq. 5.288). The finite statement was that the flows of such a field are Poisson maps: $`\varphi _\tau ^{}\{F,G\}=\{\varphi _\tau ^{}F,\varphi _\tau ^{}G\}`$ (eq. 5.287).
But the converse implication fails: an infinitesimal Poisson automorphism on a Poisson manifold need not be locally Hamiltonian. For example, make $`\mathrm{I}\mathrm{R}^2`$ a Poisson manifold by defining the Poisson structure
$$\{F,H\}=x\left(\frac{F}{x}\frac{H}{y}\frac{H}{x}\frac{F}{y}\right);$$
(6.318)
then the vector field $`X=/y`$ in a neighbourhood of a point on the $`y`$-axis is a non-Hamiltonian infinitesimal Poisson automorphism.
This point will affect the formulation of Noether’s theorem for Poisson manifolds, in Section 6.2.
Nevertheless, we shall from now on be interested in cases where for all $`\xi `$, $`\xi _M`$ is globally Hamiltonian. This means there is a map $`J:𝔤(M)`$ such that
$$X_{J(\xi )}=\xi _M$$
(6.319)
for all $`\xi 𝔤`$. There are three points we need to note about this condition.
(1): Since the right hand side of eq. 6.319 is linear in $`\xi `$, we can require such a $`J`$ to be a linear map. For given any $`J`$ obeying eq. 6.319, we can take a basis $`e_1,\mathrm{},e_m`$ of $`𝔤`$ and define a new linear $`\overline{J}`$ by setting, for any $`\xi =\xi ^ie_i`$, $`\overline{J}(\xi ):=\xi ^iJ(e_i)`$.
(2): Eq. 6.319 does not determine $`J(\xi )`$. For by the linearity of the map $`𝖡:dJ(\xi )X_{J(\xi )}`$, we can add to such a $`J(\xi )`$ any distinguished function, i.e. an $`F:M\mathrm{I}\mathrm{R}`$ such that $`X_F=0`$. That is: $`X_{J(\xi )+F}X_{J(\xi )}`$. (Of course, in the symplectic case, the only distinguished functions are constants.)
(3): It is worth expressing eq. 6.319 in terms of Poisson brackets. Recalling that for any $`F,H(M)`$, we have $`X_H(F)=\{F,H\}`$, this equation becomes
$$\{F,J(\xi )\}=\xi _M(F),F(M),\xi 𝔤.$$
(6.320)
We will also need the following result:
$$X_{J([\xi ,\eta ])}=X_{\{J(\xi ),J(\eta )\}_M}.$$
(6.321)
To prove this, we just apply two previous results, each giving a Lie algebra anti-homomorphism.
(i): Result (4) at the end of Section 4.4: for any left action of Lie group $`G`$ on any manifold $`M`$, the map $`\xi \xi _M`$ is a Lie algebra anti-homomorphism between $`𝔤`$ and the Lie algebra $`𝒳_M`$ of all vector fields on $`M`$:
$$(a\xi +b\eta )_M=a\xi _M+b\eta _M;[\xi _M,\eta _M]=[\xi ,\eta ]_M\xi ,\eta 𝔤,\mathrm{and}a,b\mathrm{I}\mathrm{R}.$$
(6.322)
(ii): The “meshing” up to a sign, just as in the symplectic case, of the Poisson bracket on scalars with the Lie bracket on vector fields, as in eq. 5.245 at the end of Section 5.2.2:
$$X_{\{F,H\}}=[X_F,X_H]=[X_H,X_F].$$
(6.323)
So for a Poisson manifold $`M`$, the map $`F(M)X_F𝒳(M)`$ is a Lie algebra anti-homomorphism.
Applying (i) and (ii), we deduce eq. 6.321 by:
$$X_{J([\xi ,\eta ])}=[\xi ,\eta ]_M=[\xi _M,\eta _M]=[X_{J(\xi )},X_{J(\eta )}]=X_{\{J(\xi ),J(\eta )\}_M}.$$
(6.324)
#### 6.1.2 Momentum maps introduced
So suppose that there is a canonical left action of $`G`$ on a Poisson manifold $`M`$. And suppose there is a linear map $`J:𝔤(M)`$ such that
$$X_{J(\xi )}=\xi _M$$
(6.325)
for all $`\xi 𝔤`$.
The two requirements—that the action be infinitesimally canonical (i.e. each $`\xi _M𝒫(M)`$) and that each $`\xi _M`$ be globally Hamiltonian—can be expressed as requiring that there be a $`J:𝔤(M)`$ such that there is a commutative diagram. Namely, the map $`\xi 𝔤\xi _M𝒫(M)`$ is to equal the composed map:
$$𝔤\stackrel{J}{}(M)\stackrel{FX_F}{}𝒫(M).$$
(6.326)
Then the map $`𝐉:M𝔤^{}`$ defined by
$$<𝐉(x);\xi >:=J(\xi )(x)$$
(6.327)
for all $`\xi 𝔤`$ and $`xM`$, is called the momentum map of the action.
Another way to state this definition is as follows. Any smooth function $`𝐉:M𝔤^{}`$ defines at each $`\xi 𝔤`$ a scalar $`J(\xi ):xM(𝐉(x))(\xi )\mathrm{I}\mathrm{R}`$. By taking $`J(\xi )`$ as a Hamiltonian function, one defines a Hamiltonian vector field $`X_{J(\xi )}`$. But since $`G`$ acts on $`M`$, each $`\xi 𝔤`$ defines a vector field on $`M`$, viz. $`\xi _M`$. So we say that $`𝐉`$ is a momentum map for the action if for each $`\xi 𝔤`$, these two vector fields are identical: $`X_{J(\xi )}=\xi _M`$.
Three further remarks by way of illustrating this definition:—
(1): An isomorphism:— One readily checks that eq. 6.327 defines an isomorphism between the space of smooth maps $`𝐉`$ from $`M`$ to $`𝔤^{}`$, and the space of linear maps $`J`$ from $`𝔤`$ to scalar functions $`(M)`$. We can take $`J`$ to define $`𝐉`$ by saying that at each $`xM`$, $`𝐉(x):\xi 𝔤𝐉(x)(\xi )\mathrm{I}\mathrm{R}`$ is to be given by the composed map
$$𝔤\stackrel{J}{}(M)\stackrel{_x}{}\mathrm{I}\mathrm{R},$$
(6.328)
where $`_x`$ means evaluation at $`xM`$. Or we can take $`𝐉`$ to define $`J`$ by saying that at each $`\xi 𝔤`$, $`J(\xi ):xMJ(\xi )(x)\mathrm{I}\mathrm{R}`$ is to be given by the composed map
$$M\stackrel{𝐉}{}𝔤^{}\stackrel{_\xi }{}\mathrm{I}\mathrm{R},$$
(6.329)
where $`_\xi `$ means evaluation at $`\xi 𝔤`$.
(2): Differential equations for the momentum map:— Using Hamilton’s equations, we can readily express the definition of momentum map as a set of differential equations. Recall that on a Poisson manifold, Hamilton’s equations are determined by eq. 5.273, which was that at each $`xM`$
$$𝖡_x(dH(x))=X_H(x);$$
(6.330)
or in local coordinates $`x^i,i=1,\mathrm{},m\mathrm{dim}(M)`$, with $`J^{ij}(x)\{x^i,x^j\}`$ the structure matrix,
$$𝖡_x(\frac{H}{x^j}dx^j)=\mathrm{\Sigma }_{i,j}J^{ij}(x)\frac{H}{x^j}\frac{}{x^i}_x;$$
(6.331)
(cf. eq. 5.275). So in local coordinates, Hamilton’s equations are given by eq. 5.256, which was:
$$\frac{dx^i}{dt}=\mathrm{\Sigma }_j^mJ^{ij}(x)\frac{H}{x^j}.$$
(6.332)
So the condition for a momentum map $`X_{J(\xi )}=\xi _M`$ is that for all $`\xi 𝔤`$ and all $`xM`$
$$𝖡_x(d(J(\xi ))(x))=\xi _M(x).$$
(6.333)
In coordinates, this is the requirement that for all $`i=1,\mathrm{},m`$
$$\mathrm{\Sigma }_j^mJ^{ij}(x)\frac{J(\xi )}{x^j}=(\xi _M)^i(x),$$
(6.334)
where—apologies!—the two $`J`$s on the left hand side have very different meanings.
In the symplectic case, $`\mathrm{dim}(M)m=2n`$ and we have Hamilton’s equations as eq. 2.15, viz.
$$𝐢_{X_H}\omega :=\omega (X_H,)=dH().$$
(6.335)
So the condition for a momentum map is that for all $`\xi `$
$$\omega (\xi _M,)=d(J(\xi ))().$$
(6.336)
In Hamiltonian mechanics, it is common to write the $`2n`$ local coordinates $`q,p`$ as $`\xi `$, i.e. to write
$$\xi ^\alpha :=q^\alpha ,\alpha =1,\mathrm{},n;\xi ^\alpha :=p_{\alpha n},\alpha =n+1,\mathrm{},2n.$$
(6.337)
So in order to express eq. 6.336 in local coordinates, let us temporarily write $`\eta `$ for the arbitrary element of $`𝔤`$. Then writing $`\eta _M=(\eta _M)^\alpha \frac{}{\xi ^\alpha }`$ and $`\omega _{\alpha \beta }:=\omega (\frac{}{\xi ^\alpha },\frac{}{\xi ^\beta })`$, eq. 6.336 becomes
$$\omega _{\alpha \beta }(\eta _M)^\alpha =\frac{J(\eta )}{\xi ^\beta }.$$
(6.338)
(3): Components: an example:— As discussed after eq. 6.317, we think of the collection of functions $`J(\xi )`$, as $`\xi `$ varies through $`𝔤`$, as the components of $`𝐉`$.
To take our standard example: the angular momentum of a particle in Euclidean space, in a state $`x=(𝐪,𝐩)`$ is $`𝐉(x):=𝐪𝐩`$. Identifying $`𝔰𝔬(3)^{}`$ with $`\mathrm{I}\mathrm{R}^3`$ so that the natural pairing is given by the dot product (cf. (3) at the end of Section 4.5.2), we get that the component of $`𝐉(x)`$ around the axis $`\xi \mathrm{I}\mathrm{R}^3`$ is $`<𝐉(x);\xi >=\xi (𝐪𝐩)`$. The Hamiltonian vector field determined by this Hamiltonian function $`x=(𝐪,𝐩)\xi (𝐪𝐩)`$ is of course the infinitesimal generator of rotations about the $`\xi `$-axis. In Section 6.3, we will see more examples of momentum maps.
### 6.2 Conservation of momentum maps: Noether’s theorem
In ordinary Hamiltonian mechanics, we saw that Noether’s theorem had a simple expression as a “one-liner” based on the antisymmetry of the Poisson bracket: namely, in eq. 2.19, which was that for any scalar functions $`F,H`$
$$X_F(H)=\{H,F\}=0\text{ iff }\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}0}=\{F,H\}=X_H(F).$$
(6.339)
In words: the Hamiltonian $`H`$ is constant under the flow induced by $`F`$ iff $`F`$ is a constant of the motion under the dynamical flow $`X_H`$.
More precisely, Section 2.1.3 vindicated this one-liner as expressing Noether’s theorem. For the one-liner respected the requirement that a symmetry should preserve the symplectic form (equivalently, the Poisson bracket), and not just (as in the left hand side of eq. 6.339) the Hamiltonian function $`H`$; for, by Cartan’s magic formula, a vector field’s preserving the symplectic form was equivalent to its being locally Hamiltonian.
For Poisson manifolds, the equivalence corresponding to this last statement fails. That is, as we noted in (2) of Section 6.1.1: an infinitesimal Poisson automorphism need not be locally Hamiltonian.
Nevertheless, most of the “one-liner” approach to Noether’s theorem carries over to the framework of Poisson manifolds. In effect, we just restrict discussion to cases where the relevant Hamiltonian vector fields exist: recall our saying after (2) of Section 6.1.1 that we would concentrate on cases where all the $`\xi _M`$ are globally Hamiltonian.
Thus, it is straightforward to show that for a Poisson manifold $`M`$, just as for symplectic manifolds: if $`F,H(M)`$, $`H`$ is constant along the integral curves of $`X_F`$ iff $`\{H,F\}=0`$ iff $`F`$ is constant along the integral curves of $`X_H`$. (We could have proved this already in Section 5.2.2; but postponed it till now, when it will be used.)
With this result as a lemma, one immediately gets
> Noether’s theorem for Poisson manifolds Suppose that $`G`$ acts canonically on a Poisson manifold $`M`$ and has a momentum map $`𝐉:M𝔤^{}`$; and that $`H`$ is invariant under $`\xi _M`$ for all $`\xi 𝔤`$, i.e. $`\{H,J(\xi )\}=\xi _M(H)=0,\xi 𝔤`$; (cf. eq. 6.320). Then $`𝐉`$ is a constant of the motion determined by $`H`$. That is:
>
> $$𝐉\varphi _\tau =𝐉$$
> (6.340)
> where $`\varphi _\tau `$ is the flow of $`X_H`$.
Proof: By the lemma, the fact that $`\{H,J(\xi )\}=\xi _M(H)=0`$ implies that $`J(\xi )`$ is constant along the flow of $`X_H`$. So by the definition of momentum map, eq. 6.327, the corresponding $`𝔤^{}`$-valued map $`𝐉`$ is also a constant of the motion. QED.
It follows immediately that $`H`$ itself, and any distinguished function, is a constant of the motion. Besides, as remarked in (2) at the end of Section 6.1.1: a constant of the motion $`J(\xi )`$ is determined only up to an arbitrary choice of a distinguished function. Indeed, though this Chapter has set aside (ever since (iii) of Section 1.2) time-dependent functions: if one considers them, then there is here an arbitrary choice of a time-dependent distinguished function.
### 6.3 Examples
We give two familiar examples; and then, as a glimpse of the general power of the theory, two abstract examples (which will not be needed later on).
(1): Total linear momentum of $`N`$ particles :—
In (3) at the end of Section 4.1.A, we showed that the left cotangent lift of the action of the translation group $`\mathrm{I}\mathrm{R}^3`$ on $`Q=\mathrm{I}\mathrm{R}^{3N}`$ to $`M=T^{}\mathrm{I}\mathrm{R}^{3N}`$, i.e. the left action corresponding to eq. 4.126, is
$$\mathrm{\Psi }_𝐱(𝐪_i,𝐩^i):=T^{}(\mathrm{\Phi }_𝐱)(𝐪_i,𝐩^i)=(𝐪_i+𝐱,𝐩^i),i=1,\mathrm{},N.$$
(6.341)
(Here we combine the discussions of examples (vi) and (ix) in Section 4.1.A.)
To find the momentum map, we: (a) compute the infinitesimal generator $`\xi _M`$ for an arbitrary element $`\xi `$ of $`𝔤=\mathrm{I}\mathrm{R}^3`$; and then (b) solve eq. 6.336, or in coordinates eq. 6.338.
(a): We differentiate eq. 6.341 with respect to $`𝐱`$ in the direction $`\xi `$, getting
$$\xi _M(𝐪_i,𝐩^i)=(\xi ,\mathrm{},\xi ,\mathrm{𝟎},\mathrm{},\mathrm{𝟎}).$$
(6.342)
(b): Any function $`J(\xi )`$ has Hamiltonian vector field
$$X_{J(\xi )}(𝐪_i,𝐩^i)=(\frac{J(\xi )}{𝐩^i},\frac{J(\xi )}{𝐪_i});$$
(6.343)
so that the desired $`J(\xi )`$ with $`X_{J(\xi )}=\xi _M`$ solves
$$\frac{J(\xi )}{𝐩^i}=\xi \mathrm{and}\frac{J(\xi )}{𝐪_i}=\mathrm{𝟎},\mathrm{\hspace{0.33em}1}iN.$$
(6.344)
Choosing constants so that $`J`$ is linear, the solution is
$$J(\xi )(𝐪_i,𝐩^i)=\left(\mathrm{\Sigma }_{i=1}^N𝐩^i\right)\xi ,\mathrm{i}.\mathrm{e}.𝐉(𝐪_i,𝐩^i)=\mathrm{\Sigma }_{i=1}^N𝐩^i;$$
(6.345)
i.e. the familiar total linear momentum.
(2): Angular momentum of a single particle :—
$`SO(3)`$ acts on $`Q=\mathrm{I}\mathrm{R}^3`$ by $`\mathrm{\Phi }_A(𝐪)=A𝐪`$. So the tangent (derivative) map is
$$T_𝐪\mathrm{\Phi }_A:(𝐪,𝐯)T\mathrm{I}\mathrm{R}_𝐪^3(A𝐪,A𝐯)T\mathrm{I}\mathrm{R}_{A𝐪}^3.$$
(6.346)
As we saw in example (vii) of Section 4.1.A, the left cotangent lift of the action to $`M=T^{}\mathrm{I}\mathrm{R}^3`$ (the lifted action “with $`g^1`$”, corresponding to eq. 4.126) is:
$$T_{A𝐪}^{}(\mathrm{\Phi }_{A^1})(𝐪,𝐩)=(A𝐪,A𝐩).$$
(6.347)
To find the momentum map, we proceed in two stages, (a) and (b), as in example (1).
(a): We differentiate eq. 6.347 with respect to $`A`$ in the direction $`\xi =\mathrm{\Theta }(\omega )𝔰𝔬(3)`$, where $`\omega \mathrm{I}\mathrm{R}^3`$ and $`\mathrm{\Theta }`$ is as in eq. 3.48 and 3.51. We get
$$\xi _M(𝐪,𝐩)=(\xi 𝐪,\xi 𝐩)=(\omega 𝐪,\omega 𝐯).$$
(6.348)
(b): So the desired $`J(\xi )`$ is the solution linear in $`\xi `$ to the Hamilton’s equations
$$\frac{J(\xi )}{𝐩}=\xi 𝐪\mathrm{and}\frac{J(\xi )}{𝐪}=\xi 𝐩.$$
(6.349)
So a solution is given by
$$J(\xi )(𝐪,𝐩)=(\xi 𝐪)𝐩=(\omega 𝐪)𝐩=(𝐪𝐩)\omega ,$$
(6.350)
so that
$$𝐉(𝐪,𝐩)=𝐪𝐩,$$
(6.351)
i.e. the familiar angular momentum.
(3): Dual of a Lie algebra homomorphism :—
We begin by stating a Lemma, which we will not prove; for details cf. Marsden and Ratiu (1999: 10.7.2, p. 372). Namely: let $`G,H`$ be Lie groups and let $`\alpha :𝔤𝔥`$ be a linear map between their Lie algebras. Then $`\alpha `$ is a Lie algebra homomorphism iff its dual $`\alpha ^{}:𝔥^{}𝔤^{}`$ is a (linear) Poisson map (where $`𝔥^{},𝔤^{}`$ are equipped with their natural Lie-Poisson brackets as in Section 5.2.4).
Now let $`G,H`$ be Lie groups, let $`A:HG`$ be a Lie group homomorphism, and let $`\alpha :𝔥𝔤`$ be the induced Lie algebra homomorphism; so that by the Lemma, $`\alpha ^{}:𝔤^{}𝔥^{}`$ is a Poisson map. We will prove that $`\alpha ^{}`$ is also a momentum map for the action of $`H`$ on $`𝔤^{}`$ given by, with $`hH,x𝔤^{}`$:
$$\mathrm{\Phi }(h,x)hx:=Ad_{A(h)^1}^{}x.$$
(6.352)
Proof: We first recall the adjoint and co-adjoint actions $`Ad_g:𝔤𝔤`$ and $`Ad_g^{}:𝔤^{}𝔤^{}`$; in particular, eq. 4.191. So the action in eq. 6.352 is:
$$x𝔤^{},\xi 𝔤:<hx;\xi >=<x;Ad_{A(h)^1}\xi >.$$
(6.353)
As usual, we compute for $`\eta 𝔥`$, the infinitesimal generator $`\eta _𝔤^{}`$ at $`x𝔤^{}`$ by differentiating eq. 6.353 with respect to $`h`$ at $`e`$ in the direction $`\eta 𝔥`$. We get (cf. eq. 4.198):
$$<\eta _𝔤^{}(x);\xi >=<x;ad_{\alpha (\eta )}\xi >=<ad_{\alpha (\eta )}^{}(x);\xi >.$$
(6.354)
We define $`𝐉(x):=\alpha ^{}(x)`$: that is,
$$J(\eta )(x)<𝐉(x);\eta >:=<\alpha ^{}(x);\eta ><x;\alpha (\eta )>;$$
(6.355)
which implies
$$_xJ(\eta )=\alpha (\eta ).$$
(6.356)
Now we recall that Hamilton’s equations for $`J(\eta )`$ as the Hamiltonian are (cf. eq. 5.264)
$$\dot{x}X_{J(\eta )}(x)=ad_{_xJ(\eta )}^{}(x).$$
(6.357)
Combining eq. 6.354 to eq. 6.357, we get:
$$X_{J(\eta )}(x)=ad_{\alpha (\eta )}^{}(x)=\eta _𝔤^{}(x);$$
(6.358)
proving that $`𝐉(x):=\alpha ^{}(x)`$ is a momentum map. QED.
(4): Momentum maps for subgroups :—
Assume that $`𝐉:M𝔤^{}`$ is a momentum map for a canonical left action of $`G`$ on $`M`$; and let $`H<G`$ be a subgroup of $`G`$. Then $`H`$ also acts canonically on $`M`$, and this action has as a momentum map the restriction of $`𝐉`$’s values to $`𝔥𝔤`$. That is: the map
$$𝐉_H:M𝔥^{}\mathrm{given}\mathrm{by}𝐉_H(x):=𝐉(x)_𝔥.$$
(6.359)
For the canonical action of $`G`$ ensures that if $`\eta 𝔥𝔤`$, then $`\eta _M=X_{J(\eta )}`$. Then $`J_H(\eta ):=J(\eta )\eta 𝔥`$ defines a momentum map for $`H`$’s action. That is
$$xM,\eta 𝔥:<𝐉_H(x);\eta >=<𝐉(x);\eta >.$$
(6.360)
### 6.4 Equivariance of momentum maps
In (1) of Section 4.2, we defined the general notion of an equivariant map $`f:MN`$ between manifolds as one that respects the actions of a group $`G`$ on $`M`$ and on $`N`$: eq. 4.144. We now develop an especially important case of this notion: the equivariance of momentum maps $`𝐉:M𝔤^{}`$, where the action on $`𝔤^{}`$ is the co-adjoint action, eq. 4.192.
For us, this notion will have two main significances:—
(i): many momentum maps that occur in examples are equivariant in this sense;
(ii): equivariance has various theoretical consequences: in particular, momentum maps for cotangent lifted actions are always equivariant (Section 6.5), and equivariance is crucial in theorems about reduction (Section 7).
In this Section, we will glimpse these points by:
(i): defining the notion, and remarking on a weakened differential version of the notion (Section 6.4.1);
(ii): proving that equivariant momentum maps are Poisson (Section 6.4.2).
#### 6.4.1 Equivariance and infinitesimal equivariance
Let $`\mathrm{\Phi }`$ be a canonical left action of $`G`$ on $`M`$, and let $`𝐉:M𝔤^{}`$ be a momentum map for it. We say $`𝐉`$ is equivariant if for all $`gG`$
$$𝐉\mathrm{\Phi }_g=Ad_{g^1}^{}𝐉;$$
(6.361)
cf. eq. 4.144 and the definition of co-adjoint action, eq. 4.193:
$$\begin{array}{c}M\\ \mathrm{\Phi }_g\\ M\end{array}\begin{array}{c}\stackrel{𝐉}{}\\ \\ \stackrel{𝐉}{}\end{array}\begin{array}{c}𝔤^{}\\ Ad_{g^1}^{}\\ 𝔤^{}\end{array}$$
(6.362)
An equivalent formulation arises by considering that we can add to the commutative square in eq. 6.362 the two commutative triangles:
$$M\stackrel{J(\xi )}{}\mathrm{I}\mathrm{R}\mathrm{is}M\stackrel{𝐉}{}𝔤^{}\stackrel{_\xi }{}\mathrm{I}\mathrm{R};$$
(6.363)
representing the fact that $`J(\xi )(x)=𝐉(x)(\xi )`$; and
$$𝔤^{}\stackrel{_\xi }{}\mathrm{I}\mathrm{R}\mathrm{is}𝔤^{}\stackrel{Ad_{g^1}^{}}{}𝔤^{}\stackrel{_{Ad_g\xi }}{}\mathrm{I}\mathrm{R};$$
(6.364)
representing the fact that for all $`\eta g^{}`$
$$<Ad_{g^1}^{}(\eta );Ad_g(\xi )>=<\eta ;Ad_{g^1}Ad_g(\xi )><\eta ;\xi >.$$
(6.365)
Eq.s 6.363 and 6.364 imply that an equivalent formulation of equivariance is that for all $`xM,gG`$ and $`\xi 𝔤`$ (and with $`gx\mathrm{\Phi }_g(x)`$)
$$𝐉(gx)(Ad_g\xi )J(Ad_g\xi )(gx)=J(\xi )(x)𝐉(x)(\xi ).$$
(6.366)
In (2) of Section 4.4, we differentiated the general notion of an equivariant map, and got the weaker differential notion that the infinitesimal generators $`\xi _M`$ and $`\xi _N`$ of the actions of $`G`$ on $`M`$ and on $`N`$ are $`f`$-related.
Here also we can differentiate equivariance, and get the notion of infinitesimal equivariance. But I will not go into details since:
(i): we will not need the notion, not least because (as mentioned above), many momentum maps are equivariant;
(i): under certain common conditions (e.g. the group $`G`$ is compact, or is connected) an infinitesimally equivariant momentum map can always be replaced by an equivariant one.
So let it suffice to say that infinitesimal equivariance is theoretically important. In particular, the result eq. 6.321, viz.
$$X_{J([\xi ,\eta ])}=X_{\{J(\xi ),J(\eta )\}_M}$$
(6.367)
implies that
$$\mathrm{\Sigma }(\xi ,\eta ):=J([\xi ,\eta ])\{J(\xi ),J(\eta )\}_M$$
(6.368)
is a distinguished function on the Poisson manifold $`M`$, and so constant on every symplectic leaf.
This makes it natural to ask when $`\mathrm{\Sigma }0`$. After all, cf. eq. 6.326. Both $`\xi \xi _M`$ and $`FX_F`$ are Lie algebra anti-homomorphisms. So it is natural to ask whether $`J`$ is a Lie algebra homomorphism, i.e. whether $`\mathrm{\Sigma }=0`$. And it turns out that infinitesimal equivariance is equivalent to $`\mathrm{\Sigma }=0`$.
#### 6.4.2 Equivariant momentum maps are Poisson
The following result is important, both as a general method of finding canonical maps between Poisson manifolds, and for the Lie-Poisson reduction theorem of Section 7.
> Equivariant momentum maps are Poisson Let $`𝐉:𝐌𝔤^{}`$ be an equivariant momentum map for a canonical left action of $`G`$ on a Poisson manifold $`M`$. Then $`𝐉`$ is a Poisson map: for all $`F_1,F_2(𝔤^{})`$,
>
> $$𝐉^{}\{F_1,F_2\}_𝔤^{}=\{𝐉^{}F_1,𝐉^{}F_2\}_M;\mathrm{i}.\mathrm{e}.\{F_1,F_2\}_𝔤^{}𝐉=\{F_1𝐉,F_2𝐉\}_M.$$
> (6.369)
Proof:— We will relate (i) the left hand side, then (ii) the right hand side of eq. 6.369 to $`J`$; and finally we will use the fact that the Poisson bracket on $`M`$ depends only on the values of the first derivatives.
(i): Let $`xM,\alpha =𝐉(x)𝔤^{}`$; and let $`\xi =F_1`$ and $`\eta =F_2`$ evaluated at $`\alpha `$, so that $`\xi ,\eta 𝔤^{}=𝔤`$. Then
$$\{F_1,F_2\}_𝔤^{}(𝐉(x))<\alpha ;[F_1,F_2]>=<\alpha ;[\xi ,\eta ]>=J([\xi ,\eta ])(x)=\{J(\xi ),J(\eta )\}(x);$$
(6.370)
where the third equation just applies the definition of $`𝐉`$, eq. 6.327, and the fourth equation uses (infinitesimal) equivariance.
(ii): We show that $`(F_1𝐉)(x)`$ and $`J(\xi )(x)`$ have equal $`x`$-derivatives. For any $`xM`$ and $`v_xT_xM`$
$$𝐝(F_1𝐉)(x)v_x=𝐝F_1(\alpha )T_x𝐉(v_x)=<T_x𝐉(v_x);F_1>=𝐝J(\xi )(x)v_x;$$
(6.371)
where the first equation uses the chain rule, and the last uses the definition of $`𝐉`$, eq. 6.327 and the fact that $`\xi =F_1`$.
Finally, since the Poisson bracket on $`M`$ depends only on the values of the first derivatives, we infer from eq. 6.371 that
$$\{F_1𝐉,F_2𝐉\}(x)=\{J(\xi ),J(\eta )\}(x).$$
(6.372)
Combining this with (i), the result follows. QED.
### 6.5 Momentum maps on cotangent bundles
Let a Lie group $`G`$ act on a manifold (“configuration space”) $`Q`$. We saw in Section 4.1.A that this action can be lifted to the cotangent bundle $`T^{}Q`$; cf. eq.s 4.121, 4.124 and 4.126. In this Section, we focus on momentum maps for such cotangent lift actions. We shall see that any such action has an equivariant momentum map, for which there is an explicit general formula. The general theory (Sections 6.5.1, 6.5.2) will need just one main new notion, the momentum function. We end with some examples (Section 6.5.3).
#### 6.5.1 Momentum functions
Given a manifold $`Q`$ and its vector fields $`𝒳(Q)`$, we define the map
$$𝒫:𝒳(Q)(T^{}Q)\mathrm{by}:(𝒫(X))(\alpha _q):=<\alpha _q;X(q)>$$
(6.373)
for $`qQ,X𝒳(Q)`$ and $`\alpha _qT_q^{}Q`$. Here, $`\alpha _q`$ is, strictly speaking, a point in the cotangent bundle above the base-point $`qQ`$: so $`\alpha _q`$ can be written as $`(q,\alpha )`$ with $`\alpha `$ a covector at $`q`$, i.e. $`\alpha T_q^{}Q`$. But as we mentioned just before defining cotangent lifts (eq. 4.121): it is harmless to (follow many presentations and) conflate a point in $`T^{}Q`$, i.e. a pair $`(q,\alpha ),qQ,\alpha T_q^{}Q`$, with its form $`\alpha `$, provided we keep track of the $`q`$ by writing the form as $`\alpha _q`$.
$`𝒫(X)`$, as defined by eq. 6.373, is called the momentum function of $`X`$. In coordinates, $`𝒫(X)`$ is given by
$$𝒫(X)(q^i,p_i)=X^j(q^i)p_j$$
(6.374)
where we sum on $`j=1,\mathrm{},n:=\mathrm{dim}Q`$. (So NB: This $`𝒫`$ is different from that in $`𝒫(M)`$, the infinitesimal Poisson automorphisms of $`M`$, discussed in Section 6.1.1.)
We also denote by $`(T^{}Q)`$ the space of smooth functions $`F:T^{}Q\mathrm{I}\mathrm{R}`$ that are linear on fibres of $`T^{}Q`$: i.e. writing the bundle points $`\alpha _q,\beta _qT_q^{}Q`$ as $`(q,\alpha )`$ and $`(q,\beta )`$, we have for $`\lambda ,\mu \mathrm{I}\mathrm{R}`$
$$F(q,(\lambda \alpha +\mu \beta ))=\lambda F((q,\alpha ))+\mu F((q,\beta )).$$
(6.375)
So functions $`F,H`$ that are in $`(T^{}Q)`$ can be written in coordinates as (summing on $`i=1,\mathrm{},n`$)
$$F(q,p)=X^i(q)p_i\mathrm{and}H(q,p)=Y^i(q)p_i$$
(6.376)
for functions $`X^i`$ and $`Y^i`$; and so any momentum function $`𝒫(X)`$ is in $`(T^{}Q)`$.
One readily checks that the standard Poisson bracket (from $`T^{}Q`$’s symplectic structure, Section 2.1.1) of such an $`F`$ and $`H`$ is also linear on the fibres of $`T^{}Q`$. In fact, eq. 6.376 implies
$$\{F,H\}(q,p):=\frac{F}{q^j}\frac{H}{p_j}\frac{H}{q^j}\frac{F}{p_j}=\left(\frac{X^i}{q^j}Y^j\frac{Y^i}{q^j}X^j\right).$$
(6.377)
So $`(T^{}Q)`$ is a Lie subalgebra of $`(T^{}Q)`$.
The next result summarizes how momentum functions relate $`𝒳(Q)`$ and Hamiltonian vector fields on $`T^{}Q`$ to $`(T^{}Q)`$.
> Three (anti)-isomorphic Lie algebras The two Lie algebras
> (i) $`(𝒳(Q),[,])`$ of vector fields on $`Q`$;
> (ii) Hamiltonian vector fields $`X_F`$ on $`T^{}Q`$ with $`F(T^{}Q)`$
> are isomorphic. And each is anti-isomorphic to
> (iii) ($`(T^{}Q),\{,\}`$).
> In particular, the map $`𝒫`$ is an anti-isomorphism from (i) to (iii), so that we have
>
> $$\{𝒫(X),𝒫(Y)\}_{T^{}Q}=𝒫([X,Y]).$$
> (6.378)
Proof: Since $`𝒫(X):T^{}Q\mathrm{I}\mathrm{R}`$ is linear on fibres, $`𝒫`$ maps $`𝒳(Q)`$ into $`(T^{}Q)`$. $`𝒫`$ is also onto $`(T^{}Q)`$: given $`F(T^{}Q)`$, we can define $`X(F)𝒳(Q)`$ by
$$<\alpha _q;X(F)(q)>:=F(\alpha _q)\alpha _qT_q^{}Q$$
(6.379)
so that $`𝒫(X(F))=F`$. $`𝒫`$ is linear and $`𝒫(X)=0`$ implies that $`X=0`$. Also, eq. 6.378 follows immediately by comparing eq. 6.377 with the Lie bracket of $`X,Y𝒳(Q)`$; cf. eq. 3.55. So $`𝒫`$ is an anti-isomorphism from $`(𝒳Q,[,])`$ to ($`(T^{}Q),\{,\}`$).
The map
$$F((T^{}Q),\{,\})X_F(\{X_FF(T^{}Q)\},[,])$$
(6.380)
is surjective by definition. It is a Lie algebra anti-homomorphism, by eq. 3.60 (i.e. result (2) in Section 3.2.2). And if $`X_F=0`$, then $`F`$ is constant on $`T^{}Q`$; and hence $`F0`$ since $`F`$ is linear on the fibres (cf. eq. 6.375). QED.
#### 6.5.2 Momentum maps for cotangent lifted actions
We begin this Subsection with a result relating the Hamiltonian flow on $`T^{}Q`$ induced by the momentum function $`𝒫(X)`$ to the Hamiltonian flow on $`X`$ induced by $`X`$. From this result, our main result—the guarantee of an equivariant momentum map for a cotangent lifted action, and an explicit formula for it—will follow directly.
> The Hamiltonian flow of a momentum function Let $`X𝒳(Q)`$ have flow $`\varphi _\tau `$ on $`Q`$; cf. Section 3.1.2.B. Then the flow of $`X_{𝒫(X)}`$ on $`T^{}Q`$ is $`T^{}\varphi _\tau `$. That is: the flow of $`X_{𝒫(X)}`$ is the cotangent lift (Section 4.1.A) of $`\varphi _\tau `$, as given by the diagram, with $`\pi _Q`$ the canonical projection:
>
> $$\begin{array}{c}Q\\ \pi _Q\\ T^{}Q\end{array}\begin{array}{c}\stackrel{\varphi _\tau }{}\\ \\ \stackrel{T^{}\varphi _\tau }{}\end{array}\begin{array}{c}Q\\ \pi _Q\\ T^{}Q\end{array}$$
> (6.381)
Proof: We differentiate the relation in eq. 6.381, i.e.
$$\pi _QT^{}\varphi _\tau =\varphi _\tau \pi _Q$$
(6.382)
at $`\tau =0`$ to get
$$T\pi _QY=X\pi _Q\mathrm{with}\alpha _qT_q^{}Q,Y(\alpha _q)=\frac{d}{d\tau }_{\tau =0}T^{}\varphi _\tau (\alpha _q);$$
(6.383)
i.e. $`T^{}\varphi _\tau `$ is the flow of $`Y`$.
Now we will show that $`Y=X_{𝒫(X)}`$, using eq. 6.383 and the geometrical formulation of Hamiltonian mechanics of Section 2.1, especially Cartan’s magic formula, eq. 2.20, applied to the canonical one-form $`\theta \theta _H`$ (defined by eq. 2.8 and 2.9).
We reported (at the start of (2) of Section 4.1.A) that the cotangent lift $`T^{}\varphi _\tau `$ preserves $`\theta \theta _H`$ on $`T^{}Q`$. So $`_Y\theta =0`$. Then the definition of $`\omega `$ as the negative exterior derivative of $`\theta `$, and Cartan’s magic formula, eq. 2.20, yields
$$𝐢_Y\omega =𝐢_Y𝐝\theta =\mathrm{𝐝𝐢}_Y\theta .$$
(6.384)
On the other hand, we also have
$$𝐢_Y\theta (\alpha _q)<\theta (\alpha _q);Y(\alpha _q)>=<\alpha _q;T\pi _Q(Y(\alpha _q))>=<\alpha _q;X(q)>=𝒫(X)(\alpha _q)$$
(6.385)
where the second equation applies the definition of the canonical one-form (eq. 2.8), the third applies eq. 6.383, and the fourth applies the definition eq. 6.373 of momentum functions.
Combining eq. 6.384 and 6.385, we have:
$$𝐢_Y\omega =𝐝𝒫(X)$$
(6.386)
which is Hamilton’s equations (eq. 2.15) telling us that $`Y=X_{𝒫(X)}`$. QED.
Accordingly the Hamiltonian vector field $`X_{𝒫(X)}`$ on $`T^{}Q`$ is called the cotangent lift of $`X𝒬`$ to $`T^{}Q`$. In local coordinates, we can write, by combining eq. LABEL:HamDelta0 and 6.374
$$X_{𝒫(X)}=\frac{𝒫(X)}{p_i}\frac{}{q^i}\frac{𝒫(X)}{q^i}\frac{}{p_i}=X^i\frac{}{q^i}\frac{X^i}{q^j}p_i\frac{}{p_j}.$$
(6.387)
Note in particular that, combining the usual sign-change between Lie algebras and Poisson brackets (eq. 3.60) with the sign-change for momentum functions (eq. 6.378), we have
$$[X_{𝒫(X)},X_{𝒫(Y)}]=X_{\{𝒫(X),𝒫(Y)\}}=X_{𝒫([X,Y])}=X_{𝒫([X,Y])}.$$
(6.388)
We can now readily prove our main result guaranteeing, and giving a formula for, equivariant momentum maps.
> Equivariant momentum maps Let $`G`$ act on the left on $`Q`$ and so by cotangent lift on $`T^{}Q`$. The cotangent lifted action has an equivariant momentum map $`𝐉:T^{}Q𝔤^{}`$ given by
>
> $$<𝐉(\alpha _q);\xi >=<\alpha _q;\xi _Q(q)>𝒫(\xi _Q)(\alpha _q).$$
> (6.389)
> In coordinates $`q^i,p_i`$ on $`T^{}Q`$ and $`\xi ^a`$ on $`𝔤`$, and with $`\xi _Q^i=\xi ^aA_a^i`$ the components of $`\xi _Q`$, this reads
>
> $$J_a\xi ^a=p_i\xi _Q^i=p_iA_a^i\xi ^a$$
> (6.390)
> so that $`J_a(q,p)=p_iA_a^i(q)`$.
Proof: The preceding result tells us that for any $`\xi 𝔤`$, the infinitesimal generator of the cotangent lifted action on $`T^{}Q`$ is $`\xi _{T^{}Q}X_{𝒫(\xi _Q)}`$. So a momentum map for this action is given by
$$J(\xi )=𝒫(\xi _Q).$$
(6.391)
This gives eq. 6.389, just by applying the definitions of the momentum map $`𝐉`$ (eq. 6.327) and of momentum function (eq. 6.373).
To prove equivariance, we argue as follows:
$`<𝐉(g\alpha _q);\xi >=<(g\alpha _q);\xi _Q(gq)>`$ (6.392)
$`=<\alpha _q;(T\mathrm{\Phi }_{g^1})\xi _Q(gq)><\alpha _q;(T_{gq}\mathrm{\Phi }_{g^1}\xi _Q\mathrm{\Phi }_g)(q).`$ (6.393)
$`=<\alpha _q;(\mathrm{\Phi }_g^{}\xi _Q)(q)>`$ (6.394)
$`=<\alpha _q;(Ad_{g^1}\xi )_Q(q)>`$ (6.395)
$`=<𝐉(\alpha _q);Ad_{g^1}\xi >=<Ad_{g^1}^{}(𝐉(\alpha _q));\xi >.`$ (6.396)
Here we have applied in succession: (i) eq. 6.389; (ii) the fact that $`g\alpha _q`$ is short for $`T^{}(\mathrm{\Phi }_{g^1})(\alpha _q)`$, cf. eq. 4.126 and 4.121; (iii) the definition of pullback, cf. eq. 4.172; (iv) result , eq. 4.167, of Section 4.5.1; (v) eq. 6.389 again; and finally, (vi) the fact that $`Ad^{}`$ is the adjoint of $`Ad`$, cf. eq. 4.191. QED.
#### 6.5.3 Examples
We discuss first our familiar examples, linear and angular momentum i.e. (1) and (2) from Section 6.3; and then the cotangent lift of left and right translations on $`G`$—an example motivated by Section 4.6’s description of kinematics on a Lie group $`G`$.
(1): Total linear momentum of $`N`$ particles:—
Since the translation group $`\mathrm{I}\mathrm{R}^3`$ acts on $`Q:=\mathrm{I}\mathrm{R}^{3N}`$ by $`\mathrm{\Phi }(𝐱,(𝐪_i))=(𝐪_i+𝐱)`$, the infinitesimal generator on $`Q`$ is
$$\xi _{\mathrm{I}\mathrm{R}^{3N}}(𝐪_i)=(\xi ,\mathrm{},\xi )(\xi N\mathrm{times})$$
(6.397)
Applying eq. 6.389, the equivariant momentum map is given by
$$J(\xi )(𝐪_i,𝐩^i)=\left(\mathrm{\Sigma }_{i=1}^N𝐩^i\right)\xi ,\mathrm{i}.\mathrm{e}.𝐉(𝐪_i,𝐩^i)=\mathrm{\Sigma }_{i=1}^N𝐩^i;$$
(6.398)
agreeing with our previous solution, eq. 6.345, based on the differential equation eq. 6.338.
(2): Angular momentum of a single particle:—
$`SO(3)`$ acts on $`\mathrm{I}\mathrm{R}^3`$ by $`\mathrm{\Phi }(A,𝐪)=A𝐪`$. Writing $`\xi 𝔰𝔬(3)`$ as $`\xi =\mathrm{\Theta }\omega `$ (cf. eq. 3.47, 3.51 and 3.105), the infinitesimal generator is
$$\xi _{\mathrm{I}\mathrm{R}^3}(𝐪)=\xi 𝐪=\omega 𝐪.$$
(6.399)
So applying eq. 6.389, the equivariant momentum map $`𝐉:T^{}\mathrm{I}\mathrm{R}^3𝔰𝔬(3)\mathrm{I}\mathrm{R}^3`$ is given by
$$<𝐉(𝐪,𝐩);\omega >=<𝐩;\omega 𝐪>=𝐩(\omega 𝐪)=\omega (𝐪𝐩),\mathrm{i}.\mathrm{e}.𝐉(𝐪,𝐩)=𝐪𝐩;$$
(6.400)
agreeing with our previous solution, eq. 6.351, based on the differential equation eq. 6.338.
(3): The cotangent lift of left and right translations on $`G`$:—
Recalling eq. 4.157, viz. that the infinitesimal generator of left translation is
$$\xi _G(g)=(T_eR_g)\xi ,$$
(6.401)
a right-invariant vector field, and applying eq. 6.389, we see that the momentum map $`𝐉_L:T^{}G𝔤^{}`$ for the cotangent lift of left translation is given by
$$<𝐉_L(\alpha _g);\xi >=<\alpha _g;\xi _G(g)>=<\alpha _g;(T_eR_g)\xi >=<(T_e^{}R_g)(\alpha _g);\xi >$$
(6.402)
where the last equation applies the definition of the cotangent lift eq. 4.121. That is: the equivariant momentum map is
$$𝐉_L(\alpha _g)=T_e^{}R_g(\alpha _g).$$
(6.403)
In words: the momentum map $`𝐉_L`$ of the cotangent lift of left translation is the cotangent lift of right translation.
In a similar way, we could consider right translation: $`R_g:hhg`$. Right translation defines a right action on $`G`$, has $`\xi _G(g)=(T_eL_g)\xi `$ as its infinitesimal generator, and so has
$$𝐉_R:T^{}G𝔤^{};𝐉_R(\alpha _g):=T_e^{}L_g(\alpha _g)$$
(6.404)
as the momentum map of its cotangent lift. Note that this momentum map is equivariant with respect to $`Ad_g^{}`$: which, as discussed after eq. 4.191, is a right action.
## 7 Reduction
### 7.1 Preamble
In this final Section, the themes of Section 2 onwards come together—at last! As announced in Section 5.1, we will concentrate on proving what is nowadays called the Lie-Poisson reduction theorem: that is, the isomorphism of Poisson manifolds
$$T^{}G/G𝔤^{}.$$
(7.405)
Here the quotient of $`T^{}G`$ is by the cotangent lift of $`G`$’s action on itself by left translation.
As it happens, this Chapter’s main sources (i.e. Abraham and Marsden (1978), Arnold (1989), Olver (2000) and Marsden and Ratiu (1999)) do not contain what is surely the most direct proof of this result. So we give it in Section 7.2. The result will follow directly from four previous main results, one from Section 5 and three from Section 6.
‘Directly’, but for one wrinkle! This relates to “flipping” between left and right translation, and their various lifts. In short: the four previous results show that $`T^{}G/G`$ is isomorphic as a Poisson manifold, not to $`𝔤^{}`$ with the Lie-Poisson bracket familiar since eq. 5.258 and 5.261, but instead to $`𝔤^{}`$ equipped with this bracket’s negative, i.e. equipped with
$$\{F,H\}_{}(x):=<x;[F(x),H(x)]>,x𝔤^{}.$$
(7.406)
But we shall (mercifully!) not reproduce, with minus signs appropriately added, our entire discussion of the Lie-Poisson bracket that ensued after eq. 5.258; (exercise for the reader!).
To avoid ambiguity, we shall sometimes write $`𝔤_+^{}`$ for $`𝔤^{}`$ equipped with the positive Lie-Poisson bracket of eq. 5.261; and $`𝔤_{}^{}`$ for $`𝔤^{}`$ equipped with the negative Lie-Poisson bracket of eq. 7.406.
In fact, it will be clearest from now on, to treat right actions on a par with left actions; despite our previous emphasis on the latter. This will mean that we will also treat right-invariant vector fields (and another notion of right-invariance defined in Section 7.3.1) on a par with left-invariant vector fields (and Section 7.3.1’s corresponding new notion of left-invariance). Indeed, we have already glimpsed this would be necessary in:
(i): Section 4.4’s result that the infinitesimal generator of left translation is a right-invariant vector field, and vice versa (eq. 4.157, 4.158); and its corollaries in Example (3) of Section 6.5.3, that
(ii): the momentum map $`𝐉_L`$ of the cotangent lift of left translation is the cotangent lift of right translation; (eq. 6.403); and
(iii): the momentum map $`𝐉_R`$ of the cotangent lift of right translation is the cotangent lift of left translation; (eq. 6.404).
So by the end of Section 7.2, we will have a short proof of the Lie-Poisson reduction theorem. But (as often happens), the most direct proof does not give very much information about the situation. So in Section 7.3 we give more information (following Marsden and Ratiu (1999)). Then in Section 7.4, we discuss the reduction of dynamics (as against Poisson structure) from $`T^{}G`$ to $`𝔤^{}`$.
Finally, in Section 7.5 we state another reduction theorem, which is cast in terms of symplectic, not Poisson, manifolds—but which uses several notions from Section 3, such as free and proper actions, and isotropy groups. But we do not prove this theorem: we include it mostly in order to emphasize our previous remark, that (despite its length!) this Chapter just scratches the surface of the subject. We also discuss the relation between it and the Lie-Poisson reduction theorem.
### 7.2 The Lie-Poisson Reduction Theorem
First we recall from the end of Section 4.6.2 (eq. 4.227) that $`\overline{}\lambda :T^{}GG\times 𝔤^{}`$ is an equivariant map relating the cotangent lifted left action of left translation on $`T^{}G`$ to the $`G`$-action on $`G\times 𝔤^{}`$ given just by left translation on the first component. So we passed to the quotients, and defined $`\widehat{\overline{}\lambda }:T^{}G/G(G\times 𝔤^{})/G`$ by eq. 4.231, viz.
$`\widehat{\overline{}\lambda }:\mathrm{Orb}(\alpha )\{\beta T^{}G\beta =T^{}L_{h^1}(\alpha ),\mathrm{some}hG\}`$ (7.407)
$`\mathrm{Orb}(\overline{}\lambda (\alpha ))\{(hg,(T_e^{}L_g)(\alpha ))\mathrm{some}hG\}\{(h,(T_e^{}L_g)\alpha )\mathrm{some}hG\}.`$ (7.408)
where $`\alpha T_g^{}G`$, so that $`T^{}L_{h^1}\alpha T_{hg}^{}G`$. Finally, we identified $`(G\times 𝔤^{})/G`$ with $`𝔤^{}`$, so that the diffeomorphism $`\widehat{\overline{}\lambda }`$ maps $`T^{}G/G`$ to $`𝔤^{}`$, as in eq. 4.232:
$$\widehat{\overline{}\lambda }:\mathrm{Orb}(\alpha )\{\beta T^{}G\beta =T^{}L_{h^1}(\alpha ),\mathrm{some}hG\}T^{}G/G(T_e^{}L_g)(\alpha )𝔤^{}.$$
(7.409)
So now, we are to show that the diffeomorphism $`\widehat{\overline{}\lambda }:T^{}G/G𝔤^{}`$ is a Poisson map, in the sense of eq. 5.286 (Section 5.3.2). So we need to show:
(i): $`T^{}G/G`$ is a Poisson manifold;
(ii): $`\widehat{\overline{}\lambda }`$ maps (i)’s Poisson structure on $`T^{}G/G`$ to that of $`𝔤^{}`$. In fact, as announced in Section 7.1, $`\widehat{\overline{}\lambda }`$ maps on to the Poisson structure of $`𝔤_{}^{}`$, i.e. as given by eq. 7.406.
Prima facie, there could be a judicious choice to be made about (i), i.e. about how to define the Poisson structure on $`T^{}G/G`$, so as to secure (ii), i.e. so that $`\widehat{\overline{}\lambda }`$ respects the Poisson structure. But in fact our previous work gives a pre-eminently obvious choice—which works. Namely: we use the Poisson structure induced on $`T^{}G/G`$ by the Poisson reduction theorem of Section 5.5. The result follows directly by combining with this theorem, three results from Section 6:
(i): that equivariant momentum maps are Poisson; eq. 6.369 in Section 6.4.2;
(ii): that a cotangent lifted left action has an equivariant momentum map; eq. 6.389 in Section 6.5.2;
(iii): that the momentum maps of the cotangent lifts of left and right translation on $`G`$ are $`𝐉_L=T_e^{}R_g`$ and $`𝐉_R=T_e^{}L_g`$; eq. 6.403 and 6.404 in Section 6.5.3.
In particular, combining (i)-(iii): one deduces (exercise!) that $`𝐉_R=T_e^{}L_g`$ is equivariant with respect to $`Ad_g^{}`$, and so Poisson with respect to the negative Lie-Poisson bracket (eq. 7.406’s bracket) on $`𝔤^{}`$. That is: it is Poisson with the codomain $`𝔤_{}^{}`$.
Thus we have the
> Lie-Poisson reduction theorem The diffeomorphism $`\widehat{\overline{}\lambda }:T^{}G/G𝔤^{}`$:
>
> $$\widehat{\overline{}\lambda }:\mathrm{Orb}(\alpha )\{\beta T^{}G\beta =T^{}L_{h^1}(\alpha ),\mathrm{some}hG\}T^{}G/G(T_e^{}L_g)(\alpha )𝔤^{}$$
> (7.410)
> is Poisson.
Proof: First, eq. 7.410 means we have a commutative triangle. For with $`\pi :T^{}GT^{}G/G`$ the canonical projection, the momentum map $`𝐉_R:T^{}G𝔤^{},\alpha _g(T_e^{}L_g)\alpha _g`$ is equal to $`\widehat{\overline{}\lambda }\pi `$:
$$T^{}G\stackrel{\pi }{}T^{}G/G\stackrel{\widehat{\overline{}\lambda }}{}𝔤^{}.$$
(7.411)
Since left translation is a diffeomorphism of $`G`$, and the cotangent lift of any diffeomorphism of a manifold to its cotangent bundle is symplectic (cf. after eq. 4.120 in Section 4.1.A), the Poisson reduction theorem of Section 5.5 applies. That is, there is a unique Poisson structure on $`T^{}G/G`$ such that $`\pi `$ is Poisson. We also know from eq. 6.389, 6.369 and 6.404 that $`𝐉_R=T_e^{}L_g`$ is Poisson with respect to eq. 7.406’s bracket on $`𝔤^{}`$.
We can now deduce that $`\widehat{\overline{}\lambda }`$ is Poisson, i.e. that for all $`xT^{}G/G`$ and all $`F,H(𝔤_{}^{})`$
$$(\{F,H\}_𝔤_{}^{}\widehat{\overline{}\lambda })(x)=\{F\widehat{\overline{}\lambda },H\widehat{\overline{}\lambda }\}_{T^{}G/G}(x).$$
(7.412)
We just use (in order) the facts that:
(i): $`\pi `$ is surjective, so that for all $`xT^{}G/G`$ there is an $`\alpha _gT^{}G`$ with $`x=\pi (\alpha _g)\mathrm{Orb}(\alpha _g)`$;
(ii): $`𝐉_R=\widehat{\overline{}\lambda }\pi `$;
(iii): $`𝐉_R`$ is Poisson; and
(iv): $`\pi `$ is Poisson:
$`(\{F,H\}_𝔤_{}^{}\widehat{\overline{}\lambda })(x)=\{F,H\}_𝔤_{}^{}(\widehat{\overline{}\lambda }\pi )(\alpha _g)`$ (7.413)
$`=\{F,H\}_𝔤_{}^{}𝐉_R(\alpha _g)=\{F𝐉_R,H𝐉_R\}_{T^{}G}(\alpha _g)`$ (7.414)
$`=\{F\widehat{\overline{}\lambda },H\widehat{\overline{}\lambda }\}_{T^{}G/G}(\pi (\alpha _g))\{F\widehat{\overline{}\lambda },H\widehat{\overline{}\lambda }\}_{T^{}G/G}(x).\mathrm{QED}.`$ (7.415)
### 7.3 Meshing with the symplectic structure on $`T^{}G`$: invariant functions
We turn to giving more information about the situation described by the Lie-Poisson reduction theorem. The general idea will be that the Lie-Poisson bracket on $`𝔤^{}`$ meshes with the canonical symplectic structure on $`T^{}G`$. This will be made precise in two ways: the first is discussed in the first two Subsections, the second is discussed in the third Subsection.
The first discussion will have three stages:
(i): we show that scalars on $`𝔤^{}`$, $`F(𝔤^{})`$, are in one-one correspondence with scalars on $`T^{}G`$ that are constant on the orbits of the cotangent lift of left translation, which will be called left-invariant functions; and similarly, for the cotangent lift of right translation (a correspondence with right-invariant functions);
(ii): we take the usual canonical Poisson bracket in $`T^{}G`$ of these left-invariant or right-invariant scalars; and restrict this bracket to $`𝔤^{}`$ regarded as the cotangent space $`T_e^{}G`$ at the identity $`eG`$; and then
(iii): we show that this restriction is the Lie-Poisson bracket on $`𝔤^{}`$: the familiar positive one for right-invariant functions, and the new negative one of eq. 7.406 for the left-invariant functions.
We do stages (i) and (ii) in Section 7.3.1. These stages will not involve the choice between the positive and negative Lie-Poisson brackets. But stage (iii), in Section 7.3.2, will involve this choice. It will be a one-liner corollary of Section 6.4.2’s result that equivariant momentum maps are Poisson maps, eq. 6.369; (unsurprisingly, in that we also used this result in Section 7.2’s proof of the reduction theorem).
In the third Subsection, we use invariant functions to show a different sense in which the Lie-Poisson bracket on $`𝔤^{}`$ meshes with the symplectic structure on $`T^{}G`$. Namely, we derive the Lie-Poisson bracket on $`𝔤^{}`$ from the Poisson reduction theorem of Section 5.5, by using the ideas of invariant functions and momentum functions.
#### 7.3.1 Left-invariant and right-invariant functions on $`T^{}G`$
We say that a function $`F:T^{}G\mathrm{I}\mathrm{R}`$ is left-invariant if for all $`gG`$, and all $`\alpha _gT_g^{}G`$
$$(FT^{}L_g)(\alpha _g)=F(\alpha _g)$$
(7.416)
where $`T^{}L_g`$ is the cotangent lift of $`L_g:GG`$. Similarly, $`F:T^{}G\mathrm{I}\mathrm{R}`$ is called right-invariant if for all $`gG`$
$$(FT^{}R_g)=F.$$
(7.417)
So if $`F:T^{}G\mathrm{I}\mathrm{R}`$ is left-invariant or right-invariant, it is determined by its values for arguments in $`T_e^{}G=𝔤^{}`$.
Since any $`\alpha 𝔤^{}`$ is mapped by $`T^{}L_{g^1}(T^{}L_g)^1`$ to an element of $`T_g^{}G`$, a function is left-invariant iff it is constant on the orbits of the various $`T^{}L_g`$ for $`gG`$, i.e. constant on the orbits of the cotangent lift of left translation. Similarly, a function is right-invariant iff it is constant on the orbits of the cotangent lift of right translation.
So left-invariant functions induce well-defined functions on the quotient space $`T^{}G/G`$; and so, by Section 7.2, on its diffeomorphic (indeed Poisson manifold) copy $`𝔤^{}`$. Similarly for right-invariant functions.
But let us for the moment consider the smooth left-invariant (or right-invariant) functions on $`T^{}G`$, rather than the induced maps on the quotient space. We will denote the space of all smooth left-invariant functions on $`T^{}G`$ by $`_L(T^{}G)`$, and similarly the space of smooth right-invariant functions by $`_R(T^{}G)`$.
Recalling (from the discussion after eq. 4.120) that cotangent lifts are symplectic maps, i.e. $`T^{}L_g`$ and $`T^{}R_g`$ are symplectic maps on $`T^{}G`$, it follows immmediately that $`_L(T^{}G)`$ and $`_R(T^{}G)`$ are each closed under the canonical Poisson bracket on $`T^{}G`$. So they are each a Lie algebra with this bracket.
Now we can use the momentum maps $`𝐉_L`$ and $`𝐉_R`$ of Example (3) of Section 6.5.3 to extend any scalar $`F:𝔤^{}\mathrm{I}\mathrm{R}`$, i.e. $`F(𝔤^{})`$, to a left-invariant, or right-invariant, scalar on $`T^{}G`$.
Thus, given $`F:𝔤^{}\mathrm{I}\mathrm{R}`$ and $`\alpha _gT_g^{}G`$, we define $`F_L_L(T^{}G)`$ by
$$F_L(\alpha _g):=(F𝐉_R)(\alpha _g)(FT_e^{}L_g)(\alpha _g).$$
(7.418)
So $`F_L`$ is by construction left-invariant, and is called the left-invariant extension of $`F`$ from $`𝔤^{}`$ to $`T^{}G`$.
One similarly defines the right-invariant extension $`F_R_R(T^{}G)`$ of any $`F(𝔤^{})`$ by
$$F_R(\alpha _g):=(F𝐉_L)(\alpha _g)(FT_e^{}R_g)(\alpha _g).$$
(7.419)
Then the maps
$$F(𝔤^{})F_L_L(T^{}G)\mathrm{and}F(𝔤^{})F_R_R(T^{}G)$$
(7.420)
are vector space isomorphisms (exercise for the reader!) whose inverse is just restriction to the fiber $`T_e^{}G=𝔤^{}`$.
This completes what we called ‘stages (i) and (ii)’: describing a correspondence between scalars on $`𝔤^{}`$ and scalars on $`T^{}G`$ that are constant on the orbits of the cotangent lifts of left and right translation; and considering the canonical Poisson bracket (on $`T^{}G`$) of these scalars, i.e. the Lie algebras $`_L(T^{}G)`$ and $`_R(T^{}G)`$.
#### 7.3.2 Recovering the Lie-Poisson bracket
We now do stage (iii): we show that the restriction of the canonical Poisson bracket on $`T^{}G`$ of the right/left invariant functions, to $`𝔤^{}`$ regarded as the cotangent space $`T_e^{}G`$ at the identity $`eG`$, is the positive/negative Lie-Poisson bracket.
Since the inverses of the maps eq. 7.420 are just restriction to the fiber $`T_e^{}G=𝔤^{}`$, it suffices to show that the maps eq. 7.420 are Lie algebra isomorphisms. More precisely:
> Recovery of the Lie-Poisson bracket Using the positive Lie-Poisson bracket on $`𝔤^{}`$ (we write $`𝔤_+^{}`$): $`FF_R`$ is a Lie algebra isomorphism.
> Similarly: using the negative Lie-Poisson bracket on $`𝔤^{}`$ (we write $`𝔤_{}^{}`$):
> $`FF_L`$ is a Lie algebra isomorphism.
>
> That is: for all $`F,H(𝔤^{})`$
>
> $$\{F,H\}_+=\{F_R,H_R\}_{T^{}G}_𝔤^{};\{F,H\}_{}=\{F_L,H_L\}_{T^{}G}_𝔤^{}$$
> (7.421)
Proof: Consider $`𝐉_L:T^{}G𝔤^{}𝔤_+^{}`$, $`𝐉_L=T_e^{}R_g`$. $`𝐉_L`$ is an equivariant momentum map. So, by the result eq. 6.369 of Section 6.4.2, it is Poisson. That is:
$$\{F,H\}_+𝐉_L=\{F𝐉_L,H𝐉_L\}_{T^{}G}=\{F_R,H_R\}_{T^{}G}.$$
(7.422)
Restricting eq. 7.422 to $`𝔤^{}`$ gives the first equation of eq. 7.421.
Similarly, one proves the second equation by using the fact that $`𝐉_R:T^{}G𝔤^{}𝔤_{}^{},𝐉_R=T_e^{}L_g`$ is an equivariant momentum map and so is Poisson. That is:
$$\{F,H\}_{}𝐉_R=\{F𝐉_R,H𝐉_R\}_{T^{}G}=\{F_L,H_L\}_{T^{}G}.$$
(7.423)
We then restrict eq. 7.423 to $`𝔤^{}`$. QED.
#### 7.3.3 Deriving the Lie-Poisson bracket
Our discussion so far, in both Section 7.2 and the two previous Subsections, has taken the Lie-Poisson bracket (whether positive or negative) as given. We now show, using invariant functions and Section 6.5.1’s idea of momentum functions, how to derive the Lie-Poisson bracket on $`𝔤^{}`$.
So this derivation will amount to another, more “constructive”, proof of the Lie-Poisson reduction theorem. As in Section 7.2’s proof, two main ingredients will be:
(a): the diffeomorphism $`\widehat{\overline{}\lambda }`$ between $`T^{}G/G`$ and $`𝔤^{}`$ (eq. 4.232 or 7.409 or 7.410), and
(b): the Poisson reduction theorem of Section 5.5, applied to $`G`$’s action on $`T^{}G`$.
But instead of Section 7.2’s proof’s using the facts that (i) the momentum maps $`𝐉_RT_e^{}L_g`$ and $`𝐉_LT_e^{}R_g`$ are equivariant and (ii) equivariant momentum maps are Poisson, we will now use the ideas of invariant functions and momentum functions.
We begin by recalling that (since left translation is a diffeomorphism of $`G`$, and the cotangent lift of any diffeomorphism of a manifold to its cotangent bundle is symplectic), the Poisson reduction theorem implies that there is a unique Poisson structure on $`T^{}G/G`$ such that $`\pi :T^{}GT^{}G/G`$ is Poisson. We now use the diffeomorphism $`\widehat{\overline{}\lambda }:T^{}G/G𝔤^{}`$ to transfer this Poisson structure to $`𝔤^{}`$. Let us call the result $`\{,\}_{}`$. Though this is not to be read (yet!) as the negative Lie-Poisson bracket, our aim now is to calculate that it is in fact this bracket.
Notice first that since the momentum map $`𝐉_R:T^{}G𝔤^{},\alpha _g(T_e^{}L_g)\alpha _g`$ is equal to $`\widehat{\overline{}\lambda }\pi `$ (eq. 7.411), we know that $`𝐉_R`$ is Poisson with respect to this induced bracket on $`𝔤^{}`$. That is
$$\{F,H\}_{}𝐉_R(\alpha _g)=\{F𝐉_R,H𝐉_R\}_{T^{}G}(\alpha _g)=\{F_L,H_L\}_{T^{}G}(\alpha _g).$$
(7.424)
To calculate the right hand side, we will apply the ideas of invariant functions and momentum functions to each argument of the bracket; in particular to the first:
$$F_L(\alpha _g)=F(T_e^{}L_g\alpha _g).$$
(7.425)
We observe that since a Poisson bracket depends only on the values of first derivatives, we can replace $`F(𝔤^{})`$ by its linearization. That is, we can assume $`F`$ is linear, so that at any point $`\alpha 𝔤^{}`$, $`F(\alpha )=<\alpha ;F>`$, where $`F`$ is a constant in $`𝔤𝔤^{}`$. Applying this, and the definition of a momentum function eq. 6.373, to eq. 7.425, we get:
$$F(T_e^{}L_g\alpha _g)=<T_e^{}L_g\alpha _g;F>=<\alpha _g;T_eL_gF>=𝒫(X_F)(\alpha _g),$$
(7.426)
where the last equation applies the definition of a momentum function to the left-invariant vector field on $`G`$, $`X_\xi (g)T_eL_g(\xi )`$, for the case $`\xi =F`$.
Now we apply to eq. 7.426, in order: eq. 6.378, the definition of the Lie algebra bracket (cf. eq. 3.74), eq. 6.373 again, and the definition of left-invariant vector fields. We get:
$`\{F_L,H_L\}_{T^{}G}(\alpha _g)=\{𝒫(X_F),𝒫(X_H)\}_{T^{}G}(\alpha _g)=𝒫([X_F,X_H])(\alpha _g)`$ (7.427)
$`=𝒫(X_{[F,H]})(\alpha _g)=<\alpha _g;X_{[F,H]}>`$ (7.428)
$`=<\alpha _g;T_eL_g([F,H])>=<T_e^{}L_g(\alpha _g);[F,H]>.`$ (7.429)
Combining eq. 7.424 and eq. 7.429, and writing $`\alpha 𝔤^{}`$ for $`(T_e^{}L_g)\alpha _g𝐉_R(\alpha _g)`$, we have our result:
$$\{F,H\}_{}(\alpha )=<\alpha ;[F,H]>.$$
(7.430)
One similarly derives the positive Lie-Poisson bracket by considering right-invariant extensions of linear functions. The minus sign coming from eq. 6.378 is cancelled by the sign reversal in the Lie bracket of right-invariant vector fields. That is, it is cancelled by a minus sign coming from eq. 3.86.
### 7.4 Reduction of dynamics
We end our account of the Lie-Poisson reduction theorem by discussing the reduction of dynamics from $`T^{}G`$ to $`𝔤^{}`$.
We can be brief since we have already stated the main idea, when discussing the Poisson reduction theorem; cf. (2)(A) in Section 5.5. Thus recall that (under the conditions of the theorem) a $`G`$-invariant Hamiltonian function on a Poisson manifold $`M`$, $`H:M\mathrm{I}\mathrm{R}`$, defines a corresponding function $`h`$ on $`M/G`$ by $`H=h\pi `$, where $`\pi `$ is the projection $`\pi :MM/G`$; and since $`\pi `$ is Poisson, and so pushes Hamiltonian flows forward to Hamiltonian flows, $`\pi `$ pushes $`X_H`$ on $`M`$ to $`X_h`$ on $`M/G`$:
$$T\pi X_H=X_h\pi .$$
(7.431)
Applying this, in particular eq. 7.431, to the Lie-Poisson reduction theorem, we get
> Reduction of dynamics Let $`H:T^{}G\mathrm{I}\mathrm{R}`$ be left-invariant. That is: the function $`H^{}:=H_𝔤^{}`$ on $`𝔤^{}`$ satisfies
>
> $$H(\alpha _g)=H^{}(𝐉_R(\alpha _g))H^{}(T_e^{}L_g\alpha _g),\alpha _gT_g^{}G.$$
> (7.432)
> Then $`𝐉_R`$ pushes $`X_H`$ forward to $`X_H^{}`$. Or in terms of the flows $`\varphi (t)`$ and $`\varphi ^{}(t)`$ of $`X_H`$ and $`X_H^{}`$ respectively:
>
> $$𝐉_R(\varphi (t)(\alpha _g))=\varphi ^{}(t)(𝐉_R(\alpha _g)).$$
> (7.433)
> Similar statements hold for a right-invariant function $`H:T^{}G\mathrm{I}\mathrm{R}`$, its restriction $`H^+:=H_𝔤^{}`$ and $`𝐉_LT_e^{}R_g`$.
Besides, we already know the vector field of $`H^{}`$ on $`𝔤^{}`$. For eq. 5.264 in (3) of Section 5.2.4 gave a basis-independent expression of Hamilton’s equations on $`𝔤^{}`$ in terms of $`ad^{}`$. We just need to note that since we are now using the negative Lie-Poisson bracket on $`𝔤^{}`$, all terms in the deduction (eq. 5.263) apart from the left hand side, get a minus sign. So writing $`\alpha 𝔤^{}`$, eq. 5.264 for the vector field $`X_H^{}`$ becomes:
$$\frac{d\alpha }{dt}=ad_{H^{}(\alpha )}^{}(\alpha ).$$
(7.434)
On the other hand, we can go in the other direction, reconstructing the dynamics on $`T^{}G`$ from eq. 7.434 on $`𝔤^{}`$. The statement of the main result, below, is intuitive, in that the “reconstruction equation” for $`g(t)T^{}G`$ is
$$g^1\dot{g}=H^{}.$$
(7.435)
This is intuitive since it returns us to the basic idea of mechanics on $`𝔤`$ and $`𝔤^{}`$, viz. that the map
$$\lambda _g:\dot{g}T_gG\lambda _g(\dot{g}):=(T_gL_{g^1})\dot{g}𝔤$$
(7.436)
maps the generalized velocity to its body representation; cf. eq. 4.205. However, the proof of this result is involved (Marsden and Ratiu (1999: theorems 13.4.3, 13.4.4, p. 423-426); so we only state the result. It is:—
> Reconstruction of dynamics Suppose given a Lie group $`G`$, a left-invariant $`H:T^{}G\mathrm{I}\mathrm{R}`$, its restriction $`H^{}:=H_𝔤^{}`$, and an integral curve $`\alpha (t)`$ of the Lie-Poisson Hamilton’s equations eq. 7.434 on $`𝔤^{}`$, with the initial condition $`\alpha (0)=T_e^{}L_{g_0}(\alpha _{g_0})`$. Then the integral curve in $`T^{}G`$ of $`X_H`$ is given by
>
> $$T_{g(t)}^{}L_{g(t)^1}(\alpha (t));$$
> (7.437)
> where $`g(t)`$ is the solution of the reconstruction equation
>
> $$g^1\dot{g}=H^{}$$
> (7.438)
> with initial condition $`g(0)=g_0`$.
### 7.5 Envoi: the Marsden-Weinstein-Meyer theorem
I emphasize that our discussion of reduction has only scratched the surface: after all this Section has been relatively short! But now that the reader is armed with the long and leisurely exposition from Section 3 onwards, they are well placed to pursue the topic of reduction; e.g. through this Chapter’s main sources, Abraham and Marsden (1978), Arnold (1989), Olver (2000) and Marsden and Ratiu (1999).
In particular, the reader can now relate the Lie-Poisson reduction theorem to another main theorem about symplectic reduction, usually called the Marsden-Weinstein-Meyer or Marsden-Weinstein theorem (after these authors’ papers in 1973 and 1974).
This theorem concerns a symplectic action of a Lie group $`G`$ on a symplectic manifold $`(M,\omega )`$. For the sake of completeness, and to orient the reader to Landsman’s discussion of this theorem (this vol., ch. 5, especially Section 4.5), it is worth stating it (as usual, for the finite-dimensional case only), together with the lemma used to prove it, and the ensuing reduction of dynamics. These statements will also round off our discussion by illustrating how some notions expounded from Section 3 onwards, but not used in this Section, are nevertheless useful—e.g. in stating the hypotheses of this theorem.
So suppose the Lie group $`G`$ acts symplectically (eq. 6.315) on the symplectic manifold $`(M,\omega )`$; and that $`𝐉:M𝔤^{}`$ is an $`Ad^{}`$-equivariant momentum map for this action (eq. 6.361 and 6.366). Assume also that $`\alpha 𝔤^{}`$ is a regular value of $`𝐉`$, i.e. that at every point $`x𝐉^1(\alpha )`$, $`T_x𝐉`$ is surjective. So the submersion theorem of (1) of Section 3.3.1 applies; in particular, $`𝐉^1(\alpha )`$ is a sub-manifold of $`M`$ with dimension dim($`M`$) - dim($`𝔤^{}`$) $``$ dim($`M`$) - dim($`G`$).
Let $`G_\alpha `$ be the isotropy group (eq. 4.148) of $`\alpha `$ under the co-adjoint action, i.e.
$$G_\alpha :=\{gGAd_{g^1}^{}\alpha =\alpha \}.$$
(7.439)
So since $`𝐉`$ is $`Ad^{}`$-equivariant under $`G_\alpha `$, the quotient space $`M_\alpha :=𝐉^1(\alpha )/G_\alpha `$ is well-defined.
Now assume that $`G_\alpha `$ acts freely and properly on $`𝐉^1(\alpha )`$, so that (Section 4.3.B) the quotient space $`M_\alpha =𝐉^1(\alpha )/G_\alpha `$ is a manifold. $`M_\alpha `$ is the reduced phase space (corresponding to the momentum value $`\alpha `$).
Now we assert:
> Marsden-Weinstein-Meyer theorem $`M_\alpha `$ has a natural symplectic form $`\omega _\alpha `$ induced from $`(M,\omega )`$ as follows. Let $`u,v`$ be two vectors tangent to $`M_\alpha `$ at some point $`pM_\alpha `$: so $`p`$ is an orbit of $`G_\alpha `$’s action on $`𝐉^1(\alpha )`$, and $`u,vT_pM_\alpha `$. Then $`u`$ and $`v`$ are obtained, respectively, from some vectors $`u^{}`$ and $`v^{}`$ tangent to $`𝐉^1(\alpha )`$ at some point $`x𝐉^1(\alpha )`$ of the orbit $`p`$, by the projection $`\pi _\alpha :𝐉^1(\alpha )M_\alpha `$. That is:
>
> $$T\pi _\alpha (u^{})=u;T\pi _\alpha (v^{})=v.$$
> (7.440)
> It turns out that the value assigned by $`M`$’s symplectic form $`\omega `$ is the same whatever choice of $`x,u^{},v^{}`$ is made. So we define the symplectic form $`\omega _\alpha `$ on $`M_\alpha `$ as assigning this value. In other words: writing $`\pi _\alpha `$ for the projection, $`i_\alpha :𝐉^1(\alpha )M`$ for the inclusion, and for pullback:
>
> $$\pi _\alpha ^{}\omega _\alpha =i_\alpha ^{}\omega .$$
> (7.441)
The proof of this theorem uses the following Lemma. Let us write $`Gx`$ for the orbit Orb($`x`$) of $`x`$ under the action of all of $`G`$, and similarly $`G_\alpha x`$ for the orbit under $`G_\alpha `$, i.e. $`\{\mathrm{\Phi }(g,x)gG_\alpha \}`$. Then the Lemma states:
> For any $`x𝐉^1(\alpha )`$:—
> (i): $`T_x(G_\alpha x)=T_x(Gx)T_x(𝐉^1(\alpha ))`$; and
> (ii): $`T_x(Gx)`$ and $`T_x(𝐉^1(\alpha ))`$ are $`\omega `$-orthogonal complements of one another in $`TM`$. That is: for all $`u^{}T_xM`$:
> $`u^{}T_x(𝐉^1(\alpha ))`$ iff $`\omega (u^{},v^{})=0`$ for all $`v^{}T_x(Gx)`$.
Both the Lemma and the theorem are each proven in some dozen lines. For details, cf. Abraham and Marsden (1978: Theorems 4.3.1-2, p. 299-300), or Arnold (1989: Appendix 5.B, p. 374-376).
Two final remarks. (1): The reduction of dynamics secured by the Marsden-Weinstein-Meyer theorem is similar to what we have seen before, for both the Poisson reduction theorem ((2) of Section 5.5), and the Lie-Poisson reduction theorem (Section 7.4). One proves, again in a few lines (Abraham and Marsden (1978: Theorems 4.3.5, p. 304):
> Marsden-Weinstein-Meyer reduction of dynamics Let $`H:M\mathrm{I}\mathrm{R}`$ be invariant under the action of $`G`$ on $`M`$, so that by Noether’s theorem for momentum maps (Section 6.2) $`𝐉`$ is conserved, i.e. $`𝐉^1(\alpha )`$ is invariant under the flow $`\varphi (t)`$ of $`X_H`$ on $`M`$. Then $`\varphi (t)`$ commutes with the action of $`G_\alpha `$ on $`𝐉^1(\alpha )`$ (i.e. $`\varphi (t)\mathrm{\Phi }_g=\mathrm{\Phi }_g\varphi (t)`$ for $`gG_\alpha `$), and so defines a flow $`\widehat{\varphi }(t)`$ on $`M_\alpha `$ such that $`\pi _\alpha \varphi (t)=\widehat{\varphi }(t)\pi _\alpha `$, i.e.
>
> $$\begin{array}{c}𝐉^1(\alpha )\\ \varphi \left(t\right)\\ 𝐉^1(\alpha )\end{array}\begin{array}{c}\stackrel{\pi _\alpha }{}\\ \\ \stackrel{\pi _\alpha }{}\end{array}\begin{array}{c}M_\alpha \\ \widehat{\varphi }\left(t\right)\\ M_\alpha \end{array}$$
> (7.442)
> The flow $`\widehat{\varphi }(t)`$ is Hamiltonian with the Hamiltonian $`H_\alpha `$ defined by $`H_\alpha \pi _\alpha =Hi_\alpha `$.
(2): I said at the start of this Subsection that the reader can now relate the Lie-Poisson reduction theorem to the Marsden-Weinstein-Meyer theorem. It is not hard to show that the former is an example of the latter. As the symplectic manifold $`M`$ one takes $`T^{}G`$, acted on symplectically by the cotangent lift of left translation. So we know (from (3) of Section 6.5.3) that $`𝐉_L:=T_e^{}R_g`$ is an $`Ad^{}`$-equivariant momentum map … and so on: I leave this as an exercise for the reader! The answer is supplied at Arnold (1989: 377, 321) and Abraham and Marsden (1978: 302). (Abraham and Marsden call it the ‘Kirillov-Kostant-Souriau theorem’.)
Suffice it to say here that this exercise gives another illustration of one of our central themes, that $`𝔤^{}`$’s symplectic leaves are the orbits of the co-adjoint representation. For the reduced phase space $`M_\alpha `$ is naturally identifiable with the co-adjoint orbit Orb($`\alpha `$) of $`\alpha 𝔤^{}`$, with the symplectic forms also naturally identified; (cf. also result (2) at the end of Section 5.4).
Acknowledgements:— I am grateful to audiences in Irvine, Oxford, Princeton and Santa Barbara; to several colleagues for encouragement; and to Gordon Belot, Klaas Landsman, David Wallace, and especially Graeme Segal, for very helpful, and patient!, conversations and correspondence.
## 8 References
R. Abraham and J. Marsden (1978), Foundations of Mechanics, second edition: Addison-Wesley.
V. Arnold (1973), Ordinary Differential Equations, MIT Press.
V. Arnold (1989), Mathematical Methods of Classical Mechanics, Springer, (second edition).
J. Barbour and B. Bertotti (1982), ‘Mach’s principle and the structure of dynamical theories’, Proceedings of the Royal Society of London A 382, p. 295-306.
G. Belot (1999), ‘Relationism rehabilitated’, International Studies in Philosophy of Science 13, p. 35-52.
G. Belot (2000), ‘Geometry and motion’, British Journal for the Philosophy of Science 51, p. 561-596.
G. Belot (2001), ‘The principle of sufficient reason’, Journal of Philosophy 98, p. 55-74.
G. Belot (2003), ‘Notes on symmetries’, in Brading and Castellani (ed.s) (2003), pp. 393-412.
G. Belot (2003a), ‘Symmetry and gauge freedom’, Studies in the History and Philosophy of Modern Physics 34, p. 189-225.
G. Belot (this volume).
G. Belot and J. Earman (2001), ‘Pre-Socratic quantum gravity’, in C. Callender and N. Huggett (ed.s) Physics meets Philosophy at the Planck Scale, Cambridge University Press, pp. 213-255.
R. Bishop and S. Goldberg (1980), Tensor Analysis on Manifolds, New York: Dover.
J. Boudri (2002), What was Mechanical about Mechanics: the Concept of Force between Metaphysics and Mechanics from Newton to Lagrange, Dordrecht: Kluwer Academic.
K. Brading and E. Castellani (ed.s) (2003), Symmetry in Physics, Cambridge University Press.
K. Brading and E. Castellani (this volume).
J. Butterfield (2004), ‘Some Aspects of Modality in Analytical mechanics’, in Formal Teleology and Causality, ed. M. Stöltzner, P. Weingartner, Paderborn: Mentis.
Available at Los Alamos arXive: http://arxiv.org/abs/physics/0210081 ; and at Pittsburgh archive: http://philsci-archive.pitt.edu/archive/00001192.
J. Butterfield (2004a), ‘Between Laws and Models: Some Philosophical Morals of Lagrangian Mechanics’; available at Los Alamos arXive: http://arxiv.org/abs/physics/0409030 ; and at Pittsburgh archive: http://philsci-archive.pitt.edu/archive/00001937/.
J. Butterfield (2005), ‘Between Laws and Models: Some Philosophical Morals of Hamiltonian Mechanics’, in preparation.
J. Butterfield (2006), ‘On Symmetry and Conserved Quantities in Classical Mechanics’, forthcoming in a Festschrift for Jeffrey Bub, ed. W. Demopoulos and I. Pitowsky, Kluwer: University of Western Ontario Series in Philosophy of Science. available at Los Alamos arXive: http://arxiv.org/abs/physics/ ; and at Pittsburgh archive: http://philsci-archive.pitt.edu/archive/00002362/
R. Darling (1994), Differential Forms and Connections, Cambridge University Press.
M. Dickson (this volume).
E. Desloge (1982), Classical Mechanics, John Wiley.
J. Earman (2003), ‘Tracking down gauge: an ode to the constrained Hamiltonian formalism’, in Brading and Castellani (ed.s) (2003), pp. 140-162.
H. Goldstein (1950), Classical Mechanics, Addison-Wesley; (1966 third printing).
I. Grattan-Guinness (2000), ‘A sideways look at Hilbert’s twenty-three problems of 1900’, Notices of the American Mathematical Society 47, p. 752-757.
I. Grattan-Guinness (2006), ‘Classical mechanics as a formal(ised) science’, in B. Loewe (ed.), Foundations of the Formal Sciences, Kluwer, to appear.
T. Hawkins (2000), Emergence of the Theory of Lie Groups: an essay in the history of mathematics 1869-1926, New York: Springer.
M. Jammer (1957), Concepts of Force, Harvard University Press.
M. Jammer (1961), Concepts of Mass in Classical and Modern Physics, Harvard University Press; republished by Dover in 1997.
M. Jammer (2000), Concepts of Mass in Contemporary Physics and Philosophy, Princeton University Press.
O. Johns (2005), Analytical Mechanics for Relativity and Quantum Mechanics, Oxford University Press, forthcoming.
A. Kirillov (1976), Elements of the Theory of Representations, Grunlehren Math. Wiss., Springer-Verlag.
H. Kragh (1999), Quantum Generations, Princeton University Press.
N. Landsman (this volume), ‘Between Classical and Quantum’.
S. Lie (1890). Theorie der Transformationsgruppen: zweiter abschnitt, Leipzig: B.G.Teubner.
J. Lutzen (1995), Denouncing Forces; Geometrizing Mechanics: Hertz’s Principles of Mechanics, Copenhagen University Mathematical Institute Preprint Series No 22.
J. Lutzen (2003), ‘Between rigor and applications: developments in the concept of function in mathematical analysis’, in Cambridge History of Science, vol. 5: The modern physical and mathematical sciences, ed. M.J. Nye, p. 468-487.
J. Lutzen (2005) Mechanistic Images in Geometric Form: Heinrich Hertz’s ’Principles of Mechanics’, Oxford University Press.
J. Marsden and T. Hughes (1982), Mathematical Foundations of Elasticity, Prentice-Hall; Dover 1994.
J. Marsden and T. Ratiu (1999), Introduction to Mechanics and Symmetry, second edition: Springer-Verlag.
E. McMullin (1978), Newton on Matter and Activity, University of Notre Dame Press.
I. Newton (1687), Principia: Mathematical Principles of Natural Philosophy, trans. I.B. Cohen and A. Whitman, Cambridge University Press (1999).
P. Olver (2000), Applications of Lie Groups to Differential Equations, second edition: Springer-Verlag.
C. Rovelli (this volume), ‘Quantum Gravity’.
T. Ryckman (2005), The Reign of Relativity: Philosophy in Physics 1915-1925 Oxford University Press.
S. Singer (2001), Symmetry in Mechanics: a Gentle Modern Introduction, Boston: Birkhauser.
E. Slovik (2002), Cartesian Spacetime: Descartes’ Physics and the Relational Theory of Space and Motion, Dordrecht: Kluwer Academic.
M. Stöltzner (2003), ‘The Principle of Least Action as the Logical Empiricist’s Shibboleth’, Studies in History and Philosophy of Modern Physics 34B, p. 285-318.
D. Wallace (2003), ‘Time-dependent Symmetries: the link between gauge symmetries and indeterminism’, in Brading and Castellani (ed.s) (2003), pp. 163-173. |
warning/0507/quant-ph0507034.html | ar5iv | text | # 1 Introduction
## 1 Introduction
Local operations and classical communications (LOCC) are basic operations in quantum information theory. Many interesting studies have arisen from the question, what we can$`\backslash `$cannot do using only LOCC. The question is highly non-trivial and difficult to solve due to the lack of simple characterization of LOCC. The necessary and sufficient condition of the deterministic convertibility of one pure state to the other was derived by Nielsen, for general bipartite systems, in . Furthermore, in , Vidal obtained the optimal probability to convert one pure state to the other, non-deterministically. However, when we start to think of simultaneous convertibility of more than one states, the problem becomes furthermore difficult, because of the fact that Lo-Popescu Theorem is not applicable there.
The local distinguishability problem is one of these questions. The problem is as follows: We investigate a combined quantum system consisting of two parts $`A`$ and $`B`$ held by separated observers (Alice and Bob). We denote the associated Hilbert space by $`_A_B`$, where $`_A`$, $`_B`$ are separable (i.e., possibly infinite dimensional) Hilbert spaces that represent the system of Alice and Bob, respectively. Let $`\psi _1,\mathrm{},\psi _M`$ be orthonormal vectors in $`_A_B`$, which represent $`M`$ pure states. Suppose that the system is in a state $`\psi `$, which is prepared to be one of $`\psi _1,\mathrm{},\psi _M`$. Alice and Bob know that $`\psi `$ is one of $`\psi _1,\mathrm{},\psi _M`$, but they don’t know which of them it is. The problem is if Alice and Bob can find out which one it is, when only LOCC is allowed.
In , Walgate et.al. proved that any two orthogonal pure states in finite dimensional systems are distinguishable. Unfortunately, because of the nature of their proof, this important result has been restricted to finite dimensional systems so far. As it is indispensable to consider infinite dimensional systems in the real world, the analogous result in infinite dimensional system is desirable. In this paper, we prove the infinite version:
###### Theorem
Any two orthogonal pure states are distinguishable by LOCC, even for infinite dimensional systems.
In spite of these simple results for two pure states, it is known that more than two pure states are not always distinguishable by LOCC. It was proved that three Bell states can not be distinguished with certainty by LOCC and four Bell states can not, even probabilistically . A set of non-entangled pure states that are not locally distinguishable was introduced in . The probability of the discrimination for the worst case was estimated in . In this paper, we give an estimate of discrimination probability for some family of more than two pure states. This result also holds for infinite dimensional systems.
In order to investigate distinguishability, we look for a suitable decomposition of the states. Let us decompose the vectors $`\psi _1,\mathrm{},\psi _M`$ with respect to an orthonornal basis $`\{e_k\}`$ of $`_B`$:
$`\psi _l={\displaystyle \underset{k}{}}\xi _k^le_k,l=1,\mathrm{},M.`$ (1)
(Here and below, if the dimension of $`_B`$ is finite $`n`$, $`_i\phi _i^{}f_i^{}`$ stands for the sum $`_{i=1}^n\phi _i^{}f_i^{}`$, while if $`_B`$ is infinite dimensional, it stands for the limit $`lim_n\mathrm{}_{i=1}^n\phi _i^{}f_i^{}`$, when the limit converges in the norm topology of $`_A_B`$.) Suppose that the vectors $`\{\xi _k^l\}`$ satisfy the orthogonal conditions for each $`k`$:
$`\xi _k^l|\xi _k^m=0lm,k.`$ (2)
This orthogonality condition does not hold in general, but if this condition holds, Alice and Bob can distinguish these states by the following LOCC: First Bob performs a projective measurement $`\{|e_ke_k|\}`$ on his side. Then he tells the result $`k`$ of his measurement to Alice by a classical communication. For each $`k`$, let $`S_k`$ be a set of $`1lM`$ such that $`\xi _k^l0`$. According to the information from Bob, Alice performs a projective measurement given by projections $`\{|\widehat{\xi }_k^l\widehat{\xi }_k^l|\}_{lS_k}`$ and $`1_{lS_k}|\widehat{\xi }_k^l\widehat{\xi }_k^l|`$. Here, a vector $`\widehat{\xi }_k^l_A`$ is the normalization of the vector $`\xi _k^l_A`$. As $`\{\xi _k^l\}_{lS_k}`$ are mutually orthogonal for each $`k`$, the projections are orthogonal. Because the initial state $`\psi `$ was prepared to be one of $`\psi _1,\mathrm{},\psi _M`$, Alice obtains one of $`\widehat{\xi }_k^l`$, $`lS_k`$. When Bob obtains $`e_k`$ and Alice obtains $`\widehat{\xi _k^l}`$, they can say the original state $`\psi `$ was $`\psi _l`$, because if $`\psi =\psi _m`$ for $`ml`$, the probability that they obtain $`e_k`$ and $`\widehat{\xi _k^l}`$ is $`0`$. Hence a deterministic local discrimination is possible when the decomposition (1) with the orthogonality condition (2) is given.
Next let us consider probabilistic discriminations. Suppose that $`\psi _1,\mathrm{},\psi _M`$ are decomposed into the form (1), but now the orthogonal condition holds only partially, i.e., just for $`k`$ larger than some $`N_p`$:
$`\xi _k^l|\xi _k^m=0lmk>N_p.`$ (3)
In this case, $`\psi _1,\mathrm{},\psi _M`$ can be distinguished by conclusive LOCC protocol, probabilistically. Let $`P_d`$ be the largest probability that can be attained. The conclusive protocol below gives the lower bound of $`P_d`$:
$`P_d1\underset{1lM}{\mathrm{max}}{\displaystyle \underset{k=1}{\overset{N_p}{}}}\xi _k^l^2.`$ (4)
First Bob performs the projective measurement $`\{|e_ke_k|\}`$ again. If he gets the result $`k>N_p`$, he tells the result to Alice. Then Alice performs the projective measurement given by projections $`\{|\widehat{\xi }_k^l\widehat{\xi }_k^l|\}_{lS_k}`$ and $`1_{lS_k}|\widehat{\xi }_k^l\widehat{\xi }_k^l|`$, and obtains one of $`\widehat{\xi }_k^l`$. If she gets $`\widehat{\xi }_k^l`$, then they can conclude $`\psi =\psi _l`$, as before. In this way, they can distinguish $`\psi _1,\mathrm{},\psi _M`$ if the result of Bob’s measurement is $`k>N_p`$. On the other hand, if Bob obtains $`kN_p`$, we regard it as an error. When $`\psi =\psi _l`$, the probability the error occurs is $`_{k=1}^{N_p}\xi _k^l^2`$. Hence $`\psi _1,\mathrm{},\psi _M`$ can be distinguished by LOCC with probability $`P_d`$, lower bounded as in (4).
The problem is if there is a decomposition (1) of $`\psi _1,\mathrm{},\psi _M`$, satisfying the orthogonality condition (2) or (3). In order to deal with this problem, we will introduce a real vector space of trace class self-adjoint operators on the Hilbert space $`_B`$, determined by the states $`\psi _1,\mathrm{},\psi _M`$. We will denote the vector space by $`𝒦`$. Let $`N`$ be the dimension of $`𝒦`$ and $`(A_1,\mathrm{},A_N)`$ a basis of $`𝒦`$. For every orthogonal projection $`P`$ on $`_B`$, we investigate the subset of $`^N`$ given by
$`\{(z,A_1z,\mathrm{},z,A_Nz):zP_B,z=1\}^N.`$
This set is the joint numerical range of operators $`(A_1,\mathrm{},A_N)`$, restricted on the sub-Hilbert space $`P_B`$. We will show that the convexity of these sets implies the existence of the decomposition (1) with the orthogonality condition (2), hence the local distinguishability of the states $`\psi _1,\mathrm{},\psi _M`$. One of the advantage of this method is that we can consider infinite systems, easily.
In this paper, we prove the infinite version of : by the convex analysis on joint numerical ranges, we show that any two pure orthogonal states can be decomposed as in (1), with the orthogonality condition (2). We also apply our method to investigate the distinguishability of more than two pure states. We show that if the dimension of $`𝒦`$ is $`3`$, the condition (3) holds for $`N_P=2`$, hence the states are distinguishable probabilistically. (Theorem 2.2)
The remainder of the paper is organized in the following way: In Section 2, we introduce a representation of a vector in $`_A_B`$ as an operator from $`_B`$ to $`_A`$. And from them, we define the real vector space $`𝒦`$. Then we represent our main results in terms of the vector space $`𝒦`$. In Section 3, by use of convex analysis on joint numerical ranges, we show the distinguishability of states.
## 2 The distinguishability of states
In this section, we introduce a representation of pure states on $`_A_B`$ as operators from $`_B`$ to $`_A`$, and describe our main results in terms of the operator representation. In finite dimensional systems, the operator representation corresponds to the well known matrix representation of states, by use of a maximal entangled state. (See for example ).
Let $`_A`$, $`_B`$ be separable (possibly infinite dimensional) Hilbert spaces. Let us fix some orthonormal basis $`\{f_i\}`$ of $`_B`$. A vector $`\psi `$ in $`_A_B`$ can be decomposed as
$`\psi ={\displaystyle \underset{i}{}}\phi _if_i,`$
in general. Here, the limit $`lim_n\mathrm{}_{i=1}^n\phi _if_i`$ converges in the norm topology of $`_A_B`$ for infinite dimensional case. The vectors $`\phi _i`$ in $`_A`$ satisfy
$`{\displaystyle \underset{i}{}}\phi _i^2=\psi ^2.`$ (5)
Now we define a bounded linear operator $`X`$ from $`_B`$ to $`_A`$ by
$`X\eta {\displaystyle \underset{i}{}}f_i|\eta \phi _i,\eta _B.`$ (6)
From (5), the sum in (6) absolutely converges in norm of $`_B`$, and we obtain $`X\psi `$. Then the vector $`\psi `$ is represented as
$`\psi ={\displaystyle \underset{i}{}}\phi _if_i={\displaystyle \underset{i}{}}(Xf_i)f_i.`$
The bounded operator $`X^{}X`$ on $`_B`$ satisfies
$`TrX^{}X={\displaystyle \underset{i}{}}\phi _i^2=\psi ^2<\mathrm{},`$ (7)
i.e., $`X^{}X`$ is a trace class operator on $`_B`$. By operating $`1|f_if_i|`$ on $`\psi `$, we see that $`X`$ is the unique operator such that $`\psi =_iXf_if_i`$. On the other hand, for any bounded linear operator $`X`$ from $`_B`$ to $`_A`$ satisfying $`TrX^{}X<\mathrm{}`$, there exists a unique vector $`_iXf_if_i`$, (i.e., there exists the limit $`lim_n\mathrm{}_{i=1}^nXf_if_i`$ for infinite dimensional case, in the norm of $`_A_B`$.) Hence we obtain the following one-to-one correspondence:
$`\psi _A_BXB(_B,_A),s.t.TrX^{}X<\mathrm{},`$
through the relation
$`\psi ={\displaystyle \underset{i}{}}(Xf_i)f_i.`$ (8)
Here $`B(_B,_A)`$ indicates the set of bounded operators from $`_B`$ to $`_A`$.
Now let us consider a set of orthonormal $`M`$ vectors $`\psi _1,\mathrm{},\psi _M`$ in $`_A_B`$. We can associate each $`\psi _l`$ with an operator $`X_l`$ through (8). As in (7), $`X_m^{}X_l`$ are trace class operators on $`_B`$ for all $`1m,lM`$ and satisfy
$`TrX_m^{}X_l=\psi _m,\psi _l=\delta _{m,l},1m,lM.`$ (9)
Let $`𝒦`$ be the real linear subspace of trace class self-adjoint operators on $`_B`$ spanned by operators $`\{X_m^{}X_l+X_l^{}X_m,i(X_m^{}X_lX_l^{}X_m)\}_{ml}`$. Let $`N`$ be the dimension of $`𝒦`$ and $`(A_1,\mathrm{},A_N)`$ an arbitrary basis of $`𝒦`$. The dimension $`N`$ is bounded as $`NM(M1)`$. Because each $`X_m^{}X_l`$ satisfies (9), we have
$`TrA_i=0,i=1,\mathrm{},N.`$ (10)
We will call $`𝒦`$ the real vector space of trace class self-adjoint operators associated with $`\psi _1,\mathrm{},\psi _M`$.
Now we are ready to state our main results. In this paper, we show the following theorems:
###### Theorem 2.1
Let $`_A`$, $`_B`$ be (possibly infinite dimensional) separable Hilbert spaces. Let $`\psi _1,\mathrm{},\psi _M`$ be a set of orthogonal pure states in $`_A_B`$ and $`𝒦`$ the associated real vector space of trace class self-adjoint operators on $`_B`$. Then if the dimension of $`𝒦`$ is $`2`$, the states $`\psi _1,\mathrm{},\psi _M`$ are distinguishable by LOCC with certainty. In particular, any pair of orthogonal pure states $`\psi _1,\psi _2`$ are distinguishable by LOCC with certainty.
###### Theorem 2.2
Let $`_A`$, $`_B`$ be (possibly infinite dimensional) separable Hilbert spaces. Let $`\psi _1,\mathrm{},\psi _M`$ be a set of orthogonal pure states in $`_A_B`$ and $`𝒦`$ the associated real vector space of trace class self-adjoint operators on $`_B`$. Suppose that the dimension of $`𝒦`$ is $`3`$. Then $`\psi _1,\mathrm{},\psi _M`$ are distinguishable by conclusive LOCC protocol with probability $`P_d`$ such that
$$P_d1\underset{1lM}{max}\left(\underset{k=1}{\overset{2}{}}p_k^l\right).$$
Here, $`p_k^l`$ represents the $`k`$-th Schmidt coefficient of $`\psi _l`$, ordered in the decreasing order.
###### Remark 2.3
The last statement of Theorem 2.1 is the extension of to infinite dimensional system. Applying the argument in , we can extend the result to multipartite systems: any two orthogonal pure states in multipartite systems are distinguishable by LOCC even in infinite dimensional systems.
###### Remark 2.4
In , S.Virmani et.al. showed that any two (even non-orthogonal) multipartite pure states in finite dimensional systems can be optimally distinguished using only LOCC. It was derived using the result of the orthogonal case in . The argument there can be applied to our infinite dimensional case. Therefore, any two bipartite pure states can be optimally distinguished using only LOCC, even for infinite dimensional system.
## 3 Proof
In this section, we prove the main theorems. We correlate the problem of the distinguishability with that of the convexity of the joint numerical ranges. Let $`(A_1,\mathrm{},A_N)`$ be bounded self-adjoint operators on a Hilbert space $``$. A subset of $`^N`$ given by
$`\{(z,A_1z,z,A_2z,\mathrm{},z,A_Nz);z,z=1\}^N`$
is called the joint numerical range of $`(A_1,\mathrm{},A_N)`$. Furthermore, for an orthogonal projection $`P`$ on $``$, we will call the set
$`C_P(A_1,\mathrm{},A_N)\{(z,A_1z,z,A_2z,\mathrm{},z,A_Nz);zP,z=1\}^N,`$
the joint numerical range of $`(A_1,\mathrm{},A_N)`$ restricted to the sub-Hilbert space $`P`$. Theorem 2.1, Theorem 2.2 are derived as corollaries of the following propositions:
###### Proposition 3.1
Let $`\psi _1,\mathrm{},\psi _M`$ be a set of orthogonal pure states in $`_A_B`$, and $`𝒦`$ the associated real vector space of trace class self-adjoint operators on $`_B`$. Let $`(A_1,\mathrm{},A_N)`$ be a basis of $`𝒦`$. Suppose that for any projection $`P`$ on $`_B`$, $`C_P(A_1,\mathrm{},A_N)`$ is convex. Then the states $`\psi _1,\mathrm{},\psi _M`$ are distinguishable by LOCC with certainty.
###### Proposition 3.2
Let $`\psi _1,\mathrm{},\psi _M`$ be a set of orthonormal pure states in $`_A_B`$, and $`𝒦`$ the associated real vector space of trace class self-adjoint operators on $`_B`$. Let $`(A_1,\mathrm{},A_N)`$ be a basis of $`𝒦`$. Suppose that for any projection $`P`$ of $`_B`$ with dimension larger than $`N_p`$, $`C_P(A_1,\mathrm{},A_N)`$ is convex. Then the states $`\psi _1,\mathrm{},\psi _M`$ are distinguishable by LOCC with the probability $`P_d`$ such that
$$P_d1\underset{1lM}{\mathrm{max}}\left(\underset{k=1}{\overset{N_p}{}}p_k^l\right).$$
Here, $`p_k^l`$ represents the $`k`$-th Schmidt coefficient of $`\psi _l`$, ordered in the decreasing order.
First we prove the Proposition 3.1. The proof consists of four steps: Step 1. First, we show that if $`_B`$ has an orthonormal basis $`\{g_k\}`$ such that $`g_k,A_ig_k=0`$ for all $`i=1,\mathrm{},N`$ and $`k`$, then, $`\psi _1,\mathrm{},\psi _N`$ are distinguishable by LOCC (Lemma 3.3). Step 2. Second, using convex analysis, we show that if the joint numerical range of $`(A_1,\mathrm{},A_N)`$ is convex, there exists at least one vector $`z_B`$ such that $`z,A_iz=0`$ for all $`i=1,\mathrm{},N`$ (Lemma 3.4). Step 3. Third, using Lemma 3.4, we show the existence of the orthonormal basis satisfying the desired condition in Step 1 (Lemma 3.6). Step 4. Finally, combining the results of Step 1 and Step 3, we obtain Proposition 3.1.
Now let us start the proof. First we show the following Lemma:
###### Lemma 3.3
Let $`(A_1,\mathrm{},A_N)`$ be a basis of $`𝒦`$ associated with $`\psi _1,\mathrm{},\psi _M`$. Suppose that there exists an orthonormal basis $`\{g_k\}`$ of $`_B`$ such that
$`g_k,A_ig_k=0,k,i=1\mathrm{}N.`$ (11)
Then the states $`\psi _1,\mathrm{},\psi _M`$ are distinguishable by LOCC.
Proof Let $`\{f_i\}`$ be the orthonormal basis fixed in Section 2. (Recall that we defined the operators $`X_l`$s in terms of $`\{f_i\}`$.) We define an antilinear operator $`J:_B_B`$ to be the complex conjugation with respect to $`\{f_i\}`$:
$`J{\displaystyle \underset{i}{}}\alpha _if_i{\displaystyle \underset{i}{}}\overline{\alpha _i}f_i.`$
As $`J`$ is an antilinear isometry, $`\{Jg_k\}`$ is an orthonormal basis of $`_B`$. Therefore, we can decompose $`\psi _1,\mathrm{},\psi _M`$ with respect to $`\{Jg_k\}`$:
$`\psi _l={\displaystyle \underset{k}{}}\xi _k^lJg_k.`$ (12)
We show that for each $`k`$, $`\{\xi _k^1,\mathrm{},\xi _k^M\}`$ are mutually orthogonal.
Let us decompose $`\psi _l`$ with respect to $`\{f_i\}`$:
$`\psi _l={\displaystyle \underset{i}{}}\phi _i^lf_i.`$ (13)
Comparing (12) and (13), we obtain
$`\xi _k^l={\displaystyle \underset{i}{}}\phi _i^lJg_k,f_i={\displaystyle \underset{i}{}}\phi _i^lf_i,g_k=X_lg_k.`$
As $`(A_,\mathrm{},A_N)`$ is a basis of $`𝒦`$, the assumption (11) implies
$`\xi _k^l,\xi _k^m=X_lg_k,X_mg_k=0lm,k.`$
Hence for each $`k`$, $`\{\xi _k^1,\mathrm{},\xi _k^M\}`$ are mutually orthogonal.
Thus (12) takes the form of (1), with the orthogonality condition (2). Therefore, from the arguments in the Introduction, we can distinguish $`\psi _1,\mathrm{},\psi _M`$ by LOCC with certainty. $`\mathrm{}`$
Next we show the following Lemma which holds on a general Hilbert space $``$:
###### Lemma 3.4
Let $`(A_1,\mathrm{},A_N)`$ be a set of trace class self-adjoint operators on a Hilbert space $``$ such that $`TrA_i=0`$ for each $`1iN`$. Suppose that the joint numerical range of $`(A_1,\mathrm{},A_N)`$ is a convex subset of $`^N`$. Then there exists a vector $`z`$ with $`z=1`$ such that
$`z,A_iz=0,i=1,\mathrm{},N.`$
Proof
Before starting the proof, we review some basic facts from convex analysis . Let $`x_1,\mathrm{},x_k`$ be elements in $`^N`$. An element $`_{i=1}^k\alpha _ix_i`$ with real coefficients $`\alpha _i`$ satisfying $`_{i=1}^k\alpha _i=1`$ is called an affine combination of $`x_1,\mathrm{},x_k`$. An affine manifold in $`^N`$ is a set containing all its affine combinations. Let $`S`$ be a nonempty subset of $`^N`$. The affine hull of $`S`$ is defined to be the smallest affine manifold containing $`S`$. We denote the affine hull of S by $`\mathrm{aff}S`$. In other words, $`\mathrm{aff}S`$ is the affine manifold generated by $`S`$. As easily seen, it is a closed plane parallel to a linear subspace in $`^N`$. Its dimension may be lower than $`N`$ in general. The relative interior of $`S`$, $`\mathrm{ri}S`$, is the interior of $`S`$ with respect to the topology relative to $`\mathrm{aff}S`$. In other words,
$`\mathrm{ri}S\{xS;\epsilon >0s.t.B(x,\epsilon )\mathrm{aff}SS\}.`$
Here, $`B(x,\epsilon )`$ is a ball of radius $`\epsilon `$, centered at $`x`$. The following fact is known:
###### Lemma 3.5
Let $`C`$ be a nonempty convex subset of $`^N`$. Then for any point $`x_0`$ in $`\mathrm{aff}C\backslash \mathrm{ri}C`$, there exists a non-zero vector $`s^N`$ parallel to $`\mathrm{aff}C`$, such that
$$s,xx_00,xC.$$
Here $`,`$ is the inner product of $`^N`$:
$`s,x{\displaystyle \underset{i=1}{\overset{N}{}}}s_ix_i.`$
Now we are ready to prove Lemma 3.4. The claim is equivalent to saying that $`0`$ is included in the joint numerical range of the operators $`(A_1,\mathrm{},A_N)`$. We denote the joint numerical range by $`C_1`$:
$`C_1\left\{(z,A_1z,z,A_2z,\mathrm{},z,A_Nz)^N;z,z=1\right\}.`$
By assumption, $`C_1`$ is a nonempty convex subset of $`^N`$. Let $`\{e_k\}`$ be an arbitrary orthonormal basis of $``$. By the definition of $`C_1`$,
$`x_k(e_k,A_1e_k,\mathrm{},e_k,A_Ne_k)`$
is an element of $`C_1`$ for each $`k`$.
The finite dimensional case $`=^n`$ is immediate. By the convexity of $`C_1`$, we obtain
$`0={\displaystyle \frac{1}{n}}(TrA_1,\mathrm{},TrA_N)={\displaystyle \frac{1}{n}}{\displaystyle \underset{k=1}{\overset{n}{}}}(e_k,A_1e_k,\mathrm{},e_k,A_Ne_k)C_1.`$
Below we prove the infinite dimensional case.
First we observe that $`0`$ is included in the closure of $`C_1`$. In particular, $`0`$ is in $`\mathrm{aff}C_1`$. To see this, note that for all $`l`$, we have
$`{\displaystyle \frac{1}{l}}{\displaystyle \underset{k=1}{\overset{l}{}}}(e_k,A_1e_k,\mathrm{},e_k,A_Ne_k)C_1.`$
As $`A_i`$ is a trace class operator, the sum $`_{k=1}^{\mathrm{}}e_k,A_ie_k`$ converges absolutely. By taking $`l\mathrm{}`$ limit, we obtain
$`0=\underset{l\mathrm{}}{lim}{\displaystyle \frac{1}{l}}{\displaystyle \underset{k=1}{\overset{l}{}}}(e_k,A_1e_k,\mathrm{},e_k,A_Ne_k)\overline{C_1}\mathrm{aff}C_1.`$
Hence $`0`$ is in $`\mathrm{aff}C_1`$.
Second, we show that $`0`$ is actually in $`\mathrm{ri}C_1`$. To prove this, assume $`0`$ is not included in $`\mathrm{ri}C_1`$. Then it is an element of $`\mathrm{aff}C_1\backslash \mathrm{ri}C_1`$. As $`C_1`$ is a nonempty convex set, from Lemma 3.5, there exists a non-zero vector $`s=(s_1,\mathrm{},s_N)^N`$ pararell to $`\mathrm{aff}C_1`$, such that
$`s,x0,xC_1.`$
As $`x_kC_1`$, we have
$`s,x_k0,`$ (14)
for all $`k`$. On the other hand, we have
$`{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}s,x_k={\displaystyle \underset{i=1}{\overset{N}{}}}s_i{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}e_k,A_ie_k={\displaystyle \underset{i=1}{\overset{N}{}}}s_iTrA_i=0.`$ (15)
From (14) and (15), we obtain $`s,x_k=0`$ for all $`k`$. As the orthonormal basis $`\{e_k\}`$ can be taken arbitrary, we obtain
$`s,x=0,xC_1.`$
As $`s`$ is a non-zero vector parallel to $`\mathrm{aff}C_1`$, this means that $`C_1`$ is included in some affine manifold that is strictly smaller than $`\mathrm{aff}C_1`$. This contradicts the definition of $`\mathrm{aff}C_1`$. (Recall that $`\mathrm{aff}C_1`$ is the smallest affine manifold including $`C_1`$.) Therefore, we obtain $`0\mathrm{ri}C_1`$. In particular, $`0C_1`$ and this completes the proof. $`\mathrm{}`$
Using Lemma 3.4, we obtain the following lemma:
###### Lemma 3.6
Let $`(A_1,\mathrm{},A_N)`$ be a set of trace class self-adjoint operators on a Hilbert space $``$ such that $`TrA_i=0`$ for each $`1iN`$. Suppose that for every orthogonal projection $`P`$ on $``$, $`C_P(A_1,\mathrm{},A_N)`$ is convex. Then there exists an orthonormal basis $`\{g_k\}`$ of $``$, such that
$$g_k,A_ig_k=0,i=1,\mathrm{}N,k.$$
Proof
We will say that a set of vectors $`Z`$ in $``$ satisfies Property * if it satisfies the following conditions:
Property *
1. $`Z`$ is a set of mutually orthogonal unit vectors of $``$.
2. $`z,A_iz=0,i=1,\mathrm{},N`$ for all $`zZ`$.
By Zorn’s lemma, there exists a maximal set of orthonormal vectors $`\{g_k\}`$ in $``$ which satisfies the Property *. It suffices to show that $`\{g_k\}`$ is complete.
Suppose that $`\{g_k\}`$ is not complete in $``$, and let $`P`$ be the orthogonal projection onto the sub-Hilbert space spanned by $`\{g_k\}`$. From Property *, we have
$`TrPA_iP={\displaystyle \underset{k}{}}g_k,A_ig_k=0,i=1,\mathrm{}N.`$
Let $`\overline{P}`$ be $`\overline{P}=1P`$. Now we regard $`(\overline{P}A_1\overline{P},\mathrm{},\overline{P}A_N\overline{P})`$ as self-adjoint trace class operators on the Hilbert space $`\overline{P}`$ such that
$`Tr_{\overline{P}}(\overline{P}A_i\overline{P})=Tr(A_i)Tr(PA_iP)=0,i=1,\mathrm{}N.`$
By the assumption, the joint numerical range of $`(\overline{P}A_1\overline{P},\mathrm{},\overline{P}A_N\overline{P})`$ on $`\overline{P}`$ is convex. Thus, applying Lemma 3.4, there exists a unit vector $`z\overline{P}`$ such that $`z,A_iz=0`$ for all $`i=1,\mathrm{},N`$. As $`z`$ is orthogonal to all $`g_k`$, the set $`\{z\}\{g_k\}`$ satisfies the Property *, and is strictly larger than $`\{g_k\}`$. This contradicts the maximality of $`\{g_k\}`$. Therefore, $`\{g_k\}`$ is complete. $`\mathrm{}`$
Now, let us complete the proof of Proposition 3.1. The basis of $`𝒦`$, $`(A_1,\mathrm{},A_N)`$ are trace class self-adjoint operators satisfying $`TrA_i=0,i=1,\mathrm{},N`$ (10). Therefore, if $`C_P(A_1,\mathrm{},A_N)`$ is a convex subset of $`^N`$ for any orthogonal projection $`P`$ on $`_B`$, there exists an orthonormal basis $`\{g_k\}`$ of $`_B`$ such that $`g_k,A_ig_k=0`$, for all $`i=1,\mathrm{}N`$ and $`k`$, from Lemma 3.6. By Lemma 3.3, this concludes that $`\psi _1,\mathrm{},\psi _M`$ are distinguishable by LOCC.$`\mathrm{}`$
Proposition 3.2 can be shown in the same way. We have the following lemma:
###### Lemma 3.7
Let $`(A_1,\mathrm{},A_N)`$ be a set of trace class self-adjoint operators on a Hilbert space $``$ such that $`TrA_i=0`$ for each $`1iN`$. Suppose that for every orthogonal projection $`P`$ on $``$ with dimension larger than $`N_p`$, $`C_P(A_1,\mathrm{},A_N)`$ is convex. Then there exists an orthonormal basis $`\{g_k\}`$ of $``$, such that
$$g_k,A_ig_k=0,i=1,\mathrm{}N,k>N_p.$$
Proof
The same as the proof of Lem 3.6. We can find a set of orthonormal vectors satisfying Property *, such that the dimension of its complementary subspace is $`N_p`$. $`\mathrm{}`$
Decomposing each $`\psi _l`$ with respect to $`\{Jg_k\}`$, we obtain
$`\psi _l={\displaystyle \underset{k}{}}\xi _k^lJg_k.`$ (16)
By the argument in the proof of Lemma 3.3, (16) takes the form of (1) with the orthogonality condition (3). Therefore, for the protocol in the Introduction, the probability that the error occurs is $`_{k=1}^{N_p}\xi _k^l^2`$ when $`\psi =\psi _l`$. It is bounded from above as follows:
$`{\displaystyle \underset{k=1}{\overset{N_p}{}}}\xi _k^l^2={\displaystyle \underset{k=1}{\overset{N_p}{}}}\left(1|Jg_kJg_k|\right)\psi _l^2`$
$`sup\{{\displaystyle \underset{k=1}{\overset{N_p}{}}}(1|z_kz_k\left|\right)\psi _l^2;\left\{z_k\right\}_{k=1}^{N_p}:\mathrm{orthonormal}\mathrm{set}\mathrm{of}_B\}={\displaystyle \underset{k=1}{\overset{N_p}{}}}p_k^l.`$
Here, $`p_k^l`$ is the $`k`$-th Schmidt coefficient of $`\psi _l`$, ordered in the decreasing order. Therefore, $`\psi _1,\mathrm{},\psi _M`$ are distinguishable by LOCC with probability $`P_d`$ such that
$$P_d1\underset{1lM}{\mathrm{max}}\left(\underset{k=1}{\overset{N_p}{}}\xi _k^l^2\right)1\underset{1lM}{\mathrm{max}}\left(\underset{k=1}{\overset{N_p}{}}p_k^l\right),$$
and we obtain Proposition 3.2.$`\mathrm{}`$
Proof of Theorem 2.1 and Theorem 2.2
Now we apply the known results about joint numerical range to Proposition 3.1, 3.2 and derive Theorem 2.1 and Theorem 2.2. For $`N=2`$ case, the following Theorem is known :
###### Theorem 3.8
For any bounded self-adjoint operators $`T_1,T_2`$ on a separable Hilbert space $``$, the set
$`\left\{(z,T_1z,z,T_2z)^2,z,z=1\right\}`$
is a convex subset of $`^2`$.
This is called Toeplitz Hausdorff Theorem. By this Theorem, $`C_P(A_1,A_2)`$ is a convex subset of $`^2`$ for any projection $`P`$ on $`_B`$. Therefore, applying Proposition 3.1, we obtain Theorem 2.1. The last statement comes from the fact $`NM(M1)=2`$ for $`M=2`$.
On the other hand, for $`N=3`$, the next Theorem is known ,.
###### Theorem 3.9
Let $``$ be a separable Hilbert space with $`dim3`$. Then for any self-adjoint operators $`T_1,T_2,T_3`$ in $``$, the set
$`\left\{(z,T_1z,z,T_2z,z,T_3z)^3,z,z=1\right\}`$
is a convex subset of $`^3`$.
By this Theorem, $`C_P(A_1,A_2,A_3)`$ is a convex subset of $`^3`$ for any projection $`P`$ on $`_B`$ with dimension larger than $`2`$. Therefore, applying Proposition 3.2, we obtain Theorem 2.2. $`\mathrm{}`$.
Acknowledgement.
This work is supported by Research Fellowships of the Japan Society for the Promotion of Science for Young Scientists. |
warning/0507/hep-th0507218.html | ar5iv | text | # Confinement of spinless particles by Coulomb potentials in two-dimensional space-time
## 1 Introduction
The potential generated by a point charge, the Coulomb potential, depends on the dimensionality of space and in a 1+1 dimension it is linear (see, e.g. ). As the time component of a Lorentz vector, the 1+1 dimensional Coulomb potential is linear and so it provides a constant electric field always pointing to, or from, the point charge. This problem is related to the confinement of fermions in the Schwinger and in the massive Schwinger models - and in the Thirring-Schwinger model . It is frustrating that, due to the tunneling effect (Klein´s paradox), there are no bound states for this kind of potential regardless of the strength of the potential -. The linear potential, considered as a Lorentz scalar, is also related to the quarkonium model in one-plus-one dimensions -. Recently it was incorrectly concluded that even in this case there is just one bound state . Later, the proper solutions for this last problem were found -. However, it is well known from the quarkonium phenomenology in the real 3+1 dimensional world that the best fit for meson spectroscopy is found for a convenient mixture of vector and scalar potentials put by hand in the equations (see, e.g., ). The same can be said about the treatment of the nuclear phenomena describing the influence of the nuclear medium on the nucleons -. The mixed vector-scalar potential has also been analyzed with the Dirac equation in 1+1 dimensions for a linear potential as well as for a general potential which goes to infinity as $`|x|\mathrm{}`$ . In both of those last references it has been concluded that there is confinement if the scalar coupling is of sufficient intensity compared to the vector coupling.
Since the pure vector Coulomb potential also frustrates the existence of bound-state solutions in the Klein-Gordon (KG) equation , the present work proposes to consider a general mixing of vector and scalar Lorentz structures. Such as in the case of the Dirac equation, this sort of mixing shows to be a powerful tool to obtain a deeper insight about the nature of the KG equation and its solutions. The problem is mapped into an exactly solvable Sturm-Liouville problem of a Schrödinger-like equation with an effective harmonic oscillator potential. In two very special circumstantes the effective potential becomes a Coulomb potential and a sort of phase transition shows up resulting in a spectrum only consisting exclusively of either particle-energy levels or antiparticle-energy levels.
## 2 The Klein-Gordon equation with mixed vector-scalar potentials in a 1+1 dimension
In the presence of vector and scalar potentials the 1+1 dimensional time-independent KG equation for a spinless particle of rest mass $`m`$ reads
$$\mathrm{}^2c^2\frac{d^2\psi }{dx^2}+\left(mc^2+V_s\right)^2\psi =\left(EV_v\right)^2\psi $$
(1)
where $`E`$ is the energy of the particle, $`c`$ is the velocity of light and $`\mathrm{}`$ is the Planck constant. The vector and scalar potentials are given by $`V_v`$ and $`V_s`$, respectively. The subscripts for the terms of potential denote their properties under a Lorentz transformation: $`v`$ for the time component of the 2-vector potential and $`s`$ for the scalar term. It is worth to note that the KG equation is covariant under $`xx`$ if $`V_v(x)`$ and $`V_s(x)`$ remain the same. Also note that $`\psi `$ remains invariant under the simultaneous transformations $`EE`$ and $`V_vV_v`$. Furthermore, for $`V_v=0`$, the case of a pure scalar potential, the negative- and positive-energy levels are disposed symmetrically about $`E=0`$.
The KG equation can also be written as
$$H_{eff}\psi =\frac{\mathrm{}^2}{2m}\psi ^{\prime \prime }+V_{eff}\psi =E_{eff}\psi $$
(2)
where
$$E_{eff}=\frac{E^2m^2c^4}{2mc^2},V_{eff}=\frac{V_s^2V_v^2}{2mc^2}+V_s+\frac{E}{mc^2}V_v$$
(3)
From this one can see that for potentials which tend to $`\pm \mathrm{}`$ as $`|x|\mathrm{}`$ the KG equation furnishes a purely discrete spectrum for $`|V_s|>|V_v|`$ , or for $`|V_s|=|V_v|`$ and $`V_s+V_vE/(mc^2)>0`$. On the other hand, if the potentials vanish as $`|x|\mathrm{}`$ the continuum spectrum is omnipresent but the necessary conditions for the existence of a discrete spectrum is not an easy task for general functional forms. The boundary conditions on the eigenfunctions come into existence by demanding that the effective Hamiltonian given (2) is Hermitian, viz.
$$_a^b𝑑x\psi _n^{}\left(H_{eff}\psi _n^{^{}}\right)=_a^b𝑑x\left(H_{eff}\psi _n\right)^{}\psi _n^{^{}}$$
(4)
where $`\psi _n`$ is an eigenfunction corresponding to an effective eigenvalue $`\left(E_{eff}\right)_n`$ and $`(a,b)`$ is the interval under consideration. In passing, note that a necessary consequence of Eq. (4) is that the eigenfunctions corresponding to distinct effective eigenvalues are orthogonal. It can be shown that (4) is equivalent to
$$\left[\psi _n^{}\frac{d\psi _n^{^{}}}{dx}\frac{d\psi _n^{}}{dx}\psi _n^{^{}}\right]_{x=a}^{x=b}=0$$
(5)
In the nonrelativistic approximation (potential energies small compared to $`mc^2`$ and $`Emc^2`$) Eq. (1) becomes
$$\left(\frac{\mathrm{}^2}{2m}\frac{d^2}{dx^2}+V_v+V_s\right)\psi =\left(Emc^2\right)\psi $$
(6)
so that $`\psi `$ obeys the Schrödinger equation with binding energy equal to $`Emc^2`$ without distinguishing the contributions of vector and scalar potentials.
It is remarkable that the KG equation with a scalar potential, or a vector potential contaminated with some scalar coupling, is not invariant under $`VV+const.`$, this is so because only the vector potential couples to the positive-energies in the same way it couples to the negative-ones, whereas the scalar potential couples to the mass of the particle. Therefore, if there is any scalar coupling the absolute values of the energy will have physical significance and the freedom to choose a zero-energy will be lost. It is well known that a confining potential in the nonrelativistic approach is not confining in the relativistic approach when it is considered as a Lorentz vector. It is surprising that relativistic confining potentials may result in nonconfinement in the nonrelativistic approach. This last phenomenon is a consequence of the fact that vector and scalar potentials couple differently in the KG equation whereas there is no such distinction among them in the Schrödinger equation. This observation permit us to conclude that even a “repulsive” potential can be a confining potential. The case $`V_v=V_s`$ presents bounded solutions in the relativistic approach, although it reduces to the free-particle problem in the nonrelativistic limit. The attractive vector potential for a particle is, of course, repulsive for its corresponding antiparticle, and vice versa. However, the attractive (repulsive) scalar potential for particles is also attractive (repulsive) for antiparticles. For $`V_v=V_s`$ and an attractive vector potential for particles, the scalar potential is counterbalanced by the vector potential for antiparticles as long as the scalar potential is attractive and the vector potential is repulsive. As a consequence there is no bounded solution for antiparticles. For $`V_v=0`$ and a pure scalar attractive potential, one finds energy levels for particles and antiparticles arranged symmetrically about $`E=0`$. For $`V_v=V_s`$ and a repulsive vector potential for particles, the scalar and the vector potentials are attractive for antiparticles but their effects are counterbalanced for particles. Thus, recurring to this simple standpoint one can anticipate in the mind that there is no bound-state solution for particles in this last case of mixing. For short, when $`V_v=\pm V_s`$ the spectrum only consists of energy levels either for particles or for antiparticles.
## 3 The mixed vector-scalar Coulomb potential
Now let us focus our attention on scalar and vector potentials in the form
$$V_s=g_s|x|,V_v=g_v|x|$$
(7)
where the coupling constants, $`g_s`$ and $`g_v`$, are real parameters. In this case the second equation of (3) transmutes into
$$V_{eff}=\frac{1}{2}Ax^2+B|x|$$
(8)
where
$$A=\frac{g_s^2g_v^2}{mc^2},B=g_s+\frac{E}{mc^2}g_v$$
(9)
Therefore, one has to search for bounded solutions in an effective shifted harmonic oscillator potential for $`g_s^2g_v^2`$, or Coulomb potential for $`g_s^2=g_v^2`$. The KG eigenvalues are obtained by inserting the effective eigenvalues into the first equation of (3). Since the effective potential is even under $`xx`$, the KG eigenfunction can be expressed as a function of definite parity. Thus, we can concentrate our attention on the positive half-line and impose boundary conditions on $`\psi `$ at $`x=0`$ and $`x=+\mathrm{}`$. From (5) one can see that in addition to $`\psi \left(\mathrm{}\right)=0`$, the boundary conditions at the origin can be met in two distinct ways: odd functions obey the Neumann condition ($`d\psi /dx|_{x=0}=0`$) whereas even functions obey the Dirichlet condition ($`\psi \left(0\right)=0`$) .
Now we move to consider a quantitative treatment of our problem by considering the two distinct classes of effective potentials.
### 3.1 The effective shifted harmonic oscillator potential ($`g_s^2g_v^2`$)
For this class, the existence of bound-state solutions requires $`|g_s|>|g_v|`$. Included into this class is the case of a pure scalar coupling, but the case of a pure vector, though, is naturally excluded. If $`g_s<g_v`$ the theory is essentially relativistic. Let us define
$$y=|x|+B/A$$
(10)
so that (2)-(3) transmute into
$$\frac{\mathrm{}^2}{2m}\frac{d^2\psi }{dy^2}+\frac{1}{2}Ay^2\psi =\left(E_{eff}+\frac{B^2}{2A}\right)\psi $$
(11)
whose square-integrable solution satisfying the boundary conditions at $`x=0`$ is given by
$`\psi (y)`$ $`=`$ $`N_n\mathrm{exp}\left(y^2/2\right)H_n\left(y\right),n=0,1,2,\mathrm{}`$
$`E_{eff}`$ $`=`$ $`\left(n+{\displaystyle \frac{1}{2}}\right)\mathrm{}\sqrt{{\displaystyle \frac{A}{m}}}{\displaystyle \frac{B^2}{2A}}`$
where $`N_n`$ is a normalization constant and $`H_n\left(y\right)`$ is a Hermite polynomial. It follows that the KG eigenenergies are the solutions of a second-degree algebraic equation:
$$E=mc^2\frac{g_v}{g_s}\pm \frac{\left(g_s^2g_v^2\right)^{3/4}}{g_s}\sqrt{\left(2n+1\right)\mathrm{}c}$$
(13)
There is an infinite sequence of allowed KG eigenenergies with alternate parities to each sign in (13). This result clearly shows that $`g_vg_v`$ makes $`EE`$ without interfering in $`\psi `$, as stated earlier. The change $`g_sg_s`$ also makes $`EE`$ but $`\psi `$ suffers a diverse sort of translation along the $`x`$-axis because $`B`$ also changes sign. Eq. (13) also shows that the KG eigenenergies are symmetrically disposed about $`E=0`$ in the case of a pure scalar coupling and tends to be so as $`g_s/|g_v|\mathrm{}`$.
Fig. 1 (Fig. 2) illustrates the three lowest energy levels for this class of effective potential as a function of $`g_v/g_s`$ ($`g_s/g_v`$). These figures show that both sorts of energy levels, for particles and antiparticles, are always present. It is evident from Fig. 1 that the bounded solutions for particles (antiparticles) are restricted to $`E>mc^2`$ ($`E<+mc^2`$). On the other hand, from Fig. 2 one can see that the bounded solutions for particles (antiparticles) are restricted to $`E>mc^2`$ ($`E<mc^2`$) for $`g_s>0`$. Despite the particle and antiparticle energy levels share the same energy in the spectral gap between $`mc^2`$ and $`+mc^2`$ for the case $`g_v/|g_s|<1`$, there is no crossing of levels.
### 3.2 The effective Coulomb potential ($`g_s^2=g_v^2`$)
For this class of effective potential $`A=0`$ and the discrete spectrum arises when $`B>0`$. These restrictions lead to the constraint relation
$$E\mathrm{sgn}(g_v)+mc^2\mathrm{sgn}(g_s)>0$$
(14)
Defining
$`z`$ $`=`$ $`a|x|+b`$
$`a`$ $`=`$ $`\left({\displaystyle \frac{2mB}{\mathrm{}^2}}\right)^{1/3},b=E_{eff}{\displaystyle \frac{a}{B}}`$
Eq. (2) turns into the Airy differential equation
$$\frac{d^2\psi }{dz^2}z\psi =0$$
(16)
which has square-integrable solutions expressed in terms of the Airy functions : $`\psi (z)=N\mathrm{A}_\mathrm{i}(z)`$, where $`N`$is a normalization constant. The boundary conditions at $`x=0`$ lead to the quantization conditions
$`\mathrm{A}_\mathrm{i}(b)`$ $`=`$ $`0,\mathrm{for}\mathrm{odd}\mathrm{parity}\mathrm{solutions}`$
$`\mathrm{A}_\mathrm{i}^{}(b)`$ $`=`$ $`0,\mathrm{for}\mathrm{even}\mathrm{parity}\mathrm{solutions}`$
These quantization conditions have solutions only for $`b<0`$ and a number of them is listed are Table I . One can see that demanding $`b<0`$ implies into an additional restriction on the KG eigenenergies: $`|E|>mc^2`$. Substitution of the roots of the Airy function and its first derivative into (LABEL:eq19a) allow us to obtain the energies as the solutions of a sixth-degree algebraic equation:
$$E^63m^2c^4E^4+\left(3m^4c^64|b|^3\mathrm{}^2g_s^2\right)c^2E^2$$
$$8|b|^3\mathrm{}^2mc^4g_s^2\mathrm{sgn}(g_sg_v)E\left(4|b|^3\mathrm{}^2g_s^2+m^4c^6\right)m^2c^6=0$$
(18)
Since there are no explicit solutions to this algebraic equation in terms of radicals, we satisfy ourselves verifying that the roots of Eq. (18) always satisfy the requirement $`|E|>mc^2`$ by using the Descartes´ rule of signs (henceforth DRS). The DRS states that an algebraic equation with real coefficients $`a_k\lambda ^k+\mathrm{}+a_1\lambda +a_0=0`$ the difference between the number of changes of signs in the sequence $`a_k,\mathrm{},a_1,a_0`$ and the number of positive real roots is an even number or zero, with a root of multiplicity $`k`$ counted as $`k`$ roots and not counting the null coefficients (see, e.g., ). The verification of the existence of solutions for $`E>mc^2`$ is made simpler if we write $`E=mc^2+\epsilon `$. We get
$$\epsilon ^6+6mc^2\epsilon ^5+12m^2c^4\epsilon ^4+8m^3c^6\epsilon ^34|b|^3\mathrm{}^2c^2g_s^2\epsilon ^2$$
$$8|b|^3\mathrm{}^2mc^4g_s^2\left[1+\mathrm{sgn}(g_sg_v)\right]\epsilon 8|b|^3\mathrm{}^2m^2c^6g_s^2\mathrm{sgn}(g_sg_v)=0$$
(19)
Observing the difference of signs among the coefficients of the leading coefficient and the lowest degree it becomes clear that there exist positive roots for $`g_v=g_s`$. In this particular case the DRS assures that there exists just one solution, since there is only one change of sign in the sequence of coefficients of (19). For $`g_v=g_s`$, though, the DRS implies that there exist zero or two positive roots. It is interesting to note that this result is true whatever the fermion mass and the coupling constants. The very same conclusion for $`E<mc^2`$ can be obtained by observing that $`g_vg_v`$ by the change $`EE`$. However, this is not the whole story because the KG eigenenergies with $`|E|>mc^2`$ are not only given by the solutions of Eq. (18). As a matter of fact, they have also to satisfy the constraint expressed by Eq. (14). As a result, there are only KG eigenenergies with $`E>mc^2`$ ($`E<mc^2`$) for $`g_v=|g_s|`$ ($`g_v=|g_s|`$). This conclusion confirms what has already been mentioned in the last paragraph of the Sec. 2: the spectrum contains either particle-energy levels or antiparticle-energy levels. Furthermore, it is not difficult to see that the spectrum for the case $`g_v=g_s`$ can be obtained from that for $`g_v=g_s`$ by changing $`E`$ by $`E`$.
## 4 Conclusions
In summary, we have succeed in the proposal for searching bounded solutions of the KG equation with a mixed vector-scalar Coulomb potential. An opportunity was given to analyze some aspects of the KG equation which have not been approached in the literature yet.
In general, there exist bounded solutions for particles and antiparticles. Nevertheless, for $`|g_s|=|g_v|`$ there are bounded solutions either for particles or antiparticles. This is a clear manifestation of the phase transition which occurs when $`|g_s|=|g_v|`$. The solutions of the KG equation with a Coulomb potential present a continuous transition as the ratio $`g_s/g_v`$ (or $`g_v/g_s`$) varies. However, when $`|g_s|=|g_v|`$the phase transition shows its face not only for the eigenenergy but also for the eigenfunction.
Finally, we draw attention to the fact that no matter how strong the potentials are, as far as $`|g_s||g_v|`$, the energy levels for particles (antiparticles) never dive into the lower (upper) continuum. Thus there is no room for the production of particle-antiparticle pairs. This all means that Klein´s paradox never comes to the scenario.
Acknowledgments
This work was supported in part by means of funds provided by CNPq and FAPESP. |
warning/0507/math0507176.html | ar5iv | text | # On classification of projective homogeneous varieties up to motivic isomorphism
## 1 Introduction
The present paper can be viewed as an application of the methods and results obtained by the authors in \[CPSZ\].
Let $`k`$ be a field of characteristic not $`2`$ and $`k_s`$ denotes its separable closure. For a variety $`X`$ over $`k`$ we denote by $`X_s`$ the base change $`X\times _kk_s`$. By $`(X)`$ we denote the Chow motive of $`X`$. Recall (see \[MPW, § 1\]) that $`X`$ is a twisted flag variety of inner type over $`k`$ if $`X={}_{\xi }{}^{}(G/P)`$ is a twisted form of the projective homogeneous variety $`G/P`$, where $`G`$ is an adjoint simple split algebraic group over $`k`$, $`P`$ its parabolic subgroup and the twisting is given by a $`1`$-cocycle $`\xi Z^1(k,G(k_s))`$.
The present paper is devoted to the following
###### Problem.
Describe all pairs $`(X,Y)`$ of non-isomorphic twisted flag varieties $`X`$ and $`Y`$ of inner type over $`k`$ which have isomorphic Chow motives.
This problem can be subdivided into two subproblems:
Describe all such pairs $`(X,Y)`$ with $`X_sY_s`$;
Describe all such pairs $`(X,Y)`$ with $`X_s\simeq ̸Y_s`$.
Let us briefly remind what is known so far. The complete solution of the problem (i) is known for quadrics and Severi-Brauer varieties due to Izhboldin, Karpenko, Merkurjev, Rost, Vishik and others (see \[Izh98\], \[Ka96\], \[Ka00\], \[Ro98\], \[Vi03\]). Concerning (ii), the example (of dimension 5) was provided by Bonnet in \[Bo03\]. It deals with twisted flag varieties of type $`\mathrm{G}_2`$. For exceptional varieties of type $`\mathrm{F}_4`$ a similar example was provided in \[NSZ\].
In the present paper we provide a complete solution of the mentioned above problem for projective homogeneous varieties of dimension less than $`6`$. Namely, we prove the following (using the notation of 2.1)
###### 1.1 Theorem.
Let $`X`$ and $`Y`$ be non-isomorphic twisted flag varieties of dimension $`5`$ of inner type over $`k`$ which have isomorphic Chow motives.
If $`X_sY_s`$, then either
$`X=\mathrm{SB}(A)`$ and $`Y=\mathrm{SB}(A^{\mathrm{op}})`$ are Severi-Brauer varieties corresponding to a central simple algebra $`A`$ and its opposite $`A^{\mathrm{op}}`$, where $`\mathrm{deg}(A)=3,4,5,6`$ and $`\mathrm{exp}(A)>2`$
or
$`X=\mathrm{SB}_{1,2}(A)`$ and $`Y=\mathrm{SB}_{1,2}(A^{\mathrm{op}})`$ are twisted forms of the flag varieties corresponding to central simple algebras $`A`$ such that $`\mathrm{deg}(A)=\mathrm{exp}(A)=4`$.
If $`X_s\simeq ̸Y_s`$, then either
$`X=\mathrm{SB}_{1,3}(A)`$ and $`Y=\mathrm{SB}_{1,2}(A^{})`$ are twisted forms of the flag varieties corresponding to central simple algebras $`A`$ and $`A^{}`$ such that $`\mathrm{deg}(A)=\mathrm{deg}(A^{})=4`$ and $`AA^{}`$ or $`A^{\mathrm{op}}`$,
or
$`X={}_{\xi }{}^{}(\mathrm{G}_2/P_1)`$ and $`Y={}_{\xi }{}^{}(\mathrm{G}_2/P_2)`$ are twisted forms of the variety $`G/P_i`$, $`i=1`$, $`2`$, where $`G`$ is a split exceptional group of type $`\mathrm{G}_2`$ and $`P_i`$ is one of its maximal parabolic subgroups,
or
$`X=^n`$ and $`Y=Q^n`$ is the projective space and the split quadric respectively, where $`n=3,5`$.
or
$`X=^5`$ and $`Y=\mathrm{G}_2/P_2`$ is the projective space and the split Fano variety of type $`\mathrm{G}_2`$.
###### 1.2 Remark.
Observe that the case $`X={}_{\xi }{}^{}(\mathrm{G}_2/P_1)`$ and $`Y={}_{\xi }{}^{}(\mathrm{G}_2/P_2)`$ of the theorem is the example of Bonnet mentioned above and, hence, is the minimal one in the sense of dimension.
###### 1.3 Remark.
The case $`X=\mathrm{SB}_{1,3}(A)`$ and $`Y=\mathrm{SB}_{2,3}(A^{})`$ with $`AA^{},A^{\mathrm{op}}`$ provides another minimal example of two non-isomorphic varieties that have isomorphic Chow motives.
Apart from Theorem 1.1, we prove the following
###### 1.4 Theorem.
Let $`X=\mathrm{SB}_{n_1,\mathrm{},n_r}(A)`$ and $`Y=\mathrm{SB}_{m_1,\mathrm{},m_r}(A^{})`$ be twisted flag varieties of inner type $`\mathrm{A}_n`$, $`n2`$, over $`k`$, where the central simple algebras $`A`$ and $`A^{}`$ have exponents $`1`$, $`2`$, $`3`$, $`4`$, or $`6`$. Assume that
1. $`(X_s)(Y_s)`$;
2. $`n_1=1`$ or $`n_r=n`$
3. $`m_1=1`$ or $`m_r=n`$.
Then $`(X)(Y)AA^{}\text{ or }A^{\mathrm{op}}`$.
As an immediate consequence we obtain
###### 1.5 Corollary.
Let $`X=\mathrm{SB}_{1,n}(A)`$ and $`Y=\mathrm{SB}_{n1,n}(A^{})`$, where $`A`$ and $`A^{}`$ are central simple algebras of degree $`n+1`$, $`n3`$, and exponent 1,2,3,4 or 6. Then
$$(X)(Y)AA^{}\text{ or }A^{\mathrm{op}}.$$
###### 1.6 Remark.
The varieties $`X`$ and $`Y`$ of 1.5 provide examples of twisted flag varieties that satisfy (b). In fact, $`X_s\simeq ̸Y_s`$, since they have different automorphism groups by \[De77\].
The paper is organized as follows. In section 2 we consider the case of a split group and state several facts which will be extensively used in the proofs. Section 3 is devoted to the case by case proof of Theorem 1.1. In the section 4 we prove Theorem 1.4 and provide several results that we need for the proof of 1.1.
## 2 Preliminaries
In the paper we use the following notation.
###### 2.1.
Let $`G`$ be a split simple algebraic group defined over a field $`k`$. We fix a maximal split torus $`T`$ of $`G`$ and a Borel subgroup $`B`$ of $`G`$ containing $`T`$ and defined over $`k`$. Denote by $`\mathrm{\Phi }`$ the root system of $`G`$, by $`\mathrm{\Pi }=\{\alpha _1,\mathrm{},\alpha _{\mathrm{rk}G}\}`$ the set of simple roots of $`\mathrm{\Phi }`$ corresponding to $`B`$, by $`W`$ the Weyl group, and by $`S=\{s_1,\mathrm{},s_{\mathrm{rk}G}\}`$ the corresponding set of fundamental reflections. Let $`P_\mathrm{\Theta }`$ be the standard parabolic subgroup corresponding to a subset $`\mathrm{\Theta }\mathrm{\Pi }`$, i.e., $`P_\mathrm{\Theta }=BW_\mathrm{\Theta }B`$, where $`W_\mathrm{\Theta }=s_\theta ,\theta \mathrm{\Theta }`$. Denote by $`P_i`$ the maximal parabolic subgroup $`P_{\mathrm{\Pi }\{\alpha _i\}}`$. By $`\mathrm{\Phi }/P_\mathrm{\Theta }`$ we denote the flag variety $`G/P_\mathrm{\Theta }`$. The root enumeration follows Bourbaki.
The notation $`\mathrm{SB}_{n_1,\mathrm{},n_r}(A)`$, $`1n_1<\mathrm{}<n_rn`$, is used for the twisted form of the variety $`\mathrm{A}_n/P_\mathrm{\Theta }`$, where $`\mathrm{\Theta }=\mathrm{\Pi }\{\alpha _{n_1},\mathrm{},\alpha _{n_r}\}`$ and $`A`$ is a central simple algebra of degree $`n+1`$ corresponding to the twisting. Observe that $`\mathrm{SB}_{n_1,\mathrm{},n_r}(A)=X(A;n_1,\mathrm{},n_r)`$ in the notation of \[MPW, Appendix\] and $`\mathrm{SB}(A)=\mathrm{SB}_1(A)`$ is the usual Severi-Brauer variety defined by $`A`$. By $`\mathrm{ind}(A)`$ we denote the index of $`A`$ and by $`\mathrm{exp}(A)`$ its exponent. A split projective quadric of dimension $`n`$ is denoted by $`Q^n`$.
###### 2.2.
According to \[Ko91\] the Chow motive of the flag variety $`X=G/P_\mathrm{\Theta }`$, when $`G`$ is a split group, is isomorphic to
$$(X)\underset{i=0}{\overset{dimX}{}}(i)^{a_i(X)},$$
where $`(i)`$ are the twists of the Lefschetz motive and the positive integers (ranks) $`a_i(X)`$ are the coefficients of the generating polynomial $`p_X(z)=_{i=0}^{dimX}a_i(X)z^i`$. The latter is defined by the following explicit formula:
$$p_X(z)=(\underset{i=1}{\overset{\mathrm{rk}G}{}}\frac{z^{d_i(W)}1}{z1})/(\underset{j=1}{\overset{m}{}}\underset{i}{}\frac{z^{d_i(W_j)}1}{z1}).$$
Here $`W_1\times \mathrm{}\times W_m`$ is the decomposition of $`W_\mathrm{\Theta }`$ into the product of Weyl groups corresponding to the irreducible root systems and $`d_i(W_j)`$ are the degrees of the respective fundamental polynomial invariants (see \[Ca72, 9.4 A\]).
The following observation follows from the above isomorphism.
###### 2.3.
The motives of flag varieties $`X`$ and $`Y`$ of dimension $`n`$ over a separably closed field are isomorphic iff the corresponding sequences of ranks $`(a_0(X),\mathrm{},a_n(X))`$ and $`(a_0(Y),\mathrm{},a_n(Y))`$ are equal.
We shall need the following two facts:
###### 2.4.
(See \[Ka00, Criterion 7.1\]) Let $`A`$, $`A^{}`$ be central simple algebras over $`k`$ and $`\mathrm{SB}(A)`$, $`\mathrm{SB}(A^{})`$ be the respective Severi-Brauer varieties. Then
$$(\mathrm{SB}(A))(\mathrm{SB}(A^{}))AA^{},A^{\mathrm{op}}.$$
###### 2.5.
(see \[Izh98, Cor. 2.9 and Prop. 3.1\]) Let $`q`$, $`q^{}`$ be regular quadratic forms of rank $`n`$ and $`X_q`$, $`X_q^{}`$ be the respective projective quadrics. If $`n`$ is odd or $`n<7`$, then
$$(X_q)(X_q^{})X_qX_q^{}.$$
Finally, we shall need the following observation:
###### 2.6.
(See \[Ka00, Proof of Lemma 2.3\]) Let $`X`$ and $`Y`$ be smooth projective varieties over $`k`$ with isomorphic Chow motives. Then there is an isomorphism of abelian groups
$$\mathrm{Coker}(\mathrm{CH}_0(X)\stackrel{res}{}\mathrm{CH}_0(X_s))\mathrm{Coker}(\mathrm{CH}_0(Y)\stackrel{res}{}\mathrm{CH}_0(Y_s)).$$
## 3 Small dimensions
In this section we classify all pairs $`(X,Y)`$ of non-isomorphic twisted flag varieties of inner type over $`k`$ of dimension $`5`$ which have isomorphic Chow motives and hence prove Theorem 1.1.
#### Dimension $`1`$.
Twisted flag varieties of dimension $`1`$ are the twisted forms of the projective line $`^1`$. The twisted forms of $`^1`$ are Severi-Brauer varieties $`\mathrm{SB}(H)`$, where $`H`$ is a quaternion algebra. By 2.4
$$(\mathrm{SB}(H))(\mathrm{SB}(H^{}))HH^{},H^{\mathrm{op}}$$
Since $`HH^{\mathrm{op}}`$, we conclude that the motives are isomorphic iff the varieties are isomorphic.
#### Dimension $`2`$.
All twisted flags of dimension $`2`$ are the twisted forms of the projective space $`^2`$ or the split quadric surface $`Q^2^1\times ^1`$. Observe that $`Q^2`$ is a projective homogeneous variety for a group of type $`\mathrm{D}_2`$ which is not simple, but semisimple. Nevertheless, we shall consider this case too.
The motives of $`^2`$ and $`Q^2`$ are not isomorphic, since the respective sequences of ranks $`(1,1,1)`$ and $`(1,2,1)`$ are different.
The twisted forms of $`Q^2`$ of inner type over $`k`$ are $`2`$-dimensional quadrics (see \[Inv, Cor. (15.12)\]). By 2.5 the motives of two quadrics of dimension $`2`$ are isomorphic iff the quadrics are isomorphic.
The twisted forms of $`^2`$ are Severi-Brauer varieties $`\mathrm{SB}(A)`$, where $`A`$ is a central simple algebra of degree $`3`$. Again by 2.4 we have
$$(\mathrm{SB}(A))(\mathrm{SB}(A^{}))AA^{},A^{\mathrm{op}}.$$
Since the varieties $`\mathrm{SB}(A)`$ and $`\mathrm{SB}(A^{\mathrm{op}})`$ are isomorphic iff $`A`$ is split, we conclude that all pairs of non-isomorphic varieties which have isomorphic motives are of the kind $`(\mathrm{SB}(A),\mathrm{SB}(A^{\mathrm{op}}))`$, where $`A`$ is a division algebra of degree $`3`$.
#### Dimension $`3`$.
Computing generating functions (see 2.2) we conclude that there are only three projective homogeneous varieties of dimension $`3`$ over $`k_s`$. Namely, the projective space $`^3`$, the quadric $`Q^3`$ and the variety of complete flags $`\mathrm{A}_2/B`$ ($`B`$ denotes a Borel subgroup). The respective sequences of ranks look as follows:
| $`^3\mathrm{A}_3/P_1`$ | : | $`(1,1,1,1)`$ |
| --- | --- | --- |
| $`Q^3\mathrm{B}_2/P_1`$ | : | $`(1,1,1,1)`$ |
| $`\mathrm{A}_2/B`$ | : | $`(1,2,2,1)`$ |
In particular, we see that the motives of $`^3`$ and $`Q^3`$ are isomorphic but the motives of $`Q^3`$ and $`\mathrm{A}_2/B`$ are not.
By 2.4 all non-isomorphic twisted forms of $`^3`$ which have isomorphic motives form pairs $`(\mathrm{SB}(A),\mathrm{SB}(A^{\mathrm{op}}))`$, where $`A`$ is a division algebra of degree 4 and exponent 4. Observe that all non-isomorphic twisted forms of $`Q^3`$ are quadrics as well and by 2.5 the motive of a quadric determines this quadric uniquely. Therefore it remains to describe all possible motivic isomorphisms between the twisted forms $`{}_{\xi }{}^{}_{}^{3}`$ and $`{}_{\zeta }{}^{}Q_{}^{3}`$ and the twisted forms $`{}_{\xi }{}^{}(\mathrm{A}_2/B)`$ and $`{}_{\zeta }{}^{}(\mathrm{A}_2/B)`$ of the variety of complete flags $`\mathrm{A}_2/B`$.
According to Corollary 4.4 there are no non-isomorphic twisted forms of $`\mathrm{A}_2/B`$ which have isomorphic Chow motives. And the next lemma shows that there are no such (non-trivial) twisted forms of $`^3`$ and $`Q^3`$.
###### 3.1 Lemma.
Let $`\xi `$, $`\zeta `$ be 1-cocycles. Then $`({}_{\xi }{}^{}_{}^{3})({}_{\zeta }{}^{}Q_{}^{3})`$ iff $`\xi `$ and $`\zeta `$ are trivial.
###### Proof.
This is a particular case of a more general result (see Lemma 4.2) proven using Index Reduction Formula. Here we give an elementary proof. It uses only well-known facts about quadrics and Severi-Brauer varieties.
Observe that any twisted form of $`^3`$ is a Severi-Brauer variety $`\mathrm{SB}(A)`$ for some central simple algebra $`A`$ of degree $`4`$ and any twisted form of $`Q^3`$ is a non-singular quadric of dimension $`3`$.
As in 2.6 for a variety $`X`$ consider the abelian group $`\mathrm{Coker}(\mathrm{CH}_0(X)\mathrm{CH}_0(X_s))`$. If $`X=\mathrm{SB}(A)`$ is a Severi-Brauer variety of a central simple algebra $`A`$, then this cokernel is equal to $`/\mathrm{ind}(A)`$ (see \[Ka00\]), where $`\mathrm{ind}(A)`$ is the index of $`A`$. In particular, this cokernel is trivial iff $`A`$ is split. If $`X`$ is a quadric then this cokernel is trivial iff $`X`$ is isotropic (see \[Sw89\]). In the case $`X`$ is an anisotropic quadric this cokernel is isomorphic to $`/2`$.
In our case we have two varieties $`X=\mathrm{SB}(A)`$ and $`Y={}_{\zeta }{}^{}Q_{}^{3}`$ which have isomorphic motives. Hence, by 2.6 the respective cokernels must be isomorphic.
Hence, if the quadric $`Y`$ is isotropic, then the algebra $`A`$ is split. The latter implies that the motive $`(\mathrm{SB}(A))`$ splits into the direct sum of Lefschetz motives and so is $`(Y)`$, i.e., $`Y`$ is split as well by 2.5.
Assume $`q`$ is anisotropic, then there exists a quadratic field extension $`l/k`$ such that the Witt index of $`Y_l=Y\times _kl`$ is one (see \[Vi03, §7.2\]). Since the motives of $`X`$ and $`Y`$ are still isomorphic over $`l`$, we conclude that $`A`$ is split over $`l`$. Then $`Y_l`$ is split as well. This leads to a contradiction. ∎
###### 3.2 Remark.
Observe that the pair of twisted forms $`({}_{\xi }{}^{}(\mathrm{B}_2/P_1),{}_{\xi }{}^{}(\mathrm{B}_2/P_2))`$ can be viewed as a low-dimensional analog of the pair $`({}_{\xi }{}^{}(\mathrm{G}_2/P_1),{}_{\xi }{}^{}(\mathrm{G}_2/P_2))`$ considered by Bonnet. The lemma says that contrary to $`\mathrm{G}_2`$-case the motives of $`{}_{\xi }{}^{}(\mathrm{B}_2/P_1)`$ and $`{}_{\xi }{}^{}(\mathrm{B}_2/P_2)`$ are not isomorphic (if $`\xi `$ is non-trivial).
#### Dimension 4.
There are three non-isomorphic projective homogeneous varieties of dimension $`4`$ over $`k_s`$. Namely, the projective space $`^4`$, the 4-dimensional quadric $`Q^4\mathrm{Gr}(2,4)`$ and the variety of complete flags $`\mathrm{B}_2/B`$. The respective sequences of ranks in these cases are all different and look as follows:
| $`^4\mathrm{A}_4/P_1`$ | : | $`(1,1,1,1,1)`$ |
| --- | --- | --- |
| $`Q^4\mathrm{A}_3/P_2`$ | : | $`(1,1,2,1,1)`$ |
| $`\mathrm{B}_2/B`$ | : | $`(1,2,2,2,1)`$ |
Hence, the motives of $`^4`$, $`Q^4`$ and $`\mathrm{B}_2/B`$ are non-isomorphic to each other.
By 2.4 all non-isomorphic twisted forms of $`^4`$ which have isomorphic motives form pairs $`(\mathrm{SB}(A),\mathrm{SB}(A^{\mathrm{op}}))`$, where $`A`$ is a division algebra of degree 5. By Corollary 4.5 there are no non-isomorphic twisted forms of $`\mathrm{B}_2/B`$ which have isomorphic Chow motives. Therefore the only case left is the case of inner twisted forms of $`Q^4`$.
The inner forms of $`Q^4`$ are the generalized Severi-Brauer varieties $`\mathrm{SB}_2(A)`$, where $`A`$ is a central simple algebra of degree $`4`$. The next lemma shows that there are no non-isomorphic forms of $`\mathrm{SB}_2(A)`$ which have isomorphic motives.
###### 3.3 Lemma.
Let $`A`$, $`A^{}`$ be central simple algebras of degree $`4`$. Then
$$(\mathrm{SB}_2(A))(\mathrm{SB}_2(A^{}))\mathrm{SB}_2(A)\mathrm{SB}_2(A^{})$$
###### Proof.
Let $`(\mathrm{SB}_2(A))(\mathrm{SB}_2(A^{}))`$. It suffices to prove that for all field extensions $`l/k`$ one has $`\mathrm{ind}(A_l)=\mathrm{ind}(A_l^{})`$. Indeed, by \[Ka00, Lemma 7.13\] $`A=A^{}`$ in $`\mathrm{Br}(k)`$, hence, $`AA^{}`$ or $`A^{\mathrm{op}}`$. But $`\mathrm{SB}_2(A)\mathrm{SB}_2(A^{\mathrm{op}})`$ for any central simple algebra $`A`$ of degree 4.
Assume that there exists a field extension $`l/k`$ such that $`\mathrm{ind}(A_l)\mathrm{ind}(A_l^{})`$. Depending on the indices of $`A`$ and $`A^{}`$ we distinguish the following cases:
##### Case 1.
$`\mathrm{ind}(A)=4`$ and $`\mathrm{ind}(A^{})=1`$ or $`2`$.
In this case $`\mathrm{SB}_2(A^{})`$ has a rational point. By \[Inv, Case $`\mathrm{A}_3=\mathrm{D}_3`$\], the variety $`\mathrm{SB}_2(A^{})`$ is isotropic, hence, the group
$$\mathrm{Coker}(\mathrm{CH}_0(\mathrm{SB}_2(A^{}))\mathrm{CH}_0(\mathrm{SB}_2(A_{k_s}^{}))$$
is trivial. By 2.6 the cokernel
$$\mathrm{Coker}(\mathrm{CH}_0(\mathrm{SB}_2(A))\mathrm{CH}_0(\mathrm{SB}_2(A_{k_s}))$$
must be trivial as well. If $`\mathrm{exp}(A)=2`$, then $`A`$ is a biquaternion algebra and by \[Inv, Cor. (15.33)\] $`\mathrm{SB}_2(A)`$ is an anisotropic quadric. Then the cokernel above must be isomorphic to $`/2`$, a contradiction. If $`\mathrm{exp}(A)=4`$, then by \[Inv, Cor. (15.33)\] $`AC^\pm (B,\sigma ,f)`$, where $`(B,\sigma ,f){}_{}{}^{1}\mathrm{D}_{3}^{}`$ and $`B`$ is a central simple algebra of degree $`6`$ and index $`2`$. By Merkurjev’s theorem (see \[Me95\]) the cokernel above must be again isomorphic to $`/2`$, a contradiction.
##### Case 2.
$`\mathrm{ind}(A)=2`$ and $`\mathrm{ind}(A^{})=1`$.
In this case $`A^{}`$ is split, hence, the corresponding variety is a split quadric. ¿From the other hand, $`\mathrm{SB}_2(A)X_q`$, where $`q`$ is some $`6`$-dimensional quadratic form and $`X_q`$ is the corresponding projective quadric. Using 2.5, we conclude that $`\mathrm{SB}_2(A)\mathrm{SB}_2(A^{})`$, a contradiction. ∎
#### Dimension $`5`$.
There are five non-isomorphic projective homogeneous varieties over $`k_s`$ of dimension $`5`$. Namely, the projective space $`^5`$, the quadric $`Q^5`$, the exceptional Fano variety $`\mathrm{G}_2/P_2`$, the flag varieties $`\mathrm{A}_3/P_{\{\alpha _1\}}`$ and $`\mathrm{A}_3/P_{\{\alpha _2\}}`$. The respective sequences of ranks look as follows:
| $`^5\mathrm{A}_5/P_1`$ | : | $`(1,1,1,1,1,1)`$ |
| --- | --- | --- |
| $`Q^5\mathrm{B}_3/P_1`$ | : | $`(1,1,1,1,1,1)`$ |
| $`\mathrm{G}_2/P_2`$ | : | $`(1,1,1,1,1,1)`$ |
| $`\mathrm{A}_3/P_{\{\alpha _1\}}\mathrm{A}_3/P_{\{\alpha _3\}}`$ | : | $`(1,2,3,3,2,1)`$ |
| $`\mathrm{A}_3/P_{\{\alpha _2\}}`$ | : | $`(1,2,3,3,2,1)`$ |
Therefore, the motives of $`^5`$, $`Q^5`$ and $`\mathrm{G}_2/P_2`$ are isomorphic and the motives of $`\mathrm{A}_3/P_{\{\alpha _1\}}`$ and $`\mathrm{A}_3/P_{\{\alpha _2\}}`$ are isomorphic.
As was mentioned before, the twisted forms of $`^5`$ and $`Q^5`$ were completely classified up to motivic isomorphisms by Karpenko and Izhboldin (see 2.4 and 2.5). Moreover, by Lemma 4.2 there there is only one pair $`({}_{\xi }{}^{}_{}^{5},{}_{\zeta }{}^{}Q_{}^{5})`$ of twisted forms that have isomorphic motives.
By the result of Bonnet \[Bo03\] the motive of the twisted form $`{}_{\xi }{}^{}(\mathrm{G}_2/P_2)`$ is isomorphic to the motive of $`{}_{\xi }{}^{}(\mathrm{G}_2/P_1)`$ which is a $`5`$-dimensional quadric.
By Corollary 1.5 the motives of the twisted forms of $`\mathrm{A}_3/P_{\{\alpha _1\}}`$ and $`\mathrm{A}_3/P_{\{\alpha _2\}}`$ are isomorphic iff the respective central simple algebras of degree $`4`$ are isomorphic or opposite. This provides the last example (see 1.1) of a pair of non-isomorphic varieties of dimension $`5`$ that have isomorphic motives.
## 4 Arbitrary dimensions
In the present section we prove several classification results. We start with the following
###### 4.1 Lemma.
Let $`X`$ and $`Y`$ be twisted flag varieties of inner type over $`k`$ which have isomorphic Chow motives. Assume $`X`$ is not of type $`\mathrm{E}_8`$ and splits over its function field $`k(X)`$, i.e., the group corresponding to $`X`$ splits over $`k(X)`$. Then $`X`$ splits over the function field of $`Y`$.
###### Proof.
Since the motives are isomorphic, there is an isomorphism of cokernels (see 2.6) and, hence, an isomorphism of cokernels over $`k(Y)`$
$$\mathrm{Coker}(\mathrm{CH}_0(X_{k(Y)})\mathrm{CH}_0(X_{k(Y)_s}))\mathrm{Coker}(\mathrm{CH}_0(Y_{k(Y)})\mathrm{CH}_0(Y_{k(Y)_s}))$$
Since $`Y_{k(Y)}`$ is isotropic, the right cokernel is trivial and so is the left one. The fact that the map $`\mathrm{res}:\mathrm{CH}_0(X_{k(Y)})\mathrm{CH}_0(X_{k(Y)_s})`$ is surjective and the group $`\mathrm{CH}_0(X_{k(Y)_s})`$ is a free abelian group of rank one generated by the class of a rational point $`[pt]`$ implies that the preimage $`\mathrm{res}^1([pt])`$ is a $`0`$-cycle of degree $`1`$ in $`\mathrm{CH}_0(X_{k(Y)})`$. Then, the variety $`X_{k(Y)}`$ is isotropic (see \[To04, Q. 0.2\]).
By \[KR94, 3.16.(iii)\] the function field $`k(X)`$ is a generic splitting field for the respective parabolic subgroup $`P`$. Since $`X_{k(Y)}`$ is isotropic, the field $`k(Y)`$ is a $`k`$-specialization of $`k(X)`$ (see \[KR94, Def. 1.2\]). Since $`X`$ splits over $`k(X)`$, $`k(X)`$ is a splitting field for the respective group $`G`$. Then, by \[KR94, 3.9.(iii)\], $`k(Y)`$ is a splitting field of $`G`$ as well, i.e., $`X_{k(Y)}`$ splits. ∎
###### 4.2 Proposition.
Let $`\gamma `$, $`\delta `$ be 1-cocycles and $`X={}_{\gamma }{}^{}_{}^{n}`$, $`Y={}_{\delta }{}^{}Q_{}^{n}`$ be the respective twisted forms for $`n>1`$ odd. Then
$$(X)(Y)\gamma \text{ and }\delta \text{ are trivial.}$$
###### Proof.
Observe that $`X`$ is a Severi-Brauer variety corresponding to a central simple algebra $`A`$ and $`Y`$ is a $`n`$-dimensional quadric.
Assume that $`(X)(Y)`$ and $`\gamma `$ is not trivial. By Lemma 4.1 applied to $`X`$ and $`Y`$, the algebra $`A_{k(Y)}`$ splits, i.e., $`\mathrm{ind}(A_{k(Y)})=1`$. From the other hand by Index Reduction Formula (see \[MPW\]) we obtain
$$\mathrm{ind}(A_{k(Y)})=\mathrm{min}\{\mathrm{ind}(A),2^{(n1)/2}\mathrm{ind}(A_kC_0(q))\}>1,$$
where $`C_0(q)`$ is the even part of the Clifford algebra of the quadric corresponding to $`Y`$. This leads to a contradiction. ∎
Note that the same proof works for twisted forms of types $`\mathrm{B}_n`$ and $`\mathrm{C}_n`$. Namely,
###### 4.3 Proposition.
Let $`\gamma `$, $`\delta `$ be 1-cocycles and $`X={}_{\gamma }{}^{}(\mathrm{C}_n/P_l)`$, $`Y={}_{\delta }{}^{}(\mathrm{B}_n/P_l)`$ be the respective twisted forms for an odd $`1l<n`$. Then
$$(X)(Y)\gamma \text{ and }\delta \text{ are trivial.}$$
###### Proof.
For any simple algebraic group $`G`$ as above consider a twisted flag variety $`W={}_{\xi }{}^{}(G/P_\mathrm{\Theta })`$ over $`k`$. On the Tits diagram (see \[Ti66\]) of $`G`$ over $`k(W)`$ all vertices corresponding to simple roots from $`\mathrm{\Pi }\mathrm{\Theta }`$ are circled.
In our case since $`l`$ is odd, this implies that $`X_{k(X)}`$ is split (see \[Ti66\] for a complete list of Tits diagrams). The rest of the proof repeats the proof of 4.2. ∎
The rest of this section is devoted to the twisted forms of flag varieties. In particular, we obtain the description of motivic isomorphisms for twisted forms of the flag varieties $`\mathrm{A}_2/B`$, $`\mathrm{B}_2/B`$ and $`\mathrm{A}_3/P_{\{\alpha _i\}}`$, $`i=1`$, $`2`$, $`3`$. We start with the proof of Theorem 1.4.
###### Proof of Theorem 1.4.
Assume $`(X)(Y)`$. Since $`X`$ and $`Y`$ are twisted forms of flags containing the subspace of dimension $`1`$ (we may assume $`n_1=m_1=1`$), the motives of $`X`$ and $`Y`$ can be decomposed into a direct sum of twisted motives of Severi-Brauer varieties (see \[CPSZ, Thm. 2.1\]). Namely,
$$(X)\underset{i}{}(\mathrm{SB}(A))(i),(Y)\underset{j}{}(\mathrm{SB}(A^{}))(j).$$
(\*)
This together with 2.6 implies the isomorphism of abelian groups
$$\mathrm{Coker}(\mathrm{CH}_0(\mathrm{SB}(A))\mathrm{CH}_0(^n))\mathrm{Coker}(\mathrm{CH}_0(\mathrm{SB}(A^{}))\mathrm{CH}_0(^n))$$
and, hence, the isomorphism $`/\mathrm{ind}(A)/\mathrm{ind}(A^{})`$, i.e., $`\mathrm{ind}(A)=\mathrm{ind}(A^{})`$. Since the motivic isomorphism is preserved under the base extensions, we obtain that $`\mathrm{ind}(A_l)=\mathrm{ind}(A_l^{})`$ for any finite field extension $`l/k`$. In fact, by \[Ka00, Lemma 7.13\] the latter is equivalent to $`A=A^{}`$ in $`\mathrm{Br}(k)`$. In particular, if $`\mathrm{exp}(A)=\mathrm{exp}(A^{})`$ is $`2,3,4,6`$, we obtain $`AA^{}`$ or $`A^{\mathrm{op}}`$.
In the opposite direction, let $`AA^{}`$ or $`A^{\mathrm{op}}`$. By conditions (i) and (iii) one has two motivic decompositions (\*) with the same sets of indices $`\{i\}`$ and $`\{j\}`$. Now according to 2.4 the motives of $`\mathrm{SB}(A)`$ and $`\mathrm{SB}(A^{})`$ are isomorphic and, hence, so are $`(X)`$ and $`(Y)`$. ∎
The following obvious consequences of Theorem 1.4 are used in the proof of Theorem 1.1.
###### 4.4 Corollary.
Let $`X=\mathrm{SB}_{1,\mathrm{},n}(A)`$ and $`Y=\mathrm{SB}_{1,\mathrm{},n}(A^{})`$ be twisted forms of the variety of complete flags of type $`\mathrm{A}_n`$. Assume the respective central simple algebras $`A`$ and $`A^{}`$ have exponents 1,2,3,4 or 6. Then
$$(X)(Y)XY.$$
###### 4.5 Corollary.
Let $`X`$ and $`Y`$ be twisted forms of the variety of complete flags $`\mathrm{B}_2/B`$. Then
$$(X)(Y)XY.$$
###### Proof.
The proof repeats the proof of 1.4 observing that the motivic decompositions (\*) is provided by \[CPSZ, Cor. 2.9\]. ∎
###### 4.6.
Consider the pseudo-abelian completion $`(G,R)`$ of the category of motives of projective $`G`$-homogeneous varieties with $`R`$-coefficients, where $`G`$ is a group of inner type $`\mathrm{A}_n`$ and $`R`$ is a ring of coefficients. Such categories were defined and extensively studied in \[CM04\]. In particular, it was proven that any object of $`(G,R)`$, where $`R`$ is a discrete valuation ring, has a unique direct sum decomposition into indecomposable objects. Modulo this result the proof of 1.4 immediately implies the following
###### 4.7 Corollary.
Let $`G`$ be an adjoint semi-simple group of inner type $`\mathrm{A}_n`$. Let $`R`$ be a ring such that any object of $`(G,R)`$ has a unique direct sum decomposition into indecomposable objects. Let $`X=\mathrm{SB}_{n_1,\mathrm{},n_r}(A)`$ and $`Y=\mathrm{SB}_{m_1\mathrm{}m_r}(A^{})`$ be two twisted flag varieties of type $`\mathrm{A}_n`$ given by central simple algebras $`A`$ and $`A^{}`$ of prime degree. Assume that $`(X_s)(Y_s)`$. Then
$$(X)(Y)\text{ in }(G,R)A=A^{}\text{ in }\mathrm{Br}(k).$$
### Acknowledgements
These notes appeared as a result of the seminar organized at the University of Bielefeld in the winter of 2004-2005. We give our thanks to the participants of the seminar V.Petrov, O.Roendigs, B.Calmès and others for their remarks. |
warning/0507/hep-th0507165.html | ar5iv | text | # 1 Introduction
## 1 Introduction
What is referred today as Kaluza–Klein theory (see for an extensive collection of papers) is Klein’s modification of the original Kaluza’s theory . These two theories are dual and the duality between them is the duality between threading and slicing decomposition of the five-dimensional spacetime — foliation with one-dimensional or one-codimensional surfaces.
The field equations of Klein’s theory (threading decomposition) express Newton’s constant as a dynamical field (dilaton) and do not allow a constant solution for the dilaton unless an unphysical restriction to the Maxwell tensor $`F_{ij}`$ is imposed: namely, $`F_{ij}F^{ij}=0`$ (Latin indexes run from 1 to 4, Greek — from 1 to 5). In 1983 Gross et al. and Sorkin found magnetic monopoles in Klein’s theory by considering four-dimensional Euclidean and periodic in time Kerr and Taub–NUT solutions which were trivially embedded into a vacuum five-dimensional Klein’s universe with timelike fifth dimension. The original Euclidean periodic time was then identified as the fifth dimension and the magnetic vector potentials — as former degrees of freedom of the four-dimensional Kerr or Taub–NUT solution. The resulting four-dimensional gravity has a non-constant dilaton and has lost the original Kerr or Taub–NUT geometry.
Unlike Klein’s theory, in the original Kaluza’s theory (slicing decomposition) the gauge degrees of freedom of the electromagnetic potentials $`A_i`$ are transferred to the dilaton . This allows us to consider four-dimensional spacetimes with constant dilaton (i.e. Newton’s constant) and fixed gauge. We show that a constant dilaton and a vacuum five-dimensional Kaluza universe necessitate a Ricci-flat four-dimensional slice. In the dual Kaluza’s setup, the Kerr or Taub–NUT geometry of the four-dimensional slice is preserved. The field equations of the original Kaluza’s theory lead to the interpretation of the four-dimensional Lorentzian Kerr and Taub–NUT solutions as resulting from static electric and magnetic charges and dipoles in the presence of ghost matter.
## 2 Field Equations of Kaluza’s Theory
The five-dimensional Kaluza’s metric is:
$`G_{\mu \nu }=\left(\begin{array}{cccc}& & & \\ & g_{ij}^{}& & A_i\\ & & & \\ & & & \\ & A_i^{}& & \varphi \end{array}\right)`$ (5)
The five-dimensional interval in mostly-plus metric is:
$`d\sigma ^2=g_{ij}(dy^i+A^ids)(dy^j+A^jds)+{\displaystyle \frac{1}{N^2}}ds^2,`$ (6)
where $`y^1t,y^5s,g^{ij}`$ is the inverse of $`g_{ij},A^i=g^{ij}A_j`$ and $`N^2=\varphi A^2`$.
The slicing lapse function is $`N^1`$, while the slicing shift vector field is given by $`A^i`$.
If one is to require $`g_{ij}`$ to be the metric of our four-dimensional world and $`A_i`$ — the electromagnetic potentials, then $`N`$ is the dilaton field and can be expressed as :
$`N^2={\displaystyle \frac{\text{det }g}{\text{det }G}}.`$ (7)
The five-dimensional Kaluza metric $`G_{\mu \nu }`$ is a solution to the five-dimensional vacuum equations $`R_{\mu \nu }=0`$, where $`R_{\mu \nu }`$ is the five-dimensional Ricci tensor. These equations were written in terms of the extrinsic curvature $`\pi _{ij}=(N/2)(_iA_j+_jA_i_sg_{ij})`$ of the four-dimensional world as follows :
$`r_{ij}{\displaystyle \frac{1}{2}}g_{ij}r`$ $`=`$ $`N_i_j{\displaystyle \frac{1}{N}}N(_\text{A}\pi _{ij}+_s\pi _{ij})+(\pi \pi _{ij}2\pi _{ik}\pi _j^k)`$ (8)
$`+{\displaystyle \frac{1}{2}}g_{ij}(\pi ^2\pi _{kl}\pi ^{kl}),`$
$`0`$ $`=`$ $`_i(\pi _j^i\delta _j^i\pi ),`$ (9)
$`{}_{}{}^{}{\displaystyle \frac{1}{N}}`$ $`=`$ $`A^i_i\pi {\displaystyle \frac{1}{N}}\pi _{ij}\pi ^{ij}_s\pi ,`$ (10)
where $`r_{ij}`$ is the four-dimensional Ricci tensor, $`r`$ is the four-dimensional scalar curvature, $`_i`$ is the four-dimensional covariant derivative, $`{}_{}{}^{}=g^{ij}_i_j,`$ $`_\text{A}`$ is the Lie derivative with respect to $`A_i`$ and and $`\pi =\pi _k^k`$.
These equations can be equivalently written as :
$`r_{ij}{\displaystyle \frac{1}{2}}g_{ij}r={\displaystyle \frac{N^2}{2}}T_{ij},`$ (11)
$`_iF^{ij}=2A_ir^{ij}+{\displaystyle \frac{2}{N^2}}(\pi ^{ij}\pi _k^kg^{ij})_iN,`$ (12)
$`_i(A_j\pi ^{ij}^i{\displaystyle \frac{1}{N}})=0,`$ (13)
where $`F_{ij}=_iA_j_jA_i`$ is the Maxwell electromagnetic tensor.
The first of these equations, (11), are the equations of general relativity with matter, the second, (12), are a generalization of Maxwell’s equations and the last, (13), is the gauge-fixing condition. The dilaton $`N`$ is related to Newton’s constant $`G_N`$ via :
$`{\displaystyle \frac{N^2}{2}}=\kappa ={\displaystyle \frac{8\pi G_N}{c^4}}.`$ (15)
The energy-momentum tensor $`T_{ij}`$ appearing in equation (11) is given by :
$`T_{ij}=T_{ij}^{\text{Maxwell}}+^k\mathrm{\Psi }_{ijk}+^k\mathrm{\Theta }_{ijk}+C_{ij}+D_{ij},`$ (16)
where:
$`T_{ij}^{\text{Maxwell}}`$ $`=`$ $`F_{ik}F_j^k{\displaystyle \frac{1}{4}}g_{ij}F_{kl}F^{kl},`$ (17)
$`\mathrm{\Psi }_{ijk}`$ $`=`$ $`A_k_jA_iA_j_kA_i+A_iF_{jk},`$ (18)
$`\mathrm{\Theta }_{ijk}`$ $`=`$ $`_i(A_kA_j)+g_{ij}(A^l_kA_lA_k_lA^l),`$ (19)
$`C_{ij}`$ $`=`$ $`g_{ij}A^kA^lr_{kl}2A^kA_jr_{ik}2A^kA_ir_{jk},`$ (20)
$`D_{ij}`$ $`=`$ $`{\displaystyle \frac{2}{N}}_i_j{\displaystyle \frac{1}{N}}{\displaystyle \frac{2}{N^2}}\pi _k^k(A_i_jN+A_j_iN)`$ (21)
$`+{\displaystyle \frac{2}{N^2}}\left[A^k\pi _{ij}+A_i\pi _j^k+A_j\pi _i^kg_{ij}(A^l\pi _l^kA^k\pi _l^l)\right]_kN.`$
The field equations reveal a very interesting relation between the type of solution of the four-dimensional general relativity and the dilaton.
We first suppose that the dilaton is constant: $`N=\text{const}`$. Let us write the five-dimensional vacuum metric $`G_{\mu \nu }`$ in a block-diagonal form: $`G_{\mu \nu }=\text{diag }(g_{ij}^{},N^2)`$. Having $`A_i=0`$ in the field equations, together with $`N=\text{const}`$, results in vanishing of the full energy-momentum tensor and, therefore in a Ricci-flat four-dimensional relativity ($`r_{ij}=0`$). One can re-introduce the electromagnetic potentials via a five-dimensional coordinate transformation. The only transformation which leaves the five-dimensional interval (6) invariant is: $`y^iy^i+sc^i,ss,`$ where $`c^i`$ are constants. Then the physical electromagnetic potentials will be $`A_j=g_{jk}c^k`$ . Under this transformation the fields in the five-dimensional interval (6) transform as $`g_{ij}^{}=g_{ij},A^i=A^i+c^i,N^{}=N`$. Thus $`r_{ij}=0`$ remains unchanged. Therefore constant dilaton and a vacuum five-dimensional Kaluza universe necessarily result in a Ricci-flat four-dimensional slice.
## 3 Flat Four-dimensional Universe
It is interesting to consider whether the converse is true, namely, if a Ricci-flat four-dimensional slice ($`r_{ij}=0`$), embedded in a vacuum five-dimensional Kaluza universe ($`R_{\mu \nu }=0`$), results in a constant dilaton ($`N=\text{const}`$). We will give an example which shows that this is not the case. Consider a flat four-dimensional slice with $`g_{ij}=\eta _{ij}=\text{diag}(1,\mathrm{\hspace{0.17em}1},\mathrm{\hspace{0.17em}1},\mathrm{\hspace{0.17em}1})`$. This is clearly a vacuum solution. Let us now see if the five-dimensional metric $`G_{\mu \nu }=\text{diag}(1,\mathrm{\hspace{0.17em}1},\mathrm{\hspace{0.17em}1},\mathrm{\hspace{0.17em}1},N^2)`$ admits a non-constant solution for $`N`$. From the field equations we see that when $`r_{ij}=0`$ and $`A_i=0`$, then $`N`$ must be a solution to:
$`_i_j{\displaystyle \frac{1}{N}}=0.`$ (22)
For the flat case, the obvious solution is: $`N=(a_ky^k+a_5)^1`$, where $`a_\mu `$ are constants. It will be very interesting to take:
$`a_2=a_3=a_4=a_5=0,a^1={\displaystyle \frac{c^2}{4\sqrt{\pi }t_0}}.`$ (23)
Then Newton’s constant will become $`G_N=c^4N^2/(16\pi )=(t_0/t)^2`$ — the gravitational attraction falling off with time from infinity. This can be explained as a purely geometric effect between a vacuum universe embedded into another vacuum universe.
The Kasner metric is:
$`d\sigma ^2=dt^2+{\displaystyle \underset{i=2}{\overset{4}{}}}\left({\displaystyle \frac{t}{t_0}}\right)^{2p_i}(dy^i)^2+\left({\displaystyle \frac{t}{t_0}}\right)^{2p_5}ds^2,`$ (24)
where $`_{i=2}^5p_i=_{i=2}^5p_i^2=1`$.
Solution (23) corresponds to the special case: $`p_2=p_3=p_4=0,p_5=1`$.
## 4 Ghost Energy-Momentum Tensor
As we are interested in solutions to the vacuum five-dimensional relativity with constant four-dimensional Newton’s constant (dilaton), we will have to consider Ricci-flat solutions to four-dimensional relativity only. For $`r_{ij}=0`$ and $`N=\text{const}`$ the field equations reduce to:
$`r_{ij}{\displaystyle \frac{1}{2}}h_{ij}r`$ $`=`$ $`{\displaystyle \frac{N^2}{2}}(T_{ij}^{\text{Maxwell}}+^k\mathrm{\Psi }_{ijk}+^k\mathrm{\Theta }_{ijk})=\mathrm{\hspace{0.17em}0},`$ (25)
$`_iF^{ij}`$ $`=`$ $`0,`$ (26)
$`_i(A_j\pi ^{ij})`$ $`=`$ $`0.`$ (27)
We further have:
$`^jT_{ij}^{\text{Maxwell}}=F_{ik}_jF^{jk}=0,`$ (28)
due to (26). Equation (28) is the conservation law for the energy and momentum resulting from Maxwell’s equations (26).
The tensor $`^k\mathrm{\Theta }_{ijk}`$ satisfies:
$`^j^k\mathrm{\Theta }_{ijk}={\displaystyle \frac{2}{N}}_j_i(A_k\pi ^{ik})=0,`$ (29)
in view of the gauge-fixing condition (27).
Considering the remaining term, $`^k\mathrm{\Psi }_{ijk}`$, we see that it satisfies:
$`^j^k\mathrm{\Psi }_{ijk}`$ $`=`$ $`{\displaystyle \frac{1}{2}}(^j^k+^k^j)\mathrm{\Psi }_{ijk}+{\displaystyle \frac{1}{2}}[^j,^k]\mathrm{\Psi }_{ijk}=0`$ (30)
in view of the antisymmetry $`\mathrm{\Psi }_{ijk}=\mathrm{\Psi }_{ikj}`$ and $`r_{ij}=0`$. Thus the tensor $`^k\mathrm{\Psi }_{ijk}`$ does not describe any dynamics.
The gauge-fixing condition (27) and equation (30) lead to the conservation law:
$`^jT_{ij}^{\text{Ghost}}=0,`$ (31)
where $`T_{ij}^{\text{Ghost}}`$ is the Belinfante symmetric energy-momentum tensor of the ghost fields:
$`T_{ij}^{\text{Ghost}}=^k(\mathrm{\Psi }_{ijk}+\mathrm{\Theta }_{ijk}).`$ (32)
For the ”haunted” Kaluza’s universe, the energy and momentum of the ghost fields compensates completely the energy and momentum of the Maxwell’s fields:
$`T_{ij}^{\text{Ghost}}+T_{ij}^{\text{Maxwell}}=0`$ (33)
and, therefore, it is possible to have matter co-existing with ghost matter in a Ricci flat universe.
## 5 Four-dimensional Lorentzian slice with Kerr <br>Geometry
We will generate the five-dimensional solution simply by starting off with a four-dimensional static Ricci-flat solution, promoting it trivially to five dimensions (by adding $`ds^2`$ in the metric) and performing a five-dimensional coordinate transformation:
$`tt+\beta s,`$ (34)
where $`\beta `$ is the inverse of the “speed of light” along the fifth, transverse dimension. This coordinate transformation will not introduce $`s`$-dependence in the four-dimensional world (as the four-dimensional metric is static and time appears only with its differential) and as a result the new five-dimensional ”observer” will “see” the electromagnetic potentials:
$`A_j=\beta g_{tj},`$ (35)
where $`j\{t,r,\theta ,\varphi \}`$.
The four-dimensional Kerr metric in Boyer–Lindquist coordinates , trivially promoted to five dimensions is:
$`d\sigma ^2`$ $`=`$ $`{\displaystyle \frac{\mathrm{\Delta }}{\rho ^2}}(dta\mathrm{sin}^2\theta d\varphi )^2+{\displaystyle \frac{\mathrm{sin}^2\theta }{\rho ^2}}[(r^2+a^2)d\varphi adt]^2`$ (36)
$`+{\displaystyle \frac{\rho ^2}{\mathrm{\Delta }}}dr^2+\rho ^2d\theta ^2+ds^2,`$
where $`\mathrm{\Delta }=r^2\alpha r+a^2`$ and $`\rho ^2=r^2+a^2\mathrm{cos}^2\theta `$. Here $`\alpha `$ and $`a`$ are integration constants which will be identified further (in the four-dimensional Kerr solution these are the mass and the angular momentum per unit mass of a black hole). We consider the physically interesting case $`\alpha >a`$ (black hole solution).
The electromagnetic potentials (35) are:
$`A_t`$ $`=`$ $`\beta (1+{\displaystyle \frac{\alpha r}{\rho ^2}}),`$ (37)
$`A_r`$ $`=`$ $`0,`$ (38)
$`A_\theta `$ $`=`$ $`0,`$ (39)
$`A_\varphi `$ $`=`$ $`{\displaystyle \frac{a\alpha \beta r\mathrm{sin}^2\theta }{\rho ^2}}.`$ (40)
For large $`r`$, the non-zero components of the vector potential are:
$`A_t`$ $``$ $`\beta (1+{\displaystyle \frac{\alpha }{r}}),`$ (41)
$`A_\varphi `$ $``$ $`{\displaystyle \frac{a\alpha \beta \mathrm{sin}^2\theta }{r}}.`$ (42)
Therefore, from (41), one can identify the constant $`\alpha \beta `$ as electric charge:
$`q=\alpha \beta .`$ (43)
Equation (42) describes the field of a magnetic dipole of strength $`a\alpha \beta `$, located at the origin :
$`m=a\alpha \beta =aq.`$ (44)
Thus one can interpret the Kerr solution as a black hole generated by an electric charge and magnetic dipole (and not by a rotating massive centre). The potentials (35) satisfy the vacuum Maxwell’s equations (26). The electric charge and the magnetic dipole are located at the singularity $`\rho =0`$ .
This is the only singularity of the Kerr spacetime and can be better understood in Cartesian coordinates :
$`x`$ $`=`$ $`\sqrt{r^2+a^2}\mathrm{sin}\theta \mathrm{cos}\varphi ,`$ (45)
$`y`$ $`=`$ $`\sqrt{r^2+a^2}\mathrm{sin}\theta \mathrm{sin}\varphi ,`$ (46)
$`z`$ $`=`$ $`r\mathrm{cos}\theta .`$ (47)
The singularity $`\rho =0`$, i.e. $`r=0`$ and $`\mathrm{cos}\theta =0`$, corresponds to the ring $`x^2+y^2=a^2`$.
One can analytically continue the Kerr solution for negative values of $`r`$ . The horizons are at:
$`r_\pm =\alpha \pm \sqrt{\alpha ^2a^2}.`$ (49)
The equations of the corresponding static horizons are:
$`r_\pm (\theta )=\alpha \pm \sqrt{\alpha ^2a^2\mathrm{cos}^2\theta }.`$ (50)
There are no timelike coordinates inside the ergosphere — the region between the event horizon and surrounding static horizon.
Then Kerr solution describes two universes which behave asymptotically as Schwarzschild universes — one with $`r>0`$ and having a positive centre $`\alpha `$, event horizon at $`r_+`$, and a Cauchy horizon at $`r_{}`$; the other — with $`r<0`$ and having a negative centre $`\alpha `$, event horizon at $`r_{}`$, and a Cauchy horizon at $`r_+`$. In our context this has the natural interpretation of a black hole solution with positive/negative charge $`q=\alpha \beta `$ and a magnetic dipole of strength $`m=a\alpha \beta `$.
## 6 Four-dimensional Lorentzian slice with Taub–NUT Geometry
We consider five-dimensional Kaluza’s universe with a four-dimensional Lorentzian Taub-NUT slice:
$`d\sigma ^2`$ $`=`$ $`V(r)(dt+2\mathrm{}\mathrm{cos}\theta d\varphi )^2+{\displaystyle \frac{1}{V(r)}}dr^2`$ (51)
$`+(r^2+\mathrm{}^2)(d\theta ^2+\mathrm{sin}^2\theta d\varphi ^2)+ds^2,`$
where
$`V(r)=12{\displaystyle \frac{\alpha r+\mathrm{}^2}{r^2+\mathrm{}^2}},`$ (52)
where $`\alpha `$ and $`\mathrm{}`$ are, again, integration constants.
This metric has conical singularities at $`\theta =0,\pi `$ (Misner string ). The event horizon is where $`V(r)`$ vanishes, i.e. at:
$`r_\pm =\alpha \pm \sqrt{\alpha ^2+\mathrm{}^2}.`$ (53)
The metric can be analytically continued for negative $`r`$ in a similar way and we will be interested in the two regions: Region I with $`\alpha 0`$ and $`r<r_{}<0`$ and Region II with $`\alpha 0`$ and $`r>r_+>0`$.
There is also an ”ergoregion”, surrounding the Misner string, inside of which $`\varphi `$ is another timelike coordinate. The equation of these horizons is:
$`\mathrm{tan}^2\theta ={\displaystyle \frac{4\mathrm{}^2V(r)}{r^2+\mathrm{}^2}},r<r_{}\text{ or }r>r_+.`$ (54)
For the electromagnetic potentials (35) introduced with the transformation (34), we get:
$`A_t`$ $`=`$ $`\beta V(r),`$ (55)
$`A_r`$ $`=`$ $`0,`$ (56)
$`A_\theta `$ $`=`$ $`0,`$ (57)
$`A_\varphi `$ $`=`$ $`2\mathrm{}\beta V(r)\mathrm{cos}\theta .`$ (58)
Asymptotically, for large $`r`$ we have:
$`A_t`$ $`=`$ $`\beta \left(1{\displaystyle \frac{2\alpha }{r}}\right),`$ (59)
$`A_\varphi `$ $`=`$ $`2\mathrm{}\beta \left(1{\displaystyle \frac{2\alpha }{r}}\right)\mathrm{cos}\theta .`$ (60)
Equation (59) is the electric potential due to charge
$`q=2\alpha \beta .`$ (61)
Equation (60) is the potential due to a magnetic monopole of charge $`m=2\beta \mathrm{}`$. This can be seen by integrating the flux of the magnetic field $`B^r`$ through the infinite sphere:
$`4\pi m=\underset{r\mathrm{}}{lim}{\displaystyle _{S_r}}B^r𝑑s_r=\underset{r\mathrm{}}{lim}{\displaystyle _{S_r}}F_{\theta \varphi }𝑑\theta 𝑑\varphi =4\pi (2\beta \mathrm{}).`$ (62)
For Kerr geometry this integral vanishes (we do not have a monopole but a magnetic dipole there).
The location of these charges appears to be unclear as $`V(r)=12(\alpha r+\mathrm{}^2)(r^2+\mathrm{}^2)^1`$, appearing in (55) and (58), is not singular outside the horizons. However, every spacelike hypersurface which is pushed between the horizons becomes singular and therefore, we can interpret the points $`ir=\pm \mathrm{}`$ (note that $`ir`$ is a spacelike coordinate between the horizons) as the loci where the images of the charges are seen by an observer with $`r>r_+`$ or $`r<r_{}`$. For the case $`\alpha =0`$, the proper distance between the origin and the location of the images is:
$`{\displaystyle \underset{i0}{\overset{\pm i\mathrm{}}{}}}{\displaystyle \frac{dr}{\sqrt{V(r)}}}=\pm \mathrm{}{\displaystyle \underset{0}{\overset{\mathrm{ln}(1+\sqrt{2})}{}}}\sqrt{1\mathrm{sinh}^2x}𝑑x\pm \mathrm{\hspace{0.17em}0.7}\mathrm{}.`$ (63)
Therefore, an observer with $`r>r_+>0`$ (Region I) will register the image of a monopole a proper distance $`|0.7\mathrm{}|`$ from the origin, while an observer with $`r<r_{}<0`$ (Region II) will register the image of a monopole a proper distance $`|0.7\mathrm{}|`$ from the origin.
## 7 Conclusions
We have presented solutions to the original Kaluza’s theory which describe static electric and magnetic fields generated by point-like electric and magnetic charges and dipoles. Unlike the dual Kaluza–Klein theory (namely, Klein’s modification of the original Kaluza’s theory) these solutions allow to have a constant Newton’s constant, as the gauge degrees of freedom are now transferred to the dilaton. The gauge-fixing of the electromagnetic potentials results in the appearance of a Belinfante ghost part in the full energy-momentum tensor which fully compensates the electromagnetic energy-momentum tensor. Four-dimensional solutions with vanishing Ricci tensor ($`r_{ij}=0`$) are the only possible solutions when the dilaton is required to be constant in a Ricci-flat five-dimensional universe. These four-dimensional gravitational Ricci-flat solutions can be interpreted from a five-dimensional Kaluza’s perspective, as solutions generated by four-dimensional electromagnetism of charges and dipoles or their images (for the case of a Taub–NUT four-dimensional slice). The integration constants $`a,\alpha \text{ and }\beta `$ (for Kerr geometry) and $`\alpha ,\mathrm{}\text{ and }\beta `$ (for Taub–NUT geometry) have interpretation as charges (see (43) and (44) for Kerr geometry and (61) and (62) for Taub–NUT geometry) and the solutions represent gravitational attraction without unphysical regions with gravitational repulsion, as, for example, in the Reissner–Nordstrøm case .
## Acknowledgements
We would like to thank Siddhartha Sen, Brian Nolan and Vesselin Gueorguiev for very helpful discussions. |
warning/0507/astro-ph0507207.html | ar5iv | text | # Observations of Selected AGN with H.E.S.S.
## 1 Introduction
Active galactic nuclei are known to emit radiation over the entire electromagnetic spectrum, from radio waves to TeV $`\gamma `$-rays. These objects, which are found in only a small fraction of the total number of observed galaxies, are very luminous, extremely compact, and can exhibit large luminosity variations on time scales ranging from less than an hour up to several years. Although AGN differ widely in their observed characteristics, a unified description (as reviewed in Urry & Padovani (1995)) has emerged in which an AGN consists of a super-massive black hole ($`10^710^{10}`$ solar masses) surrounded in the inner regions by an accretion disk, and in the outer regions by a thick torus of gas and dust. In some AGN (the radio-loud population, $``$10%), a highly relativistic outflow of energetic particles exists approximately perpendicular to the accretion disk and torus plane. This flow forms collimated radio-emitting jets which generate the non-thermal emission observed from radio to $`\gamma `$-rays. It is believed that some of the numerous AGN classifications result from viewing these objects at various orientation angles with respect to the torus plane. Essentially all AGN detected at VHE energies (shown in order of redshift with references in Table 1) are radio-loud objects of the BL Lacertae (BL Lac) type, which have their jet pointed close to the observer’s line of sight. An exception to this exists with the detection of the Fanaroff-Riley type I radio galaxy M 87 at VHE energies. Although M 87 is believed to be a mis-aligned BL Lac M87\_misalign (Tsvetanov et al. 1998), it is not clear whether the VHE emission comes from the jet or the central object.
The known VHE AGN have helped to constrain significantly the models for production of VHE $`\gamma `$-rays through spectral and variability studies. However, there are still many differing models that describe the present data, making a larger sample of known VHE AGN necessary to make more definitive conclusions. Also, VHE photons are absorbed by interactions on the extragalactic background light (EBL) leading to an energy dependent horizon for viewing VHE sources. The energy spectrum of VHE AGN may exhibit characteristics, such as steepening of the spectrum and a cutoff, as a result of this absorption. Interpretation of such features can be used as a probe of the EBL in the optical and near-IR regimes (Stecker, de Jager & Salamon (1992); Schroedter (2005)), for which direct measurements are dominated by large systematic uncertainties. Since such an interpretation is complicated by discerning which features are a result of EBL absorption and which are intrinsic to the object, a large data set of VHE AGN at differing redshifts are needed to ascertain which effects can be attributed to the EBL.
A large sample of AGN located at z$`<`$0.333 was observed by H.E.S.S. in 2003 and 2004. Most of these objects are BL Lacs, many of which are suggested as good candidates for detection as VHE emitters (Costamante & Ghisellini (2002); Perlman (1999); Stecker, de Jager & Salamon (1996)). A sample of nearby non-blazar AGN, like M 87, was also observed with the hope of extending the known VHE-bright AGN to other classes. These include a set of famous radio-loud galaxies, characterized by resolved radio, optical and X-ray jets (Cen A, Pictor A, 3C 120, and the quasar 3C 273) and a sample of radio-weak objects (the Seyfert galaxies NGC 1068, NGC 3783 and NGC 7469). The detections resulting from the H.E.S.S. AGN observation program have been reported elsewhere (see Table 1 for references). These include the confirmation of the VHE emission seen from M 87 and PKS 2155$``$304, the detection of Markarian 421 using large-zenith-angle observations, and the discovery of VHE emission from PKS 2005$``$489. Flux upper limits, the strongest ever produced, from the non-detection of the remaining objects are presented here.
## 2 H.E.S.S. Detector
The H.E.S.S. experiment, a square array (120 m side) of four imaging atmospheric-Cherenkov telescopes located in the Khomas Highland of Namibia (23 16’ 18” S, 16 30’ 1” E, 1800 m above sea level), uses stereoscopic observations of $`\gamma `$-ray induced air showers to search for astrophysical $`\gamma `$-ray emission above $``$100 GeV. Each telescope has a 107 m<sup>2</sup> tessellated mirror dish and a 5 field-of-view (f.o.v.) camera consisting of 960 individual photomultiplier pixels. The sensitivity of H.E.S.S. (5$`\sigma `$ in 25 hours for a 1% Crab Nebula flux source at 20 zenith angle) allows for detection of VHE emission from objects at previously undetectable flux levels. More details on H.E.S.S. can be found in Bernlöhr et al. (2003), Funk et al. (2004), Hofmann (2003), and Vincent et al. (2003).
## 3 Observations
The H.E.S.S. observations of AGN in 2004 use the full four-telescope array. For some of the data, individual telescopes were excluded from the observations or analysis due to hardware problems. Also, 2003 observations of 1ES 0323+022 were made prior to the completion of the array and thus use only two or three telescopes. While the sensitivity of H.E.S.S. is less during observations with fewer telescopes, it is still unprecedented. Table 2 shows the candidate AGN observed by H.E.S.S. and gives details of the observations that pass selection criteria which remove data for which the weather conditions were poor or the hardware was not functioning properly. The data were taken in 28 minute runs using Wobble mode, i.e. the source direction is offset, typically by $`\pm `$0.5, relative to the center of the f.o.v. of the camera during observations, which allows for both on-source observations and simultaneous estimation of the background induced by charged cosmic rays. As the energy threshold of H.E.S.S. observations increases with zenith angle, the mean zenith angle of the exposure for each of the AGN along with the corresponding average energy threshold (after selection cuts) of those observations is also shown in Table 2. It should be noted that the H.E.S.S. Monte Carlo simulations show that the azimuthal angle at which an object is observed has a small effect on the energy threshold of observations. Sources which culminate in the south (i.e. those with declination less than the latitude of H.E.S.S.) have slightly higher energy thresholds (e.g. compare 1ES 0323+022 with Pictor A).
## 4 Analysis Technique
The data passing the run selection criteria are calibrated as detailed in Aharonian et al. (2004b), and the event reconstruction and background rejection are performed as described in Aharonian et al. (2005a), with some minor improvements discussed in Aharonian et al. (2005b). The background is estimated using all events passing selection cuts in a number of circular off-source regions offset by the same distance, relative to the center of the f.o.v., in the sky as the on-source region (for more details see Aharonian et al. (2001)). The on-source region, the size of which is optimized for the detection of point sources, is a circle of radius $``$0.11 centered on the source, and each off-source region has approximately<sup>1</sup><sup>1</sup>1The off-source data are first placed into a pixelated two-dimensional map and then integrated in an approximate circle for each region. The difference in total area is of order 1%. the same area as the on-source region. The maximum number of non-overlapping off-source regions fitting in the field of view are used. An area around the on-source position, completely containing the H.E.S.S. point-spread-function, is excluded to eliminate possible contamination from poorly reconstructed $`\gamma `$-rays. For the typical on-source offset of 0.5, eleven off-source regions are possible. In the case of observations of Cen A, offset by 0.7, sixteen regions are used. The statistical error on the background measurement is reduced by the use of a larger background sample, and there is no need for a radial acceptance correction, which accounts for the strongest acceptance change across the f.o.v., since the off-source regions are offset by the same radial distance as the on-source region. The significance of any excess is calculated following the method of Equation (17) in Li & Ma (1983) and all upper limits are determined using the method of Feldman & Cousins (1998).
## 5 Results
Figure 1 shows the distribution of the significance observed from the direction of each of the twenty AGN. No significant excess of VHE $`\gamma `$-rays is found from any of the AGN in the given exposure time ($`<`$8 hrs each), with the possible exception of Markarian 501 (3.1$`\sigma `$). Specific details of the results for each AGN are shown in Table 2. Additionally, a search for serendipitous source discoveries in the H.E.S.S. f.o.v. centered on each of the AGN yielded no significant excess.
Given that it is well established that Markarian 501 is a VHE $`\gamma `$-ray emitter, the excess (3.1$`\sigma `$) from the only night (MJD 53172) of observations of Markarian 501 can be treated as significant and a flux calculated. Assuming the spectrum measured above 1.5 TeV by the High Energy Gamma Ray Astronomy (HEGRA) experiment HEGRA\_501 (Bradbury et al. 1997), a power law with a photon index of $`\mathrm{\Gamma }`$=2.6, the corresponding integral flux above the 1.65 TeV energy threshold is I($`>`$1.65 TeV) = (1.5$`\pm `$0.6<sub>stat</sub>$`\pm `$0.3<sub>syst</sub>) $`\times `$ 10<sup>-12</sup> cm<sup>-2</sup> s<sup>-1</sup> or $``$15% of the H.E.S.S. Crab Nebula flux above this threshold. While the VHE flux from Markarian 501 is known to be highly variable, the measured flux is similar to the value reported in Bradbury et al. (1997).
For the remaining undetected AGN, 99.9% upper limits on the integral flux (assuming a power law spectrum with $`\mathrm{\Gamma }`$=3.0) above the energy threshold of the observations, and references to previously published limits (when available), are shown in Table 3. The photon index, $`\mathrm{\Gamma }`$=3.0, was chosen for two reasons: First, the recently measured VHE spectra of several AGN (e.g. PKS 2155$``$304, PKS 2005$``$489) are softer than the Crab Nebula-like index of $`\mathrm{\Gamma }`$=2.5 often used for VHE upper limits in past publications. Second, the softer index was chosen to account for the possible steepening of the observed spectra of the AGN due to the absorption of $`\gamma `$-rays on the EBL. Assuming a different photon index (i.e. $`\mathrm{\Gamma }`$ between 2.5 and 3.5) has less than a $``$10% effect on the reported limits, and the systematic error on the upper limits is estimated to be $``$20%. The percentage of the Crab Nebula flux shown in Table 3 is calculated relative to the integral flux, above the same threshold, determined from the H.E.S.S. Crab Nebula spectrum. The H.E.S.S. limits are considerably ($`>`$5 times) stronger than any reported to date. However, due to the generally variable nature of AGN emission, these upper limits constrain the maximum average brightness of the AGN only during the observation time. Hence they are limits on the steady-component or quiescent flux from the AGN. Future flaring behavior may increase the VHE flux from any of these AGN to significantly higher levels.
A search for VHE flux variability from each observed AGN was also performed. Here the nightly integral flux above the average energy threshold was calculated assuming a photon index of $`\mathrm{\Gamma }`$=3.0 and fit by a constant. Any flaring behavior would be demonstrated in the form of a poor $`\chi ^2`$ probability for the fit. Table 4 shows the dates each AGN was observed and the resulting $`\chi ^2`$ probability. As can be seen, no evidence for VHE flux variability is found.
The lack of any significant VHE detection or flaring behavior is perhaps expected from the beahvior of the individual AGN in the X-ray regime. Quick-look results provided by ASM/RXTE team show that none of the AGN (for which all-sky monitor data exists) were particularly active during the dates of the H.E.S.S. observations. On these dates, the measured daily average count rate from each AGN never deviated by more than $``$2$`\sigma `$ from the mean value averaged over the whole X-ray data set.
## 6 Discussion
Since AGN are known to emit radiation in all wavebands, understanding and modelling their emission must take into account their entire spectral energy distribution (SED). Constraining any model is difficult as only a limited number of high-energy measurements currently exist (see Fichtel et al. (1994) for EGRET upper limits on blazars, Seyfert galaxies and radio-loud galaxies). This is especially true at VHE energies, making the upper limits presented in Table 3 quite useful due to their unprecedented strength. While such modelling is beyond the scope of this paper, the applicability and usefulness of the limits for each of the three classes of observed AGN are discussed.
### 6.1 BL Lacs
BL Lacs belong to the sub-class of radio-loud AGN known as blazars, which are AGN thought to possess a jet which is viewed close to the line of sight (Urry & Padovani (1995)). The distinction between BL Lacs and other blazars is primarily based on their optical spectra which are characterized by weak or absent emission lines. As mentioned in the introduction, almost all VHE bright AGN belong to this class. These AGN have dominantly non-thermal emission and are characterized by a double-humped SED. The low-energy component is widely accepted as originating from synchrotron radiation of relativistic electrons in the magnetic field around the object. However, the origin of the high-energy component is the subject of much debate. Various models involving either leptonic or hadronic processes have been proposed and can be constrained using the H.E.S.S. results. However, some caveats are required for interpreting a blazar SED with the H.E.S.S. upper limits.
Blazars are known to be highly variable at all wavelengths, typically characterized by low-emission quiescent states interrupted by periods of flaring behavior where the flux increases dramatically. In some cases this increase is several orders of magnitude. Due to this extreme variability, it has been shown that fitting the SED of blazars has very large uncertainties when non-simultaneous multiwavelength (MWL) data are used (see e.g. Böttcher, Mukherjee & Reimer (2002)). As a result the usefulness of non-simultaneous upper limits, as is the case for the H.E.S.S. observations, in modelling these object is limited. The H.E.S.S. upper limits, in the absence of simultaneous MWL data, are only relevant for modelling the quiescent state of the blazar using archival low-state MWL data. An additional problem using these upper limits arises due to the absorption of VHE photons on the EBL. The upper limit on the flux intrinsic to the object can be significantly higher than those presented in Table 3 depending on the redshift. As a result the upper limits must have the effects of the EBL removed before they can be used for modelling. Unfortunately, parameterizations of the EBL are poorly constrained leading to numerous models with dramatically different behaviors, adding another significant uncertainty when using VHE upper limits to help model blazar emission. Given the wide range of EBL interpretations, this deabsorption is not performed here.
Taking note of the caveats regarding the effects of the EBL and the issues with non-simultaneous observations, a comparison of the upper limits is made, where possible, to three sets of VHE flux predictions based on the SEDs of blazars. The first set (Stecker, de Jager & Salamon (1996)), referred to as SDS henceforth, uses simple scaling arguments to predict VHE fluxes for Einstein Slew survey objects. In the case of the SDS flux predictions the effects of EBL absorption are already accounted for with an ”averaged” model. The other two sets of predictions are taken from Costamante & Ghisellini (2002). The first, referred to as FOS, uses a phenomenological description of the average SED of blazars based on their bolometric luminosity Fossati (Fossati et al. 1998), modified by Donato et al. (2001), and derives predictions on the basis of the individual blazar’s radio luminosity and synchrotron peak frequency. The second, referred to as CG, uses fits of a synchrotron self-Compton model to existing multiwavelength data. Both the FOS and CG predictions do not have the effects of EBL absorption accounted for. This could change the flux predictions by factors of $``$5 above 300 GeV and by factors $`>`$100 above 1 TeV for objects at $`z`$$``$0.2.
Table 5 shows the 99.9% H.E.S.S. flux upper limits extrapolated (assuming $`\mathrm{\Gamma }=3.0`$) to above 300 GeV and above 1 TeV, as well as which predictions are available above these thresholds. The H.E.S.S. upper limits are below the SDS predictions above 300 GeV in three of the five cases (factors ranging from $``$2 to $``$5), and below two of the five predictions above 1 TeV (factors of $``$1.3 and $``$5). Even if the EBL absorption effects are accounted for in the SDS predictions, the discrepancies can easily be accounted for by the aforementioned simultaneity caveats and thus the H.E.S.S. upper limits do not make any strong statements regarding the SDS predictions. All the FOS predictions are above the H.E.S.S. upper limits, from factors of $``$1.4 to $``$16 for the predictions above 300 GeV and factors of $``$5 to $``$40 for the predictions above 1 TeV. While at first this seems severe, accounting for TeV absorption can reduce these discrepancies dramatically. In addition the FOS predictions are claimed to be more suitable for ”high” state VHE flux predictions, whereas the H.E.S.S. upper limits are most appropriate for constraining the quiescent state of the AGN. Given that variability of up to two orders of magnitude have been seen in VHE blazars such as Markarian 421, it is clear that it is again difficult to test these predictions with the H.E.S.S. upper limits. However, the disagreement suggests that different sets of parameters might be necessary to account for the quiescent state of the source. The CG predictions, which are claimed to be more appropriate for the quiescent state of the AGN tested by the H.E.S.S. upper limits, are all below the upper limits.
### 6.2 Other Radio-Loud Galaxies
Speculation exists for detectable levels of VHE emission from the jets of AGN without doppler boosting along the line of sight (see e.g. Aharonian, Coppi & Völk (1994)). Therefore, the H.E.S.S. observation program also included four other radio-loud AGN. Like BL Lacs they all possess jets. One of these, 3C 273, meets some, but not all, of the phenomenological criteria for classification as a blazar. However, it is most accurately characterized as a quasar. It is the brightest and one of the most nearby quasars. The other three AGN, Cen A, 3C 120 and Pictor A, are found in Fanaroff-Riley (FR) radio galaxies. These galaxies fall into two classes, FR I and FR II. The distinction is based on their radio morphology FR-morph (Fanaroff & Riley 1974). FR I objects, such as Cen A (the prototype), 3C 120 and the VHE-emitter M 87, show extended jets with no distinct termination point, and FR II objects, like Pictor A, have narrow, collimated jets with terminal ”hotspots.” These FR objects differ from BL Lacs mainly due to a large viewing angle (50$`{}_{}{}^{}`$80) with respect to the jet axis.
Chandra observations (for a review see e.g. Harris (2001)) show that Pictor A, 3C 120, and 3C 273 all possess bright X-ray features like knots and hot spots in their large-scale extragalactic jets. The X-ray fluxes of these features are at least a factor of 10 larger than the radio and optical fluxes. This behavior is the opposite of the predictions from synchrotron self-Compton and inverse-Compton models and requires alternative theoretical explanations (see e.g. Aharonian (2002)). Use of the H.E.S.S. upper limits for these objects should aid in constraining some of the presented scenarios. However, they are still subject to the aforementioned variability and EBL absorption (mainly for 3C 273) caveats.
Located at a distance of 3.4 Mpc, Cen A (NGC 5128) is the closest radio-loud AGN. It is one of the best-studied extragalactic objects due to its large apparent brightness in all wavebands (for a recent review see Israel (1998)). The proximity of Cen A means that the intrinsic spectrum of the object is unaffected by absorption on the EBL, considerably simplifying the use of the H.E.S.S. upper limit in the modelling of its VHE emission. However, the lack of simultaneous observations is still an issue as Cen A, like blazars, exhibits large flux variability, albeit on much longer time scales (years).
During the early days of VHE astronomy, a detection of emission above 300 GeV from Cen A was claimed using a non-imaging Cherenkov system Grindlay (Grindlay et al. 1975) during a historically high emission state. The flux reported, I($`>`$300 GeV) = (4.4$`\pm `$1.0) $`\times `$ 10<sup>-11</sup> cm<sup>-2</sup> s<sup>-1</sup>, is over an order of magnitude above the H.E.S.S. 99.9% flux upper limit extrapolated to above 300 GeV, I($`>`$300 GeV) $`<`$ 2.3 $`\times `$ 10<sup>-12</sup> cm<sup>-2</sup> s<sup>-1</sup>. The H.E.S.S. result does not contradict the claimed detection as RXTE ASM observations show that Cen A was in a low emission state when observed by H.E.S.S.. During a similar low emission state, EGRET detected $`>`$100 MeV $`\gamma `$-ray emission from Cen A EGRET\_cenA (Sreekumar et al. 1999). This is the only EGRET detection associated with an AGN that is not a member of the blazar class. Extrapolating the EGRET spectrum to above the H.E.S.S. threshold yields I($`>`$190 GeV) = 3.5 $`\times `$ 10<sup>-12</sup> cm<sup>-2</sup> s<sup>-1</sup> which is $``$60% of the upper limit shown in Table 3. The H.E.S.S. limit is similar, 5.5 $`\times `$ 10<sup>-12</sup> cm<sup>-2</sup> s<sup>-1</sup>, when assuming the measured EGRET spectrum of $`\mathrm{\Gamma }`$=2.40. These results imply that future identification of a high-emission state in Cen A should motivate further VHE observations.
### 6.3 Radio-Weak Galaxies
All of the radio-weak AGN observed by H.E.S.S. are located in Seyfert galaxies which differ from the galaxies previously discussed in many respects. Like the other AGN, they have outflows, albeit typically with low velocity and uncollimated, approximately perpendicular to the accretion disk. Some even have collimated jets that emit synchrotron radiation. However, the jets are neither as collimated as in radio-loud AGN, nor do they show any indications of relativistic motion. Two kinds of Seyfert galaxies (types I and II) exist whose differences can be explained in terms of viewing angle Seyfert (Antonucci & Miller 1985). It is believed that Seyfert I galaxies are viewed ”face on” and thus the nuclear regions are directly visible, whereas Seyfert II galaxies are viewed ”edge on” causing the nuclear regions to be obscured by material (the torus or warped disk). Currently, no Seyfert galaxies are known to be VHE emitters.
Three bright well-studied Seyfert galaxies were observed by H.E.S.S.: NGC 1068, NGC 3783, and NGC 7469. NGC 1068 (M 77), the prototypical type II object, is the brightest and closest known Seyfert galaxy and as such is perhaps the best candidate for detection of this class at VHE energies. Here it should be noted that since the emission from Seyferts is not beamed, their orientation is not as important as with blazars. NGC 3783, a classical type I object, is also one of the brightest and closest Seyfert galaxies, and one of the most well studied. It is also interesting in that exceptionally deep measurements made using the Chandra X-ray Observatory reveal a fast ($`>10^6`$ km hr<sup>-1</sup>) wind of highly ionized atoms blowing away from the galaxy’s suspected central black hole chandra\_ngc3783 (Kaspi et al. 2002). NGC 7469, also type I, is unusual in that it has an inner ring of gas very close to the nucleus that is undergoing massive star formation starburst (Genzel et al. 1995).
None of these objects were detected and the upper limits shown in Table 3 are quite constraining. While Seyfert-type galaxies are not necessarily expected to emit VHE $`\gamma `$-rays at observable levels, the H.E.S.S. results easily provide constraints for modelling. This is because these AGN generally show less variability than blazars, and all the ones observed are close enough to only have minimal effects from the absorption of VHE photons on the EBL. The H.E.S.S. results could be interpreted as implying that Seyfert-type AGN are not significant emitters of VHE photons. However, the observed sample and exposure times are small, making it premature to rule the class out all together.
## 7 Conclusions
H.E.S.S. observed greater than twenty AGN in 2003 and 2004 as part of a campaign to identify new VHE-bright AGN. Several significant detections from this campaign have been reported elsewhere (see Table 1 for references). Results presented here detail the AGN observations for which no significant excess was found, apart from a marginal signal from the well-known VHE-emitter Markarian 501. Despite the limited exposure ($`<`$8 hours) for each of these AGN, the upper limits on the VHE flux determined by H.E.S.S. are the most stringent to date, demonstrating the unprecedented sensitivity of the instrument. Clearly the strength of the limits makes them quite useful, yet it must again be stressed that any interpretation using the H.E.S.S. limits must take into account both the EBL and the state of the source using simultaneous data at different wavelengths.
The H.E.S.S. AGN observation program is not complete as many proposed candidates have not yet been observed. Further, more time is scheduled for observations of some of the AGN presented here as part of a monitoring effort for blazars. H.E.S.S. has already detected $`\gamma `$-ray emission from four AGN, including one (PKS 2005$``$489) never previously detected in the VHE regime. Clearly the prospects of finding additional VHE-bright AGN are excellent.
###### Acknowledgements.
The support of the Namibian authorities and of the University of Namibia in facilitating the construction and operation of H.E.S.S. is gratefully acknowledged, as is the support by the German Ministry for Education and Research (BMBF), the Max-Planck-Society, the French Ministry for Research, the CNRS-IN2P3 and the Astroparticle Interdisciplinary Programme of the CNRS, the U.K. Particle Physics and Astronomy Research Council (PPARC), the IPNP of the Charles University, the South African Department of Science and Technology and National Research Foundation, and by the University of Namibia. We appreciate the excellent work of the technical support staff in Berlin, Durham, Hamburg, Heidelberg, Palaiseau, Paris, Saclay, and in Namibia in the construction and operation of the equipment. |
warning/0507/hep-th0507177.html | ar5iv | text | # 1 Introduction
## 1 Introduction
Logarithmic conformal field theory is essentially based on the appearance of conformal Jordan cells in the spectrum of fields. We refer to for the first systematic study of logarithmic conformal field theory, and to for recent reviews on the subject. The number of fields making up a conformal Jordan cell is called the rank of the cell. We will focus on conformal Jordan cells of rank two.
We consider the case where the logarithmic fields in the conformal Jordan cells are quasi-primary, and discuss the conformal Ward identities which follow. Without making any simplifying assumptions about the operator-product expansions of the fields, we find the general solutions for two- and three-point functions. Our results thus cover all the possible cases based on primary fields not belonging to conformal Jordan cells, primary fields belonging to conformal Jordan cells, and the logarithmic partner fields completing the conformal Jordan cells.
We also study the generality of two observations made under certain simplifying assumptions. The first observation concerns the expressibility of the correlators in terms of conformal weights with nilpotent parts . This is a non-trivial point as it a priori presumes that the general solutions to the conformal Ward identities factor accordingly. We demonstrate that they do.
The second observation concerns a hierarchical structure for the set of correlators where the links are based on computing derivatives of the correlators with respect to the conformal weights . Also in this case, we find that the basic idea extends from the simpler set-up to our general situation.
This paper proceeds as follows. After a short introduction to the conformal Ward identities, we work out the general solutions for two- and three-point functions. We then affirm the assertions about conformal weights with nilpotent parts and the hierarchical structure. We conclude with some comments on further extensions.
## 2 Correlators in logarithmic conformal field theory
A conformal Jordan cell of rank two consists of two fields: a primary field, $`\mathrm{\Phi }`$, of conformal weight $`\mathrm{\Delta }`$ and its non-primary, ‘logarithmic’ partner field, $`\mathrm{\Psi }`$, on which the Virasoro algebra generated by $`\{L_n\}`$ does not act diagonally. With a conventional relative normalization of the fields, we have
$`[L_n,\mathrm{\Phi }(z)]`$ $`=`$ $`\left(z^{n+1}_z+\mathrm{\Delta }(n+1)z^n\right)\mathrm{\Phi }(z)`$
$`[L_n,\mathrm{\Psi }(z)]`$ $`=`$ $`\left(z^{n+1}_z+\mathrm{\Delta }(n+1)z^n\right)\mathrm{\Psi }(z)+(n+1)z^n\mathrm{\Phi }(z)`$ (1)
It has been suggested by Flohr to describe these fields in a unified way by introducing a nilpotent, yet even, parameter $`\theta `$ satisfying $`\theta ^2=0`$. We will follow this idea here, though use an approach closer to the one employed in . We thus define the field or unified cell
$$\mathrm{{\rm Y}}(z,\theta )=\mathrm{\Phi }(z)+\theta \mathrm{\Psi }(z)$$
(2)
which is seen to be ‘primary’ of conformal weight $`\mathrm{\Delta }+\theta `$ as the commutators (1) are replaced by
$$[L_n,\mathrm{{\rm Y}}(z,\theta )]=\left(z^{n+1}_z+(\mathrm{\Delta }+\theta )(n+1)z^n\right)\mathrm{{\rm Y}}(z,\theta )$$
(3)
A primary field belonging to a conformal Jordan cell is referred to as a ‘cellular’ primary field. A primary field not belonging to a conformal Jordan cell may be represented as $`\mathrm{{\rm Y}}(z,0)`$, and we will reserve this notation for these non-cellular primary fields. To avoid ambiguities, we will therefore refrain from considering unified cells $`\mathrm{{\rm Y}}(z,\theta )`$, as defined in (2), for vanishing $`\theta `$.
### 2.1 Conformal Ward identities
We will consider quasi-primary fields only, ensuring the projective invariance of their correlators constructed by sandwiching the fields between projectively invariant vacua. That is, insertion of any of the three generators $`L_1,L_0,L_1`$ into a correlator annihilates the correlator. When expressed in terms of the differential operators (3), this is known as the conformal Ward identities which are given here for $`𝒩`$-point functions:
$`0`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{𝒩}{}}}_{z_i}\mathrm{{\rm Y}}_1(z_1,\theta _1)\mathrm{}\mathrm{{\rm Y}}_𝒩(z_𝒩,\theta _𝒩)`$
$`0`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{𝒩}{}}}\left(z_i_{z_i}+\mathrm{\Delta }_i+\theta _i\right)\mathrm{{\rm Y}}_1(z_1,\theta _1)\mathrm{}\mathrm{{\rm Y}}_𝒩(z_𝒩,\theta _𝒩)`$
$`0`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{𝒩}{}}}\left(z_i^2_{z_i}+2(\mathrm{\Delta }_i+\theta _i)z_i\right)\mathrm{{\rm Y}}_1(z_1,\theta _1)\mathrm{}\mathrm{{\rm Y}}_𝒩(z_𝒩,\theta _𝒩)`$ (4)
To simplify the notation we introduce the differential operator
$$_1^𝒩=\underset{i=1}{\overset{𝒩}{}}\left(z_i^2_{z_i}+2\mathrm{\Delta }_iz_i\right)$$
(5)
in terms of which the third conformal Ward identity reads
$$0=\left(_1^𝒩+2\underset{i=1}{\overset{𝒩}{}}\theta _iz_i\right)\mathrm{{\rm Y}}_1(z_1,\theta _1)\mathrm{}\mathrm{{\rm Y}}_𝒩(z_𝒩,\theta _𝒩)$$
(6)
It is noted that a correlator satisfying the first and third Ward identities (4) automatically satisfies the second Ward identity. This follows readily from the commutator $`[L_1,L_1]=2L_0`$.
The first conformal Ward identity merely imposes translation invariance on the correlators, allowing us to express them solely in terms of differences, $`z_iz_j`$, between the coordinates.
It is stressed that some solutions for correlators involving non-cellular primary fields $`\mathrm{{\rm Y}}_i(z_i,0)`$ may be lost if one simply sets the corresponding $`\theta _i`$ equal to zero in the solutions for non-vanishing $`\theta _i`$. This will be illustrated in the following.
Before proceeding, let us indicate how one extracts information on the individual correlators from solutions to the conformal Ward identities involving unified cells. In the case of
$$\mathrm{{\rm Y}}_1(z_1,\theta _1)\mathrm{{\rm Y}}_2(z_2,0)\mathrm{{\rm Y}}_3(z_3,\theta _3)$$
(7)
for example, the identity (6) reads
$$0=\left(_1^3+2(\theta _1z_1+\theta _3z_3)\right)\mathrm{{\rm Y}}_1(z_1,\theta _1)\mathrm{{\rm Y}}_2(z_2,0)\mathrm{{\rm Y}}_3(z_3,\theta _3)$$
(8)
A solution to the full set of conformal Ward identities is an expression expandable in $`\theta _1`$ and $`\theta _3`$. The term proportional to $`\theta _1`$ but independent of $`\theta _3`$, for example, should then be identified with $`\mathrm{\Psi }_1(z_1)\mathrm{{\rm Y}}_2(z_2,0)\mathrm{\Phi }_3(z_3)`$.
By construction, and as illustrated by this example, correlators involving unified cells and non-cellular primary fields may thus be regarded as generating-function correlators whose expansions in the nilpotent parameters give the individual correlators involving combinations of cellular primary fields, non-cellular primary fields, and logarithmic fields. Our focus will therefore be on correlators of combinations of unified cells and non-cellular primary fields. To the best of our knowledge, most results found in the literature pertain to correlators involving unified cells only or non-cellular primary fields only, though Ref. does contain a discussion of three-point functions involving so-called twist fields as examples of so-called ‘pre-logarithmic’ fields in the $`c=2`$ conformal field theory. Those particular results are in accordance with our general results. Furthermore, studies of three-point functions involving unified cells only are most often based on a simplifying, though physically motivated, assumption to which we will return in due time.
### 2.2 Two-point functions
We have three situations to analyze, distinguished by the number of unified cells appearing in the correlator. The case with non-cellular primary fields only is as in ordinary conformal field theory and we have the well-known result
$$\mathrm{{\rm Y}}_1(z_1,0)\mathrm{{\rm Y}}_2(z_2,0)\frac{\delta _{\mathrm{\Delta }_1,\mathrm{\Delta }_2}}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2}}$$
(9)
To simplify the notation, we have introduced the standard abbreviation $`z_{ij}=z_iz_j`$.
We now turn to the situation with at least one unified cell (i.e., one or two) in the two-point function. Motivated by the known results for two-point functions of unified cells only, we consider the following common ansatz
$$\mathrm{{\rm Y}}_1(z_1,\theta _1)\mathrm{{\rm Y}}_2(z_2,\theta _2)=\frac{A(\theta _1,\theta _2)+B(\theta _1,\theta _2)\mathrm{ln}z_{12}}{z_{12}^{2h}}$$
(10)
where the dependence of the structure constants $`A`$ and $`B`$ on $`\theta _1`$ or $`\theta _2`$ vanishes if we consider the non-cellular primary field $`\mathrm{{\rm Y}}_1(z_1,0)`$ or $`\mathrm{{\rm Y}}_2(z_2,0)`$, respectively. The general expansion of $`A`$ reads
$$A(\theta _1,\theta _2)=A^0+A^1\theta _1+A^2\theta _2+A^{12}\theta _1\theta _2$$
(11)
and similarly for $`B`$. Imposing (6) results in
$`\mathrm{{\rm Y}}_1(z_1,\theta _1)\mathrm{{\rm Y}}_2(z_2,0)`$ $`=`$ $`\delta _{\mathrm{\Delta }_1,\mathrm{\Delta }_2}{\displaystyle \frac{A^1\theta _1}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2}}}`$
$`\mathrm{{\rm Y}}_1(z_1,0)\mathrm{{\rm Y}}_2(z_2,\theta _2)`$ $`=`$ $`\delta _{\mathrm{\Delta }_1,\mathrm{\Delta }_2}{\displaystyle \frac{A^2\theta _2}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2}}}`$
$`\mathrm{{\rm Y}}_1(z_1,\theta _1)\mathrm{{\rm Y}}_2(z_2,\theta _2)`$ $`=`$ $`\delta _{\mathrm{\Delta }_1,\mathrm{\Delta }_2}{\displaystyle \frac{A^1\delta _{A^1,A^2}\left(\theta _1+\theta _22\theta _1\theta _2\mathrm{ln}z_{12}\right)+A^{12}\theta _1\theta _2}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2}}}`$ (12)
which in terms of individual two-point functions corresponds to
$`\mathrm{\Phi }(z_1)\mathrm{{\rm Y}}(z_2,0)`$ $`=`$ $`\mathrm{{\rm Y}}(z_1,0)\mathrm{\Phi }(z_2)=\mathrm{\Phi }_1(z_1)\mathrm{\Phi }_2(z_2)=0`$
$`\mathrm{\Psi }(z_1)\mathrm{{\rm Y}}(z_2,0)`$ $``$ $`{\displaystyle \frac{\delta _{\mathrm{\Delta }_1,\mathrm{\Delta }_2}}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2}}}`$
$`\mathrm{{\rm Y}}(z_1,0)\mathrm{\Psi }(z_2)`$ $``$ $`{\displaystyle \frac{\delta _{\mathrm{\Delta }_1,\mathrm{\Delta }_2}}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2}}}`$
$`\mathrm{\Phi }_1(z_1)\mathrm{\Psi }_2(z_2)`$ $`=`$ $`\mathrm{\Psi }_1(z_1)\mathrm{\Phi }_2(z_2)=\delta _{\mathrm{\Delta }_1,\mathrm{\Delta }_2}{\displaystyle \frac{A^1}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2}}}`$
$`\mathrm{\Psi }_1(z_1)\mathrm{\Psi }_2(z_2)`$ $`=`$ $`\delta _{\mathrm{\Delta }_1,\mathrm{\Delta }_2}{\displaystyle \frac{A^{12}2A^1\mathrm{ln}z_{12}}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2}}}`$ (13)
Explicit relations similar to the one between $`A^1`$ and $`A^2`$ represented by the delta function in (12) will be omitted in the following. As indicated above, the solution (9) would have been lost if one were to set $`\theta _1=\theta _2=0`$ in (12), whereas the first two solutions in (12) neatly follow from the last solution in (12) if one sets $`\theta _2=0`$ or $`\theta _1=0`$, respectively.
### 2.3 Three-point functions
We now have four situations to analyze, again characterized by the number of unified cells appearing in the correlator. As for two-point functions, the case with non-cellular primary fields only is as in ordinary conformal field theory and we have the well-known result
$$\mathrm{{\rm Y}}_1(z_1,0)\mathrm{{\rm Y}}_2(z_2,0)\mathrm{{\rm Y}}_3(z_3,0)\frac{1}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }_3}z_{23}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2+\mathrm{\Delta }_3}z_{13}^{\mathrm{\Delta }_1\mathrm{\Delta }_2+\mathrm{\Delta }_3}}$$
(14)
For the combined three-point functions, associativity and the results on two-point functions suggest that we consider the following ansatz
$`\mathrm{{\rm Y}}_1(z_1,\theta _1)\mathrm{{\rm Y}}_2(z_2,\theta _2)\mathrm{{\rm Y}}_3(z_3,\theta _3)`$ (15)
$`=`$ $`\{A(\theta _1,\theta _2,\theta _3)+B_1(\theta _1,\theta _2,\theta _3)\mathrm{ln}z_{12}+B_2(\theta _1,\theta _2,\theta _3)\mathrm{ln}z_{23}+B_3(\theta _1,\theta _2,\theta _3)\mathrm{ln}z_{13}`$
$`+D_{11}(\theta _1,\theta _2,\theta _3)\mathrm{ln}^2z_{12}+D_{12}(\theta _1,\theta _2,\theta _3)\mathrm{ln}z_{12}\mathrm{ln}z_{23}+D_{13}(\theta _1,\theta _2,\theta _3)\mathrm{ln}z_{12}\mathrm{ln}z_{13}`$
$`+D_{22}(\theta _1,\theta _2,\theta _3)\mathrm{ln}^2z_{23}+D_{23}(\theta _1,\theta _2,\theta _3)\mathrm{ln}z_{23}\mathrm{ln}z_{13}+D_{33}(\theta _1,\theta _2,\theta _3)\mathrm{ln}^2z_{13}\}`$
$`\times `$ $`\left(z_{12}^{h_1}z_{23}^{h_2}z_{13}^{h_3}\right)`$
Here $`h_i`$ is $`\theta `$-independent while
$$A(\theta _1,\theta _2,\theta _3)=A^0+A^1\theta _1+A^2\theta _2+A^3\theta _3+A^{12}\theta _1\theta _2+A^{23}\theta _2\theta _3+A^{13}\theta _1\theta _3+A^{123}\theta _1\theta _2\theta _3$$
(16)
and similarly for $`B_i`$ and $`D_{ij}`$. Imposing the Ward identities (i.e., on this ansatz, (6) suffices), corresponds to the following conditions, obtained from considering the part independent of logarithms
$`0`$ $`=`$ $`(2\mathrm{\Delta }_1h_1h_3+2\theta _1)A(\theta _1,\theta _2,\theta _3)+B_1(\theta _1,\theta _2,\theta _3)+B_3(\theta _1,\theta _2,\theta _3)`$
$`0`$ $`=`$ $`(2\mathrm{\Delta }_2h_1h_2+2\theta _2)A(\theta _1,\theta _2,\theta _3)+B_1(\theta _1,\theta _2,\theta _3)+B_2(\theta _1,\theta _2,\theta _3)`$
$`0`$ $`=`$ $`(2\mathrm{\Delta }_3h_2h_3+2\theta _3)A(\theta _1,\theta _2,\theta _3)+B_2(\theta _1,\theta _2,\theta _3)+B_3(\theta _1,\theta _2,\theta _3)`$ (17)
the part linear in logarithms
$`0`$ $`=`$ $`(2\mathrm{\Delta }_1h_1h_3+2\theta _1)B_1(\theta _1,\theta _2,\theta _3)+2D_{11}(\theta _1,\theta _2,\theta _3)+D_{13}(\theta _1,\theta _2,\theta _3)`$
$`0`$ $`=`$ $`(2\mathrm{\Delta }_2h_1h_2+2\theta _2)B_1(\theta _1,\theta _2,\theta _3)+2D_{11}(\theta _1,\theta _2,\theta _3)+D_{12}(\theta _1,\theta _2,\theta _3)`$
$`0`$ $`=`$ $`(2\mathrm{\Delta }_3h_2h_3+2\theta _3)B_1(\theta _1,\theta _2,\theta _3)+D_{12}(\theta _1,\theta _2,\theta _3)+D_{13}(\theta _1,\theta _2,\theta _3)`$
$`0`$ $`=`$ $`(2\mathrm{\Delta }_1h_1h_3+2\theta _1)B_2(\theta _1,\theta _2,\theta _3)+D_{12}(\theta _1,\theta _2,\theta _3)+D_{23}(\theta _1,\theta _2,\theta _3)`$
$`0`$ $`=`$ $`(2\mathrm{\Delta }_2h_1h_2+2\theta _2)B_2(\theta _1,\theta _2,\theta _3)+D_{12}(\theta _1,\theta _2,\theta _3)+2D_{22}(\theta _1,\theta _2,\theta _3)`$
$`0`$ $`=`$ $`(2\mathrm{\Delta }_3h_2h_3+2\theta _3)B_2(\theta _1,\theta _2,\theta _3)+2D_{22}(\theta _1,\theta _2,\theta _3)+D_{23}(\theta _1,\theta _2,\theta _3)`$
$`0`$ $`=`$ $`(2\mathrm{\Delta }_1h_1h_3+2\theta _1)B_3(\theta _1,\theta _2,\theta _3)+D_{13}(\theta _1,\theta _2,\theta _3)+2D_{33}(\theta _1,\theta _2,\theta _3)`$
$`0`$ $`=`$ $`(2\mathrm{\Delta }_2h_1h_2+2\theta _2)B_3(\theta _1,\theta _2,\theta _3)+D_{13}(\theta _1,\theta _2,\theta _3)+D_{23}(\theta _1,\theta _2,\theta _3)`$
$`0`$ $`=`$ $`(2\mathrm{\Delta }_3h_2h_3+2\theta _3)B_3(\theta _1,\theta _2,\theta _3)+D_{23}(\theta _1,\theta _2,\theta _3)+2D_{33}(\theta _1,\theta _2,\theta _3)`$ (18)
and the part quadratic in logarithms
$`0`$ $`=`$ $`(2\mathrm{\Delta }_1h_1h_3+2\theta _1)D_{ij}(\theta _1,\theta _2,\theta _3),1ij3`$
$`0`$ $`=`$ $`(2\mathrm{\Delta }_2h_1h_2+2\theta _2)D_{ij}(\theta _1,\theta _2,\theta _3),1ij3`$
$`0`$ $`=`$ $`(2\mathrm{\Delta }_3h_2h_3+2\theta _3)D_{ij}(\theta _1,\theta _2,\theta _3),1ij3`$ (19)
These apply whether or not the individual $`\theta `$s vanish, even if $`\theta _1=\theta _2=\theta _3=0`$ as in (14). In the further analysis, one should distinguish between the different numbers of unified cells, that is, the numbers of non-vanishing $`\theta `$s. Also, it is understood that an $`A^1`$, for example, appearing in the study of one set of correlators (related through one or several Jordan-cell structures) a priori is independent of an $`A^1`$ appearing in a different set (not related to the former through a Jordan-cell structure).
Now, it is not hard to show that we in every case have
$$h_1=\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }_3,h_2=\mathrm{\Delta }_1+\mathrm{\Delta }_2+\mathrm{\Delta }_3,h_3=\mathrm{\Delta }_1\mathrm{\Delta }_2+\mathrm{\Delta }_3$$
(20)
meaning that these identities apply to all combinations of vanishing or non-vanishing $`\theta `$s. In the case where $`\theta _1=\theta _2=\theta _3=0`$, there is only one solution to the conditions (17-19) and one recovers (14) with $`A^0`$ as the proportionality constant.
In the case where $`\theta _10`$ while $`\theta _2=\theta _3=0`$, we find
$$\mathrm{{\rm Y}}_1(z_1,\theta _1)\mathrm{{\rm Y}}_2(z_2,0)\mathrm{{\rm Y}}_3(z_3,0)=\frac{A^0+A^1\theta _1+A^0\theta _1(\mathrm{ln}z_{12}+\mathrm{ln}z_{23}\mathrm{ln}z_{13})}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }_3}z_{23}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2+\mathrm{\Delta }_3}z_{13}^{\mathrm{\Delta }_1\mathrm{\Delta }_2+\mathrm{\Delta }_3}}$$
(21)
which in terms of the individual correlators reads
$`\mathrm{\Phi }_1(z_1)\mathrm{{\rm Y}}_2(z_2,0)\mathrm{{\rm Y}}_3(z_3,0)`$ $`=`$ $`{\displaystyle \frac{A^0}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }_3}z_{23}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2+\mathrm{\Delta }_3}z_{13}^{\mathrm{\Delta }_1\mathrm{\Delta }_2+\mathrm{\Delta }_3}}}`$
$`\mathrm{\Psi }_1(z_1)\mathrm{{\rm Y}}_2(z_2,0)\mathrm{{\rm Y}}_3(z_3,0)`$ $`=`$ $`{\displaystyle \frac{A^1A^0\mathrm{ln}\frac{z_{12}z_{13}}{z_{23}}}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }_3}z_{23}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2+\mathrm{\Delta }_3}z_{13}^{\mathrm{\Delta }_1\mathrm{\Delta }_2+\mathrm{\Delta }_3}}}`$ (22)
The other two cases with only one unified cell are treated similarly and the corresponding correlators may be obtained from (21) and (22) by appropriately permuting the indices. We note that there in each case are two a priori independent structure constants. Before commenting on the structure of these results, let us complete the analysis of the conditions (17-19).
In the case where $`\theta _1,\theta _20`$ while $`\theta _3=0`$, we find
$`\mathrm{{\rm Y}}_1(z_1,\theta _1)\mathrm{{\rm Y}}_2(z_2,\theta _2)\mathrm{{\rm Y}}_3(z_3,0)`$ (23)
$`=`$ $`\{A^0+A^1\theta _1+A^2\theta _2+A^{12}\theta _1\theta _2+(A^0\theta _1A^0\theta _2(A^1+A^2)\theta _1\theta _2)\mathrm{ln}z_{12}`$
$`+`$ $`\left(A^0\theta _1A^0\theta _2+(A^1+A^2)\theta _1\theta _2\right)\mathrm{ln}z_{23}+\left(A^0\theta _1+A^0\theta _2+(A^1A^2)\theta _1\theta _2\right)\mathrm{ln}z_{13}`$
$`+`$ $`A^0\theta _1\theta _2(\mathrm{ln}^2z_{12}\mathrm{ln}^2z_{23}\mathrm{ln}^2z_{13}+2\mathrm{ln}z_{23}\mathrm{ln}z_{13})\}`$
$`\times `$ $`z_{12}^{\mathrm{\Delta }_1\mathrm{\Delta }_2+\mathrm{\Delta }_3}z_{23}^{\mathrm{\Delta }_1\mathrm{\Delta }_2\mathrm{\Delta }_3}z_{13}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }_3}`$
which in terms of the individual correlators reads
$`\mathrm{\Phi }_1(z_1)\mathrm{\Phi }_2(z_2)\mathrm{{\rm Y}}_3(z_3,0)`$ $`=`$ $`{\displaystyle \frac{A^0}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }_3}z_{23}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2+\mathrm{\Delta }_3}z_{13}^{\mathrm{\Delta }_1\mathrm{\Delta }_2+\mathrm{\Delta }_3}}}`$
$`\mathrm{\Psi }_1(z_1)\mathrm{\Phi }_2(z_2)\mathrm{{\rm Y}}_3(z_3,0)`$ $`=`$ $`{\displaystyle \frac{A^1A^0\mathrm{ln}\frac{z_{12}z_{13}}{z_{23}}}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }_3}z_{23}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2+\mathrm{\Delta }_3}z_{13}^{\mathrm{\Delta }_1\mathrm{\Delta }_2+\mathrm{\Delta }_3}}}`$
$`\mathrm{\Phi }_1(z_1)\mathrm{\Psi }_2(z_2)\mathrm{{\rm Y}}_3(z_3,0)`$ $`=`$ $`{\displaystyle \frac{A^2A^0\mathrm{ln}\frac{z_{12}z_{23}}{z_{13}}}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }_3}z_{23}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2+\mathrm{\Delta }_3}z_{13}^{\mathrm{\Delta }_1\mathrm{\Delta }_2+\mathrm{\Delta }_3}}}`$
$`\mathrm{\Psi }_1(z_1)\mathrm{\Psi }_2(z_2)\mathrm{{\rm Y}}_3(z_3,0)`$ $`=`$ $`{\displaystyle \frac{A^{12}A^1\mathrm{ln}\frac{z_{12}z_{23}}{z_{13}}A^2\mathrm{ln}\frac{z_{12}z_{13}}{z_{23}}+A^0\mathrm{ln}\frac{z_{12}z_{23}}{z_{13}}\mathrm{ln}\frac{z_{12}z_{13}}{z_{23}}}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }_3}z_{23}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2+\mathrm{\Delta }_3}z_{13}^{\mathrm{\Delta }_1\mathrm{\Delta }_2+\mathrm{\Delta }_3}}}`$ (24)
The other two cases with two unified cells are treated similarly and the corresponding correlators may be obtained from (23) and (24) by an appropriate permutation of the indices. We note that there in each case are four a priori independent structure constants.
In the case with three unified cells, that is, $`\theta _1,\theta _2,\theta _30`$, we find
$`\mathrm{{\rm Y}}_1(z_1,\theta _1)\mathrm{{\rm Y}}_2(z_2,\theta _2)\mathrm{{\rm Y}}_3(z_3,\theta _3)`$ (25)
$`=`$ $`\{A^1\theta _1+A^2\theta _2+A^3\theta _3+A^{12}\theta _1\theta _2+A^{23}\theta _2\theta _3+A^{13}\theta _1\theta _3+A^{123}\theta _1\theta _2\theta _3`$
$`+`$ $`((A^1A^2)\theta _1\theta _2+(A^2A^3)\theta _2\theta _3+(A^1A^3)\theta _1\theta _3`$
$`+(A^{12}A^{23}A^{13})\theta _1\theta _2\theta _3)\mathrm{ln}z_{12}`$
$`+`$ $`((A^1+A^2)\theta _1\theta _2+(A^2A^3)\theta _2\theta _3+(A^1+A^3)\theta _1\theta _3`$
$`+(A^{12}+A^{23}A^{13})\theta _1\theta _2\theta _3)\mathrm{ln}z_{23}`$
$`+`$ $`((A^1A^2)\theta _1\theta _2+(A^2+A^3)\theta _2\theta _3+(A^1A^3)\theta _1\theta _3`$
$`+(A^{12}A^{23}+A^{13})\theta _1\theta _2\theta _3)\mathrm{ln}z_{13}`$
$`+`$ $`(A^1A^2+A^3)\theta _1\theta _2\theta _3\mathrm{ln}^2z_{12}+2A^2\theta _1\theta _2\theta _3\mathrm{ln}z_{12}\mathrm{ln}z_{23}`$
$`+`$ $`2A^1\theta _1\theta _2\theta _3\mathrm{ln}z_{12}\mathrm{ln}z_{13}+(A^1A^2A^3)\theta _1\theta _2\theta _3\mathrm{ln}^2z_{23}`$
$`+`$ $`2A^3\theta _1\theta _2\theta _3\mathrm{ln}z_{23}\mathrm{ln}z_{13}+(A^1+A^2A^3)\theta _1\theta _2\theta _3\mathrm{ln}^2z_{13}\}`$
$`\times `$ $`z_{12}^{\mathrm{\Delta }_1\mathrm{\Delta }_2+\mathrm{\Delta }_3}z_{23}^{\mathrm{\Delta }_1\mathrm{\Delta }_2\mathrm{\Delta }_3}z_{13}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }_3}`$
which in terms of the individual correlators reads
$`\mathrm{\Phi }_1(z_1)\mathrm{\Phi }_2(z_2)\mathrm{\Phi }_3(z_3)`$ $`=`$ $`0`$
$`\mathrm{\Psi }_1(z_1)\mathrm{\Phi }_2(z_2)\mathrm{\Phi }_3(z_3)`$ $`=`$ $`{\displaystyle \frac{A^1}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }_3}z_{23}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2+\mathrm{\Delta }_3}z_{13}^{\mathrm{\Delta }_1\mathrm{\Delta }_2+\mathrm{\Delta }_3}}}`$
$`\mathrm{\Phi }_1(z_1)\mathrm{\Psi }_2(z_2)\mathrm{\Phi }_3(z_3)`$ $`=`$ $`{\displaystyle \frac{A^2}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }_3}z_{23}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2+\mathrm{\Delta }_3}z_{13}^{\mathrm{\Delta }_1\mathrm{\Delta }_2+\mathrm{\Delta }_3}}}`$
$`\mathrm{\Phi }_1(z_1)\mathrm{\Phi }_2(z_2)\mathrm{\Psi }_3(z_3)`$ $`=`$ $`{\displaystyle \frac{A^3}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }_3}z_{23}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2+\mathrm{\Delta }_3}z_{13}^{\mathrm{\Delta }_1\mathrm{\Delta }_2+\mathrm{\Delta }_3}}}`$
$`\mathrm{\Psi }_1(z_1)\mathrm{\Psi }_2(z_2)\mathrm{\Phi }_3(z_3)`$ $`=`$ $`{\displaystyle \frac{A^{12}A^1\mathrm{ln}\frac{z_{12}z_{23}}{z_{13}}A^2\mathrm{ln}\frac{z_{12}z_{13}}{z_{23}}}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }_3}z_{23}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2+\mathrm{\Delta }_3}z_{13}^{\mathrm{\Delta }_1\mathrm{\Delta }_2+\mathrm{\Delta }_3}}}`$
$`\mathrm{\Phi }_1(z_1)\mathrm{\Psi }_2(z_2)\mathrm{\Psi }_3(z_3)`$ $`=`$ $`{\displaystyle \frac{A^{23}A^2\mathrm{ln}\frac{z_{23}z_{13}}{z_{12}}A^3\mathrm{ln}\frac{z_{12}z_{23}}{z_{13}}}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }_3}z_{23}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2+\mathrm{\Delta }_3}z_{13}^{\mathrm{\Delta }_1\mathrm{\Delta }_2+\mathrm{\Delta }_3}}}`$
$`\mathrm{\Psi }_1(z_1)\mathrm{\Phi }_2(z_2)\mathrm{\Psi }_3(z_3)`$ $`=`$ $`{\displaystyle \frac{A^{13}A^1\mathrm{ln}\frac{z_{23}z_{13}}{z_{12}}A^3\mathrm{ln}\frac{z_{12}z_{13}}{z_{23}}}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }_3}z_{23}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2+\mathrm{\Delta }_3}z_{13}^{\mathrm{\Delta }_1\mathrm{\Delta }_2+\mathrm{\Delta }_3}}}`$
$`\mathrm{\Psi }_1(z_1)\mathrm{\Psi }_2(z_2)\mathrm{\Psi }_3(z_3)`$ $`=`$ $`(A^{123}A^{12}\mathrm{ln}{\displaystyle \frac{z_{23}z_{13}}{z_{12}}}A^{23}\mathrm{ln}{\displaystyle \frac{z_{12}z_{13}}{z_{23}}}A^{13}\mathrm{ln}{\displaystyle \frac{z_{12}z_{23}}{z_{13}}}`$
$`+A^1\mathrm{ln}{\displaystyle \frac{z_{12}z_{23}}{z_{13}}}\mathrm{ln}{\displaystyle \frac{z_{23}z_{13}}{z_{12}}}+A^2\mathrm{ln}{\displaystyle \frac{z_{12}z_{13}}{z_{23}}}\mathrm{ln}{\displaystyle \frac{z_{23}z_{13}}{z_{12}}}`$
$`+A^3\mathrm{ln}{\displaystyle \frac{z_{12}z_{23}}{z_{13}}}\mathrm{ln}{\displaystyle \frac{z_{12}z_{13}}{z_{23}}})z_{12}^{\mathrm{\Delta }_1\mathrm{\Delta }_2+\mathrm{\Delta }_3}z_{23}^{\mathrm{\Delta }_1\mathrm{\Delta }_2\mathrm{\Delta }_3}z_{13}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }_3}`$
We note that there are seven a priori independent structure constants. In the literature, on the other hand, one deals with three structure constants only (see , for example). This discrepancy is due to an assumption usually made in available studies of three-point functions. It concerns a particular property of the cellular primary fields which we will address presently.
Primary fields are called proper primary if their operator-product expansions with each other cannot produce a logarithmic field. It is argued in (see also ) that correlators not involving improper primary fields satisfy
$`\mathrm{\Psi }_1(z_1)\mathrm{\Phi }_2(z_2)\mathrm{}\mathrm{\Phi }_𝒩(z_𝒩)`$ $`=`$ $`\mathrm{\Phi }_1(z_1)\mathrm{\Psi }_2(z_2)\mathrm{\Phi }_3(z_3)\mathrm{}\mathrm{\Phi }_𝒩(z_𝒩)`$ (27)
$`\mathrm{}`$
$`=`$ $`\mathrm{\Phi }_1(z_1)\mathrm{}\mathrm{\Phi }_{𝒩1}(z_{𝒩1})\mathrm{\Psi }_𝒩(z_𝒩)`$
in particular, and that the general form of the individual three-point functions of logarithmic fields and cellular primary fields hence read
$`\mathrm{\Phi }_1(z_1)\mathrm{\Phi }_2(z_2)\mathrm{\Phi }_3(z_3)`$ $`=`$ $`0`$
$`\mathrm{\Psi }_1(z_1)\mathrm{\Phi }_2(z_2)\mathrm{\Phi }_3(z_3)`$ $`=`$ $`\mathrm{\Phi }_1(z_1)\mathrm{\Psi }_2(z_2)\mathrm{\Phi }_3(z_3)=\mathrm{\Phi }_1(z_1)\mathrm{\Phi }_2(z_2)\mathrm{\Psi }_3(z_3)`$
$`=`$ $`{\displaystyle \frac{C_{123;1}}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }_3}z_{23}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2+\mathrm{\Delta }_3}z_{13}^{\mathrm{\Delta }_1\mathrm{\Delta }_2+\mathrm{\Delta }_3}}}`$
$`\mathrm{\Psi }_1(z_1)\mathrm{\Psi }_2(z_2)\mathrm{\Phi }_3(z_3)`$ $`=`$ $`{\displaystyle \frac{C_{123;2}2C_{123;1}\mathrm{ln}z_{12}}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }_3}z_{23}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2+\mathrm{\Delta }_3}z_{13}^{\mathrm{\Delta }_1\mathrm{\Delta }_2+\mathrm{\Delta }_3}}}`$
$`\mathrm{\Phi }_1(z_1)\mathrm{\Psi }_2(z_2)\mathrm{\Psi }_3(z_3)`$ $`=`$ $`{\displaystyle \frac{C_{123;2}2C_{123;1}\mathrm{ln}z_{23}}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }_3}z_{23}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2+\mathrm{\Delta }_3}z_{13}^{\mathrm{\Delta }_1\mathrm{\Delta }_2+\mathrm{\Delta }_3}}}`$
$`\mathrm{\Psi }_1(z_1)\mathrm{\Phi }_2(z_2)\mathrm{\Psi }_3(z_3)`$ $`=`$ $`{\displaystyle \frac{C_{123;2}2C_{123;1}\mathrm{ln}z_{13}}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }_3}z_{23}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2+\mathrm{\Delta }_3}z_{13}^{\mathrm{\Delta }_1\mathrm{\Delta }_2+\mathrm{\Delta }_3}}}`$
$`\mathrm{\Psi }_1(z_1)\mathrm{\Psi }_2(z_2)\mathrm{\Psi }_3(z_3)`$ $`=`$ $`\{C_{123;3}C_{123;2}(\mathrm{ln}z_{12}+\mathrm{ln}z_{23}+\mathrm{ln}z_{13})`$ (28)
$`+C_{123;1}(2\mathrm{ln}z_{12}\mathrm{ln}z_{23}+2\mathrm{ln}z_{12}\mathrm{ln}z_{13}+2\mathrm{ln}z_{23}\mathrm{ln}z_{13}`$
$`\mathrm{ln}^2z_{12}\mathrm{ln}^2z_{23}\mathrm{ln}^2z_{13})\}`$
$`\times `$ $`z_{12}^{\mathrm{\Delta }_1\mathrm{\Delta }_2+\mathrm{\Delta }_3}z_{23}^{\mathrm{\Delta }_1\mathrm{\Delta }_2\mathrm{\Delta }_3}z_{13}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }_3}`$
We may recover this result from (LABEL:33ind) by relating the structure constants appearing there as
$$A^1=A^2=A^3,A^{12}=A^{23}=A^{13}$$
(29)
in which case the correlators (LABEL:33ind) are seen to reduce to (28) with $`C_{123;1}=A^1`$, $`C_{123;2}=A^{12}`$, and $`C_{123;3}=A^{123}`$.
Regarding the reduction in the number of unified cells in a three-point function, it is observed that setting $`\theta _3=0`$ in (25) does not reproduce the full expression (23) but only the part independent of $`A^0`$. Setting $`\theta _2=0`$ in (23) or $`\theta _1=0`$ in (21), on the other hand, neatly reproduces the expressions (21) and (14), respectively.
According to the general results above, a logarithmic singularity may appear in a three-point function involving only one logarithmic field as long as at least one of the other two (primary) fields is non-cellular. This is in contrast to the situation based on conformal Jordan cells only, where at least two logarithmic fields are required to have a logarithmic singularity. Likewise, a singularity quadratic in logarithms may appear in a three-point function with two logarithmic fields and one non-cellular primary field, while such a singularity cannot appear if the primary field is cellular.
### 2.4 In terms of weights with nilpotent parts
It has been discussed how the correlators of unified cells only may be represented compactly if one considers the nilpotent parameter $`\theta _i`$ as part of a generalized conformal weight given by $`\mathrm{\Delta }_i+\theta _i`$ . A general version of this assertion is of course very natural from the point of view of the extended Virasoro action (3). It nevertheless presumes that the general solution to the conformal Ward identities may be factored accordingly. This has been shown to be the case when the simplifying assumption about the cellular primary fields being proper primary is imposed. The extension to our general set-up is discussed in the following and is found to affirm the assertion.
The two-point functions may thus be represented as
$`\mathrm{{\rm Y}}_1(z_1,0)\mathrm{{\rm Y}}_2(z_2,0)`$ $`=`$ $`\delta _{\mathrm{\Delta }_1,\mathrm{\Delta }_2}{\displaystyle \frac{A^0}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2}}}`$
$`\mathrm{{\rm Y}}_1(z_1,\theta _1)\mathrm{{\rm Y}}_2(z_2,0)`$ $`=`$ $`\delta _{\mathrm{\Delta }_1,\mathrm{\Delta }_2}{\displaystyle \frac{A^1\theta _1}{z_{12}^{(\mathrm{\Delta }_1+\theta _1)+\mathrm{\Delta }_2}}}`$
$`\mathrm{{\rm Y}}_1(z_1,\theta _1)\mathrm{{\rm Y}}_2(z_2,\theta _2)`$ $`=`$ $`\delta _{\mathrm{\Delta }_1,\mathrm{\Delta }_2}{\displaystyle \frac{A^1(\theta _1+\theta _2)+A^{12}\theta _1\theta _2}{z_{12}^{(\mathrm{\Delta }_1+\theta _1)+(\mathrm{\Delta }_2+\theta _2)}}}`$ (30)
The similar expression for the correlator $`\mathrm{{\rm Y}}_1(z_1,0)\mathrm{{\rm Y}}_2(z_2,\theta _2)`$ is obtained from the second one by interchanging the indices.
It is straightforward to verify that the three-point functions may be represented as
$`\mathrm{{\rm Y}}_1(z_1,0)\mathrm{{\rm Y}}_2(z_2,0)\mathrm{{\rm Y}}_3(z_3,0)`$ $`=`$ $`{\displaystyle \frac{A^0}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }_3}z_{23}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2+\mathrm{\Delta }_3}z_{13}^{\mathrm{\Delta }_1\mathrm{\Delta }_2+\mathrm{\Delta }_3}}}`$
$`\mathrm{{\rm Y}}_1(z_1,\theta _1)\mathrm{{\rm Y}}_2(z_2,0)\mathrm{{\rm Y}}_3(z_3,0)`$ $`=`$ $`{\displaystyle \frac{A^0+A^1\theta _1}{z_{12}^{(\mathrm{\Delta }_1+\theta _1)+\mathrm{\Delta }_2\mathrm{\Delta }_3}z_{23}^{(\mathrm{\Delta }_1+\theta _1)+\mathrm{\Delta }_2+\mathrm{\Delta }_3}z_{13}^{(\mathrm{\Delta }_1+\theta _1)\mathrm{\Delta }_2+\mathrm{\Delta }_3}}}`$ (31)
and
$`\mathrm{{\rm Y}}_1(z_1,\theta _1)\mathrm{{\rm Y}}_2(z_2,\theta _2)\mathrm{{\rm Y}}_3(z_3,0)`$
$`={\displaystyle \frac{A^0+A^1\theta _1+A^2\theta _2+A^{12}\theta _1\theta _2}{z_{12}^{(\mathrm{\Delta }_1+\theta _1)+(\mathrm{\Delta }_2+\theta _2)\mathrm{\Delta }_3}z_{23}^{(\mathrm{\Delta }_1+\theta _1)+(\mathrm{\Delta }_2+\theta _2)+\mathrm{\Delta }_3}z_{13}^{(\mathrm{\Delta }_1+\theta _1)(\mathrm{\Delta }_2+\theta _2)+\mathrm{\Delta }_3}}}`$
$`\mathrm{{\rm Y}}_1(z_1,\theta _1)\mathrm{{\rm Y}}_2(z_2,\theta _2)\mathrm{{\rm Y}}_3(z_3,\theta _3)`$
$`={\displaystyle \frac{A^1\theta _1+A^2\theta _2+A^3\theta _3+A^{12}\theta _1\theta _2+A^{23}\theta _2\theta _3+A^{13}\theta _1\theta _3+A^{123}\theta _1\theta _2\theta _3}{z_{12}^{(\mathrm{\Delta }_1+\theta _1)+(\mathrm{\Delta }_2+\theta _2)(\mathrm{\Delta }_3+\theta _3)}z_{23}^{(\mathrm{\Delta }_1+\theta _1)+(\mathrm{\Delta }_2+\theta _2)+(\mathrm{\Delta }_3+\theta _3)}z_{13}^{(\mathrm{\Delta }_1+\theta _1)(\mathrm{\Delta }_2+\theta _2)+(\mathrm{\Delta }_3+\theta _3)}}}`$ (32)
The remaining four combinations are obtained by appropriate permutations in the indices.
As already indicated, it is not clear a priori that the general solutions to the conformal Ward identities (4) based on the ansätze (10) and (15) reduce to expressions which may be factored as in (30), (31) and (32). Our analysis has demonstrated that this is indeed the case.
### 2.5 Derivatives with respect to the conformal weights
Acting on either
$$W_2=\frac{\delta _{\mathrm{\Delta }_1,\mathrm{\Delta }_2}}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2}}$$
(33)
or
$$W_3=\frac{1}{z_{12}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }_3}z_{23}^{\mathrm{\Delta }_1+\mathrm{\Delta }_2+\mathrm{\Delta }_3}z_{13}^{\mathrm{\Delta }_1\mathrm{\Delta }_2+\mathrm{\Delta }_3}}$$
(34)
we may substitute derivatives with respect to the conformal weights by multiplicative factors according to
$$_{\mathrm{\Delta }_1}=_{\mathrm{\Delta }_2}2\mathrm{ln}z_{12}$$
(35)
or
$$_{\mathrm{\Delta }_1}\mathrm{ln}\frac{z_{12}z_{13}}{z_{23}},_{\mathrm{\Delta }_2}\mathrm{ln}\frac{z_{12}z_{23}}{z_{13}},_{\mathrm{\Delta }_3}\mathrm{ln}\frac{z_{23}z_{13}}{z_{12}}$$
(36)
respectively. This simple observation allows us to represent the correlators involving logarithmic fields as follows:
$`\mathrm{\Psi }_1(z_1)\mathrm{{\rm Y}}_2(z_2,0)`$ $`=`$ $`A^1W_2`$
$`\mathrm{\Psi }_1(z_1)\mathrm{\Phi }_2(z_2)`$ $`=`$ $`A^1W_2`$
$`\mathrm{\Psi }_1(z_1)\mathrm{\Psi }_2(z_2)`$ $`=`$ $`\left(A^{12}+A^2_{\mathrm{\Delta }_1}+A^1_{\mathrm{\Delta }_2}\right)W_2`$
$`\mathrm{\Psi }_1(z_1)\mathrm{{\rm Y}}_2(z_2,0)\mathrm{{\rm Y}}_3(z_3,0)`$ $`=`$ $`\left(A^1+A^0_{\mathrm{\Delta }_1}\right)W_3`$
$`\mathrm{\Psi }_1(z_1)\mathrm{\Phi }_2(z_2)\mathrm{{\rm Y}}_3(z_3,0)`$ $`=`$ $`\left(A^1+A^0_{\mathrm{\Delta }_1}\right)W_3`$
$`\mathrm{\Psi }_1(z_1)\mathrm{\Psi }_2(z_2)\mathrm{{\rm Y}}_3(z_3,0)`$ $`=`$ $`\left(A^{12}+A^1_{\mathrm{\Delta }_2}+A^2_{\mathrm{\Delta }_1}+A^0_{\mathrm{\Delta }_1}_{\mathrm{\Delta }_2}\right)W_3`$
$`\mathrm{\Psi }_1(z_1)\mathrm{\Phi }_2(z_2)\mathrm{\Phi }_3(z_3)`$ $`=`$ $`A^1W_3`$
$`\mathrm{\Psi }_1(z_1)\mathrm{\Psi }_2(z_2)\mathrm{\Phi }_3(z_3)`$ $`=`$ $`\left(A^{12}+A^2_{\mathrm{\Delta }_1}+A^1_{\mathrm{\Delta }_2}\right)W_3`$
$`\mathrm{\Psi }_1(z_1)\mathrm{\Psi }_2(z_2)\mathrm{\Psi }_3(z_3)`$ $`=`$ $`(A^{123}+A^{23}_{\mathrm{\Delta }_1}+A^{13}_{\mathrm{\Delta }_2}+A^{12}_{\mathrm{\Delta }_3}`$ (37)
$`+A^3_{\mathrm{\Delta }_1}_{\mathrm{\Delta }_2}+A^1_{\mathrm{\Delta }_2}_{\mathrm{\Delta }_3}+A^2_{\mathrm{\Delta }_1}_{\mathrm{\Delta }_3})W_3`$
in addition to expressions obtained by appropriately permuting the indices. One may therefore represent the correlators hierarchically as
$`\mathrm{\Psi }_1(z_1)\mathrm{{\rm Y}}_2(z_2,0)`$ $`=`$ $`A^1W_2+_{\mathrm{\Delta }_1}\mathrm{\Phi }_1(z_1)\mathrm{{\rm Y}}_2(z_2,0)`$
$`\mathrm{\Psi }_1(z_1)\mathrm{\Phi }_2(z_2)`$ $`=`$ $`A^1W_2+_{\mathrm{\Delta }_1}\mathrm{\Phi }_1(z_1)\mathrm{\Phi }_2(z_2)`$
$`\mathrm{\Psi }_1(z_1)\mathrm{\Psi }_2(z_2)`$ $`=`$ $`A^{12}W_2+_{\mathrm{\Delta }_1}\mathrm{\Phi }_1(z_1)\mathrm{\Psi }_2(z_2)+_{\mathrm{\Delta }_2}\mathrm{\Psi }_1(z_1)\mathrm{\Phi }_2(z_2)`$ (38)
$``$ $`_{\mathrm{\Delta }_1}_{\mathrm{\Delta }_2}\mathrm{\Phi }_1(z_1)\mathrm{\Phi }_2(z_2)`$
in the case of two-point functions, and
$`\mathrm{\Psi }_1(z_1)\mathrm{{\rm Y}}_2(z_2,0)\mathrm{{\rm Y}}_3(z_3,0)`$ $`=`$ $`A^1W_3+_{\mathrm{\Delta }_1}\mathrm{\Phi }_1(z_1)\mathrm{{\rm Y}}_2(z_2,0)\mathrm{{\rm Y}}_3(z_3,0)`$
$`\mathrm{\Psi }_1(z_1)\mathrm{\Phi }_2(z_2)\mathrm{{\rm Y}}_3(z_3,0)`$ $`=`$ $`A^1W_3+_{\mathrm{\Delta }_1}\mathrm{\Phi }_1(z_1)\mathrm{\Phi }_2(z_2)\mathrm{{\rm Y}}_3(z_3,0)`$
$`\mathrm{\Psi }_1(z_1)\mathrm{\Psi }_2(z_2)\mathrm{{\rm Y}}_3(z_3,0)`$ $`=`$ $`A^{12}W_3+_{\mathrm{\Delta }_1}\mathrm{\Phi }_1(z_1)\mathrm{\Psi }_2(z_2)\mathrm{{\rm Y}}_3(z_3,0)`$
$`+`$ $`_{\mathrm{\Delta }_2}\mathrm{\Psi }_1(z_1)\mathrm{\Phi }_2(z_2)\mathrm{{\rm Y}}_3(z_3,0)`$
$``$ $`_{\mathrm{\Delta }_1}_{\mathrm{\Delta }_2}\mathrm{\Phi }_1(z_1)\mathrm{\Phi }_2(z_2)\mathrm{{\rm Y}}_3(z_3,0)`$
$`\mathrm{\Psi }_1(z_1)\mathrm{\Phi }_2(z_2)\mathrm{\Phi }_3(z_3)`$ $`=`$ $`A^1W_3+_{\mathrm{\Delta }_1}\mathrm{\Phi }_1(z_1)\mathrm{\Phi }_2(z_2)\mathrm{\Phi }_3(z_3)`$
$`\mathrm{\Psi }_1(z_1)\mathrm{\Psi }_2(z_2)\mathrm{\Phi }_3(z_3)`$ $`=`$ $`A^{12}W_3+_{\mathrm{\Delta }_1}\mathrm{\Phi }_1(z_1)\mathrm{\Psi }_2(z_2)\mathrm{\Phi }_3(z_3)`$
$`+`$ $`_{\mathrm{\Delta }_2}\mathrm{\Psi }_1(z_1)\mathrm{\Phi }_2(z_2)\mathrm{\Phi }_3(z_3)_{\mathrm{\Delta }_1}_{\mathrm{\Delta }_2}\mathrm{\Phi }_1(z_1)\mathrm{\Phi }_2(z_2)\mathrm{\Phi }_3(z_3)`$
$`\mathrm{\Psi }_1(z_1)\mathrm{\Psi }_2(z_2)\mathrm{\Psi }_3(z_3)`$ $`=`$ $`A^{123}W_3+_{\mathrm{\Delta }_1}\mathrm{\Phi }_1(z_1)\mathrm{\Psi }_2(z_2)\mathrm{\Psi }_3(z_3)`$ (39)
$`+`$ $`_{\mathrm{\Delta }_2}\mathrm{\Psi }_1(z_1)\mathrm{\Phi }_2(z_2)\mathrm{\Psi }_3(z_3)+_{\mathrm{\Delta }_3}\mathrm{\Psi }_1(z_1)\mathrm{\Psi }_2(z_2)\mathrm{\Phi }_3(z_3)`$
$``$ $`_{\mathrm{\Delta }_1}_{\mathrm{\Delta }_2}\mathrm{\Phi }_1(z_1)\mathrm{\Phi }_2(z_2)\mathrm{\Psi }_3(z_3)`$
$``$ $`_{\mathrm{\Delta }_2}_{\mathrm{\Delta }_3}\mathrm{\Psi }_1(z_1)\mathrm{\Phi }_2(z_2)\mathrm{\Phi }_3(z_3)`$
$``$ $`_{\mathrm{\Delta }_1}_{\mathrm{\Delta }_3}\mathrm{\Phi }_1(z_1)\mathrm{\Psi }_2(z_2)\mathrm{\Phi }_3(z_3)`$
$`+`$ $`_{\mathrm{\Delta }_1}_{\mathrm{\Delta }_2}_{\mathrm{\Delta }_3}\mathrm{\Phi }_1(z_1)\mathrm{\Phi }_2(z_2)\mathrm{\Phi }_3(z_3)`$
in the case of three-point functions. As above, the remaining correlators may be obtained by appropriately permuting the indices. Similar results in the particular case of proper primary fields (28) have already appeared in the literature , see also .
We finally wish to re-address the conformal Ward identities in the realm of these hierarchical structures. Since the latter are the same in all the cases, we will focus on the most complex scenario, the one involving the three-point function of three logarithmic fields. The conformal Ward identity following from inserting $`L_1`$ into such a correlator may be written
$$0=\left(_1^3+2z_1\delta _1+2z_2\delta _2+2z_3\delta _3\right)\mathrm{\Psi }_1(z_1)\mathrm{\Psi }_2(z_2)\mathrm{\Psi }_3(z_3)$$
(40)
As is conventional in logarithmic conformal field theory, we have introduced here the operator $`\delta _i`$ acting (in the case of a conformal Jordan cell of rank two) on the fields in a correlator as
$$\delta _i\mathrm{\Psi }_j(z_j)=\delta _{ij}\mathrm{\Phi }_j(z_j),\delta _i\mathrm{\Phi }_j(z_j)=0$$
(41)
in addition to $`\delta _i\mathrm{{\rm Y}}_j(z_j,0)=0`$. This means that the conformal Ward identity (40) reads
$`0`$ $`=`$ $`_1^3\mathrm{\Psi }_1(z_1)\mathrm{\Psi }_2(z_2)\mathrm{\Psi }_3(z_3)+2z_1\mathrm{\Phi }_1(z_1)\mathrm{\Psi }_2(z_2)\mathrm{\Psi }_3(z_3)`$ (42)
$`+`$ $`2z_2\mathrm{\Psi }_1(z_1)\mathrm{\Phi }_2(z_2)\mathrm{\Psi }_3(z_3)+2z_3\mathrm{\Psi }_1(z_1)\mathrm{\Psi }_2(z_2)\mathrm{\Phi }_3(z_3)`$
This condition is easily verified using (37). It is stressed, though, that it is only with hindsight that these structures appear natural.
## 3 Conclusion
We have studied the conformal Ward identities for quasi-primary fields appearing in logarithmic conformal field theory based on conformal Jordan cells of rank two. Even though our results are based on an ansatz, it appears natural to suspect that they constitute the general solution for two- and three-point functions.
We anticipate that one, in a straightforward manner, may extract general information about the operator-product expansions underlying the correlators we have found. This is an interesting enterprise we intend to undertake.
As already mentioned, our results pertain to conformal Jordan cells of rank two. We hope to study the case of general rank elsewhere. Partial results in this direction may be found in . Conformal Jordan cells of infinite rank have been introduced in .
We have found that the results presented in this paper may be extended to affine Jordan cells appearing in certain logarithmic extensions of Wess-Zumino-Witten models . The general solutions in these models also satisfy the Knizhnik-Zamolodchikov equations and are found to reduce, by hamiltonian reduction, to the solutions provided in the present paper.
Another natural extension of the present work which would be interesting to pursue, is the general solution to the superconformal Ward identities appearing in logarithmic superconformal field theory. Results in this direction may be found in .
Acknowledgements The author thanks Michael Flohr for very helpful comments. |
warning/0507/hep-th0507248.html | ar5iv | text | # 1 Introduction
## 1 Introduction
Myers effect shows that, in the background of RR potential the multiple D0-branes will expand into a fuzzy D2-brane with a spherical profile. This is one of the general properties that the D(p+2)-brane could be realized by the fuzzy configurations of the Dp-branes \[2-4\]. The phenomena are described by the non-abelian Dirac-Born-Infeld action and have been extended to the system in the curved spacetimes , in which the fuzzy configurations of sphere and cylinder have been studied.
The fuzzy spike in the curved background, in which the “radius” $`r(z)`$ depends on the coordinate of $`z`$ direction has not yet been discussed. The problem of finding the solution is investigated in this paper. We will see that, although a simple formula of the Lagrangian for the multiple D0 in a class of the curved background could be obtained it is difficult to obtain the exact solution therein. A solution we found is the system in the geometry of the NS5-branes background. From our investigations it indicates that the size of matrices for a fuzzy spike is finite when it is in the curved spacetimes. We also in this paper re-examine the system in the flat spacetime and show that, in addition to the fuzzy tube and fuzzy spike which were found in the previous literature, there is the fuzzy wormhole solution. Note that using the abelian Dirac-Born-Infeld action the BIons with spike and wormhole profiles have been found by Emparan . Our investigations in his paper have used the method of Hyakutake while extend it to the curved spacetimes.
In section II we simplify the non-abelian Dirac-Born-Infeld action in a class of the curved background. The detailed calculations are present. After choosing the matrix elements as the coordinates of the D0-branes we obtain a simple formula of the Lagrangian for the system in the class of the curved background. Using the formula we first in section III review the fuzzy solutions in flat spacetime, i.e. the fuzzy tube and fuzzy spike which were found in the previous literature. We then present a new fuzzy wormhole solution. In section IV, we apply the formula to the system in the geometry of the NS5-branes background. We present a solution of fuzzy BIon of spike profile. From our investigations it is seen that the size of the matrices is finite for the fuzzy spike in the curved spacetimes. We make a discussion in the last section. Note that in recent many authors have investigated the dynamics of D-branes in the NS5-branes . Their results are used to show that the radial mode of the BPS D-brane in the NS5-branes backgrounds resembles the tachyon rolling dynamics of unstable D-brane . The solution found in this paper could be used to discuss the relevant problem.
## 2 Non-abelian Dirac-Born-Infeld Action in Curved Spacetimes
The Myers T-dual non-abelian Born-Infeld action describing N coincident Dp-branes is given by .
$$S_{BI}=T_pd^{p+1}\sigma STr\left(e^\varphi \sqrt{det(P[E_{ab}+E_{ai}(Q^1\delta )^{ij}E_{jb}]+\lambda F_{ab})det(Q^i{}_{j}{}^{})}\right),$$
$`(2.1)`$
in which $`F_{ab}`$ is the DBI 2-form strength on the Dp-brane and
$$E_{\mu \nu }=G_{\mu \nu }+B_{\mu \nu }^{NS},$$
$`(2.2)`$
$$Q^i{}_{j}{}^{}=\delta ^i{}_{j}{}^{}+\frac{i}{\lambda }[X^i,X^k]E_{kj},$$
$`(2.3)`$
where $`G_{\mu \nu }`$ is the metric of the spacetime, $`B_{\mu \nu }^{NS}`$ the NS 2-form, and $`X^i`$ are the $`N\times N`$ matrices describing the coordinates of the N Dp branes. The pull-back of the bulk spacetime tensors $`V_{\mu _1\mathrm{}\mu _n}`$ to the D-brane world-volume is denoted by the symbol $`P[V_{\alpha _1\mathrm{}\alpha _n}]`$ in which
$$P[V_{\alpha _1\mathrm{}\alpha _n}]=V_{\mu _1..\mu _n}D_{\alpha _1}X^{\mu _1}\mathrm{}D_{\alpha _n}X^{\mu _n},$$
$`(2.4)`$
with $`D_aX^i=_aX^i+i[A_a,X^i]`$ . The brane tension is defined by
$$T_p=\frac{2\pi }{g_s\left(2\pi \mathrm{}_s\right)^{p+1}},$$
$`(2.5)`$
and $`\lambda 2\pi \mathrm{}_s^2`$ in which $`\mathrm{}_s`$ is the string length scale.
A class of curved spacetime we will consider is described by
$$G_{\mu \mu }=(g_{00},g_{xx},g_{yy},g_{zz}),$$
$`(2.6a)`$
$$𝐁^{NS}=B_{0z}dtdz.$$
$`(2.6b)`$
In this paper we will consider the system without the RR potential and thus the Wess-Zumino term does not appear.
For the case of N coincident D0-branes in the class of the curved background described by (2.6) the relevant quantities become
$$Q^i{}_{j}{}^{}=\left(\begin{array}{ccc}1& i\lambda ^1[X^1,X^2]g_{yy}& i\lambda ^1[X^1,X^3]g_{zz}\\ i\lambda ^1[X^2,X^1]g_{xx}& 1& i\lambda ^1[X^2,X^3]g_{zz}\\ i\lambda ^1[X^3,X^1]g_{xx}& i\lambda ^1[X^3,X^2]g_{yy}& 1\end{array}\right).$$
$`(2.7)`$
$$det(Q^i{}_{j}{}^{})\stackrel{STr}{=}1\lambda ^2[X^1,X^2]^2g_{xx}g_{yy}\lambda ^2[X^2,X^3]^2g_{yy}g_{zz}\lambda ^2[X^3,X^1]^2g_{zz}g_{xx},$$
$`(2.8)`$
The notation $`\stackrel{STr}{=}`$ is used to emphasize that the above equation holds under the symmetrized trace prescription . The pull-back of the bulk spacetime tensors are
$$P[E_{00}]=g_{00}g_{xx}[A_0,X^1]^2g_{yy}[A_0,X^2]^2[A_0,X^3]^2g_{zz}.$$
$`(2.9)`$
$$P\left[E_{0i}(Q^1\delta )^{ij}E_{j0}\right]=B_{0z}(Q^1\delta )^z{}_{z}{}^{}g_{zz}^{1}+iB_{0z}(Q^1\delta )^z{}_{i}{}^{}[A_0,X^i]$$
$$iB_{0z}E_{ki}(Q^1\delta )^i{}_{z}{}^{}[A_0,X^k]g_{zz}^1E_{mi}(Q^1\delta )^i{}_{n}{}^{}[A_0,X^m][A_0,X^n],$$
$`(2.10)`$
and finally we have a simple relation
$$det\left(P[E_{00}+E_{0i}(Q^1\delta )^{ij}E_{j0}]\right)det(Q^i{}_{j}{}^{})=g_{00}g_{00}g_{xx}g_{yy}\lambda ^2[X^1,X^2]^2$$
$$g_{00}g_{yy}g_{zz}\lambda ^2[X^2,X^3]^2g_{00}g_{zz}g_{xx}\lambda ^2[X^3,X^1]^2g_{yy}B_{0z}^2\lambda ^2[X^2,X^3]^2$$
$$g_{xx}\lambda ^2B_{0z}^2[X^3,X^1]^2g_{xx}[A_0,X^1]^2g_{yy}[A_0,X^2]^2g_{zz}[A_0,X^3]^2.$$
$`(2.11)`$
in which we have neglected terms containing higher power of commutative matrices. This is the first generally useful formula presented in this paper. Note that many interesting backgrounds fall into the class of eq.(2.6), for example, the $`AdS_3\times S^3`$, NS5-branes and macroscopic string backgrounds.
To proceed, we will consider the fuzzy surface with axial symmetry around the $`x^3(=z)`$ direction, thus the matrices is chosen as
$$X_{mn}^1=\frac{1}{2}\rho _{m+1/2}\delta _{m+1,n}+\frac{1}{2}\rho _{m1/2}\delta _{m,n+1},$$
$`(2.12a)`$
$$X_{mn}^2=\frac{i}{2}\rho _{m+1/2}\delta _{m+1,n}\frac{i}{2}\rho _{m1/2}\delta _{m,n+1},$$
$`(2.12b)`$
$$X_{mn}^3=z_m\delta _{m,n},$$
$`(2.12c)`$
where $`m,n`$ are the set of integers. $`z_m`$ is interpreted as a position of the $`m`$th segment in the $`z`$ direction. Because that
$$\left(X^1\right)_{mn}^2+\left(X^2\right)_{mn}^2=\frac{1}{2}\left(\rho _{m+1/2}^2+\rho _{m1/2}^2\right)\delta _{m,n}r_m^2\delta _{m,n},$$
$`(2.13)`$
the function $`r_m\frac{1}{2}\left(\rho _{m+1/2}^2+\rho _{m1/2}^2\right)^{1/2}`$ is naturally interpreted as a position at $`m`$th segments in the radial direction. Values of $`m,n`$ run an infinite set of integers for the open surface and a finite set of integers for the closed surface. A possible fuzzy spike is plotted in figure 1.
. Figure 1. The profile of a fuzzy spike.
The commutation relations of $`X^i`$ are evaluated as
$$[X^1,X^2]_{mn}=\frac{i}{2}\left(\rho _{m+1/2}^2\rho _{m1/2}^2\right)\delta _{m,n},$$
$`(2.14a)`$
$$[X^2,X^3]_{mn}=\frac{i}{2}\rho _{m+1/2}\left(z_{m+1}z_m\right)\delta _{m+1,n}+\frac{i}{2}\rho _{m1/2}\left(z_mz_{m1}\right)\delta _{m,n+1},$$
$`(2.14b)`$
$$[X^3,X^1]_{mn}=\frac{1}{2}\rho _{m+1/2}\left(z_{m+1}z_m\right)\delta _{m+1,n}+\frac{1}{2}\rho _{m1/2}\left(z_mz_{m1}\right)\delta _{m,n+1}.$$
$`(2.14c)`$
Thus
$$\left([X^1,X^2]\right)_{mn}^2=\frac{1}{4}\left(\rho _{m+1/2}^2\rho _{m1/2}^2\right)^2\delta _{m,n},$$
$`(2.15a)`$
$$\left([X^2,X^3]\right)_{mn}^2=\left([X^3,X^1]\right)_{mn}^2=\frac{1}{4}[\rho _{m+3/2}\rho _{m+1/2}(z_{m+2}z_{m+1})(z_{m+1}z_m)\delta _{m+2,n}$$
$$+\left(\rho _{m+1/2}^2(z_{m+1}z_m)^2+\rho _{m1/2}^2(z_mz_{m1})^2\right)\delta _{m,n}$$
$$+\rho _{m1/2}\rho _{m3/2}(z_mz_{m1})(z_{m1}z_{m2})\delta _{m2,n}],$$
$`(2.15b)`$
We will consider the system with
$$A_0=a_0X^3.$$
$`(2.16)`$
This means that there is an electric flux along the $`z`$ direction and the system have the fundamental strings along the z axial.
Substituting (2.15) into (2.11) the Lagrangian for the N coincident D0-branes in the class of the curved background becomes
$$L=T_0e^\varphi (N(g_{00})^{1/2}\underset{m}{}[\frac{\lambda ^2}{4}(g_{00})^{1/2}(\frac{1}{2}g_{xx}g_{yy}(\rho _{m+1/2}^2\rho _{m1/2}^2)^2+(g_{yy}g_{zz}+g_{zz}g_{xx})$$
$$\times \rho _{m+1/2}^2(z_{m+1}z_m)^2)\frac{1}{4}(g_{00})^{1/2}(\lambda ^2B_{0z}^2+a_0^2)(g_{xx}+g_{yy})\rho _{m+1/2}^2(z_{m+1}z_m)^2]).$$
$`(2.17)`$
This relation is the main formula using in this paper.
Notice that in the curved spacetime the metric $`g_{\mu \nu }`$ in (2.17) will be the function of the spacetime. For the fuzzy surface with axial symmetry around the $`z`$ direction the metric will depend on the radius $`r`$ which is defined by
$$r^2=\frac{1}{N}Tr\left(\left(X^1\right)^2+\left(X^2\right)^2\right)=\frac{1}{N}\underset{m}{}\left[\frac{1}{2}\left(\rho _{m+1/2}^2+\rho _{m1/2}^2\right)\right]\frac{1}{N}\underset{m}{}r_m^2,$$
$`(2.18)`$
in which we have used the relation (2.13) in which $`r_m`$ is interpreted as the position at $`m`$th segment in the radial direction.
## 3 Fuzzy BIons in Vacuum
In the vacuum the main formula (2.17) becomes
$$L=\underset{m}{}\left[\frac{\lambda ^2}{8}\left(\rho _{m+1/2}^2\rho _{m1/2}^2\right)^2\frac{\lambda ^2}{2}\rho _{m+1/2}^2\left(z_{m+1}z_m\right)^2+\frac{1}{2}a_0^2\rho _{m+1/2}^2\left(z_{m+1}z_m\right)^2\right].$$
$`(3.1)`$
After the variation with respective to the $`\rho _{m+1/2}`$ and $`z_m`$ we have the equations
$$\left(\rho _{m+3/2}^22\rho _{m+1/2}^2+\rho _{m+1/2}^2\right)2(1a_0^2\lambda ^2)\left(z_{m+1}z_{m+1}\right)^2=0.$$
$`(3.2)`$
$$\rho _{m+1/2}^2\left(z_{m+1}z_m\right)\rho _{m1/2}^2\left(z_{m1}z_{m1}\right)=0.$$
$`(3.3)`$
Searching the solution from above equations is an interesting problem and the result represents the fuzzy BIon in the vacuum. Note that the above equations had been obtain by Hyakutake from the SFSS matrix theory.
There are two nontrivial solutions presented in the previous literature.
(I) Fuzzy tube : The fuzzy tube solution is described by
$$\rho _{m+1/2}=r,z_m=Lm.$$
$`(3.4)`$
In this case, the fuzzy tube has a constant radius while its position of the $`m`$th segment in the $`z`$ direction is a discrete value with equal spacing. The solution of eq.(3.4) implies that the commutation relations (2.14) become
$$[X^1,X^2]=0,[X^2,X^3]=iLX^1,[X^3,X^1]=iLX^2,$$
$`(3.5)`$
which is just the matrix tube found in . The tube solution was first found in by using the abelian Dirac-Born-Infeld action. Note that Eq.(3.4) shows that $`\mathrm{}<m<\mathrm{}`$, thus the size of the matrices is infinite.
(II) Fuzzy spike : The fuzzy spike found in is a solution of the following matrix elements
$$\rho _{m+1/2}=2\alpha m,$$
$`(3.6a)`$
$$z_m=c+L\underset{i=1}{\overset{m1}{}}\frac{1}{i},$$
$`(3.6b)`$
$$a_0=\lambda ^1.$$
$`(3.6c)`$
Thus, increasing the $`m`$ the position $`z_m`$ is increasing and the radius of the BIon increasing also. The profile of BIon is spike as that plotted in figure 1. Eq.(3.6b) shows that $`1<m<\mathrm{}`$, thus the size of the matrices of fuzzy spike is infinite.
(II) Fuzzy wormhole : We now present the fuzzy wormhole solution. It is described by the following matrix elements
$$\rho _{m+1/2}=(1\lambda ^2a_0^2)L^2\left(m+\frac{3}{2}\right)^3\left(m+\frac{7}{2}\right),$$
$`(3.7a)`$
$$z_m=c+L\underset{i=m_0}{\overset{m}{}}\left(i+\frac{1}{2}\right),$$
$`(3.7b)`$
in which $`m_0`$ is an arbitrary integral. Note that form (3.7a) we find that the radius “$`r_m`$” at $`m`$th segment (defined in (2.13)) is
$$r_m^2=\frac{1}{2}L^2\left[(m+\frac{5}{2})^3(m+\frac{7}{2})+(m+\frac{3}{2})^2(m+\frac{5}{2})\right]$$
$`(3.8)`$
It is an easy work to check that $`r_m^2>0`$ for any integral $`m`$, thus the size of the matrices of fuzzy wormhole is infinite. Eq.(3.8) show that $`r_m\mathrm{}`$ at $`m=\pm \mathrm{}`$ and a minimum is at $`m=0`$, thus the profile of the fuzzy BIon is a wormhole.
## 4 Fuzzy BIons in NS5-branes Background
In this section we will use the main formula (2.17) to find the fuzzy spike in NS5-branes background. The background fields around $`N_5`$ NS5-branes are given by the CHS solution . The metric, dilaton and NS-NS $`B^{NS}`$ field are
$$ds^2=dx_\mu dx^\mu +h(x^n)dx^mdx^m,$$
$$e^{2(\varphi )}=h(x^n),$$
$$H_{mnp}=ϵ_{mnp}^q_q\varphi .$$
$`(4.1)`$
Here $`h(x^n)`$ is the harmonic function describing fivebranes, and $`H_{mnp}`$ is the field strength of the NS-NS $`B^{NS}`$ field. For the case of coincident $`N_5`$ fivebranes one has
$$h(r)=1+\frac{N_5}{r^2},$$
$`(4.2)`$
where $`r=|\stackrel{}{x}|`$ is the radial coordinate away from the fivebranes in the transverse space labeled by $`(x^6,\mathrm{},x^9)`$.
Using the above metric and fields we can easily from (2.17) show that the Lagrangian of $`N`$ coincident D0-branes in the NS5-branes background becomes
$$L=\frac{N}{\sqrt{h}}\frac{\lambda ^2}{8}h\sqrt{h}\underset{m}{}\left(\rho _{m+1/2}\rho _{m1/2}\right)^2+\frac{1}{2}\left(a_0^2\lambda ^2\right)\sqrt{h}\underset{m}{}\rho _{m+1/2}^2\left(z_{m+1}z_m\right)^2.$$
$`(4.3)`$
After the variation with respective to the $`\rho _{m+1/2}`$ and $`z_m`$ we have the equations
$$\frac{}{rr}\left(\frac{N}{\sqrt{h}}\right)\frac{}{rr}\left(h\sqrt{h}\right)\frac{\lambda ^2}{8}\underset{m}{}\left(\rho _{m+1/2}\rho _{m1/2}\right)^2\frac{\lambda ^2}{2}h\sqrt{h}\left(\rho _{m+3/2}^22\rho _{m+1/2}^2+\rho _{m+1/2}^2\right)$$
$$+\frac{1}{2}\left(a_0^2\lambda ^2\right)\left[\frac{}{rr}\left(\sqrt{h}\right)\left(\underset{m}{}\rho _{m+1/2}^2\left(z_{m+1}z_{m+1}\right)\right)+2\sqrt{h}\left(z_{m+1}z_{m+1}\right)^2\right]=0.$$
$`(4.4)`$
$$\rho _{m+1/2}^2\left(z_{m+1}z_m\right)\rho _{m1/2}^2\left(z_{m1}z_{m1}\right)=0.$$
$`(4.5)`$
The fuzzy spike in the NS5-branes background can be obtained by choosing the following matrix elements
$$\rho _{m+1/2}^2=2\alpha (m+\beta ),$$
$`(4.6a)`$
$$z_m=c+L\underset{i=1}{\overset{m1}{}}\frac{1}{i+\beta },$$
$`(4.6b)`$
$$a_0=\lambda ^1.$$
$`(4.6c)`$
which becomes the fuzzy spike solution in the vacuum if $`\beta =0`$. Above solution automatically satisfies eq.(4.5) and after substituting them into eq.(4.4) we find that
$$r^2=\frac{N_5}{\sqrt{\frac{8N\lambda ^2}{3\alpha }}1}.$$
$`(4.7)`$
Now, as shown in (2.18) the value “r” shall be defined by
$$r^2=\frac{1}{N}\underset{m}{}\rho _{m+1/2}^2=\frac{1}{N}\underset{m}{}\alpha (m+\beta ),$$
$`(4.8)`$
the value of $`\beta `$ could therefore be found for a fixed $`\alpha `$. The value $`N`$ specifies the size of the matrices of the fuzzy spike in the curved spacetimes.
We now make following comments to discuss the physical properties of the parameters $`\alpha `$, $`L`$, and $`N`$.
(I) Substituting (3.6a) or (4.6a) into (2.14a) we see that
$$[X^1,X^2]_{mn}=i\alpha \delta _{m,n}.$$
$`(4.9)`$
Thus, the value $`\alpha `$ represents the fuzziness of the D0-branes. Comparing to the approach of abelian Dirac-Born-Infeld, it is known that the value of $`\alpha `$ shall be proportional to the density of D0-branes, thus it will be proportional to the magnetic flux of the system.
(II) The value of $`L`$ in (3.6b) or (4.6b) could not be fixed by the equation. This means that the scale of fuzzy BIon along “z” axial is arbitrary. The fact may be traced to the conformal symmetry of the string. Another arbitrary constant $`c`$ appearing (3.6b) or (4.6b) reveals the translation symmetry along the $`z`$ direction.
(III) From eq.(4.8) we see that the value $`N`$, which specifies the size of the matrices of the fuzzy spike in the curved spacetimes, shall be a finite number. This is the general property for a fuzzy spike in the curved background, in contrast to that in flat spacetime (see section III). This is because that if a fuzzy spike in the curved background has infinite dimension, then the value $`\rho _{m+1/2}^2`$ shall run to infinite and thus the definition of “r” in eq.(4.8) will become infinite in general. However, the value of “r” is a variable of metric $`g_{\mu \nu }`$ and it shall be a finite value, thus we conclude that the size of the matrices shall be finite for the fuzzy spike in the curved spacetimes. It is easy to see that the some conclusion could also be found for the fuzzy wormhole in the curved spacetimes, if it exist.
## 5 Conclusion
In this paper we use the non-abelian Dirac-Born-Infeld action to find the fuzzy spike solution in NS5-branes background. We have simplified the non-abelian Dirac-Born-Infeld action in a class of the curved background and presented the detailed calculations to obtain the simple form of the Lagrangian. Using the formula, after reviewing the solutions of fuzzy tube and fuzzy spike in flat spacetime, which were found in the previous literature, we first present a new fuzzy wormhole solution. We next apply the formula to the system in the geometry of the NS5-branes background and present a fuzzy BIon of spike profile. Our investigations indicates that the size of the matrices is finite for the fuzzy spike in the curved spacetimes.
We make following comments to conclude this paper.
(1) Although the final form of the Lagrangian is obtained in this paper, it is difficult to find the fuzzy BIon solution in general. For example, we have investigates the system in $`AdS_3\times S^2`$ and macroscopic string background , however, the consistent fuzzy spike or wormhole solution has not yet been found. In fact, even in the NS5-branes background it is difficult to find the fuzzy wormhole solution.
(2) As mentioned in section I, many authors have investigated the dynamics of D-branes in the NS5-branes, as the results could be used to show that the radial mode of the BPS D-brane in the NS5-branes backgrounds resembles the tachyon rolling dynamics of unstable D-brane . The problem of unstable fuzzy BIons in curved spacetimes have not yet discussed. It would be interesting to use the solution found in this paper to study the dynamics of fuzzy spike in the NS5-branes. The works remain to be studied in the future.
References
1. R.C. Myers, “Dielectric-Branes”, JHEP 9912 (1999) 022 \[hep-th/9910053\]; “Nonabelian Phenomena on D-branes”, Class.Quant.Grav. 20 (2003) S347-S372 \[hep-th/0303072\].
2. Y. Hyakutake, “Notes on the Construction of the D2-brane from Multiple D0-branes”, Nucl. Phys. B675 (2003) 241 \[hep-th/0302190\]; “Fuzzy BIon”, Phys.Rev. D68 (2003) 046003 \[hep-th/0305019\].
3. M. Asano “Non-commutative branes in D-brane backgrounds”, Int.J.Mod.Phys. A17 (2002) 4733-4748 \[hep-th/0106253\]; D. K. Park, S. Tamaryan, H. J. W. Müller-Kirsten, “General Criterion for the existence of Supertube and BIon in Curved Target Space”, Phys.Lett. B563 (2003) 224-230 \[hep-th/0302145\]; Y. Hyakutake, “Gravitational Dielectric Effect and Myers Effect” , Phys.Rev. D71 (2005) 046007 \[hep-th/0401026\]; Wung-Hong Huang ; “Fuzzy Rings in D6-Branes and Magnetic Field Background” , JHEP 0407 (2004) 012 \[hep-th/0404202\].
4. Y. Kimura, “Nonabelian gauge field and dual description of fuzzy sphere”, JHEP 0404 (2004) 058 \[hep-th/0402044\]; “Myers Effect and Tachyon Condensation”, Nucl.Phys. B692 (2004) 394-416 \[hep-th/0309082\]; Y. Hyakutake, “Torus-like Dielectric D2-brane”, JHEP 0105 (2001) 013 \[ hep-th/0103146\]; J. de Boer, E. Gimon, K. Schalm, J. Wijnhout, “Evidence for a gravitational Myers effect”, Annals Phys. 313 (2004) 402-424 \[hep-th/0212250\]; B. Janssen, Y. Lozano, “Non-Abelian Giant Gravitons”, Class. Quant. Grav. 20 (2003) S517-S524 \[hep-th/0212257\].
5. R. Emparan, “Born-Infeld strings tuuneling to D-branes”, Phys.Lett. B423 (1998) 71-78 \[hep-th/9711106\]; C. G. Callen and J. M. Maldacena, “Brane dynamics from the Born-Infeld action”, Nucl. Phys. B513 (1998) 198 \[hep-th/9708147\].
6. D. Kutasov, “D-brane dynamics near NS5-brane,” \[hep-th/0405058\]; “A geometric interpretation of the open string tachyon,” \[hep-th/0408073\]; Y. Nakayama, Y. Sugawara and H. Takayanagi, “Boundary states for the rolling D-brane in NS5 background,” JHEP 0407(2004) 020 \[hep-th/0406173\]; D. A. Sahakyan, “Comments on D-brane dynamics near NS5-brane,” JHEP 0410 (2004) 008 (2004) \[hep-th/0408070\].
7. J. Kluson, “Non-BPS D-brane near NS5-brane,” JHEP 0411 (2004) 013 \[hep-th/0409298\]; “Non-BPS Dp-brane in the background of NS5-brane on transverse R\**3 x S\**1,” \[hep-th/0411014\]; J. Kluson and K. Panigrahi, “Supertube dynamics in diverse backgrounds”, \[hep-th/0506012\]; S. Thomas and J. Ward, “D-brane dynamics and NS5 rings,” JHEP 0502 (2005) 015 \[hep-th/0411130\]; “D-brane dynamics near compactified NS5-branes” \[hep-th/0501192\] ; B. Chen, M. Li and B. Sun, “D-brane near NS5-brane: With electromagnetic field,” JHEP 0412 (2004) 057 \[hep-th/0412022\];Y. Nakayama, K. L. Panigrahi, S. J. Rey and H. Takayanagi, “Rolling down the throat in NS5-brane background: The case of electrified D-brane,” JHEP 0501 (2005) 052 \[hep-th/0412038\].
8. A. Sen, “Time and tachyon,” Int. J. Mod. Phys. A 18 (2003) 4869 \[hep-th/0209122\]; “Tachyon matter,” JHEP 0207 (2002) 065 \[hep-th/0203265\]; “Rolling tachyon,” JHEP 0204 (2002) 048 \[hep-th/0203211\]; “Tachyon dynamics in open string theory,” \[hep-th/0410103\].
9. D. Bak, K. M. Lee, “Noncommutative Supersymmetric Tubes”, Phys. Lett. B509 (2001) 168 \[hep-th/0103148\]; D. Bak and S. W. Kim, “Junction of Supersymmetric Tubes,” Nucl. Phys. B622 (2002) 95 \[hep-th/0108207\].
10. D. Mateos and P. K. Townsend, “Supertubes”, Phys. Rev. Lett. 87 (2001) 011602 \[hep-th/0103030\]; R. Emparan, D. Mateos and P. K. Townsend, “Supergravity Supertubes”, JHEP 0107 (2001) 011 \[hep-th/0106012\].
11. C. G. . Callan, J. A. Harvey and A. Strominger, “Supersymmetric string solitons,” \[hep-th/9112030\].
12. C. Bachas and M. Petropoulos, “Anti-de-Sitter D-branes”, JHEP 0102, 025 (2001) \[hep-th/0012234\]; P. M. Petropoulos, S. Ribault, “Some comments on Anti-de Sitter D-branes” , JHEP 0107 (2001) 036 \[hep-th/0105252\]; A. Giveon, D. Kutasov and A. Schwimmer, “Comments on D-branes in AdS(3)”, Nucl. Phys. B 615, 133 (2001) \[hep-th/0106005\]; Wung-Hong Huang, “Anti-de Sitter D-branes in Curved Backgrounds” JHEP 0507 (2005) 031 \[hep-th/0504013\].
13. G. W. Gibbons and K. Maeda, “Black holes and membranes in higher dimensional theories with dilaton fields,” Nucl. Phys. B298 (1988) 741; D. Bak, S. J. Rey and H. U. Yee, “Exactly soluble dynamics of (p,q) string near macroscopic fundamental strings,” JHEP 0412 (2004) 008 \[hep-th/0411099\]; Wung-Hong Huang, “Tubular Solutions in NS5-brane, Dp-brane and Macroscopic String Background” JHEP 0502 (2005) 061 \[hep-th/0502023\]. |
warning/0507/cond-mat0507627.html | ar5iv | text | # Gradient and Amplitude Scattering in Surface-Corrugated Waveguides
## Abstract
We investigate the interplay between amplitude and square-gradient scattering from the rough surfaces in multi-mode waveguides (conducting quantum wires). The main result is that for any (even small in height) roughness the square-gradient terms in the expression for the wave scattering length (electron mean free path) are dominant, provided the correlation length of the surface disorder is small enough. This important effect is missed in existing studies of the surface scattering.
1. The problem of wave transport (both classical and quantum) through the guiding surface-disordered systems has a quite long history and remains a hot topic till now (see, e.g., Refs. BFb79, ; RytKrTat89, ; McGM84, ; SMFY9899, ; Mig77, ; Konrady74, ; TJM86, ; TrAsh88, ; MeyStep949597989900, ; LunKyReiKr96, ; IsPuzFuks9091, ; MT9801, ; LunMenIz00, ; VoronB94, ; KMYa91, ; MakMorYam95, ; Tatar93, and references therein). One of the main tools to treat this problem is the reduction of the wave/electron surface scattering to the bulk one in such a way that the latter can be described by an effective Hamiltonian with a complicated potential, however, with flat boundaries. Applying this approach, one can reasonably discriminate between the so-called amplitude and gradient scattering, and analyze their interplay explicitly.
To the best of our knowledge, originally the idea of this approach was discussed by Migdal Mig77 . After, it was frequently used in the theories of classical and quantum wave/electron scattering, see, e.g., Refs. Konrady74, ; TJM86, ; TrAsh88, ; MeyStep949597989900, ; LunKyReiKr96, ; IsPuzFuks9091, ; MT9801, ; LunMenIz00, . But in the majority of them Konrady74 ; TJM86 ; TrAsh88 ; MeyStep949597989900 ; LunKyReiKr96 the study was restricted to the lowest order in the roughness height $`\sigma `$. Other methods IsPuzFuks9091 ; MT9801 ; LunMenIz00 ; VoronB94 ; KMYa91 ; MakMorYam95 ; Tatar93 were mainly based on the principal assumption that the surface roughness is sufficiently smooth.
In this contribution we present the theory of wave scattering from rough surfaces, which takes into account both the amplitude and gradient scattering mechanisms. The important point is that we do not assume any special restrictions on the model parameters except for general conditions of weak scattering. The latter provides us with an appropriate perturbative approach in scattering potential, however, is not restricted by the smoothness of surfaces.
The main attention is paid to the competition between the amplitude and gradient scattering. One of our main results is that at fixed r.m.s. roughness height $`\sigma `$, the less correlation length $`R_c`$ of a random surface profile, the larger contribution of the gradient mechanism. Thus, by passing from the smooth to white-noise profiles, the gradient scattering becomes to prevail. We have analyzed this crossover and obtained the estimates describing the transition to the dominating gradient scattering. In particular, we show that this transition is located within the region of small roughness slopes where $`\sigma /R_c1`$.
2. In what follows we consider an open plane waveguide (or conducting quasi-one-dimensional quantum wire) of the average width $`d`$, stretched along the $`x`$-axis. For simplicity, one (lower) surface of the waveguide is assumed to be flat, $`z=0`$, while the other (upper) surface has a rough profile $`z=w(x)=d+\sigma \xi (x)`$ with $`\xi (x)=0`$ and $`\xi ^2(x)=1`$. The average $`\mathrm{}`$ is performed over different realizations of a statistically homogeneous and isotropic Gaussian random function $`\xi (x)`$. We also assume that its pair correlator $`\xi (x)\xi (x^{})=𝒲(xx^{})`$ decreases on a scale $`R_c`$, with the normalization $`𝒲(0)=1`$. The roughness-height power (RHP) spectrum $`W(k_x)`$ is defined by
$$W(k_x)=_{\mathrm{}}^{\mathrm{}}𝑑x\mathrm{exp}(ik_xx)𝒲(x).$$
(1)
Since $`𝒲(x)`$ is an even function of $`x`$, its Fourier transform (1) is even, real and nonnegative function of $`k_x`$. The RHP spectrum has maximum at $`k_x=0`$ with $`W(0)R_c`$, and decreases on the scale $`R_c^1`$.
In order to analyze the surface scattering for our model, we employ the method of the retarded Green’s function $`𝒢(x,x^{};z,z^{})`$. Specifically, we start with the Dirichlet boundary-value problem
$`\left({\displaystyle \frac{^2}{x^2}}+{\displaystyle \frac{^2}{z^2}}+k^2\right)𝒢(x,x^{};z,z^{})`$
$`=\delta (xx^{})\delta (zz^{}),`$ (2)
$`𝒢(x,x^{};z=0,z^{})=𝒢(x,x^{};z=w(x),z^{})=0.`$ (3)
Here the wave number $`k`$ is equal to $`\omega /c`$ for an electromagnetic wave of the frequency $`\omega `$ and TE polarization, propagating through a waveguide with perfectly conducting walls. As for an electron quantum wire, $`k`$ is the Fermi wave number within the isotropic Fermi-liquid model. In order to express the surface scattering as a bulk one, we perform the transformation to new coordinates,
$$x_{new}=x_{old},z_{new}=z_{old}d/[d+\sigma \xi (x)],$$
(4)
in which both waveguide surfaces are flat. Correspondingly, we introduce the canonically conjugate Green’s function, $`𝒢_{new}=d^1\sqrt{w(x)w(x^{})}𝒢_{old}`$ and omit the subscript “new” in what follows. As a result, we arrive at the equivalent boundary-value problem governed by the equation with a “bulk” perturbation potential,
$`\{{\displaystyle \frac{^2}{x^2}}+{\displaystyle \frac{^2}{z^2}}+k^2[1{\displaystyle \frac{d^2}{w^2(x)}}]{\displaystyle \frac{^2}{z^2}}`$
$`{\displaystyle \frac{\sigma }{w(x)}}\left[\xi ^{}(x){\displaystyle \frac{}{x}}+{\displaystyle \frac{}{x}}\xi ^{}(x)\right]\left[{\displaystyle \frac{1}{2}}+z{\displaystyle \frac{}{z}}\right]`$
$`+{\displaystyle \frac{\sigma ^2\xi _{}^{}{}_{}{}^{2}(x)}{w^2(x)}}[{\displaystyle \frac{3}{4}}+3z{\displaystyle \frac{}{z}}+z^2{\displaystyle \frac{^2}{z^2}}]\}𝒢(x,x^{};z,z^{})`$
$`=\delta (xx^{})\delta (zz^{}),`$ (5)
$`𝒢(x,x^{};z=0,z^{})=𝒢(x,x^{};z=d,z^{})=0.`$ (6)
Here the prime stands for the derivative over $`x`$.
We emphasize that Eq. (Gradient and Amplitude Scattering in Surface-Corrugated Waveguides) is exact and valid for any form of $`w(x)`$. As one can see, the scattering potential depends both on the roughness profile $`\sigma \xi (x)`$ and on its gradient $`\sigma \xi ^{}(x)`$. Moreover, the potential contains the term with the square gradient $`\sigma ^2\xi _{}^{}{}_{}{}^{2}(x)`$. This term is proportional to $`\sigma ^2`$ and for this reason was neglected in all previous studies of transport properties in the surface-disordered waveguides. However, as a matter of fact, the square gradient introduces the operator $`\widehat{𝒱}(x)=\xi _{}^{}{}_{}{}^{2}(x)\xi _{}^{}{}_{}{}^{2}(x)`$, which plays a special role. Its pair correlator,
$$\widehat{𝒱}(x)\widehat{𝒱}(x^{})=2\xi ^{}(x)\xi ^{}(x^{})^2=2𝒲_{}^{\prime \prime }{}_{}{}^{2}(xx^{}),$$
(7)
determines the square-gradient power (SGP) spectrum
$$T(k_x)=_{\mathrm{}}^{\mathrm{}}𝑑x\mathrm{exp}(ik_xx)𝒲_{}^{\prime \prime }{}_{}{}^{2}(x).$$
(8)
One should stress that although by integration by parts the power spectrum of the roughness gradients $`\sigma \xi ^{}(x)`$ can be reduced to the RHP spectrum $`W(k_x)`$, this is not possible for the SGP spectrum $`T(k_x)`$. This very fact reflects a highly non-trivial role of the square-gradient scattering, giving rise to the competition with the well known amplitude scattering, in spite of the seeming smallness of the term $`\sigma ^2\xi _{}^{}{}_{}{}^{2}(x)`$.
To proceed, we pass from Eq. (Gradient and Amplitude Scattering in Surface-Corrugated Waveguides) to the Dyson-type equation, performing the ensemble averaging with the use of the technique developed in Ref. McGM84, . The method allows one to develop the consistent perturbative approach with respect to the scattering potential, which takes adequately into account the multiple scattering from the corrugated boundary. After quite cumbersome calculations we have obtained the average Green’s function which in the normal-mode representation has the form
$`𝒢(x,x^{};z,z^{})={\displaystyle \underset{n=1}{\overset{N_d}{}}}\mathrm{sin}\left({\displaystyle \frac{\pi nz}{d}}\right)\mathrm{sin}\left({\displaystyle \frac{\pi nz^{}}{d}}\right)`$
(9)
$`\times {\displaystyle \frac{\mathrm{exp}(ik_n|xx^{}|)}{ik_nd}}\mathrm{exp}\left({\displaystyle \frac{|xx^{}|}{2L_n}}\right).`$ (10)
Here $`k_n=\sqrt{k^2(\pi n/d)^2}`$ corresponds to the unperturbed lengthwise wave number $`k_x`$, and $`N_d=[kd/\pi ]`$ is the number of propagating modes (or conducting electron channels) determined by the integer part $`[\mathrm{}]`$ of the ratio $`kd/\pi `$.
3. Our interest is in the attenuation length or total mean free path $`L_n`$ of the $`n`$-th mode. Its inverse value is given by the imaginary part of the proper self-energy and, in accordance with the form of the scattering potential, consists of two terms describing different scattering mechanisms,
$$\frac{1}{L_n}=\frac{1}{L_n^{(1)}}+\frac{1}{L_n^{(2)}}.$$
(11)
The first length $`L_n^{(1)}`$ is determined by the expression
$`{\displaystyle \frac{1}{L_n^{(1)}}}`$ $`=`$ $`\sigma ^2{\displaystyle \frac{(\pi n/d)^2}{k_nd}}{\displaystyle \underset{n^{}=1}{\overset{N_d}{}}}{\displaystyle \frac{(\pi n^{}/d)^2}{k_n^{}d}}`$ (12)
$`\times \left[W(k_n+k_n^{})+W(k_nk_n^{})\right].`$
Its diagonal term is formed by the amplitude scattering while the off-diagonal terms result from the gradient one. Eq. (12) coincides with that previously obtained by different methods (see, e.g., Ref. BFb79, ).
The second length $`L_n^{(2)}`$ is associated solely with the square-gradient mechanism due to the operator $`\widehat{𝒱}(x)`$,
$$\frac{1}{L_n^{(2)}}=\underset{n^{}=1}{\overset{N_d}{}}\frac{1}{L_{nn^{}}^{(2)}}.$$
(13)
Its diagonal term controls the intramode scattering,
$`{\displaystyle \frac{1}{L_{nn}^{(2)}}}`$ $`=`$ $`{\displaystyle \frac{\sigma ^4}{2}}{\displaystyle \frac{(\pi n/d)^4}{k_n^2}}\left[{\displaystyle \frac{1}{3}}+{\displaystyle \frac{1}{(2\pi n)^2}}\right]^2`$ (14)
$`\times \left[T(2k_n)+T(0)\right].`$
The off-diagonal partial length $`L_{nn^{}}^{(2)}`$ describes the intermode scattering (from $`n`$ to $`n^{}n`$ channels),
$`{\displaystyle \frac{1}{L_{nn^{}}^{(2)}}}`$ $`=`$ $`{\displaystyle \frac{8\sigma ^4}{\pi ^4}}{\displaystyle \frac{(\pi n/d)^2}{k_n}}{\displaystyle \frac{(\pi n^{}/d)^2}{k_n^{}}}{\displaystyle \frac{(n^2+n^2)^2}{(n^2n^2)^4}}`$ (15)
$`\times \left[T(k_n+k_n^{})+T(k_nk_n^{})\right].`$
To the best of our knowledge, in the surface-scattering problem for multi-mode waveguides the operator $`\widehat{𝒱}(x)`$ was never taken into account, and, as a result, the square-gradient attenuation length $`L_n^{(2)}`$ was missed in previous studies.
Let us analyze the conditions under which Eqs. (11) – (15) are derived. We stress that the Dyson-type equation for the average Green’s function was obtained within the second-order approximation in the perturbation potential. This means that the self-energy in this equation contains the binary correlator of the surface-scattering potential and the unperturbed Green’s function. In terms of the diagrammatic technique this is similar to the “simple vortex” or, the same, Bourret approximation Bourret62 . Following the ideas discussed in Ref. RytKrTat89, , one can show that this approximation is justified when the broadening $`1/2L_n`$ of the quantum wave number $`k_n`$ is much less than both the correlation scale $`R_c^1`$ and the spacing $`|k_nk_{n\pm 1}||k_n/n|`$ between neighboring quantum wave numbers. The same conditions also arise due to another approximation which is the use of the unperturbed value $`k_n`$ in the expression for the self-energy, instead of the perturbed one. Now we take into account that $`|k_n/n|\mathrm{\Lambda }_n^1`$, where $`\mathrm{\Lambda }_n`$ is the distance between two successive reflections of a wave from the rough boundary inside the $`n`$-th channel. As a result, we come to the following conditions of a weak surface scattering
$$\mathrm{\Lambda }_n=2k_nd/(\pi n/d)2L_n,R_c2L_n.$$
(16)
These inequalities imply that the electron/wave weakly attenuates on both the correlation length $`R_c`$ and the cycle length $`\mathrm{\Lambda }_n`$.
As one can see, the expressions (12) and (13) – (15) represent, respectively, basic contributions from principally different surface-scattering mechanisms related to the amplitude, gradient and square-gradient terms. It should be emphasized that the corrections proportional to $`\sigma ^4`$, originated from higher order approximations in the amplitude and gradient terms of the perturbation potential, are smaller than the main contribution (12) under the conditions (16). Contrary, the square-gradient terms give rise to the $`\sigma ^4`$-terms in Eqs.(14)-(15) which should not be neglected due to a specific dependence on the correlation length $`R_c`$. Note that Eq. (16) implicitly includes the requirement for the surface corrugations be small in height ($`\sigma d`$), but does not restrict the value $`\sigma /R_c`$ of the roughness slope.
4. Since $`L_n^{(1)}`$ and $`L_n^{(2)}`$ depend on as many as four dimensionless parameters $`(k\sigma )^2`$, $`kR_c`$, $`kd/\pi `$, and $`n`$, the complete analysis appears to be quite complicated. For this reason, below we restrict ourselves by the analysis of the interplay between $`L_n^{(1)}`$ and $`L_n^{(2)}`$ as a function of the dimensionless correlation length $`kR_c`$ for $`N_dkd/\pi 1`$.
As follows from Eq. (12), the inverse value of the amplitude-scattering length typically increases with an increase of $`kR_c`$. Specifically, in the case of the small-scale roughness ($`kR_c1k\mathrm{\Lambda }_n`$) we have $`1/L_n^{(1)}kR_c`$. Then, within the intermediate region where $`1kR_ck\mathrm{\Lambda }_n`$, the increase of $`1/L_n^{(1)}`$ slows down, or can even be replaced by the decrease for some values of the model parameters. Finally, for large-scale roughness and strong correlations ($`1k\mathrm{\Lambda }_nkR_c`$) the value of $`1/L_n^{(1)}`$ again starts to increase linearly with $`kR_c`$.
In contrast with $`1/L_n^{(1)}`$, the inverse square-gradient scattering length $`1/L_n^{(2)}`$ reveals a monotonous decrease as the parameter $`kR_c`$ increases. At small ($`kR_c1k\mathrm{\Lambda }_n`$) and extremely large ($`1k\mathrm{\Lambda }_nkR_c`$) values of $`kR_c`$, this decrease obeys the law $`1/L_n^{(2)}(kR_c)^3`$, due to $`T(0)R_c^3`$.
¿From this analysis it becomes clear that the curves displaying $`1/L_n^{(1)}`$ and $`1/L_n^{(2)}`$ must intersect, and one can observe the crossover from the square-gradient to amplitude surface scattering. To the left from the crossing point $`(kR_c)_{}`$ the square-gradient scattering length prevails, $`L_n^{(2)}L_n^{(1)}`$. To its right the main contribution is due to the well known amplitude scattering, $`L_n^{(1)}L_n^{(2)}`$. If the crossing point falls onto the interval of the small-scale roughness ($`kR_c1`$), its dependence on the model parameters is described by
$$(kR_c)_{}^2(k\sigma )n/\sqrt{k_nd}.$$
(17)
This estimate shows that the crossing point is smaller for smaller values of the dimensionless roughness height $`k\sigma `$, as well as for smaller mode indices $`n`$, or for larger values of the parameter $`kd/\pi `$.
In Fig. 1 we display the dependence of $`\mathrm{\Lambda }_n/2L_n`$ as a function of $`kR_c`$ assuming the Gaussian binary correlator $`𝒲(x)=\mathrm{exp}(x^2/2R_c^2)`$ for random surface profile $`\xi (x)`$. The curves are plotted starting from such values of $`kR_c`$ for which $`\mathrm{\Lambda }_n/2L_n^{(2)}=1`$, according to the first condition of Eq. (16). Taking into account the second condition restricting the maximal value of $`kR_c`$, we plot every curve within the range where $`R_c<2L_n^{(1)}`$. As one can see, all curves have the crossover from the square-gradient to amplitude surface scattering. The first (lowest) one with $`(k\sigma )^2=10^4`$ has the crossing point $`(kR_c)_{}0.2`$ located within the interval of small-scale roughness, and the crossover reveals a small dip centered at $`(kR_c)_{}`$. The curve obeys the asymptotic behavior $`(kR_c)^3`$ to the left from $`(kR_c)_{}`$ due to the main contribution from $`\mathrm{\Lambda }_n/2L_n^{(2)}`$. Then the quantity $`\mathrm{\Lambda }_n/2L_n^{(1)}`$ becomes dominating in the sum (11), therefore, the curve starts to rise. Firstly, the linear dependence on $`kR_c`$ on the right deep-side (where $`kR_c<1`$) is replaced with a smoother one (for $`kR_c>1`$). Finally, for $`R_c>\mathrm{\Lambda }_n`$ (strong correlations) the linear dependence restores.
The crossing points of the second, third and fourth curves have the values of the order one. Here the total attenuation length $`L_n`$ within the whole small-scale region is formed by the square-gradient scattering length $`L_n^{(2)}`$. In full agreement with Eq. (17) the presented curves display that the smaller parameter $`(k\sigma )^2`$ the smaller value of the crossing point $`(kR_c)_{}`$.
Note that for all curves in Fig.1 the roughness height is small, $`\sigma /d1`$. Furthermore, for the amplitude-dominated scattering (to the right from the point $`(kR_c)_{}`$ where $`\mathrm{\Lambda }_n/2L_n^{(1)}`$ mainly contributes), the average corrugation slope is also small for all data, $`\sigma /R_c1`$. The roughness slope remains to be small at the crossing points too, but increases to their left with the decrease of $`kR_c`$. As a result, to the left from the crossing point where the square-gradient term $`\mathrm{\Lambda }_n/2L_n^{(2)}`$ prevails, the slope reaches the values of the order one, or even larger for the first tree curves.
In conclusion, we have discovered the principal importance of the square-gradient surface-scattering mechanism which was never taken into account in the literature. We have shown that at any fixed value of the roughness height $`\sigma `$ one can indicate the region of small values of the correlation length $`R_c`$ where the new square-gradient scattering length $`L_n^{(2)}`$ predominates over the known amplitude scattering length $`L_n^{(1)}`$ ($`L_n^{(2)}L_n^{(1)}`$). The predominance occurs in spite of the fact that $`1/L_n^{(1)}`$ is proportional to $`\sigma ^2`$ while $`1/L_n^{(2)}`$ is proportional to $`\sigma ^4`$. This happens since the two lengths are determined by the substantially different roughness-height $`W(k_x)`$ and roughness-square-gradient $`T(k_x)`$ power spectra, that have vastly different dependencies on $`R_c`$. It is remarkable that the square-gradient mechanism prevails in the commonly used region $`kR_c1`$ of a small-scale boundary perturbation, where the surface roughness is typically described via the white-noise potential.
This research was partially supported by the CONACYT (México) grant No 43730, and by the VIEP-BUAP (México) under the grant II 104-05/ING/G. |
warning/0507/astro-ph0507431.html | ar5iv | text | # CHANDRA reveals galaxy cluster with the most massive nearby cooling core, RXCJ1504.1-0248
## 1 Introduction
One of the currently most debated questions concerning the structure of the X-ray luminous, hot intracluster plasma of clusters of galaxies is the consequence of the small cooling time of this plasma in those clusters with dense central cores (e.g. Fabian 1994). XMM-Newton X-ray spectroscopy has shown that in spite of its short cooling time the gas is not cooling at the high expected rates in the absence of heating processes (e.g. Peterson 2001, 2003; Matsushita et al. 2002; Böhringer et al. 2002; Molendi 2002). The most popular scenario which allows for a self-regulated heating of the hot plasma in cluster cooling cores and prevents massive cooling is the heating of the intracluster medium (ICM) by the jets and radio lobes of the AGN in the central cluster galaxies (e.g. Churazov et al. 2000, 2001; McNamara et al. 2000; David et al. 2001; Fabian 2003; Forman et al. 2004). It was shown that this interaction provides enough power in many nearby cooling flows to at least balance cooling. For example the current kinetic energy output of the inner radio lobes in M87 in Virgo and NGC 1275 in the Perseus cluster is with estimated values of $`10^{44}`$ erg s<sup>-1</sup> and $`10^{45}`$ erg s<sup>-1</sup>, respectively, in both these cases about an order of magnitude higher than the cooling power in the cooling core (Churazov et al. 2000, 2003; Birzan et al. 2004). Recently further very deep and detailed X-ray observations have given some insight into the details of the heating process, which is proposed to occur through the dissipation of sound waves set off by the interaction of the radio lobes with the ICM (Fabian, 2003) or by shock fronts (Forman et al. 2004, Nulsen et al. 2005a, 2005b, McNamara et al. 2005). A promising picture seems to be emerging where the recycling of AGN energy with powers in the order of $`10^{44}`$ to $`10^{45}`$ erg s<sup>-1</sup> in cluster cooling cores of nearby clusters could explain the observed phenomena.
Searching through larger volumes of our Universe, more extreme cases of cluster cooling cores can be found, where the rate of AGN energy recycling is even higher by up to an order of magnitude. The cluster with the largest cooling core detected so far, RXCJ1347.5-1144 at z=0.4516 was discovered in the REFLEX survey (Schindler et al., 1995; Böhringer et al. 2004a, ) with a formally derived cooling flow mass deposition rate of the order of 3000 M yr<sup>-1</sup> (for a Hubble constant of $`H_0=50`$ km s<sup>-1</sup> Mpc<sup>-1</sup>; Schindler et al. 1997; Allen 2000) which corresponds to an energy dissipation of $`10^{46}`$ erg s<sup>-1</sup>. This cluster is unfortunately much more distant than the well studied nearby clusters and does not lend itself easily to a detailed observational study. We have now discovered a more nearby, similarly striking massive cooling core cluster in the REFLEX Survey, RXCJ1504.1-0248 at a redshift of $`z=0.2153`$. This cluster was flagged as a cluster candidate due to a galaxy overdensity detected in the COSMOS data base (McGillivray & Stobie, 1984) and six concordant galaxy redshifts found in our subsequent follow-up observations confirmed the existence of a cluster. Three of these cluster galaxies show AGN-like spectra.
Serious doubts remained about the cluster identification of this X-ray emitter, because the X-ray source appeared much too compact for its high luminosity, compared to the other clusters in the REFLEX sample in the same distance and luminosity range. This could possibly be attributed to a contaminating central AGN. The certain identification clearly required a higher resolution X-ray observation, which could recently be made through a CHANDRA snap-shot exposure yielding a perfect cluster image without significant contamination by point sources (Fig.1). With these source properties it becomes immediately clear that RXCJ1504.1-0248 must have an extremely bright core and is potentially a very interesting cooling core cluster.
In this paper we study the structure of this cluster in more detail. In section 2 and 3 we present the observational results. Section 4 is devoted to the modeling of the mass profile and the determination of the parameters in the frame of a classical cooling flow model. In section 5 we discuss further phenomenological features related to the cooling core of the cluster and section 6 provides the conclusion. We will adopt a cosmological model with $`\mathrm{\Omega }_m=0.3`$, $`\mathrm{\Omega }_\mathrm{\Lambda }=0.7`$ and $`H_0=70`$ km s<sup>-1</sup> Mpc<sup>-1</sup>, except if it is noted otherwise. Thus 1 arcmin at the distance of RXCJ1504-0248 corresponds to 209 kpc.
## 2 Observation and Data preparation
RXCJ1504.1-0248 was observed with the CHANDRA ACIS-I on January 7, 2004 for 13463 sec. The observation was hardly disturbed by times of high background and the net exposure after standard cleaning procedures is 13 298 sec. Fig. 2 shows an image of the cluster in the 0.5 - 2 keV energy band superposed on an optical R-band image taken as 5 min exposure in our REFLEX redshift survey at the ESO La Silla 3.6m telescope. The total ACIS-I count rate in the region $`r6`$ arcmin in the 0.5 to 2 kev band is 2.03 cts s<sup>-1</sup> implying an unabsorbed flux of about $`1.18`$ and $`1.910^{11}`$ erg s<sup>-1</sup> cm<sup>-2</sup> in the $`0.52`$ kev and $`0.12.4`$ keV energy bands, respectively. In the ROSAT All-Sky Survey we found a flux of $`2.2(\pm 0.11)10^{11}`$ erg s<sup>-1</sup> cm<sup>-2</sup> in the $`0.12.4`$ keV band within an aperture of 12 arcmin radius, in reasonable agreement with the new results from the much deeper image.
The X-ray cluster center coincides well with a central dominant galaxy that can be identified with LCRS B150131.5-023636 at the J2000 position 15 04 07.5 -02 48 16 at z=0.216917 (Shechtman et al. 1996). The optical image of the cluster and the spectrum of the galaxy B150131.5-023636 described in section 5 have been obtained with EFOSC2 at the 3.6m telescope of ESO La Silla on Aug. 14 and 20, 2001, respectively.
## 3 Analysis and Results
The above noted fluxes imply an X-ray luminosity of $`L_x=2.310^{45}h_{70}^1`$ erg s<sup>-1</sup> in the $`0.12.4`$ keV band and a bolometric X-ray luminosity of $`L_x=4.310^{45}h_{70}^1`$ erg s<sup>-1</sup>. This makes this cluster the most prominent X-ray luminous cluster in the southern sky at redshifts below $`z=0.34`$, with only two galaxy clusters at larger distances in the REFLEX catalogue having a higher X-ray luminosity. The X-ray image in Fig. 1 shows a high degree of regularity with a slightly elliptical shape and a major axis approximately along a position angle of about 40 degrees (North to East). As seen in Fig. 2 the center of the X-ray emission is marked by a dominant giant galaxy in the optical.
The azimuthally averaged X-ray surface brightness profile from which the background was subtracted is well described by a $`\beta `$-model out to a radius of about 300 arcsec, outside which the background subtraction uncertainties become significant (Fig. 3). The background was estimated either from the outer region of the CCD or from an external background field. Remarkable is the small core radius of $`r_c30h_{70}^1`$ kpc and the high central gas density of $`n_{e0}0.16h_{70}^{1/2}`$ cm<sup>-3</sup>. The slope parameter, $`\beta `$, has the very typical value of $`0.6`$.
The temperature profile was determined by fitting a spectral MEKAL model with fixed galactic absorption with a column density of $`610^{20}`$ cm<sup>-2</sup> (as measured from 21 cm observations, Dickey & Lockman 1990) to the spectra obtained from the photons extracted from concentric rings around the cluster center. The background for subtraction was taken either from a background region at the outer parts of the detector (radial zone 3.8 to 5.7 arcmin) or from a background field in the same rings as taken for the target spectral extraction. Since there is some faint X-ray emission from the cluster almost throughout the entire detector, the on-target background subtraction prevents an accurate temperature determination at larger radii. Fig 4 provides a comparison of both methods, showing that the two approaches yield practically identical results out to a radius of 1 arcmin, but the analysis based on an external background field can be extended to a radius of 3 arcmin. While the bulk temperature of the cluster is about 10.5 keV, we note a strong temperature drop towards the center to a value below 5 keV. Such a temperature drop by a factor of 2 or 3 is observed in many cooling core clusters (e.g. De Grandi & Molendi 2002; Fabian 2003; Ikebe et al. 2004; Sanders et al. 2004). A possible temperature drop to larger radii indicated by the data cannot be established with the present photon statistics. Fig. 5 shows the spectrum of the innermost circle (0 - 15 arcsec). For the given photon statistics the spectrum is well fit by a one-temperature MEKAL model. We do not note any features which could indicate the Fe L line complex observed at lower temperatures, which would indicate the presence of cooler temperature phases.
In Fig. 4 we show 3 rough fits of analytic expressions to the temperature profile which were chosen to approximately bracket the inner and outer gradients of the temperature profile. The analytical expressions are:
$`T_1(\mathrm{keV})=3.5+0.44r^{0.9}0.044r^{1.4}+0.0007r^2`$ (for $`r<150`$)
and $`T_1=11.0760.00544r`$ (for $`r>150`$)
$`T_2(\mathrm{keV})=3.0+0.305r^{0.9}0.0192r^{1.4}`$ (for $`r<100`$)
and $`T_2=9.80+0.003322r`$ (for $`r>100`$)
$`T_3(\mathrm{keV})=3.5+0.52r^{0.9}0.0526r^{1.4}+0.00083r^2`$ (for $`r<150`$)
and $`T_3=12.9320.01364r`$ (for $`r>150`$)
where the radius, r, is in units of arcsec.
The spectral fits also indicate abundances of heavy elements (dominated by the fit to the Fe K line) of about 0.3 to 0.4 solar with large errors of about 0.1 to 0.2 in solar units (abundances based on Anders & Grevesse 1989). To our surprise we do not find a strong increase of the iron abundance towards the center, as seen in many cooling flow clusters (DeGrandi & Molendi 2001), but a deeper observation is necessary to draw a firm conclusion.
## 4 Modeling the cluster structure
### 4.1 Mass profile
From an analytical deprojection of the $`\beta `$-model fit to the X-ray surface brightness profile of the cluster and the temperature profile we can obtain the cluster mass profile under the assumption of hydrostatic equilibrium and spherical symmetry. The resulting mass profile is shown in Fig. 6 together with the profile of the gas mass, as derived from the $`\beta `$-model fit. For the gravitational mass profile we also indicate the typical uncertainties determined from the local minima and maxima of the mass profile for the different analytical temperature fits shown in Fig. 4. An additional scaling uncertainty of 15% for $`T_1`$ and $`T_2`$ and 5% for $`T_3`$ is included. At a radius of $`3h_{70}^1`$ Mpc the total mass is $`1.8(+0.35,0.28)10^{15}h_{70}^1`$ M. For the radius $`r_{200}=2.3`$ Mpc we find a mass of $`1.510^{15}h_{70}^1`$ M. Thus RXCJ1504-0248 is among the most massive clusters known. The gas-to-total mass ratio for the two fiducial radii is $`0.17(+0.03,0.07)`$ and $`0.15(+0.03,0.05)`$, respectively.
### 4.2 Cooling flow analysis
To gain an understanding of the processes occuring in the central ICM of this cluster we start with a classical cooling flow analysis (e.g. Fabian et al. 1984; Thomas et al. 1987). We take two approaches. For the more simple model A we equate the energy loss by radiation inside a given radius, $`r`$, with the enthalpy influx from outside through the sphere with radius $`r`$. In model B we formulate the energy balance in a local differential way and include the gain of gravitational energy of the material flowing in from the outer regions.
$$\dot{M}(r)=4\pi r^2\frac{n_e^2\mathrm{\Lambda }(T)+\frac{\dot{M}(r+\mathrm{\Delta }r)}{\mathrm{\Delta }r4\pi r^2}\frac{5k_BT}{2\mu m_p}}{\frac{5k_B}{2\mu m_p}\left(\frac{T}{\mathrm{\Delta }r}+\frac{dT}{dr}\right)+\frac{M(r)G}{r}+\frac{\mathrm{\Phi }}{\mathrm{\Delta }r}}$$
where $`\mathrm{\Lambda }(T)`$ is the cooling function normalized to the electron density squared, $`\mathrm{\Delta }r`$ is the shell width in the numerical calculation, $`\mathrm{\Phi }`$ is the gravitational potential, and the other symbols have their usual meaning.
Fig. 7 shows the cooling time as a function of the cluster radius. If we take the often used fiducial value of $`10^{10}`$ years, we find a cooling radius of $`140(\pm 5)`$ kpc (for a Hubble constant of $`h_{100}=H_0/(100`$ km s<sup>-1</sup> Mpc$`{}_{}{}^{1})=0.7`$) and $`165(\pm 5)`$ kpc (for $`h_{100}=0.5`$). Here the uncertainty is determined from the minima and maxima for the different adopted fits to the temperature profile. Fig. 8 then shows the mass flow rates determined for model A and B and the cooling radius is indicated by vertical lines. For this adopted cooling radius we find mass flow rates of 1400 and 1900 M yr<sup>-1</sup> (for $`h_{100}=0.7`$) and 2300 and 2930 (for $`h_{100}=0.5`$) for model B and A, respectively.
These high formal cooling flow mass deposition rates make RXCJ1504-0248 the most prominent cooling flow cluster next to the most luminous cluster known, RXCJ1347-1144 at $`z=0.45`$ for which also a mass deposition rate of the order of 3000 M yr<sup>-1</sup> (for $`h_{100}=0.5`$) was deduced (Schindler et al. 1997; Allen 2000). This makes RXCJ1504-0248 a very interesting target to study the cooling core phenomenon under the most extreme conditions.
## 5 Discussion
In the new scenario of the physics of cooling core clusters, the large radiative cooling rates are compensated by the energy released from a central AGN (e.g. Churazov et al. 2000, 2001; McNamara et al. 2000; David et al. 2001; Böhringer et al. 2002; Fabian 2003; Forman et al. 2004). To keep the balance such that neither massive mass condensation nor a dispersion of the dense gaseous core occurs, the heating has to be fine-tuned. This is achieved by a self-regulation system where large mass deposition rates in the center lead to an increased feedback from the AGN which limits the cooling rate. Seen from the perspective of the central AGN, its accretion rate is limited by the amount of cooling that can occur, that is the energy that can be dissipated by the ICM in the cooling core region (Churazov et al. 2002). In this cooling core scenario a high radiative power of the central ICM indicates a fast accretion of the interacting AGN.
The case of RXCJ1504-0248 is an extreme case in this scenario. Since the core radius is so small, actually much smaller than the cooling radius by a factor of about four, the major part of the total X-ray luminosity (about 72%) originates from inside the cooling radius. This corresponds to a total radiation power of $`310^{45}`$ erg s<sup>-1</sup>. So far we have no direct indication that this cooling rate is balanced by AGN heating in this system. Some support for this scenario is discussed below. Assuming that the above sketched cooling core scenario applies and that the observed cooling power inside the cooling radius is balanced by the energy output from the AGN we can calculate further interesting system parameters. If the radiation power is replenished by accretion power from the AGN and if we assume an energy return efficiency, $`\eta =0.1`$ from accretion onto the AGN black hole we can imply an accretion rate of the order of $`0.5`$ M yr<sup>-1</sup>. Thus, in this mode the central black hole can gain a considerable mass of the order of the most massive black holes known over cosmic times.
Therefore it is very interesting to see if this scenario actually applies to RXCJ1504-0248. So far the observational evidence is far less detailed as for the best observed nearby cooling core clusters and the implications indicated above remain very speculative. But the few additional features known, point in the right direction. The central dominant galaxy is known to harbor a radio source with a brightness of 62 mJy at 1.4 GHz (Bauer et al. 2000) and thus it presumably contains a massive black hole. The radio source image obtained from the NVSS survey is, however, unresolved and featureless. A zoom into the central region of the CHANDRA image (Fig. 9) shows indications of an asymmetric distortion which could be due to the interaction of the AGN jets with the intracluster medium. Better photon statistics is needed to draw a more firm conclusion. Therefore deeper X-ray observations have been scheduled for CHANDRA and a detailed VLA radio study has been proposed for this cluster.
The central dominant galaxy is extremely large. The Gunn r magnitude of 16.4 determined in the Las Campanas redshift survey (Shechtman et al. 1996) translates into an absolute magnitude of $`M_r24`$ which corresponds to about $`3\times 10^{11}`$ L in this band. It places this galaxy in the upper few percent of the luminosity distribution of cluster central galaxies as found for example by comparison to the survey by Lauer and Postman (1994).
The optical spectrum of the central galaxy, shown in Fig. 10, shows narrow, low excitation emission lines, notably strong \[O II\], weaker \[O III\], and bright H$`\alpha `$/\[NII\] lines, similar to what has been observed for many massive cooling flows (e.g. Hu, Cowie & Wang 1985; Johnstone, Fabian & Nulsen 1987, Heckman et al. 1989, Donahue, Stocke & Gioia 1989; McNamara & O’Connell 1993; Crawford et al. 1999). This provides another hint that this cluster resembles a scaled-up version of the known nearby cooling core clusters.
A very similar spectrum has been observed and studied in detail in the central galaxy of the cooling core galaxy cluster A2597 by Voit & Donahue (1997). In the spectrum of A2597 the \[OII\] line is even more dominant compared to the other lines. Voit & Donahue provide a very comprehensive discussion on the origin of this spectrum. They conclude that hot stars constitute the best fitting source of ionization, but to match the spectral properties with an photoionization nebular model an additional heat source has to be assumed which may well be the heating source of the cooling core. The same discussion most probably applies also to this object, but to perform the same analysis a deeper spectroscopic observation is required to get accurate line fluxes for more diagnostic lines. Very similar spectra can also be found among the spectra compiled from central cluster galaxies by Crawford et al. (1999). It is very striking that the spectra which best match the spectrum of RXCJ1504-0248 are those from the prominent well known cooling flow clusters, such as Z3146, A1835, A2204, and A2390. Overall these spectra show some variation in the degree of ionization with \[OII\] and \[OI\] lines being more prominent compared to \[OIII\] in Z3146 and \[OIII\] being relatively more prominent in A2390 than in the present case. But the close resemblance is very obvious. Within the classification scheme of AGN the present observation features as a LINER spectrum.
The equivalent width of the \[OII\] line is about $`151\pm 2.43`$ Å corresponding to a line flux of about $`9\times 10^{14}`$ erg s<sup>-1</sup> cm<sup>-2</sup>. Together with the total luminosity of the galaxy and the relation of \[OII\] luminosity and star formation rate as given by Kennicutt (1992) we get formally a star formation rate of the order of about 50 M yr<sup>-1</sup>. This is not untypical for a massive cooling core cluster (e.g. McNamara 1997). Since there are surely other ionization sources present, this simple modeling is certainly an oversimplification.
## 6 Conclusion
The CHANDRA observation reveals the X-ray source RXCJ1504-0248 as a galaxy cluster that has extreme and surprising properties in two respects. It is found to be a very massive cluster and the most luminous cluster known in the southern sky at redshifts lower than $`z0.3`$. Secondly, the cluster appears extremely compact with a very dense central region. Thus the high X-ray luminosity is the result of both, the large cluster mass and the high central density.
These properties would make the cluster a cooling flow with one of the largest mass deposition rates ever inferred. But the observation of a radio AGN in the cluster center and the absence of low temperature signatures in the central X-ray spectrum leads us to suspect that the dense central ICM region is heated by the AGN like it is implied from the X-ray observations for nearby clusters. The heating source required in RXCJ1504-0248 needs to have a power about an order of magnitude larger than that of the nearby well studied cooling core clusters. This extreme energy recycling will surely make RXCJ1504-0248 a very interesting study object for many future investigations. This X-ray source has been accepted for observations with ASTRO-E2 which should for example provide insight into the degree of turbulence prevailing in the cooling core region.
This study was made possible by the highly advaced capabilities of the NASA CHANDRA Observatory. We also acknowledge support for our international collaboration by NASA under the grant G04-5142X. |
warning/0507/math0507220.html | ar5iv | text | # Percolation theory
## Introduction
Percolation as a mathematical theory was introduced by Broadbent and Hammersley , as a stochastic way of modeling the flow of a fluid or gas through a porous medium of small channels which may or may not let gas or fluid pass. It is one of the simplest models exhibiting a phase transition, and the occurrence of a critical phenomenon is central to the appeal of percolation. Having truly applied origins, percolation has been used to model the fingering and spreading of oil in water, to estimate whether one can build non-defective integrated circuits, to model the spread of infections and forest fires. From a mathematical point of view percolation is attractive because it exhibits relations between probabilistic and algebraic/topological properties of graphs.
To make the mathematical construction of such a system of channels, take a graph $`𝒢`$ (which originally was taken as $`^d`$), with vertex set $`𝒱`$ and edge set $``$, and make all the edges independently *open* (or passable) with probability $`p`$ or *closed* (or blocked) with probability $`1p`$. Write $`P_p`$ for the corresponding probability measure on the set of configurations of open and closed edges — that model is called *bond percolation*. The collection of open edges thus forms a random subgraph of $`𝒢`$, and the original question stated by Broadbent was whether the connected component of the origin in that subgraph is finite or infinite.
A *path* on $`𝒢`$ is a sequence $`v_1,v_2,\mathrm{}`$ of vertices of $`𝒢`$, such that for all $`i1`$, $`v_i`$ and $`v_{i+1}`$ are adjacent on $`𝒢`$. A path is called *open* if all the edges $`\{v_i,v_{i+1}\}`$ between successive vertices are open. The infiniteness of the cluster of the origin is equivalent to the existence of an unbounded open path starting from the origin.
There is an analogous model, called *site percolation*, in which all edges are assumed being passable, but the vertices are independently open or closed with probability $`p`$ or $`1p`$, respectively. An open path is then a path along which all vertices are open. Site percolation is more general than bond percolation in the sense that the existence of a path for bond percolation on a graph $`𝒢`$ is equivalent to the existence of a path for site percolation on the covering graph of $`𝒢`$. However, site percolation on a given graph may not be equivalent to bond percolation on any other graph.
All graphs under consideration will be assumed to be connected, locally finite and quasi-transitive. If $`A,B𝒱`$, then $`AB`$ means that there exists an open path from some vertex of $`A`$ to some vertex of $`B`$; by a slight abuse of notation, $`uv`$ will stand for the existence of a path between sites $`u`$ and $`v`$, *i.e.* the event $`\{u\}\{v\}`$. The *open cluster* $`C(v)`$ of the vertex $`v`$ is the set of all open vertices which are connected to $`v`$ by an open path:
$$C(v)=\{u𝒱:uv\}.$$
The central quantity of the percolation theory is the *percolation probability*:
$$\theta (p):=P_p\{\text{0}\mathrm{}\}=P_p\{|C(\text{0})|=\mathrm{}\}.$$
The most important property of the percolation model is that it exhibits a *phase transition*, *i.e.* there exists a threshold value $`p_c[0,1]`$, such that the global behavior of the system is substantially different in the two regions $`p<p_c`$ and $`p>p_c`$. To make this precise, observe that $`\theta `$ is a non-decreasing function. This can be seen using Hammersley’s joint construction of percolation systems for all $`p[0,1]`$ on $`𝒢`$: Let $`\{U(v),v𝒱\}`$ be independent random variables, uniform in $`[0,1]`$. Declare $`v`$ to be $`p`$-open if $`U(v)p`$, otherwise it is declared $`p`$-closed. The configuration of $`p`$-open vertices has the distribution $`P_p`$ for each $`p[0,1]`$. The collection of $`p`$-open vertices is non-decreasing in $`p`$, and therefore $`\theta (p)`$ is non-decreasing as well. Clearly $`\theta (0)=0`$ and $`\theta (1)=1`$.
The *critical probability* is defined as
$$p_c:=p_c(𝒢)=sup\{p:\theta (p)=0\}.$$
By definition, when $`p<p_c`$ the open cluster of the origin is $`P_p`$-a.s. finite, hence all the clusters are also finite. On the other hand, for $`p>p_c`$ there is a strictly positive $`P_p`$-probability that the cluster of the origin is infinite. Thus, from Kolmogorov’s zero-one law it follows that
$$P_p\{|C(v)|=\mathrm{}\text{for some }v𝒱\}=1\text{for }p>p_c.$$
Therefore, if the intervals $`[0,p_c)`$ and $`(p_c,1]`$ are both non-empty, there is a phase transition at $`p_c`$.
Using a so-called Peierls argument it is easy to see that $`p_c(𝒢)>0`$ for any graph $`𝒢`$ of bounded degree. On the other hand, Hammersley proved that $`p_c(^d)<1`$ for bond percolation as soon as $`d2`$, and a similar argument works for site percolation and various periodic graphs as well. But for some graphs $`𝒢`$ it is not so easy to show that $`p_c(𝒢)<1`$. One says that the system is in the *subcritical (resp. supercritical) phase* if $`p<p_c`$ (resp. $`p>p_c`$).
It was one of the most remarkable moments in the history of percolation when Kesten proved that the critical parameter for bond-percolation on $`^2`$ is equal to $`1/2`$. Nevertheless the exact value of $`p_c(𝒢)`$ is known only for a handful of graphs, all of them periodic and two-dimensional — see below.
## 1 Percolation in $`^d`$
The graph on which most of the theory was originally built is the cubic lattice $`^d`$, and it was not before the late 20th century that percolation was seriously considered on other kinds of graphs (such as *e.g.* Cayley graphs), on which specific phenomena can appear, such as the coexistence of multiple infinite clusters for some values of the parameter $`p`$. In all this section, the underlying graph is thus assumed to be $`^d`$ for $`d2`$, although most of the results still hold in the case of a periodic $`d`$-dimensional lattice.
### 1.1 The sub-critical regime
When $`p<p_c`$, all open clusters are finite almost surely. One of the greatest challenges in percolation theory has been to prove that $`\chi (p):=E_p\{|C(v)|\}`$ is finite if $`p<p_c`$ ($`E_p`$ stands for the expectation with respect to $`P_p`$). For that one can define another critical probability as the threshold value for the finiteness of the expected cluster size of a fixed vertex:
$$p_T(𝒢):=sup\{p:\chi (p)<\mathrm{}\}.$$
It was an important step in the development of the theory to show that $`p_T(𝒢)=p_c(𝒢)`$. The fundamental estimate in the subcritical regime, which is a much stronger statement than $`p_T(𝒢)=p_c(𝒢)`$, is the following:
###### Theorem 1 (Aizenman and Barsky, Menshikov)
Assume that $`𝒢`$ is periodic. Then for $`p<p_c`$ there exist constants $`0<C_1,C_2<\mathrm{}`$, such that
$$P_p\{|C(v)|n\}C_1e^{C_2n}.$$
The last statement can be sharpened to a “local limit theorem” with the help of a subadditivity argument : For each $`p<p_c`$ there exists a constant $`0<C_3(p)<\mathrm{}`$, such that
$$\underset{n\mathrm{}}{lim}\frac{1}{n}\mathrm{log}P_p\{|C(v)|=n\}=C_3(p).$$
### 1.2 The super-critical regime
Once an infinite open cluster exists, it is natural to ask how it looks like, and how many infinite open clusters exist. It was shown by Newman and Schulman that for periodic graphs, for each $`p`$, exactly one of the following three situations prevails: If $`N_+\{\mathrm{}\}`$ is the number of infinite open clusters, then $`P_p(N=0)=1`$, or $`P_p(N=1)=1`$, or $`P_p(N=\mathrm{})=1`$.
Aizenman, Kesten and Newman showed that the third case is impossible on $`^d`$. By now several proofs exist, perhaps the most elegant proof of that is due to Burton and Keane, who prove that indeed there cannot be infinitely many infinite open clusters on any amenable graph. However, there are some graphs, such as regular trees, on which coexistence of several infinite clusters is possible.
The geometry of the infinite open cluster can be explored in some depth by studying the behavior of a random walk on it. When $`d=2`$, the random walk is recurrent, and when $`d3`$ is a.s. transient. In all dimensions $`d2`$ the walk behaves diffusively, and the Central Limit Theorem and the Invariance principle were established in both the annealed and quenched cases.
#### Wulff droplets
In the supercritical regime, aside from the infinite open cluster, the configuration contains finite clusters of arbitrary large sizes. These large finite open clusters can be thought of as droplets swimming in the areas surrounded by an infinite open cluster. The presence at a particular location of a large finite cluster is an event of low probability, namely, on $`^d`$, $`d2`$, for $`p>p_c`$, there exist positive constants $`0<C_4(p),C_5(p)<\mathrm{}`$, such that
$$C_4(p)\frac{1}{n^{(d1)/d}}\mathrm{log}P_p\{|C(v)|=n\}C_5(p)$$
for all large $`n`$. This estimate is based on the fact that the occurrence of a large finite cluster is due to a surface effect. The typical structure of the large finite cluster is described by the following theorem:
###### Theorem 2
Let $`d2`$, and $`p>p_c`$. There exists a bounded, closed, convex subset $`W`$ of $`^d`$ containing the origin, called the *normalized Wulff crystal* of the Bernoulli percolation model, such that, under the conditional probability $`P_p\{n^d|C(\text{0})|<\mathrm{}\}`$, the random measure
$$\frac{1}{n^d}\underset{xC(\text{0})}{}\delta _{x/n}$$
(where $`\delta _𝐱`$ denotes a Dirac mass at $`𝐱`$) converges weakly in probability towards the random measure $`\theta (p)\mathrm{𝟙}_W(xM)dx`$ (where $`M`$ is the rescaled center of mass of the cluster $`C(\mathrm{𝟎})`$). The deviation probabilities behave as $`\mathrm{exp}\{cn^{d1}\}`$ (*i.e.* they exhibit large deviations of surface order).
This result was proved in dimension $`2`$ by Alexander, Chayes and Chayes , and in dimensions $`3`$ and more by Cerf .
### 1.3 Percolation near the critical point
#### 1.3.1 Percolation in slabs
The main macroscopic observable in percolation is $`\theta (p)`$, which is positive above $`p_c`$, $`0`$ below $`p_c`$, and continuous on $`[0,1]\{p_c\}`$. Continuity at $`p_c`$ is an open question in the general case; it is known to hold in two dimensions (cf. below) and in high enough dimension (at the moment $`d19`$ though the value of the critical dimension is believed to be $`6`$) using lace expansion methods. The conjecture that $`\theta (p_c)=0`$ for $`3d18`$ remains one of the major open problems.
Efforts to prove that led to some interesting and important results. Barsky, Grimmett and Newman solved the question in the half-space case, and simultaneously showed that the slab percolation and half-space percolation thresholds coincide. This was complemented by Grimmett and Marstrand showing that
$$p_c(slab)=p_c(^d).$$
#### 1.3.2 Critical exponents
In the sub-critical regime, exponential decay of the correlation indicates that there is a finite *correlation length* $`\xi (p)`$ associated to the system, and defined (up to constants) by the relation
$$P_p(0n𝐱)\mathrm{exp}(\frac{n\phi (x)}{\xi (p)})$$
where $`\phi `$ is bounded on the unit sphere (this is known as *Ornstein-Zernike decay*). The phase transition can then also be defined in terms of the divergence of the correlation length, leading again to the same value for $`p_c`$; the behavior at or near the critical point then has no finite characteristic length, and gives rise to scaling exponents (conjecturally in most cases).
The most usual critical exponents are defined as follows, if $`\theta (p)`$ is the percolation probability, $`C`$ the cluster of the origin, and $`\xi (p)`$ the correlation length:
$`{\displaystyle \frac{^3}{p^3}}E_p[|C|^1]`$ $`|pp_c|^{1\alpha }`$
$`\theta (p)`$ $`(pp_c)_+^\beta `$
$`\chi ^f(p):=E_p[|C|\mathrm{𝟙}_{|C|<\mathrm{}}]`$ $`|pp_c|^\gamma `$
$`P_{p_c}[|C|=n]`$ $`n^{11/\delta }`$
$`P_{p_c}[xC]`$ $`|x|^{2d\eta }`$
$`\xi (p)`$ $`|pp_c|^\nu `$
$`P_{p_c}[\mathrm{diam}(C)=n]`$ $`n^{11/\rho }`$
$`{\displaystyle \frac{E_p[|C|^{k+1}\mathrm{𝟙}_{|C|<\mathrm{}}]}{E_p[|C|^k\mathrm{𝟙}_{|C|<\mathrm{}}]}}`$ $`|pp_c|^\mathrm{\Delta }`$
These exponents are all expected to be universal, *i.e.* to depend only on the dimension of the lattice, although this is not well understood at the mathematical level; the following *scaling relations* between the exponents are believed to hold:
$$2\alpha =\gamma +2\beta =\beta (\delta +1),\mathrm{\Delta }=\delta \beta ,\gamma =\nu (2\eta ).$$
In addition, in dimensions up to $`d_c=6`$, two additional *hyperscaling relations* involving $`d`$ are strongly conjectured to hold:
$$d\rho =\delta +1,d\nu =2\alpha ,$$
while above $`d_c`$ the exponents are believed to take their mean-field value, *i.e.* the ones they have for percolation on a regular tree:
$$\alpha =1,\beta =1,\gamma =1,\delta =2,$$
$$\eta =0,\nu =\frac{1}{2},\rho =\frac{1}{2},\mathrm{\Delta }=2.$$
Not much is known rigorously on critical exponents in the general case. Hara and Slade () proved that mean field behavior does happen above dimension $`19`$, and the proof can likely be extended to treat the case $`d7`$. In the two-dimensional case on the other hand, Kesten () showed that, assuming that the exponents $`\delta `$ and $`\rho `$ exist, then so do $`\beta `$, $`\gamma `$, $`\eta `$ and $`\nu `$, and they satisfy the scaling and hyperscaling relations where they appear.
#### 1.3.3 The incipient infinite cluster
When studying long-range properties of a critical model, it is useful to have an object which is infinite at criticality, and such is not the case for percolation clusters. There are two ways to condition the cluster of the origin to be infinite when $`p=p_c`$: The first one is to condition it to have diameter at least $`n`$ (which happens with positive probability) and take a limit in distribution as $`n`$ goes to infinity; the second one is to consider the model for parameter $`p>p_c`$, condition the cluster of $`0`$ to be infinite (which happens with positive probability) and take a limit in distribution as $`p`$ goes to $`p_c`$. The limit is the same in both cases, it is known as the *incipient infinite cluster*.
As in the super-critical regime, the structure of the cluster can be investigated by studying the behavior of a random walk on it, as was suggested by de Gennes; Kesten proved that in two dimensions, the random walk on the incipient infinite cluster is sub-diffusive, *i.e.* the mean square displacement after $`n`$ steps behaves as $`n^{1\epsilon }`$ for some $`\epsilon >0`$.
The construction of the incipient infinite cluster was done by Kesten in two dimensions , and a similar construction was performed recently in high dimension by Van der Hofstad and Jarai ().
## 2 Percolation in two dimensions
As is the case for several other models of statistical physics, percolation exhibits many specific properties when considered on a two-dimensional lattice: Duality arguments allow for the computation of $`p_c`$ in some cases, and for the derivation of *a priori* bounds for the probability of crossing events at or near the critical point, leading to the fact that $`\theta (p_c)=0`$. On another front, the scaling limit of critical site-percolation on the two-dimensional triangular lattice can be described in terms of SLE processes.
### 2.1 Duality, exact computations and RSW theory
Given a planar lattice $``$, define two associated graphs as follows. The *dual lattice* $`^{}`$ has one vertex for each face of the original lattice, and an edge between two vertices if and only if the corresponding faces of $``$ share an edge. The *star graph* $`^{}`$ is obtained by adding to $``$ an edge between any two vertices belonging to the same face ($`^{}`$ is not planar in general; $`(,^{})`$ is commonly known as a *matching pair*). Then, a result of Kesten is that, under suitable technical conditions,
$$p_c^{\mathrm{bond}}()+p_c^{\mathrm{bond}}(^{})=p_c^{\mathrm{site}}()+p_c^{\mathrm{site}}(^{})=1.$$
Two cases are of particular importance: The lattice $`^2`$ is isomorphic to its dual; the triangular lattice $`𝒯`$ is its own star graph. It follows that
$$p_c^{\mathrm{bond}}(^2)=p_c^{\mathrm{site}}(𝒯)=\frac{1}{2}.$$
The only other critical parameters that are known exactly are $`p_c^{\mathrm{bond}}(𝒯)=2\mathrm{sin}(\pi /18)`$ (and hence also $`p_c^{\mathrm{bond}}`$ for $`𝒯^{}`$, *i.e.* the hexagonal lattice), and $`p_c^{\mathrm{bond}}`$ for the bow-tie lattice which is a root of the equation $`p^56p^3+6p^2+p1=0`$. The value of the critical parameter for site-percolation on $`^2`$ might on the other hand never be known, it is even possible that it is “just a number” without any other signification.
Still using duality, one can prove that the probability, for bond-percolation on the square lattice with parameter $`p=1/2`$, that there is a connected component crossing an $`(n+1)\times n`$ rectangle in the longer direction is exactly equal to $`1/2`$. This and clever arguments involving the symmetry of the lattice lead to the following result, proved independently by Russo and by Seymour and Welsh and known as the RSW theorem:
###### Theorem 3 (Russo ; Seymour-Welsh )
For every $`a,b>0`$ there exist $`\eta >0`$ and $`n_0>0`$ such that for every $`n>n_0`$, the probability that there is a cluster crossing an $`na\times nb`$ rectangle in the first direction is greater than $`\eta `$.
The most direct consequence of this estimate is that the probability that there is a cluster going around an annulus of a given modulus is bounded below independently of the size of the annulus; in particular, almost surely there is *some* annulus around $`0`$ in which this happens, and that is what allows to prove that $`\theta (p_c)=0`$ for bond-percolation on $`^2`$.
### 2.2 The scaling limit
RSW-type estimates give positive evidence that a scaling limit of the model should exist; it is indeed essentially sufficient to show convergence of the crossing probabilities to a non-trivial limit as $`n`$ goes to infinity. The limit, which should depend only on the ration $`a/b`$, was predicted by Cardy using conformal fields theory methods. A most celebrated result of Smirnov is the proof of Cardy’s formula in the case of site-percolation on the triangular lattice $`𝒯`$:
###### Theorem 4 (Smirnov )
Let $`\mathrm{\Omega }`$ be a simply connected domain of the plane with four points $`a`$, $`b`$, $`c`$, $`d`$ (in that order) marked on its boundary. For every $`\delta >0`$, consider a critical site-percolation model on the intersection of $`\mathrm{\Omega }`$ with $`\delta 𝒯`$ and let $`f_\delta (ab,cd;\mathrm{\Omega })`$ be the probability that it contains a cluster connecting the arcs $`ab`$ and $`cd`$. Then:
1. $`f_\delta (ab,cd;\mathrm{\Omega })`$ has a limit $`f_0(ab,cd;\mathrm{\Omega })`$ as $`\delta 0`$;
2. The limit is conformally invariant, in the following sense: If $`\mathrm{\Phi }`$ is a conformal map from $`\mathrm{\Omega }`$ to some other domain $`\mathrm{\Omega }^{}=\mathrm{\Phi }(\mathrm{\Omega })`$, and maps $`a`$ to $`a^{}`$, $`b`$ to $`b^{}`$, $`c`$ to $`c^{}`$ and $`d`$ to $`d^{}`$, then $`f_0(ab,cd;\mathrm{\Omega })=f_0(a^{}b^{},c^{}d^{};\mathrm{\Omega }^{})`$;
3. In the particular case when $`\mathrm{\Omega }`$ is an equilateral triangle of side length $`1`$ and vertices $`a`$, $`b`$ and $`c`$, and if $`d`$ is on $`(ca)`$ at distance $`x(0,1)`$ from $`c`$, then $`f_0(ab,cd;\mathrm{\Omega })=x`$.
Point 3. in particular is essential since it allows to compute the limiting crossing probabilities in any conformal rectangle. In the original work of Cardy, he made his prediction in the case of a rectangle, for which the limit involves hypergeometric functions; the remark that the equilateral triangle gives rise to nicer formulae is originally due to Carleson.
To precisely state the convergence of percolation to its scaling limit, define the random curve known as the *percolation exploration path* (see fig. 3) as follows: In the upper half-plane, consider a site-percolation model on a portion of the triangular lattice and impose the boundary conditions that on the negative real half-line all the sites are open, while on the other half-line the sites are closed. The exploration curve is then the common boundary of the open cluster spanning from the negative half-line, and the closed cluster spanning from the positive half-line; it is an infinite, self-avoiding random curve in the upper half-plane.
As the mesh of the lattice goes to $`0`$, the exploration curve then converges in distribution to the trace of an $`SLE`$ process, as introduced by Schramm, with parameter $`\kappa =6`$ — see fig. 4. The limiting curve is not simple anymore (which corresponds to the existence of pivotal sites on large critical percolation clusters), and it has Hausdorff dimension $`7/4`$. For more details on $`SLE`$ processes, see *e.g.* the related entry in the present volume.
As an application of this convergence result, one can prove that the critical exponents described in the previous section do exist (still in the case of the triangular lattice), and compute their exact values, except for $`\alpha `$, which is still listed here for completeness:
$$[\alpha =\frac{2}{3},]\beta =\frac{5}{36},\gamma =\frac{43}{18},\delta =\frac{91}{5},$$
$$\eta =\frac{5}{24},\nu =\frac{4}{3},\rho =\frac{48}{5},\mathrm{\Delta }=\frac{91}{36}.$$
These exponents are expected to be *universal*, in the sense that they should be the same for percolation on any two-dimensional lattice; but at the time of this writing this phenomenon is far from being understood on a mathematical level.
The rigorous derivation of the critical exponents for percolation is due to Smirnov and Werner ; the dimension of the limiting curve was obtained by Beffara .
## 3 Other lattices and percolative systems
Some modifications or generalizations of standard Bernoulli percolation on $`^d`$ exhibit an interesting behavior and as such provide some insight into the original process as well; there are too many mathematical objects which can be argued to be percolative in some sense to give a full account of all of them, so the following list is somewhat arbitrary and by no means complete.
### 3.1 Percolation on non-amenable graphs
The first modification of the model one can think of is to modify the underlying graph and move away from the cubic lattice; phase transition still occurs, and the main difference is the possibility for infinitely many infinite clusters to coexist. On a regular tree, such is the case whenever $`p(p_c,1)`$, the first non-trivial example was produced by Grimmett and Newman as the product of $``$ by a tree: There, for some values of $`p`$ the infinite cluster is unique, while for others there is coexistence of infinitely many of them. The corresponding definition, due to Benjamini and Schramm, is then the following: If $`N`$ is as above the number of infinite open clusters,
$$p_u:=inf\{p:P_p(N=1)=1\}p_c.$$
The main question is then to characterize graphs on which $`0<p_c<p_u<1`$.
A wide class of interesting graphs is that of *Cayley graphs* of infinite, finitely generated groups. There, by a simultaneous result by Häggström and Peres and by Schonmann, for every $`p(p_c,p_u)`$ there are $`P_p`$-a.s. infinitely many infinite cluster, while for every $`p(p_u,1]`$ there is only one — note that this does not follow from the definition since new infinite components could appear when $`p`$ is increased. It is conjectured that $`p_c<p_u`$ for any Cayley graph of a non-amenable group (and more generally for any quasi-transitive graph with positive Cheeger constant), and a result by Pak and Smirnova is that every infinite, finitely generated, non-amenable group has a Cayley graph on which $`p_c>p_u`$; this is then expected not to depend on the choice of generators. In the general case, it was recently proved by Gaboriau that if the graph $`𝒢`$ is unimodular, transitive, locally finite and supports non-constant harmonic Dirichlet functions (*i.e.* harmonic functions whose gradient is in $`\mathrm{}^2`$), then indeed $`p_c(𝒢)<p_u(𝒢)`$.
For reference and further reading on the topic, the reader is advised to refer to the review paper by Benjamini and Schramm , the lecture notes of Peres, and the more recent article of Gaboriau .
### 3.2 Gradient percolation
Another possible modification of the original model is to allow the parameter $`p`$ to depend on the location; the porous medium may for instance have been created by some kind of erosion, so that there will be more open edges on one side of a given domain than on the other. If $`p`$ still varies smoothly, then one expects some regions to look subcritical and others to look supercritical, with interesting behavior in the vicinity of the critical level set $`\{p=p_c\}`$. This particular model was introduced by Sapoval et al. under the name of *gradient percolation*; see fig. 5.
The control of the model away from the critical zone is essentially the same as for usual Bernoulli percolation, the main question being how to estimate the width of the phase transition. The main idea is then the same as in scaling theory: If the distance between a point $`v`$ and the critical level set is less than the correlation length for parameter $`p(v)`$, then $`v`$ is in the phase transition domain. This of course makes sense only asymptotically, say in a large $`n\times n`$ square with $`p(x,y)=1y/n`$ as is the case in the figure: The transition then is expected to have width of order $`n^a`$ for some exponent $`a>0`$.
### 3.3 First-passage percolation
First passage percolation (also known as Eden or Richardson model) was introduced by Hammersley and Welsh in 1965 as a time-dependent model for the passage of fluid through a porous medium. To define the model, with each edge $`e(^d)`$ is associated a random variable $`T(e)`$, which can be interpreted as being the time required for fluid to flow along $`e`$. $`T(e)`$ are assumed to be independent non-negative random variables having common distribution $`F`$. For any path $`\pi `$ we define the passage time $`T(\pi )`$ of $`\pi `$ as
$$T(\pi ):=\underset{e\pi }{}T(e).$$
The *first passage time* $`a(x,y)`$ between vertices $`x`$ and $`y`$ is given by
$$a(x,y)=inf\{T(\pi ):\pi \text{ a path from }x\text{ to }y\};$$
and we can define
$$W(t):=\{x^d:a(0,x)t\},$$
the set of vertices reached by the liquid by time $`t`$. It turns out that $`W(t)`$ grows approximately linearly as time passes, and that there exists a non-random limit set $`B`$ such that either $`B`$ is compact and
$$(1\epsilon )B\frac{1}{t}\stackrel{~}{W}(t)(1+\epsilon )B,\text{eventually a.s.}$$
for all $`\epsilon >0`$, or $`B=^d`$, and
$$\{x^d:|x|K\}\frac{1}{t}\stackrel{~}{W}(t),\text{eventually a.s.}$$
for all $`K>0`$. Here $`\stackrel{~}{W}(t)=\{z+[1/2,1/2]^d:zW(t)\}`$.
Studies of first passage percolation brought many fascinating discoveries, including Kingman’s celebrated sub-additive ergodic theorem. In recent years interest has been focused on study of fluctuations of the set $`\stackrel{~}{W}(t)`$ for large $`t`$. In spite of huge effort and some partial results achieved, it still remains a major task to establish rigorously conjectures predicted by Kardar-Parisi-Zhang theory about shape fluctuations in first passage percolation.
### 3.4 Contact processes
Introduced by Harris and conceived with biological interpretation, the *contact process* on $`^d`$ is a continuous-time process taking values in the space of subsets of $`^d`$. It is informally described as follows: Particles are distributed in $`^d`$ in such a way that each site is either empty or occupied by one particle. The evolution is Markovian: Each particle disappears after an exponential time of parameter $`1`$, independently from the others; at any time, each particle has a possibility to create a new particle at any of its empty neighboring sites, and does so with rate $`\lambda >0`$, independently of everything else.
The question is then whether, starting from a finite population, the process will die out in finite time or whether it will survive forever with positive probability. The outcome will depend on the value of $`\lambda `$, and there is a critical value $`\lambda _c`$, such that for $`\lambda \lambda _c`$ process dies out, while for $`\lambda >\lambda _c`$ indeed there is survival, and in this case the shape of the population obey a shape theorem similar to that of first-passage percolation.
The analogy with percolation is strong, the corresponding percolative picture being the following: In $`_+^{d+1}`$, each edge is open with probability $`p(0,1)`$, and the question is whether there exists an infinite *oriented* path $`\pi `$ (*i.e.* a path along which the sum of the coordinates is increasing), composed of open edges. Once again, there is a critical parameter customarily denoted by $`\stackrel{}{p}_c`$, at which no such path exists (compare this to the open question of the continuity of the function $`\theta `$ at $`p_c`$ in dimensions $`3d18`$). This variation of percolation lies in a different universality class than the usual Bernoulli model.
### 3.5 Invasion percolation
Let $`X(e):e`$ be independent random variables indexed by the edge set $``$ of $`^d`$, $`d2`$, each having uniform distribution in $`[0,1]`$. One constructs a sequence $`C=\{C_i,i1\}`$ of random connected subgraphs of the lattice in the following iterative way: The graph $`C_0`$ contains only the origin. Having defined $`C_i`$, one obtains $`C_{i+1}`$ by adding to $`C_i`$ an edge $`e_{i+1}`$ (with its outer lying end-vertex), chosen from the outer edge boundary of $`C_i`$ so as to minimize $`X(e_{i+1})`$. Still very little is known about the behavior of this process.
An interesting observation, relating $`\theta (p_c)`$ of usual percolation with the invasion dynamics, comes from C.M. Newman:
$$\theta (p_c)=0P\{xC\}0\text{as }|x|\mathrm{}.$$
## Further reading
For a much more in-depth review of percolation on lattices and the mathematical methods involved in its study, and for the proofs of most of the results we could only point at, we refer the reader to the standard book of Grimmett ; another excellent general reference, and the only place to find some of the technical graph-theoretical details involved, is the book of Kesten . More information in the case of graphs that are not lattices can be found in the lecture notes of Peres .
For curiosity, the reader can refer to the first mention of a problem close to percolation, in the problem section of the first volume of the American Mathematical Monthly . References on more specific topics are given at the end of each section.
## See also
Introductory article: Statistical mechanics; 2D Ising model; Wulff droplets; Stochastic Loewner evolutions.
## Keywords
Percolation, random medium, porous medium, random graph, phase transition, critical exponent, shape theorem, correlation length, scaling, scaling relation, scaling limit, exploration path, Cayley graph, contact process.
Vincent BEFFARA
UMPA – ENS Lyon
46 Allée d’Italie
69364 Lyon Cedex 07
FRANCE
Vladas SIDORAVICIUS
IMPA
Estrada Dona Castorina 110
Rio de Janeiro 22460-320
BRASIL |
warning/0507/cs0507035.html | ar5iv | text | # Enhancing Global SLS-Resolution with Loop Cutting and Tabling Mechanisms
## 1 Introduction
There are two types of semantics for a logic program: a declarative semantics and a procedural semantics. The declarative semantics formally defines the meaning of a logic program by specifying an intended model among all models of the logic program, whereas the procedural semantics implements/computes the declarative semantics by providing an algorithm for evaluating queries against the logic program. Most existing procedural semantics are built upon the well-known resolution rule created by Robinson .
Prolog is the first yet the most popular logic programming language . It adopts SLDNF-resolution as its procedural semantics . One of the best-known properties of SLDNF-resolution is its linearity of derivations, i.e., its query evaluation (i.e., SLDNF-derivations) constitutes a search tree, called an SLDNF-tree, which can be implemented easily and efficiently using a simple stack-based memory structure . However, SLDNF-resolution suffers from two serious problems. First, its corresponding declarative semantics, i.e. the predicate completion semantics , is based on two truth values (either true or false) and thus incurs inconsistency for some logic programs like $`P=\{p(a)\neg p(a)\}`$ . Second, it may generate infinite loops and a large amount of redundant sub-derivations .
To overcome the first problem with SLDNF-resolution, the well-founded semantics is introduced as an alternative to the predicate completion semantics. A well-founded model accommodates three truth values: true, false and undefined, so that inconsistency is avoided by letting atoms that are recursively connected through negation undefined. Several procedural semantics have been developed as an alternative to SLDNF-resolution to compute the well-founded semantics, among the most representative of which are Global SLS-resolution and SLG-resolution .
Global SLS-resolution is a direct extension of SLDNF-resolution. It evaluates queries under the well-founded semantics by generating a search tree, called an SLS-tree, in the same way as SLDNF-resolution does except that infinite derivations are treated as failed and infinite recursions through negation as undefined. Global SLS-resolution retains the linearity property of SLDNF-resolution, but it also inherits the problem of infinite loops and redundant computations. Moreover, Global SLS-resolution handles negation as follows: A ground atom $`A`$ is false when all branches of the SLS-tree for $`A`$ are either infinite or end at a failure leaf. Infinite branches make Global SLS-resolution not effective in general .
To resolve infinite loops and redundant computations, the tabling technique is introduced . The main idea of tabling is to store intermediate answers of subgoals and then apply them to solve variants of the subgoals. With tabling no variant subgoals will be recomputed by applying the same set of clauses, so infinite loops can be avoided and redundant computations be substantially reduced. There are two typical ways to make use of tabling to compute the well-founded semantics. One is to directly enhance SLDNF-resolution or Global SLS-resolution with tabling while the other is to create a new tabling mechanism with a different derivation structure. SLG-resolution results from the second way . Due to the use of tabling, SLG-resolution gets rid of infinite loops and reduces redundant computations. However, it does not have the linearity property since its query evaluation constitutes a search forest instead of a search tree. As a result, it cannot be implemented in the same way as SLDNF-resolution using a simple stack-based memory structure .
In an attempt is made to directly enhance SLDNF-resolution with tabling to compute the well-founded semantics, which leads to a tabling mechanism, called SLT-resolution. SLT-resolution retains the linearity property, thus is referred to as a linear tabling mechanism. Due to the use of tabling, it is free of infinite loops and has fewer redundant computations than SLDNF-resolution. However, SLT-resolution has the following two major drawbacks: (1) It defines positive loops and negative loops based on the same ancestor-descendant relation, which makes loop detection and handling quite costly since a loop may go across several (subsidiary) SLT-trees. (2) It makes use of answer iteration to derive all answers of looping subgoals, but provides no answer completion criteria for pruning redundant derivations. Note that answer completion is the key to an efficient tabling mechanism.
In this paper, we develop a new procedural semantics, called SLTNF-resolution, for the well-founded semantics by enhancing Global SLS-resolution with tabling and loop cutting mechanisms. SLTNF-resolution retains the linearity property and makes use of tabling to get rid of all loops and reduce redundant computations. It defines positive and negative loops in terms of two different ancestor-descendant relations, one on subgoals within an SLS-tree and the other on SLS-trees, so that positive and negative loops can be efficiently detected and handled. It employs two effective criteria for answer completion of tabled subgoals so that redundant derivations can be pruned as early as possible. All these mechanisms are integrated into an algorithm quite like that for generating SLS-trees.
The paper is organized as follows. Section 2 reviews Global SLS-resolution. Section 3 defines ancestor-descendant relations for identifying positive and negative loops, develops an algorithm for generating SLTNF-trees, establishes criteria for determining answer completion of tabled subgoals, and proves the correctness of SLTNF-resolution. Section 4 mentions some related work, and Section 5 concludes.
## 2 Preliminaries and Global SLS-Resolution
In this section, we review some standard terminology of logic programs and recall the definition of Global SLS-Resolution. We do not repeat the definition of the well-founded model here; it can be found in and many other papers.
Variables begin with a capital letter, and predicate, function and constant symbols with a lower case letter. By a variant of a literal $`L`$ we mean a literal $`L^{}`$ that is identical to $`L`$ up to variable renaming.
###### Definition 2.1
A general logic program (logic program for short) is a finite set of clauses of the form
$`AL_1,\mathrm{},L_n`$
where $`A`$ is an atom and $`L_i`$s are literals. $`A`$ is called the head and $`L_1,\mathrm{},L_n`$ is called the body of the clause. When $`n=0`$, the “$``$” symbol is omitted. If a logic program has no clause with negative literals in its body, it is called a positive logic program.
###### Definition 2.2
A goal $`G`$ is a headless clause $`L_1,\mathrm{},L_n`$ where each $`L_i`$ is called a subgoal. A goal is also written as $`G=Q`$ where $`Q=L_1,\mathrm{},L_n`$ is called a query. A computation rule (or selection rule) is a rule for selecting one subgoal from a goal.
Let $`G_i=L_1,\mathrm{},L_j,\mathrm{},L_n`$ be a goal with $`L_j`$ a positive subgoal. Let $`C=LF_1,\mathrm{},F_m`$ be a clause such that $`L`$ and $`L_j`$ are unifiable, i.e. $`L\theta =L_j\theta `$ where $`\theta `$ is an mgu (most general unifier). The resolvent of $`G_i`$ and $`C`$ on $`L_j`$ is a goal $`G_k=(L_1,\mathrm{},L_{j1},F_1,\mathrm{},F_m,L_{j+1},\mathrm{},L_n)\theta `$. In this case, we say that the proof of $`G_i`$ is reduced to the proof of $`G_k`$.
The initial goal, $`G_0=L_1,\mathrm{},L_n`$, is called a top goal. Without loss of generality, we shall assume throughout the paper that a top goal consists only of one atom (i.e. $`n=1`$ and $`L_1`$ is a positive literal).
Trees are used to depict the search space of a top-down query evaluation procedure. For convenience, a node in such a tree is represented by $`N_i:G_i`$ where $`N_i`$ is the node name and $`G_i`$ is a goal labeling the node. Assume no two nodes have the same name, so we can refer to nodes by their names.
Let $`P`$ be a logic program and $`G_0=Q`$ a top goal. Global SLS-resolution is the process of constructing SLS-derivations from $`P\{G_0\}`$ via a computation rule $`R`$. An SLS-derivation is a partial branch beginning at the root $`N_0:G_0`$ of an SLS-tree. Every leaf of an SLS-tree is either a success leaf or a failure leaf or a flounder leaf or an undefined leaf.<sup>1</sup><sup>1</sup>1In , an undefined leaf is called a non-labeled leaf. $`Q`$ is a non-floundering query if no SLS-tree for evaluating $`Q`$ under $`R`$ contains a flounder leaf.
An SLS-tree is successful if it has a success leaf. It is failed if all of its branches are either infinite or end at a failure leaf. It is floundered if it contains a floundered leaf and is not successful. An SLS-tree is undefined if it is neither successful nor failed nor floundered.
There are two slightly different definitions of an SLS-tree: Przymusinski’s definition and Ross’ definition . Przymusinski’s definition requires a level mapping (called strata) to be associated with literals and goals, while Ross’ definition requires the computation rule to be preferential, i.e. positive subgoals are selected ahead of negative ones and negative subgoals are selected in parallel. Both of the two definitions allow infinite branches and infinite recursion through negation. The following definition of an SLS-tree is obtained by combining the two definitions.
###### Definition 2.3 (SLS-trees )
Let $`P`$ be a logic program, $`G_0`$ a top goal, and $`R`$ a computation rule. The SLS-tree $`T_{N_0:G_0}`$ for $`P\{G_0\}`$ via $`R`$ is a tree rooted at $`N_0:G_0`$ such that for any node $`N_i:G_i`$ in the tree with $`G_i=L_1,\mathrm{},L_n`$:
1. If $`n=0`$ then $`N_i`$ is a success leaf, marked by $`\mathrm{}_t`$.
2. If $`L_j`$ is a positive literal selected by $`R`$, then for each clause $`C`$ in $`P`$ whose head is unifiable with $`L_j`$, $`N_i`$ has a child $`N_k:G_k`$ where $`G_k`$ is the resolvent of $`C`$ and $`G_i`$ on $`L_j`$. If no such a clause exists in $`P`$, then $`N_i`$ is a failure leaf, marked by $`\mathrm{}_f`$.
3. Let $`L_j=\neg A`$ be a negative literal selected by $`R`$. If $`A`$ is not ground then $`N_i`$ is a flounder leaf, marked by $`\mathrm{}_{fl}`$, else let $`T_{N_{i+1}:A}`$ be an (subsidiary) SLS-tree for $`P\{A\}`$ via $`R`$. We consider four cases:
1. If $`T_{N_{i+1}:A}`$ is failed then $`N_i`$ has only one child that is labeled by the goal $`L_1,\mathrm{},L_{j1},L_{j+1},\mathrm{},L_n`$.
2. If $`T_{N_{i+1}:A}`$ is successful then $`N_i`$ is a failure leaf, marked by $`\mathrm{}_f`$.
3. If $`T_{N_{i+1}:A}`$ is floundered then $`N_i`$ is a flounder leaf, marked by $`\mathrm{}_{fl}`$.
4. Otherwise (i.e. $`T_{N_{i+1}:A}`$ is undefined), we mark $`L_j`$ in $`G_i`$ as skipped and use the computation rule $`R`$ to select a new literal $`L_k`$ from $`G_i`$ and apply the resolution steps 2 and 3 to the goal $`G_i`$. If all literals in $`G_i`$ were already marked as skipped then $`N_i`$ is an undefined leaf, marked by $`\mathrm{}_u`$.
We make two remarks. First, the level mapping/strata used in Przymusinski’s definition is implicit in Definition 2.3. That is, in case 3 the level/stratum of $`A`$ is less than the level/stratum of $`G_i`$ if and only if either case 3a or case 3b or case 3c holds. Second, the preferential restriction of Ross’ definition to the computation rule is relaxed by marking undefined subgoals as skipped and then continuing to select new subgoals from the remaining subgoals in $`G_i`$ for evaluation (see case 3d). A leaf is undefined if and only if all its subgoals are marked as skipped.
###### Definition 2.4
A successful (resp. failed or undefined) derivation for a goal $`G`$ is a branch beginning at the root of the SLS-tree for $`G`$ and ending at a success (resp. failure or undefined) leaf. A correct answer substitution for $`G`$ is the substitution $`\theta =\theta _1\mathrm{}\theta _n`$, where $`\theta _i`$s are the most general unifiers used at each step along the derivation, restricted to the variables in $`G`$.
It has been shown that Global SLS-resolution is sound and complete with respect to the well-founded semantics for non-floundering queries.
###### Theorem 2.1 ()
Let $`P`$ be a logic program, $`R`$ a computation rule, and $`G_0Q`$ be a top goal with $`Q`$ a non-floundering query under $`R`$. Let $`WF(P)`$ be the well-founded model of $`P`$.
1. $`WF(P)(Q)`$ if and only if the SLS-tree for $`P\{G_0\}`$ via $`R`$ is successful.
2. $`WF(P)(Q\theta )`$ if and only if there exists a correct answer substitution for $`G_0`$ more general than the substitution $`\theta `$.
3. $`WF(P)\neg (Q)`$ if and only if the SLS-tree for $`P\{G_0\}`$ is failed.
###### Definition 2.5
Let $`N_i:G_i`$ be a node in an SLS-tree $`T_{N_r:G_r}`$ where $`A`$ is the selected positive subgoal in $`G_i`$. The partial branches of $`T_{N_r:G_r}`$ beginning at $`N_i`$ that are used to evaluate $`A`$ constitute sub-derivations for $`A`$. All such sub-derivations form a sub-SLS-tree for $`A`$ at $`N_i`$.
By Theorem 2.1, for any correct answer substitution $`\theta `$ built from a successful sub-derivation for $`A`$, $`WF(P)(A\theta )`$.
Since Global SLS-resolution allows infinite derivations as well as infinite recursion through negation, we may need infinite time to generate an SLS-tree. This is not feasible in practice. In the next section, we resolve this problem by enhancing Global SLS-resolution with both loop cutting and tabling mechanisms.
## 3 SLTNF-Resolution
We first define an ancestor-descendant relation on selected subgoals in an SLS-tree. Informally, $`A`$ is an ancestor subgoal of $`B`$ if the proof of $`A`$ depends on (or in other words goes via) the proof of $`B`$. For example, let $`M:A,A_1,\mathrm{},A_m`$ be a node in an SLS-tree, and $`N:B_1\theta ,\mathrm{},B_n\theta ,A_1\theta ,\mathrm{},A_m\theta `$ be a child node of $`M`$ that is generated by resolving $`M`$ on the subgoal $`A`$ with a clause $`A^{}B_1,\mathrm{},B_n`$ where $`A\theta =A^{}\theta `$. Then $`A`$ at $`M`$ is an ancestor subgoal of all $`B_i\theta `$s at $`N`$. However, such relationship does not exist between $`A`$ at $`M`$ and any $`A_j\theta `$ at $`N`$. It is easily seen that all $`B_i\theta `$s at $`N`$ inherit the ancestor subgoals of $`A`$ at $`M`$, and that each $`A_j\theta `$ at $`N`$ inherits the ancestor subgoals of $`A_j`$ at $`M`$. Note that subgoals at the root of an SLS-tree have no ancestor subgoals.
Let $`N_i:G_i`$ and $`N_k:G_k`$ be two nodes and $`A`$ and $`B`$ be the selected subgoals in $`G_i`$ and $`G_k`$, respectively. When $`A`$ is an ancestor subgoal of $`B`$, we refer to $`B`$ as a descendant subgoal of $`A`$, $`N_i`$ as an ancestor node of $`N_k`$, and $`N_k`$ as a descendant node of $`N_i`$. Particularly, if $`A`$ is both an ancestor subgoal and a variant, i.e. an ancestor variant subgoal, of $`B`$, we say the derivation goes into a loop, where $`N_i`$ and $`N_k`$ are respectively called an ancestor loop node and a descendant loop node, and $`A`$ (at $`N_i`$) and $`B`$ (at $`N_k`$) are respectively called an ancestor loop subgoal and a descendant loop subgoal.
The above ancestor-descendant relation is defined over subgoals and will be applied to detect positive loops, i.e. loops within an SLS-tree. In order to handle negative loops (i.e. loops through negation like $`A\neg B`$ and $`B\neg A`$) which occur across SLS-trees, we define an ancestor-descendant relation on SLS-trees. Let $`N_i:\neg A,\mathrm{}`$ be a node in $`T_{N_r:G_r}`$, with $`\neg A`$ the selected subgoal, and let $`T_{N_{i+1}:A}`$ be an (subsidiary) SLS-tree for $`P\{A\}`$ via $`R`$. $`T_{N_r:G_r}`$ is called an ancestor SLS-tree of $`T_{N_{i+1}:A}`$, while $`T_{N_{i+1}:A}`$ is called a descendant SLS-tree of $`T_{N_r:G_r}`$. Of course, the ancestor-descendant relation is transitive.
A negative loop occurs if an SLS-tree has a descendant SLS-tree, with the same goal at their roots. For convenience, we use dotted edges to connect parent and child SLS-trees, so that negative loops can be clearly identified. Let $`G_0`$ be a top goal. We call $`T_{N_0:G_0}`$ together with all of its descendant SLS-trees a generalized SLS-tree, denoted $`GT_{P,G_0}`$ (or simply $`GT_{G_0}`$ when no confusion would arise). Therefore, a branch of a generalized SLS-tree may come across several SLS-trees through dotted edges. A generalized SLS-derivation is a partial branch beginning at the root of a generalized SLS-tree.
Assume that all loops are detected and cut based on the ancestor-descendant relations. This helps Global SLS-resolution get rid of infinite derivations and infinite recursion through negation. However, applying such loop cutting mechanism alone is not effective since some answers would be lost. In order to guarantee the completeness of Global SLS-resolution with the loop cutting mechanism, we introduce a tabling mechanism into SLS-derivations, leading to a tabulated SLS-resolution.
In tabulated resolutions, the set of predicate symbols in a logic program is partitioned into two groups: tabled predicate symbols and non-tabled predicate symbols. Subgoals with tabled predicate symbols are then called tabled subgoals. A dependency graph is used to make such classification. Informally, for any predicate symbols $`p`$ and $`q`$, there is an edge $`pq`$ in the dependency graph $`G_P`$ of a logic program $`P`$ if and only if $`P`$ contains a clause whose head contains $`p`$ and whose body contains $`q`$. $`p`$ is a tabled predicate symbol if $`G_P`$ contains a cycle involving $`p`$. It is trivial to show that subgoals involved in any loops in SLS-trees must be tabled subgoals.
Intermediate answers of tabled subgoals will be stored in tables once they are produced at some derivation stages. Such answers are called tabled answers. For convenience of presentation, we organize a table into a compound structure like $`struct`$ in pseudo $`C^{++}`$ language. That is, the table of an atom $`A`$, denoted $`TB_A`$, is internally an instance of the data type TABLE defined as follows:
| | typ | edef struct { |
| --- | --- | --- |
| | | string | $`atom`$; //for $`TB_A`$, $`atom=A`$. |
| | | int | $`comp`$; //status of $`atom`$ indicating if all answers have been tabled. |
| | | set | $`ans`$; //tabled answers of $`atom`$. |
| | } TABLE; |
Answers of a tabled subgoal $`A`$ are stored in $`TB_Aans`$. We say $`TB_A`$ is complete if $`TB_Aans`$ contains all answers of $`A`$. We use $`TB_Acomp=1`$ to mark the completeness of tabled answers. Clearly, the case $`TB_Acomp=1`$ and $`TB_Aans=\mathrm{}`$ indicates that $`A`$ is false.
We introduce a special subgoal, $`u^{}`$, which is assumed to occur neither in logic programs nor in top goals. $`u^{}`$ will be used to substitute for some ground negative subgoals whose truth values are temporarily undefined (i.e., whether they are true or false cannot be determined at the current stage of derivation). We assume such a special subgoal will not be selected by a computation rule.
We also use a special subgoal, $`LOOP`$, to mark occurrence of a loop.
Augmenting SLS-trees with the loop cutting and tabling mechanisms leads to the following definition of SLTNF-trees. Here “SLTNF” stands for “Linear Tabulated resolution using a Selection/computation rule with Negation as Finite Failure.”
###### Definition 3.1 (SLTNF-trees)
Let $`P`$ be a logic program, $`G_0`$ a top goal, and $`R`$ a computation rule. Let $`𝒯_𝒫`$ be a set of tables each of which contains a finite set of tabled answers. The SLTNF-tree $`T_{N_0:G_0}`$ for $`(P\{G_0\},𝒯_𝒫)`$ via $`R`$ is a tree rooted at $`N_0:G_0`$ such that for any node $`N_i:G_i`$ in the tree with $`G_i=L_1,\mathrm{},L_n`$:
1. If $`n=0`$ then $`N_i`$ is a success leaf, marked by $`\mathrm{}_t`$, else if $`L_1=u^{}`$ then $`N_i`$ is a temporarily undefined leaf, marked by $`\mathrm{}_u^{}`$, else if $`L_1=LOOP`$ then $`N_i`$ is a loop leaf, marked by $`\mathrm{}_{loop}`$.
2. If $`L_j=p(.)`$ is a positive literal selected by $`R`$, we consider two cases:
1. If $`TB_{L_j}𝒯_𝒫`$ with $`TB_{L_j}comp=1`$, then for each tabled answer $`A`$ in $`TB_{L_j}ans`$, $`N_i`$ has a child node $`N_k:G_k`$ where $`G_k`$ is the resolvent of $`A`$ and $`G_i`$ on $`L_j`$. In case that $`TB_{L_j}ans=\mathrm{}`$, $`N_i`$ is a failure leaf, marked by $`\mathrm{}_f`$.
2. Otherwise, for each tabled answer $`A`$ in $`TB_{L_j}ans`$ $`N_i`$ has a child node $`N_k:G_k`$ where $`G_k`$ is the resolvent of $`A`$ and $`G_i`$ on $`L_j`$, and
1. If $`N_i`$ is a descendant loop node then it has a child node $`N_l:LOOP`$.
2. Otherwise, for each clause $`C`$ in $`P`$ whose head is unifiable with $`L_j`$ $`N_i`$ has a child node $`N_l:G_l`$ where $`G_l`$ is the resolvent of $`C`$ and $`G_i`$ on $`L_j`$. If there are neither tabled answers nor clauses applicable to $`N_i`$ then $`N_i`$ is a failure leaf, marked by $`\mathrm{}_f`$.
3. Let $`L_j=\neg A`$ be a negative literal selected by $`R`$. If $`A`$ is not ground then $`N_i`$ is a flounder leaf, marked by $`\mathrm{}_{fl}`$, else we consider the following cases:
1. If $`TB_A𝒯_𝒫`$ with $`TB_Acomp=1`$ and $`TB_Aans=\mathrm{}`$, then $`N_i`$ has only one child node $`N_k:G_k`$ with $`G_k=L_1,\mathrm{},L_{j1},L_{j+1},\mathrm{},L_n`$.
2. If $`TB_A𝒯_𝒫`$ with $`TB_Acomp=1`$ and $`TB_Aans=\{A\}`$, then $`N_i`$ is a failure leaf, marked by $`\mathrm{}_f`$.
3. Otherwise, if the current SLTNF-tree or one of its ancestor SLTNF-trees is with a goal $`A`$ at the root, $`N_i`$ has only one child node $`N_k:G_k`$ where if $`L_nu^{}`$ then $`G_k=L_1,\mathrm{},L_{j1},L_{j+1},\mathrm{},`$ $`L_n,u^{}`$ else $`G_k=L_1,\mathrm{},L_{j1},L_{j+1},\mathrm{},L_n`$.
4. Otherwise, let $`T_{N_r:A}`$ be an (subsidiary) SLTNF-tree for $`(P\{A\},𝒯_𝒫)`$ via $`R`$. We have the following cases:
1. If $`T_{N_r:A}`$ has a success leaf then $`N_i`$ is a failure leaf, marked by $`\mathrm{}_f`$.
2. If $`T_{N_r:A}`$ has no success leaf but a flounder leaf then $`N_i`$ is a flounder leaf, marked by $`\mathrm{}_{fl}`$.
3. (Negation As Finite Failure) If all branches of $`T_{N_r:A}`$ end at either a failure or a loop leaf where for each loop leaf generated from a descendant loop subgoal $`V`$, no successful sub-derivation for its ancestor loop subgoal has a correct answer substitution $`\theta `$ such that $`V\theta `$ is not in $`𝒯_𝒫`$, then $`N_i`$ has only one child node $`N_k:G_k`$ with $`G_k=L_1,\mathrm{},L_{j1},L_{j+1},\mathrm{},L_n`$.
4. Otherwise, $`N_i`$ has only one child node $`N_k:G_k`$ where if $`L_nu^{}`$ $`G_k=L_1,\mathrm{},L_{j1},L_{j+1},\mathrm{},`$ $`L_n,u^{}`$ else $`G_k=L_1,\mathrm{},L_{j1},L_{j+1},\mathrm{},L_n`$.
Note that some commonly used concepts, such as derivations (for goals), sub-derivations (for subgoals), sub-trees (for subgoals), generalized trees, and correct answer substitutions, have the same meanings as in SLS-trees (see Section 2).
Positive loops are broken simply by disallowing descendant loop nodes to apply clauses in $`P`$ for expansion (see case 2b), while negative loops are broken by substituting $`u^{}`$ for looping negative subgoals (see case 3c). This guarantees that SLTNF-trees are finite for logic programs with the bounded-term-size property (see Definition 3.2 and Theorem 3.1).
Note that $`u^{}`$ is only introduced to signify existence of subgoals whose truth values are temporarily non-determined because of occurrence of positive or negative loops. So keeping only one $`u^{}`$ in a goal is enough for such a purpose. From case 1 of Definition 3.1 we see that goals with $`u^{}`$ cannot lead to a success leaf. However, $`u^{}`$ may well appear in a failure leaf since one of the other subgoals may fail regardless of what truth values the temporarily undefined subgoals would take. This achieves the effect of what a preferential computation rule is supposed to achieve, although our computation rule is not necessarily preferential.
Observe that SLTNF-trees implement an Negation As Finite Failure (NAF) rule (see case 3(d)iii): A ground subgoal $`\neg A`$ fails if $`A`$ succeeds, and succeeds if $`A`$ finitely fails after exhausting all answers of the loop subgoals involved in evaluating $`A`$. This NAF rule is the same as that used in SLDNF-resolution except that loop leaves are considered.
The following example illustrates the process of constructing SLTNF-trees.
###### Example 3.1
Consider the following program and let $`G_0=p(a,Y)`$ be the top goal.
| $`P_1`$: | $`p(X,Y)p(X,Z),e(Z,Y).`$ $`C_{p_1}`$ |
| --- | --- |
| | $`p(X,Y)e(X,Y),\neg r.`$ $`C_{p_2}`$ |
| | $`e(a,b).`$ $`C_{e_1}`$ |
| | $`e(b,c)`$ $`C_{e_2}`$ |
| | $`rs,r.`$ $`C_r`$ |
| | $`s\neg s.`$ $`C_s`$ |
Let $`𝒯_{𝒫}^{}{}_{1}{}^{}=\mathrm{}`$, and for convenience, let us choose the widely-used left-most computation rule (i.e. we always select the left-most subgoal from a goal). The generalized SLTNF-tree $`GT_{p(a,Y)}`$ for $`(P_1\{p(a,Y)\},\mathrm{})`$ is shown in Figure 1,<sup>2</sup><sup>2</sup>2 For simplicity, in depicting SLTNF-trees we omit the “$``$” symbol in goals. which consists of three finite SLTNF-trees that are rooted at $`N_0`$, $`N_5`$ and $`N_8`$, respectively. Note that two positive loops are cut at $`N_1`$ and $`N_{11}`$, respectively, and one negative loop is cut at $`N_9`$.
$`T_{N_5:r}`$ has only one branch, which ends at a loop leaf $`N_{12}`$. There is no successful sub-derivation for the ancestor loop subgoal $`r`$ at $`N_5`$, so the NAF rule is applicable. Thus, $`\neg r`$ at $`N_4`$ succeeds, leading to a successful sub-derivation for $`p(a,Y)`$ at $`N_0`$ with a correct answer substitution $`\{Y/b\}`$.
###### Definition 3.2 ()
A logic program has the bounded-term-size property if there is a function $`f(n)`$ such that whenever a top goal $`G_0`$ has no argument whose term size exceeds $`n`$, then no subgoals and tabled answers in any generalized SLTNF-tree $`GT_{G_0}`$ have an argument whose term size exceeds $`f(n)`$.
The following result shows that the construction of SLTNF-trees is always terminating for logic programs with the bounded-term-size property.
###### Theorem 3.1
Let $`P`$ be a logic program with the bounded-term-size property, $`G_0`$ a top goal and $`R`$ a computation rule. The generalized SLTNF-tree $`GT_{G_0}`$ for $`(P\{G_0\},𝒯_𝒫)`$ via $`R`$ is finite.
Proof: First note that $`GT_{G_0}`$ contains no negative loops (see case 3c). The bounded-term-size property guarantees that no term occurring on any path of $`GT_{G_0}`$ can have size greater than $`f(n)`$, where $`n`$ is a bound on the size of terms in the top goal $`G_0`$. Assume, on the contrary, that $`GT_{G_0}`$ is infinite. Since the branching factor of $`GT_{G_0}`$ (i.e. the average number of children of all nodes in the tree) is bounded by the finite number of clauses in $`P`$, $`GT_{G_0}`$ either contains an infinite number of SLTNF-trees or has an infinite derivation within some SLTNF-tree. Note that $`P`$ has only a finite number of predicate, function and constant symbols. If $`GT_{G_0}`$ contains an infinite number of SLTNF-trees, there must exist negative loops in $`GT_{G_0}`$, a contradiction. If $`GT_{G_0}`$ has an infinite derivation within some SLTNF-tree, some positive subgoal $`A_0`$ selected by $`R`$ must have infinitely many variant descendants $`A_1,A_2,\mathrm{},A_i,\mathrm{}`$ on the path such that the proof of $`A_0`$ needs the proof of $`A_1`$ that needs the proof of $`A_2`$, and so on. That is, $`A_i`$ is an ancestor loop subgoal of $`A_j`$ for any $`0i<j`$. This contradicts the fact that any descendant loop subgoal in $`GT_{G_0}`$ has only one ancestor loop subgoal because a descendant loop subgoal cannot generate descendant loop subgoals since no clauses will be applied to it for expansion (see case 2b of Definition 3.1). $`\mathrm{}`$
Consider Figure 1 again. Observe that if we continued expanding $`N_1`$ (like Global SLS-resolution) by applying $`C_{p_1}`$ and $`C_{p_2}`$, we would generate another correct answer substitution $`\{Y/c\}`$ for $`G_0`$. This indicates that applying loop cutting alone would result in incompleteness.
We use answer iteration to derive all answers of loop subgoals. Here is the basic idea: We first build a generalized SLTNF-tree for $`(P\{G_0\},𝒯_{𝒫}^{}{}_{}{}^{0})`$ with $`𝒯_{𝒫}^{}{}_{}{}^{0}=\mathrm{}`$ while collecting all new tabled answers (for all tabled subgoals) into $`NEW^0`$. Then we build a new generalized SLTNF-tree for $`(P\{G_0\},𝒯_{𝒫}^{}{}_{}{}^{1})`$ with $`𝒯_{𝒫}^{}{}_{}{}^{1}=𝒯_{𝒫}^{}{}_{}{}^{0}NEW^0`$ while collecting all new tabled answers into $`NEW^1`$. Such an iterative process continues until no new tabled answers are available.
The key issue with answer iteration is answer completion, i.e, how to determine if the table of a subgoal is complete at some derivation stages. Careful reader may have noticed that we have already used a completion criterion for ground subgoals in defining the NAF rule (see case 3d of Definition 3.1). We now generalize this criterion to all subgoals.
###### Theorem 3.2
Let $`GT_{G_0}`$ be the generalized SLTNF-tree for $`(P\{G_0\},𝒯_𝒫)`$ and $`NEW`$ contain all new tabled answers in $`GT_{G_0}`$. The following completion criteria hold.
1. For a ground tabled positive subgoal $`A`$, $`TB_A𝒯_𝒫NEW`$ is complete for $`A`$ if $`TB_Aans=\{A\}`$.
2. For any tabled positive subgoal $`A`$, $`TB_A𝒯_𝒫NEW`$ is complete for $`A`$ if there is a node $`N_i:G_i`$ in $`GT_{G_0}`$, where $`A`$ is the selected subgoal in $`G_i`$ and let $`T_A`$ be the sub-SLTNF-tree for $`A`$ at $`N_i`$, such that (1) $`T_A`$ has no temporarily undefined leaf, and (2) for each loop leaf in $`T_A`$, the sub-SLTNF-tree for its ancestor loop subgoal $`V`$ has neither temporarily undefined leaf nor success leaf with a correct answer substitution $`\theta `$ such that $`V\theta `$ is not in $`𝒯_𝒫`$.
Proof: The first criterion is straightforward since $`A`$ is ground. We now prove the second. Note that there are only two cases in which a tabled subgoal $`A`$ may get new answers via iteration. The first is due to that some temporarily undefined subgoals in the current round would become successful or failed in the future rounds of iteration. This case is excluded by conditions (1) and (2). The second case is due to that some loop subgoals in $`T_A`$ in the current round would produce new answers in the future rounds of iteration. Such new answers are generated in an iterative way, i.e., in the current round descendant loop subgoals in $`T_A`$ consume only existing tabled answers in $`𝒯_𝒫`$ and help generate new answers (which are not in $`𝒯_𝒫`$) for their ancestor loop subgoals. These new answers are then tabled (in $`NEW`$) for the descendant loop subgoals to consume in the next round. In this case, $`T_A`$ must contain at least one descendant loop subgoal $`V^{}`$ such that the sub-SLTNF-tree for its ancestor loop subgoal $`V`$ has a success leaf with a new correct answer substitution not included in $`𝒯_𝒫`$ (this new answer is not consumed by $`V^{}`$ in the current round but will be consumed in the next round). Obviously, this case is excluded by condition (2). As a result, conditions (1) and (2) together imply that further iteration would generate no new answers for $`A`$. Therefore, $`TB_A`$ is complete for $`A`$ after merging $`𝒯_𝒫`$ with the new tabled answers $`NEW`$ in $`GT_{G_0}`$. $`\mathrm{}`$
###### Example 3.2
Consider Figure 1. We cannot apply Theorem 3.2 to determine the completeness of $`TB_{p(a,Y)}`$ since the ancestor loop subgoal $`p(a,Y)`$ at $`N_0`$ has a successful sub-derivation with an answer $`p(a,b)`$ not in $`𝒯_{𝒫}^{}{}_{1}{}^{}`$. As we can see, applying this new answer to the descendant loop subgoal at $`N_1`$ would generate another new answer $`p(a,c)`$. The completeness of $`TB_s`$ is not determinable either, since both the two sub-SLTNF-trees for $`s`$ (rooted at $`N_6`$ and $`N_8`$, respectively) contain a temporarily undefined leaf. However, by Theorem 3.2, $`TB_r`$ is complete.
###### Definition 3.3 (SLTNF-resolution)
Let $`P`$ be a logic program, $`G_0=A`$ a top goal with $`A`$ an atom, and $`R`$ a computation rule. Let $`𝒯_{𝒫}^{}{}_{}{}^{0}=\mathrm{}`$. SLTNF-resolution evaluates $`G_0`$ by calling the function $`SLTNF(P,G_0,R,𝒯_{𝒫}^{}{}_{}{}^{0})`$, defined as follows.
| function $`SLTNF(P,G_0,R,𝒯_{𝒫}^{}{}_{}{}^{i})`$ returns a table $`TB_A`$ |
| --- |
| $`\{`$ |
| | Build | a generalized SLTNF-tree $`GT_{G_0}^i`$ for $`(P\{G_0\},𝒯_{𝒫}^{}{}_{}{}^{i})`$ while collecting |
| | | all new tabled answers into $`NEW^i`$; |
| | $`𝒯_{𝒫}^{}{}_{}{}^{i+1}=𝒯_{𝒫}^{}{}_{}{}^{i}NEW^i`$; |
| | Check completeness of all tables in $`𝒯_{𝒫}^{}{}_{}{}^{i+1}`$ and update their status; |
| | if $`NEW^i=\mathrm{}`$ or $`TB_Acomp=1`$ then return $`TB_A`$; |
| | return $`SLTNF(P,G_0,R,𝒯_{𝒫}^{}{}_{}{}^{i+1})`$; |
| $`\}`$ |
###### Example 3.3 (Cont. of Example 3.1)
First execute $`SLTNF(P_1,G_0,R,𝒯_{𝒫}^{}{}_{1}{}^{0})`$ where $`𝒯_{𝒫}^{}{}_{1}{}^{0}=\mathrm{}`$, $`G_0=p(a,Y)`$ and $`R`$ is the left-most computation rule. The procedure builds a generalized SLTNF-tree for $`(P_1\{p(a,Y)\},\mathrm{})`$ as shown in Figure 1. It also collects the following new tabled answer into $`NEW^0`$: $`p(a,b)`$ for $`TB_{p(a,Y)}`$. Moreover, it has $`TB_r`$ completed by setting $`TB_rcomp`$ to 1 (note that $`TB_rans=\mathrm{}`$).
Next execute $`SLTNF(P_1,G_0,R,𝒯_{𝒫}^{}{}_{1}{}^{1})`$ where $`𝒯_{𝒫}^{}{}_{1}{}^{1}=𝒯_{𝒫}^{}{}_{1}{}^{0}NEW^0`$. It builds a generalized SLTNF-tree $`GT_{p(a,Y)}^1`$ for $`(P_1\{p(a,Y)\},𝒯_{𝒫}^{}{}_{1}{}^{1})`$ as shown in Figure 2, and collects the following new tabled answer into $`NEW^1`$: $`p(a,c)`$ for $`TB_{p(a,Y)}`$.
Finally execute $`SLTNF(P_1,G_0,R,𝒯_{𝒫}^{}{}_{1}{}^{2})`$ where $`𝒯_{𝒫}^{}{}_{1}{}^{2}=𝒯_{𝒫}^{}{}_{1}{}^{1}NEW^1`$. The procedure builds a generalized SLTNF-tree $`GT_{p(a,Y)}^2`$ for $`(P_1\{p(a,Y)\},𝒯_{𝒫}^{}{}_{1}{}^{2})`$ in which no new tabled answer is produced. Therefore, it returns with two tabled answers, $`p(a,b)`$ and $`p(a,c)`$, to the top goal $`G_0`$.
###### Theorem 3.3
Let $`P`$ be a logic program with the bounded-term-size property, $`G_0`$ a top goal and $`R`$ a computation rule. $`SLTNF(P,G_0,R,\mathrm{})`$ terminates in finite time.
Proof: Let $`n`$ be the maximum size of arguments in any top goal. Since $`P`$ has the bounded-term-size property, neither subgoals nor tabled answers have arguments whose size exceeds $`f(n)`$ for some function $`f`$. Let $`s=f(n)`$. Since $`P`$ has a finite number of predicate symbols, the number of distinct subgoals (up to variable renaming) occurring in all $`GT_{G_0}^i`$s is bounded by a finite number $`N(s)`$. Therefore, SLTNF-resolution performs at most $`N(s)`$ iterations (i.e. generates at most $`N(s)`$ generalized SLTNF-trees). By Theorem 3.1, each iteration terminates in finite time, hence SLTNF-resolution terminates in finite time. $`\mathrm{}`$
###### Theorem 3.4
Let $`P`$ be a logic program with the bounded-term-size property, $`A`$ an atom, and $`G_0=A`$ a top goal with $`A`$ a non-floundering query. Let $`TB_A`$ be the tabled answers returned from $`SLTNF(P,G_0,R,\mathrm{})`$, and let $`T_{N_0:G_0}`$ be the SLS-tree for $`P\{G_0\}`$ via $`R`$.
1. $`A\theta `$ is in $`TB_A`$ if and only if there is a correct answer substitution $`\theta `$ for $`G_0`$ in $`T_{N_0:G_0}`$.
2. $`TB_Acomp=1`$ and $`TB_Aans=\mathrm{}`$ if and only if $`T_{N_0:G_0}`$ is failed.
Proof: We first prove that SLS-trees with negative loops can be transformed into equivalent SLS-trees without negative loops. Let $`T_{N_i:B}`$ be an SLS-tree with a descendant SLS-tree $`T_{N_j:B}`$. Obviously, this is a negative loop. Observe that $`B`$ at $`N_i`$ being successful or failed must be independent of the loop SLS-tree $`T_{N_j:B}`$, for otherwise the truth value of $`B`$ would depend on $`\neg B`$ so that $`B`$ is undefined. This strongly suggests that using a temporarily undefined value $`u^{}`$ as the truth value of $`T_{N_j:B}`$ does not change the answer of $`B`$ at $`N_i`$. In other words, any SLS-trees with negative loops can be transformed into equivalent SLS-trees where all descendant loop SLS-trees are assumed to return a temporarily undefined value $`u^{}`$.
Let $`T_{N_0:G_0}^i`$ and $`GT_{G_0}^i`$ be respectively the SLTNF-tree and the generalized SLTNF-tree for $`(P\{G_0\},𝒯_{𝒫}^{}{}_{}{}^{i})`$, where $`𝒯_{𝒫}^{}{}_{}{}^{0}=\mathrm{}`$ and for each $`i0`$, $`𝒯_{𝒫}^{}{}_{}{}^{i+1}=𝒯_{𝒫}^{}{}_{}{}^{i}NEW^i`$ where $`NEW^i`$ contains all new tabled answers collected from $`GT_{G_0}^i`$. We prove this theorem by showing that answers over SLS-derivations can be extracted in an iterative way and such iterations are the same as those of SLTNF-resolution. Therefore, both resolutions extract the same set of answers to $`G_0`$. We distinguish between three cases:
1. For any answer $`A\theta `$ that is generated without going through any loops, we must have the same successful derivations for $`A`$ in $`T_{N_0:G_0}^0`$ as in $`T_{N_0:G_0}`$.
2. Let us consider answers to $`G_0`$ that are generated without going through any negative loops. Without loss of generality, assume the SLS-derivations for the answers involve positive loops as shown in Figure 3, where for any $`j>k0`$, $`B^k`$ is an ancestor loop subgoal of $`B^j`$ and each $`T^k`$ together with the branch leading to $`N_{i^{k+1}}`$ is a sub-SLS-tree for $`B^k`$ at $`N_{i^k}`$. Obviously, all $`T^k`$s are identical up to variable renaming and thus they have the same set $`S_{B^0}`$ of correct answer substitutions for $`B^k`$ (up to variable renaming).
Observe that besides $`S_{B^0}`$, the other possible correct answer substitutions for $`B^k`$ must be generated via the infinite loops in an iterative way: For any $`l>0`$, the correct answer substitutions for $`B^l,E_1^l,\mathrm{},E_n^l`$ at $`N_{i^l}`$ combined with $`\delta ^l`$, when restricted to the variables in $`B^{l1}`$, are also correct answer substitutions for $`B^{l1}`$ at $`N_{i^{l1}}`$. These substitutions are obtained by applying each correct answer substitution $`\theta ^l`$ for $`B^l`$ to $`E_1^l,\mathrm{},E_n^l`$ and then evaluating $`(E_1^l,\mathrm{},E_n^l)\theta ^l`$. Since $`P`$ has the bounded-term-size property, no correct answer substitution requires performing an infinite number of such iterations. That is, there must exist a depth bound $`d`$ such that any correct answer substitution $`\theta `$ for $`B^0`$ is in the following closure (fixpoint):
* The initial set of correct answer substitutions is $`S_d=S_{B^0}`$.
* For each $`0<ld`$, the set of correct answer substitutions for $`B^{l1}`$ at $`N_{i^{l1}}`$ is $`S_{l1}=S_l\{\theta |\theta ^lS_l`$ and $`\theta =\delta ^l\theta ^l\alpha `$ where $`\alpha `$ is a correct answer substitution for $`(E_1^l,\mathrm{},E_n^l)\theta ^l\}`$.
Apparently, SLTNF-resolution performs the same iterations by making use of the loop cutting and tabling mechanisms: In the beginning, $`TB_{B^0}`$ is empty. The loop is cut at $`N_{i^1}`$, so $`TB_{B^0}=S_d=S_{B^0}`$ after $`T_{N_0:G_0}^0`$ is generated (note $`B^0`$ and $`B^k`$ (resp., $`T^0`$ and $`T^k`$) are variants). Then for the $`l`$-th iteration $`(0<ld)`$ $`TB_{B^0}`$ obtains new answers by applying the already tabled answers to $`B^1`$ at $`N_{i^1}`$ in $`T_{N_0:G_0}^l`$; i.e., $`TB_{B^0}=S_{l1}`$. As a result, SLS-resolution and SLTNF-resolution derive the same set of correct answer substitutions for all subgoals involving no negative loops.
3. Let us now consider answers to $`G_0`$ that are generated involving negative loops. As we discussed earlier, loop descendant SLS-trees $`T_{N_i:B}`$ can be removed by assuming they return a temporarily undefined value $`u^{}`$. Then we get equivalent SLS-trees without any negative loops. By point 2 above, we can exhaust all answers to $`G_0`$ from these (negative loop free) SLS-trees in an iterative way, as SLTNF-resolution does. If no single answer to $`A`$ in $`G_0`$ is generated after the iteration, we have two cases. The first is that no SLS-derivation for $`A`$ at $`N_0`$ ends at a leaf with $`u^{}`$. This means that the truth value of $`A`$ does not depend on any negative loop subgoal, so $`T_{N_0:G_0}`$ is failed and thus $`TB_Acomp=1`$ and $`TB_Aans=\mathrm{}`$. The second case is that some SLS-derivation for $`A`$ at $`N_0`$ ends at a leaf with $`u^{}`$. This means that the truth value of $`A`$ recursively depends on some negative loop subgoal, so $`A`$ is undefined. In this case, SLTNF-resolution stops with $`TB_Acomp=0`$ and $`TB_Aans=\mathrm{}`$. $`\mathrm{}`$
Since Global SLS-resolution is sound and complete w.r.t. the well-founded semantics (see Theorem 2.1), we have the following immediate corollary.
###### Corollary 3.5
Let $`P`$ be a logic program, $`R`$ a computation rule, and $`G_0Q`$ be a top goal with $`Q`$ a non-floundering query under $`R`$. SLTNF-resolution is sound and complete w.r.t. the well-founded semantics.
## 4 Related Work
Existing procedural semantics for the well-founded model can be divided into two groups in terms of the way they make derivations: (1) bottom-up approaches, such as the alternating fixpoint approach , the magic sets approach and the transformation-based bottom-up approach , and (2) top-down approaches. Our method belongs to the second group. Existing top-down methods can be further divided into two groups: (1) non-tabling methods, such as Global SLS-resolution, and (2) tabling methods. Our method is one with tabling. Several tabling methods for positive logic programs have been proposed, such as OLDT-resolution , TP-resolution and the DRA tabling mechanism . However, to the best of our knowledge, only SLG-resolution and SLT-resolution use tabling to compute the well-founded semantics for general logic programs.
SLG-resolution is the state-of-the-art tabling mechanism. It is based on program transformations, instead of on standard tree-based formulations like SLDNF- or Global SLS-resolution. Starting from the predicates of the top goal, it transforms (instantiates) a set of clauses, called a system, into another system based on six basic transformation rules. Such a system corresponds to a forest of trees with each tree rooted at a tabled subgoal. A special class of literals, called delaying literals, is used to represent and handle temporarily undefined negative literals. Negative loops are identified by maintaining an additional dependency graph of subgoals . In contrast, SLTNF-resolution generates an SLTNF-tree for the top goal in which the flow of the query evaluation is naturally depicted by the ordered expansions of tree nodes. Such a tree-style formulation is quite easy for users to understand and keep track of the computation. It can also be implemented efficiently using a simple stack-based memory structure. The disadvantage of SLTNF-resolution is that it is a little more costly in time than SLG-resolution due to the use of answer iteration in exchange for the linearity of derivations.
SLT-resolution is a tabling mechanism with the linearity property. Like SLTNF-resolution, it expands tree nodes by first applying tabled answers and then applying clauses. It also uses answer iteration to derive missing answers caused by loop cuttings. However, it is different from SLTNF-resolution both in loop handling and in answer completion (note that loop handling and answer completion are two key components of a tabling system).
Recall that SLT-resolution defines positive and negative loops based on the same ancestor-descendant relation: Let $`A`$ be a selected positive subgoal and $`B`$ be a subgoal produced by applying a clause to $`A`$, then $`B`$ is a descendant subgoal of $`A`$ and inherits all ancester subgoals of $`A`$; let $`\neg A`$ be a selected ground subgoal with $`T_{N_r:A}`$ being its subsidiary SLT-tree, then the subgoal $`A`$ at $`N_r`$ inherits all ancester subgoals of $`\neg A`$. A (positive or negative) loop occurs when a selected subgoal has an ancestor loop subgoal. Observe that the ancestor and descendant subgoals may be in different SLT-trees.
When a positive loop occurs, SLTNF-resolution will apply no clauses to the descendant loop subgoal for node expansion, which guarantees that any ancestor loop subgoal has just one descendant loop subgoal. However, SLT-resolution will continue expanding the descendant loop subgoal by applying those clauses that have not yet been applied by any of its ancestor loop subgoals. As an illustration, in Figure 1, SLT-resolution will apply $`C_{p_2}`$ to expand $`N_1`$, leading to a child node $`N_1^{}`$ with a goal $`e(a,Z),\neg r,e(Z,Y)`$. Observe that if the subgoal $`e(a,Z)`$ at $`N_1^{}`$ were $`p(a,Z)`$, another loop would occur between $`N_0`$ and $`N_1^{}`$. This suggests that in SLT-resolution, an ancestor loop subgoal may have several descendant loop subgoals. Due to this, SLT-resolution is more complicated and costly than SLTNF-resolution in handling positive loops.
SLT-resolution is also more costly than SLTNF-resolution in handling negative loops. It checks negative loops in the same way as positive loops by comparing a selected subgoal with all of its ancester subgoals across all of its ancestor SLT-trees. However, in SLTNF-resolution a negative loop is checked simply by comparing a selected ground negative subgoal with the root goals of its ancestor SLTNF-trees. Recall that a negative loop occurs if a negative ground subgoal $`\neg A`$ is selected such that the root of the current SLTNF-tree or one of its ancestor SLTNF-trees is with a goal $`A`$.
SLT-resolution provides no mechanism for answer completion except that when a generalized SLT-tree $`GT_{G_0}^i`$ is generated which contains no new tabled answers, it evaluates each negative ground subgoal $`\neg A`$ in $`GT_{G_0}^i`$ in a way such that (1) $`\neg A`$ fails if $`A`$ is a tabled answer, and (2) $`\neg A`$ succeeds if (i) all branches of its subsidiary SLT-tree $`T_{N_r:A}`$ end with a failure leaf and (ii) for each loop subgoal in $`T_{N_r:A}`$, all branches of the sub-SLT-trees for its ancestor loop subgoals end with a failure leaf. Not only is this process complicated, it is also quite inefficient since the evaluation of $`\neg A`$ may involve several ancestor SLT-trees. In contrast, SLTNF-resolution provides two criteria for completing answers of both negative and positive subgoals. On the one hand, the criteria are applied during the construction of generalized SLT-trees so that redundant derivations can be pruned as early as possible. On the other hand, checking the completion of a subgoal involves only one SLTNF-tree.
## 5 Conclusions and Further Work
Global SLS-resolution and SLG-resolution represent two typical styles in top-down computing the well-founded semantics; the former emphasizes the linearity of derivations as SLDNF-resolution does while the latter focuses on making full use of tabling to resolve loops and redundant computations. SLTNF-resolution obtains the advantages of the two methods by enhancing Global SLS-resolution with loop cutting and tabling mechanisms. It seems that the existing linear tabling mechanism SLT-resolution has similar advantages, but SLTNF-resolution is simpler and more efficient due to its distinct mechanisms for loop handling and answer completion.
Due to its SLDNF-tree like structure, SLTNF-resolution can be implemented over a Prolog abstract machine such as WAM or ATOAM . In particular, it can be implemented over existing linear tabling systems for positive logic programs such as , simply by adding two more mechanisms, one for identifying negative loops and the other for checking answer completion of tabled subgoals. We are currently working on the implementation. Experimental analysis of SLTNF-resolution will then be reported in the near future.
## Acknowledgment
We thank the anonymous referees for their helpful comments. Yi-Dong Shen is supported in part by Chinese National Natural Science Foundation and Trans-Century Training Programme Foundation for the Talents by the Chinese Ministry of Education. |
warning/0507/hep-ph0507194.html | ar5iv | text | # 1 Introduction
## 1 Introduction
While the existence of dark matter on galactic and cosmological scales has been firmly established, its microscopic nature is still unknown. According to the “WIMP hypothesis”, dark matter consists of stable, weakly interacting massive particles (WIMPs) with masses roughly within the 10 GeV – 10 TeV range. From the theoretical point of view, this hypothesis is perhaps the most attractive among the proposed candidate theories. There is as yet no direct evidence for its validity; however, it does predict several potentially observable new phenomena. In particular, pairs of WIMPs accumulated in the Milky Way and other galaxies should occasionally annihilate into lighter particles. These lighter particles (or their decay products) can then be found in cosmic rays, providing an “indirect” signature of galactic WIMPs.
The same process, pair annihilation of WIMPs into lighter particles, is also responsible for maintaining the thermal equilibrium between the WIMPs and the rest of the cosmic fluid in the early universe. As a result, the temperature at which the WIMPs decouple depends sensitively on the pair annihilation cross section. This implies that a measurement of the present dark matter density (currently known with an accuracy of about 10% ) provides a determination of the annihilation cross section under the conditions prevailing at the time of decoupling. Since WIMPs are non-relativistic at decoupling, it is useful to expand the total annihilation cross section as a power series in terms of the WIMP relative velocity $`v`$:
$$\sigma v=a+bv^2+\mathrm{}.$$
(1)
In a generic situation, one of the two terms in this equation dominates the cross section at decoupling ($`v^23T/M0.1`$): if $`s`$ wave annihilation is unsuppressed, the cross section is dominated by the $`a`$ term, whereas if the annihilation predominantly occurs in a $`p`$ wave, the $`b`$ term dominates. Therefore, a measurement of the present dark matter density determines the quantity $`\sigma _{\mathrm{an}}`$ defined in Ref. as the coefficient of the dominant term (i.e. $`\sigma _{\mathrm{an}}=a`$ for $`s`$-annihilators and $`\sigma _{\mathrm{an}}=b`$ for $`p`$-annihilators). This result, shown in Fig. 1, is completely independent of the particle physics model responsible for the WIMPs; the only requirement is that the spectrum be generic, which ensures that co-annihilation processes and resonances are unimportant<sup>1</sup><sup>1</sup>1The analysis can also be extended to the case of superWIMP dark matter .. Moreover, $`\sigma _{\mathrm{an}}`$ is largely independent of the WIMP mass and spin: roughly, $`\sigma _{\mathrm{an}}^s=0.85`$ pb for $`s`$-annihilators and $`\sigma _{\mathrm{an}}^p=7`$ pb for $`p`$-annihilators.
In this article, we extend the model-independent approach of Ref. to predict the fluxes of anomalous cosmic rays due to WIMP annihilation. Indirect WIMP searches predominantly concentrate on three signatures: anomalous high-energy gamma rays, antimatter (positrons, antiprotons, etc.), and neutrinos . While the dark matter density measurement determines the total cross section of WIMP annihilation, the distribution between the various possible final states ($`e^+e^{}`$, $`q\overline{q}`$, $`\gamma \gamma `$, $`W^+W^{}`$, etc.) is not constrained. In order to keep the analysis as model-independent as possible, we focus on the signatures that are least sensitive to this distribution, i.e. those that appear for the maximal number of final states. High-energy neutrinos and positrons are only produced if the WIMPs annihilate directly into $`\nu \overline{\nu }`$ or $`e^+e^{}`$ pairs, respectively, or (in smaller numbers) if the primary annihilation final state contains $`W/Z`$ bosons. Gamma rays, on the other hand, are produced almost independently of the primary final state (with $`\nu \overline{\nu }`$ being the only exception among two-body final states), and we therefore concentrate on this signature. There are several ways in which gamma rays can be produced in WIMP annihilation events. Two well known processes are the direct annihilation to photon-photon or photon-$`Z`$ pairs ($`\chi \chi \gamma \gamma ,\gamma Z`$, where $`\chi `$ denotes the WIMP) and fragmentation following WIMP annihilation into final states containing quarks and/or gluons. While these processes can be easily described within our approach, we will concentrate on another source of photons, the final state radiation (FSR), which has until now received far less attention in the literature<sup>2</sup><sup>2</sup>2A recent discussion of the FSR flux in the context of a specific model (universal extra dimensions) and a subset of primary final states (charged leptons) is contained in .. The FSR component of the gamma ray spectrum has several important advantages. First, FSR photons are produced whenever the primary products of WIMP annihilation are charged: e.g. charged leptons, quarks or $`W`$ bosons. Even if the WIMPs annihilate into $`ZZ`$, $`Zh`$, or $`hh`$ pairs, the charged decay products of these particles will contribute to the FSR flux; only the $`\nu \overline{\nu }`$ channel does not contribute. In contrast, the monochromatic photons are only produced when the WIMPs annihilate into $`\gamma \gamma `$ or $`\gamma Z`$ pairs; since these processes can only occur at one-loop level , only a small fraction of WIMP annihilation events results in these funal states. The fragmentation photons are not produced for leptonic final states. In this sense, out of the three components of the photon flux, the FSR component is the most robust. Second, even though the energy spectrum of the FSR photons is broad, in many cases (whenever the WIMPs annihilate directly into charged fermion pairs) the spectrum contains a sharp edge feature at an energy close to the WIMP mass . This feature can be extremely useful in differentiating the WIMP signal from the astrophysical background: while no detailed theoretical understanding of the background is available, it seems very unlikely that such a feature in the relevant energy range can be produced by conventional physics. This is in sharp contrast with the fragmentation photons, whose broad and featureless spectrum makes it difficult to rule out a more conventional astrophysical explanation if an excess over the expected background is observed.
This article is organized as follows. In Section 2 we present the model-independent approximate formulas for the energy spectrum of the FSR photons produced in WIMP annihilation events. We test the accuracy of our analytical results against explicit numerical calculations in specific models. In Section 3, we use these results to predict the gamma ray fluxes from WIMP dark matter annihilation in the Milky Way. After discussing the relevant backgrounds in Section 4, we estimate the sensitivity reach of the typical space-based and ground-based gamma ray telescopes in Section 5. We reserve Section 6 for our conclusions.
## 2 Final State Radiation in WIMP Annihilation
If a WIMP pair can annihilate into a pair of charged particles, $`X`$ and $`\overline{X}`$, annihilation into a three-body final state $`X\overline{X}\gamma `$ is always also possible. As long as the $`X`$ particles in the final state are relativistic, the cross section of this reaction is dominated by the photons that are approximately collinear with either $`X`$ or $`\overline{X}`$. These are referred to as the “final state radiation” (FSR) photons. In this kinematic regime, the cross section factorizes into the short-distance part, $`\sigma (\chi \chi X\overline{X})`$, and a universal collinear factor:
$$\frac{d\sigma (\chi \chi X\overline{X}\gamma )}{dx}\frac{\alpha Q_X^2}{\pi }_X(x)\mathrm{log}\left(\frac{s(1x)}{m_X^2}\right)\sigma (\chi \chi X\overline{X}),$$
(2)
where $`\alpha `$ is the fine structure constant, $`Q_X`$ and $`m_X`$ are the electric charge and the mass of the $`X`$ particle, $`s`$ is the center-of-mass energy ($`s4m_\chi ^2`$ for non-relativistic WIMPs), and $`x=2E_\gamma /\sqrt{s}`$. The splitting function $``$ is independent of the short-distance physics, depending only on the spin of the $`X`$ particles. If $`X`$ is a fermion, the splitting function is given by
$$_f(x)=\frac{1+(1x)^2}{x},$$
(3)
whereas if $`X`$ is a scalar particle,
$$_s(x)=\frac{1x}{x}.$$
(4)
If $`X`$ is a $`W`$ boson, the Goldstone boson equivalence theorem implies that $`_W(x)_s(x)`$. (The applicability of the Goldstone boson equivalence theorem is guaranteed whenever the collinear factorization in Eq. (2) is a good approximation, since both require $`m_\chi m_W`$.)
Does Eq. (2) provide a good approximation of the FSR photon spectrum from galactic WIMP annihilation in a realistic situation? To address this question, we compare the FSR photon spectrum obtained by a direct calculation in a specific model with the prediction of Eq. (2) with the appropriate parameters. For this comparison, we have used the minimal universal extra dimension (UED) model . We computed the cross section of the process $`B_1B_1e^+e^{}\gamma `$ using the CompHEP package . ($`B_1`$, the first Kaluza-Klein excitation of the hypercharge gauge boson, plays the role of the WIMP dark matter candidate in the UED model .) We have fixed the radius of the extra dimension to be $`R=(499.07`$ GeV$`)^1`$, corresponding to $`B_1`$ mass of 500 GeV. While Eq. (2) holds for any WIMP momentum, we have chosen the colliding WIMPs to be nonrelativistic ($`\sqrt{s}=1001`$ GeV), to approximate the kinematics typical of galactic WIMP collisions. The result of the direct cross section calculation is shown by the red histogram in Fig. 2. The blue (continuous) line corresponds to the prediction of Eq. (2) with the same $`\sqrt{s}=1001`$ GeV, $`X=e`$, and the appropriate value of $`\sigma (\chi \chi e^+e^{})5.67`$ pb. The good agreement between the line and the histogram proves the validity of the collinear approximation for the total cross section. Remarkably, the spectrum has a sharp step-like edge feature at the endpoint, $`EM_\chi `$. The origin of the feature is obvious from Eqs. (2) and (3): ignoring the $`x`$ dependence of the logarithm in Eq. (2), which only has a small effect on the spectrum, it is easy to see that the differential cross section approaches a non-zero constant value at $`x1`$, whereas it obviously has to vanish for $`x>1`$. Since it is difficult to imagine an astrophysical process providing a similarly sharp endpoint feature at the relevant energy scales, observing the step would provide a strong evidence for WIMPs .
If the primary product of WIMP annihilation is a lepton pair ($`e^+e^{}`$, $`\mu ^+\mu ^{}`$), the FSR mechanism discussed above is the dominant source of secondary photons. On the other hand, if the WIMPs annihilate into quark-antiquark or $`\tau ^+\tau ^{}`$ pairs, an additional contribution to the secondary photon flux arises from hadronization and fragmentation. This contribution is dominated by the decays of neutral pions. While the fragmentation photons are more numerous than the FSR photons, they tend to be softer. The spectrum close to the endpoint is still dominated by the FSR component, and can be predicted using Eq. (2), with an appropriate choice of the “effective” value of $`m_X`$ in the logarithm. This is illustrated in Figure 3, which shows the secondary photon fluxes from a primary $`u`$ quark of 250 GeV energy. The upper (blue) histogram shows the total $`\gamma `$ spectrum, including both fragmentation and FSR components, calculated using the PYTHIA package ; the lower (red) histogram shows the PYTHIA prediction for the FSR flux alone. The total flux above about 225 GeV ($`x=0.9`$) is dominated by the FSR component, and exhibits the expected edge feature. The FSR spectrum predicted by PYTHIA is consistent with the prediction of Eq. (2); however, to obtain a good fit, the quantity $`m_u`$ in the logarithm should be replaced with the “effective mass” $`m_u^{\mathrm{eff}}`$, which takes into account soft gluon radiation and other effects of strong interactions. The black dashed line in Fig. 3 represents the prediction of Eq. (2) using the best-fit value of $`m_u^{\mathrm{eff}}=20`$ GeV. An excellent fit to the PYTHIA output is obtained. We conclude that the sharp endpoint with a shape given by Eqs. (2), (3) exists whenever the primary WIMP annihilation products are charged fermions: it does not matter whether they are leptons or quarks.
Based on the above discussion, we will replace the bare mass with the “effective” mass whenever we apply Eq. (2) to light quarks; this substitution will be implicit for the rest of the paper. For simplicity, we will assume $`m_u^{\mathrm{eff}}=m_d^{\mathrm{eff}}=m_s^{\mathrm{eff}}=0.2`$ GeV, since the scale for the effective mass is set by the QCD confinement scale. In reality, the situation is more complicated, since the effective mass may depend on the energy and flavor of the primary quark. Note also that the values needed to fit the PYTHIA predictions are substantially higher than $`\mathrm{\Lambda }_{\mathrm{QCD}}`$, but the interpretation of this result is not clear due to large uncertainties inherent in the showering algorithm and the fit. However, since the dependence on the mass is logarithmic, changing $`m^{\mathrm{eff}}`$ has only a moderate effect on the photon flux predictions. We have confirmed that replacing our simple assumption with an energy-dependent value of $`m^{\mathrm{eff}}`$ based on a fit to PYTHIA predictions does not substantially affect any of the estimates of photon fluxes and telescope sensitivities made below.
Unfortunately, a model-independent prediction of the sharp endpoint is not valid if the primary annihilation products of the WIMP are bosons, such as $`W^+W^{}`$ pairs or the charged Higgs bosons in the minimal supersymmetric standard model (MSSM). According to Eq. (4), $`lim_{x1}_s(x)=0`$; because of this, the flux near the endpoint is dominated by the model-dependent non-collinear contributions, and no firm model-independent prediction of the shape of the endpoint spectrum is possible.
FSR photons will also be produced when the WIMPs annihilate into neutral, unstable particles, whose decay products are charged: in the Standard Model, these could be $`ZZ`$, $`Zh`$, or $`hh`$ pairs. For example, consider the process $`\chi \chi ZZ`$ with the subsequent $`Z`$ decay into charged fermions ($`\mathrm{}^+\mathrm{}^{}`$ or $`q\overline{q}`$), which in turn emit an FSR photon. The photon spectrum in the $`Z`$ rest frame is given by Eq. (2), with the substitutions $`sm_Z^2`$ and $`\sigma (\chi \chi X\overline{X})2\sigma (\chi \chi ZZ)\mathrm{Br}(ZX\overline{X})`$. Performing the boost to return to the laboratory frame yields
$$\frac{d\sigma }{dx}=\frac{\alpha }{\pi }\sigma (\chi \chi ZZ)\mathrm{\Psi }_Z(x),$$
(5)
where
$$\mathrm{\Psi }_Z=2\theta (m_\chi m_Z)\frac{1}{x}\left[1+\frac{1x}{v}\frac{2x^2}{v(v+1)}+\frac{2x}{v}\mathrm{log}\frac{2xv}{1+v}\right]\underset{X}{}Q_X^2\mathrm{Br}(ZX\overline{X})\mathrm{log}\left(\frac{m_Z^2}{m_X^2}\right).$$
(6)
In this equation, $`x=E_\gamma /m_\chi `$, $`v=\sqrt{1m_Z^2/m_\chi ^2}`$ is the velocity of the $`Z`$ boson, and the sum runs over all the charged fermion pairs that $`Z`$ can decay into. We have ignored the corrections that are not enhanced by $`\mathrm{log}(m_Z^2/m_X^2)`$. If $`m_\chi m_Z`$, the $`Z`$ bosons are relativistic and the spectrum is given by
$$\mathrm{\Psi }_Z=2\frac{2x+2x\mathrm{log}xx^2}{x}\underset{X}{}Q_X^2\mathrm{Br}(ZX\overline{X})\mathrm{log}\left(\frac{m_Z^2}{m_X^2}\right).$$
(7)
For the $`Zh`$ and $`hh`$ final states, we obtain expressions analogous to (5). The corresponding functions $`\mathrm{\Psi }_{Zh}`$ and $`\mathrm{\Psi }_h`$ can be easily obtained from Eq. (6) by replacing the parameters of the $`Z`$ boson with those of the Higgs where appropriate. In particular, in the limit $`m_\chi m_h`$ we obtain
$`\mathrm{\Psi }_{hZ}(x)`$ $`=`$ $`{\displaystyle \frac{2x+2x\mathrm{log}xx^2}{x}}{\displaystyle \underset{X}{}}Q_X^2[\mathrm{Br}(ZX\overline{X})\mathrm{log}\left({\displaystyle \frac{m_Z^2}{m_X^2}}\right)`$
$`+\mathrm{Br}(hX\overline{X})\mathrm{log}\left({\displaystyle \frac{m_h^2}{m_X^2}}\right)],`$
$`\mathrm{\Psi }_h(x)`$ $`=`$ $`2{\displaystyle \frac{2x+2x\mathrm{log}xx^2}{x}}{\displaystyle \underset{X}{}}Q_X^2\mathrm{Br}(hX\overline{X})\mathrm{log}\left({\displaystyle \frac{m_h^2}{m_X^2}}\right).`$ (8)
These expressions include only fermionic decays of the Higgs; we assumed that the Higgs is too light to decay into $`W/Z`$ pairs. The analysis can be straightforwardly generalized to include these decays. Unfortunately, it is clear from the above equations that the spectrum of FSR photons produced in $`Z/h`$ decays does not possess a sharp endpoint; instead, it approaches 0 gradually in the $`x1`$ limit. This means that the non-universal, model-dependent contributions may become dominant near the endpoint.
## 3 FSR Photon Flux Estimates
In general, the differential $`\gamma `$ flux from WIMP annihilations observed by a telescope can generally be written as
$$\frac{d^2\mathrm{\Phi }}{dEd\mathrm{\Omega }}=\left(\underset{i}{}\frac{d\sigma _i}{dE}vB_i\right)\frac{1}{4\pi m_\chi ^2}_\mathrm{\Psi }\rho ^2(l)𝑑l,$$
(9)
where the sum runs over all possible annihilation channels containing photons, and $`\sigma _i`$ and $`B_i`$ are the annihilation cross section and the number of photons per event in a given channel, respectively. The average is over the thermal ensemble of WIMPs in the galaxy. The integral is computed along a line of sight in the direction parametrized by $`\mathrm{\Psi }=(\theta ,\varphi )`$, and $`\rho (l)`$ is the mass density of WIMP dark matter at a distance $`l`$ from the observer. To obtain the FSR photon flux, we substitute the differential cross section for the final states containing such photons, given in Eqs. (2) and (5), into Eq. (9), and take into account that $`B_i=1`$ for these final states. We obtain
$`{\displaystyle \frac{d^2\mathrm{\Phi }_{\mathrm{FSR}}}{dEd\mathrm{\Omega }}}`$ $`=`$ $`[{\displaystyle \underset{X}{}}\theta (m_\chi m_X)Q_X^2\sigma _Xv_X(x)\mathrm{log}\left({\displaystyle \frac{4m_\chi ^2(1x)}{m_X^2}}\right)+\sigma _Zv\mathrm{\Psi }_Z(x)`$ (10)
$`+\sigma _{hZ}v\mathrm{\Psi }_{hZ}(x)+\sigma _hv\mathrm{\Psi }_h(x)]\times {\displaystyle \frac{\alpha }{\pi }}{\displaystyle \frac{1}{4\pi m_\chi ^3}}{\displaystyle }_\mathrm{\Psi }\rho ^2dl,`$
where $`x=E/m_\chi `$. The sum runs over all possible two-body final states with charged particles $`X`$ and $`\overline{X}`$, and $`\sigma _X=\sigma (\chi \chi X\overline{X})`$. To simplify notation, we have also defined $`\sigma _Z=\sigma (\chi \chi ZZ)`$, $`\sigma _h=\sigma (\chi \chi hh)`$, and $`\sigma _{hZ}=\sigma (\chi \chi Zh)`$.
In the spirit of Ref. , we define the total WIMP annihilation cross section, $`\sigma _0=\sigma (\chi \chi \mathrm{anything})`$, and the “annihilation fractions” for the two-particle final states, $`\kappa _X=\sigma _Xv/\sigma _0v`$. (Note that $`_X\kappa _X=1`$, up to a small correction due to the contribution of the processes with three or more particles in the final state.) With these definitions, the FSR flux can be written as
$$\frac{d^2\mathrm{\Phi }_{\mathrm{FSR}}}{dEd\mathrm{\Omega }}=\frac{\alpha }{\pi }\frac{1}{4\pi m_\chi ^3}\sigma _0v𝒢(x)_\mathrm{\Psi }\rho ^2𝑑l,$$
(11)
where
$$𝒢(x)=\underset{X}{}\theta (m_\chi m_X)Q_X^2\kappa _X_X(x)\mathrm{log}\left(\frac{4m_\chi ^2(1x)}{m_X^2}\right)+\kappa _Z\mathrm{\Psi }_Z(x)+\kappa _{hZ}\mathrm{\Psi }_{hZ}(x)+\kappa _h\mathrm{\Psi }_h(x).$$
(12)
Notice that almost all WIMP annihilation channels, with the exception of $`\nu \overline{\nu }`$ and $`gg`$ final states, contribute to the FSR photon flux; only the details of the flux depend on the distribution of the cross section among the channels.
The photon flux prediction is subject to large uncertainties in the distribution of dark matter in the galaxy. These uncertainties are conventionally parametrized by a dimensionless function
$$\overline{J}(\mathrm{\Psi },\mathrm{\Delta }\mathrm{\Omega })\frac{1}{8.5\mathrm{kpc}}\left(\frac{1}{0.3\mathrm{GeV}/\mathrm{cm}^3}\right)^2\frac{1}{\mathrm{\Delta }\mathrm{\Omega }}_{\mathrm{\Delta }\mathrm{\Omega }}𝑑\mathrm{\Omega }_\mathrm{\Psi }\rho ^2𝑑l,$$
(13)
where $`\mathrm{\Delta }\mathrm{\Omega }`$ denotes the field of view of a given experiment. The values of $`\overline{J}`$ depend on the galactic halo model. The optimal line of sight for WIMP searches is towards the galactic center; in this case, the uncertainty is particularly severe due to the possibility of a sharp density enhancement at the center. For example, at $`\mathrm{\Delta }\mathrm{\Omega }=10^3`$ sr, typical values of $`\overline{J}`$ range from $`10^3`$ for the NFW profile to about $`10^5`$ for the profile of Moore et.al. , and can be further enhanced by a factor of up to $`10^2`$ due to the effects of adiabatic compression .
Using Eq. (11) and the above definition of $`\overline{J}`$ yields the FSR flux integrated over the field of view:
$$\frac{d\mathrm{\Phi }_{\mathrm{FSR}}}{dE}=\mathrm{\Phi }_0\left(\frac{\sigma _0v}{1\mathrm{pb}}\right)\left(\frac{100\mathrm{GeV}}{m_\chi }\right)^3𝒢(x)\overline{J}(\mathrm{\Psi },\mathrm{\Delta }\mathrm{\Omega })\mathrm{\Delta }\mathrm{\Omega },$$
(14)
where $`\mathrm{\Phi }_0=1.4\times 10^{14}`$ cm<sup>-2</sup> sec<sup>-1</sup> GeV<sup>-1</sup>.
While Eq. (14) provides a complete description of the FSR photon spectrum, the shape and the normalization of the flux for the most energetic photons (close to $`x=1`$) is of particular interest due to the possibility of the observable edge feature. In this region, the flux is dominated by the photons radiated by fermion products of WIMP annihilation. Neglecting the $`x`$ dependence in the logarithm, whose only effect is to slightly smooth out the edge in the region $`1x1`$, the flux is approximately given by
$$\frac{d\mathrm{\Phi }_{\mathrm{FSR}}}{dE}=\mathrm{\Phi }_0g\left(\frac{100\mathrm{GeV}}{m_\chi }\right)^3_f(x)\overline{J}(\mathrm{\Psi },\mathrm{\Delta }\mathrm{\Omega })\mathrm{\Delta }\mathrm{\Omega }.$$
(15)
Here, the dimensionless parameter $`g`$ contains all the information about the primary WIMP annihilation processes:
$$g=\left(\frac{\sigma _0v}{1\mathrm{pb}}\right)\underset{f}{}Q_f^2\kappa _f\mathrm{log}\left(\frac{4m_\chi ^2}{m_f^2}\right),$$
(16)
where the sum runs over all kinematically accessible fermionic final states. Depending on the microscopic model giving rise to the WIMP, the parameter $`g`$ can vary between 0 (if, for example, the WIMPs can only annihilate into neutral states) and about 35 in the most favorable case of very heavy WIMPs annihilating into electron-positron pairs in an $`s`$ wave.
The normalization of the FSR photon flux is determined by the quantity $`\sigma _0v`$. As we argued in the Introduction, the measurement of the present cosmological abundance of dark matter determines the total WIMP annihilation cross section at decoupling ($`v^21/10`$). A typical relative velocity of galactic WIMPs is much smaller, $`v10^3`$. In models where the $`s`$-wave annihilation is unsuppressed, the quantity $`\sigma v`$ is velocity-independent at low $`v`$, allowing us to make a robust model-independent prediction:
$$\sigma _0v=\sigma _{\mathrm{an}}^s\mathrm{\hspace{0.17em}0.85}\mathrm{pb}.$$
(17)
If, on the other hand, the cross section at decoupling is dominated by the $`b`$ term, no firm prediction for the quantity $`\sigma _0v`$ is possible: even a small $`a`$ term, if present, may become dominant for galactic WIMPs due to the low value of $`v`$. If no $`a`$ term is present, we estimate $`\sigma _0v=\sigma _{\mathrm{an}}^pv^210^5`$ pb; a larger cross section is possible if an $`a`$ term is present, with the upper bound provided by Eq. (17). Given the uncertainty present in the $`p`$-annihilator case, we will use the $`s`$-annihilator WIMP examples to illustrate our approach in the remainder of this paper.
## 4 Background Fluxes
Estimating the sensitivity of WIMP searches also requires the knowledge of background fluxes. The searches for anomalous cosmic $`\gamma `$ rays are conducted both by space-based telescopes and ground-based atmospheric Cerenkov telescopes (ACTs). The space-based telescopes observe photons directly, and the only source of irreducible background in this case is the cosmic $`\gamma `$ rays of non-WIMP origin. The ACTs observe the Cerenkov showers created when a cosmic ray strikes the upper atmosphere, and are subject to the additional backgrounds of Cerenkov showers from leptonic and hadronic cosmic rays. In our estimates of the experiments’ sensitivities, we will use simple power-law extrapolations of the background fluxes measured at low energies. For the non-WIMP photon flux, we use
$$\frac{d^2\mathrm{\Phi }_{\gamma ,\mathrm{bg}}}{dEd\mathrm{\Omega }}=4\times 10^{12}N_0(\mathrm{\Psi })\left(\frac{100\mathrm{GeV}}{E}\right)^{2.7}\mathrm{cm}^2\mathrm{s}^1\mathrm{GeV}^1\mathrm{sr}^1,$$
(18)
where the function $`N_0`$ describes the angular distribution of the photons (an approximation is given in Refs. .) In our analysis, we will replace $`N_0(\mathrm{\Psi })\mathrm{max}N_089`$. This generally overestimates the background; however, the effect is small, especially for the line of sight close to the direction to the galactic center. The non-photonic background flux for the ACTs is estimated as
$`{\displaystyle \frac{d^2\mathrm{\Phi }_{\mathrm{lep}}}{dEd\mathrm{\Omega }}}`$ $`=`$ $`1.73\times 10^8\left({\displaystyle \frac{100\mathrm{GeV}}{E}}\right)^{3.3}\mathrm{cm}^2\mathrm{s}^1\mathrm{GeV}^1\mathrm{sr}^1,`$
$`{\displaystyle \frac{d^2\mathrm{\Phi }_{\mathrm{had}}}{dEd\mathrm{\Omega }}}`$ $`=`$ $`4.13\times 10^8ϵ_{\mathrm{had}}\left({\displaystyle \frac{100\mathrm{GeV}}{E}}\right)^{2.7}\mathrm{cm}^2\mathrm{s}^1\mathrm{GeV}^1\mathrm{sr}^1,`$
where $`ϵ_{\mathrm{had}}`$ is the telescope-dependent probability that a hadronic Cerenkov shower will be misidentified as a photonic shower, normalized so that it is equal to one for the Whipple telescope (see Ref. ).
It is worth emphasizing that the fluxes (18) and (LABEL:act\_bgd) are merely extrapolations; in both cases the background cannot be accurately predicted from theory. While we will use these fluxes in our estimates, one should keep in mind that there are large uncertainties associated with them. This is why merely observing a flux enhancement is in general not sufficient to provide convincing evidence for WIMPs; a discovery of the step-like edge feature in the spectrum would greatly strengthen the case.
## 5 Sensitivity Reach of Future Telescopes
To illustrate the prospects for observational discovery of the FSR edge, we will use two toy scenarios. In the first scenario, the annihilation fractions for two-body final states are taken to scale as $`Y^4N_c`$, where $`Y`$ is the hypercharge of the final state particles, and $`N_c=3`$ for quarks and $`1`$ for other states<sup>3</sup><sup>3</sup>3Note that Eq. (2) can be applied to polarized final states. Therefore, accounting for the different hypercharge of left-handed and right-handed fermions is straightforward.. An explicit example in which this scenario is realized is provided by the model with universal extra dimensions , and we will therefore label it as UED. In the second scenario, the WIMPs do not annihilate into bosonic final states, while the annihilation fractions $`\kappa _i`$ for all kinematically accessible fermion final states are equal (up to a factor of $`N_c`$). We will refer to this scenario as “democratic”. In both scenarios, we assume that the WIMPs can only annihilate into Standard Model particles, and use a Higgs mass of 120 GeV. The values of the quantity $`g`$, defined in Eq. (16), as a function of the WIMP mass $`m_\chi `$, in the two scenarios under consideration are shown in Figure 4. In both cases, we assume that WIMPs are $`s`$-annihilators, with the total annihilation cross section given by Eq. (17).
The magnitude of the FSR photon flux in each scenario is easily estimated using Eq. (14). As an example, Fig. 5 shows the number of events per 100 GeV bin expected to be observed at an ACT with an exposure time $`T=50`$ hrs, and a field of view $`\mathrm{\Delta }\mathrm{\Omega }=4\times 10^3`$ sr. (These parameters are similar to those of the VERITAS and HESS telescope arrays.) The effective collection area of the ACTs depends on the photon energy; in our analysis, we use an analytic fit to the effective area of the VERITAS array shown in Fig. 4, Ref. :
$$A(E)=\mathrm{\hspace{0.17em}1.2}\mathrm{exp}\left[0.513\left(\mathrm{log}\frac{E}{5\mathrm{TeV}}\right)^2\right]\times 10^9\mathrm{cm}^2.$$
(20)
We assumed the UED scenario with an 800 GeV WIMP. We have further assumed $`\overline{J}(\mathrm{\Delta }\mathrm{\Omega })=10^5`$, which is the case in the NFW galactic profile with an adiabatic compression enhancement factor of about a 100 , or in the profile of Moore et. al. with no adiabatic compression. It is clear from the figure that the edge feature due to the FSR photon emission following WIMP annihilation should be easily discernible in this data set.
An analogous plot illustrating the observability of the edge feature at the GLAST space telescope is shown in Fig. 6. We have assumed a collection area $`A=10000`$ cm<sup>2</sup>, an exposure time $`T=2`$ years, and a field of view<sup>4</sup><sup>4</sup>4The field of view at GLAST can be varied between about $`5\times 10^6`$ sr (the angular resolution of the telescope) and 2.3 sr (the full field of view). While larger values of $`\mathrm{\Delta }\mathrm{\Omega }`$ are advantageous from the point of view of statistics, focusing narrowly on the galactic center can lead to improved signal/background ratio if the dark matter density has a sharp peak at the center. However, reducing $`\mathrm{\Delta }\mathrm{\Omega }`$ substantially below 10<sup>-3</sup> typically results in insufficient statistics with the assumed collection area and exposure time. $`\mathrm{\Delta }\mathrm{\Omega }=10^3`$ sr. We have further assumed the “democratic” scenario with a 100 GeV WIMP, and a galactic model with $`\overline{J}(\mathrm{\Delta }\mathrm{\Omega })=5\times 10^4`$. Again, the edge feature would be easily discernible for these parameters.
In addition to the FSR photon flux plotted in Figs. 5 and 6, photons are also expected to be produced both by quark fragmentation and loop-induced $`\chi \chi \gamma \gamma ,\gamma Z`$ annihilation processes. As we showed in Section 2, the fragmentation component is subdominant to the FSR flux near the endpoint, and therefore will not affect the edge feature. However, this component may dominate the flux at lower energies, in which case the edge feature would be accompanied by a sharp change in the slope of the spectrum. The monochromatic photon flux from $`\chi \chi \gamma \gamma `$ will contribute to the signal in the bin containing $`E_\gamma =m_\chi `$. This contribution is also generally subdominant since $`\sigma (\chi \chi \gamma \gamma )/\sigma (\chi \chi X\overline{X}\gamma )\alpha 10^2`$. If present, the line contribution will make the edge feature even sharper than our predictions based on the FSR flux alone.
To observe the FSR edge feature in the photon spectrum, the experiments need to search for a large drop in the number of events between two neighboring energy bins. A statistically significant discovery requires that the drop be larger than what can be expected from a fit to the rest of the spectrum. While a detailed analysis of the reach of any particular telescope is beyond the scope of this article, a simple estimate of the reach can be obtained as follows. Consider the energy bin $`[m_\chi (1\delta ),m_\chi (1+\delta )]`$, where $`\delta `$ is the fractional energy resolution of a telescope<sup>5</sup><sup>5</sup>5The assumption that the bin is centered at $`m_\chi `$ represents the worst-case scenario for the reach; the reach can be improved by up to a factor of $`\sqrt{2}`$ by optimizing the binning to maximize the significance. In addition, our estimates ignore the possible monochromatic photon flux from $`\chi \chi \gamma \gamma `$, which would appear in the same bin. The fragmentation photon flux, which is subdominant to the FSR component but could still enhance the signal, is also ignored. In this sense, our reach estimates are rather conservative.. The number of signal events in this bin is
$$N_{\mathrm{sig}}1.4\times 10^{12}g\delta \left(\frac{100\mathrm{GeV}}{m_\chi }\right)^2\overline{J}(\mathrm{\Delta }\mathrm{\Omega })A_{\mathrm{cm}^2}T_{\mathrm{sec}}\mathrm{\Delta }\mathrm{\Omega },$$
(21)
where $`A_{\mathrm{cm}^2}`$ and $`T_{\mathrm{sec}}`$ are the area of the telescope in cm<sup>2</sup> and the collection time in sec, respectively. Assuming that the fit to the high energy part of the spectrum ($`E>m_\chi `$) produces an estimate of the background consistent with Eqs. (18) and (LABEL:act\_bgd), the expected number of background events $`N_{\mathrm{bg}}`$ in the energy bin $`[m_\chi (1\delta ),m_\chi (1+\delta )]`$ can be computed. Requiring
$$N_{\mathrm{sig}}3\sqrt{N_{\mathrm{bg}}}$$
(22)
for a statistically significant discovery of the step, we find that a discovery at a space-based telescope is possible if
$$g\overline{J}(\mathrm{\Delta }\mathrm{\Omega })\mathrm{\hspace{0.17em}6}\times 10^8(A_{\mathrm{cm}^2}T_{\mathrm{sec}}\delta \mathrm{\Delta }\mathrm{\Omega })^{1/2}\left(\frac{m_\chi }{100\mathrm{GeV}}\right)^{1.15}.$$
(23)
This condition, together with the “minimal signal” requirement,
$$N_{\mathrm{sig}}10,$$
(24)
can be used to determine the reach of the GLAST telescope. The reach is shown in Fig. 7, where we plot the minimal value of $`\overline{J}`$ required for the discovery, as a function of the WIMP mass $`m_\chi `$, in the UED and “democratic” scenarios. The reach is shown for two values of $`\mathrm{\Delta }\mathrm{\Omega }`$: 2.3 sr, corresponding to utilizing the full field of view of the telescope, and 10<sup>-3</sup> sr, corresponding to focusing narrowly on the galactic center. (We assume the collection area $`A=10000`$ cm<sup>2</sup>, the exposure time $`T=2`$ years, and the energy resolution $`\delta =10`$%.) For $`\mathrm{\Delta }\mathrm{\Omega }=2.3`$ sr, the minimal signal criterion (24) is always weaker than the 3$`\sigma `$ requirement in Eq. (22), and we do not plot it. For $`\mathrm{\Delta }\mathrm{\Omega }=10^3`$ sr, on the other hand, the minimal signal criterion (24) dominates the reach determination for large masses; the dotted lines in Fig. 7 indicate the minimal value of $`\overline{J}`$ for which it is satisfied. Note that, while the reach in terms of $`\overline{J}`$ is clearly higher for the larger $`\mathrm{\Delta }\mathrm{\Omega }`$ due to higher statistics, the values of $`\overline{J}`$ in most galactic halo models are substantially enhanced at low values of $`\mathrm{\Delta }\mathrm{\Omega }`$.
The discovery reach for an ACT, assuming that the background is dominated by leptonic showers<sup>6</sup><sup>6</sup>6The leptonic background is dominant over the entire range of WIMP masses of interest, provided that ϵhad
<0.1
<subscriptitalic-ϵhad0.1\epsilon_{\rm had}\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}0.1., is given by
$$g\overline{J}(\mathrm{\Delta }\mathrm{\Omega })\mathrm{\hspace{0.17em}4}\times 10^9(A_{\mathrm{cm}^2}T_{\mathrm{sec}}\delta \mathrm{\Delta }\mathrm{\Omega })^{1/2}\left(\frac{m_\chi }{100\mathrm{GeV}}\right)^{0.85}.$$
(25)
To estimate the discovery potential of the VERITAS and HESS ACT arrays, consider an ACT with the collection area given in Eq. (20), an exposure time $`T=50`$ hrs, and the energy resolution $`\delta =15`$%. The discovery reach for such a telescope is shown in Fig. 8. Dashed contours correspond to an experiment utilizing the full field of view of the ACT, assumed to be $`4\times 10^3`$ sr. Solid contours indicate the reach of an experiment focusing narrowly on the galactic center, with the angular resolution of $`0.07^{}`$ corresponding to $`\mathrm{\Delta }\mathrm{\Omega }=5\times 10^6`$ sr. (Galactic halo model predictions for $`\overline{J}`$ for this value of $`\mathrm{\Delta }\mathrm{\Omega }`$ range from about $`10^4`$ to a few$`\times 10^8`$.) In the latter case, the minimal signal requirement (24) dominates the reach determination for large $`m_\chi `$, and is shown in the figure using the dotted lines.
Given a model for galactic halo profile and a set of assumptions about the relevant annihilation fractions, Figs. 7 and 8 can be used to estimate the reach of the telescopes in terms of the highest value of the WIMP mass for which the edge feature can be observed. The estimate indicates that the prospects for observing the feature are quite good. For example, with a rather conservative assumption $`\overline{J}(\mathrm{\Delta }\mathrm{\Omega }=4\times 10^3\mathrm{sr})=10^4`$, an ACT with the parameters used in our study would be able to discover the feature for $`m_\chi `$ up to about 2 TeV in the UED model, covering the entire range where the model is cosmologically consistent . Comparing Figs. 7 and 8 indicates that the VERITAS and HESS arrays have a sensitivity comparable to GLAST. The experiments are complementary in terms of the range of WIMP masses that can be covered: the ACT will be sensitive to values of $`m_\chi `$ between about 50 GeV and 10 TeV, while GLAST can observe the FSR edge if $`m_\chi `$ is in the 10 – 250 GeV mass range. We conclude that both space based telescopes and ACTs could provide sufficient sensitivity in the near future to discover the edge feature in the $`\gamma `$ flux if WIMPs are $`s`$-annihilators and the galactic halo profile and annihilation fractions are favorable.
Since the edge feature appears at $`E_\gamma =m_\chi `$, an observation of this feature would provide a direct measurement of the WIMP mass, with an accuracy determined by the energy resolution of the telescope, potentially better than 10%. This is especially interesting because this parameter would be difficult to measure in a collider experiment, since WIMPs are pair-produced and escape the detector without interacting. Thus, observation of the edge feature would provide information complementary to what will be obtained at the LHC. For example, in the case of supersymmetry, the LHC can often determine the mass differences between some of the superpartners and the lightest neutralino, but not always the overall mass scale . This ambiguity could be resolved if the edge feature in the gamma ray spectrum is observed.
## 6 Conclusions
In this article, we have obtained a prediction for the flux of photons produced as final state radiation in galactic WIMP annihilation processes. The prediction relies on the determination of the total WIMP annihilation cross section, which is provided by the measurement of the current cosmological dark matter abundance. As emphasized in , this determination does not require any assumptions about the fundamental physics giving rise to the WIMP, apart from the mild condition of a generic mass spectrum. While the distribution of the cross section among various possible final states is not constrained by cosmological arguments, the FSR photons are produced for almost every possible final state (with the exception of $`\nu \overline{\nu }`$ and $`gg`$), making this signature quite model-independent. Moreover, if the final state of WIMP annihilation is a pair of charged fermions (leptons or quarks), the FSR flux has a well-defined step-like edge feature, dropping abruptly at the energy equal to the WIMP mass. Observing such a feature would provide strong evidence for the WIMP-related nature of the flux distortion, and yield a measurement of the WIMP mass.
If WIMPs are $`s`$-annihilators, the predicted FSR fluxes can be quite sizable, and the edge feature can be easily discernible above the expected background. Using a rough statistical criterion, we have shown that both ground-based ACTs such as HESS and VERITAS and space-based gamma telescopes such as GLAST have a good chance of observing the edge feature. It is likely that our simplified analysis underestimates the ability of the experiments to observe a step-like feature in the photon spectrum; a more sophisticated statistical analysis is clearly needed to obtain a more realistic estimate of the reach.
In the $`p`$-annihilator WIMP case, the fluxes are expected to be lower, and it is difficult to make model-independent predictions due to the possible presence of an $`a`$ term in the annihilation cross section which would not affect the WIMP relic abundance, but could dominate galactic WIMP annihilation. Nevertheless, it would be interesting to analyze if observable FSR photon fluxes can be produced in models with $`p`$-annihilator WIMPs, such as the bino-like neutralino in supersymmetric models.
In summary, the flux of FSR photons emitted in the process of WIMP annihilation in the center of Milky Way could be observable. An observation of the step-like edge feature characteristic of this flux could provide the first robust signature of WIMP dark matter. We encourage the collaborations involved in the analysis of the data coming from ground-based and space-based gamma ray telescopes to perform systematic searches for this important signature in a model-independent fashion as presented here.
## Acknowledgments
We are grateful to the Aspen Center for Physics where part of this work has been completed. MP and AS are supported by the NSF grant PHY-0355005. KM and AB are supported by a US DoE OJI award under grant DE-FG02-97ER41029. |
warning/0507/hep-ph0507032.html | ar5iv | text | # Advantages and Distinguishing Features of Focus Point Supersymmetry
## I Introduction
The standard model of particle physics is fine-tuned. Quantum corrections to the scalar Higgs boson mass<sup>2</sup> are quadratically divergent, so that a natural estimate of their magnitude is $`\alpha M^2`$, where $`M`$ is a cutoff mass. If we associate the cutoff with unification scale or Planck scale physics, we find that the quantum corrections are much larger than the desired net result. This blemish has been a prime motivation for proposing supersymmetric extensions to the standard model. In models with low-energy supersymmetry, naturalness can be restored by having superpartners with approximately weak-scale masses Maiani . Low-energy supersymmetry facilitates several other theoretically desirable ideas, including, very notably, quantitatively accurate unification of gauge couplings Dimopoulos:1981yj . It also provides an excellent dark matter candidate Goldberg:1983nd .
Unfortunately, straightforward breaking of supersymmetry at the weak scale also opens the door to various difficulties. Together with many new particles it introduces many new possibilities for couplings, which generically induce unacceptable violations of observed approximate symmetries. Conservation of $`R`$-parity removes the most severe of these difficulties, but significant challenges remain. Superpartners are accompanied by many new flavor mixing angles and $`CP`$-violating phases. If those mixings and phases are of order unity, then constraints on flavor-changing neutral currents and the $`ϵ`$ parameter require some superpartner masses to be at or above $`10\text{TeV}`$ and 100 TeV, respectively Ciuchini:1998ix . If flavor mixing is suppressed, but $`CP`$-violating phases aren’t, the electron and neutron electric dipole moments still require some superpartners to have masses above $`2\text{TeV}`$ Pospelov:2005pr ; Olive:2005ru . Finally, bounds arising from theoretical estimates of proton decay and the Higgs boson mass are most easily obeyed if some superpartners have masses well above the weak scale Goto:1998qg ; Ambrosanio:2001xb . While none of these constraints is completely watertight, taken together they put considerable pressure on models that attempt to keep all superpartner masses close to the weak scale.
An alternative is to take the data at face value and explore the most straightforward interpretation: that some superpartners are superheavy, with masses well above the weak scale. Here we briefly compare and contrast conceptual frameworks for superheavy supersymmetry: focus point supersymmetry Feng:1999hg ; Feng:1999mn ; Feng:2000gh ; Feng:2000bp , which is our primary emphasis, inverted hierarchy models Drees:1985jx ; Dvali:1996rj , and split supersymmetry Arkani-Hamed:2004fb ; Giudice:2004tc . Operationally, below and even at LHC energies, they appear rather similar, for in all, the central proposal is to allow squark and slepton masses to be large, while keeping gaugino masses relatively small. Philosophically, however, they are quite different: focus point supersymmetry retains naturalness of the weak scale as a guiding principle and implements it through a dynamical mechanism, inverted hierarchy models retain naturalness for the weak scale and implement it by hypothesizing a specific family-dependent pattern of supersymmetry breaking masses, while split supersymmetry explicitly abandons naturalness.
Since the robust phenomenological and cosmological features of the focus point and split supersymmetry frameworks, first examined in detail in Refs. Feng:1999hg ; Feng:1999mn ; Feng:2000gh ; Feng:2000bp , are so similar, refined measurements will be needed to decide between them. We outline how measurements of superoblique parameters and other practical observables can accomplish that task. If we discover, through the appearance of gauginos but not squarks and sleptons at the Large Hadron Collider (LHC), that a structured form of supersymmetry breaking holds in nature, it will be important to carry out such measurements to elucidate the conceptual meaning of the discovery.
## II Focus Point Supersymmetry
Focus point supersymmetry is defined by the hypothesis that all squarks and sleptons are superheavy, with masses at the TeV scale or higher, while gauginos and Higgsinos remain at the weak scale, and the hypothesis that the weak scale arises naturally. There is tension between these hypotheses, but no contradiction Feng:1999hg ; Feng:1999mn . The naturalness requirement, that the electroweak potential is insensitive to small relative changes in the fundamental supersymmetry breaking parameters, can either be met straightforwardly, by having all these parameters small, or through focusing. In the latter alternative, renormalization group evolution focuses a large range of initial values, defined by the fundamental parameters at the unification scale, into a relatively small range of effective values for the phenomenologically relevant parameters at the weak scale.
In practice, insensitivity of the weak scale to variations in the fundamental parameters is largely guaranteed if focusing occurs for the up-type Higgs boson mass. It will occur if the soft scalar masses at the unification scale are in the ratio Feng:1999mn
$$(m_{H_u}^2,m_{\stackrel{~}{t}_R}^2,m_{\stackrel{~}{t}_L}^2)(1,1+x,1x)$$
(1)
for moderate values of $`\mathrm{tan}\beta `$, and
$$(m_{H_u}^2,m_{\stackrel{~}{t}_R}^2,m_{\stackrel{~}{t}_L}^2,m_{\stackrel{~}{b}_R}^2,m_{H_d}^2)(1,1+x,1x,1+xx^{},1+x^{})$$
(2)
for large values of $`\mathrm{tan}\beta `$, where $`x`$ and $`x^{}`$ are arbitrary constants. A universal scalar mass obviously satisfies both Eqs. (1) and (2), but in principle more general possibilities are allowed. Given Eq. (1) or Eq. (2), focusing occurs for any weak-scale gaugino masses and $`A`$-parameters, any moderate or large value of $`\mathrm{tan}\beta `$, and any top quark mass within existing experimental bounds. Note that focusing makes the weak scale insensitive to variations in parameters introduced to explain the weak scale, the supersymmetry breaking parameters, but not to variations in other parameters, such as the top quark Yukawa coupling. Of course, the fact that the measured top quark mass is compatible with focusing for simple boundary conditions is tantalizing, if preliminary, quantitative evidence for focus point supersymmetry.
Focus point supersymmetry has been studied in great detail for the specific case of minimal supergravity. For top quark mass $`m_t=174(178)\text{GeV}`$, the region in which all phenomenological constraints are satisfied and relic neutralino dark matter has the observed density is at $`m_03(8)\text{TeV}`$ Feng:1999mn ; Paige:2003mg . Such superheavy squarks and sleptons sufficiently suppress one-loop contributions to the electron and neutron electric dipole moments even for $`𝒪(1)`$ phases. Two-loop effects are dominant and might be within experimental reach in the near future Chang:1998uc . The high sfermion masses, together with additional suppression from squark and slepton degeneracy as occurs in unified focus point models, comfortably solve all problems with flavor-violation and flavor-violating CP-violation Feng:2000bp . Of course, given the Tevatron Run I average top mass of $`m_t=178.0\pm 4.3\text{GeV}`$ Azzi:2004rc and the most recent average including preliminary Run II results of $`m_t=174.3\pm 3.4\text{GeV}`$ Group:2005cg , values of $`m_t`$ higher than $`178\text{GeV}`$ are still well within current constraints. For such top masses, the focus point region moves to values of $`m_010\text{TeV}`$. In this regime the heaviness of squarks and sleptons can remove all the flavor and CP problems associated with low-energy supersymmetry without the need for flavor degeneracy or additional assumptions.
A broad variety of phenomenological implications and virtues of the focus point spectrum has been explored more generally in Refs. Feng:1999hg ; Feng:1999mn ; Feng:2000gh ; Feng:2000bp :
* A noteworthy feature is that radiative corrections to the predicted value of the Higgs boson mass arising from loops containing heavy top and bottom squarks can raise the Higgs boson mass well above current bounds Feng:2000bp . This feature does not occur for inverted hierarchy models Drees:1985jx ; Dvali:1996rj , in which the light fermions have superheavy partners, while the heavy fermions have light (weak-scale) superpartners. Like focus point supersymmetry, inverted hierarchy models resolve many of the phenomenological difficulties generically associated with low-energy supersymmetry without sacrificing naturalness, because experimental constraints are stringent only for observables involving the first two generations, while naturalness constraints are stringent only for fields with large couplings to the Higgs sector Drees:1985jx .
* Gauge unified focus point models naturally obey constraints on proton decay as well Feng:2000bp . Viewed in isolation, suppression of proton decay does not pose a critical problem: the dangerous processes involve virtual exchange of both standard model superpartners and unification-scale particles, especially the color triplet Higgs superpartners, and they can always be satisfied by raising the masses of the latter. But if we want to maintain the impressive quantitative success of the unification of couplings, which is a major motivation for low-energy supersymmetry, then obtaining sufficient suppression of proton decay is problematic Bajc:2002pg . Coupling constant unification constrains unification-scale threshold effects, which in simple unification models implies upper bounds on GUT-scale masses. With superheavy squarks and sleptons, this difficulty is resolved, and one is left with viable (and interesting!) expectations for proton decay.
* In focus point models the lightest supersymmetric particle (LSP) is a neutralino that provides a dark matter candidate with excellent prospects for detection Feng:2000gh . In this context, the neutralino cannot be pure Bino, because in that case it annihilates through $`\stackrel{~}{B}\stackrel{~}{B}f\overline{f}`$ with a $`t`$-channel sfermion $`\stackrel{~}{f}`$, and these processes become inefficient for $`m_{\stackrel{~}{f}}`$ in the multi-TeV range or above, leading to an overabundant relic density. For neutralinos with significant Wino or Higgsino component, however, $`\chi \chi WW`$ and $`\chi \chi ZZ`$ become efficient, and the LSP’s relic density is naturally in the desired range. For similar reasons, mixed neutralinos give rise to relatively large direct and indirect detection rates.
## III Abandoning Naturalness?
The confluence of the existing failure to explain the anomalously small value of the cosmological term in a natural way, the suggestion from inflationary scenarios that on ultra-ultra-large scales the Universe might be drastically inhomogeneous, and the longstanding indications that consistent solutions of the equations of string theory provide a plethora of candidate macroscopic universes Bousso:2000xa have rekindled interest in the possibility that selection effects (random or anthropic) play a more central role, and the program of explanation through symmetry and naturalness a less central role, than traditionally has been assumed in theoretical physics. While it is certainly logically possible that one will be driven in that direction, we feel that it is a wise methodological principle to attempt to maintain the tightest available explanatory framework until forced to abandon it. Moreover, in several specific instances, including the unification of couplings, the smallness of the $`\theta `$ term in QCD, and the extremely long lifetime of the proton, it is difficult to conceive of plausible selection effects that could supplant symmetry as an explanation of the observed phenomena.
The central proposal of split supersymmetry is to drop any direct connection between low-energy supersymmetry and the solution of the weak scale hierarchy problem Arkani-Hamed:2004fb ; Giudice:2004tc . On the face of it, that idea would suggest that all superpartners acquire unification or Planck-scale masses, if indeed one has supersymmetry at all. To preserve desirable features of low-energy supersymmetry, i.e., quantitative unification of couplings and the existence of a good dark matter candidate, however, additional residual symmetries (and fine-tunings, see below) are postulated to ensure that there are gauginos and Higgsinos with weak-scale masses. Thus, phenomenologically, split supersymmetry is very similar to focus point supersymmetry, but one no longer requires Eq. (1) or Eq. (2), and the squark and slepton masses are allowed to become arbitrarily large.
Are the distinctions testable? The answer is not immediately obvious, because those distinctions lie in the masses of the superheavy superpartners, which are beyond the reach of currently planned colliders and largely decouple from low energy observables.
## IV Tests of Naturalness
One might hope to distinguish focus point and split supersymmetry by finding evidence for extremely large squark and slepton masses. Extremely heavy sfermions lead, through radiative corrections, to large Higgs boson masses, for example. An even more striking prediction is that, for extremely heavy squarks, gluinos become long-lived, with lifetime Kilian:2004uj
$$\tau _{\stackrel{~}{g}}10^{12}\text{s}\left[\frac{m_{\stackrel{~}{q}}}{10^6\text{GeV}}\right]^4\left[\frac{1\text{TeV}}{m_{\stackrel{~}{g}}}\right]^5.$$
(3)
Long-lived, weak-scale gluinos have been studied in Refs. Farrar:1978xj . They arise in theories with weak-scale supersymmetry breaking where the gluino is the LSP or decays only to a gravitino LSP. Those studies motivated discussions of the accompanying collider phenomenology and appropriate triggers long before the proposal of split supersymmetry. Nevertheless, coexistence of long-lived gluinos with lighter neutralinos and charginos could provide an unambiguous signal of superheavy sfermions.
Unfortunately, for Eq. (3) to yield a practically detectable lifetime, sfermion masses probably must exceed $`10^6\text{GeV}`$. Such large masses pose a significant challenge, because Weyl anomaly-mediated contributions Randall:1998uk ; Giudice:1998xp require gaugino/Higgsino masses to be suppressed relative to sfermion masses by no more than a factor of $`g^2/(16\pi ^2)`$. If such contributions are present, then, the natural range for the superheavy sfermion masses is constrained to be at or below $`10^5\text{GeV}`$. Of course, given the few guiding principles in split supersymmetry, there is no requirement that anomaly-mediated contributions be present at the expected order of magnitude.
Both split supersymmetry and focus point supersymmetry can accommodate superheavy superpartner masses in the $`10^4`$ to $`10^5\text{GeV}`$ range. As noted above, the focus point mechanism preserves naturalness for $`m_t=178\text{GeV}`$ for scalar masses $`10\text{TeV}`$ and weak-scale gauginos and Higgsinos. However, the preferred sfermion mass range depends on the top quark mass and increases rapidly for larger $`m_t`$. A careful analysis of renormalization group equations and electroweak symmetry breaking is required to determine the exact relation. However, given the currently favored range of top quark masses, large sfermion masses above 10 TeV are certainly a possibility, and the mere presence of sfermion masses in this range cannot be used to distinguish between natural and fine-tuned theories.
A far more incisive method for differentiating superheavy particle spectra is through superoblique parameters Cheng:1997sq . Superoblique parameters measure splittings between dimensionless couplings and their supersymmetric analogues. Exact supersymmetry demands equality of these couplings, but split supermultiplets introduce corrections Chankowski:1989du ; Hikasa:1996bw . As with their electroweak analogues, the oblique corrections Peskin:1990zt , superoblique corrections are non-decoupling: they become large for highly split supermultiplets. They can be determined by precise measurements of the properties of light superpartners, which are kinematically accessible in both focus point and split supersymmetry frameworks. These properties imply that superoblique parameters are likely to play an essential role in the experimental exploration of any supersymmetric theory in which some superpartners are beyond direct detection.
The full set of possible superoblique parameters has been cataloged Cheng:1997vy , and their measurement at colliders has been explored in detail in several studies Cheng:1997vy ; Feng:1995zd ; Nojiri:1996fp ; Nojiri:1997ma ; Katz:1998br ; Kiyoura:1998yt ; Mahanta:1999hx .<sup>1</sup><sup>1</sup>1The super-oblique parameters have also recently been discussed again in the context of split supersymmetry, for example in Ref. Arkani-Hamed:2004fb , where a subset of them have been reparametrized and denoted $`\kappa `$. In the leading logarithm approximation, the superoblique parameters are
$$\stackrel{~}{U}_i\frac{h_i}{g_i}1\frac{g_i^2}{16\pi ^2}\left(b_{g_i}b_{h_i}\right)\times \mathrm{ln}R,$$
(4)
where $`i=1,2,3`$ denotes the gauge group U(1), SU(2), or SU(3), $`g_i`$ is the standard model gauge coupling, $`h_i`$ its supersymmetric analogue, and $`R`$ is the ratio between the effective superheavy superpartner mass scale and the weak scale. The coefficients $`b_{g_i}`$ and $`b_{h_i}`$ are the one-loop beta function coefficients for $`g_i`$ and $`h_i`$ for the effective theory between the superheavy and weak scales; $`b_{g_i}b_{h_i}`$ is therefore the contribution from standard model particles whose superpartners are superheavy. For focus point supersymmetry and split supersymmetry in which all sfermions are superheavy, the superheavy particles are in complete multiplets of SU(5), and so $`b_{g_i}b_{h_i}`$ is independent of $`i`$. Numerically, $`b_{g_i}b_{h_i}=4`$, and
$`\stackrel{~}{U}_1`$ $``$ $`1.2\%\mathrm{log}_{10}R`$ (5)
$`\stackrel{~}{U}_2`$ $``$ $`2.5\%\mathrm{log}_{10}R`$ (6)
$`\stackrel{~}{U}_3`$ $``$ $`8.3\%\mathrm{log}_{10}R.`$ (7)
In focus point supersymmetry and split supersymmetry, the superoblique parameters can be measured in a number of ways. As an example, consider the chargino mass matrix
$$M_{\chi ^\pm }=\left(\begin{array}{cc}M_2& \frac{1}{\sqrt{2}}h_2v\mathrm{sin}\beta \\ \frac{1}{\sqrt{2}}h_2v\mathrm{cos}\beta & \mu \end{array}\right).$$
(8)
In the limit of exact supersymmetry, the $`Whh`$ and $`\stackrel{~}{W}\stackrel{~}{h}h`$ couplings are identical, and so $`h_2`$ is equal to $`g_2`$, the SU(2) gauge coupling constant. Superheavy superpartners break this degeneracy, and predict a non-vanishing superoblique parameter $`\stackrel{~}{U}_2`$. Dark matter constraints require significant mixing in the neutralino and chargino sectors, and so it is likely that both charginos and all four neutralinos will be produced at the Large Hadron Collider and the International Linear Collider.
The possibility of measuring superoblique parameters at the International Linear Collider in scenarios with mixed charginos and neutralinos has been discussed in Refs. Feng:1995zd ; Cheng:1997vy ; Kiyoura:1998yt . Supersymmetric parameters may be constrained by measuring chargino and neutralino masses and bounding the polarized cross sections for chargino and neutralino pair production. The sensitivity to the superheavy mass scale entering through the dependence of the chargino mass matrix on $`\stackrel{~}{U}_2`$ may be quite large. For example, in the mixed scenario studied in Ref. Kiyoura:1998yt , the cross section $`\sigma _R=\sigma (e_R^{}e^+\chi _1^+\chi _1^{})`$ varies from $`50\text{fb}`$ to 62 fb as the superheavy scalar mass scale varies from 1 to 10 TeV. Given an integrated luminosity of $`50\text{fb}^1`$, the statistical uncertainty in $`\sigma _R`$ is $`2\%`$, corresponding to an uncertainty in the superheavy mass scale of $`\mathrm{\Delta }\mathrm{log}_{10}R0.1`$. Of course, this precision will be compromised by systematic experimental uncertainties and uncertainties in other supersymmetry parameters. The size of these effects depends on the underlying supersymmetry scenario realized in nature, the final properties of the International Linear Collider, and the success with which other experiments may be used to constrain supersymmetry parameters, such as $`\mathrm{tan}\beta `$. Nevertheless, barring the possibility that these effects completely degrade the statistical precision, constraints on the superheavy superpartner mass scale to within an order of magnitude ($`\mathrm{\Delta }\mathrm{log}_{10}R1`$) appear possible.
Fine structure within the superheavy superpartner mass spectrum may be constrained by precise measurements of branching fractions mediated by virtual superheavy superpartners. The branching fractions $`B(\stackrel{~}{g}q_R\overline{q}_R\chi )`$ and $`B(\stackrel{~}{g}q_L\overline{q}_L\chi )`$ are sensitive to the fourth powers of $`m_{\stackrel{~}{q}_R}`$ and $`m_{\stackrel{~}{q}_L}`$, respectively. For the cases of greatest interest here, where $`q=t,b`$, these branching fractions with polarized final states can be distinguished through the energy distributions of $`q`$ decay products. Splittings in the superheavy spectrum also result in different effective $`R`$ parameters for the different superoblique parameters, and so additional rough constraints on fine structure can also be obtained if the superoblique parameters can be measured in more than one way. Finally, $`m_{H_u}^2`$ and $`m_{H_d}^2`$ can be determined by precise measurements of $`\mu `$, $`\mathrm{tan}\beta `$, and other parameters entering the Higgs potential.
These weak scale parameters can then be extrapolated to high scales to determine the fundamental soft supersymmetry-breaking parameters. This program is challenging. However, if superheavy masses above 100 TeV are realized in nature, even rough constraints on the superheavy mass scale will likely provide evidence for fine-tuning or, alternatively, motivate focusing or other mechanisms different from those discussed so far. On the other hand, consistency with superheavy mass scales below 100 TeV and with the predictions of Eqs. (1) and (2) would constitute striking evidence for focus point supersymmetry and naturalness. It would further motivate mechanisms of supersymmetry breaking that explain Eqs. (1) and (2), providing essential guidance for the next step to more fundamental theories.
## Acknowledgments
The work of JLF is supported in part by NSF CAREER grant No. PHY-0239817, NASA Grant No. NNG05GG44G, and the Alfred P. Sloan Foundation. The work of FW is supported in part by funds provided by the U.S. Department of Energy under cooperative research agreement DE-FC02-94ER40818. |
warning/0507/astro-ph0507115.html | ar5iv | text | # Astrometric Methods and Instrumentation to Identify and Characterize Extrasolar Planets: A Review
## 1 Introduction
The astrophysics of planetary systems is a good example of a branch of science in which theory is mostly driven by observations. Hardly any of the properties of the sample of $`150`$ extrasolar giant planets discovered to-date<sup>1</sup><sup>1</sup>1See for example the continuously updated Extrasolar Planet Encyclopedia website (http://www.obspm.fr/encycl/encycl.html), and references to published papers therein. Objects with masses $`M_p`$ below the Deuterium-burning threshold of $`13`$ $`M_J`$ (where $`M_J`$ is the mass of Jupiter; Oppenheimer et al., 2000) are defined as planets and included in the catalog, with the exception of companions with masses as large as 20 $`M_J`$ in systems already containing one planetary mass companion. (Jupiters on few-days orbits, very high eccentricities, objects with masses five to ten times the mass of Jupiter) had been predicted a priori by theoretical models. Correlations among planetary orbital and physical parameters had not been anticipated. The dependence of the frequency and properties of planetary systems on some of the characteristics of the parent stars (mass, metallicity) had not been foreseen.
The unexpected properties of extrasolar planets have sparked new enthusiasm among theorists, who have engaged in fruitful intellectual confrontations, with the aim to move from a set of models describing separately different aspects of the physics of the formation and evolution of planetary systems to a plausible, unified theory capable of making robust and testable predictions. By analogy, a number of new as well as old techniques of astronomy has been fuelled by the new discoveries, with the twin goals to follow-up and better characterize the extrasolar planet sample and cover new areas of the discovery space. The result is an ongoing, positive, creative tension between theory and observation that will put to the test the most basic ideas of how planets form and evolve.
Among the detection techniques, astrometry is the oldest. Nonetheless, no planet discovery announcement has been ascribed to it yet, and a few confirmations of previously detected systems have made the news well after other novel methods, considered completely unrealistic a few years back, had already achieved important results. However, the contribution of astrometric measurements of sufficient precision is potentially very relevant for a continued improvement of our understanding of the formation and evolution of planetary systems, and possibly for the identification of the first Earth-sized objects worthy of follow-up observations to search for signs of the presence of essential ‘bio-markers’, i.e. of the existence of life as we know it outside of our Solar System.
It is the aim of this review paper to describe the current and future sensitivity of astrometric measurements, both from the ground and in space, and to delineate the areas of planetary science in which astrometry will be able to make significant contributions, by comparison with other direct and indirect methods for the detection and characterization of planetary systems. In Section 2 I summarize the observed properties of extrasolar planets, in connection with renewed theoretical efforts in the fields of planet formation and evolution. I describe in Section 3 the astrometric methods and instrumentation utilized to hunt for planets, in light of a number of technological as well as astrophysical challenges to be faced in order to achieve the required degree of measurement precision. Section 4 contains a brief review of past and present efforts to detect planets with astrometry. In Section 5 I discuss the planet finding capabilities of future astrometric observatories. Finally, in Section 6 I outline a number of experiments that can be conducted with high-precision astrometry, to illustrate its potential for important contributions to planetary science.
## 2 Emerging Statistical Properties of Planetary Systems
Ten years after the announcement of the first Jupiter-sized object orbiting a star other than the Sun (Mayor & Queloz, 1995), the number of extrasolar planets announced has increased by a few tens per year. Based on the large datasets mostly made available by ground-based Doppler surveys (e.g, Marcy et al. 2004b, , and references therein; Mayor et al., 2004, and references therein), estimates of the frequency $`f_p`$ of giant planets around solar-type (late F, G, and early K spectral types) stars in the solar neighborhood ($`D50`$ pc) are all in fair agreement with each other (Zucker & Mazeh 2001a, ; Tabachnik & Tremaine, 2002; Lineweaver & Grether, 2003; Marcy et al. 2004b, ; Naef et al., 2004). Quoted values range between $`f_p4\%5\%`$ and $`f_p6\%9\%`$ for planet masses in the range $`1M_JM_p10M_J`$, and orbital radii $`a3`$ AU, and $`0.5M_JM_p10M_J`$, and $`a4`$ AU, respectively.
On the observational side, the startling diversity of planetary systems, when compared with the properties of our own Solar System, has, if possible, become more evident with the addition of newly discovered planets. On the theoretical side, the only concept that has not yet undergone significant revision or criticism is the paradigmatic statement that planets form within gaseous disks around young T Tauri stars. Many old ideas have been revisited or revived, and a number of completely new ones has been proposed in an attempt to confront and explain the observational data on extrasolar planets.
### 2.1 Mass, Period, and Eccentricity Distributions
The extrasolar planet sample exhibits many interesting and surprising characteristics. The distribution of minimum masses <sup>2</sup><sup>2</sup>2Only the product $`M_p\mathrm{sin}i`$ between the actual companion mass and the unknown inclination of the orbital plane with respect to the line of sight can be derived from Doppler-shift measurements. rises towards lower masses (e.g., Tabachnik & Tremaine, 2002, and references therein; Lineweaver & Grether, 2003; Marcy et al., 2004a ). Incompleteness below $`M_p\mathrm{sin}i0.5`$ $`M_J`$ becomes increasingly important, although recently objects with minimum masses as low as the mass of Neptune have been discovered (Butler et al., 2004; McArthur et al., 2004; Santos et al. 2004b, ). The sharp cut-off at the high-mass tail of the distribution, above $`M_p\mathrm{sin}i1015`$ $`M_J`$, with very few low-mass companions to solar-type stars in the range $`10M_JM_p\mathrm{sin}i80M_J`$, and for orbital periods up to a decade, is often referred to as the “brown dwarf desert”.
Theoretical predictions of the mass distribution of extrasolar planets within the context of the core-accretion model (e.g., Lissauer, 1993; Pollack et al., 1996) of giant planet formation have recently been proposed (Alibert et al., 2005; Ida & Lin, 2004a , 2005), which qualitatively reproduce the observed one (particularly for objects with $`M_p5`$ $`M_J`$). Rice et al. () and Rafikov (2005) argue instead that giant planets formed by disk instability (e.g., Boss, 1997, 2001, 2004; Mayer et al., 2002) should preferentially populate the high-mass tail ($`M_p5`$ $`M_J`$) of the planet mass distribution, while Mayer et al. (2004) indicate the possibility that the range of masses of giant planets formed via core accretion and disk instability could significantly overlap.
The period ($`P`$) and eccentricity ($`e`$) distributions also contain interesting features (Tabachnik & Tremaine, 2002, and references therein; Udry et al., 2003; Marcy et al., 2004a ; Halbwachs et al., 2005). Orbital periods are found in the range $`1.5P5400`$ days. About 20% of the planet sample, the so-called ”hot” Jupiters, are found orbiting within 0.1 AU. The number of planets increases with orbital period for $`60P2000`$ days, with increasing incompleteness for orbital radii $`3`$ AU. The median of the eccentricity distribution of extrasolar planets is $`0.3`$. The orbits of extrasolar planets span the whole range of available eccentricities, and they can be extremely elongated (Naef et al., 2001). Planets orbiting within 0.1 AU are all found with $`e0.0`$, a feature usually explained in terms of tidal circularization (Goldreich & Soter, 1966).
Predictions on the actual orbital distance distribution of giant planets have been made within the context of core-accretion models which include mechanisms of inward orbital migration due to tidal interactions between a gaseous disk and an embedded planet (e.g., Goldreich & Tremaine, 1979, 1980; Lin & Papaloizou, 1993; Ward, 1986, 1997). Trilling et al. (2002) and Armitage et al. (2002) were able to qualitatively reproduce the observed semi-major axis distribution of giant planets, for $`a>0.1`$ AU. Similar results were obtained recently by Alibert et al. (2004, 2005) and by Ida & Lin (, 2005). However, these models largely neglect the difficult problem of identifying general mechanisms capable of stopping orbital migration (e.g., Terquem, 2003, and references therein). In the context of the disk instability model of giant planet formation, migration efficiency might not be very effective (Rice et al. 2003a, , ; Mayer et al., 2004), thus planets formed by this mechanism should be found on not too close-in orbits (Rice et al., 2003b ; Rafikov, 2005).
The large spread of orbital eccentricities is difficult to explain by the standard core-accretion model. Several mechanisms have been proposed to reproduce the observed $`e`$ distribution, which are based on dynamical interactions of various nature, such as interactions between the planet and a gaseous or planetesimal disk, planet-planet resonant interactions, close encounters between planets, or secular interactions with a distant companion star (for a review see Tremaine & Zakamska, 2004, and references therein), but none of them can represent alone the observed distribution in a natural way. Furthermore, in multiple-planet systems different eccentricity excitation mechanisms induce different evolution of the orbital alignment, and planetary orbits could be significantly non-coplanar. The alternative mode of planet formation by disk instability gives rise to eccentric orbits, but no clear prediction of the final distribution of eccentricities has been provided yet (e.g., Papaloizou & Terquem, 2001; Terquem & Papaloizou, 2002; Mayer et al., 2004).
### 2.2 Correlations
With improved statistics, in recent years a number of studies has been carried out to find evidence of correlations among orbital parameters and masses, and between planet characteristics and stellar host properties.
As initially pointed out by Zucker & Mazeh (2002), Udry et al. (2003) and more recently by Eggenberger et al. (2004), the extrasolar planet sample exhibits a statistically significant lack of massive, close-in planets. These objects are the easiest to detect with the Doppler method <sup>3</sup><sup>3</sup>3Recall that the radial-velocity amplitude $`KP^{1/3}M_p\mathrm{sin}i`$, thus the paucity of high-mass planets on short-period orbits is real, and not due to selection effects.
Regardless of the formation mode, orbital migration effects are the likely responsible for the observed $`M_p\mathrm{sin}iP`$ correlation. Many models can reproduce such results, including reduced migration efficiency due to gap opening (Ward, 1997; Trilling et al., 2002), substantial mass-loss through Roche lobe overflow (Trilling et al., 1998; Gu et al., 2003), and accelerated orbital decay due to enhanced tidal interactions with the host stars (Pätzold & Rauer, 2002). Finally, Ida & Lin () have derived a theoretical mass-period diagram that closely resembles the one of the extrasolar planet sample, and predicted a paucity of planets in the intermediate mass range $`0.05M_p0.5`$ $`M_J`$, for orbital distances $`<3`$ AU.
The possibility that super-solar metallicity could correspond to a higher likelihood of a given star to harbor a planet has been the subject of a large number of studies (for a detailed review see Gonzalez, 2003). Recent works (e.g., Santos et al., 2001, ; Fischer & Valenti, 2005) have conclusively shown that planet occurrence correlates strongly with the host stars’ primordial metallicity. Up to $`20\%`$ of metal-rich (\[Fe/H\] $`0.3`$) F-G-K stars harbor planets, while less than 3% of metal-poor stars (\[Fe/H\] $`0.0`$) have been found to be planet hosts.
Based on the core accretion model, Kornet et al. (2005) and Ida & Lin () have quantified the dependence of planetary frequency on stellar metallicity, in qualitatively good agreement with the observed trend. The alternative scenario of giant planet formation via disk instability, however, is mostly insensitive to the primordial metal content of the protoplanetary disk (Boss, 2002; Rice et al., 2003b ), thus planet occurrence should not be hampered around metal-poor stars. The observed trend suggests that giant planet formation by core accretion predominates in the metal-rich regime (\[Fe/H\]$`0.0`$). However, due to the low numbers of metal-poor stars (\[Fe/H\]$`0.5`$) surveyed to-date, no definitive conclusion can be drawn, except that maybe both mechanisms operate (Beer et al., 2004).
Several authors have searched in the past for possible correlations between stellar metallicity and planet properties. Udry et al. (2002), Santos et al. (2001, 2003), and Fischer et al. (2002) searched for correlations in the $`M_p\mathrm{sin}i`$-\[Fe/H\] and $`e`$-\[Fe/H\] diagrams, but concluded no statistically significant trend can be found. The $`P`$-\[Fe/H\] diagram deserves instead more attention. Gonzalez (1998) and Queloz et al. (2000) initially argued that metal-rich stars seem to possess an excess of very short-period planets with respect to other planet hosts. In later works (Santos et al., 2001, 2003; Laws et al., 2003) no trend was found. However, after removing some potential sources of bias, Sozzetti (2004) has shown how this trend is still present in the data, specifically when one restricts the analysis to single planets orbiting single stars.
If real, the $`P`$\[Fe/H\] correlation could either reflect the fact that migration rates are slowed down in metal-poor protoplanetary disks (Livio & Pringle, 2003), or it might be indicative of longer timescales for giant planet formation around metal-poor stars, and thus reduced migration efficiency before the disk dissipates (Ida & Lin, 2004a ).
Finally, planet frequency appears to correlate with the primary mass. In particular, as pointed out by e.g. Butler et al. (2004), the occurrence rate of giant planets orbiting within 2 AU around M dwarfs ($`0.3M_{}`$ $`M_{}0.6`$ $`M_{}`$) seems suppressed by about an order of magnitude with respect to that of analogous planets around F- and G-type dwarfs ($`0.8M_{}`$ $`M_{}1.3`$ $`M_{}`$).
The presently small number of giant planets discovered around M dwarfs might still be partly an artifact due to the small-number statistics (Butler et al., 2004). However, the observed trend is supported by theoretical arguments (Laughlin et al., 2004; Ida & Lin, 2005) for a strong dependence of planet occurrence rates on stellar mass, highlighted by a significantly suppressed probability of forming giant planets by core accretion around M dwarfs, and by an enhanced likelihood for M dwarfs to harbor Neptune-sized objects. Planet occurrence, on the other hand, may not be a strong function of the primary mass for objects formed by disk instability (e.g., Boss, 2000).
### 2.3 Multiple Systems and Planets in Stellar Systems
Some 10% of the planet hosts are found to be orbited by multiple systems, containing up to 4 planets, while $`12\%`$ of the planet-bearing stars are themselves components of wide multiple stellar systems, and in two of the latter cases the stars harbor multiple-planet systems.
A few authors have searched for differences between the distributions of orbital elements and masses of planets orbiting single and multiple stars and between those of single- and multiple-planet systems. Zucker & Mazeh (2002) and Eggenberger et al. (2004) presented evidence for no correlation between masses and periods of planets found in stellar systems, while Marcy et al. () compared visually the eccentricity and mass distributions of single planets and planetary systems, and concluded that no significant difference was apparent. Finally, Mazeh & Zucker (2003) have recently presented arguments for a correlation between mass ratios and period ratios among pairs of planets in multiple systems (assuming coplanarity of the orbits).
From a theoretical viewpoint, the overall impact of the presence of a secondary star on the efficiency of planet formation and migration is far from being clear. For example, Nelson (2000) and Mayer et al. (2005) argue that giant planet formation by either core accretion or disk instability can be strongly inhibited in relatively close binary systems with separation of order of a few tens of AUs. Boss (1998), however, comes to opposite conclusions. Due to enhanced migration and gas accretion rates (Kley & Burkert, 2000; Kley, 2000, 2001; Nelson, 2003) planets formed around binaries should not show evidence for a mass-period correlation. These predictions appear to agree with the observed trend.
Theoretical investigations of the long-term dynamical evolution of multiple-planet systems (e.g., Kiseleva-Eggleton et al., 2002; Ji et al., 2003, and references therein; Correia et al., 2005; Barnes & Quinn, 2004; Goździewski & Konacki, 2004, and references therein) have allowed to divide such systems in three broad classes: $`a)`$ “hierarchical” planetary systems, with widely separated orbits, in which dynamical interactions appear negligible; $`b)`$ planetary systems subject to strong secular interactions; $`c)`$ planetary systems locked in mean motion resonances, which in some cases exhibit important variations of the orbital elements on time-scales comparable to the time-span of the radial-velocity monitoring. In multiple-planet systems, regions of dynamical stability do exist inside the parent stars’ Habitable Zones <sup>4</sup><sup>4</sup>4The Habitable Zone of any star is defined as the range of orbital distances at which a potential water reservoir, the primary ingredient for the development of life as we know it, would be found in liquid form (e.g., Kasting et al., 1993), where Earth-sized planets may be found (e.g., Menou & Tabachnik, 2003, and references therein; Jones et al., 2005, and references therein). Furthermore, the detected giant planets in binaries are likely to reside in stable orbital configurations, and there are margins for the presence of rocky planets in the Habitable Zone of close binaries (e.g., Holman & Wiegert, 1999; Pilat-Lohinger et al., 2003, and references therein; Marzari et al., 2005; Musielak et al., 2005, and references therein), although the formation of terrestrial planets in such environments does not easily occur in the first place (Thébault et al., 2002; Raymond & Barnes, 2004; Barnes & Raymond, 2004; Raymond et al., 2004). However, the lack of information on the actual mutual inclination angles between pairs of planetary orbits somewhat limits the generality of these findings.
### 2.4 Planetary Radii and Atmospheres
A handful of hot Jupiters ($`P4`$ days) have been discovered by means of photometric transit surveys (Udalski et al., 2002a , 2002b, 2003; Alonso et al., 2004), and confirmed by high-resolution spectroscopic measurements (Torres et al., 2004; Bouchy et al., 2004; Moutou et al., 2004; Pont et al., 2004; Sozzetti et al., 2004; Konacki et al., 2003, 2004, 2005), while one, HD 209458b, was observed transiting (Charbonneau et al., 2000; Brown et al., 2001) subsequently to the detection of its gravitational pull on the star (Mazeh et al., 2000; Henry et al., 2000).
The combination of the Doppler-shift and transit photometry measurements allows to derive estimates of the true mass and radius of the planet. These two critically interesting parameters can then be used for directly constraining structural models of irradiated giant planets (see Guillot, 2005, and references therein, for a detailed review). The measured radii for six of the seven transiting planets provide good agreement with theoretical expectations, while all models seem to systematically underestimate the radius of HD 209458b. The recent successful detection of thermal emission in the infrared from the planet (Deming et al., 2005), and in particular the timing of the secondary eclipse, clearly suggests that the planet revolves on an orbit with no significant eccentricity, essentially ruling out mechanisms invoked to provide additional heat/power sources in the core, such as tidal dissipation of a nonzero eccentricity induced by the gravitational perturbation of an undetected long-period companion (Bodenheimer et al., 2001, 2003; Laughlin et al., 2005).
Transmission spectroscopy during transits has allowed to detect absorption features in the spectrum of HD 209458 which are indicative of the presence of various constituents in the planet’s atmosphere, notably sodium, hydrogen, oxygen, and carbon (Charbonneau et al., 2002; Vidal-Madjar et al., 2003, 2004). Also, the planet appears to have an extended atmosphere, presumably due to evaporation effects. In two cases, detection of the planet’s thermal emission (Charbonneau et al., 2005; Deming et al., 2005) has permitted to estimate the planets’ effective temperatures and Bond albedos, and to infer the presence of atmospheric water vapor and carbon monoxide.
Theoretical predictions of the atmospheric composition, temperature, and circulation of irradiated giant planets (Burrows et al., 2005; Fortney et al., 2005. For a review see Burrows, 2005) are in fair agreement with the first infrared direct detections, however a proper understanding of the fine details of the emergent spectra of TrES-1 and HD 209458b will require both improved modelling and larger, high-quality datasets. Finally, studies of the phenomenon of atmospheric escape from hot Jupiters (Lammer et al., 2003; Gu et al., 2003, 2004; Lecavelier des Etangs et al., 2004; Baraffe et al., 2004; Grießmeier et al., 2004) predict that, under strong irradiation, these objects, depending on their mass and orbital distance, could undergo significant evaporation of their gaseous envelope, in reasonable agreement with observations.
### 2.5 Toward a Unified Picture
The observational data on extrasolar planets show such striking properties that one must infer that planet formation and evolution is a very complex process. Indeed, the confrontation between theory and observations indicates that there are numerous problems in connection with the elucidation of planetary formation and evolution processes.
An ideal theory of planet formation and evolution should be capable of explaining in a self-consistent way, be it deterministic or probabilistic, all the different properties of planetary systems discussed above. To this end, the help from future data obtained with a variety of different techniques will be crucial. Ultimately, both theory and observation will have to provide answers to a number of fundamental questions, that can be summarized as follows. (1) Where are the earth-like planets, and what is their frequency? (2) What is the preferred method of gas giant planet formation? (3) Under which conditions does migration occur and stop? (4) What is the origin of the large planetary eccentricities? (5) Are multiple-planet orbits coplanar? (6) How many families of planetary systems can be identified from a dynamical viewpoint? (7) What are the atmospheres, inner structure and evolutionary properties of gas giant planets? (8) Do stars with circumstellar dust disks actually shelter planets? (9) What are the actual mass and orbital elements distributions of planetary systems? (10) How do planet properties and frequencies depend upon the characteristics of the parent stars (spectral type, age, metallicity, binarity/multiplicity)?
With the above questions in mind, I focus next on what the contribution of astrometry from ground and in space will be, by presenting a summary of methods and instrumentation, by reviewing past and present efforts, by discussing future prospects and by putting this technique in perspective with other planet-detection methods.
## 3 Astrometric Planet Detection Techniques
Astrometric detection of extrasolar planets can be conducted with instrumentation on the ground or in space. I describe in this Section the generic approach to planet detection and measurement with astrometry, in terms of what type of data are needed, how to extract and model the planet signal from the data in presence of a number of noise sources, and how to assess the significance of a detection. I will conclude the Section by summarizing results from a set of ground-based and space-borne experiments aimed at demonstrating the theoretical predictions on the achievable astrometric precision under a variety of conditions.
The general analysis methods can be applied to astrometric observables appropriately defined for both monopupil and diluted-aperture telescopes, both from the ground and in space. To this end, I will describe the techniques in terms of the basic observable and noise models and the estimation process. The observable model produces theoretical values for the data as a function of adjustable parameters. The noise model describes errors that corrupt the astrometric data. The estimation process finds parameter values that produce the closest agreement between the observable model estimates and the data in light of the noise model.
### 3.1 Observable Model
The astrometric observable is generally defined as the angular position of a star as measured by a given instrument in its local frame of reference. The measurement could be for example intrinsically one-dimensional, as is the case for space missions such as ESA’s $`Hipparcos`$ (Perryman et al., 1997) and $`Gaia`$ (Perryman et al., 2001), which are designed to perform angular position measurements in their sensitive directions by centroiding their diffraction-limited images. Or, it could be a set of two coordinates on the focal plane of the instrument, as is the case for ground-based telescopes (e.g, Gatewood, 1987; Dekany et al., 1994; Pravdo & Shaklan, 1996). Finally, it could be either the optical path-length difference between the two arms of an interferometer on the ground (Shao et al., 1988; Armstrong et al., 1998; Colavita et al., 1999; van Belle et al., 1998; Glindemann et al., 2003) or in space, such as NASA’s Space Interferometry Mission ($`SIM`$; Danner & Unwin, 1999), or the normalized difference between the signals of two photomultiplier tubes (the Transfer Function) of a space-borne interferometer, such as $`HST`$/FGS (Taff, 1990).
Both from the ground and in space, astrometric measurements can be performed in wide-angle mode, i.e relative to a local frame of reference composed of a set of one or more reference stars at typical angular distances of several degrees from the target object. If the selected local reference frame lies at $`1^{}`$, the data are said to be collected while operating in narrow-angle mode. From space, without the limiting presence of atmospheric turbulence, which induces large-scale wavefront distortions (Lindegren, 1980), the astrometric observable can be determined with respect to a global inertial reference frame by accurately bridging together multi-epoch observations of objects distributed everywhere in the sky (and thus separated by typically tens of degrees) and by adopting a global closure condition over the whole celestial sphere. The combination of such an observing scenario and data reduction method is called global astrometric mode (Kovalevsky, 1980).
Regardless of the mode of operation and of the instrument utilized to carry out the measurements, four categories of information should be identified for inclusion in the observable model utilized to calculate theoretical values of the observable with negligible errors: (1) the location and motion of the target (if working in global astrometric mode) and a possible set of reference stars (if working in wide-angle or narrow-angle mode), (2) the location and motion of the observing instrument (if on the ground) or the attitude of the spacecraft (if in space), (3) the number, masses, and orbital parameters of companions to the target (and reference stars where applicable), and (4) any physical effects that modify the apparent positions of stars.
#### 3.1.1 Stellar and Instrumental Parameters
The star information consists of the five basic astrometric parameters–position on the celestial sphere (2 parameters, say $`\lambda `$ and $`\beta `$), proper motion (2 parameters, say $`\mu _\lambda `$ and $`\mu _\beta `$), and parallax (1 parameter, say $`\pi `$)–plus the radial velocity $`v_r`$, which can be determined by auxiliary measurements or from a sufficiently large secular acceleration.
The stellar locations and motions are usually determined in the Solar-System barycentric frame in which the global frame is defined. A variety of transformations can be utilized to connect the instrument-specific observational frame to the stellar frame.
If the object’s position in the instrument and barycentric frame are described by the vectors $`𝐙`$ and $`𝐒_{}`$, respectively, then for an all-sky survey instrument such as $`Hipparcos`$ or $`Gaia`$ the mapping is specified by the $`3\times 3`$ rotation matrix $`𝖠`$:
$$𝐙=𝖠𝐒_{}$$
(1)
From this relation, the along-scan angular coordinate of the object, which constitutes the actual observable, can be solved for in terms of $`𝖠`$ and $`𝐒_{}`$. The matrix $`𝖠`$ is a continuous function of time that specifies the spacecraft attitude. It could be defined by a set of nine functions $`A_{ij}(t)`$, $`i,j=1,2,3`$. Or, it could be expressed in terms of three Euler angles ($`\varphi (t)`$, $`\theta (t)`$, $`\psi (t)`$), as it was done for $`Hipparcos`$. Alternatively, it could be described by means of the quaternion representation, as is presently envisioned for $`Gaia`$.
For an interferometer operating in wide-angle mode, both on the ground and in space, the measured optical path-length delay $`d_{}`$ corresponds to the instantaneous three-dimensional position of the target on the sky projected onto the interferometer baseline:
$$d_{}=𝐁𝐒_{}+C$$
(2)
The baseline vector $`𝐁=B𝐮_𝐛`$ of length $`B`$ describes the spacecraft attitude, while $`C`$ is a constant term representing residual internal optical path differences. In the narrow-angle regime, this expression is modified as follows:
$$\mathrm{\Delta }d_{,j}=𝐁(𝐒_{}𝐒_𝐣)$$
(3)
The relative optical path-length delay $`\mathrm{\Delta }d_{,j}`$ is then the instantaneous angular distance between the target and its $`j`$th reference star projected onto $`𝐁`$. Due to the differential nature of the measurement, the constant term $`C`$ cancels out, to first order.
Finally, in order to relate the detector frame of a ground-based monolithic telescope or $`HST`$/FGS to the actual coordinates of an object in the sky, a plate-reduction transformation is applied (e.g., Kovalevsky & Seidelmann, 2004). In this case, the two-dimensional standard cartesian coordinate vector $`𝐬(s_1,s_2)`$ describing the position of the target in the plane of the sky is mapped into the two-dimensional vector of measured coordinates on the detector $`𝐫(r_1,r_2)`$ via the transformation:
$$𝐫=𝖬𝐩$$
(4)
In the above expression, $`𝖬`$ is the model matrix:
$$𝖬=\left(\begin{array}{ccccccc}s_1& s_2& 1& 0& 0& 0& \mathrm{}\\ 0& 0& 0& s_1& s_2& 1& \mathrm{}\end{array}\right)$$
(5)
The column vector $`𝐩=(p_1,p_2,\mathrm{},p_n)`$ contains the so-called plate constants. A minimum of six is required, to describe scale and rotation factors and offsets of coordinate origins between the two frames of reference, but it is not uncommon to include focal-plane tilt and other optical distortion terms, in addition to terms dependent on the magnitude and color index of the star observed. The same relation holds for all the objects used as reference stars.
#### 3.1.2 Planet Parameters
Masses and orbits of companions to the target object (and reference stars where applicable) come from fitting a model of Keplerian orbital motion to the data. The Keplerian orbit of each companion is described by seven parameters: semi-major axis $`a`$ with respect to the center of mass of the system, period $`P`$, eccentricity $`e`$, inclination $`i`$, position angle of the line of nodes $`\mathrm{\Omega }`$, argument of pericenter $`\omega `$, and epoch of pericenter passage $`\tau `$.
The observable model computes the star’s reflex motion projected on the plane of the sky due to the gravitational pull of such companions, that might be stellar or sub-stellar (brown dwarfs and planets) in nature. If the primary mass is $`M_{}`$ and the secondary is a planet of mass $`M_p`$, then, assuming a perfectly circular orbit, the apparent amplitude of the perturbation, i.e. the stellar orbital radius around the center of mass of the system scaled by the distance from the observer, is the so-called astrometric signature:
$$\alpha =\frac{M_\mathrm{p}}{M_{}}\frac{a}{D}$$
(6)
If $`M_\mathrm{p}`$ and $`M_{}`$ are given in solar mass units, $`a`$ in AU, and $`D`$ in parsec, then $`\alpha `$ is in arcsec.
Table 1 summarizes the values of $`\alpha `$ for a range of planet masses at different orbital radii from a 1-$`M_{}`$ star at 10 pc, compared to typical values of parallax and proper motion for stars in the solar neighborhood.
As one can see, planetary signatures are a higher-order effect for astrometry. Jupiter-sized planets already produce perturbative effects whose size is smaller than the typical $`Hipparcos`$ milli-arcsecond (mas) measurement precision. Detection of orbital motion induced by terrestrial planets necessarily implies an improvement of 2-3 orders of magnitude in precision, down to the few micro-arcseond ($`\mu `$as) regime.
Finally, in the case for example of a multiple-planet system, simply considering independent Keplerian orbits might not be sufficient, whenever secular or resonant gravitational perturbations among planets in the systems (due to the presence of large mass ratios, highly eccentric orbits, commensurabilities between orbital periods, and significantly non-coplanar orbits) are strong enough to induce measurable variations of orbital elements over time-scales comparable to the time-span of observations. For these cases, additional information must be fed to the observable model, such as approximate analytical expressions describing the gravitational perturbations and consequent time variations of the orbital elements, or fully self-consistent fitting algorithms which include the direct solution of the equations of motion of an N-body system.
#### 3.1.3 Physical Effects
A variety of physical effects that cause the apparent coordinates of observed stars to differ from the transformed values of their true barycentric coordinates can be taken into account in principle. In order to understand which effects are more relevant, the driver is the limiting single-measurement precision the adopted instrument is designed to achieve. The 1 mas state-of-the-art astrometric precision has been set by $`Hipparcos`$ and $`HST`$/FGS. The expected improvement in measurement precision by a few orders of magnitude envisioned for future ground-based and space-borne instrumentation such as VLTI, Keck-I, $`SIM`$, and $`Gaia`$ will sensibly increase the order of higher approximations.
Higher-order perturbations can be classical in nature, such as additional secular variations in the target space motion with respect to the observer, or intrinsically relativistic, such as corrections to classical effects due to the motion of the observer itself, or contributions coming from the gravitational fields of massive bodies in the vicinity of the observer. Many of these effects are well-known in pulsar timing work, and are included in detailed models of pulse arrival times (e.g., Hellings, 1986; Wolszczan & Frail, 1992; Stairs et al., 1998, 2002, and references therein). However, they are often neglected in astrometric data reduction with mas-level precision.
Secular changes in proper motion (perspective acceleration) and annual parallax can be quantified as time-derivatives of these two astrometric parameters (e.g., Dravins et al., 1999, and references therein):
$`{\displaystyle \frac{\mathrm{d}\mu }{\mathrm{d}t}}`$ $`=`$ $`{\displaystyle \frac{2v_r}{AU}}\mu \pi `$ (7)
$`{\displaystyle \frac{\mathrm{d}\pi }{\mathrm{d}t}}`$ $`=`$ $`{\displaystyle \frac{v_r}{AU}}\pi ^2`$ (8)
The above quantities are expressed in arcsec yr<sup>-2</sup> and arcsec yr<sup>-1</sup>, respectively, if the radial velocity $`v_r`$ is in km s<sup>-1</sup>, $`\pi `$ is in arcsec, $`\mu `$ is in arcsec yr<sup>-1</sup>, and the astronomical unit $`AU=9.77792\times 10^5`$ arcsec km yr s<sup>-1</sup>.
I show in Figure Astrometric Methods and Instrumentation to Identify and Characterize Extrasolar Planets: A Review the values of $`\mathrm{d}\mu /\mathrm{d}t`$ and $`\mathrm{d}\pi /\mathrm{d}t`$ as a function of $`v_r`$ and of the product $`v_t\times v_r`$, where the tangential velocity $`v_t=AU\mu /\pi `$ (with the astronomical unit now defined as $`AU=4.74`$ km yr s<sup>-1</sup>). As one can see, the effect of changing annual parallax is below a few $`\mu `$as/yr for stars more distant than a few pc from the Sun, even assuming large values of $`v_r`$. Its inclusion in the observable model might then be limited to the nearest stars. The contribution from perspective acceleration drops below a few $`\mu `$as/yr<sup>2</sup> only for objects farther away than a few tens of pc, thus such corrective term might have to be taken into account in the observable model for a few hundred nearby, high velocity stars.
Relativistic corrections to the motion of the observer with respect to the solar-system barycenter (aberration) can be quantified through the formula (e.g., Klioner, 2003; Kovalevsky & Seidelmann, 2004):
$`\alpha _{\mathrm{aberr}}`$ $``$ $`{\displaystyle \frac{v}{c}}\mathrm{sin}\vartheta {\displaystyle \frac{1}{4}}{\displaystyle \frac{v^2}{c^2}}\mathrm{sin}2\vartheta `$ (9)
$`+{\displaystyle \frac{1}{6}}{\displaystyle \frac{v^3}{c^3}}\mathrm{sin}\vartheta (1+2\mathrm{sin}^2\vartheta )+O(c^4),`$
where $`\vartheta `$ is the angular distance between the direction to the target and the observer’s space velocity vector, $`v`$ is the modulus of the space velocity vector of the observer, and $`c`$ is the speed of light.
The magnitude of the classic aberration term (first order in $`v/c`$) is $`2030`$ arcsec, while the second-order relativistic correction amounts to $`13`$ mas. For a targeted measurement precision of 1 $`\mu `$as, terms of order $`(v/c)^3`$ must be retained, but may be dropped for less stringent requirements. In addition, for relativistic stellar aberration to be properly accounted for, the spacecraft’s velocity will need to be determined to an accuracy of 20 mm/sec or better.
The purely general relativistic effect of deflection of light by Solar-System objects is instead defined by the expression (e.g., Klioner, 2003):
$$\alpha _{\mathrm{defl}}=\frac{(1+\gamma )GM}{R_0c^2}\mathrm{cot}\frac{\psi }{2},$$
(10)
where $`\psi `$ is the angular distance between the given Solar-System body and the source, $`G`$ is the gravitational constant, $`R_0`$ is the distance between the observer and the Sun, $`M`$ is the mass of the perturbing body, and $`\gamma =1`$ is the Post-Post-Newtonian (PPN) parameter.
I summarize in Table 2 the magnitudes of the gravitational effects on limb-grazing light rays induced by all Solar-System planets and some of the largest moons. If the targeted measurement precision is of order of 1-10 $`\mu `$as, the observable model should consider deflection by all these bodies, when observing close to their directions, and even quadrupole effects induced by the non-oblateness of the major Solar-System planets may have to be considered (e.g., Klioner, 2003).
Indeed, several attempts have been made in the past years (Soffel, 1989; Brumberg, 1991; Klioner & Kopeikin, 1992; de Felice et al., 1998, 2001, 2004; Vecchiato et al., 2003; Klioner, 2003) to develop schemes for the reduction of astrometric observations at the $`\mu `$as precision level directly within the framework of General Relativity, either employing non-perturbative approaches or the PPN formulation (Will, 1993). If a complete general relativistic observable model is implemented, other more subtle effects arise, such as the need to re-define parallax, proper motion, and radial velocity, depending on the given choice of the local reference system of the observer (whose motion must also be described in physically adequate relativistic terms, see for example Klioner, 2004), and the fact that these parameters, when higher-order terms are included, cannot be considered anymore independent from each other. Furthermore, a number of possible effects that may be caused by gravitational fields produced outside of the Solar System might have to be considered, such as weak gravitational lensing by distant sources (e.g., Belokurov & Evans, 2002), binary systems viewed edge-on which contain compact objects such as neutron stars and/or black holes (Doroshenko & Kopeikin, 1995), and gravitational waves (Kopeikin et al., 1999).
### 3.2 Noise Model
Astrometric data contain correlated and uncorrelated instrumental, atmospheric (if operating from the ground), and astrophysical noise. The noise model describes these uncertainties for use in the estimation process, which, particularly by taking proper account of correlated errors, provides the most accurate and sensitive results. When a noise source can be identified and modelled deterministically, such as a newly-found companion to a reference star, its predictable effects can be incorporated into the observable model and any provision in the noise model is deleted. When uncertainties can be quantified but not modelled, they are accounted for in the noise model.
I summarize below a variety of known sources of astrometric noise. Where applicable, i.e. in the case of instrumental and atmospheric noise, the different implications for filled-aperture telescopes and interferometers will be discussed separately. For instrumental noise, a further distinction will be made depending on whether astrometric measurements are performed from the ground or in space. The relative importance of any given source of noise will be gauged bearing in mind the goal of achieving a final astrometric precision of order of a few $`\mu `$as.
#### 3.2.1 Instrumental Noise
The various sources of instrumental uncertainties in astrometric measurements can be described in terms of the two general classes of random, photon errors $`\sigma _{\mathrm{ph}}`$ and systematic errors $`\sigma _{\mathrm{sys}}`$.
The expression for the photometric noise of a monopupil telescope is (Lindegren, 1978):
$$\sigma _{\mathrm{ph}}=\frac{\lambda }{4\pi A}\frac{1}{SNR}$$
(11)
Here $`A`$ is the telescope aperture and $`\lambda `$ the monochromatic wavelength of the observations (both in meters), while $`SNR`$ is the signal-to-noise ratio of the target, including sky background and detector noise ($`SNR\sqrt{N}\sqrt{t}`$, where $`N`$ is the number of photoelectrons and $`t`$ is the exposure time in seconds). For a $`m_v=15`$ solar-type star observed near the zenith with good (better than 0.5 arcsec) seeing conditions, and assuming an overall system efficiency $`\epsilon =0.4`$ (including CCD quantum efficiency, atmospheric and optics transmission), the contribution from $`\sigma _{\mathrm{ph}}`$ over small fields of view ($`<1`$ arcmin) can be $`300`$ $`\mu `$as, $`30`$ $`\mu `$as, and $`3`$ $`\mu `$as in 1 hr integration for $`A=1`$ m, $`A=10`$ m, $`A=100`$ m, respectively (see for example Allen, 2000).
For an interferometer, the photometric error is expressed as (Shao & Colavita, 1992):
$$\sigma _{\mathrm{ph}}=\frac{\lambda }{2\pi B}\sqrt{\frac{t_c}{t}}\frac{1}{SNR},$$
(12)
with $`A`$ replaced by $`B`$ in Eq. 11. The atmospheric coherence time $`t_c`$, with average seeing conditions, is typically a few tens of ms in the near-infrared $`K`$ band (Shao & Colavita, 1992; Quirrenbach et al., 1994; Lane & Colavita, 2003), while the signal-to-noise ratio per coherence time $`SNR`$ is a measure of the uncertainty $`\sigma _\varphi `$ in the measurement of the phase of the interferometric fringes ($`\sigma _\varphi (\mathrm{SNR})^1`$, see for example Wyant, 1975). The value of $`SNR`$ in this case is not only a function of the number of counts $`N`$, dark count and background, and read-noise (as in the filled-aperture case), but also of the square of the complex visibility $`V^2`$ ($`SNR\sqrt{NV^2}`$. See for example Quirrenbach et al., 1994; Colavita, 1999).
If the interferometer sensitivity is limited by the actual atmospheric coherence time, accurate measurements of the fringe phase, and thus fringe tracking, can be performed only on very bright targets (typically $`m_k5.0`$), for which enough photons are collected in a coherence volume $`t_cr_c^2`$ (where $`r_c`$ is the coherence diameter). When the interferometer is used in phase-referencing mode (e.g., Lane & Colavita, 2003, and references therein), $`t_c`$ can be artificially extended by more than an order of magnitude, thus allowing to improve the limiting magnitude of the instrument, or allowing for higher $`SNR`$ at the same magnitude. The fundamental requirement is that target and reference object be separated by less than an isoplanatic angle $`\theta _i`$ <sup>5</sup><sup>5</sup>5The isoplanatic angle is the angle in the sky over which atmosphere-induced motion is well-correlated, usually of order of tens of arcsecs (e.g., Shao & Colavita, 1992). The isoplanatic angle $`\theta _ir_c/h\lambda ^{6/5}`$ (with $`h`$ the effective height of the turbulence profile. See for example Colavita, 1994), thus, as chances of finding reference objects are increased in larger fields, the obvious choice is to use the instrument in the infrared. It is however with long baselines that interferometers have a major photon-noise advantage. For a $`m_k=13`$ star, $`B=100`$ m, $`SNR=5`$, $`t_c=100`$ ms, $`\sigma _{\mathrm{ph}}1`$ $`\mu `$as in 1 hr integration can be achieved (e.g., Shao & Colavita, 1992).
The instrumental systematic term for a monolithic telescope can be for example (Pravdo & Shaklan, 1996) expressed as:
$$\sigma _{\mathrm{sys}}=\sqrt{\sigma _{\mathrm{CCD}}^2+\sigma _{\mathrm{OP}}^2}$$
(13)
The systematic limitations imposed by the CCD detectors (charge transfer efficiency, deviations from uniformity or from linear pixel response, etc.) through $`\sigma _{\mathrm{CCD}}`$ and optics imperfections (optical aberrations and distortions, pixelization, etc.) through $`\sigma _{\mathrm{OP}}`$ can be overcome with improvements in image detector and optics technology. Centroid accuracies of 1/100 of a pixel are today readily achievable (Monet et al., 1992), translating in a typical value of $`\sigma _{\mathrm{CCD}}50`$ $`\mu `$as. Future developments promise improvements of about one order of magnitude in CCD image location accuracy (Gai et al., 2001), with hopes to keep $`\sigma _{\mathrm{CCD}}510`$ $`\mu `$as. Optical aberrations and distortions for large apertures ($`>5`$ m) and small fields of view ($`<1`$ arcmin) are typically small. Pravdo & Shaklan (1996) have shown how for the Keck 10-m telescope $`\sigma _{\mathrm{OP}}5`$ $`\mu `$as, or less.
For narrow-angle measurements with an interferometer, the systematic term will read (e.g., Shao & Colavita, 1992):
$$\sigma _{\mathrm{sys}}=\sqrt{\sigma _\mathrm{l}^2+\sigma _\mathrm{B}^2}$$
(14)
The two main sources of systematic errors arise from the uncertainty $`\sigma _\mathrm{l}=\delta l/B`$ with which optical delay lines in long-baseline interferometers can control internal optical path delays, and from the uncertainty on the knowledge of the interferometric baseline $`\sigma _\mathrm{B}=(\delta B/B)\vartheta `$, where $`\vartheta `$ is the angular separation between the target and a reference star. As for the former, in order to reach a positional measurement precision of $`10`$ $`\mu `$as with $`B=200`$ m (the maximum baseline of the VLTI), measurements of optical paths must be made with an accuracy of $`\sigma _\mathrm{l}<10`$ nm, a challenging but not impossible achievement with today’s technology (see for example Shao et al., 1988). Due to the differential nature of the measurement, instead, knowledge of the instrument baseline does not need to be very precise. For $`B=200`$ m, $`\vartheta 20`$ arcsec, the goal of 10 $`\mu `$as precision is achieved by determining the baseline stability with an uncertainty $`\sigma _\mathrm{B}50`$ $`\mu `$m, a requirement relaxed by a few orders of magnitude with respect to a wide-angle measurement (e.g., Shao et al., 1990).
Finally, if the astrometric measurements are performed in space, additional random and systematic error sources must be taken into account, which are introduced by the satellite operations and environment. For example, a class of relevant error sources is related to the determination of the spacecraft attitude. Attitude errors can occur due to perturbations produced by the solar wind, micrometeorites, particle radiation, and radiation pressure. Thermal drifts and spacecraft jitter can also induce significant errors in attitude determination. However, these noise sources are hard to quantify a priori in a very general fashion. Detailed error models must be developed case by case (see Section 4.5). Thus, in the design of a space-borne observatory for high-precision astrometry, be it a monolithic telescope or an interferometer, ad hoc calibration procedures must be devised, in order to attain the goal of limiting the magnitude of such instrumental error sources at the few $`\mu `$as level.
#### 3.2.2 Atmospheric Noise
For ground-based instrumentation, the atmosphere constitutes an additional source of noise through both its turbulent layers (a random component) and due to the differential chromatic refraction (DCR) effect (a systematic component).
The problems caused by the DCR effect can be very difficult to overcome. The magnitude of the effect depends on zenith distance $`z_0`$, air temperature and pressure, spectral band and star color, and even the non-sphericity of the Earth (e.g., Gubler & Tytler, 1998). The precision of theoretical and empirical DCR correction formulae degrades very quickly with $`z_0`$. Even at small zenith distances, for a monolithic telescope the goal of $`\mu `$as astrometry is unlikely to be attained. Typical uncertainties for small values of $`z_0`$ can be of order of $`\sigma _{\mathrm{DCR}}13`$ mas, and increase by over an order of magnitude close to the horizon (Kovalevsky & Seidelmann, 2004).
For conventional narrow-angle astrometric measurements with separations of $`1030`$ arcmin, the positional error $`\sigma _{\mathrm{atm}}`$ due to atmospheric turbulence is weakly dependent on separation and does not depend on $`A`$ (or $`B`$). This prevents the achievement of sub-mas astrometric precision (e.g., Lindegren, 1980; Han, 1989). For separations $`<110`$ arcmin, the situation improves. In this regime, the astrometric error due to anisoplanatism for a filled-aperture telescope has been calculated in the past (Lindegren, 1980; Shao & Colavita, 1992) as:
$$\sigma _{\mathrm{atm}}300A^{2/3}\vartheta t^{1/2}$$
(15)
In this case $`\vartheta `$ is in radians, $`A`$ is expressed in meters, and the factor 300 is obtained directly from the phase-structure function describing the turbulence, assuming standard Kolmogorov-Hufnagel (Hufnagel, 1974) atmospheric and wind-speed profiles for good seeing conditions (FWHM $`0.5`$ arcsec), typical of a site such as the Keck Observatory (e.g., Shao & Colavita, 1992). The rather weak power dependency of $`\sigma _{\mathrm{atm}}`$ on target-reference star separation and objective diameter implies that a typical value $`\sigma _{\mathrm{atm}}1`$ mas is obtained with $`A=1`$ m in $`t=1`$ hr of integration with a separation $`\vartheta =1`$ arcmin. In order to suppress $`\sigma _{\mathrm{atm}}`$ by two-three orders of magnitude one would require unrealistically large $`A`$ and $`t`$, unless $`\vartheta `$ is limited to uselessly small angles.
Interferometers can in principle get much closer to the limits in precision set by the atmosphere. For a diluted architecture, in fact, Eq. 15 now reads $`\sigma _{\mathrm{atm}}300B^{2/3}\vartheta t^{1/2}`$. Values of $`B`$ of order of 100-200 meters are more easily attainable than the equivalent filled-aperture size. Thus, for a 20-arcsec star separation and a 200-m baseline, a 1-hr integration would allow to achieve $`\sigma _{\mathrm{atm}}10`$ $`\mu `$as (Shao & Colavita, 1992). In conditions of extremely favorable seeing, large isoplanatic angles, and long atmospheric coherence times, such as those reported above Dome C in Antarctica (Lloyd et al., 2002; Lawrence et al., 2004), atmospheric errors in image motion maybe reduced by another order of magnitude.
However, if phase-referencing is used to artificially increase $`t_c`$ and the limiting $`m_k`$, additional noise sources are introduced. In particular, coherence losses occur due to fluctuations in the fringe position during integration, which induce in turn a visibility reduction by a factor $`\eta =\mathrm{e}^{\sigma _{\mathrm{\Delta }\varphi }^2}`$ (where $`\sigma _{\mathrm{\Delta }\varphi }`$ is a measure of jitter in the referenced phase. See for example Colavita, 1994). This in turn limits the achievable $`SNR`$ in a given $`t_c`$, thus contributing to increase $`\sigma _{\mathrm{ph}}`$. These time-dependent effects can be divided into two classes, namely instrument-specific errors in the determination of the phase, and those that are due one more time to atmospheric propagation effects. The dominant effect (Quirrenbach et al., 1994) is again induced by DCR. For large values of $`z_0`$, $`\eta 0`$, thus applications of this technique are likely to be restricted to moderate zenith angles, depending on wavelength and seeing.
#### 3.2.3 Astrophysical Noise
While the abovementioned error sources can be to some extent dealt with and reduced, in order to progress toward the goal of a few $`\mu `$as precision, astrophysical noise sources (due to the environment or intrinsic to the target) cannot be easily minimized, if present.
For example, I show in figure Astrometric Methods and Instrumentation to Identify and Characterize Extrasolar Planets: A Review the comparison between the gravitational perturbations (as seen along one of the two axes in the plane of the sky as a function of time) induced by a 5 $`M_J`$ and a 0.1 $`M_{}`$ companion to a 1 $`M_{}`$ star with periods of 5 yr and 100 yr, respectively, with the system at a distance $`D=100`$ pc from the Sun. As one can see from the right panel of the Figure, on a time-scale short compared to the orbital period of the stellar companion, the astrometric signature of the planet is superposed to a large effect. The additional signal in the data could be easily misinterpreted as an extra proper motion component or as a significant acceleration, depending on the orbital characteristics of the long-period companion and epochs of observation.
As I mentioned earlier, if the orbital characteristics of the perturbing stellar companion around a target and/or reference star(s) are known to exist in advance, they can be modelled within the context of the observable model. The dynamical effect of a previously unknown stellar companion constitutes otherwise a significant source of noise that might hamper the reliability of orbit reconstruction for a newly detected planet.
Such a problem will affect primarily future high-precision space-borne global astrometric missions such as $`Gaia`$, which will not have the luxury to pre-select the list of targets. In the case of relative astrometry, if feasible, one could require that the objects composing the local frame of reference not be orbited by stellar companions inducing unmodelled signatures larger than a few $`\mu `$as. A typical strategy to achieve this, adopted for example for the selection of the grid stars for $`SIM`$, is to look for reference objects that are K giants, and pre-select them on the basis of medium-precision radial-velocity monitoring. In this case, the typically large distance of these stars (1 kpc) implies (provided they are not too faint, otherwise photon noise becomes an issue) a significant suppression of any astrometric signature that might significantly pollute the potential planetary signal from the target (e.g., Gould, 2001, and references therein).
Another source of astrophysical noise due to the environment is the presence of a circumstellar disk. The motion of the center of mass of the disk, provoked by the excitation of spiral density waves by an embedded planet, induces an additional wobble in the stellar position, while time-variable, asymmetric starlight scattering by the disk can introduce shifts in the photocenter position.
Takeuchi et al. (2005) have recently studied these effects assuming Jupiter-mass planets embedded in gravitationally stable circumstellar disks around young solar-type stars at the distance of the Taurus-Auriga star-forming region ($`D140`$ pc). They conclude that the additional stellar motion caused dynamically by the disk’s gravity is negligible (sub-$`\mu `$as) with respect to the signature from the planet ($`36`$ $`\mu `$as if the planet’s semi-major axis is 5 AU). Variable disk illumination can induce peak-to-peak photocenter variations of order 10-100 $`\mu `$as, but they claim that $`SIM`$ would not be sensitive to such excursions. Finally, Boss (1998) and Rice et al. () have quantified the magnitude of the astrometric displacement induced dynamically by a marginally unstable disk. They found that in this case the effect can be as large as $`50100`$ $`\mu `$as, but the typical time-scale of this perturbation would be of order of decades, as compared to a few years of observations with $`SIM`$ or $`Gaia`$, thus such source of astrometric noise should not constitute a major cause of concern.
The last important class of astrophysical noise sources that can cause shifts in the observed photocenter is not due to stellar environment but rather intrinsic to the target. Such noise sources include a variety of surface temperature inhomogeneities such as spots and flares, and just like disks, they are primarily characteristic of rapidly rotating, young stellar objects (e.g., Bouvier et al., 1995; Schuessler et al., 1996, and references therein).
In the context of a study of the effects of the variety of astrophysical sources of astrometric noise on the planet detection capabilities of $`Gaia`$ (Sozzetti et al., in preparation), I have implemented a numerical model to calculate the photometric and astrometric effects of a distribution of spots over the surface of a rotating star. The model is based on the analytical theories developed by Dorren (1987), Eker (1994), and Eaton et al. (1996). It incorporates a broad range of spot and star parameters, including stellar limb-darkening, and it allows for the presence of multiple spots of any shape, including umbra-penumbra effects.
The key result of the numerical analysis (Figure Astrometric Methods and Instrumentation to Identify and Characterize Extrasolar Planets: A Review) is that a photometric variation in the visual of $`\mathrm{\Delta }F/F=10\%`$ (rms) corresponds to an astrometric variation of $`3`$ $`\mu `$as (rms) in the position of a 1 $`R_{}`$ pre-main sequence star at the distance of Taurus ($`D=140`$ pc). The magnitude of the spot-induced photocenter motion on a T Tauri star is thus comparable to the gravitational effect of a Jupiter-mass object orbiting the star at 0.5 AU ($`5`$ $`\mu `$as). The effect scales with distance just like the astrometric signature of a planet, thus for a nearby, less active sun-like star at 10 pc its magnitude could still be of the same order (e.g., Woolf & Angel, 1998), while the amplitude of the planet-induced stellar motion would be at least one order of magnitude larger.
However, astrometric signatures induced for example by a few Earth-mass planet on a 1 AU orbit around a solar-type star at a distance of 10 pc covered by spots causing a change in photospheric flux of $`1\%`$ could be comparable in size. The non-uniformity of illumination of the stellar disk might then jeopardize successful Earth-sized planet detection with astrometry around the nearest stars, as well the detection of Jupiter-sized objects in nearby star-forming regions. Fortunately, large spotted areas on solar-type stars are relatively uncommon (the Sun itself, at its peak of activity, is covered by spots for up to $`0.1`$% of its visible surface. See for example Allen, 2000). Furthermore, the spot-induced photocenter variation has a period that is strongly correlated to the photometric excursion as well as the stellar rotation period (of order of a few days for T Tauri stars, up to several weeks for solar-type objects). With the help of careful photometric monitoring, the two sources of astrometric signal might then be successfully disentangled.
Ultimately, in order to keep environmental and intrinsic astrophysical noise sources at the few $`\mu `$as level, an important cause of concern primarily for giant planet searches in star-forming regions, it would be beneficial to avoid stars with large photometric variations and objects with particularly large, flared disks.
### 3.3 Estimation Process
The estimation process applies the observable model and noise model to the data. The estimation process includes several functions, such as search techniques, hypothesis testing, and parameter estimation. The observable model provides the estimation process with a parametric description of the expected data. The estimation process finds the observable model parameters that best match the data, with deviations weighted by the noise model. The estimation process may be a generalized least-squares method that takes advantage of the full noise covariance matrix constructed from the noise model, as I briefly describe below.
Suppose we have performed $`n`$ measurements of the quantity $`y`$ collected in the vector Y$`(y_1,y_2,\mathrm{},y_n)`$, with associated measurement uncertainties $`𝚺(\sigma _1,\sigma _2,\mathrm{},\sigma _n)`$. Furthermore, call X$`(x_1,x_2,\mathrm{},x_p)`$ the vector of $`p`$ unknown quantities that we want to determine. Let F(X) be the actual functional form of the observable model. The method of least squares will attempt to find a solution to the equation $`𝐘+𝚺=𝐅(𝐱)`$ in terms of the unknowns in the model.
Under the assumptions that the unknowns are normally distributed and are sufficiently small, the set of equations can be expanded to first order in the unknowns. The resulting system of equations of condition can be written as:
$$𝚫𝐘=𝐘𝐅(𝐗)=𝖣\delta +𝚺,$$
(16)
with $`\delta (\delta x_1,\delta x_2,\mathrm{},\delta x_n)`$ the vector of new unknowns, and $`𝖣`$ the design matrix containing all partial derivatives of the observable model with respect to the unknowns. In order for the method to be applicable, the number of equations of condition must be larger than the number of unknowns, usually at least $`n>2p`$.
The objective of the least squares technique is to determine the vector $`\delta `$ that minimizes the sum of the squares of the components of the vector of uncertainties $`𝚺`$:
$$\underset{i=1}{\overset{n}{}}\sigma _i^2=𝚺^\mathrm{T}𝚺=(𝚫𝐘𝖣\delta )^\mathrm{T}(𝚫𝐘𝖣\delta ),$$
(17)
where the superscript T indicates transposed. The solution is given by:
$$\delta _\mathrm{𝟎}=(𝖣^\mathrm{T}𝖣)^1𝖣^\mathrm{T}𝚫𝐘$$
(18)
In a weighted least squares solution, Eq. 16 is multiplied on both sides by a square $`n\times n`$ matrix $`𝐆`$ containing zeroes except for the elements on the main diagonal, which are $`g_i=\sigma _i^1`$. The formal solution now becomes:
$$\delta _\mathrm{𝟎}=(𝖣^\mathrm{T}\mathrm{𝖶𝖣})^1𝖣^\mathrm{T}𝖶𝚫𝐘=\mathrm{𝖢𝖣}^\mathrm{T}𝖶𝚫𝐘,$$
(19)
with the weight matrix $`𝖶=𝖦^\mathrm{T}𝖦`$, and the covariance matrix of the solution $`𝖢=(𝖣^\mathrm{T}\mathrm{𝖶𝖣})^1`$. The formal standard deviations of any unknown $`\delta x_i`$ in the solution will be given by the square-root of the diagonal terms of the covariance matrix $`c_i`$.
The weight matrix maybe generalized in the case where correlations between the unknowns are present. In this case, non diagonal terms will not be zero. This method is probably superior to a more conventional weighted-least-squares technique, in which difference observations are formed to cancel correlated errors, and each difference is assigned a weight commensurate with the expected uncorrelated portion of the error.
The latter approach fails to take full advantage of the knowledge of the correlations, and residual correlated errors between successive differences persist. With a more rigorous approach, not only one can take into account any correlations that might exist, regardless of their temporal or angular scale and dependence on other observing parameters, but the correlation itself becomes part of the solution, and the self-consistency of the solution can be determined to establish to what extent the noise model is satisfactory.
If the functional form $`𝐅(𝐗)`$ is non-linear, as is often the case, multiple iterations of the linearized least squares procedure must be carried out. In this case, the quality of the solution ultimately obtained, i.e. how close the minimum of the functional form adopted for the observable model is to the real one, will strongly depend on the point at which the equations are linearized. Good starting guesses of the parameters of the model would be highly desirable in order to favor the convergence of the iterative procedure. To this end, several techniques could be adopted, such as local and global minimization strategies, including the simplex method, simulated annealing, or genetic algorithms (e.g., Press et al., 1992).
Once a solution for the vector of parameters $`\delta _\mathrm{𝟎}`$ is obtained, it is necessary to assess whether the observable model employed is indeed representative of the reality. To this end, a number of tests can be conducted.
The general procedure consists in defining a test statistic ($`\chi ^2`$ or its transformation $`F2`$, Fisher’s $`F`$, Kolmogorov-Smirnov’s $`D`$) which is some function of the data measuring the distance between the hypothesis and the data, and then calculating the probability of obtaining data which have a still larger value of this test statistic than the value observed, assuming the hypothesis is true. This probability is called the confidence level. Small probabilities (say, less than one percent) indicate a poor fit. Especially high probabilities (close to one) correspond to a fit which is too good to happen very often, and may indicate a mistake in the way the test was applied, such as treating data as independent when they are correlated.
For the purpose of planet detection, upon rejection of an observable model which assumes the star is single by means of a goodness-of-fit test, observations residuals should be inspected for the presence of hidden periodicities in the measurements. A possible approach is as follows.
First, a period search is conducted. To this end, the Lomb-Scargle periodogram analysis could be performed (e.g., Lomb, 1976; Scargle, 1982; Horne & Baliunas, 1986). Alternatively, the observable model could be expanded to incorporate a perfectly sinusoidal term (circular orbit), and a star+circular orbit fit performed adopting a dense grid of trial periods. It should then be possible to test if the new observable model (star+circular orbit) provides a significant improvement with respect to the single-star model. For example, a statistical test of the goodness-of-fit of the single-star and star+circular orbit model could be adopted, such as the likelihood-ratio test.
If the star+circular orbit fit performs significantly better in a measurable way, the observable model is then further expanded, to include a full Keplerian orbit. In the presence of multiple planetary signals, the procedure is carried out until not further periodicities can be uncovered and the observation residuals are fully consistent with noise.
### 3.4 Ground-Based Experiments
The confirmation of theoretical predictions that astrometry with ground-based monopupil telescopes is limited to the mas precision regime has been given by numerous experiments. Gatewood (1987) derived $`\sigma _{\mathrm{atm}}3`$ mas/h with a reference frame of $`10^{}20^{}`$. Han (1989) showed that a 1 mas/h precision can be reached for a double star with $`\vartheta =1^{}`$. Similar conclusions were derived by Gatewood (1991) and Dekany et al. (1994).
If instead of a single reference star, a dense star field is used as a frame of reference, the situation improves somewhat. The theoretical predictions of $`\sigma _{\mathrm{atm}}A^1\vartheta ^{4/3}t^{1/2}`$ (Lindegren, 1980) and $`\sigma _{\mathrm{atm}}A^{3/2}\vartheta ^{11/6}t^{1/2}`$ (Lazorenko, 2002) have been substantially confirmed by Pravdo & Shaklan (1996), who demonstrated $`\sigma _{\mathrm{atm}}150`$ $`\mu `$as/h with the 5-m Palomar telescope using 15 reference objects in a field of 90 arcsec. These authors showed also that $`\sigma _{\mathrm{DCR}}60100`$ $`\mu `$as within 1 hour of meridian crossing and at declinations within $`45^{}`$ of the zenith.
It is worth noting that, assuming apodization of the entrance pupil and enhanced symmetrization of the reference field, achieved assigning a specific weight to each reference star, Lazorenko & Lazorenko (2004) have recently generalized the theoretical expression for the astrometric error due to the atmosphere:
$$\sigma _{\mathrm{atm}}A^{k/2+1/3}\vartheta ^{k\mu /2}t^{1/2},$$
(20)
with $`2k\sqrt{8N_r+1}1`$, limited by the number $`N_r`$ of reference objects, and $`\mu 1`$ a term dependent on $`k`$ and the magnitude and distribution on the sky of the field stars. The classic result by Lindegren (1980) is recovered for $`N_r=1`$. However, the Lazorenko & Lazorenko (2004) expression predicts $`\sigma _{\mathrm{atm}}1060`$ $`\mu `$as (depending on stellar field density) for a 10-m telescope, very good seeing (FWHM = 0.4 arcsec), and $`t=600`$ sec. This is about a factor 2-5 improvement with respect to the prediction of Eq. 15, which would have the goal of $`\sigma _{\mathrm{atm}}60`$ $`\mu `$as reached in $`t=1`$ hr, for $`A=10`$ meters. The improvement due to this new approach to the astrometric measurement process (which however neglects DCR effects) still awaits experimental confirmation.
The promise of long-baseline optical/infrared interferometry for high-precision astrometry has been tested by a number of experiments in the past. The Mark-III and NPOI interferometers have achieved long-term wide-angle astrometric precision at the 10 mas level (Hummel et al., 1994). Short-term astrometric performance at the 100 $`\mu `$as level has been demonstrated with Mark-III and PTI (Colavita, 1994; Shao et al., 1999; Lane et al., 2000), for moderately close (30 arcsec) pairs of bright ($`m_v25`$) stars. Recently, Lane & Muterspaugh (2004) have demonstrated that 10-$`\mu `$as short-term very narrow-angle astrometry is possible for a sample of close, sub-arcsec binaries observed with PTI in phase-referencing mode.
The predicted astrometric performances of Keck-I (Boden et al., 1999) and VLTI (Derie et al, 2003) will presumably reach the actual limits of this technique from the ground (unless such an instrument is built at the South Pole). The two instruments have quoted limiting magnitudes in the near-infrared (2-2.4 $`\mu `$m) of $`m_k1718`$ for narrow-angle astrometry at the 30 $`\mu `$as and 10-20 $`\mu `$as level, respectively, between pairs of objects separated by $`<2030`$ arcsec. <sup>6</sup><sup>6</sup>6For astrometric planet searches conducted with these instruments, the probability of finding a reference star with $`m_k=13`$, or fainter within 20-30 arcsec from a target object is about 50%-60%, or greater, if the target is not very far from the galactic plane (Derie et al, 2003).
### 3.5 Space-Borne Experiments
Relative, narrow-angle astrometry from space has been performed so far with the Fine Guidance Sensors aboard $`HST`$, while global astrometric measurements have been carried out for the first time by $`Hipparcos`$.
For $`HST`$/FGS, the data reduction of the two-dimensional interferometric measurements entails a number of ad hoc calibration and data reduction procedures to remove a variety of random and systematic error sources from the astrometric reference frame (e.g., Taff, 1990; Bradley et al., 1991; Benedict et al., 1994, 1999, and references therein). The calibration of random and systematic, long- and short-term error sources for $`HST`$/FGS includes removing intra-observation spacecraft jitter, compensating for temperature variations and temperature-induced changes in the secondary mirror position, applying constant and time-dependent optical field angle distortion calibrations, correcting for intra-orbit drift, and applying lateral color corrections during the orbit-to-orbit astrometric modeling (e.g., Benedict et al., 1994, 1999).
The global single-measurement error budget for $`HST`$/FGS astrometry with respect to a set of reference objects near the target (within the $`5\times 5`$ arcsec instantaneous field of view of FGS) had received a pre-launch estimate of $`2.7`$ mas by Bahcall & O’Dell (1980). Benedict et al. (1994, 1999) confirmed the overall performance level of the instrument, with single-measurement uncertainties of 1-2 mas down to $`m_v=16`$. The limiting factor is the spacecraft jitter. A single-measurement precision below 0.5-1 mas is out of reach for $`HST`$/FGS.
For $`Hipparcos`$, a calibration and iterative reduction procedure in five main steps (Lindegren & Kovalevsky, 1989) is devised in order to derive values of positions, proper motions, and parallaxes simultaneously for $`120,000`$ stars by combining one-dimensional angular measurements along the satellite’s instantaneous scanning direction into a global astrometric solution over the whole celestial sphere. These steps include: 1) the determination of the satellite attitude; 2) the estimation of stellar coordinates relative to the main focal grid; 3) the reference great circle (RGC) reduction, to determine the abscissae of stars on each RGC; 4) the sphere solution, to determine the correction to the great circle origins using a set of instrumental calibration parameters (including chromatic terms); 5) the determination of the five astrometric parameters with respect to the rigid reference sphere of RGCs, using all RGC abscissae, the RGC origins, and instrumental parameters (Lindegren & Kovalevsky, 1989).
Typical uncertainties on the RGC abscissae are of order of 1.0 mas for bright objects ($`m_v<7`$), and degrade up to $`4.5`$ mas for $`m_v11`$ (e.g., Kovalevsky, 2002). <sup>7</sup><sup>7</sup>7The errors on the final astrometric parameters are not only a function of $`m_v`$, but also of the ecliptic latitude $`\beta `$, a normal consequence of the adopted scanning law (stars at low latitudes were observed significantly less often). These agree well with pre-launch predictions by Lindegren (1989). Without the presence of the atmosphere, and similarly to $`HST`$/FGS, the best-achievable single-measurement precision is limited by the uncertainties in the determination of the along-scan attitude.
The ability to suppress systematics by at least two orders of magnitude for a space-borne instrument is a major technological goal. Both $`SIM`$ and $`Gaia`$ promise to achieve this level of astrometric precision. For the purpose of planet detection with $`SIM`$, in order to deliver 1 $`\mu `$as narrow-angle astrometry in 1 hr integration time down to $`m_v1112`$, an accuracy on the position of the delay line of 50 pm with a 10 m baseline must be achieved (Shaklan et al., 1998). Furthermore, a positional stability of internal optical pathlengths of $`10`$ nm is required, in order to ensure maintenance of the fringe visibility (Neat et al., 1998). For a $`Gaia`$-like instrument, the success in meeting the goal of $`510`$ $`\mu `$as single-measurement astrometric precision to hunt for planets around bright stars ($`m_v<1112`$) will depend on a) the ability to attain CCD centroiding errors not greater than 1/1000 of a pixel in the along-scan direction (Gai et al., 2001) and b) the capability to limit instrumental uncertainties (thermo-mechanical stability of telescope and focal plane assembly, metrology errors in the monitoring of the basic angle) and calibration errors (chromaticity, satellite attitude, focal plane-to-field coordinates transformation) down to the few $`\mu `$as level (e.g., Perryman et al., 2001).
## 4 Planet Detection with Astrometry: Past and Present Efforts
Astronomers have long sought to find astrometric perturbations in a star’s motion due to orbiting planet-sized companions. Many attempts have failed, some have produced more or less significant upper limits, a few have been successful. I review in turn the history of these efforts.
### 4.1 Unfinished Tales: Barnard’s Star and Lalande 21185
During the 1960s, based on the analysis of over 2000 photographic plates of the Sproul Observatory covering 24 years (1938-1962) van de Kamp (1963, , ) announced the discovery of perturbations in the motion of Barnard’s Star (GJ 699) that could be explained initially with the presence of a 1.6-1.7 $`M_J`$ planet on a 24-25 years, eccentric orbit, and then instead in terms of two Jupiter-sized objects on coplanar, circular orbits with periods of 11.5 and 22 years, respectively. Through the years, van de Kamp refined his results, extending the time duration of the photographic plate observations up to 43 years (1938-1981), and publishing two more papers (van de Kamp, 1975, 1982). In his last interview on the subject (Schilling, 1985), he still claimed that Barnard’s Star was orbited by two massive planets of 0.7 $`M_J`$ and 0.5 $`M_J`$, co-revolving on circular, coplanar orbits with periods of 12 and 20 years, respectively.
Neither planet was ever confirmed, however. Initial claims by Jensen & Ulrych (1973), that observations were compatible with the presence of up to five planets, were not verified by Gatewood & Eichhorn (1973), who could not detect any additional motion perturbing Barnard’s Star. The existence of giant planets orbiting the star was further cast in doubt by Hershey (1973), and Heintz (1976), who explained van de Kamp’s results in terms of a number of unrecognized systematics, including telescope internal motions due to two phases of cleaning and remounting of the telescope lens 25 years after he began his observations. Years later Frederick & Ianna (1980), Harrington (1986), and Harrington & Harrington (1987) reported other independent studies of Barnard’s star, in which no wobble was detected, although van de Kamp’s results were not totally discounted. Croswell (1988), on the other hand, addressed again the issue of the misinterpretation of incorrect Sproul Observatory data, concluding that unknown telescope systematics were the more likely explanation.
In a more recent study, Gatewood (1995) ruled out the presence of massive planets or brown dwarfs ($`M_p>10`$ $`M_J`$) around Barnard’s star, while no conclusion was reached on the existence of objects of order of the mass of Jupiter or smaller. Using $`HST`$/FGS astrometry, Benedict et al. (1999) ruled out the presence of Jupiter-mass planets with orbital periods $`P<3`$ years, but their observations were not taken for a sufficiently long amount of time to address the period range of the putative planets discovered by van de Kamp. Schroeder et al. (2000) conducted a photometric study of Barnard’s Star, and did not find any supporting evidence for the presence of massive planets and brown dwarfs at large orbital radii, in agreement with Gatewood’s (1995) findings. The latest study of this star has been undertaken by Kürster et al. (2003) using precision radial-velocity measurements, which allowed them to rule out the presence of planets down to the terrestrial mass regime (a few $`M_{}`$) for objects within 1 AU. Although studies of Barnard’s Star have spanned over half a century, no definitive confirmation or disproval has been established.
The possible existence of a giant planet companion (at least several Jupiter masses) to Lalande 21185 (HD 95735) was first discussed by Lippincott (Lippincott 1960a, , ) on the basis of photographic plates covering a time-span of 47 years taken with the Sproul telescope. Gatewood (1974) did not find any evidence of a planetary signature at the suggested 8-yr period, but Hershey & Lippincott (1982) claimed the planetary mass companion did exist, although on a different, longer period. Based on a limited dataset covering four years, Gatewood et al. (1992) were not able to detect any significant perturbation in the star’s proper motion. However, four years later Gatewood (1996), in examining 50 years or radial-velocity data of Lalande 21185, as well as a more sophisticated set of astrometric observations, first concluded that the star is indeed orbited by a 2.0 $`M_J`$ planet at $`10`$ AU, and then suggested the existence of two giant planets, the second body being less massive than Jupiter and orbiting at around 3 AU from the parent star. Also in this case, an independent confirmation has yet to be made on either of the two planets.
### 4.2 Upper Limits and Controversial Mass Determinations
Prompted by the success of Doppler surveys for giant planets of nearby stars and by the need to find a method to break the $`M_pi`$ degeneracy intrinsic to radial-velocity measurements, several authors have re-analyzed in recent years the $`Hipparcos`$ Intermediate Astrometric Data (IAD), in order to either detect the planet-induced stellar astrometric motion of the bright hosts, most of which had been observed by the satellite, or place upper limits to the magnitude of the perturbation, in the case of no detections. The $`Hipparcos`$ IAD have been re-processed alone, or in combination with either the spectroscopic information or with additional ground-based astrometric measurements.
The first such analysis was performed by Perryman et al. (1996), who failed to detect the astrometric motion of the first three planet-bearing stars announced, 51 Peg, 47 Uma, and 70 Vir. Based on the size of the astrometric residuals to a single-star fit, they report upper limits on companions masses in the sub-stellar regime (7-65 $`M_J`$, depending on confidence levels) for 47 UMa and 70 Vir, while limits on the companion to 51 Peg, due to its very short period, are less stringent.
Orbital fits to the $`Hipparcos`$ IAD can be performed by using the values of $`P`$, $`\tau `$, $`e`$, and $`\omega `$ derived from spectroscopy and by solving for $`a`$, $`i`$, and $`\mathrm{\Omega }`$, with the additional constraint (Pourbaix & Jorissen, 2000):
$$\frac{a\mathrm{sin}i}{\pi _{}}=\frac{PK\sqrt{1e^2}}{2\pi \times 4.7405}$$
(21)
With this approach, Mazeh et al. (1999) and Zucker & Mazeh (2000) published astrometric orbits for the outermost planet in the $`\upsilon `$ And system and for the planet orbiting HD 10697. The derived semi-major axes of $`1.4\pm 0.6`$ mas and $`2.1\pm 0.7`$ mas, respectively, imply companion masses of $`10.1_{4.6}^{+4.7}`$ $`M_J`$ and $`38\pm 13`$ $`M_J`$, respectively. These values depart significantly from the minimum masses from spectroscopy, as a consequence of the small inclination angles obtained by the fitting procedures ($`i=24^{}`$ and $`i=10^{}`$, respectively).
However, two subsequent studies by Gatewood et al. (2001) and Han et al. (2001) contributed to spark a controversy over the reliability of the determination of sub-stellar companion masses with milli-arcsecond astrometry. In the first work, Gatewood et al. (2001) combined the $`Hipparcos`$ IAD with Multichannel Astrometric Photometer (MAP; Gatewood, 1987) observations of $`\varrho `$ CrB in an astrometric orbital solution that yielded a semi-major axis of $`1.66\pm 0.35`$ mas, an inclination of $`0^{}.5`$, and a derived companion mass of $`0.14\pm 0.05`$ $`M_{}`$. The follow-up paper by Han et al. (2001) presented $`Hipparcos`$-based preliminary astrometric masses for 30 stars with at least one spectroscopically detected giant planet. The main conclusion of this work is that a significant fraction ($`40\%`$) of the planet candidates are instead stars, and the remainder sub-stellar companions are in most cases brown dwarfs rather than planets. The results stem from the derivation of a vast majority of quasi-face-on orbits, with 60% of the sample having $`i<5^{}`$ and $`27\%`$ having $`i<1^{}`$.
On the one hand, if orbits are isotropically oriented in space, the probability of finding one with $`i<1^{}`$ is $`1\times 10^4`$, thus Han et al. (2001) come to the conclusion that the sample of planet-bearing stars is severely biased towards small inclination angles. On the other hand, rather than having to reject the planet hypothesis for a substantial fraction of the Doppler candidates, the systematically very small inclination angles, and thus very large actual companion masses, could arise as an artifact of the fitting procedure. This thesis was indeed put forth by Pourbaix (2001) and Pourbaix & Arenou (2001) (and later by Zucker & Mazeh ()). Using different statistical approaches aimed at assessing the robustness of the derived $`Hipparcos`$ astrometric orbits, these authors demonstrated that the $`Hipparcos`$ IAD do not have enough precision to actually reject the planet hypothesis in essentially all cases (although a few border line cases do exist). Thus, essentially all the preliminary astrometric masses derived for stars with planets observed with $`Hipparcos`$ (Mazeh et al., 1999; Zucker & Mazeh, 2000; Gatewood et al., 2001; Han et al., 2001) do not survive close statistical scrutiny.
The $`Hipparcos`$ IAD can still be used however to put upper limits on the size of the astrometric perturbations, as done by Perryman et al. (1996) and by Zucker & Mazeh (), who could rule out at the $`2\sigma `$ level the hypothesis of low-mass stellar companions disguised as planets for over two dozen objects. This, combined with the fact that the same analysis of $`Hipparcos`$ data reveals that instead a significant fraction of the proposed brown dwarf companions from spectroscopy is stellar in nature, is interpreted as further evidence of the existence of the brown dwarf desert that separates stellar and planetary mass secondaries.
In the end, the only firm upper limits on the mass of a spectroscopically detected extrasolar planet are those placed by McGrath et al. (2002, 2004) who failed to reveal astrometric motion of the $`M_p\mathrm{sin}i=0.88`$ $`M_J`$ object on a 14.65-day orbit in the $`\varrho ^1`$ Cnc multiple-planet system using $`HST`$/FGS astrometry. With a nominal single-measurement precision of 0.5 mas, the failed attempt at detecting any reflex motion in the data implies that the 1.15 mas preliminary $`Hipparcos`$-based mass estimate by Han et al. (2001) is ruled out at the 3-5 $`\sigma `$ level, thus establishing an updated mass upper limit of $`30`$ $`M_J`$ and firmly confirming that the object is sub-stellar in nature.
### 4.3 Actual Measurements and Work in Progress
It was not until two years after the first confirmation of the planetary nature of the companion to HD 209458 via detection of its transits across the disk of the parent star (Charbonneau et al., 2000; Henry et al., 2000) that astrometric techniques finally provided the first undisputed value of the actual mass of a Doppler-detected planet. Narrow-field relative astrometry of the multiple-planet host star GJ 876 was carried out by Benedict et al. (2002) using $`HST`$/FGS.
The goal of this project was to determine the astrometric wobble induced on the parent star by the outer planet. At a distance $`D=4.7`$ pc, and with a nominal primary mass of the M4 dwarf star $`M_{}=0.32`$ $`M_{}`$, the planet with a projected mass $`M_p\mathrm{sin}i2`$ $`M_J`$ on a $`P=60`$ days orbit was predicted to produce a minimum gravitational perturbation of $`270`$ $`\mu `$as, which was deemed detectable by the typical 0.5 mas single-measurement precision of $`HST`$/FGS. Benedict et al. (2002) utilized five reference stars within a few arc-minutes from the target, and derived the perturbation size, inclination angle, and mass of GJ 876b from a combined fit to the available astrometry and spectroscopy. They found $`\alpha =250\pm 60`$ $`\mu `$as, $`i=84^{}\pm 6^{}`$, and $`M_p=1.89\pm 0.34`$ $`M_J`$.
In the recent announcement (McArthur et al., 2004) of the discovery of a Neptune-sized planet on a 2.8 days orbit in the $`\varrho ^1`$ Cnc system (which brought the number of planets in the systems to a total of four), $`HST`$/FGS astrometry played again an important role. The authors in fact re-analyzed the available data on $`\varrho ^1`$ Cnc which had allowed McGrath et al. (2002, 2004) to put stringent upper limits on the mass of the 14.65-day period planet, and estimated, from the small arc of the orbit covered in the limited $`HST`$ dataset, a perturbation size ($`1.94\pm 0.4`$ mas) and inclination ($`53^{}\pm 6^{}.8`$) for the outermost planet, orbiting at $`5.9`$ AU. Under the assumption of perfect coplanarity of all planets in the system, this implies an actual mass for the innermost planet of $`17.7\pm 5.57`$ $`M_{}`$.
Currently, Benedict et al. (, , 2004) are monitoring with $`HST`$/FGS the stars $`\upsilon `$ And and $`\epsilon `$ Eri, and plan to combine the data with the available radial-velocity datasets and with lower-per-measurement precision ground-based astrometry. The predicted minimum perturbation sizes of the long-period (3.51 yr and 6.85 yr, respectively) planets orbiting these stars ($`\alpha _{\upsilon \mathrm{And}}540`$ $`\mu `$as and $`\alpha _{\epsilon \mathrm{Eri}}1120`$ $`\mu `$as, respectively) should be clearly detectable with $`HST`$/FGS, provided a sufficient time baseline for the observations.
## 5 Future Prospects
A number of authors have tackled the problem of evaluating the sensitivity of the astrometric technique required to detect extrasolar planets and reliably measure their orbital elements and masses. In particular, the works by Casertano et al. (1996), Lattanzi et al. (1997, , , 2002, 2005), and Sozzetti et al. (2000, 2001, ) were specifically tailored to $`Gaia`$, those of Casertano & Sozzetti (1999), Sozzetti et al. (2002, ), Ford & Tremaine (2003), Ford (2004), and Marcy et al. (2005) where instead centered on $`SIM`$. Black & Scargle (1982) and Eisner & Kulkarni (2001, 2002) studied the general problem of the detectability of periodic signals with the astrometric technique alone or in combination with spectroscopic measurements, while Konacki et al. (2002) and Pourbaix (2002) explored to some extent the reliability of orbit reconstruction of future astrometric missions when all parameters have to be derived from scratch, in the limit of high and low signal-to-noise ratios.
The abovementioned exploratory works which provided a first assessment of the planet detection capabilities of $`Gaia`$ and $`SIM`$ adopted a qualitatively correct description of the measurements each mission will carry out. For $`Gaia`$, the then-current scanning law was adopted, while for $`SIM`$ reference stars and realistic observation overheads were included. The authors implemented realistic data analysis techniques based on both the $`\chi ^2`$ test and periodogram search for estimating detection probabilities as well as non-linear least squares fits to the data to determine orbital parameters and planet masses ranging from 1 $`M_J`$ down to 1 $`M_{}`$.
From the point of view of data simulation, the major simplifying assumption of these studies is the idealization of the adopted instrument. Measurement errors assume simple gaussian distributions, and knowledge of the spacecraft attitude is assumed perfect, with no additional instrumental effects, measurement biases, and calibration imperfections. In terms of data analysis procedures, the most relevant simplification is the adoption of perturbations of the true values of all parameters as initial guesses for the non-linear fits, largely neglecting the difficult problem of identifying adequate configurations of starting values from scratch.
### 5.1 Planet Detection
Detection probabilities are determined based on a $`\chi ^2`$ test of the null hypothesis that there is no planet. Five-parameter, single-star fits to the simulated data-sets are carried out, and observation residuals are inspected. Residuals large compared to the assumed single-measurement precision will induce a failure of the $`\chi ^2`$ test, at a given confidence level.
The two parameters upon which detection probabilities mostly depend are the astrometric signal-to-noise ratio $`\alpha /\sigma _m`$ and the period $`P`$, while eccentricity and orientation in the plane of the sky do not significantly affect planet detectability. Figure Astrometric Methods and Instrumentation to Identify and Characterize Extrasolar Planets: A Review shows iso-probability contours for $`SIM`$ as a function of $`\alpha /\sigma _m`$ and $`P`$, based on a $`\chi ^2`$ test with a confidence level of 95%.
For both instruments, assuming a realistic number of data points throughout the nominal mission lifetimes $`T=5`$ years, $`\alpha /\sigma _m2`$ is sufficient to detect planetary signatures for $`PT`$. As orbital sampling gets increasingly worse for $`P>T`$, the required signal rises sharply, especially for high detection probabilities. The same qualitative behavior of generic detection curves was recovered by Eisner & Kulkarni (2001), who also provided analytical expressions for the behavior of the astrometric sensitivity to planetary signatures in the two regimes.
### 5.2 Orbit Reconstruction and Mass Determination
Upon detection of its signal, the goal of determining a planet’s orbital characteristics and mass requires the adoption of observable models with at least 12 parameters (5 astrometric + 7 describing the full Keplerian motion). For $`SIM`$, the model is further complicated by the simultaneous solution for the astrometric parameters of the local frame of reference (5 for each astrometrically clean reference star). The simultaneous fit to both astrometric and orbital parameters strongly reduces the covariance between proper motion and astrometric signature pointed out by Black & Scargle (1982), in particular for $`PT`$.
A standard metric to understand how well the observable model performs on the simulated data is the convergence probability, i.e. the fraction of the final values of each parameter that falls within a given fractional error of the true values. I show in Figure Astrometric Methods and Instrumentation to Identify and Characterize Extrasolar Planets: A Review the $`Gaia`$ convergence probability to 10% fractional uncertainty for $`a`$, $`P`$, $`e`$, and $`i`$ as a function of the distance from the Sun, for a Jupiter-Sun system with $`PT`$, $`PT`$, and $`PT`$.
As a general result, $`\alpha /\sigma _m5`$ is required for orbit reconstruction and mass determination at the 20-30% accuracy level, while $`\alpha /\sigma _m1015`$ is necessary for a more stringent 10% accuracy requirement.
As it can be seen in Figure Astrometric Methods and Instrumentation to Identify and Characterize Extrasolar Planets: A Review, orbital periods twice as long as the mission duration induce significant degradation in the quality of the orbit reconstruction, although different parameters are affected differently. For example, the correct period of the signal is more easily identified as it is independent of the Keplerian nature of the problem (e.g., Monet, 1979). Short periods also cause a degradation of the results, due to the increasingly smaller amplitude of the perturbation, an effect that overruns the increasingly larger number of orbital revolutions sampled during the mission duration.
Orbital eccentricity also plays a very significant role, when attempting to obtain an orbital solution. The deterioration of orbit determination is especially prominent for long-period planets, where the limited orbital sampling couples with large values of $`e`$, with the result that orbits are increasingly more unlikely to be sampled during pericenter passage, and the correct orbit size and geometry become very difficult to identify correctly.
On the other hand, the inclination of the orbital plane does not impact very significantly the ability to accurately determine the orbital parameters and mass of a planet, unless $`i90^{}`$. In quasi edge-on configurations, in fact, the projected stellar motion is reduced to one dimension, and a considerable amount of information is lost. However, this effect is already negligible for configurations departing from exactly edge-on by a few degrees (Eisner & Kulkarni, 2002; Ford, 2004).
Finally, unlike $`Gaia`$, $`SIM`$ will have the leisure to choose the number $`N_o`$ and timing of the observations as well as the number $`N_r`$ of reference objects. Both detection probabilities and the quality of orbit determination are sensitive to these parameters with simple parameterizations given by $`\sqrt{N_o}`$ and $`\sqrt{N_r}`$ (Sozzetti et al., 2002). Ford (2004) has studied in detail a wide range of possible observing schedules, and concluded that both planet detection and orbit reconstruction are relatively insensitive to the specific choice of the distribution of observations.
### 5.3 Multiple-Planet Systems
The limiting ability to detect and characterize planetary systems with $`\mu `$as astrometry has been estimated by Sozzetti et al. (2001, ), utilizing as test-cases the then-current lists of multiple-planet systems discovered by Doppler surveys. In their works, the authors neglected any complications deriving from significant perturbations of the planetary orbits due to strong planet-planet secular or resonant dynamical interactions.
Under the assumption of sufficient data redundancy with respect to the number of parameters in the observable model fitted to the observations<sup>8</sup><sup>8</sup>8If $`N_{\mathrm{pl}}`$ is the number of planets in the system, then at least $`N_o>2\times (5+7\times N_{\mathrm{pl}})`$ for $`Gaia`$, and $`N_o>2\times (5+7\times N_{\mathrm{pl}}+5\times N_r)`$ for $`SIM`$ is required, the detection of additional components in a system will be reliably carried out. Only border-line cases, in which a signal with $`\alpha /\sigma _m1`$ is not properly modelled and subtracted, will produce a significant increase in the false detection rates. For such cases, and in the limit for $`PT`$, a period search would add robustness to the detection method, while the least squares technique combined with Fourier analysis would arguably be preferred when attempting to detect signals with $`P>T`$.
The typical accuracy of multiple-planet orbit reconstruction and mass determination will be degraded by 30-40% with respect to the single-planet case, a relatively modest deterioration particularly for well-sampled, well-spaced orbits with $`\alpha /\sigma _m10`$.
The ability of astrometry to determine the full set of orbital parameters implies that for favorable multiple-planet configurations it should be possible to derive a meaningful estimate of the relative inclination angle (e.g., Kells et al., 1942):
$$\mathrm{cos}i_{\mathrm{rel}}=\mathrm{cos}i_{\mathrm{in}}\mathrm{cos}i_{\mathrm{out}}+\mathrm{sin}i_{\mathrm{in}}\mathrm{sin}i_{\mathrm{out}}\mathrm{cos}(\mathrm{\Omega }_{\mathrm{out}}\mathrm{\Omega }_{\mathrm{in}}),$$
(22)
where $`i_{\mathrm{in}}`$ and $`i_{\mathrm{out}}`$, $`\mathrm{\Omega }_{\mathrm{in}}`$ and $`\mathrm{\Omega }_{\mathrm{out}}`$ are the inclinations and lines of nodes of the inner and outer planet, respectively.
I show in Figure Astrometric Methods and Instrumentation to Identify and Characterize Extrasolar Planets: A Review the estimated accuracy with which $`SIM`$ could determine the coplanarity (i.e., $`i_{\mathrm{rel}}0.0`$) between pairs of planetary orbits as a function of the common inclination angle, for 11 known multiple-planet systems.
For configurations in which all components produce $`\alpha /\sigma _m10`$, coplanarity could be established with typical uncertainties of a few degrees, for periods up to twice the mission duration. In systems where at least one component has $`\alpha /\sigma _m1`$, accurate coplanarity measurements are compromised, and mutual inclinations can only be determined with uncertainties of several tens of degrees.
Finally, if combined radial velocity + astrometric solutions were to be carried out on single- or multiple-planet systems, the quality of orbit reconstruction and mass determination would be significantly improved, especially in the long period regime ($`P>T`$) and for edge-on configurations, while well-sampled, well-measured orbits ($`PT`$, $`\alpha /\sigma _m1`$) would be only marginally improved by radial velocity + astrometric solutions (Eisner & Kulkarni, 2002; Sozzetti et al., 2003a ).
### 5.4 The Search for Good Starting Values
The convergence of non-linear fitting procedures and the quality of orbital solutions can both be significantly affected by the choice of the starting guesses. In the absence of any kind of a priori information on the actual presence of planets around a given target, all orbital parameters will have to be derived from scratch. The results of, e.g., Han et al. (2001) already provided a word of caution on the reliability of low $`S/N`$ astrometric orbits, even when constraints on some of the parameters are available from spectroscopy. It is thus crucial to investigate new strategies in the fitting procedure to maximize the robustness of the solutions obtained.
Pourbaix (2002) tackled the problem in the context of a work on the precision achievable on the orbital parameters of astrometric binaries from two- and one-dimensional observations, in the case of low $`S/N`$. He proposed a two-dimensional global grid search approach in the ($`e`$, $`\tau `$) space coupled to a guess on $`P`$ by means of a period search technique (e.g., Horne & Baliunas, 1986), while fitting a linearized model in the four Thiele-Innes elements (e.g., Green, 1985).
Konacki et al. (2002) applied a ‘frequency decomposition’ method to simulated $`SIM`$ observations of $`\upsilon `$ And. This approach is based on a Fourier expansion of the Keplerian motion, in which the coefficients of the successive harmonics are functions of all orbital elements. The values of the latter obtained from the linear least squares solution performed with the Fourier expansion are then utilized as starting guesses of a local minimization of the non-linear problem. This method avoids the complications of a global-search approach in several dimensions, which can be computationally very intensive. However, the authors did not attempt to validate their approach in cases departing from the favorable ($`PT`$, $`\alpha /\sigma _m1`$) configuration studied.
The most detailed study on this subject is the one currently carried out by the $`Gaia`$ Planetary Systems Working Group. Lattanzi et al. (2005) have recently presented preliminary results of an on-going, large-scale double-blind test campaign that has been set up in order to provide a realistic assessment of the $`Gaia`$ capabilities in detecting extra-solar planets.
The double-blind test protocol envisions three distinct groups of participants. The Simulators define and generate simulated observations of stars with and without planets with a $`Gaia`$-like satellite; the Solvers define detection tests, with levels of statistical significance of their own choice, and orbital fitting algorithms, using any local, global, or hybrid solution method that they devise is best; the Evaluators compare simulations and solutions and draw a first set of conclusions on the process.
As an illustrative example, Figure Astrometric Methods and Instrumentation to Identify and Characterize Extrasolar Planets: A Review shows one of the results of a simulation of 50,000 stars orbited by a single planet having $`0.2P12`$ yr, and producing $`2\alpha /\sigma _m100`$. The present-day $`Gaia`$ scanning-law is utilized, with a single-measurement precision $`\sigma _m=8`$ $`\mu `$as. The plot shows how the derived periods by one of the Solvers compare to the true simulated ones. The most striking result is the ability to derive very accurate estimates of the period for $`P6`$ yr, for the full range of $`\alpha /\sigma _m`$ and for all possible values of $`0e1`$ and $`0^{}i90^{}`$. For periods exceeding the mission duration by over 20%, it becomes increasingly difficult to identify the correct value of $`P`$. In this case, part of the signal can be absorbed in the stellar proper motion, with the net result that the size and period of the perturbation are systematically underestimated.
However, the preliminary findings by Lattanzi et al. (2005) show that ‘mission-ready’ detection and orbital fit packages (including reliable estimates of the covariance matrix of the solutions) tailored to future high-precision astrometric observatories, requiring no a priori knowledge of the orbital elements, can already achieve good performances.
## 6 Discussion: Astrometry in Perspective
The classic way to gauge the effectiveness of different planet search techniques is to compare their respective discovery spaces, defined in terms of the planets of given mass and period each method will be able to detect. As an illustrative example, I show in Figure Astrometric Methods and Instrumentation to Identify and Characterize Extrasolar Planets: A Review the $`M_p`$-$`P`$ diagram with the plotted present-day sensitivities of transit photometry and radial-velocity, and with the expected $`SIM`$ and $`Gaia`$ detection thresholds at 10 pc and 150 pc, respectively. For the radial-velocity detection curve the simulations by Sozzetti et al. (2005) were utilized, while for $`SIM`$ and $`Gaia`$ the Sozzetti et al. (2002, ) results were used. The sensitivity for transit photometry was derived based on the Gaudi et al. (2005) analytical dependence of the detectable planet radius $`R_p`$ on $`P^{1/6}`$ (converted to $`M_pP^{1/2}`$ assuming constant planet density), under the (naive) hypothesis of uniform sampling.
Simply taking at face value the curves of Figure Astrometric Methods and Instrumentation to Identify and Characterize Extrasolar Planets: A Review, however, can lead to important misunderstandings on the intrinsic relevance of the different techniques to planetary science. For example, the sensitivity of photometric techniques to transiting planets with $`P10`$ days is strongly suppressed, and this detection method is useless if the planet does not transit. However, the information this technique provides for the very close-in objects discovered is extremely valuable, and it cannot be provided by radial-velocity and astrometry.
A more effective way to proceed is then to gauge the relative importance of different planet detection techniques by looking at their discovery potential not per se, but rather in connection to outstanding questions to be addressed and answered in the science of planetary systems, such as those I illustrated at the end of Section 2. I summarize below some of the most important issues for which $`\mu `$as astrometry will play a key role.
### 6.1 The hunt for Other Earths
The holy grail in extra-solar planet science is clearly the direct detection and characterization of Earth-sized, habitable planets, with atmospheres where bio-markers (e.g., Lovelock, 1965; Ford et al. 2001b, ; Des Marais et al., 2002; Selsis et al., 2002; Seager & Ford, 2005) might be found that could give clues on the possible presence of life forms. Imaging terrestrial planets is presently the primarily science goal of the coronagraphic and interferometric configurations of the Terrestrial Planet Finder (TPF; Beichman et al., 2002), and of the Darwin Mission (Fridlund, 2000).
Space-borne transit photometry carried out with $`Corot`$ (Baglin et al., 2002) or $`Kepler`$ (Borucki et al., 2003) has the potential to be the first technique to make such a detection. However, astrometry of all nearby stars within 10-20 pc from the Sun at the $`\mu `$as level (with $`SIM`$ and $`Gaia`$ in space, and possibly with Keck-I and VLTI from the ground) will be an essential ingredient in order to be able to provide Darwin/TPF with a) systems containing bona fide terrestrial, habitable planets (Ford & Tremaine, 2003; Sozzetti et al., 2002; Marcy et al., 2005), and b) a comprehensive database of F-G-K-M stars with and without detected giant planets orbiting out to a few AU from which to choose additional targets based on the presence or absence of Jupiter signposts (Sozzetti et al., 2003a ). Such measurements will uniquely complement ongoing and planned radial-velocity programs aiming at $`1`$ m s<sup>-1</sup> precision (e.g., Santos et al. 2004b, ), and exo-zodiacal dust emission observations from the ground with Keck-I, LBTI, and VLTI.
### 6.2 Statistical Properties and Correlations
As discussed in Sections 2.1 and 2.2, planet properties (orbital elements and mass distributions, and correlations amongst them) and frequencies are likely to depend upon the characteristics of the parent stars (spectral type, age, metallicity, binarity/multiplicity). It is thus desirable to be able to provide as large a database as possible of stars screened for planets.
The size of the stellar sample screened for planets by an all-sky astrometric survey such as $`Gaia`$ (Lattanzi et al., 2000a ) could be of order of a few hundred thousand relatively bright ($`m_v<13`$) stars with a wide range of spectral types, metallicities, and ages out to $`150`$ pc. The sample-size is thus comparable to that of planned space-borne transit surveys, such as $`Corot`$ and $`Kepler`$. The statistical value of such a sample is better understood when one considers that, depending on actual giant planet frequencies as a function of spectral type and orbital distance, at least a few thousand planets could be detected and measured (Lattanzi et al., 2002). This number is comparable to the present-day size of the target lists of ground-based Doppler surveys. Finally, the ranges of orbital parameters and planet host characteristics probed by an all-sky astrometric planet survey would crucially complement both transit observations (which strongly favor short orbital periods and are subject to stringent requisites on favorable orbital alignment), and radial-velocity measurements (which can be less effectively carried out for stars covering a wide range of spectral types, metallicities, and ages and do not allow to determine either the true planet mass or the full three-dimensional orbital geometry).
### 6.3 Tests of Giant Planet Formation and Migration
The competing giant-planet formation models make very different predictions regarding formation time-scales, planet mass ranges, and planet frequency as a function of host star characteristics. Furthermore, correlations between orbital elements and masses, and possibly between the former and some of the host star characteristics (metallicity) might reflect the outcome of a variety of migration processes and their possible dependence on environment (see Sections 2.1 and 2.2). These predictions could be tested on firm statistical grounds by extending planet surveys to large samples of PMS objects and field metal-poor stars.
The full sample of $`1500`$ relatively bright ($`m_v<13`$), nearby ($`D150200`$ pc), field metal-poor stars presently known could be screened for giant planets on wide orbits by $`Gaia`$ or $`SIM`$, thus complementing the shorter-period ground-based spectroscopic surveys (Sozzetti et al., 2005), which are also limited in the sample sizes due to the intrinsic faintness and weakness of the spectral lines of the targets. These data combined would allow for improved understanding of the behavior of the probability of planet formation in the low-metallicity regime, by direct comparison between large samples of metal-poor and metal-rich stars, in turn putting stringent constraints on the proposed planet formation models. Disproving or confirming the existence of the $`P`$-\[Fe/H\] correlation would also help to understand whether metallicity plays a significant role in the migration scenarios for giant planets.
High-precision astrometric measurements of at least a few hundred relatively bright ($`m_v<1314`$) PMS stars in a dozen of nearby ($`D<200`$ pc) star-forming regions could be carried out with $`SIM`$ and $`Gaia`$, searching for planets orbiting at 1-5 AU. The possibility to determine the epoch of giant planet formation in the protoplanetary disk would provide the definitive observational test to distinguish between the proposed theoretical models. These data would uniquely complement near- and mid-infrared imaging surveys (e.g., Burrows, 2005, and references therein) for direct detection of young, bright, wide-separation ($`a>30100`$ AU) giant planets.
### 6.4 Dynamical interactions in Multiple-Planet Systems
The different sources of dynamical interactions proposed to explain the highly eccentric orbits of planetary systems (see Section 2.1) give rise to significantly different orbital alignments. An effective way to understand their relative roles would involve measuring the relative inclination angle between pairs of planetary orbits. Studies addressing the long-term dynamical stability issue for multiple-planet systems, as well as the possibility of formation and survival of terrestrial planets in the Habitable Zone of the parent star (see Section 2.3), would also greatly benefit from knowledge of whether pairs of planetary orbits are coplanar or not.
The only way to provide meaningful estimates of the full three-dimensional geometry of any planetary system (without restrictions on the orbital alignment with respect to the line of sight) is through direct estimates of the mutual inclinations angles using high-precision astrometry (Sozzetti et al., 2001, ). For a $`Gaia`$-like, all-sky survey instrument, the database of potential targets out to 50-60 pc is of order of a few tens of thousand objects (Sozzetti et al., 2001). These data, combined with those available from Doppler measurements and transit photometry and transit timing (e.g., Miralda-Escudé, 2002; Holman & Murray, 2005; Agol et al., 2005), would then allow to put studies of the dynamical evolution of planetary systems on firmer grounds.
### 6.5 Concluding Remarks
Despite several decades of attempts, and a few recent successes, astrometric measurements with milli-arcsecond precision have so far proved of limited utility when employed as a tool to search for planetary mass companions orbiting nearby stars. However, an improvement of 2-3 orders of magnitude in achievable measurement precision, down to the few $`\mu `$as level, would allow this technique to achieve in perspective the same successes of the Doppler method, for which the improvement from the km s<sup>-1</sup> to the few m s<sup>-1</sup> precision opened the doors for ground-breaking results in planetary science.
In this paper I have reviewed a series of technological, statistical, and astrophysical issues that future ground-based as well as space-borne efforts will have to face in their attempts to discover planets. At the $`\mu `$as precision level, independently on the type of instrument utilized (either filled- or diluted aperture telescopes), a number of important modifications to the standard definition of astrometric observable (the stellar position in the instrument-specific reference frame) will have to be introduced, such as subtle effects due to general relativity. Astrophysical noise sources will have to be taken into account, which may mimic the presence of a planet, such as significant stellar surface activity. Several tools will have to be considered when attempting to derive reliable orbital solutions, such as optimized strategies to find good initial configurations for the orbital parameters. However, the greatest challenge will be to build instrumentation, both from ground (Keck-I, VLTI) and in space ($`SIM`$ and $`Gaia`$), capable of attaining the technologically demanding requirements to achieve a targeted single-measurement precision $`\sigma _m110`$ $`\mu `$as. Provided these will be met, astrometry during the next decade bears the potential to provide critical contributions to planetary science, which are crucially needed in order to complement the expectations from other indirect and direct planet detection methods, and refined theoretical understanding, for continuous improvements in the field of the formation and evolution of planetary systems.
During the preparation of this manuscript, I have benefited from very fruitful discussion with many colleagues. I am particularly grateful to A. P. Boss, S. Casertano, D. Charbonneau, D. W. Latham, M. G. Lattanzi, G. W. Marcy, M. Mayor, D. Pourbaix, D. Queloz, D. D. Sasselov, and G. Torres for helpful comments and insights. I thank an anonymous referee for a very critical revision of an earlier version of this paper, and for illuminating suggestions and recommendations that greatly improved the manuscript. The author acknowledges support from the Smithsonian Astrophysical Observatory through the SAO Predoctoral Fellowship program and the Keck PI Data Analysis Fund (JPL 1262605). This research has made use of NASA’s Astrophysics Data System Abstract Service and of the SIMBAD database, operated at CDS, Strasbourg, France. |
warning/0507/hep-ph0507181.html | ar5iv | text | # I Best fit values of the parameters of the Sivers functions. Notice that the errors generated by MINUIT are strongly correlated, and should not be taken at face value; the significant fluctuations in our results are shown by the shaded areas in Figs. and .
Extracting the Sivers function from polarized SIDIS data
and making predictions
M. Anselmino<sup>1</sup>, M. Boglione<sup>1</sup>, U. D’Alesio<sup>2</sup>, A. Kotzinian<sup>3</sup>, F. Murgia<sup>2</sup>, A. Prokudin<sup>1</sup>
<sup>1</sup>Dipartimento di Fisica Teorica, Università di Torino and
INFN, Sezione di Torino, Via P. Giuria 1, I-10125 Torino, Italy
<sup>2</sup>INFN, Sezione di Cagliari and Dipartimento di Fisica, Università di Cagliari,
C.P. 170, I-09042 Monserrato (CA), Italy
<sup>3</sup>Dipartimento di Fisica Generale, Università di Torino and
INFN, Sezione di Torino, Via P. Giuria 1, I-10125 Torino, Italy
## Abstract
The most recent data on the weighted transverse single spin asymmetry $`A_{UT}^{\mathrm{sin}(\varphi _h\varphi _S)}`$ from HERMES and COMPASS collaborations are analysed within LO parton model with unintegrated parton distribution and fragmentation functions; all transverse motions are taken into account, with exact kinematics, in the elementary interactions. The overall quality of the data is such that, for the first time, a rather well constrained extraction of the Sivers function for $`u`$ and $`d`$ quarks is possible and is performed. Comparisons with models are made. Based on the extracted Sivers functions, predictions for $`A_{UT}^{\mathrm{sin}(\varphi _h\varphi _S)}`$ asymmetries at JLab are given; suggestions for further measurements at COMPASS, with a transversely polarized hydrogen target and selecting favourable kinematical ranges, are discussed. Predictions are also presented for Single Spin Asymmetries (SSA) in Drell-Yan processes at RHIC and GSI.
1. Introduction In a recent paper we have discussed the role of intrinsic motions in inclusive and Semi-Inclusive Deep Inelastic Scattering (SIDIS) processes, both in unpolarized and polarized $`\mathrm{}p\mathrm{}hX`$ reactions. The LO QCD parton model computations have been compared with the experimental dependence of the unpolarized cross section on the azimuthal angle, around the virtual photon direction, between the leptonic and the hadronic planes (Cahn effect ); at small transverse momentum $`P_T`$ of the produced hadron $`h`$, such an effect is dominantly related to intrinsic motions and it allows an estimate of the average values of the transverse momenta of quarks inside a proton, $`𝒌_{}`$, and of final hadrons inside the fragmenting quark jet, $`𝒑_{}`$, with the best fit results:
$$k_{}^2=0.25(\mathrm{GeV}/c)^2p_{}^2=0.20(\mathrm{GeV}/c)^2.$$
(1)
More detail, both about the kinematical configurations and conventions and the fitting procedure can be found in Ref. . We only notice here that the above values have been derived from sets of data collected at different energies and in different ranges of the kinematical variables $`x_B`$, $`Q^2`$ and $`z_h`$, looking at the combined production of all charged hadrons in SIDIS processes; constant and flavour independent values of $`k_{}^2`$ and $`p_{}^2`$ have been assumed, avoiding at this stage complications related to possible $`x,z`$ and $`Q^2`$ dependences. Rather than a definite derivation, the above results are better considered as a consistent simple estimate and a convenient parameterization of the true intrinsic motion of quarks in nucleons and of hadrons in jets, supported by the available experimental information.
Equipped with such estimates, in Ref. we have studied the transverse single spin asymmetries $`A_{UT}^{\mathrm{sin}(\varphi _\pi \varphi _S)}`$ observed by HERMES collaboration ; that allowed a first rough extraction of the Sivers function
$$\mathrm{\Delta }^Nf_{q/p^{}}(x,k_{})=\frac{2k_{}}{m_p}f_{1T}^q(x,k_{}),$$
(2)
defined by
$`f_{q/p^{}}(x,𝒌_{})`$ $`=`$ $`f_{q/p}(x,k_{})+{\displaystyle \frac{1}{2}}\mathrm{\Delta }^Nf_{q/p^{}}(x,k_{})𝑺(\widehat{𝑷}\times \widehat{𝒌}_{})`$ (3)
$`=`$ $`f_{q/p}(x,k_{})f_{1T}^q(x,k_{}){\displaystyle \frac{𝑺(\widehat{𝑷}\times 𝒌_{})}{m_p}},`$ (4)
where $`f_{q/p}(x,k_{})`$ is the unpolarized $`x`$ and $`k_{}`$ dependent parton distribution ($`k_{}=|𝒌_{}|`$); $`m_p`$, $`𝑷`$ and $`𝑺`$ are respectively the proton mass, momentum and transverse polarization vectors ($`\widehat{𝑷}`$ and $`\widehat{𝒌}_{}`$ denote unit vectors).
The Sivers function extracted from HERMES data in was shown to be consistent with preliminary COMPASS data on $`A_{UT}^{\mathrm{sin}(\varphi _h\varphi _S)}`$ obtained on a deuteron target, in a different kinematical region .
While the preliminary HERMES data offered a definite indication of a non zero Sivers effect, their amount and quality were not yet such that an accurate extraction of the Sivers functions was possible; that reflects in the values of the parameters of the Sivers functions given in Table I of Ref. , which have large uncertainties.
New HERMES data are now available ; they are consistent with the previous ones, with much smaller errors. Similarly, COMPASS collaboration published their preliminary results. We consider here these whole new sets of HERMES and COMPASS data and perform a novel fit of the Sivers functions. It turns out that the new data constrain much better the parameters, thus offering the first direct significant estimate of the Sivers functions – for $`u`$ and $`d`$ quarks – active in SIDIS processes. The sea quark contributions are found to be negligible, at least in the kinematical region of the available data.
The modeled and extracted Sivers functions $`\mathrm{\Delta }^Nf_{u/p^{}}(x,k_{})`$ and $`\mathrm{\Delta }^Nf_{d/p^{}}(x,k_{})`$ are used to compute, and thus predict, the values of $`A_{UT}^{\mathrm{sin}(\varphi _h\varphi _S)}`$ expected at COMPASS, for scattering off a polarized hydrogen (rather than deuteron) target, which avoids cancellations between the opposite $`u`$ and $`d`$ contributions. Suggestions for the selection of favourable kinematic regions, where the asymmetry is sizeable, are discussed. Similar predictions, with strongly encouraging results, are given for polarized SIDIS processes at JLab.
Finally, we exploit the QCD prediction
$$f_{1T}^q(x,k_{})|_{\mathrm{DIS}}=f_{1T}^q(x,k_{})|_{\mathrm{D}\mathrm{Y}}$$
(5)
and compute a single spin asymmetry, which can only originate from the Sivers mechanism , for Drell-Yan processes at RHIC and GSI.
2. Extracting the Sivers functions Following Ref. , the inclusive ($`\mathrm{}p\mathrm{}X`$) unpolarized DIS cross section in non collinear LO parton model is given by
$$\frac{d^2\sigma ^{\mathrm{}p\mathrm{}X}}{dx_BdQ^2}=\underset{q}{}d^2𝒌_{}f_q(x,k_{})\frac{d\widehat{\sigma }^{\mathrm{}q\mathrm{}q}}{dQ^2}J(x_B,Q^2,k_{}),$$
(6)
and the semi-inclusive one ($`\mathrm{}p\mathrm{}hX`$) by
$$\frac{d^5\sigma ^{\mathrm{}p\mathrm{}hX}}{dx_BdQ^2dz_hd^2𝑷_T}=\underset{q}{}d^2𝒌_{}f_q(x,k_{})\frac{d\widehat{\sigma }^{\mathrm{}q\mathrm{}q}}{dQ^2}J\frac{z}{z_h}D_q^h(z,p_{}),$$
(7)
where
$$J=\frac{\widehat{s}^2}{x_B^2s^2}\frac{x_B}{x}\left(1+\frac{x_B^2}{x^2}\frac{k_{}^2}{Q^2}\right)^1$$
(8)
and
$$\frac{d\widehat{\sigma }^{\mathrm{}q\mathrm{}q}}{dQ^2}=e_q^2\frac{2\pi \alpha ^2}{\widehat{s}^2}\frac{\widehat{s}^2+\widehat{u}^2}{Q^4}$$
(9)
$`Q^2`$, $`x_B`$ and $`y=Q^2/(x_Bs)`$ are the usual leptonic DIS variables and $`z_h,𝑷_T`$ the usual hadronic SIDIS ones, in the $`\gamma ^{}`$$`p`$ c.m. frame; $`x`$ and $`z`$ are light-cone momentum fractions, with (see Ref. for exact relationships and further detail):
$$x=x_B+𝒪\left(\frac{k_{}^2}{Q^2}\right)z=z_h+𝒪\left(\frac{k_{}^2}{Q^2}\right)𝒑_{}=𝑷_Tz_h𝒌_{}+𝒪\left(\frac{k_{}^2}{Q^2}\right).$$
(10)
The elementary Mandelstam variables are given by
$`\widehat{s}^2`$ $`=`$ $`{\displaystyle \frac{Q^4}{y^2}}\left[14{\displaystyle \frac{k_{}}{Q}}\sqrt{1y}\mathrm{cos}\phi \right]+𝒪\left({\displaystyle \frac{k_{}^2}{Q^2}}\right)`$ (11)
$`\widehat{u}^2`$ $`=`$ $`{\displaystyle \frac{Q^4}{y^2}}(1y)^2\left[14{\displaystyle \frac{k_{}}{Q}}{\displaystyle \frac{\mathrm{cos}\phi }{\sqrt{1y}}}\right]+𝒪\left({\displaystyle \frac{k_{}^2}{Q^2}}\right),`$ (12)
where $`\phi `$ is the azimuthal angle of the quark transverse momentum, $`𝒌_{}`$ = $`k_{}(\mathrm{cos}\phi ,`$ $`\mathrm{sin}\phi ,0)`$. Regarding angle definitions and notations we adopt throughout the paper the so-called “Trento conventions” (see also Fig. 3 of Ref. ).
The $`\mathrm{sin}(\varphi _h\varphi _S)`$ weighted transverse single spin asymmetry, measured by HERMES and COMPASS, which singles out the contribution of the Sivers function (2), is given by:
$`A_{UT}^{\mathrm{sin}(\varphi _h\varphi _S)}=`$ (13)
$`{\displaystyle \frac{{\displaystyle \underset{q}{}}{\displaystyle 𝑑\varphi _S𝑑\varphi _hd^2𝒌_{}\mathrm{\Delta }^Nf_{q/p^{}}(x,k_{})\mathrm{sin}(\phi \varphi _S)\frac{d\widehat{\sigma }^{\mathrm{}q\mathrm{}q}}{dQ^2}J\frac{z}{z_h}D_q^h(z,p_{})\mathrm{sin}(\varphi _h\varphi _S)}}{{\displaystyle \underset{q}{}}{\displaystyle 𝑑\varphi _S𝑑\varphi _hd^2𝒌_{}f_{q/p}(x,k_{})\frac{d\widehat{\sigma }^{\mathrm{}q\mathrm{}q}}{dQ^2}J\frac{z}{z_h}D_q^h(z,p_{})}}}`$
We shall use Eq. (13), in which we insert a parameterization for the Sivers functions, to fit the experimental data.
We adopt the usual (and convenient) gaussian factorization for the distribution and fragmentation functions:
$$f_{q/p}(x,k_{})=f_q(x)\frac{1}{\pi k_{}^2}e^{k_{}^2/k_{}^2}$$
(14)
and
$$D_q^h(z,p_{})=D_q^h(z)\frac{1}{\pi p_{}^2}e^{p_{}^2/p_{}^2},$$
(15)
with the values of $`k_{}^2`$ and $`p_{}^2`$ of Eq. (1). Isospin and charge–conjugation relations imply
$`D_u^{\pi ^+}(z)=D_d^\pi ^{}(z)=D_{\overline{u}}^\pi ^{}(z)=D_{\overline{d}}^{\pi ^+}(z)D_{\mathrm{fav}}(z)`$
$`D_u^\pi ^{}(z)=D_d^{\pi ^+}(z)=D_{\overline{u}}^{\pi ^+}(z)=D_{\overline{d}}^\pi ^{}(z)D_{\mathrm{unfav}}(z).`$ (16)
The integrated parton distribution and fragmentation functions $`f_q(x)`$ and $`D_q^h(z)`$ are taken from the literature, at the appropriate $`Q^2`$ values of the experimental data .
We parameterize, for each light quark flavour $`q=u,d,\overline{u},\overline{d}`$, the Sivers function in the following factorized form:
$$\mathrm{\Delta }^Nf_{q/p^{}}(x,k_{})=2𝒩_q(x)h(k_{})f_{q/p}(x,k_{}),$$
(17)
where
$`𝒩_q(x)=N_qx^{a_q}(1x)^{b_q}{\displaystyle \frac{(a_q+b_q)^{(a_q+b_q)}}{a_q^{a_q}b_q^{b_q}}},`$ (18)
$`h(k_{})={\displaystyle \frac{2k_{}M_0}{k_{}^2+M_0^2}}`$ (19)
$`N_q`$, $`a_q`$, $`b_q`$ and $`M_0`$ (GeV/$`c`$) are free parameters. $`f_{q/p}(x,k_{})`$ is the unpolarized distribution function defined in Eq. (14). Since $`h(k_{})1`$ and since we allow the constant parameter $`N_q`$ to vary only inside the range $`[1,1]`$ so that $`|𝒩_q(x)|1`$ for any $`x`$, the positivity bound for the Sivers function is automatically fulfilled:
$$\frac{|\mathrm{\Delta }^Nf_{q/p^{}}(x,k_{})|}{2f_{q/p}(x,k_{})}1.$$
(20)
We have first attempted a fit of the HERMES and COMPASS data, taking into account 4 Sivers functions (for $`u,d,\overline{u}`$ and $`\overline{d}`$ quarks), for a total of 13 parameters, like in Ref. . However, it turns out that the available data are almost insensitive to the sea quark (and, in general, small $`x`$) contributions, which leads to largely undetermined parameters of the corresponding Sivers functions. Indeed, we have explicitely checked that various choices of $`\mathrm{\Delta }^Nf_{\overline{u}/p^{}}`$ and $`\mathrm{\Delta }^Nf_{\overline{d}/p^{}}`$ do not significantly affect the computation of $`A_{UT}^{\mathrm{sin}(\varphi _h\varphi _S)}`$, Eq. (13), in the kinematical regions of the performed experiments. We have then neglected the contributions of these functions and considered only the contributions of $`\mathrm{\Delta }^Nf_{u/p^{}}`$ and $`\mathrm{\Delta }^Nf_{d/p^{}}`$, for a total of 7 free parameters:
$$N_ua_ub_uN_da_db_dM_0.$$
(21)
The results of our fits are shown in Figs. 1 and 2. The weighted SSA $`A_{UT}^{\mathrm{sin}(\varphi _h\varphi _S)}`$ is plotted as a function of one variable at a time, either $`z_h`$ or $`x_B`$ or $`P_T`$; an integration over the other variables has been performed consistently with the cuts of the corresponding experiment (see Ref. for further detail). The resulting best fit values of the parameters are reported in Table I. The shaded area in Figs. 1 and 2 corresponds to one-sigma deviation at 90% CL and was calculated using the errors (Table I) and the correlation matrix generated by MINUIT, minimizing and maximizing the function under consideration, in a 7-dimensional parameter space hyper-volume corresponding to one-sigma deviation.
In Fig. 1 we also show predictions, obtained using the extracted Sivers functions, for $`\pi ^0`$ and $`K`$ production; data on these asymmetries might be available soon from HERMES collaboration.
3. Comparison with models and predictions for SIDIS processes
The extracted Sivers functions for $`u`$ and $`d`$ quarks are shown in Fig. 3, where we plot, for comparison with other results, the first $`𝒌_{}`$ moment
$$\mathrm{\Delta }^Nf_q^{(1)}(x)d^2𝒌_{}\frac{k_{}}{4m_p}\mathrm{\Delta }^Nf_{q/p^{}}(x,k_{})=f_{1T}^{(1)q}(x).$$
(22)
The solid line corresponds to the central values in Table I and the shaded area corresponds to varying the parameters within the shaded areas in Figs. 1 and 2. The other curves show results from models or fits to different data , as discussed below.
* The $`x`$-dependences of both $`\mathrm{\Delta }^Nf_{u/p^{}}`$ and $`\mathrm{\Delta }^Nf_{d/p^{}}`$ – as modeled in Eqs. (17-19) and shown in Fig. 3 – appear to be rather well determined, keeping in mind that the data are essentially confined in the region $`0.01\text{ }<x_B\text{ }<0.2`$. We notice that a 13-parameter fit – including $`\overline{u}`$ and $`\overline{d}`$ contributions – would lead to similar results; however, the values of $`\mathrm{\Delta }^Nf_{\overline{u}/p^{}}`$ and $`\mathrm{\Delta }^Nf_{\overline{d}/p^{}}`$, within their shaded areas, would be consistent with zero in the kinematical region of HERMES and COMPASS experiments. That is why we have not considered these contributions here.
* The large-$`x`$ behaviour of the Sivers functions cannot be fixed by the existing data. According to the counting rules of Ref. , and keeping in mind that $`\mathrm{\Delta }^Nf_{q/p^{}}`$ originates from the interference between distribution amplitudes with different proton helicities , one expects the large-$`x`$ behaviours
$$\mathrm{\Delta }^Nf_{u/p^{}}\mathrm{\Delta }^Nf_{d/p^{}}(1x)^4.$$
(23)
JLab data will cover the appropriate region to help checking this prediction.
* The dot–dashed line in Fig. 3 shows fit I of Ref. , where the $`q_T/m_p`$ weighted SIDIS asymmetries were fitted and the large $`N_c`$ relation was adopted:
$$f_{1T}^u(x,k_{})=f_{1T}^d(x,k_{}).$$
(24)
Notice that their results are in qualitative agreement with ours and that Eq. (24) naturally turns out to be approximately true in our fit.
* The dashed line in Fig. 3 plots the first $`𝒌_{}`$ moment of the Sivers functions obtained in Ref. , by fitting $`A_N`$ data in $`p^{}p\pi X`$ processes; these data are mainly sensitive to large $`x`$ value, where – again – approximately opposite values of $`\mathrm{\Delta }^Nf_{u/p^{}}`$ and $`\mathrm{\Delta }^Nf_{d/p^{}}`$ seem to be favoured. However, as discussed in Ref. , the universality of the Sivers functions active in SIDIS and $`p^{}p\pi X`$ processes is still an open issue.
* Most theoretical models give a Sivers function for $`u`$ quarks much larger, in magnitude, than for $`d`$ quarks; this can be seen, for example, from the dotted curve in Fig. 3, taken from the computation, in the MIT bag model, of Ref. . The same is true for the Sivers functions obtained, within a spectator model with diquarks, in Refs. , and .
3.1 $`A_{UT}^{\mathrm{sin}(\varphi _h\varphi _S)}`$ at COMPASS with polarized hydrogen target
By inspection of Eq. (13) it is easy to understand our numerical results for the $`u`$ and $`d`$ Sivers functions. In fact one can see that for scattering off a hydrogen target (HERMES), one has
$$\left(A_{UT}^{\mathrm{sin}(\varphi _h\varphi _S)}\right)_{\mathrm{hydrogen}}4\mathrm{\Delta }^Nf_{u/p^{}}D_u^h+\mathrm{\Delta }^Nf_{d/p^{}}D_d^h,$$
(25)
while, for a scattering off a deuterium target (COMPASS),
$$\left(A_{UT}^{\mathrm{sin}(\varphi _h\varphi _S)}\right)_{\mathrm{deuterium}}\left(\mathrm{\Delta }^Nf_{u/p^{}}+\mathrm{\Delta }^Nf_{d/p^{}}\right)\left(4D_u^h+D_d^h\right).$$
(26)
Opposite $`u`$ and $`d`$ Sivers contributions suppress COMPASS asymmetries for any hadron $`h`$. These opposite contributions do not affect the $`\pi ^+`$ asymmetry measured off a a hydrogen target, Eq. (25): in this case the charge factor 4 and the favourite fragmentation function ($`D_u^{\pi ^+}>D_d^{\pi ^+}`$) combine to make the first term of Eq. (25) larger than the second one. The cancellation between the two terms is stronger for the $`\pi ^{}`$ asymmetry, because in this case the charge factor 4 in the first term of Eq. (25) couples to the unfavoured fragmentation function ($`D_u^\pi ^{}<D_d^\pi ^{}`$). Similar arguments hold for the production of kaons and, in general, for the production of charged hadrons, which is dominated by pions.
However, the COMPASS collaboration will soon be taking data with a transversely polarized hydrogen target. We can easily compute the expected results: adopting the same experimental cuts which were used for the deuterium target we obtain the predictions shown in the upper panel of Fig. 4. The asymmetry is found to be around 5%. These expected values can be further increased by properly selecting the experimental data, thus excluding kinematical regions whose contribution to the asymmetry is negligible. For example, selecting events with
$$0.4z_h10.2P_T1\mathrm{GeV}/c0.02x_B1,$$
(27)
yields the predictions shown in the lower panel of Fig. 4. The asymmetry for positively charged hadrons becomes larger, and, provided that enough statistics can be gathered, one expects a clear observation of a sizeable azimuthal asymmetry also for the COMPASS experiment.
3.2 $`A_{UT}^{\mathrm{sin}(\varphi _h\varphi _S)}`$ at JLab with polarized hydrogen target
Also JLab experiments are supposed to measure the SIDIS azimuthal asymmetry for the production of pions on a transversely polarised hydrogen target, at incident beam energies of 6 and 12 GeV. The kinematical region of this experiment is very interesting, as it will supply information on the behaviour of the Sivers functions in the large-$`x_B`$ domain, up to $`x_B0.6`$. The experimental acceptance for JLab events at 6 GeV is constrained by :
$`0.4z_h0.70.02P_T1\mathrm{GeV}/c0.1x_B0.6`$ (28)
$`0.4y0.85Q^21(\mathrm{GeV}/c)^2W^24\mathrm{GeV}^21E_h4\mathrm{GeV}.`$
while, with an incident beam energy of 12 GeV, this becomes:
$`0.4z_h0.70.02P_T1.4\mathrm{GeV}/c0.05x_B0.7`$ (29)
$`0.2y0.85Q^21(\mathrm{GeV}/c)^2W^24\mathrm{GeV}^21E_h7\mathrm{GeV}.`$
Imposing these experimental cuts we obtain the predictions shown in Fig. 5. A large and healthy azimuthal asimmetry for $`\pi ^+`$ production should be observed. Similar results have been obtained also in an approach based on a Monte Carlo event generator . However, one relevant comment is in order:
* As the region of high $`x_B`$ is not covered by HERMES and COMPASS experiments, the predictions for the large-$`x_B`$ dependence of the asymmetry are very sensitive to the few large-$`x_B`$ data points of these two experiments. As a consequence, the results for JLab experiments may still change drastically in the region $`0.4\text{ }<x_B\text{ }<0.6`$, and the asymmetry might be much smaller than presented in Fig. 5. This reflects in the wide shaded area at large $`x_B`$ values. Conversely, the results on $`P_T`$ and $`z_h`$ dependences are more stable as they depend on the $`x_B`$-integrated Sivers function. Notice also the little dependence on the beam energy, consistent with the approximate factorized form of the numerator and denominator of Eq. (13), which leads to cancellations in their ratio.
4. Transverse single spin asymmetries in Drell-Yan processes
Let us now consider the transverse single spin asymmetry,
$$A_N=\frac{d\sigma ^{}d\sigma ^{}}{d\sigma ^{}+d\sigma ^{}},$$
(30)
for Drell-Yan processes, $`p^{}p\mathrm{}^+\mathrm{}^{}X`$, $`p^{}\overline{p}\mathrm{}^+\mathrm{}^{}X`$ and $`\overline{p}^{}p\mathrm{}^+\mathrm{}^{}X`$, where $`d\sigma `$ stands for
$$\frac{d^4\sigma }{dydM^2d^2𝒒_T}$$
(31)
and $`y`$, $`M^2`$ and $`𝒒_T`$ are respectively the rapidity, the squared invariant mass and the transverse momentum of the lepton pair in the initial nucleon c.m. system. The cross section can eventually be integrated over some of these variables, according to the kinematical configurations of the experiments.
In such a case the single spin asymmetry (30) can only originate from the Sivers function and is given (selecting the region with $`q_T^2M^2,q_Tk_{}`$) by
$$A_N=\frac{_qe_q^2d^2𝒌_qd^2𝒌_{\overline{q}}\delta ^2(𝒌_q+𝒌_{\overline{q}}𝒒_T)\mathrm{\Delta }^Nf_{q/p^{}}(x_q,𝒌_q)f_{\overline{q}/p}(x_{\overline{q}},𝒌_{\overline{q}})}{2_qe_q^2d^2𝒌_qd^2𝒌_{\overline{q}}\delta ^2(𝒌_q+𝒌_{\overline{q}}𝒒_T)f_{q/p}(x_q,𝒌_q)f_{\overline{q}/p}(x_{\overline{q}},𝒌_{\overline{q}})},$$
(32)
where $`q=u,\overline{u},d,\overline{d},s,\overline{s}`$ and
$$x_q=\frac{M}{\sqrt{s}}e^yx_{\overline{q}}=\frac{M}{\sqrt{s}}e^y.$$
(33)
Eq. (32) explicitely refers to $`p^{}p`$ processes, with obvious modifications for $`p^{}\overline{p}`$ and $`\overline{p}^{}p`$ ones.
Inserting into Eq. (32) the Sivers functions extracted from our fit to SIDIS data and reversed in sign according to Eq. (5), we obtain the predictions shown in Figs. 6 and 7. Fig. 6 shows the value of $`A_N`$ as a function of $`M`$ and $`x_F=x_qx_{\overline{q}}`$, for RHIC configurations: the lepton pair transverse momentum $`𝒒_T`$ has been integrated in the range $`0q_T1`$ GeV/$`c`$, while the rapidity variable $`y`$ and the lepton pair invariant mass $`M`$ have been integrated according to the experimental situations, as indicated in the legenda. The integration over the azimuthal angle of $`𝒒_T`$ has been performed, to avoid cancellations, as in Ref. , that is integrating over $`\varphi _{q__T}`$ in the range $`[0,\pi /2]`$ only, or, alternatively, taking into account the change of sign in the different production quadrants. In either case, the $`\varphi _{q__T}`$ integration gives an overall factor $`2/\pi `$. As the shaded areas in previous figures, the closed areas correspond to the uncertainty in our determination of the Sivers functions.
Fig. 7 shows the same plots for the PAX experiment planned at the proposed asymmetric $`p\overline{p}`$ collider at GSI: $`𝒒_T`$ has been integrated over the same range as for RHIC predictions, while $`y`$ and $`M`$ as indicated in the legenda. Results for $`p^{}\overline{p}`$ and $`\overline{p}^{}p`$ processes are identical, due to charge conjugation invariance. Notice that in our configuration the polarized proton or antiproton always moves along the $`(+\widehat{𝒛})`$–direction.
The correct interpretation of these results requires some further considerations.
* In our computations we have used the value of $`k_{}^2=0.25(\mathrm{GeV}/c)^2`$, obtained from an analysis of SIDIS data ; such a value is certainly appropriate for consistently computing spin asymmetries in SIDIS processes, in the $`\gamma ^{}p`$ c.m. frame, as we have done in Section 3.1 and 3.2. This value naturally corresponds to the intrinsic motion of partons confined in a nucleon, simply according to uncertainty principle, and describes well the $`𝑷_T`$ dependences of measured cross sections, up to $`P_T1`$ GeV/$`c`$. In addition, as we have seen, it allows an understanding of the azimuthal asymmetries, which would otherwise vanish.
* However, when considering other processes, as the inclusive production of hadrons or leptons in $`pp`$ or $`p\overline{p}`$ interactions, we know that higher order QCD corrections, like the threshold resummation of large logarithms due to soft gluon emission , lead to large $`K`$-factor enhancements of the cross sections. Our $`𝒌_{}`$ unintegrated approach to the description of hard scattering processes within a generalization of the QCD factorization theorem , can be considered as an effective model which not only takes into account the original partonic intrinsic motion (related to parton confinement), but also, to some extent, the intrinsic $`𝒌_{}`$ built via soft gluon emission. Indeed, the values of $`k_{}^2`$ used in Ref. in order to describe the data on the unpolarized $`pp\pi X`$ processes are higher than the values used here, and those requested for the Drell-Yan cross-section might be even higher. The average $`k_{}^2`$ estimate of $`0.25(\mathrm{GeV}/c)^2`$ might be at most adequate to explain the Drell-Yan cross section up to $`q_T\text{ }<`$ 1 GeV/$`c`$, but would badly fail above that value.
* For the reasons explained above, consistently with our approach expected to hold in the $`k_{}P_Tq_T`$ region, in our predictions for $`A_N`$, Figs. 6 and 7, we have integrated over $`𝒒_T`$ up to $`q_T`$ = 1 GeV/$`c`$. In addition, we notice that the value of $`A_N`$, as given by Eq. (32), is little sensitive to the chosen value of $`k_{}^2`$: while both the numerator and the denominator of Eq. (32) greatly vary with $`k_{}^2`$, their ratio does much less so. We then consider our predictions for $`A_N`$, assuming the validity of the relation (5), safe and significant.
5. Comments and conclusions
We have considered the most recent data from polarized SIDIS processes which single out the Sivers effect, namely the $`A_{UT}^{\mathrm{sin}(\varphi _h\varphi _S)}`$ transverse single spin azimuthal asymmetry, measured by HERMES and COMPASS collaborations for charged hadron production. Assuming a Gaussian factorization of the $`k_{}`$ and $`p_{}`$ dependence of all distribution and fragmentation functions, together with a most simple parameterization of the $`x`$-dependence of the unknown Sivers functions, we have exploited the data to extract information on $`\mathrm{\Delta }^Nf_{u/p^{}}(x,k_{})`$ and $`\mathrm{\Delta }^Nf_{d/p^{}}(x,k_{})`$.
For the first time, the amount and quality of the experimental results allow a significant, although still limited in $`x`$-range, estimate of the Sivers functions for $`u`$ and $`d`$ quarks; these turn out to be definitely different from zero, well inside the positivity bound of Eq. (20) and almost opposite to each other. This last feature, predicted theoretically in some models , explains naturally and is related to the small asymmetry observed by COMPASS in scatterings of muons off a deuteron target.
According to the general strategy of combining new experimental information with the computation and prediction of new expected results, the extracted functions have been used to compute $`A_{UT}^{\mathrm{sin}(\varphi _h\varphi _S)}`$ in other experiments. It turns out that, contrary to the results obtained off a polarized deuteron target, a sizeable $`h^+`$ asymmetry should be measured by COMPASS collaboration once they switch, as planned, to a transversely polarized hydrogen target; a careful choice of the kinematical region of the selected events would help in further increasing the numerical value of the asymmetry for positively charged hadron production.
Large values of $`A_{UT}^{\mathrm{sin}(\varphi _h\varphi _S)}`$ are expected at JLab, both in the 6 and 12 GeV operational modes, for $`\pi ^+`$ inclusive production; in particular the $`z_h`$ and $`P_T`$ dependence of the asymmetry seems to be stable and reliable, while the $`x_B`$ dependence shows large uncertainties due to the lack of HERMES and COMPASS information in this region. JLab experiments have the unique features of exploring the large $`x_B`$ behaviour of the quark distribution functions, where predictions from QCD counting rules, Eq. (23), could be tested.
The Sivers effect was believed for some time to be forbidden by QCD time reversal properties ; however, this proved to be incorrect after an explicit model showed the existence of a non zero Sivers function . The original proof of the vanishing of the Sivers effect turned into the relation of Eq. (5), which predicts opposite values for the Sivers functions measured in SIDIS and Drell-Yan processes. We have then used this basic QCD relation and computed the single spin asymmetries in Drell-Yan processes given in Eq. (32); these can only be generated by the Sivers functions, since no fragmentation functions are needed to describe this process. We have used the same functions as extracted from SIDIS data, with opposite signs. The predicted $`A_N`$ could be measured at RHIC in $`pp`$ collisions and, in the long range, at the proposed PAX experiment at GSI , in $`p\overline{p}`$ interactions. It would provide a clear and stringent test of basic QCD properties.
A phenomenological study of the Sivers asymmetry – the correlation between the intrinsic $`𝒌_{}`$ of partons and the proton spin – is now possible, thanks to the existing experimental information and more which will soon be available. Basic properties of the QCD proton structure can and will be clarified. A good control of the Sivers mechanism will help in learning and understanding about other fundamental partonic spin properties, like the transversity distribution and the Collins mechanism .
Acknowledgements We would like to thank Elke Aschenauer, Harut Avakian and Delia Hasch for fruitful discussions. We acknowledge the support of the European Community–Research Infrasctructure Activity under the FP6 “Structuring the European Research Area” programme (HadronPhysics, contract number RII3-CT-2004-506078). U.D. and F.M. acknowledge partial support by MIUR (Ministero dell’Istruzione, dell’Università e della Ricerca) under Cofinanziamento PRIN 2003. |
warning/0507/physics0507035.html | ar5iv | text | # Income Distribution Dependence of Poverty Measure: A Theoretical Analysis
## 1 Introduction
Since the paradigmatic contribution of Sen and Atkinson , a remarkable amount of effort has been undertaken in theoretically understanding the economics of poverty and inequality. The studies range from being aptly mathematical in nature to a qualitative characterisation of such population dialectics. Pradhan and Ravallion have used qualitative assessments of perceived consumption adequacy based on a household survey. They claim that perceived consumption needs can be a more promising approach than the subjective income-based poverty line. This consumption norm can correspond to a saturation level of consumption, below which the individual could be considered to be in poverty. Further, in this paper, our approach is rather complementary to a lemma-based mathematical model in that we use survey based consumption data to quantify the dependence of a well-known poverty function on the mean and variance of the income distribution. To this end, we use income-expenditure data from a ‘developing nation’ (India in our case) and utilise the well established technique of data fitting to define the per capita consumption as a function of income. Here the implicit assumption is that of a near equilibrium situation such that the time dependence of both income and consumption variables can be considered as transients without much effect on the asymptotic distributions. Deaton has discussed the ambiguity that arises using survey data versus national accounts data for individual consumption or income levels. Although the survey consumption data seem to understate the true consumption levels, we are however using the data as a backup to our analytical results thereby restricting our claims to being qualitative in nature. Such comparisons with real data help us have approximate ideas of the values of the unknown parameters, two in our model, although the general conclusions are remarkably independent of these parameter values.
Assuming that the income distribution can be characterised by a two-parameter function, such as a log-normal distribution, in the first section of the paper we study the effects of changes in the mean and variance of the underlying income distribution on poverty. The results of this analysis indicate that an increase in mean income and a reduction in the variance of income distribution can reduce poverty. It also hints towards a trade-off, in that while an increase in average income reduces poverty, a simultaneous increase in income variance can escalate poverty. This result is likely to suggest that reducing income inequality should be the precondition for lowering poverty. These general results are then contrasted in the following section using a different model for the income distribution, the pareto distribution. The objective is basically to probe whether the results obtained are universal in nature and if not, then which distribution defines a better measure of poverty. <sup>1</sup><sup>1</sup>1Sen (1976) introduced the notion of deprivation in the income distribution literature, and criticised the use of the head-count ratio as a measure of poverty. Rao (1981) suggested broadening the scope of poverty measurement to nutritional norms as opposed to monetary measures. If poverty is to be regarded as negative welfare, it makes sense to relate it to consumption deprivation resulting from an uneven income distribution rather than to the income distribution alone as is done by the traditional poverty ratio index (Kumar et al., 1996).
## 2 Poverty impact of changes in log-normal income distribution
Poverty equals consumption deprivation on an essential food. The necessity of defining poverty as a multidimensional concept rather than relying on income or consumption expenditures per capita has been well documented. Although it is important to assess deprivation with more than one attribute (see ), we consider the case of most essential food item that is required for survival, in an attempt to include deprivation into the poverty index. Such an index would suggest that a person can be considered poor if the individual’s consumption falls within the deprivation area in the diagram (see lower panel of Fig.1), that is, the cumulative difference between the saturation consumption level of cereal and actual cereal consumption by the community as a whole. The upper panel of Fig.1 shows positive consumption even at zero income level, which makes our formulation more realistic than the non-linear function used in Kumar et al. . The non-linear function used in our paper that allows a saturation level of consumption norm for food-grains is as follows:
$$C(y)=\frac{V\mathrm{exp}(y)}{K+\mathrm{exp}(y)}$$
(1)
where $`C`$ is the consumption expenditure on food-grains, $`y`$ is income and the parameters $`V,K(>0)`$ represent the saturation level of real food-grain consumption expenditure or the bliss level and the level of income needed to consume one half of the saturation level respectively. Consumption deprivation (CD) or poverty (P) can be defined as the shortfall of actual consumption expenditure relative to saturation level V, or $`CD=VC`$. Thus the non-linear CD function is derived as:
$$CD=\frac{VK}{K+\mathrm{exp}(y)}$$
(2)
This function, being a convex decreasing function of income provides a direct measure of poverty based on nutritional norms, while $`V`$ and $`K`$ are parameters of a concave Engel curve. Here $`CV`$ represents the idealistic limit where there is no deprivation or poverty corresponding to a static equilibrium in the social dialectics mathematically represented by $`y=y^{}`$. In what follows, we would consider two asymptotic regimes - $`y0`$ and $`y\mathrm{}`$ \- physically which correspond to the low and high income groups respectively. Naturally our focus would be on the $`y0`$ limit, that is on the low income section although the analysis would encompass both limits.
If consumption of the most essential food item follows a concave non-linear functional form and if individual poverty is measured as the difference between the saturation level of consumption of the essential food item and its actual level, assumption of a log-normal income distribution implies a reduction in poverty with the increase of mean income of the population and an increase in inequality with increasing poverty. This new measure of poverty is based on the notion of consumption deprivation of a very essential staple food such as rice or wheat (cereal), derived from a nonlinear, monotonically increasing concave consumption function varying with the income, albeit with no specific reference to a subjective poverty line. The standard log-normal probability density function (pdf) is defined as
$$\mathrm{f}(y/\mu ,\sigma ^2)=\frac{1}{y\sigma \sqrt{2\pi }}\mathrm{exp}[\frac{(lny\mu )^2}{2\sigma ^2}]$$
(3)
where y is normally distributed with mean $`\mu `$ and variance $`\sigma ^2`$ (both positive real numbers). With this log-normal pdf for the income y, the poverty equation can be rewritten as follows:
$`P`$ $`=`$ $`{\displaystyle _0^{\mathrm{}}}CD(y)\mathrm{f}(y)𝑑y`$ (4)
$`=`$ $`{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{VK}{K+\mathrm{exp}(y)}}{\displaystyle \frac{1}{y\sigma \sqrt{2\pi }}}\mathrm{exp}[{\displaystyle \frac{(lny\mu )^2}{2\sigma ^2}}]𝑑y`$
Partial derivatives of the above equation (4) with respect to $`\mu `$ and $`\sigma ^2`$ give
$`{\displaystyle \frac{P}{\mu }}`$ $`=`$ $`{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{VK}{K+\mathrm{exp}(y)}}{\displaystyle \frac{1}{y\sigma \sqrt{2\pi }}}`$
$`\times `$ $`\mathrm{exp}[{\displaystyle \frac{(lny\mu )^2}{2\sigma ^2}}]{\displaystyle \frac{lny\mu }{\sigma ^2}}dy`$
$`{\displaystyle \frac{P}{\sigma ^2}}`$ $`=`$ $`{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{VK}{K+\mathrm{exp}(y)}}{\displaystyle \frac{1}{2y\sigma ^3\sqrt{2\pi }}}`$ (5)
$`\times `$ $`\mathrm{exp}[{\displaystyle \frac{(lny\mu )^2}{2\sigma ^2}}][{\displaystyle \frac{(lny\mu )^2}{\sigma ^2}}1]dy`$
P satisfies the three standard axioms of a poverty index <sup>2</sup><sup>2</sup>2See Foster et al , Kakwani , Atkinson , and Foster and Shorrocks for the different axioms of a poverty index., namely the monotonicity, transfer, and transfer sensitivity axioms that any such index must satisfy.
### 2.1 Asymptotic solutions of the poverty function
This section deals with asymptotic solutions of the poverty functions for extremely low ($`y0`$) to moderate values of the income distribution. This is mathematically categorised in the following manner:
For moderate incomes, $`C(y)=\frac{V\mathrm{exp}(y)}{K+\mathrm{exp}(y)}=C_{\mathrm{mod}}(y)`$, say,
1. $`C_{\mathrm{mod}}(y0)=\frac{V}{K+1}`$ $`\&`$
2. $`C_{\mathrm{mod}}(y\mathrm{})=V`$
whereas for very low income groups, $`C(y)=\frac{Vy}{K+y}=C_{\mathrm{low}}(y)`$, say,
1. $`C_{\mathrm{low}}(y0)=0`$ $`\&`$
2. $`C_{\mathrm{low}}(y\mathrm{})=V`$
The above comparison clearly shows that although both definitions of the consumption function are generally equivalent in the low income limit, for the absolutely needy groups, $`C_{\mathrm{mod}}(y)`$ predicts a non-zero ($`\frac{V}{K+1}`$) lower limit of income which is more realistic than $`C_{\mathrm{low}}(y0)=0`$. A linear stability analysis of $`C_{\mathrm{low}}(y)`$ also shows that $`y=0`$ is an unstable fixed point, which further strengthens this conviction. Henceforth our attention will mainly be focused towards the lowest income groups defined by $`C_{\mathrm{low}}(y)`$, although we would flip back and forth between the moderate to the low income classes for comparisons. Before proceeding any further, though, we first derive the values of the parameters $`V,K`$ by fitting the function $`C_{\mathrm{mod}}`$ with actual survey data obtained from National Sample Survey, 1999-2000, 55th Round, India. We would be using these values of $`V,K`$ in all analyses in this paper. Fig.2 portrays the shape of an Engle curve, graphing real cereal expenditure against the total expenditure - a surrogate for income.
The above exact data fitting conclusively shows that the parameters $`V`$ and $`K`$ have the respective values 45 and 15 in Indian currency (Rupees). These are roughly equivalent to 1.0 USD and 0.33 USD respectively. Now using these values, we study the case for typically the lowest income classes defined by the consumption function $`C_{\mathrm{low}}(y)`$. In this case, however, we need to focus on both low and high limits of the variance. Upto first order in $`\sigma ^2`$, we find that
$$P_{\mathrm{low}}^{}{}_{\sigma 0}{}^{}=\frac{V}{K}[K\mathrm{exp}(\mu \frac{\sigma ^2}{2})]$$
(6)
The poverty dependence on the mean for this asymptotic regime can be understood from figure 3.
Fig. 3 tells us that poverty is a monotonically decreasing function of variance for a fixed mean (taken to be 2.773 for a direct comparison with Fig. 4 later). On the other hand, for a fixed variance (0.001), poverty increases with mean and then saturates after a critical value. This result is very remarkable but needs to be taken with a pinch of salt, especially since this is true only in the asymptotic ($`\sigma 0`$) regime. We will revisit this problem in the following section where we discuss the situation when both the mean and the variance of the income distribution are simultaneously varying.
### 2.2 Overall impact of simultaneous changes in mean and variance
Here we show what effect any change, either increase or decrease, in the income distribution has on the overall poverty function when the distribution is log-normal and when both mean and variance are varying. Since our focus is on the low income group, we will be using $`C_{\mathrm{low}}(y)`$ as our definition for the consumption function. The attention here would be to decipher the joint variation of the poverty function $`P(\mu ,\sigma ^2)`$ with respect to $`\mu `$ and $`\sigma ^2`$. Once again using a $`1/y`$ expansion <sup>3</sup><sup>3</sup>3This might sound confusing since we are discussing small income but in effect, all that we are doing is to use a well known 1/y expansion prevalent in statistical mechanics. It is generally valid for a considerable range involving large to moderate values of the variable y. We have checked this result using $`C_{\mathrm{mod}}`$ and the qualitative results remain altogether unaltered. upto the first order, we find that the joint poverty function reads as
$`dP(\mu ,\sigma ^2)`$ $`=`$ $`{\displaystyle \frac{P}{\mu }}d\mu +{\displaystyle \frac{P}{\sigma ^2}}d\sigma ^2`$ (7)
$`=`$ $`{\displaystyle \frac{VK}{\sigma ^2}}\mathrm{exp}[(\mu {\displaystyle \frac{\sigma ^2}{2}})]d\mu `$
$`+`$ $`{\displaystyle \frac{VK}{2}}\mathrm{exp}[(\mu {\displaystyle \frac{\sigma ^2}{2}})A]d\sigma ^2`$
This equation suggests that poverty is a decreasing function of changes in $`\mu `$ and an increasing function of changes in $`\sigma `$<sup>2</sup>. For a fixed variance, d$`\sigma `$<sup>2</sup>=0, and hence the first component of , reflecting change in $`\mu `$, will provide convergence; and with a fixed mean, d$`\mu `$=0, the second component, exhibiting change in $`\sigma `$<sup>2</sup>, will give convergence of the equation. When both the mean and variance of the income distribution change as a result of changes in macroeconomic policies, their effect on poverty can be evaluated via equation (7). The notable point here is the fundamental qualitative difference with the prediction from equation (6). As opposed to the earlier asymptotic result where increase of the mean income was expected to generate a positive augmentation in poverty (for fixed variance) followed by a saturation at a particular value $`\mu _c`$, equation (7) with a fixed $`\sigma `$ clearly suggests that poverty decreases with increase of the mean income. This apparent dichotomy can be understood once we analyse the physical meaning hidden in equation (6). It says that in a relatively large group of low earning population, a very small variance between the earners contributes to an increase in poverty for very low to moderate values of the mean income. However, once the mean income reaches a critical value, this spurious effect saturates off. This can be contrasted with the prediction from the last equation which holds true for moderate to large values of $`\sigma `$. We would like to specifically point out here that both predictions from equations (6, 7) are true but in their respective regimes defined by small to large values of $`\sigma `$.
## 3 Poverty impact of changes in pareto income distribution
In this section, our objective is to study the mean and variance dependence of the poverty function, replacing the log-normal probability distribution, previously assumed, with a pareto distribution and contrast the findings later. Once again we would conform to the same consumption and deprivation functions (1,2) and try to understand the qualitative changes in the poverty function of a growing economy with respect to changes in the mean and variance of the overall income distribution.
The standard pareto probability density function $`f_{\mathrm{pareto}}`$ defined over the interval $`yb`$ is given by
$$\mathrm{f}_{\mathrm{pareto}}(y)=\frac{ab^a}{y^{a+1}}$$
(8)
where the mean $`\mu `$ and the variance $`\sigma ^2`$ can be easily shown to be as follows
$`\mu `$ $`=`$ $`{\displaystyle \frac{ab}{a1}}`$
$`\sigma ^2`$ $`=`$ $`{\displaystyle \frac{ab^2}{(a1)^2(a2)}}`$ (9)
With this pareto probability density function, the poverty function $`P_{\mathrm{pareto}}`$ reads as follows
$`P_{\mathrm{pareto}}(a,b)`$ $`=`$ $`{\displaystyle _0^{\mathrm{}}}CD(y)\mathrm{f}_{\mathrm{pareto}}(y)𝑑y`$ (10)
$`=`$ $`VKab^a{\displaystyle _b^{\mathrm{}}}{\displaystyle \frac{dy}{(K+y)y^{a+1}}}`$
$`=`$ $`{\displaystyle \frac{V}{K}}[1ab^a{\displaystyle _b^{\mathrm{}}}𝑑y{\displaystyle \frac{1}{(K+y)y^a}}]`$
Defining the identity $`I(a)=_b^{\mathrm{}}\frac{dy}{(K+y)y^{a+1}}`$, and taking recourse to a bit of algebra one can deduce a recursive relation
$`I(a)`$ $`=`$ $`{\displaystyle \frac{1}{Kab^a}}[1{\displaystyle \frac{ab}{K(a1)}}+{\displaystyle \frac{ab^2}{K^2(a2)}}]`$ (11)
$``$ $`{\displaystyle \frac{1}{K^3}}I(a3,b)`$
This equation (11) can be correlated with a hypergeometric $`{}_{2}{}^{}F_{1}^{}`$ series <sup>4</sup><sup>4</sup>4A hypergeometric series is an algebraic power series in which the ratio of successive coefficients $`r_n/r_{n1}`$ is a rational function of $`n`$. The hypergeometric series that we are using here is due to Gauss and has the mathematical definition $`{}_{2}{}^{}F_{1}^{}(a,b;c;,z)=\frac{\mathrm{\Gamma }(c)}{\mathrm{\Gamma }(b)\mathrm{\Gamma }(cb)}_0^1dtt^{b1}(1t)^{cb1}(1tz)^a`$. In our case, $`I(a)=\frac{b^{1a}}{1+a}{}_{2}{}^{}F_{1}^{}(1,1+a;2+a;K/b)`$ for $`b>0,b+K0,\mathrm{Re}[a]>1,\mathrm{Im}[K]0`$. and for specified values of the parameters can be solved numerically. For our purpose though, we consider the limit $`a\mathrm{}`$ to have a first hand impression of the situation
$$P(a\mathrm{},b)=\frac{V}{K}[1\frac{1}{K}\frac{1}{1+\frac{b}{K}}]$$
(12)
We would now directly evaluate the poverty function in a more physical limit. Without any loss of generality we choose the limit $`K0`$ which is akin to the 1/y expansion we did in deriving the poverty function for the log-normal distribution. We would shortly see that in this case, this basic expansion allows us to have an ‘exact’ derivation of the poverty function as opposed to its log-normal counterpart. Upto the first order in 1/y and utilising equation (9), we find
$$P(a,b)=VK[\frac{1}{\mu }\frac{a^2}{a^21}\frac{1}{\mu ^2}\frac{a^3}{(a1)^2(a+2)}]$$
(13)
where $`a=2+\sqrt{1+\frac{\mu ^2}{\sigma ^2}}`$ and $`b=\frac{a1}{a}\mu `$. A numerical solution of the above equation (13) <sup>5</sup><sup>5</sup>5To evaluate the inflexion points, we used the software mathematica and later checked the result using another software called maple. The results were once again cross-checked using a self-generated fortran code. All numerical results that we cite in this article have been cross-checked using three different and independent numerical techniques. shows that it has a pair of inflexion points <sup>6</sup><sup>6</sup>6The inflexion point is defined through the numerical solution of the coupled equations $`\frac{^2P}{\mu ^2}=0`$ and $`\frac{^2P}{s^2}=0`$, where $`s=\sigma ^2`$. Out of the two pairs of solution, only one turns out to be physical. The other solution gives a negative value of $`s`$. We work with the physical solution only., out of which the physical pair is at $`\mu =3.05139\&\sigma ^2=0.0692138`$. Solving around this inflexion point, we now come across one of the most remarkable results of this article, the fact that poverty initially decreases with increasing variance until it reaches a critical value $`\sigma ^2=\sigma _{c}^{}{}_{}{}^{2}`$ beyond which the poverty starts increasing with variance followed by a dip once again.
Fig. 4 has been drawn using $`\sigma ^2=2.773`$, a value reasonably close to the inflexion point. The plot shows that poverty decreases until it reaches the point $`\mu _c0.25`$ after which it starts increasing approximately until $`\sigma ^2=2`$ and then it starts decreasing again. This result is in marked contrast with the log-normal case where the poverty rather uninterestingly decreases with increasing mean for a fixed variance, and increases with for a fixed mean. It is now not difficult to pinpoint the detailed meaning of this result. Referring to Fig. 4, zone A defines a rather ‘underdeveloped’ economy, zone B stands for a ‘developing’ economy, our case in study, while the final zone C clearly indicates what one would expect in the case of an economically ‘developed’ nation. We can probably claim without much ambiguities that a pareto distribution has the power to encapsulate all three modes of economies and is the ideal candidate for all future studies involving poverty measure. Further, zones B and C appear to suggest an inverted-U hypothesis similar to Kuznets that poverty increases in the early stages of development and subsequently it declines with higher level of economic progress even though such development is associated with higher inequality.
## 4 Conclusion
This paper made use of a poverty function, which is different from the conventional poverty indices in the following manner: (1) the CD index does not depend on an arbitrarily chosen poverty line, (2) it depends on the observed and measurable consumption behaviour of people, (3) the index satisfies the standard axioms of a poverty index. Having used such a consumption deprivation function as a measure of poverty, this paper has shown analytically that for a log-normal income distribution, an increase in mean income, ceteris paribus, will decrease poverty while an increase in the variance of the income distribution, ceteris paribus, will increase poverty although somewhat contradictory information was obtained for the limiting case of earners with extremely low variance in their income distribution. In this case, poverty was found to decrease with increasing variance for a fixed mean, while when plotted against the mean (Fig. 3), it was found to initially increase and then saturate after a critical value of the mean which we could determine theoretically.
These observations were later contrasted with observations made from a pareto distribution. Here we found that for very low earning groups in a developing economy, poverty initially decreases with increasing variance but beyond a critical value of the variance, it starts increasing later to decrease again. In the process, this defines all three economies characterised by individual parametric regimes. The conclusion that we derive from these joint analyses is that the variance dependence of poverty is not unequivocally simplistic, in that one distribution (log-normal) predicts an increase in poverty with increasing variance (although the limiting $`\sigma ^20`$ case was somewhat qualitatively identical to zone B for the pareto distribution) while the other (pareto) shows the existence of an inflexion point in the poverty function. This means that the poverty-variance graph in a pareto distribution has a critical point, on one side (zone A) of which poverty decreases with increasing variance, while on the other side it is just the reverse.
Our contribution here is to prove that a pareto distribution offers the more realistic measure of poverty in a developing economy. This is because it condones the very realistic fact that for very low income groups a slight increase in the variance only serves to decrease poverty whereas for high earning groups, greater the variation in earning greater is the probability of an escalation in poverty up to another critical point, beyond which poverty declines with any further increase in variance of wealth distribution in a society. This phase seems to reflect the case of a very developed economy, one which we identify as the supra-economic behaviour. In macroeconomic sense, this phase suggests that close to an equilibrium dynamics, higher inequality could contribute to higher savings and thereby higher growth and reduced poverty. In a follow-up work , shortly to be communicated, we have shown that in the non-stationary case, where both income and consumption are functions of time, the consumption deprivation dynamics can be mapped to the paradigmatic Burgers’ equation, <sup>7</sup><sup>7</sup>7Burger’s equation is a 1+1 dimensional equation which generally represents the time change in the velocity of a fluid flowing under constant pressure thereby bestowing us with the ability to make quantitative predictions on the poverty of a developing economy as a function of income and time. |
warning/0507/math0507317.html | ar5iv | text | # A Continuous Field of 𝐶^∗-algebras and the Tangent Groupoid for Manifolds with Boundary
## Introduction
It is a central idea of semi-classical analysis to consider Planck’s constant $`\mathrm{}`$ as a small real variable and to study the relation between systems in mechanics and systems in quantum mechanics by associating to a function $`f=f(x,\xi )`$ on the cotangent bundle of a manifold the $`\mathrm{}`$-scaled pseudodifferential operator $`\mathrm{op}_{\mathrm{}}(f)`$ with symbol $`f(x,\mathrm{}\xi )`$ and analyzing their relation as $`\mathrm{}0`$.
For $`f𝒮(T^{}^n)`$, for example, a basic estimate states that
$$\underset{\mathrm{}0}{lim}\mathrm{op}_{\mathrm{}}(f)=f_{\mathrm{sup}}.$$
(0.1)
Moreover, given a second symbol $`g𝒮(T^{}^n)`$ we have
$$\underset{\mathrm{}0}{lim}\mathrm{op}_{\mathrm{}}(f)\mathrm{op}_{\mathrm{}}(g)\mathrm{op}_{\mathrm{}}(fg)=0;$$
(0.2)
in other words, the map $`\mathrm{op}_{\mathrm{}}`$ is asymptotically multiplicative.
As both statements concern the asymptotic behavior of pseudodifferential operators, it is somewhat surprising that they can be proven within the framework of continuous fields of $`C^{}`$-algebras associated to amenable Lie groupoids, more precisely, the $`C^{}`$-algebra of the so-called tangent groupoid $`𝒯M`$, cf. Connes \[6, Section II.5\].
For a boundaryless manifold $`M`$, $`𝒯M`$ is constructed by gluing the tangent space $`TM`$ to the cartesian product $`M\times M\times ]0,1]`$ via the map $`TM\times [0,1](m,v,\mathrm{})(m,\mathrm{exp}_m(\mathrm{}v),\mathrm{})`$. It has the natural cross-sections $`𝒯M(\mathrm{})`$, $`0\mathrm{}1`$, given by $`TM`$ for $`\mathrm{}=0`$ and by $`M\times M\times \{\mathrm{}\}`$ for $`\mathrm{}0`$.
The basic observation, establishing the link between $`\mathrm{}`$-scaled pseudodifferential operators and the tangent groupoid, is the following: In the Fourier transformed picture, the $`\mathrm{}`$-scaled pseudodifferential operator $`\mathrm{op}_{\mathrm{}}(f)`$ becomes the convolution operator $`\rho _{\mathrm{}}(\widehat{f})`$ acting by
$$\rho _{\mathrm{}}(\widehat{f})\xi (x)=\frac{1}{\mathrm{}^n}\widehat{f}(x,\frac{xy}{\mathrm{}})\xi (y)𝑑y,\xi L^2(^n),$$
and for $`\mathrm{}0`$, the mappings $`\rho _{\mathrm{}}`$ (or better their generalization to the manifold case) coincide with the natural representations of $`C_c^{\mathrm{}}(𝒯M(\mathrm{}))`$ by convolution operators.
The $`\rho _{\mathrm{}}`$, $`\mathrm{}0`$, are complemented by the representation $`\pi _0`$ of $`C_c^{\mathrm{}}(TM)`$ on $`L^2(TM)`$ via convolution in the fiber which in turn coincides with the natural representation of $`C_c^{\mathrm{}}(𝒯M(0)).`$
Now the tangent groupoid is additionally amenable, so that, according to a theorem by Anantharaman-Delaroche and Renault , the reduced $`C^{}`$-algebra $`C_r^{}(𝒯M)`$, defined as the closure of $`C_c^{\mathrm{}}(𝒯M)`$ with respect to the natural representations, and the full $`C^{}`$-algebra $`C^{}(𝒯M)`$, i.e., the closure with respect to all involutive Hilbert space representations, are isomorphic.
The crucial fact then is that $`C_r^{}(𝒯M)`$ is a continuous field of $`C^{}`$-algebras over $`[0,1]`$; the fiber over $`\mathrm{}`$ is $`C_r^{}(𝒯M(\mathrm{}))`$. An elegant way to establish the continuity is to show upper semi-continuity and lower semi-continuity separately, noticing that upper semi-continuity is easily proven in $`C^{}(𝒯M)`$ while lower semi-continuity is not difficult to show in $`C_r^{}(𝒯M)`$. As both $`C^{}`$-algebras are isomorphic, continuity follows. For a good account of these facts see by Landsman and Ramazan. The identities (0.1) and (0.2) are then an immediate consequence of the continuity of the field.
In the present paper we consider manifolds with boundary. The analog of the usual pseudodifferential calculus here is Boutet de Monvel’s calculus for boundary value problems . In order to obtain an operator algebra, one cannot work with pseudodifferential operators alone, but has to introduce an additional class of operators, the so-called singular Green operators. The reason is the way pseudodifferential operators act on functions defined on a half space: One first extends the function (by zero) to the full space, then applies the pseudodifferential operator and finally restricts the result to the half space again – one often speaks of trucated pseudodifferential operators. Given two pseudodifferential operators $`P`$ and $`Q`$, the ‘leftover operator’ $`L(P,Q)=(PQ)_+P_+Q_+`$, i.e. the difference between the trucated pseudodifferential operator $`(PQ)_+`$ associated to the composition $`PQ`$ and the composition of the truncated operators $`P_+`$ and $`Q_+`$ associated with $`P`$ and $`Q`$ is a typical example of such a singular Green operator. The singular Green operators ‘live’ at the boundary. They are smoothing operators in the interior, while, close to the boundary, they can be viewed as operator-valued pseudodifferential operators along the boundary, acting like smoothing operators in the normal direction.
In the full algebra which consists – at least in the slightly simplified picture we have here – of sums of (truncated) pseudodifferential operators and singular Green operators, the singular Green operators form an ideal.
With this picture in our mind, we construct an analog of Connes’ tangent groupoid for a manifold $`X`$ with boundary. Our semi-groupoid $`𝒯^{}X`$ consists of the groupoid $`X\times X\times ]0,1]`$ to which we glue, with the same map as above, the half-tangent space $`T^{}X`$, which comprises all those tangent vectors to $`X`$ for which $`\mathrm{exp}_m(\mathrm{}v)`$ lies in $`X`$ for small $`\mathrm{}`$ (note that this condition is only effective at the boundary of $`X`$). As before, we have natural cross-sections $`𝒯^{}X(\mathrm{})`$, coinciding with $`X\times X\times \{\mathrm{}\}`$ for $`\mathrm{}0`$ and with $`T^{}X`$ for $`\mathrm{}=0`$.
For $`\mathrm{}0`$, the operators $`\rho _{\mathrm{}}`$ (with integration now restricted to $`X`$), are the natural representations of the groupoid $`𝒯^{}X(\mathrm{})`$. At $`\mathrm{}=0`$ we use two mappings. The first, $`\pi _0`$ is the analog of the above map $`\pi _0`$. It acts on the tangent space of $`X`$ by convolution. The second, $`\pi _0^{}`$, acts on the half tangent space over the boundary by half-convolution: $`\pi _0^{}:C_c^{\mathrm{}}(T^{}X)(L^2(T^{}X|_X)`$ is given by
$$\pi _0^{}(f)\xi (m,v)=_{T_m^{}X}f(m,vw)\xi (m,w)𝑑w.$$
In order to avoid problems concerning the topology of $`𝒯^{}X`$, we denote by $`C_c^{\mathrm{}}(𝒯^{}X)`$ the space of all restrictions of functions in $`C_c^{\mathrm{}}(𝒯\stackrel{~}{X})`$ to $`𝒯^{}X`$; here $`\stackrel{~}{X}`$ is a boundaryless manifold containing $`X`$.
The reduced $`C^{}`$-algebra $`C_r^{}(𝒯^{}X)`$ is then defined as the $`C^{}`$-closure of $`C_c^{\mathrm{}}(𝒯^{}X)`$ with respect to the $`\rho _{\mathrm{}}`$, $`\mathrm{}0`$, and $`\pi _0,\pi _0^{}`$ for $`\mathrm{}=0`$. For the full $`C^{}`$-algebra we use all involutive representations.
We show that $`C_r^{}(𝒯^{}X)`$ is a continuous field of $`C^{}`$-algebras over $`[0,1]`$, where the fiber over $`\mathrm{}0`$ is $`C_r^{}(𝒯^{}X(\mathrm{}))`$, and the fiber over $`\mathrm{}=0`$ is the $`C^{}`$-closure of $`C_c^{\mathrm{}}(T^{}X)`$ with respect to $`\pi _0`$ and $`\pi _0^{}`$.
The proof of continuity is again split up into showing upper semi-continuity and lower semi-continuity. According to an idea by Rieffel , lower semi-continuity is established using strongly continuous representations. The basic idea for the proof of upper semi-continuity would be to infer an isomorphism between $`C_r^{}(𝒯^{}X)`$ and $`C^{}(𝒯^{}X)`$ from the amenability of $`𝒯^{}X`$. However, as $`T^{}X`$ is only a semi-groupoid, we make a little detour: Using short exact sequences and the amenability of the tangent groupoids for boundaryless manifolds we prove that $`C_r^{}(T^{}X)`$ is isomorphic to the closure of $`C_c^{\mathrm{}}(T^{}X)`$ with respect to the upper semi-continuous norm.
The present study should be seen as a step towards fitting Boutet de Monvel’s calculus for boundary value problems into the framework of deformation quantization and groupoids, in the spirit of Connes , Monthubert and Pierrot , Nest and Tsygan , , Nistor and Weinstein and Xu , Eventually one could hope to develop an algebraic index theory for these deformations in the spirit of Nest and Tsygan.
The structure of the paper is as follows: In the first section we review the case of boundaryless manifolds. We introduce the basic notions and show how (0.1) and (0.2) are derived with the help of the continuous field of $`C^{}`$-algebras associated to the tangent groupoid.
We then consider a manifold $`X`$ with boundary. In order to make the presentation more transparent, we first study the case where $`X=_+^n=\{(x_1,\mathrm{},x_n)|x_n0\}`$. Here all relevant features show up, but computations are easier to perform. We then go over to the general case.
In Section 3 we determine the $`K`$-theory of the symbol algebra $`C_r^{}(T^{}X)`$. Starting from the short exact sequence
$$0C_r^{}(TX^{})C_r^{}(T^{}X)Q0$$
we show that the quotient $`Q`$ can be identified with $`C_0(T^{}X)𝒯_0`$, where $`𝒯_0`$ is an ideal in the Toeplitz algebra with vanishing $`K`$-theory. In particular, we obtain the isomorphism
$$K_i(C_r^{}(T^{}X))K_i(C_0(T^{}X)),i=0,1.$$
The appearance of the Toeplitz operators can be seen as a feature inherent in the geometry of the problem. In fact, the construction of an algebra of pseudodifferential operators on a closed (Riemannian) manifold amounts to the construction of a suitably completed operator algebra, generated by multivariable functions of vector fields and the operators of multiplication by smooth functions.
In the boundaryless case, one can localize to $`^n`$ and reduce the task essentially to defining $`f(D)`$ for a classical symbol $`f`$ and $`D=(D_1,\mathrm{},D_n)`$ with the vector fields $`D_j=i_{x_j}`$. One convenient way of achieving this is to use the operator families $`e^{itD_j}`$ and to let
$$f(D)=(2\pi )^n\widehat{f}(\xi )e^{i\xi D}𝑑\xi $$
with the Fourier transform $`\widehat{f}`$ of $`f`$ and $`\xi D=\xi _1D_1+\mathrm{}+\xi _nD_n`$. Note that the use of the $`e^{i\xi D}`$ is purely geometric and only relies on the fact that vector fields integrate to flows.
On a manifold with boundary, one will have vector fields transversal to the boundary which do not integrate to flows. In this case, one has two possibilities: The first is to restrict the class of admissible vector fields to those which do integrate. This is a basic idea in the pseudodifferential calculi introduced by Melrose , see also Ammann, Lauter, Nistor .
In Boutet de Monvel’s calculus, on the other hand, transversal vector fields are admitted. After localization to $`\overline{}_+^n`$, we may focus on $`D_n`$. One of the functions one would certainly like to define is the Cayley transform (recall that the Cayley transform $`C(A)`$ of an operator $`A`$ is given by $`C(A)=(Ai)(A+i)^1=12i(A+i)^1`$).
Now it is well-known that the Cayley transform $`C(A)`$ is an isometry, and that it is a unitary if and only if $`A`$ is selfadjoint. As there is no selfadjoint extension of $`D_n`$, its Cayley transform will be a proper isometry. Hence by a theorem of Coburn , the algebra generated by it (which becomes part of the calculus), is the Toeplitz algebra.
While the pseudodifferential calculus for closed manifolds is commutative modulo lower order terms, this calculus is not. From a geometric point of view, the resulting algebra can thus be seen as a noncommutative completion of the manifold with boundary.
Remark on the notation. A variety a representations naturally comes up in this context. In order to distinguish their origin, we will apply the following rule. Representations related to the groupoid structure are denoted by $`\pi `$ (possibly indexed), asymptotic pseudodifferential operators by $`\rho _{\mathrm{}}`$ and the asymptotic Green operators (introduced in Section 2) by $`\kappa _{\mathrm{}}`$.
## 1 The Classical Case
### Groupoids
A groupoid $`G`$ is a small category where all the morphisms are invertible. We will denote by $`G^{(0)}`$ the set of objects in $`G`$ and by $`G^{(1)}`$ the set of morphisms. We will also call $`G^{(0)}`$ the base and the elements in $`G^{(1)}`$ the arrows. On $`G^{(1)}`$ there are two maps $`r,s`$ into $`G^{(0)}`$. The first map, $`r`$, is the range object of a morphism and the second, $`s`$, the source. For $`xG^{(0)}`$ we define $`G^x=r^1(x)`$ and $`G_x=s^1(x)`$. There is an embedding $`\iota `$ of $`G^{(0)}`$ into $`G^{(1)}`$ given by mapping an object to the identity morphism on this object. Furthermore we define $`G^{(2)}`$ to be the subset of composable morphisms of $`G^{(1)}\times G^{(1)}`$.
###### 1.1
Definition. A Lie groupoid $`G`$ is a groupoid together with a manifold structure on $`G^{(0)}`$ and $`G^{(1)}`$ such that the maps $`r,s`$ are submersions, the map $`\iota `$ and the the composition map $`G^{(2)}G^{(1)}`$ are smooth.
To a given a smooth manifold $`M`$ without boundary there are associated two canonical Lie groupoids. The first is the tangent bundle $`TM`$ of $`M`$. The groupoid structure is given by
$`G^{(0)}=M,`$ $`G^{(1)}=TM`$
$`r(m,X)=m,`$ $`s(m,X)=m`$
$`(m,X)(m,Y)`$ $`=`$ $`(m,X+Y)`$
The second one is the pair groupoid $`M\times M`$ with
$`G^{(0)}=M,`$ $`G^{(1)}=M\times M`$
$`r(m_1,m_2)=m_1,`$ $`s(m_1,m_2)=m_2`$
$`(m_1,m_2)(m_2,m_3)`$ $`=`$ $`(m_1,m_3)`$
Both are clearly Lie groupoids.
###### 1.2
Haar systems. A smooth left Haar system on a Lie groupoid is a family of measures $`\{\lambda ^x\}_{xG^{(0)}}`$ on $`G`$ with $`\text{supp}\lambda ^x=G^x`$ which is left invariant, i.e. $`\gamma (\lambda ^{s(\gamma )})=\lambda ^{r(\gamma )}`$, and for each $`fC_c^{\mathrm{}}(G^{(1)})`$, the function on $`G^{(0)}`$ defined by
$$xf𝑑\lambda ^x,fC_c^{\mathrm{}}(G^{(1)})$$
is smooth. In \[10, Proposition 3.4\], it is proven that all Lie groupoids possess a smooth left Haar system. Similarly, a right Haar system $`\{\lambda _x\}`$ is given by $`\lambda _x=(\lambda ^x)^1`$.
###### 1.3
Definition. A Lie groupoid $`G`$ with a smooth left Haar system $`\lambda ^x`$ is called topologically amenable if there exists a net of nonnegative continuous functions $`\{f_i\}`$ on $`G^{(1)}`$ such that
1. For all $`i`$ and for $`xG^{(0)}`$, $`f_i𝑑\lambda ^x=1`$.
2. The functions $`\gamma |f_i(\gamma ^1\gamma ^{})f_i(\gamma ^{})|𝑑\lambda ^{r(\gamma )}(\gamma ^{})`$ converge uniformly to zero on compact subsets of $`G^{(1)}`$.
It is easy to verify that the two groupoids $`TM`$ and $`M\times M`$ are topologically amenable.
###### 1.4
Connes’ tangent groupoid. Let $`M`$ be a smooth manifold. Connes tangent groupoid $`𝒯M`$ is a blow up of the diagonal in $`M\times M`$. More specifically:
Let $`𝒯M=TM(M\times M\times ]0,1])`$ as a set. The groupoid structure is just the fiberwise groupoid structure coming from the groupoid structure on $`TM`$ and $`M\times M`$. The manifold structure on $`M\times M\times ]0,1]`$ is obvious. We next glue $`TM`$ to $`M\times M\times ]0,1]`$ to get a manifold structure on $`𝒯M`$. To this end we choose a Riemannian metric on $`M`$ and glue with the charts
$$TM\times [0,1]U(m,v,\mathrm{})\{\begin{array}{cc}(m,v)& \text{for }\mathrm{}=0\\ (m,\mathrm{exp}_m(\mathrm{}v),\mathrm{})& \text{for }\mathrm{}0,\end{array}$$
where $`\mathrm{exp}_m`$ denotes the exponential map and $`U`$ is an open neighborhood of $`M\times \{0\}TM\times \{0\}`$; here, $`M`$ is embedded as the zero section.
Here, $`G^{(0)}=M\times [0,1]`$. For $`\mathrm{}0`$ and $`x=(\stackrel{~}{m},\stackrel{~}{\mathrm{}})G^{(0)}`$, we have $`G^x=\{(\stackrel{~}{m},m,\stackrel{~}{\mathrm{}}):mM\}`$; for $`x=(\stackrel{~}{m},0)`$, $`G^x=T_{\stackrel{~}{m}}M`$. Fixing the measure $`\mu `$ on $`M`$ induced by the metric, we obtain a Haar system $`\{\lambda ^x\}_{xG^{(0)}}`$ by $`\lambda ^{(\stackrel{~}{m},\stackrel{~}{\mathrm{}})}=\mathrm{}^n\mu `$, $`\mathrm{}0`$; for $`\mathrm{}=0`$, we let $`\lambda ^{(\stackrel{~}{m},0)}`$ be the measure on $`T_mM`$ given by the metric.
This makes $`𝒯M`$ a Lie groupoid, see .
###### 1.5
C\*-algebras associated to groupoids. Let $`G`$ be a Lie groupoid with a smooth left Haar system $`\lambda `$. On $`C_c^{\mathrm{}}(G^{(1)})`$ we define a $``$-algebra structure by
$`(fg)(\gamma )`$ $`=`$ $`{\displaystyle _{G^{s(\gamma )}}}f(\gamma \gamma _1)g(\gamma _1^1)𝑑\lambda ^{s(\gamma )}(\gamma _1)`$ (1.3)
$`f^{}(\gamma )`$ $`=`$ $`\overline{f(\gamma ^1)}`$ (1.4)
There are involutive representations $`\pi _x`$, $`xG^{(0)}`$, of this $``$-algebra on the Hilbert spaces $`L^2(G_x,\lambda _x)`$ given by
$$\pi _x(f)\xi (\gamma )=_{G^x}f(\gamma \gamma _1)\xi (\gamma _1^1)𝑑\lambda ^x(\gamma _1),\xi L^2(G_x,\lambda _x).$$
(1.5)
###### 1.6
Definition. The full $`C^{}`$-algebra $`C^{}(G)`$ of a groupoid is the $`C^{}`$-completion of the $``$-algebra $`C_c^{\mathrm{}}(G^{(1)})`$ with respect to all involutive Hilbert space representations.
The reduced $`C^{}`$-algebra $`C_r^{}(G)`$ of $`G`$ is the $`C^{}`$-completion of $`C_c^{\mathrm{}}(G)`$ with respect to the representations (1.5).
Note that, by universality, we have a quotient map from $`C^{}(G)`$ to $`C_r^{}(G)`$.
###### 1.7
Remark. Although the construction of the $``$-algebra structure on $`C_c^{\mathrm{}}(G^{(1)})`$ and the representations (1.5) use a smooth Haar system, the algebra is independent of the choice. See for a detailed exposition.
###### 1.8
Example. For the tangent bundle $`TM`$ of a manifold, the space $`G_m`$ is just $`T_mM`$ and the representation is
$$\pi _m(f)\xi (v)=_{T_mM}f(m,vw)\xi (w)𝑑w\xi L^2(T_mM).$$
By Fourier transform in each fiber $`T_mM`$, the $`C^{}`$-algebra $`C_r^{}(TM)`$ becomes isomorphic to $`C_0(T^{}M)`$, the continuous functions on $`T^{}M`$ vanishing at infinity.
The importance of topological amenability lies in the following result from :
###### 1.9
Proposition. When $`G`$ is topologically amenable the quotient map from $`C^{}(G)`$ to $`C_r^{}(G)`$ is an isomorphism.
### Continuous Fields and $`\mathrm{}`$-Scaled Pseudodifferential Operators
###### 1.10
Definition. A continuous field of $`C^{}`$-algebras $`(A,\{A(\mathrm{}),\phi _{\mathrm{}}\}_{\mathrm{}[0,1]})`$ over $`[0,1]`$ consists of a $`C^{}`$-algebra $`A`$, $`C^{}`$-algebras $`A(\mathrm{})`$, $`\mathrm{}[0,1]`$, with surjective homomorphisms $`\phi _{\mathrm{}}:AA(\mathrm{})`$ and an action of $`C([0,1])`$ on $`A`$ such that for all $`aA`$
1. The function $`\mathrm{}\phi _{\mathrm{}}(a)`$ is continuous;
2. $`a=sup_{\mathrm{}[0,1]}\phi _{\mathrm{}}(a)`$;
3. For $`fC([0,1])`$, $`\phi _{\mathrm{}}(fa)=f(\mathrm{})\phi _{\mathrm{}}(a)`$.
###### 1.11
Theorem. For the tangent groupoid $`𝒯M`$ we define $`𝒯M(0)=TM`$ and $`𝒯M(\mathrm{})=M\times M\times \{\mathrm{}\}`$ for $`\mathrm{}0`$. The pullback under the inclusion $`𝒯M(\mathrm{})𝒯M`$ induces a map $`\phi _{\mathrm{}}:C_c^{\mathrm{}}(𝒯M)C_c^{\mathrm{}}(𝒯M(\mathrm{}))`$ which extends by continuity to a surjective $``$-homomorphism $`\phi _{\mathrm{}}:C_r^{}(𝒯M)C_r^{}(𝒯M(\mathrm{}))`$. The $`C^{}`$-algebras $`A=C_r^{}(𝒯M)`$ and $`A(\mathrm{})=C_r^{}(𝒯M(\mathrm{}))`$ with the maps $`\phi _{\mathrm{}}`$ form a continuous field over $``$.
Proof. Together with the amenability of $`𝒯M`$ and Proposition 1.9 this is immediate from Theorem 6.4 in $``$
###### 1.12
$`\mathrm{}`$-scaled pseudodifferential operators. For $`0<\mathrm{}1`$ define $`\rho _{\mathrm{}}:C_c^{\mathrm{}}(T^n)(L^2(^n))`$ by
$`\rho _{\mathrm{}}(f)\xi (x)`$ $`=`$ $`{\displaystyle f(x,w)\xi (x\mathrm{}w)𝑑w}=\mathrm{}^n{\displaystyle f(x,\frac{xw}{\mathrm{}})\xi (w)𝑑w},\xi L^2(^n)`$ (1.6)
We complement this by the map $`\pi _0:C_c^{\mathrm{}}(T^n)(L^2(T^n))`$
$`\pi _0(f)\xi (x,v)`$ $`=`$ $`{\displaystyle f(x,w)\xi (x,vw)𝑑w}.`$ (1.7)
###### 1.13
Remark. (a) We can define $`\stackrel{~}{\rho }_{\mathrm{}}:C_c^{\mathrm{}}(T^n)(L^2(T^n))`$, $`\mathrm{}0`$, by
$$\stackrel{~}{\rho }_{\mathrm{}}(f)\xi (x,v)=f(x,w)\xi (x\mathrm{}w,vw)𝑑w$$
and then obtain a more consistent representation. Note that for $`h>0`$ the representations $`\rho _{\mathrm{}}`$ and $`\stackrel{~}{\rho }_{\mathrm{}}`$ are unitarily equivalent.
(b) On a smooth Riemannian manifold $`M`$ we define $`\rho _{\mathrm{}}`$ by
$`(\rho _{\mathrm{}}f)\xi (x)`$ $`=`$ $`{\displaystyle \psi (x,\mathrm{exp}_x(\mathrm{}w))f(x,w)\xi (\mathrm{exp}_x(\mathrm{}w))𝑑w}`$ (1.8)
$`=`$ $`\mathrm{}^n{\displaystyle \psi (x,y)f(x,\mathrm{exp}^1(x,y)/\mathrm{})\xi (y)𝑑y}.`$
Here $`\psi C^{\mathrm{}}(M\times M)`$ is a function, which is one on a neighborhood of the diagonal, $`0\psi 1`$ and such that
$$\mathrm{exp}:TMM\times M,(m,v)(m,\mathrm{exp}_mv),$$
maps a neighborhood of the zero section diffeomorphically to the support of $`\psi `$; a similar construction applies to $`\stackrel{~}{\rho }`$.
Note that for two representations $`\rho _{\mathrm{}}^1,\rho _{\mathrm{}}^2`$, defined with cut-off functions $`\psi _1`$ and $`\psi _2`$, the norm $`\rho _{\mathrm{}}^1(f)\rho _{\mathrm{}}^2(f)`$ tends to zero as $`\mathrm{}0`$.
###### 1.14
Lemma. To each $`fC_c^{\mathrm{}}(TM)`$ we associate a function $`\stackrel{~}{f}C^{\mathrm{}}(𝒯M)`$ by
$$\begin{array}{cc}\stackrel{~}{f}(x,v,0)=f(x,v)\hfill & \text{ for }\mathrm{}=0,xM,vT_xM;\hfill \\ \stackrel{~}{f}(x,y,\mathrm{})=\psi (x,y)f(x,\mathrm{exp}^1(x,y)/\mathrm{})\hfill & \text{ for }\mathrm{}0,x,yM.\hfill \end{array}$$
By (1.8)
$$\pi _{(x,\mathrm{})}(\stackrel{~}{f})=\rho _{\mathrm{}}(f)\text{ and }\phi _{\mathrm{}}(\stackrel{~}{f})_{𝒯M(\mathrm{})}=\underset{x}{sup}\pi _{(x,\mathrm{})}(\stackrel{~}{f})_{(L^2(G_{(x,\mathrm{})},\lambda _{(x,\mathrm{})}))}=\rho _{\mathrm{}}(f).$$
###### 1.15
Theorem. We denote by $`\widehat{f}`$ the Fourier transform of $`f`$ with respect to the covariable. Then
(a) $`lim_{h0}\rho _h(f)=\widehat{f}_{sup}`$.
(b) $`lim_{h0}\rho _h(f)\rho _h(g)\rho _h(fg)=0`$.
Proof. We have
$$\underset{\mathrm{}0}{lim}\rho _{\mathrm{}}(f)=\underset{\mathrm{}0}{lim}\phi _{\mathrm{}}(\stackrel{~}{f})=\phi _0(\stackrel{~}{f})=\pi _0(f)=\widehat{f}_{sup}$$
and, for arbitrary $`x`$,
$`\rho _{\mathrm{}}(f)\rho _{\mathrm{}}(g)\rho _{\mathrm{}}(fg)`$ $`=`$ $`\pi _{(x,\mathrm{})}(\stackrel{~}{f})\pi _{(x,\mathrm{})}(\stackrel{~}{g})\pi _{(x,\mathrm{})}(\stackrel{~}{fg})`$
$`=`$ $`\pi _{(x,\mathrm{})}(\stackrel{~}{f}\stackrel{~}{g}\stackrel{~}{fg})\pi _0(\stackrel{~}{f}\stackrel{~}{g}\stackrel{~}{fg})=0.`$
$``$
## 2 Manifolds with Boundary
In the following, we shall denote by $`X`$ a smooth $`n`$-dimensional manifold with boundary, $`X`$. We assume that $`X`$ is embedded in a boundaryless manifold $`\stackrel{~}{X}`$ and write $`X^{}`$ for the interior of $`X`$. We also fix a Riemannian metric on $`X`$, so that we have $`L^2`$ spaces. We will show later on that the construction is independent of the choice of the metric. First of all, however, it is helpful to study the case where $`X=_+^n=\{(x_1,\mathrm{},x_n)|x_n0\}`$ (including $`x_n=0`$!). We adopt the usual notation by writing an element $`x_+^n`$ as $`x=(x^{},x_n)`$.
### Local Computation of the Asymptotic Green Term
We change the formula for the $`\mathrm{}`$-scaled boundary pseudodifferential operators with Fourier transformed symbol $`fC_c^{\mathrm{}}(T_+^n)`$ to
$`\rho _{\mathrm{}}(f)\xi (x)`$ $`=`$ $`{\displaystyle _{x_n\mathrm{}v_n}}f(x,v)\xi (x\mathrm{}v)𝑑v`$ (2.1)
$`=`$ $`\mathrm{}^n{\displaystyle _{w_n0}}f(x,{\displaystyle \frac{xw}{\mathrm{}}})\xi (w)𝑑w,\xi L^2(_+^n).`$
A straightforward computation shows that
$`(\rho _{\mathrm{}}(f)\rho _{\mathrm{}}(g)\xi )(x)`$ $`=`$ $`{\displaystyle _{x_n\mathrm{}w_n}}\left({\displaystyle _{x_n\mathrm{}v_n}}f(x,v)g(x\mathrm{}v,wv)𝑑v\right)\xi (x\mathrm{}w)𝑑w`$
$`=`$ $`{\displaystyle _{x_n\mathrm{}w_n}}({\displaystyle }f(x,v)g(x\mathrm{}v,wv)dv`$
$`{\displaystyle _{x_n\mathrm{}v_n}}f(x,v)g(x\mathrm{}v,wv)dv)\xi (x\mathrm{}w)dw,`$
where in the last line $`f,g`$ have to be understood as extended to functions in $`C_c^{\mathrm{}}(T^n)`$. The term
$`f_hg={\displaystyle f(x,v)g(x\mathrm{}v,wv)𝑑v}`$ (2.2)
is just the usual composition of Fourier transformed symbols of pseudodifferential operators on manifolds without boundary. We call the remainder, i.e. the operator which maps $`\xi `$ to
$`x`$ $``$ $`{\displaystyle _{x_n\mathrm{}w_n}}{\displaystyle _{x_n\mathrm{}v_n}}f(x,v)g(x\mathrm{}v,wv)𝑑v\xi (x\mathrm{}w)𝑑w`$ (2.3)
$`=`$ $`{\displaystyle _{y_n0}}{\displaystyle _{x_n\mathrm{}v_n}}f(x,v)g(x\mathrm{}v,y^{}v^{},{\displaystyle \frac{x_n}{\mathrm{}}}y_nv_n)𝑑v\xi (x^{}\mathrm{}y^{},\mathrm{}y_n)𝑑y`$
the “asymptotic Green” term, because it corresponds to the leftover term in the composition of two truncated pseudodifferential operators in Boutet de Monvel’s calculus, which is a singular Green operator, cf. . In order to analyze it, we introduce the following notation:
###### 2.1
Definition. For $`0<\mathrm{}1`$ define
$$\kappa _{\mathrm{}}:C_c^{\mathrm{}}(T^{n1}\times _+\times _+\times [0,1])(L^2(_+^n))$$
by
$$\kappa _{\mathrm{}}(K)\xi (x)=_{y_n0}K(x^{},y^{},\frac{x_n}{\mathrm{}},y_n,\mathrm{})\xi (x^{}\mathrm{}y^{},\mathrm{}y_n)𝑑v.$$
The asymptotic Green thus is of the form $`\kappa _{\mathrm{}}(l_{\mathrm{}}(f,g))`$ with
$`l_{\mathrm{}}(f,g)(x^{},y^{},x_n,y_n)`$ (2.4)
$`=`$ $`{\displaystyle _{x_nv_n}}f(x^{},\mathrm{}x_n,v)g(x^{}\mathrm{}v^{},\mathrm{}(x_nv_n),y^{}v^{},x_ny_nv_n)𝑑v.`$
As $`\mathrm{}0`$ this tends to
$`l(f,g)(x^{},y^{},x_n,y_n)={\displaystyle _{x_nv_n}}f(x^{},0,y^{}v^{},v_n)g(x^{},0,v^{},x_nv_ny_n)𝑑v.`$ (2.5)
In fact, the difference $`l_{\mathrm{}}(f,g)l(f,g)`$ is an element of $`C_c^{\mathrm{}}(T^{n1}\times _+\times _+\times [0,1])`$ which vanishes for $`\mathrm{}=0`$. Similarly, the difference $`f_{\mathrm{}}gfgC_c^{\mathrm{}}(T_+^n\times [0,1])`$ vanishes for $`\mathrm{}=0`$.
In order to extend Theorem 1.15 to manifolds with boundary, the asymptotic Green term has to be taken into account.
###### 2.2
Definition. For $`\mathrm{}=0`$ we introduce
$$\pi _0^{}:C_c^{\mathrm{}}(T_+^n\times [0,1])C_c^{\mathrm{}}(T^{n1}\times _+\times _+\times [0,1])(L^2(T^{n1}\times _+))$$
given by
$`\pi _0^{}(fK)\xi (x^{},v^{},v_n)=`$
$`=`$ $`{\displaystyle _{w_n0}}\left(f(x^{},0,v^{}w^{},v_nw_n,0)+K(x^{},v^{}w^{},v_n,w_n,0)\right)\xi (x^{},w^{},w_n)𝑑w`$
We complement $`\pi _0^{}`$ by the map $`\pi _0:C_c^{\mathrm{}}(T_+^n)(L^2(T_+^n))`$ in (1.7).
The crucial point is:
###### 2.3
Lemma. The map
$$(\pi _0,\pi _0^{}):C_c^{\mathrm{}}(T_+^n)C_c^{\mathrm{}}(T^{n1}\times _+\times _+)(L^2(T_+^n)L^2(T_+^{n1}\times _+))$$
given by
$$(\pi _0,\pi _0^{})(fK)=(\pi _0(f),\pi _0^{}(fK))$$
turns $`C_c^{\mathrm{}}(T_+^n)C_c^{\mathrm{}}(T_+^{n1}\times _+\times _+)`$ into an algebra. We denote this product with $`^{}`$. Note that $`f^{}g=fg+l(f,g)`$.
It is clear that Theorem 1.15(b) will not remain true literally. Instead we obtain:
###### 2.4
Theorem. For two symbols $`f,gC_c^{\mathrm{}}(T_+^n)`$ the following holds
$$\underset{\mathrm{}0}{lim}\rho _{\mathrm{}}(f)\rho _{\mathrm{}}(g)\rho _{\mathrm{}}(fg)\kappa _{\mathrm{}}(l(f,g))=0.$$
As in the case of manifolds without boundary, this will be related to the continuity of a field of $`C^{}`$-algebras which we will now introduce
###### 2.5
Definition. We denote by $`A`$ the $`C^{}`$-completion of
$$A^{\mathrm{}}=C_c^{\mathrm{}}(T_+^n\times [0,1])C_c^{\mathrm{}}(T^{n1}\times _+\times _+\times [0,1])$$
in the representation $`\rho _{\mathrm{}}+\kappa _{\mathrm{}}`$, for $`\mathrm{}0`$ and $`(\pi _0,\pi _0^{})`$ for $`\mathrm{}=0`$, i.e. under the mappings
$$fK\{\begin{array}{cc}\rho _{\mathrm{}}(f)+\kappa _{\mathrm{}}(K),\hfill & \mathrm{}0;\hfill \\ \pi _0(f)\pi _0^{}(fK)),\hfill & \mathrm{}=0.\hfill \end{array}$$
There are obvious maps
$$\phi _{\mathrm{}}:AA(\mathrm{}),$$
where $`A(\mathrm{})`$ is the completion of $`C_c^{\mathrm{}}(T_+^n)C_c^{\mathrm{}}(T^{n1}\times _+\times _+)`$ with respect to the specific representation in $`\mathrm{}`$.
We will show:
###### 2.6
Theorem. The triple $`(A,\{A(\mathrm{}),\phi _{\mathrm{}}\}_{\mathrm{}[0,1]})`$ is a continuous field of $`C^{}`$-algebras with $`A(\mathrm{})`$ isomorphic to the compact operators for $`\mathrm{}0`$.
For fixed $`\mathrm{}0`$, the operators $`\rho _{\mathrm{}}(f)+\kappa _{\mathrm{}}(K)`$ are compact, because they are integral operators with a square integrable kernel, so $`A(\mathrm{})`$ is isomorphic to the compact operators.
We shall next analyze the field in more detail. We abbreviate
$$T=T_+^n\text{ and }𝒯_{}=T^{n1}\times _+\times _+$$
and start with the following observation:
###### 2.7
Proposition. As a subset of $`A`$, $`A^{\mathrm{}}`$ is a is a $``$-algebra.
Proof. First we prove closure under multiplication. The product of $`K_1,K_2C_c^{\mathrm{}}(𝒯_{}\times [0,1])`$ is just the convolution product of the two functions on the groupoid $`𝒯^{n1}\times _+\times _+`$, thus again a function in $`C_c^{\mathrm{}}(𝒯_{}\times [0,1])`$.
For $`f,gC_c^{\mathrm{}}(T)`$ we have already computed, cf. (2.2), (2.4):
$$\rho _{\mathrm{}}(f)\rho _{\mathrm{}}(g)=\rho _{\mathrm{}}(\stackrel{~}{f}_{\mathrm{}}\stackrel{~}{g})+\kappa _{\mathrm{}}(l_{\mathrm{}}(\stackrel{~}{f},\stackrel{~}{g})).$$
where $`\stackrel{~}{f},\stackrel{~}{g}`$ are smooth extensions of $`f,g`$ to functions in $`C_c^{\mathrm{}}(T^n\times [0,1])`$. Since
$$(\pi _0,\pi _0^{})(f)(\pi _0,\pi _0^{})(g)=(\pi _0(fg),\pi _0^{}(fg)+\pi _0^{}(l(f,g))$$
we see the closure under products of $`f,g`$.
Checking the closure under products of $`f`$’s with $`K`$’s is straightforward. The same is true for the closure under involution. $``$
### The Algebra in 0
The algebra in zero, $`A(0),`$ is the completion of
$$A(0)^{\mathrm{}}:=(C_c^{\mathrm{}}(T)C_c^{\mathrm{}}(𝒯_{}),^{})$$
in the representation $`(\pi _0,\pi _0^{})`$. The summand $`C_c^{\mathrm{}}(𝒯_{})`$ becomes an ideal in $`A(0)^{\mathrm{}}`$. We thus get the short exact sequence
$$0C_c^{\mathrm{}}(𝒯_{})A(0)^{\mathrm{}}\stackrel{q}{}C_c^{\mathrm{}}(T)0.$$
(2.6)
As noted in the proof of Proposition 2.7, the algebra structure on $`C_c^{\mathrm{}}(𝒯_{})`$ comes from the groupoid structure on $`𝒯_{}`$, where $`_+\times _+`$ carries the pair groupoid structure. Likewise, the algebra structure on $`C_c^{\mathrm{}}(T)`$ stems from the groupoid structure on $`T`$. Note that both groupoids are amenable.
###### 2.8
Lemma. We have a short exact sequence of $`C^{}`$-algebras
$$0C_r^{}(𝒯_{})A(0)C_r^{}(T)0.$$
(2.7)
Proof. In the short exact sequence (2.6), the projection $`q`$, mapping $`fK`$ to $`f`$, is a $``$-homomorphism. The trivial estimate
$$\pi _0(f)_{(L^2(T))}\pi _0(f)\pi _0^{}(fK)_{(L^2(T)L^2(T^{n1}\times _+))},$$
shows that $`\pi `$ extends to a map $`A(0)C_r^{}(T)`$ with $`C_r^{}(𝒯_{})`$ in its kernel. Since we may estimate the norm of $`\pi _0^{}(f)`$ by the norm of $`\pi _0(f)`$, we obtain (2.7). $``$
Alternatively, the lemma may be proven using only the amenability of the groupoids, similarly as in the proof of Theorem 2.11, below. Note that, via the Fourier transform,
$$C_r^{}(𝒯_{})C_0(T^{}^{n1})𝒦(L^2(_+))$$
and
$$C_r^{}(T)C_0(T^{}_+^n).$$
### Upper Semi-continuity
###### 2.9
Definition. On $`A`$ we define
$$a_{as}=\mathrm{max}(\underset{\mathrm{}0}{lim\; sup}\phi _{\mathrm{}}(a),\phi _0(a)).$$
This is a $`C^{}`$-seminorm which is continuous with respect to the norm of $`A`$. The quotient
$$A[0]=A/I,\text{where}I=\{aA|a_{as}=0\},$$
therefore carries two norms: the quotient norm and $`_{as}`$. Both are equivalent by \[7, Proposition 1.8.1\], so that $`A[0]`$ is a $`C^{}`$-algebra with norm $`_{as}`$.
Since $`a_{as}\phi _0(a)`$ we have a natural map
$$\mathrm{\Phi }:A[0]A(0).$$
###### 2.10
Lemma. Elements in $`A^{\mathrm{}}`$ which are $`0`$ for $`\mathrm{}=0`$ belong to $`I`$.
Proof. For $`fKA^{\mathrm{}}`$ it is easy to estimate
$$\rho _{\mathrm{}}(f)M_ff(,\mathrm{})_{\mathrm{}}\text{ and }\kappa _{\mathrm{}}(K)M_KK(,\mathrm{})_{\mathrm{}},$$
where $`M_f`$ and $`M_K`$ are constants depending on $`f`$ and $`K`$, respectively, but not on $`\mathrm{}`$. $``$
###### 2.11
Theorem. The field $`(A,\{A(\mathrm{}),\phi _{\mathrm{}}\}_{\mathrm{}[0,1]})`$ is upper semi-continuous in $`0`$.
Proof. We denote by $`R`$ the closure of the range of the natural map $`\gamma :C_c^{\mathrm{}}(𝒯_{})A[0]`$. This is an ideal in $`A[0]`$: Indeed, $`C_c^{\mathrm{}}(𝒯_{})`$ is an ideal in $`A(0)^{\mathrm{}}`$, and the extension (e.g. constant in $`\mathrm{}`$) of functions in $`A(0)^{\mathrm{}}`$ to functions in $`A^{\mathrm{}}`$ furnishes an embedding of $`A(0)^{\mathrm{}}`$ into $`A[0]`$ with dense range.
Since $`𝒯_{}`$ is amenable, the quotient map $`C^{}(𝒯_{})C_r^{}(𝒯_{})`$ is an isomorphism. It factorizes through $`R`$, since $`R`$ gives us a Hilbert space representation of $`𝒯_{}`$, while $`a_{as}\phi _0(a)`$.
This leads to a commutative diagram of natural maps
$`C^{}(𝒯_{})`$
$``$ $``$
$`C_c^{\mathrm{}}(𝒯_{})`$ $`\stackrel{}{}`$ $`RA[0],`$
$``$ $``$
$`C_r^{}(𝒯_{})`$
where the upper vertical arrow is surjective, since the inclusion has dense range. The invertibility of the quotient map implies that the lower vertical arrow is an isomorphism.
Next we define a map $`\stackrel{~}{q}:A[0]C_r^{}(T)`$: By definition, $`A[0]`$ is the set of equivalence classes of Cauchy sequences in $`A^{\mathrm{}}`$ with respect to $`_{as}`$. Given such a Cauchy sequence $`a_k=(f_kK_k)`$, we may evaluate at $`\mathrm{}=0`$ and obtain a sequence $`(f_k^0K_k^0)`$ in $`A(0)^{\mathrm{}}`$. As $`a_k_{as}\phi _0(a_k)`$, the sequence $`(f_k^0)`$ is a Cauchy sequence in $`C_r^{}(T)`$; moreover, the mapping $`(a_k)(f_k^0)`$ is well-defined and continuous. In view of Lemma 2.10 its kernel is $`R`$.
Combining this with the short exact sequence (2.7) we obtain the following commutative diagram of short exact sequences
$$\begin{array}{ccccccccc}0& & C_r^{}(𝒯_{})& & A[0]& \stackrel{\stackrel{~}{q}}{}& C_r^{}(T)& & 0\\ & & & & \mathrm{\Phi }& & & & \\ 0& & C_r^{}(𝒯_{})& & A(0)& & C_r^{}(T)& & 0\end{array}.$$
(2.8)
We conclude from the five lemma that $`\mathrm{\Phi }`$ is an isomorphism, and therefore
$$\underset{\mathrm{}0}{lim\; sup}\phi _{\mathrm{}}(a)\phi _0(a),$$
i.e. the field is upper semi-continuous in $`0`$$``$
What is still missing is the proof of the lower semi-continuity of the field $`A`$. It will be given at the end of Section 2, since there is no simplification for the half-space case.
### The Tangent Groupoid for a Manifold with Boundary
###### 2.12
Definition. We denote by $`T^{}X`$ the subset of $`T\stackrel{~}{X}`$ formed by all vectors $`(m,v)T\stackrel{~}{X}|_X`$ for which $`\mathrm{exp}_m(\epsilon v)X`$ for sufficiently small $`\epsilon >0`$. This is a semi-groupoid with addition of vectors. Note that $`T^{}X=TX^{}T^{}X|_X`$
We define $`𝒯^{}X`$ as the disjoint union $`T^{}X(X\times X\times ]0,1])`$, endowed with the fiberwise semi-groupoid structure induced by the semi-groupoid structure on $`T^{}X`$ and the groupoid structure on $`X\times X`$. As in the boundaryless case, we glue $`T^{}X`$ to $`X\times X\times ]0,1]`$ via the charts
$$T^{}X\times [0,1]U(m,v,\mathrm{})\{\begin{array}{cc}(m,v)& \text{for }\mathrm{}=0\\ (m,\mathrm{exp}_m(\mathrm{}v),\mathrm{})& \text{for }\mathrm{}0\end{array}$$
and let $`𝒯^{}X(0)=T^{}X`$ and $`𝒯^{}X(\mathrm{})=X\times X\times \{\mathrm{}\}`$.
In order to avoid problems with the topology of $`𝒯^{}X`$ (which is in general not a manifold with corners) we let $`C_c^{\mathrm{}}(𝒯^{}X)=C_c^{\mathrm{}}(𝒯\stackrel{~}{X})|_{𝒯^{}X}`$.
### C\*-algebras Associated to the Semi-groupoids $`T^{}X`$ and $`𝒯^{}X`$
We start with $`T^{}X`$. Let $`C_c^{\mathrm{}}(T^{}X)`$ denote the smooth functions on $`T^{}X`$ which have compact support in $`T^{}X`$. In analogy with Definition 2.1 we introduce
$`\pi _0`$ $`:`$ $`C_c^{\mathrm{}}(T^{}X)(L^2(TX^{}))\text{and}`$
$`\pi _0^{}`$ $`:`$ $`C_c^{\mathrm{}}(T^{}X)(L^2(T^{}X|_X))`$
acting by
$`\pi _0(f)\xi (m,v)={\displaystyle _{T_mX}}f(m,vw)\xi (m,w)𝑑w,`$ (2.9)
$`\pi _0^{}(f)\xi (m,v)={\displaystyle _{T_m^+X}}f(m,vw)\xi (m,w)𝑑w.`$ (2.10)
Note that due to its compact support in $`T^{}X`$, the function $`f`$ naturally extends (by zero) to $`TX`$.
###### 2.13
Definition. We denote by $`C_r^{}(T^{}X)`$ the $`C^{}`$-algebra generated by $`\pi _0`$ and $`\pi _0^{}`$, i.e. by the map $`C_c^{\mathrm{}}(T^{}X)f(\pi _0(f),\pi _0^{}(f))(L^2(TX^{})L^2(T^{}X|_X))`$.
At first glance, this definition seems to overlook the operators of the form $`\pi _0^{}(K)`$ in 2.1 and operators of the form $`\pi _0(f)`$ and $`\pi _0^{}(f)`$, where $`fC_c^{\mathrm{}}(T\stackrel{~}{X})|_{TX}`$. In fact, this is not the case. The second type of operators belongs to $`C_r^{}(T^{}X)`$, because we take the closure under the adjoint operation and addition. The reason that the first type of operators is in $`C_r^{}(T^{}X)`$, is the well-known relation between operators of half-convolution and Toeplitz operators, which we recall, below. We denote by $`𝔗`$ the algebra of all Toeplitz operators on $`L^2(S^1)`$ and by $`𝔗_0`$ the ideal of all operators whose symbol vanishes in $`1`$.
###### 2.14
Lemma. Let $`fC_c^{\mathrm{}}().`$ Then the operator
$$L^2(_+)\xi \left(s_0^{\mathrm{}}f(sw)\xi (w)𝑑w\right)L^2(_+)$$
is unitarily equivalent to the Toeplitz operator $`T_\phi `$ with symbol $`\phi (z)=\widehat{f}(i(z1)/(z+1)).`$ Note that $`\phi (1)=\widehat{f}(\mathrm{})=0`$.
The $`C^{}`$-algebra generated by the operators in the image of $`C_c^{\mathrm{}}()`$ under this map is precisely the ideal $`𝔗_0`$, while the compact operators in $`𝔗`$ are generated by their commutators.
Proof. Plancherel’s theorem shows that the above operator of half convolution is the truncated pseudodifferential operator with symbol $`\widehat{f}`$, mapping $`\xi L^2(_+)`$ to $`\mathrm{op}(\widehat{f})_+\xi (s)=e^{ist}\widehat{f}(t)\widehat{(e^+\xi )}(t)dt`$, where $`e^+\xi `$ is the extension (by zero) of $`\xi `$ to $``$.
Now one observes that the unitary $`U:L^2(S^1)L^2(_+)`$ given by $`Ug(t)=\frac{\sqrt{2}}{1+it}g\left(\frac{1it}{1+it}\right)`$ maps the Hardy space $`H^2`$ to $`F(L^2(_+))`$ with the Fourier transform $`F`$, and that $`\mathrm{op}(\widehat{f})_+`$ is $`F^1UT_\phi U^1F`$. See \[18, Section 2\] for details.
For the second statement, one first notes that the $`C^{}`$-algebra generated by these operators is a subalgebra of $`𝔗_0`$. On the other hand, $`𝔗_0`$ consists of the operators of the form $`T_\phi +C`$, where $`\phi C(S^1)`$ vanishes in $`1`$, and $`C`$ is compact. According to \[8, Proposition 7.12\], the commutators of all $`T_\phi `$, $`\phi C(S^1)`$, generate the compacts, hence so do the commutators of those $`T_\phi `$, where $`\phi `$ vanishes in $`1`$. As these $`T_\phi `$ can be approximated by elements in the image of $`C_c^{\mathrm{}}()`$, the proof is complete. $``$
###### 2.15
Lemma. We have a representation $`\pi _0^{}`$ of $`C_c^{\mathrm{}}(TX\times _+\times _+)`$ on $`L^2(T^{}X|_X)`$ via
$`\pi _0^{}(K)\xi (m,v^{},v_n)={\displaystyle K(m,v^{}w^{},v_n,w_n)\xi (m,w^{},w_n)𝑑w^{}𝑑w_n}.`$ (2.11)
The closure of its range is isomorphic to
$$J=C_0(T^{}X)𝒦(L^2(_+)).$$
$`J`$ is an ideal in $`C_r^{}(T^{}X)`$ generated by commutators of elements of the form $`\pi _0^{}(f)`$.
Proof. The algebraic tensor product $`C_c^{\mathrm{}}(TX)C_c^{\mathrm{}}(_+\times _+)`$ is dense in $`C_c^{\mathrm{}}(TX\times _+\times _+)`$. Due to the continuity of
$$\pi _0^{}:C_c^{\mathrm{}}(TX\times _+\times _+)(L^2(T^{}X|_X)$$
it is sufficient to determine the closure of $`\pi _0^{}(C_c^{\mathrm{}}(TX)C_c^{\mathrm{}}(_+\times _+))`$.
It is clear that $`\pi _0^{}(C_c^{\mathrm{}}(TX)C_c^{\mathrm{}}(_+\times _+))J`$. In fact, we have equality, since the Fourier transform gives an isomorphism $`C_r(TX)C_0(T^{}X)`$ and since a compact operator on $`L^2(_+)`$ can be approximated by a Hilbert-Schmidt operator, thus by an integral operator with kernel in $`C_c^{\mathrm{}}(_+\times _+)`$.
In order to see that $`J`$ is contained in $`C_r^{}(T^{}X)`$, it is sufficient to approximate both factors of a pure tensor $`hc`$, where $`hC_0(T^{}X)`$ and $`c𝒦(L^2(_+))`$. For the first task we choose a function in $`C_c^{\mathrm{}}(TX)`$ whose fiberwise Fourier transform is close to $`h`$ in sup-norm. For the second, we refer to Lemma 2.14. In particular, we see that $`J`$ also is generated by commutators.
A direct computation shows that $`J`$ is an ideal in $`C_r^{}(T^{}X)`$. $``$
###### 2.16
Definition. We let
$$C_{tc}^{\mathrm{}}(T^{}X)=C_c^{\mathrm{}}(TX)C_c^{\mathrm{}}(TX\times _+\times _+).$$
This is a dense $``$-subalgebra of $`C_r^{}(T^{}X)`$. We will denote the product in this subalgebra by $`^{}`$.
For $`\mathrm{}0`$ we obtain representations of $`C_c^{\mathrm{}}(𝒯^{}X)=C_c^{\mathrm{}}(𝒯\stackrel{~}{X})|_{𝒯^{}X}`$ in $`(L^2(X))`$ by:
$$\pi _{\mathrm{}}(f)\xi (m)=\frac{1}{\mathrm{}^n}f(m,\stackrel{~}{m},\mathrm{})\xi (\stackrel{~}{m})𝑑\stackrel{~}{m}.$$
(2.12)
Note that these are the natural groupoid representations for $`X\times X\times ]0,1]`$.
We denote by $`C_r^{}(𝒯^{}X)`$ the reduced $`C^{}`$-algebra generated by $`\pi _{\mathrm{}}`$, $`0\mathrm{}1`$, and $`\pi _0^{}`$.
For $`X=_+^n`$ we have $`C_{tc}^{\mathrm{}}(T^{}X)=A(0)^{\mathrm{}}`$, $`C_r^{}(T^{}X)=A(0)`$ and $`C_r^{}(𝒯^{}X)=A`$. Also there are evaluation maps
$$\phi _{\mathrm{}}:C_r^{}(𝒯^{}X)C_r^{}(𝒯^{}X)(\mathrm{}).$$
###### 2.17
Theorem. We have
$`C_r^{}(𝒯^{}X(\mathrm{}))`$ $`=`$ $`𝒦(L^2(X)),\mathrm{}0;`$
$`C_r^{}(𝒯^{}X(0))`$ $`=`$ $`C_r^{}(T^{}X).`$
Moreover: $`(C_r^{}(𝒯^{}X),\{C_r^{}(𝒯^{}X(\mathrm{})),\phi _{\mathrm{}}\}_{\mathrm{}[0,1]})`$ is a continuous field of $`C^{}`$-algebras.
The first two statements are obvious. For the proof of upper semi-continuity, we will essentially follow the ideas for the half-space case. Our first task is the construction of a representation of $`C_{tc}^{\mathrm{}}(T^{}X)`$. To this end, we will simply extend $`fC_c^{\mathrm{}}(TX)`$ and $`KC_c^{\mathrm{}}(TX\times _+\times _+)`$ to functions $`\stackrel{~}{f}`$ and $`\stackrel{~}{K}`$ on $`𝒯^{}X`$ as described below, then apply (2.12).
Choose a function $`\psi C^{\mathrm{}}(X\times X)`$ which is one on a neighborhood of the diagonal, $`0\psi 1`$, such that
$$\mathrm{exp}:T^{}XX\times X$$
maps a neighborhood of the zero section diffeomorphically to the support of $`\psi `$.
For $`fC_c^{\mathrm{}}(TX)`$ we define $`\stackrel{~}{f}C_c^{\mathrm{}}(𝒯^{}X)`$ by
$$\stackrel{~}{f}(m,\stackrel{~}{m},\mathrm{})=\psi (m,\stackrel{~}{m})f(m,\frac{\mathrm{exp}^1(m,\stackrel{~}{m})}{\mathrm{}}).$$
(2.13)
We next identify a neighborhood $`U`$ of $`X`$ in $`X`$ with $`X\times [0,1[`$ and write $`Um=(m^{},m_n)`$ with $`m^{}X`$ and $`m_n0`$. We also choose a function $`\chi C_c^{\mathrm{}}(X)`$ supported in $`U`$ with $`0\chi 1`$ and $`\chi 1`$ near $`X`$. For $`KC_c^{\mathrm{}}(TX\times _+\times _+)`$ we then define $`\stackrel{~}{K}C_c^{\mathrm{}}(X\times X\times ]0,1])`$ by
$`\stackrel{~}{K}(m,\stackrel{~}{m},\mathrm{})=\chi (m)\chi (\stackrel{~}{m})\psi (m,\stackrel{~}{m})K(m^{},{\displaystyle \frac{\mathrm{exp}^1(m^{},\stackrel{~}{m}^{})}{\mathrm{}}},{\displaystyle \frac{m_n}{\mathrm{}}},{\displaystyle \frac{\stackrel{~}{m}_n}{\mathrm{}}}).`$ (2.14)
###### 2.18
Remark. In the half-space case with the flat metric we have, for fixed $`f`$ and $`K`$,
$$\pi _{\mathrm{}}(\stackrel{~}{f})=\rho _{\mathrm{}}(f)\text{and}\pi _{\mathrm{}}(\stackrel{~}{K})=\kappa _{\mathrm{}}(K)$$
provided $`\mathrm{}`$ is sufficiently small.
###### 2.19
Corollary. We then obtain the analog of Property (0.1):
$`\underset{\mathrm{}0}{lim}\pi _{\mathrm{}}(\stackrel{~}{f})+\pi _{\mathrm{}}(\stackrel{~}{K})=\mathrm{max}\{\pi _0(f),\pi _0^{}(fK)\}.`$ (2.15)
###### 2.20
Metrics. The construction of $`C_r^{}(𝒯^{}X)`$ and the extensions (2.13), (2.14) used a metric, but $`C_r^{}(𝒯^{}X)`$ is independent of the choice: Let $`\nu _1,\nu _2`$ be two different metrics on $`X`$, and denote by $`\mu _1,\mu _2`$ the associated measures on $`X`$ as well as the fiberwise measures in $`TX`$. Let $`kC^{\mathrm{}}(X)`$ be given by
$$\mu _1=k\mu _2.$$
Multiplication by $`\sqrt{k}`$ yields a unitary
$$U:(L^2(X),\mu _1)(L^2(X),\mu _2),$$
and multiplication by $`\sqrt{k(m)}`$ a family of unitaries
$$U_m:(L^2(T_m^{}X),\mu _1)(L^2(T_m^{}X),\mu _2).$$
We define
$$\varphi :C_c^{\mathrm{}}(𝒯^{}X)C_c^{\mathrm{}}(𝒯^{}X)$$
taking $`f(m,v,0)`$ to $`f(m,v,0)k(m)`$ for $`\mathrm{}=0`$ and $`f(m,\stackrel{~}{m},\mathrm{})`$ to $`f(m,\stackrel{~}{m},\mathrm{})\sqrt{k(m)k(\stackrel{~}{m})}`$, $`\mathrm{}0`$. Then $`\pi _{\mathrm{}}^1(f)=U^1\pi _{\mathrm{}}^2(\varphi (f))U`$, where $`\pi _{\mathrm{}}^1`$ and $`\pi _{\mathrm{}}^2`$ are the representations induced by $`\mu _1`$ and $`\mu _2`$. A corresponding relation holds for $`\pi _0^{}`$. Hence $`C_r^{}(𝒯^{}X)`$ is independent of the metric.
The following lemma clarifies the influence of the extension by different metrics.
###### 2.21
Lemma. Let $`fC_c^{\mathrm{}}(TX)`$. Denote by $`\stackrel{~}{f}^i`$ the extension of $`f`$ with respect to the metric $`\nu _i`$, $`i=1,2`$. Then
$$\pi _{\mathrm{}}(\varphi (\stackrel{~}{f}^1))\pi _{\mathrm{}}(\stackrel{~}{\varphi (f)}^2)0\text{ for }\mathrm{}0.$$
Here $`\pi _{\mathrm{}}`$ is understood with respect to $`\mu _2`$.
Proof. This follows from Lemma 2.10, since $`\varphi (\stackrel{~}{f}^1)\stackrel{~}{\varphi (f)}^2`$ is a function in $`C_c^{\mathrm{}}(𝒯^{}X)`$ which is zero at $`\mathrm{}=0`$. $``$
A similar statement holds if we start with $`KC_c^{\mathrm{}}(TX\times _+\times _+)`$.
### Upper Semi-continuity
We again use the seminorm
$$a_{as}=\mathrm{max}\{\phi _0(a),\underset{\mathrm{}0}{lim\; sup}\phi _{\mathrm{}}(a)\}$$
for elements in $`C_r^{}(𝒯^{}X)`$ and introduce the analog of $`A[0]`$:
$$C_\mathrm{?}^{}(T^{}X)=C_r^{}(𝒯^{}X)/I,$$
where
$$I=\{aC_r^{}(𝒯^{}X)|a_{as}=0\}.$$
The notation $`C_\mathrm{?}^{}(TX^{})`$ is justified by the following:
###### 2.22
Proposition. The mappings $`f\stackrel{~}{f}`$ and $`K\stackrel{~}{K}`$ induce a $``$-homomorphism $`\mathrm{\Psi }`$ from $`(C_c^{\mathrm{}}(T^{}X),^{})`$ to $`C_\mathrm{?}^{}(T^{}X)`$ with dense range, and we have
$`\underset{\mathrm{}0}{lim}\pi _{\mathrm{}}(\stackrel{~}{f})\pi _{\mathrm{}}(\stackrel{~}{g})\pi _{\mathrm{}}(\stackrel{~}{f^{}g)}=0,f,gC_c^{\mathrm{}}(TX).`$ (2.16)
Proof. Choose an open covering $`\{U_i\}`$ of $`X`$, where each $`U_i`$ can be identified with an open subset of $`^n`$ or $`_+^n`$. By possibly shrinking the $`U_i`$, we may assume that the function $`\psi `$ used in (2.13) and (2.14) equals $`1`$ on $`U_i\times U_i`$ and that the function $`\chi `$ is $`1`$ on $`U_i`$ whenever $`U_i`$ intersects the boundary. We also fix a subordinate partition of unity $`\{\psi _i\}C_c^{\mathrm{}}(U_i)`$.
For $`f,gC_c^{\mathrm{}}(TX)`$ we have $`(\psi _if)^{}g=(\psi _if)^{}(\eta _ig)`$ for each $`\eta _iC_c^{\mathrm{}}(U_i)`$ with $`\psi _i\eta _i=\psi _i`$. Moreover, $`\pi _{\mathrm{}}(\stackrel{~}{\psi _if})\pi _{\mathrm{}}(\stackrel{~}{g})=\pi _{\mathrm{}}(\stackrel{~}{\psi _if})\pi _{\mathrm{}}(\stackrel{~}{\theta _ig})`$ for suitable $`\theta _iC_c^{\mathrm{}}(U_i)`$, provided $`\mathrm{}`$ is small. Hence
$`\pi _{\mathrm{}}(\stackrel{~}{f^{}g})\pi _{\mathrm{}}(\stackrel{~}{f})\pi _{\mathrm{}}(\stackrel{~}{g}){\displaystyle \left(\pi _{\mathrm{}}(\stackrel{~}{(\psi _if)^{}g})\pi _{\mathrm{}}(\stackrel{~}{\psi _if})\pi _{\mathrm{}}(\stackrel{~}{g})\right)}`$ (2.17)
$`=`$ $`{\displaystyle }\pi _{\mathrm{}}(\stackrel{~}{\psi _if^{}\eta _ig}))\pi _{\mathrm{}}(\stackrel{~}{\psi _if})\pi _{\mathrm{}}(\stackrel{~}{\theta _ig}).`$
For sufficiently small $`\mathrm{}`$, all operators will have support in $`U_i\times U_i\times [0,1]`$ so that we are working on Euclidean space. According to Lemma 2.21 we can also, modulo terms converging to zero as $`\mathrm{}0`$, use the Euclidean metric. So we are precisely in the situation considered at the beginning of the section. The explicit computation shows that
$`\pi _{\mathrm{}}(\stackrel{~}{f^{}g}))\pi _{\mathrm{}}(\stackrel{~}{f})\pi _{\mathrm{}}(\stackrel{~}{g})=\rho _{\mathrm{}}(fgf_{\mathrm{}}g)+\kappa _{\mathrm{}}(l(f,g)l_{\mathrm{}}(f,g)).`$ (2.18)
As $`fgf_{\mathrm{}}gC_c^{\mathrm{}}(T_+^n\times [0,1])`$ and $`l(f,g)l_{\mathrm{}}(f,g)C^{\mathrm{}}(T^{n1}\times _+\times _+\times [0,1])`$ vanish for $`\mathrm{}=0`$, the difference (2.18) is in $`I`$ by Lemma 2.10. Hence (2.17) tends to zero, and $`\mathrm{\Psi }(f^{}g)=\mathrm{\Psi }(f)\mathrm{\Psi }(g)`$. The remaining $``$-algebra properties are checked similarly.
In order to see that the image of $`\mathrm{\Psi }`$ is dense in $`C_\mathrm{?}^{}(T^{}X)`$, we simply note that the evaluation at $`\mathrm{}=0`$ associates to an element $`F`$ in $`C_c^{\mathrm{}}(𝒯^{}X)`$ an element in $`C_{tc}^{\mathrm{}}(T^{}X)`$ whose extension via (2.13), (2.14) induces the same element in $`C_\mathrm{?}^{}(𝒯^{}X)`$ by Lemma 2.10$``$
###### 2.23
Remark. Property (2.16) is the analog of the asymptotic multiplicativity (0.2) in the case of manifolds with boundary. In particular, we have established Theorem 2.4.
With Proposition 2.22, the proof of the following theorem is analogous to that of Theorem 2.11.
###### 2.24
Theorem. $`(C_r^{}(𝒯^{}X),\{C_r^{}(𝒯^{}X)(\mathrm{}),\phi _{\mathrm{}}\}_{\mathrm{}[0,1]})`$ is upper semi-continuous in $`0`$.
### Lower Semi-continuity
As in the classical case lower semi-continuity is proven by introducing strongly continuous representations using the groupoid structure. We split the representations into two: One taking care of the contribution from the interior of the manifold, i.e. the convolution part, and one taking care of the boundary part, i.e. half convolution and kernels on the boundary.
For the lemmata, below, we note that – by construction – $`\pi _0`$ and $`\pi _0^{}`$ extend to $`C_r^{}(𝒯^{}X)`$.
###### 2.25
Lemma. $`lim\; inf_\mathrm{}0\phi _{\mathrm{}}(a)\pi _0(a)`$ for all $`aC_r^{}(𝒯^{}X)`$.
Proof. According to Proposition 2.22 it is sufficient to show that
$$\underset{\mathrm{}0}{lim}\rho _{\mathrm{}}(\stackrel{~}{f}+\stackrel{~}{K})\pi _0(f)\text{for}fKC_{tc}^{\mathrm{}}(T^{}X).$$
(2.19)
For $`gC_c^{\mathrm{}}(𝒯^{}X)`$ define
$$g_\mathrm{},\mathrm{}^2=\underset{mX}{sup}\left\{\frac{1}{\mathrm{}^n}_X|g(x,m,\mathrm{})|^2𝑑x\right\}\text{ for }\mathrm{}0,$$
and
$$g_{\mathrm{},0}^2=\underset{mX}{sup}\left\{_{T_mX}|g(m,v,0)|^2𝑑v\right\},\text{ for }\mathrm{}=0.$$
Set
$$g_{\mathrm{}}=\underset{\mathrm{}[0,1]}{sup}g_\mathrm{},\mathrm{}.$$
It is easily checked that
$`\pi _{\mathrm{}}(\stackrel{~}{f}+\stackrel{~}{K})=sup\{{\displaystyle \frac{1}{\mathrm{}^n}}{\displaystyle }(\stackrel{~}{f}(,m,\mathrm{})+\stackrel{~}{K}(,m,\mathrm{}))g(m,,\mathrm{})dm_\mathrm{},\mathrm{}|g_{\mathrm{}}1\}`$ (2.20)
for $`\mathrm{}0`$, and
$`\pi _0(f)=sup\{{\displaystyle }f(,v,0)g(,v,0)dv_{\mathrm{},0}|g_{\mathrm{}}1\}:`$ (2.21)
In fact, for (2.20) we note that “$``$” follows from the estimate
$`{\displaystyle \frac{1}{\mathrm{}^n}}{\displaystyle \stackrel{~}{f}(m_1,m,\mathrm{})g(m,m_2,\mathrm{})𝑑m}_\mathrm{},\mathrm{}^2={\displaystyle \frac{1}{\mathrm{}^n}}\pi _{\mathrm{}}(\stackrel{~}{f})g(,m_2,\mathrm{})_\mathrm{},\mathrm{}^2`$
$`=`$ $`\underset{m_2X}{sup}{\displaystyle \frac{1}{\mathrm{}^n}}\pi _{\mathrm{}}(\stackrel{~}{f})g(,m_2,\mathrm{})_{L^2(X)}^2\pi _{\mathrm{}}(\stackrel{~}{f})^2\underset{m_2X}{sup}{\displaystyle \frac{1}{\mathrm{}^n}}g(,m_2,\mathrm{})_{L^2(X)}^2`$
$`=`$ $`\pi _{\mathrm{}}(\stackrel{~}{f})^2g_\mathrm{},\mathrm{}^2\pi _{\mathrm{}}(\stackrel{~}{f})^2g_{\mathrm{}}^2.`$
For the reverse inequality we choose $`g(x,m,\mathrm{})=s(m)\xi (x)\mathrm{}^n\phi (\mathrm{})`$, where $`sC_c^{\mathrm{}}(X)`$, $`s1`$, $`\xi _{L^2(X)}=1`$ with $`\pi _{\mathrm{}}(\stackrel{~}{f})\xi \pi _{\mathrm{}}(\stackrel{~}{f})\epsilon `$, and $`\phi C_c^{\mathrm{}}(]0,1])`$ is equal to one outside a neighborhood of zero. Equation (2.21) follows by a similar argument.
Now suppose that $`gC_c^{\mathrm{}}(𝒯^{}X)`$ and $`g(x,m,h)=0`$ for $`xX`$. Then the weak convergence of $`\stackrel{~}{K}`$ towards zero implies that
$`\underset{\mathrm{}0}{lim}{\displaystyle \frac{1}{\mathrm{}^n}}{\displaystyle (\stackrel{~}{f}(,m,\mathrm{})+\stackrel{~}{K}(,m,\mathrm{}))g(m,,\mathrm{})𝑑m}_\mathrm{},\mathrm{}`$ $`=`$ $`{\displaystyle }f(,v,0)g(,v,0)dv_{\mathrm{},0}.`$
As the set of these $`g`$ is dense in $`\{gC_c^{\mathrm{}}(𝒯^{}X)|g_{\mathrm{}}1\}`$, (2.19) follows. $``$
###### 2.26
Lemma. $`lim\; inf_\mathrm{}0\phi _{\mathrm{}}(a)\pi _0^{}(a)`$ for all $`aC_r^{}(𝒯^{}X)`$.
Proof. As in the proof of Lemma 2.25 we only have to show that
$$\underset{\mathrm{}0}{lim\; inf}\rho _{\mathrm{}}(\stackrel{~}{f}+\stackrel{~}{K})\pi _0^{}(fK),$$
(2.22)
for $`fC_c^{\mathrm{}}(TX)`$ and $`KC_c^{\mathrm{}}(TX\times _+\times _+)`$.
We let $`P_{\mathrm{}}`$ be the projection in $`L^2(X)`$ given by multiplication by the characteristic function of $`X\times [0,a_{\mathrm{}}[`$, where
$$a_{\mathrm{}}0\text{ for }\mathrm{}0\text{ and }\frac{a_{\mathrm{}}}{\mathrm{}}\mathrm{}\text{ for }\mathrm{}0.$$
As $`P_{\mathrm{}}\pi _{\mathrm{}}(\stackrel{~}{f}+\stackrel{~}{K})P_{\mathrm{}}\pi _{\mathrm{}}(\stackrel{~}{f}+\stackrel{~}{K})`$, it is enough to show that
$$\underset{\mathrm{}0}{lim\; inf}P_{\mathrm{}}\pi _{\mathrm{}}(\stackrel{~}{f}+\stackrel{~}{K})P_{\mathrm{}}\pi _0^{}(fK).$$
Since we are free too choose a metric, we fix a metric on $`X`$ and the standard metric on $`[0,a_{\mathrm{}}[`$.
As in the proof of Lemma 2.25, we equip the space $`C_c^{\mathrm{}}(𝒯X\times [0,\mathrm{}[)`$ with norms $`_\mathrm{},\mathrm{}`$, $`_{\mathrm{}}`$, which are just like the norms before, on $`𝒯X`$ instead of $`𝒯^{}X`$, combined with the $`L^2`$-norm on $`[0,\mathrm{}[`$. For $`fC_c^{\mathrm{}}(𝒯^{}X)`$ and $`KC_c^{\mathrm{}}(TX\times _+\times _+)`$ we define representations on $`C_c^{\mathrm{}}(𝒯X\times [0,\mathrm{}[)`$ by
$`\eta _{\mathrm{}}(f)g(m_1,m_2,\mathrm{},b)`$ $`=`$ $`{\displaystyle \frac{1}{\mathrm{}^{n1}}}{\displaystyle _{a[0,\frac{a_{\mathrm{}}}{\mathrm{}}]}}f(m_1,\mathrm{}b,m,\mathrm{}a,\mathrm{})g(m,m_2,\mathrm{},a)dmda,b[0,{\displaystyle \frac{a_{\mathrm{}}}{\mathrm{}}}[,\mathrm{}0;`$
$`\eta _0(f)g(m_1,v,0,b)`$ $`=`$ $`{\displaystyle _{T_{m_1}X\times _+}}f(m_1,0,vw,ba,0)g(m_1,w,0,a)𝑑w𝑑a;`$
$`\eta _0(K)(m_1,v,0,b)`$ $`=`$ $`{\displaystyle _{T_{m_1}X\times _+}}K(m_1,vw,b,a)g(m_1,w,0,a)𝑑w𝑑a.`$
Note that $`P_{\mathrm{}}\pi _{\mathrm{}}(f)P_{\mathrm{}}=D_{\mathrm{}}P_{\mathrm{}}\pi _{\mathrm{}}(f)P_{\mathrm{}}D_\mathrm{}^1=sup\{\eta _{\mathrm{}}(f)g_\mathrm{},\mathrm{}|g_{\mathrm{}}1\}`$, where $`D_{\mathrm{}}`$ is the dilation operator in the normal direction, given by $`D_{\mathrm{}}f(x^{},x_n)=f(x^{},\mathrm{}x_n)`$.
As before
$$\pi _0^{}(fK)=sup\{(\eta _0(fK))g_{\mathrm{},0}|g_{\mathrm{}}1\}.$$
Plugging in the definitions of $`\stackrel{~}{f}`$ and $`\stackrel{~}{K}`$ (omitting the cut off functions) we get
$$\eta _{\mathrm{}}(\stackrel{~}{f})g(m_1,m_2,\mathrm{},b)=\frac{1}{\mathrm{}^{n1}}_{[0,\frac{a_{\mathrm{}}}{\mathrm{}}]}f(m_1,\mathrm{}b,\frac{\mathrm{exp}^1(m_1,m)}{\mathrm{}},ba)g(m,m_2,\mathrm{},a)𝑑m𝑑a$$
$$\eta _{\mathrm{}}(\stackrel{~}{K})g(m_1,m_2,\mathrm{},b)=\frac{1}{\mathrm{}^{n1}}_{[0,\frac{a_{\mathrm{}}}{\mathrm{}}]}K(m_1,b,\frac{\mathrm{exp}^1(m_1,m)}{\mathrm{}},a)g(m,m_2,\mathrm{},a)𝑑m𝑑a$$
Using dominated convergence and the fact that $`g`$ for small $`\mathrm{}`$ looks like $`g_0(m,\frac{\mathrm{exp}^1(m,m_2)}{\mathrm{}},\mathrm{},a)`$, $`g_0C_c^{\mathrm{}}(TX\times [0,1]\times [0,\mathrm{}[)`$, we get
$$\underset{\mathrm{}0}{lim}(\eta _{\mathrm{}}(\stackrel{~}{f}+\stackrel{~}{K})g_\mathrm{},\mathrm{}=(\eta _0(fK)g_{\mathrm{},0},$$
and (2.22) follows. $``$
Lemma 2.25 and 2.26 imply that $`lim\; inf_\mathrm{}0\phi _{\mathrm{}}(a)\phi _0(a),`$ i.e.
###### 2.27
Theorem. $`(C_r^{}(𝒯X^{}),\{C_r^{}(𝒯X^{})(\mathrm{}),\phi _{\mathrm{}}\}_{\mathrm{}[0,1]})`$ is lower semi-continuous in $`0`$.
This finishes the proof of Theorem 2.17.
## 3 $`K`$-theory of the Symbol Algebra $`C_r^{}(T^{}X)`$
$`C_c^{\mathrm{}}(TX^{})`$ with the fiberwise convolution product is a $``$-ideal of $`C_{tc}^{\mathrm{}}(T^{}X)`$. After completion, $`C_r^{}(TX^{})`$ becomes a $`C^{}`$-ideal of $`C_r^{}(T^{}X)`$, and we have a short exact sequence
$$0C_r^{}(TX^{})C_r^{}(T^{}X)C_r^{}(T^{}X)/C_r^{}(TX^{})0.$$
(3.1)
###### 3.1
Proposition. The quotient $`Q=C_r^{}(T^{}X)/C_r^{}(TX^{})`$ is naturally isomorphic to $`C_0(T^{}X)𝔗_0`$ for the ideal $`𝔗_0`$ of the Toeplitz algebra introduced before Lemma 2.14.
Proof. Define
$$\mathrm{\Psi }:C_c^{\mathrm{}}(T^{}X)(L^2(T^{}X|_X))\text{ by }\mathrm{\Psi }(fK)=\pi _0^{}(f)+\pi _0^{}(K)$$
with the maps in (2.10) and (2.11). This is a $``$-homomorphism with respect to $`^{}`$, and $`C_c^{\mathrm{}}(TX^{})`$ is in its kernel. We first show that $`\mathrm{ker}\mathrm{\Psi }=C_r^{}(TX^{})`$: Since $`C_r^{}(T^{}X)`$ is the closure of $`C_c^{\mathrm{}}(T^{}X)`$ with respect to the norm
$$fK=\mathrm{max}\{|\pi _0(f),\pi _0^{}(f)+\pi _0^{}(K)\},$$
and $`C_r^{}(TX^{})`$ is the closure of $`C_c^{\mathrm{}}(TX^{})`$ with respect to $`\pi _0(f)`$, we have $`C_r^{}(TX^{})\mathrm{ker}\mathrm{\Psi }`$.
On the other hand, suppose that $`a\mathrm{ker}\mathrm{\Psi }`$; i.e., $`a`$ is the equivalence class of a Cauchy sequence $`(f_kK_k)C_c^{\mathrm{}}(T^{}X)`$ with $`\pi _0^{}(f_k)+\pi _0^{}(K_k)0`$. We next note that
$`\pi _0(f_k)`$ $`=`$ $`sup\{|\widehat{f}_k(m,\sigma )||(m,\sigma )T^{}X\}:\text{ and}`$
$`\pi _0^{}(f_k)+\pi _0^{}(K_k)`$ $``$ $`sup\{|\widehat{f}_k(m,\sigma )||(m,\sigma )T^{}X|_X\}`$
Indeed the first inequality follows from the fact that, via fiberwise Fourier transform, $`\pi _0(f_k)`$ is equivalent to multiplication by $`\widehat{f}_k(m,\sigma )`$. For the second, we observe first that $`\pi _0^{}(f_k)=sup\{|\widehat{f}_k|\}`$ as a consequence of the fact that translation of $`\xi =\xi (m,w)`$ in the direction of $`w_n`$ preserves $`\pi _0^{}(f_k)\xi `$ in $`L^2(T^{}X|_X)`$. On the other hand, $`\pi _0^{}(K_k)\xi =0`$ provided we translate sufficiently far. Hence $`\pi _0^{}(f_k)+\pi _0^{}(K_k)\pi _0^{}(f_k)`$.
We conclude that the fiberwise Fourier transforms $`\widehat{f}_k`$ tend to zero uniformly on $`T^{}X|_X`$. Hence the Cauchy sequence $`(f_k)`$ may be replaced by an equivalent Cauchy sequence $`(g_k)`$ with $`g_kC_c^{\mathrm{}}(TX^{})`$. We conclude that $`\pi _0^{}(K_k)0`$ so that $`(K_k)0`$, and therefore $`\mathrm{ker}\mathrm{\Psi }C_r^{}(TX^{})`$.
Hence $`\mathrm{\Psi }`$ descends to an injective $`C^{}`$-morphism on $`Q`$; in particular, it has closed range.
Now we observe that we have a natural identification of $`TX|_X`$ with $`TX\times `$ and consequently of $`T^{}X|_X`$ with $`TX\times _{}`$. Hence $`(L^2(T^{}X|_X))(L^2(TX)L^2(_{})).`$
Suppose that, at the boundary, $`fC_c^{\mathrm{}}(TX)`$ is of the form $`f(x^{},0,v^{},v_n)=g(x^{},v^{})h(v_n)`$ with $`gC_c^{\mathrm{}}(TX)`$ and $`hC_c^{\mathrm{}}()`$. Then $`\pi _0^{}(f)=\pi _0^{,0}(g)\pi _0^{,n}(h)`$, where $`\pi _0^{,0}`$ is the convolution operator by $`g`$, acting on $`L^2(TX)`$, while $`\pi _0^{,n}(h)`$ is the operator of half convolution acting on $`L^2(_{})`$ (note that $`_{}T^{}_+|_{\{0\}}`$). Via Fourier transform, the operator $`\pi _0^{,0}(g)`$ is unitarily equivalent to multiplication by $`\widehat{g}C_0(T^{}X)`$, while, according to Lemma 2.14, $`\pi _0^{,n}(h)`$ is unitarily equivalent to a Toeplitz operator in $`𝔗_0`$. The closure of the image of the span of the pure tensors thus gives us $`C_0(T^{}X)𝔗_0`$.
We know already from Lemma 2.15 that – via the Fourier transform – the image of $`C_c^{\mathrm{}}(TX\times _+\times _+)`$ can also be identified with a subset of $`C_0(T^{}X)𝒦C_0(T^{}X)𝔗_0`$. This completes the argument. $``$
###### 3.2
Theorem. Via fiberwise Fourier transform $`C_r^{}(TX^{})`$ can be identified with $`C_0(T^{}X^{})`$ and the inclusion $`C_0(T^{}X^{})C_r^{}(TX^{})C_r^{}(T^{}X)`$ induces an isomorphism of K-groups
$$K_i(C_r^{}(T^{}X))K_i(C_0(T^{}X^{})),i=0,1.$$
Proof. It is well-known (or easily checked) that $`K_i(𝔗_0)=0`$, $`i=0,1`$. Thus it follows from the Künneth formula that $`K_i(C_0(T^{}X)𝔗_0)=0`$, $`i=0,1`$. The result now is a consequence of (3.1) and the associated six term exact sequence. $``$
Johannes Aastrup, Institut für Mathematik, Universität Hannover, Welfengarten 1, 30167 Hannover, Germany, email: aastrup@math.uni-hannover.de
Ryszard Nest, Department of Mathematics, Copenhagen University, Universitetsparken 5, 2100 Copenhagen, Denmark, email: rnest@math.ku.dk
Elmar Schrohe, Institut für Mathematik, Universität Hannover, Welfengarten 1, 30167 Hannover, Germany, email: schrohe@math.uni-hannover.de |
warning/0507/nucl-ex0507005.html | ar5iv | text | # ICANS-XVII 17th Meeting of the International Collaboration on Advanced Neutron Sources April 25-29, 2005 Santa Fe, New Mexico The NPDGamma Experiment at LANSCE
## 1 Introduction
Since 1980, the weak parity-violating nucleon-nucleon interaction has typically been described by a meson-exchange potential involving seven weak meson-nucleon coupling constants . The weak interaction changes the parity and isospin ($`\mathrm{\Delta }I=0,1,2`$) of the nucleon-nucleon pair and perturbatively introduces parity violating admixtures in nuclear wave functions. The study of the hadronic weak interaction is of great relevance for low energy, non-perturbative QCD. The hadronic weak couplings probe short range correlations between quarks because the quark-quark weak interaction occurs when the distance between quarks is $`2\times 10^3\mathrm{fm}`$. The electro-weak Standard Model predicts that charged current contributions to the weak N-N interaction are suppressed and that, therefore, the measurement of a zero asymmetry in the $`\stackrel{}{n}+pd+\gamma `$ reaction would suggest that neither neutral currents nor the strange quark pair sea contribute significantly to the hadronic weak interaction. A non-zero $`\stackrel{}{n}+pd+\gamma `$ asymmetry, on the other hand, would then establish weak neutral currents as the dominant factor with possibly significant contributions from strange quarks.
The NPDGamma experiment is under commissioning at the Los Alamos Neutron Scattering Center (LANSCE). It is the first experiment designed for the new pulsed high flux cold beam line, flight path 12, at LANSCE. NPDGamma will determine the very small weak pion-nucleon coupling constant $`f_\pi `$ in the nucleon-nucleon interaction . This coupling constant is directly proportional to the parity-violating up-down asymmetry $`A_\gamma `$ in the angular distribution of 2.2-MeV $`\gamma `$-rays with respect to the neutron spin direction (eqn. 1) in the reaction $`\stackrel{}{n}+pd+\gamma `$.
$$\frac{d\sigma }{d\mathrm{\Omega }}\frac{1}{4\pi }\left(1+A_\gamma \mathrm{cos}\theta \right)$$
(1)
The asymmetry has a predicted size of $`5\times 10^8`$ and our goal is to measure it to 10% accuracy. The small size of the asymmetry and the high proposed measurement precision impose heavy requirements on the performance of the beam line and apparatus. It is necessary to achieve high counting statistics while at the same time suppressing any systematic errors below the statistical limit. The experiment was designed to satisfy these requirements .
During commissioning the radiative neutron capture on various target materials was investigated to look for any asymmetry which may enter as a systematic effect while taking data with the hydrogen target. The measurements concentrated on materials that can be found within the experimental apparatus and which are interacting with the neutrons. The targets included Al, Cu, In, and $`\mathrm{B}_4\mathrm{C}`$. In addition, the known asymmetry in $`\mathrm{CCl}_4`$ was used to demonstrate that the array functions as designed and is capable of measuring non-zero asymmetries. Boron is used throughout the experiment, for neutron shielding and to collimate the beam. Aluminum is used in most of the equipment and the beam encounters several millimeter of it, primarily in the windows of the hydrogen target. Cu and In are also used in the target. It is therefore necessary to establish the size of the $`\gamma `$-asymmetry due to neutron capture on each of these elements.
## 2 The Experiment
The NPDGamma experiment is located on flight path 12 at the Manuel Lujan Jr. Neutron Scattering Center at LANSCE. The statistical accuracy that can be reached in the NPDGamma experiment at LANSCE is limited by the available cold neutron flux. In a spallation neutron source, the neutron flux depends on the proton current, the energy incident on the spallation target, the moderator performance (brightness) and the neutron guide performance . The LANSCE linear accelerator delivers $`800`$ MeV protons to a proton storage ring, which compresses the beam to $`250`$ ns wide pulses at the base, at a rate of $`20`$ Hz. The protons from the storage ring are incident on a split Tungsten target and the resulting spallation neutrons are cooled by and backscattered from a cold $`\mathrm{H}_2`$ moderator with a surface area of $`12\times 12`$ $`\mathrm{cm}^2`$ .
### 2.1 Beam Line
Figure (1) illustrates the flight path and experiment cave. The distance between the moderator and target is about $`22`$ meters. The flight path 12 beam line is $`19.5`$ m long and consists of $`4`$ m of in pile guide, a $`2`$ m long shutter, a frame overlap beam chopper and $`13`$ meters of neutron guide. The pulsed neutron source allows us to know the neutron time of flight or energy and the installed beam chopper allows us to select a range of neutron energies. In the experiment cave, the beam is transversely polarized by transmission through a polarized <sup>3</sup>He cell. Three <sup>3</sup>He ion chambers are used to monitor beam intensity and measure beam
polarization through transmission ratios. A radio frequency spin flipper is used to reverse the neutron spin direction on a pulse by pulse basis. For the production experiment the neutrons will capture in a 21 liter liquid para-hydrogen target. The $`\gamma `$-rays from the capture are detected by an array of 48 CsI(Tl) detectors operated in current mode . The entire apparatus is located in a homogeneous 10 Gauss with a field gradient of less than $`1`$ mG/cm which is required to maintain the neutron spin downstream of the polarizer.
For the ideal experiment, with 100% beam polarization, 100% spin flip efficiency, point sources and detectors, and no systematic effects resulting in background asymmetries or beam depolarization, the yield from a single detector is given by Eqn. 1, with the intensity essentially determined by the neutron flux. However, the actual signal in the detectors is more closely related to the number of neutrons that capture on the target, the target size, the neutron capture location in the target along the beam direction and the average detector solid angle. All components between the guide exit and the target have been optimized to ensure that they can properly function while attenuating the beam as little as possible. This includes optimization of the monitor and spin filter thicknesses, reduction of the aluminum windows on the monitors and the spin flipper and the overall reduction of the experiment length to reduce beam divergence. In addition to maximizing the number of captured neutrons, a successful asymmetry measurement requires a stable, polarized beam and the ability to reverse the beam polarization without significant losses. The emitted $`\gamma `$-rays have to be detected with high efficiency and with reasonably good angular resolution, which is limited by the finite size of the detectors and targets .
The FP12 beam guide was installed to deliver the maximum possible number of low energy neutrons to the experimental apparatus. FP12 uses an m=3 guide with a $`9.5\times 9.5`$ $`\mathrm{cm}^2`$ cross-sectional area. The guide is coated with hundreds of layers of $`{}_{}{}^{58}\mathrm{Ni}`$ and $`{}_{}{}^{47}\mathrm{Ti}`$. It allows neutrons with 3 times the normal perpendicular velocity to be transmitted, resulting in a large increase in neutron flux as compared to standard guides which employ only $`{}_{}{}^{58}\mathrm{Ni}`$ coating. A detailed description of our measurement of the FP12 moderator brightness and performance of the neutron guide is given in . The measured brightness has a maximum of $`1.25\times 10^8`$ $`\mathrm{n}/(\mathrm{s}\mathrm{cm}^2\mathrm{sr}\mathrm{meV}\mu \mathrm{A})`$ for neutrons with an energy of $`3.3`$ meV.
### 2.2 Beam Chopper
The beam chopper incorporates two blades which rotate independently at up to $`1200`$ rpm. The chopper is located $`9.38`$ m from the surface of the moderator. Since the flight path is about $`21`$ m long and the pulse period is $`50`$ ms the slowest neutrons that reach the end of the guide in each pulse have an energy of about $`1`$ meV. The blades are coated with a layer of $`\mathrm{Gd}_2\mathrm{O}_3`$ which was determined to be fully absorbing for neutron energies up to $`30`$ meV. This is used to block the slow neutrons at the tail end of the time-of-flight spectrum when either one or both of the blades cover the beam opening. The diameter of the blades is $`1.024`$ m and each blade covers $`4.38`$ radians of a full circle. The ability to select only part of the neutron spectrum is an important tool to control systematic errors since it provides the ability to effectively polarize, spin-flip and capture the neutrons. It prevents the overlapping of very slow neutrons from a previous pulse with the faster ones from the following pulse. At a distance of about $`22`$ m from the moderator, the time required to fully open or close the beam aperture is $`4`$ ms. During the 2004 commissioning run, the chopper rotation was phased with the beam pulses such that it began opening at each pulse onset $`(0\mathrm{ms})`$, was fully open $`4`$ ms after pulse onset, began closing about $`30`$ ms after pulse onset and was completely closed about $`4`$ ms later (Fig. 2). This allowed us to take beam-off (pedestal) data for $`6`$ ms at the end of each neutron pulse, which is needed for pedestal and background studies. The chopper control feedback loop kept the chopper in phase with the beam pulse to within $`30\mu \mathrm{s}`$.
### 2.3 Beam Monitors
The experiment uses three parallel plate ion chambers, each with a $`12\times 12\mathrm{cm}^2`$ active area as beam monitors. The first (upstream) monitor is located immediately after the neutron guide exit. The second monitor is located downstream of the $`{}_{}{}^{3}\mathrm{He}`$ polarizer to allow in situ absolute beam polarization measurements to be made. The third monitor is located downstream of the target and detector array. It was used during the 2004 commissioning run to study the spin flipper efficiency and will be used to monitor neutron depolarization in the hydrogen target.
### 2.4 Neutron Spin Filter
After exiting the neutron guide, the neutrons are spin filtered by passing through a 12 cm diameter glass cell containing polarized $`{}_{}{}^{3}\mathrm{He}`$. $`{}_{}{}^{3}\mathrm{He}`$ spin filters have a number of desirable features . They have large acceptance angles, do not require high magnetic fields as is the case with supermirrors and neutron capture on $`{}_{}{}^{3}\mathrm{He}`$ does not create a $`\gamma `$-ray background. The $`{}_{}{}^{3}\mathrm{He}`$ polarization and therefore the neutron polarization can be reversed without changing the direction of the holding field, by adiabatic fast passage of the $`{}_{}{}^{3}\mathrm{He}`$ spin. The neutron polarization can be measured with $`23`$% accruracy and without introducing large magnetic field gradients. The cross-section for capture $`(\sigma _a)`$ of neutrons with spin parallel to the $`{}_{}{}^{3}\mathrm{He}`$ nuclear spin has been measured to be $`0.01\pm 0.03`$ of the total absorption cross-section , but is assumed to be zero on theoretical grounds. So neutrons with spin anti-parallel to the $`{}_{}{}^{3}\mathrm{He}`$ nuclear spin are absorbed while those with spin parallel are mostly transmitted. If the time of flight (energy) of the neutrons is known, then the neutron polarization can be determined directly from the neutron transmission measurements . For NPDGamma, the figure of merit is the statistical accuracy that can be reached for a certain running time, which is a product of both, the neutron transmission and polarization $`P_n\sqrt{T_n}`$. The neutron transmission increases as a function of energy (decreases as a function of time of flight), whereas the neutron polarization decreases as a function of energy (Fig. (3)). In the analysis of the data taken during the 2004 commissioning run the neutron polarization was calculated for each run. The transmission spectrum obtained for each run was fitted with $`P_n=\mathrm{tanh}(\sigma _anlP_{He})`$, using a $`{}_{}{}^{3}\mathrm{He}`$ thickness of $`nl=4.84\mathrm{bar}\mathrm{cm}`$, which was measured with an unpolarized cell, during the commissioning run. For each run the $`{}_{}{}^{3}\mathrm{He}`$ polarization was extracted as a fit parameter.
### 2.5 Spin Flipper
In this experiment the asymmetry is measured continuously since the signals from opposite detectors in a pair are measured simultaneously for each spin state. However, the efficiency of the $`\gamma `$-ray detectors will change slowly due to a number of effects including temperature and crystal activation and the detector gains cannot be matched to an accuracy that would allow an asymmetry measurement to be made for each individual neutron pulse. In addition, an asymmetry measurement cannot be made to the required level of accuracy by simply measuring the signal in a given detector for one spin state and the corresponding signal in the same detector for the opposing spin state, made some time later after the neutron spin has been reversed using the neutron spin filter. This possibility is precluded because of pulse-to-pulse fluctuations in the beam current. Both situations lead to false asymmetries.
The primary technique for reducing false asymmetries generated by gain non-uniformities, slow efficiency changes and beam fluctuations is fast neutron spin reversal. This allows asymmetry measurements to be made for opposing detectors, removing sensitivity to beam fluctuations, and for consecutive pulses with different spin states, removing the sensitivity to detector gain differences, drifts, and fluctuations. The asymmetries can then be measured very close together in time, before significant drift occurs. By carefully choosing the sequence of spin reversal, the effects of drifts up to second order can be further reduced. To achieve this fast neutron spin reversal, the experiment employs a radio frequency neutron spin flipper (RFSF) which operates according to the principles of NMR, using a $`30`$ kHz magnetic field with an amplitude of a few Gauss. The neutron spin direction is reversed when the RFSF is on and is unaffected when it is off. During the 2004 commissioning run the spin flip efficiency was measured to be about 95% averaged over the beam cross-section.
### 2.6 Liquid Hydrogen Target
The NPDGamma liquid hydrogen target consists of a cylindrical $`20`$ l target vessel containing the liquid hydrogen, surrounded by a vacuum chamber. The hydrogen itself and the heat radiation shield, located around the vessel, are cooled by two cryogenic refrigerators. In the cooling process, the hydrogen is converted to liquid para-hydrogen, from its usual state of mostly ortho-hydrogen, to prevent the depolarization of the neutron spin in the target via spin-flip scattering. Monte Carlo calculations indicate that the $`30`$ cm diameter and $`30`$ cm long hydrogen vessel is large enough to capture about 60% of the incident neutron beam. The beam entrance windows in the vacuum chamber, radiation shield, and target vessel are as thin as possible to efficiently transmit the neutron beam and create minimal prompt capture radiation.
### 2.7 Detector Array
The detector array consists of 48 CsI(Tl) cubes arranged in a cylindrical pattern in 4 rings of 12 detectors each around the liquid hydrogen target. In addition to the conditions set on the detector array by the need to preserve statistical accuracy and suppress systematic effects, the array was also designed to satisfy criteria of sufficient spatial and angular resolution, high efficiency, and large solid angle coverage . To measure $`A_\gamma `$ to an accuracy of $`5\times 10^9`$ the experiment must detect at least $`10^{17}`$ $`\gamma `$-rays from $`\stackrel{}{n}+pd+\gamma `$ capture with high efficiency. The average rate of $`\gamma `$-rays deposited in the detectors for any reasonable run-time is therefore high. Because of the high rates and for a number of other reasons discussed in , the detector array uses accurate current mode $`\gamma `$-ray detection. Current mode detection is performed by converting the scintillation light from CsI(Tl) detectors to current signals using vacuum photo diodes (VPD), and the photocurrents are converted to voltages and amplified by low-noise solid-state electronics .
In current mode detection, counting statistics appears as the RMS width in the sample distribution, due to the fluctuation in the number of electrons produced at the photo-cathode of the VPD, as a result of the quasi instantaneous amount of energy deposited in the CsI crystal. During beam on measurements, the shot noise RMS width is given by
$$\sigma _{I_{\mathrm{shot}}}=\sqrt{2qI}\sqrt{f_B},$$
(2)
where $`q`$ is the amount of charge created by the photo cathode per detected $`\gamma `$-ray, $`I`$ is the average photo-current per detector and $`f_B`$ is the sampling bandwidth, set by the $`0.4`$ ms time bin width in the time of flight spectrum. Figure 4 shows the RMS width for a typical detector, as seen at the preamplifier output. The width from counting statistics is compared to the RMS width seen for beam-off electronic noise.
To ensure a timely and accurate measurement of the $`\gamma `$-ray asymmetry the detector array must operate at counting statistics which requires the RMS width from electronic noise to be significantly smaller than the width observed from real events. For a given $`q`$ and $`I`$, the RMS width expected from the detector design criteria at the preamplifier output is $`5.7\pm 0.3`$ mV. The error on the expected width is dominated by the accuracy to which we know the efficiency (number of photo-electrons per MeV) of the detector. Using the known proton current and an appropriate Monte Carlo model for neutron capture on $`\mathrm{B}_4\mathrm{C}`$ the number of photons entering a detector, per time bin, is $`3\times 10^4`$. The corresponding RMS width expected due to neutron counting statistics is $`6\pm 0.5`$ mV.
## 3 2004 Commissioning Targets Asymmetries
To determine the parity violating asymmetry in neutron-proton capture to the proposed accuracy, any possible false asymmetry from neutron capture on other materials must be measured. These asymmetries form a background which introduces a shift in the measured $`\stackrel{}{n}+pd+\gamma `$ asymmetry if they are non-zero and, at the very least, produce a dilution of the asymmetry, even if they are zero. The degree of shift or dilution in the asymmetry is proportional to the size of the background signal, relative to the signal of interest. We refer to these false asymmetries as neutron capture related or induced systematic effects. There are also instrumental systematic effects which arise due to changing equipment properties which may be correlated with the neutron spin. Some of the possible instrumental systematic effects have been briefly mentioned earlier and those related to the detector array in particular are discussed in detail in . A more detailed discussion of systematic effects related to neutron capture and scattering is provided in . It is difficult to model or calculate the level of parity-violation in these targets and to establish an upper level of their contribution it must be measured.
### 3.1 Asymmetry Definition
Due to the $`10`$ G holding field, surrounding the experimental apparatus, the neutrons are polarized vertically after leaving the <sup>3</sup>He spin filter. While taking hydrogen data, the parity violating asymmetry in n-p capture is therefore seen in a difference of the number of $`\gamma `$-rays going up and down. For the ideal experiment, the $`\gamma `$-ray cross section is proportional to $`Y=1+A_\gamma \mathrm{cos}\theta `$, where $`\theta `$ is the angle between the neutron polarization and the momentum of the emitted photon. A third term is introduced if a left-right (LR) asymmetry exists $`Y=1+A_\gamma \mathrm{cos}\theta +A_{\gamma ,LR}\mathrm{sin}\theta `$. However, as discussed earlier, the basic expression for the $`\gamma `$-ray yield is modified due to limitations in the properties of the experimental apparatus and interaction of neutrons with elements other than hydrogen. In calculating the final combined asymmetry one asymmetry was calculated for each detector pair and time bin and over any valid sequence of 8 macro pulses with the correct neutron spin state pattern. A so-called valid 8 step sequence of spin states is defined as ($``$). This pattern suppresses first and second order gain drifts within the sequence. The measured (raw) asymmetry $`(A_{raw}^{j,p})`$ for each detector pair $`(p)`$ can be extracted by forming a ratio of differences between detectors in a pair divided by their sum. After all correction factors have been applied, the final physics asymmetry for a given detector pair, spin sequence $`(j)`$, and neutron time of flight $`(t_i)`$ is given by
$`\left(A_{UD}^{j,p}(t_i)+\beta A_{UD,b}^{j,p}(t_i)\right)G_{UD}(t_i)`$ $`+`$ $`\left(A_{LR}^{j,p}(t_i)+\beta A_{LR,b}^{j,p}(t_i)\right)G_{LR}(t_i)`$
$`=`$ $`{\displaystyle \frac{\left(A_{raw}^{j,p}A_g^pA_f(t_i)A_{noise}^p\right)}{ϵ(t_i)P_n(t_i)S(t_i)}}`$
Where the background asymmetries $`(A_{UD,b}^{j,p},A_{LR,b}^{j,p})`$ and the relative signal level $`(\beta )`$ from the elements that cause them must be measured in auxiliary measurements. The relative, target out, background levels for the various targets were $`7\%\beta 17\%`$, depending on the shielding collimation and target geometries. Asymmetries from target out runs were measured to be zero. $`A_g^p`$ is the gain asymmetry between the detector pair and $`A_f(t_i)`$ is the asymmetry from pulse to pulse beam fluctuations. The neutron energy and detection efficiency weighted spatial average detector cosine (up-down asymmetry) with respect to the vertical is given by $`G_{UD}(t_i)cos(\theta )`$, while the detector sine (left-right asymmetry) is given by $`G_{LR}(t_i)sin(\theta )`$. Also included are the correction factors due to the neutron beam polarization $`(P_n(t_i))`$, the spin flip efficiency $`(ϵ(t_i))`$ and the level of beam depolarization in the target $`(S_n)`$. The beam depolarization for the targets we report on here was modeled and values for $`(S_n)`$ range from $`0.95`$ to $`1`$. It is important to realize that signal fluctuations that are not correlated with the switching of the neutron polarization direction will average out and don’t contribute to any asymmetry. It is, however, essential that these signals have an RMS width that is small compared to the RMS width in the asymmetries of interest (driven by counting statistics) so that they do not dilute the result and are averaged to zero quickly compared to the time it takes to measure the asymmetry to the desired accuracy. The product of the gain and beam fluctuation asymmetries was measured to be consistent with zero with a statistical error that was typically two orders of magnitude smaller than the error on the raw asymmetry (see RMS width in table 1). Possible false asymmetries due to electronic pickup and possible magnetic field induced gain changes in the detector VPDs have previously been measured and are consistent with zero to within $`5\times 10^9`$ .
The detector pair physics asymmetries as represented by eqn. LABEL:eqn:TBPHASY can then be combined in error weighted averages over the neutron time of flight spectrum to form a single asymmetry for the entire detector array for a single sequence of beam pulses. If beam intensity levels are monitored to be reasonably stable over the measurement time these sequence asymmetries can be histogrammed. Typical run lengths were $`8.3`$ minutes and included 10000 beam pulses or 1250 8-step sequences and the asymmetry measurements performed so far usually extended over several hundred runs.
### 3.2 Results
The known asymmetry in $`\mathrm{CCl}_4`$ was used to verify that a nonzero asymmetry can, in fact, be measured with this experimental setup. The $`\mathrm{CCl}_4`$ asymmetry was also used to verify the geometrical dependence of the pair asymmetries. For this purpose, each of the 24 pair asymmetries, extracted from the histogrammed 8-step sequence asymmetries from all data obtained with that target, were multiplied by its mean geometry factor and plotted versus its corresponding mean error. The resulting graph is shown in Fig. 5. The fit function used to extract the total array asymmetry is $`A_{UD}\mathrm{cos}\theta +A_{LR}\mathrm{sin}\theta `$. The $`{}_{}{}^{35}\mathrm{Cl}`$ asymmetries obtained from the $`\mathrm{CCl}_4`$measurements were previously measured by this collaboration and others. M. Avenier and collaborators found an Up-Down asymmetry of $`(21.2\pm 1.7)\times 10^6`$. The results of the asymmetry measurements performed during 2004 commissioning run are summarized in table 1. Systematic errors from the correction factors discussed above are less than 10% and are scaled by the asymmetry.
## 4 Conclusion
The NPDGamma experiment successfully completed a commissioning run in April 2004. It was shown here that each component in the experiment, except the hydrogen target, was commissioned during the 2004 run cycle and each component performed as designed. To establish the level of false asymmetries that may be present for the hydrogen production runs, several measurements were performed. Possible false asymmetries due to instrumental systematic effects involving the detector array and spin flipper were measured to be zero at the $`5\times 10^9`$ level. Asymmetries due to beam fluctuations were measured using the beam monitors. The beam asymmetry enters into the main data asymmetry as a product with the detector pair gain asymmetry with a combined RMS width of $`10^5`$, which is negligible. False asymmetries due to neutron capture on materials other than hydrogen were measured for Al, Cu, In and $`\mathrm{B}_4\mathrm{C}`$. These asymmetries were found to be consistent with zero. It is clear from the results obtained so far that NPDGamma incorporates a powerful experimental setup that can be used to measure very small parity violating asymmetries and that the experiment meets all criteria needed to perform a successful measurement of the weak parity-violating $`\gamma `$ asymmetry with an accuracy of $`5\times 10^9`$ in the neutron capture reaction $`\stackrel{}{n}+pd+\gamma `$ . |
warning/0507/gr-qc0507077.html | ar5iv | text | # Simple model of big-crunch/big-bang transition
## I Introduction
Presently available cosmological data suggest that our universe emerged from a state with extremally high density of physical fields Spergel:2003cb ; Bahcall:1999xn . It is called the cosmic singularity. For modelling the very early universe it is necessary to understand the nature of the singularity. We also believe that the cosmic singularity problem is inseparable from the problem of dark energy, which seems to be the most fundamental problem of contemporary physics (see, e.g. rbj ; Peebles:2002gy ; Padmanabhan:2002ji and references therein).
It is attractive to assume that the singularity consists of contraction and expansion phases. This way one opens door for the cyclic universe models Khoury:2001bz ; Steinhardt:2001vw ; Steinhardt:2001st ; Steinhardt:2004gk ; Bojowald:2004kt ; Gasperini:2002bn ; Lidsey:1999mc , which are expected to be free of the problem of beginning/end of the universe , i.e. creation/annihilation of space-time-matter-field from/into ‘nothing’. It is also proposed to use such models to get rid of the cosmic inflation Linde:2005ht ; Linde:2002ws , because of its puzzling features and difficulties Turok:2004yx ; Turok:2002yq ; GDS .
There exist at least two frameworks used for modelling the singularity: general relativity (GR) and string/M (SM) theory<sup>1</sup><sup>1</sup>1Since string/M theory is far from being complete, we make distinction between SM and GR.. One of the simplest models of singularity offered by GR is described by de Sitter’s space<sup>2</sup><sup>2</sup>2It is only symbolically called here the singularity; de Sitter space with topology $`^1\times 𝕊^1`$ has neither incomplete geodesics nor blowing up Riemann tensor invariants.. It is of the ‘big-bounce’ type, which may be treated as a gentle singularity. The Milne space, considered recently within SM scheme Khoury:2001bz , is the simplest spacetime modelling the ‘big-crunch/big-bang’ type singularity. It represents a violent model of the contraction/expansion transition.
Any reasonable model of the cosmic singularity should be able to describe quantum propagation of a fundamental object (e.g. particle, string, membrane,…) from the pre-singularity to post-singularity epoch. It is the most elementary criterion that should be satisfied. Some insight into the problem may be already achieved by studying dynamics of a test particle in low dimensional spacetime. Recently, we presented results concerning dynamics of a particle in the two-dimensional de Sitter space WP ; Piechocki:2003hh . The model has passed the above test. Classical and quantum dynamics of a particle is well defined in the entire universe including the big-bounce period.
Results of the present paper concern evolution of a particle in the compactified Milne space. We analyse classical and quantum dynamics of a free particle in two-dimensional spacetime. In Section II we specify geometry, topology and symmetry of the Milne space. Classical dynamics of a particle is carried out in Section III. We find constraint for dynamics, determine dynamical integrals, identify observables and introduce phase space of the system. Next, we analyse particle’s motion in the Milne space and specify local symmetry of the phase space. We consider four models of evolution of a particle across the singularity. Section IV is devoted to quantization of the classical system. Quantization is carried out by finding a self-adjoint representation of the algebra of observables. In Section V we suggest the way our results can be linked to cosmological solutions of some higher-dimensional effective field theories. We conclude in Section VI.
## II The compactified Milne space
### II.1 Geometry and topology
Let us define two quadrants, $``$, of the two-dimensional Minkowski space by
$$=\{(x^+,x^{})^2|x^+x^{}>0x^+=0=x^{}\},x^\pm :=x^0\pm x^1,$$
(1)
where $`x^0`$ (time) and $`x^1`$ (space) are coordinates of the Minkowski space with the signature defined by the line element $`ds^2=(dx^0)^2+(dx^1)^2`$. Next, let us introduce the finite boost transformation $`B`$ on $``$
$$B:(x^+,x^{})(e^{2\pi r}x^+,e^{2\pi r}x^{}),$$
(2)
where $`r`$ defines a boost. The factor space $`/B`$ is called the compactified Milne space, $`_C`$. Since specification of $`r`$ identifies the $`_C`$ space, the set of the compactified Milne spaces is uncountable.
It is convenient to introduce new coordinates $`t`$ and $`\theta `$ defined by
$$x^0=:t\mathrm{cosh}\theta ,x^1=:t\mathrm{sinh}\theta .$$
(3)
The line element in $`_C`$ reads
$$ds^2=dt^2+t^2d\theta ^2,$$
(4)
where $`(t,\theta )^1\times 𝕊^1`$.
To visualize the compactified Milne space, we present the isometric embedding of $`_C`$ into three-dimensional Minkowski space. It may be defined by the mapping
$$y^0(t,\theta )=t\sqrt{1+r^2},y^1(t,\theta )=rt\mathrm{sin}(\theta /r),y^2(t,\theta )=rt\mathrm{cos}(\theta /r),$$
(5)
and one has
$$\frac{r^2}{1+r^2}(y^0)^2(y^1)^2(y^2)^2=0.$$
(6)
Eq. (6) presents two cones with a common vertex at $`(y^0,y^1,y^2)=(0,0,0)`$. One may verify that the induced metric on (6) coincides with the metric (4).
The space $`_C`$ is locally isometric with the Minkowski space at each point, but at the vertex $`t=0`$. The neighborhood of that very special point cannot be made homeomorphic to an open circle in $`^2`$. For this reason $`_C`$ is not a manifold, but orbifold. Obviously, the Riemann tensor is not well defined there. At the locus $`t=0`$ the orbifold $`_C`$ has space-like singularity. However, it is of removable type because any time-like geodesic from the lower cone ($`t<0`$) linked with the vertex ($`t=0`$) can be extended to the time-like geodesic of the upper-cone ($`t>0`$). It is clear that such an extension cannot be unique because at $`t=0`$ the Cauchy problem for the geodesic equation is not well defined, due to the disappearance of space dimension.
The coefficient of $`d\theta ^2`$ in (4) disappears as $`t0`$. Therefore, one may use the $`_C`$ space to model a two-dimensional universe with the ‘big-crunch/big-bang’ singularity.
### II.2 Local symmetry
Solution to the Killing field equations with the metric (4) reads
$$\eta _1=\mathrm{cosh}\theta \frac{}{t}\frac{\mathrm{sinh}\theta }{t}\frac{}{\theta },\eta _2=\mathrm{sinh}\theta \frac{}{t}\frac{\mathrm{cosh}\theta }{t}\frac{}{\theta },\eta _3=\frac{}{\theta }.$$
(7)
One may easily verify that the Killing vectors (7) satisfy the algebra
$$[\eta _1,\eta _2]=0,[\eta _3,\eta _2]=\eta _1,[\eta _3,\eta _1]=\eta _2,$$
(8)
which is the $`iso(1,1)`$ Lie algebra Vil . The algebra (8) is well defined locally everywhere in $`_C`$ with exception of the singularity $`t=0`$.
## III Classical dynamics of a particle
An action integral, $`𝒜`$, describing a relativistic test particle of mass $`m`$ in a gravitational field $`g_{kl},(k,l=0,1)`$ may be defined by
$$𝒜=𝑑\tau L(\tau ),L(\tau ):=\frac{m}{2}(\frac{\dot{x}^k\dot{x}^l}{e}g_{kl}e),\dot{x}^k:=dx^k/d\tau ,$$
(9)
where $`\tau `$ is an evolution parameter, $`e(\tau )`$ denotes the ‘einbein’ on the world-line, $`x^0`$ and $`x^1`$ are time and space coordinates, respectively.
In case of $`_C`$ space the Lagrangian reads<sup>3</sup><sup>3</sup>3It is not well defined for $`t=0`$, unless one can give meaning to $`\dot{\theta }`$ at $`t=0`$; equations (11) and (12) suffer from the same problem..
$$L(\tau )=\frac{m}{2e}(t^2\dot{\theta }^2\dot{t}^2e^2).$$
(10)
The action (9) is invariant under reparametrization with respect to $`\tau `$. This gauge symmetry leads to the constraint
$$\mathrm{\Phi }:=(p_\theta /t)^2(p_t)^2+m^2=0,$$
(11)
where $`p_t:=L/\dot{t}`$ and $`p_\theta :=L/\dot{\theta }`$ are canonical momenta.
Variational principle applied to (9) gives at once the equations of motion of a particle
$$\frac{d}{d\tau }p_t\frac{L}{t}=0,\frac{d}{d\tau }p_\theta =0,\frac{L}{e}=0.$$
(12)
Thus, during evolution of the system $`p_\theta `$ is conserved. Owing to the constraint (11), $`p_t`$ blows up as $`t0`$ for $`p_\theta 0`$. This is a real problem, i.e. it cannot be avoid by a suitable choice of coordinates. It is called the ’blue-shift’ effect. However, trajectories of a test particle, i.e. nonphysical particle, coincide (by definition) with time-like geodesics of an empty spacetime, and there is no obstacle for such geodesics to reach/leave the singularity<sup>4</sup><sup>4</sup>4We continue this discussion in the subsections A, B, C and D, and in the conclusion section..
It is commonly known that Killing vectors of a spacetime may be used to find dynamical integrals of a particle, i.e. quantities which do not change during the motion of a point mass. In our case there exist three dynamical integrals and they can be determined as follows
$$I_1:=p_t\eta _1^t+p_\theta \eta _1^\theta =p_t\mathrm{cosh}\theta p_\theta \frac{\mathrm{sinh}\theta }{t},$$
(13)
$$I_2:=p_t\eta _2^t+p_\theta \eta _2^\theta =p_t\mathrm{sinh}\theta p_\theta \frac{\mathrm{cosh}\theta }{t},$$
(14)
$$I_3:=p_t\eta _3^t+p_\theta \eta _3^\theta =p_\theta ,$$
(15)
where $`\eta _a^T`$ and $`\eta _a^\theta `$ are components of the Killing vectors $`\eta _a(a=1,2,3)`$. Making use of (13)-(15) we may rewrite the constraint (11) in the form
$$\mathrm{\Phi }=I_2^2I_1^2+m^2=0.$$
(16)
For further analysis we introduce the phase space. It is defined to be the space of all particle geodesics. To describe a geodesic uniquely one may use two independent dynamical integrals. In case only one part of the Milne space is available for particle dynamics, for example with $`t<0`$, the phase space, $`\mathrm{\Gamma }`$, could be defined as
$$\mathrm{\Gamma }=\{(I_1,I_2,I_3)|I_2^2I_1^2+m^2=0,I_3=p_\sigma \}.$$
(17)
For the choice (17) the phase space may be parametrized by two variables $`\sigma `$ and $`p_\sigma `$ in the following way
$$I_1=m\mathrm{cosh}\sigma ,I_2=m\mathrm{sinh}\sigma ,I_3=p_\sigma .$$
(18)
One can easily check that
$$\{I_1,I_2\}=0,\{I_3,I_2\}=I_1,\{I_3,I_1\}=I_2,$$
(19)
where the Poisson bracket is defined as
$$\{,\}=\frac{}{p_\sigma }\frac{}{\sigma }\frac{}{\sigma }\frac{}{p_\sigma }.$$
(20)
Thus the dynamical integrals (13)-(15) and the Killing vectors (7) satisfy the same algebra. Using properties of the Poisson bracket we get
$$\{\mathrm{\Phi },I_a\}=0,a=1,2,3.$$
(21)
We define classical observables to be real functions on phase space which are: (i) gauge invariant, (ii) specify all time-like geodesics of a particle, and (iii) their algebra corresponds to the local symmetry of the phase space. It is clear, due to (16) and (21), that all dynamical integrals are gauge invariant. There exist two functionally independent combinations of them which specify all time-like geodesics (see (61) of appendix). We use them to represent particle observables (one may verify that they are gauge invariant).
### III.1 Specification of phase space and observables based on continuous symmetries
Let us denote by $`𝒮_{}`$ the part of spacetime $`_C`$ with $`t<0`$, the big-crunch/big-bang singularity by $`𝒮`$, and the part of $`_C`$ with $`t>0`$ by $`𝒮_{}`$.
By definition, a test particle with constant mass does not modify a background spacetime. Hence, we postulate that a particle arriving at the singularity $`𝒮`$ from $`𝒮_{}`$ is ‘annihilated’ at $`𝒮`$ and next, ‘created’ into $`𝒮_{}`$. There are four interesting cases of propagation depending on the way a particle may go across $`𝒮`$. In each case the propagation must be consistent with the constraint equation (11) and have constant $`p_\theta `$ (due to (12)) in $`𝒮_{}`$ and $`𝒮_{}`$. At $`𝒮`$ both (11) and (12) are not well defined.
In this subsection we consider the following propagation: particle following spiral geodesics winding clockwise the cone $`𝒮_{}`$ continues to move along clockwise spirals in $`𝒮_{}`$ (the same concerns propagation along anticlockwise spirals). Obviously, for $`p_\theta =0`$ particle trajectories are just straight lines both in $`𝒮_{}`$ and $`𝒮_{}`$. Apart from this we take into account the rotational invariance (with respect to the axis which coincides with the $`y^0`$-axis of 3d Minkowski frame defining (5)) of the space of particle trajectories which occur independently in $`𝒮_{}`$ and $`𝒮_{}`$.
It results from (58) of Appendix A that the set of all particle trajectories can be determined by two parameters $`(c_1,c_2)^1\times [0,2\pi [`$. Thus, the phase space $`\mathrm{\Gamma }_{}`$ of a particle in $`𝒮_{}`$ has topology $`^1\times 𝕊^1`$. The transition of a particle across $`𝒮`$ makes the dynamics in $`𝒮_{}`$ and $`𝒮_{}`$ to be, to some extent, independent so the phase space $`\mathrm{\Gamma }_{}`$ of a particle in $`𝒮_{}`$ has also the $`^1\times 𝕊^1`$ topology. Therefore, the phase space $`\mathrm{\Gamma }_C`$ of the entire system has the topology $`𝕊^1\times ^1\times 𝕊^1`$.
Now let us specify the local symmetry of either $`\mathrm{\Gamma }_{}`$ or $`\mathrm{\Gamma }_{}`$ by defining the Lie algebra of particle observables. The system has two independent degrees of freedom represented by the observables $`c_1`$ and $`c_2`$. Equation (58) tells us that $`c_2`$ has interpretation of position coordinate, whereas $`c_1`$ plays the role of momentum, owing to (59). With such an interpretation, it is natural to postulate the following Lie algebra for either $`\mathrm{\Gamma }_{}`$ or $`\mathrm{\Gamma }_{}`$.
$$\{c_1,c_2\}=1,\{,\}:=\frac{}{c_1}\frac{}{c_2}\frac{}{c_2}\frac{}{c_1}.$$
(22)
Heuristic reasoning we use to introduce the algebra (22) may be replaced by derivation. It is presented at the end of Appendix A.
Suppose the observables $`c_1`$ and $`c_2`$ describe dynamics in $`𝒮_{}`$, and let us assume that propagations in $`𝒮_{}`$ and $`𝒮_{}`$ are independent. In such case it would be convenient to introduce two new observables $`c_4`$ and $`c_3`$ in $`𝒮_{}`$ corresponding to $`c_1`$ and $`c_2`$. The Lie algebra in $`\mathrm{\Gamma }_C`$ would be defined as follows
$$\{c_1,c_2\}=1,\{c_4,c_3\}=1,\{c_i,c_j\}=0,\text{where}i=1,2\text{and}j=3,4$$
(23)
with the Poisson bracket
$$\{,\}:=\frac{}{c_1}\frac{}{c_2}+\frac{}{c_4}\frac{}{c_3}\frac{}{c_2}\frac{}{c_1}\frac{}{c_3}\frac{}{c_4}.$$
(24)
But from the discussion above it results that $`\mathrm{\Gamma }_C`$ has only three independent variables. We can encode this property modifying (23) and (24) by the condition $`c_4=c_1`$. Finally, we get
$$\{c_1,c_2\}=1,\{c_1,c_3\}=1,\{c_2,c_3\}=0,$$
(25)
with the Poisson bracket
$$\{,\}=\frac{}{c_1}\frac{}{c_2}+\frac{}{c_1}\frac{}{c_3}\frac{}{c_2}\frac{}{c_1}\frac{}{c_3}\frac{}{c_1}.$$
(26)
The type of propagation we have considered so far is consistent with the isometry (i.e., continuous symmetry) of the compactified Milne space. In the next subsection we increase respected symmetries to include the space inversion (i.e., discrete symmetry).
### III.2 Specification based on continuous and discrete symmetries
We take into account (as in case considered in subsection A) that $`𝒮_{}`$ and $`𝒮_{}`$ have the (clockwise and anticlockwise) rotational symmetry quite independently. Apart from this we assume that the singularity $`𝒮`$ may ‘change’ the clockwise type geodesics into anticlockwise ones, and vice-versa. From mathematical point of view such case is allowed because at $`𝒮`$ the space dimension disappears, thus $`p_\theta `$ is not well defined there, so it may have different signs in $`𝒮_{}`$ and $`𝒮_{}`$. Therefore, the space of geodesics has reflection type of symmetry independently in $`𝒮_{}`$ and $`𝒮_{}`$, which is equivalent to the space inversion separately in $`𝒮_{}`$ and $`𝒮_{}`$. The last symmetry is of discrete type, so it is not the isometry of the compactified Milne space. It is clear that the phase space $`\mathrm{\Gamma }_C`$ has the topology $`𝕊^1\times ^1\times 𝕊^1\times _2`$.
Proposed type of propagation of a particle through $`𝒮`$ may be characterized by the conservation of $`|p_\theta |`$ (instead of $`p_\theta `$ required in subsection A). The consequence is that now $`|c_1|=|c_4|`$ (instead of $`c_1=c_4`$ of subsection A). To obtain the algebra of observables we propose to put $`c_4=\epsilon c_1`$, where $`\epsilon =\pm 1`$ is a new descrete variable, into (23) and (24). Thus the algebra reads
$$\{c_1,c_2\}=1,\{c_1,c_3\}=\epsilon ,\{c_2,c_3\}=0,$$
(27)
with the Poisson bracket
$$\{,\}=\frac{}{c_1}\frac{}{c_2}+\epsilon \frac{}{c_1}\frac{}{c_3}\frac{}{c_2}\frac{}{c_1}\epsilon \frac{}{c_3}\frac{}{c_1}.$$
(28)
### III.3 The case trajectories in pre- and post-singularity epochs are independent
Now, we assume that there is no connection at all between trajectories in the upper and lower parts of the Milne space. For instance, spiral type geodesic winding the cone in $`𝒮_{}`$ may be ‘turned’ by $`𝒮`$ into straight line in $`𝒮_{}`$, and vice-versa. It means that we consider the case equations (11) and (12) are satisfied both in $`𝒮_{}`$ and $`𝒮_{}`$, but not necessarily at $`𝒮`$ (as in case considered in subsection B). In addition we propose that $`p_\theta `$ may equal zero either in $`𝒮_{}`$ or in $`𝒮_{}`$. Justification for such choices are the same as in the preceding subsection. Obviously, the present case also includes transitions of spiral geodesics into spiral ones, and straight line into straight line geodesics.
It is clear that now the algebra of observables coincides with (23) and (24), and the entire phase space $`\mathrm{\Gamma }_C`$ has the topology $`\mathrm{\Gamma }_{}\times \mathrm{\Gamma }_{}:=(𝕊^1\times ^1)\times (^1\times 𝕊^1)`$.
### III.4 The case space of trajectories has reduced form of rotational invariance
There is one more case we would like to consider: it can be obtain from the case considered in subsection A by ignoring the rotational invariance of the space of solutions assumed to exist separately in $`𝒮_{}`$ and $`𝒮_{}`$. Now we assume that the invariance does occur, but in the entire spacetime. Consequently, the algebra of observables is defined by Eq. (22).
Such type of symmetry of the space of geodesics appears, e.g. in case of propagation of a particle in two-dimensional one-sheet hyperboloid embedded in three-dimensional Minkowski space WP (2d de Sitter space with topology $`^1\times 𝕊^1`$).
## IV Quantization
By quantization we mean finding a self-adjoint representation of the algebra of classical observables<sup>5</sup><sup>5</sup>5We do not need the observables to be well defined globally, which would be required for finding an unitary representation of the corresponding Lie group.. We find that our quantization method is sufficient for analysis of evolution of a quantum particle across the vertex of $`_C`$. Such method was used in our previous papers WP ; Piechocki:2003hh dealing with dynamics of a particle in de Sitter space<sup>6</sup><sup>6</sup>6Lifting of self-adjoint representation of the algebra to the unitary representation of the corresponding Lie group was possible in case of the spacetime topology $`^1\times 𝕊^1`$, but could not be done in case of topology $`^2`$.. Applying the same quantization method in both cases enables the comparison of results.
In this paper our genuine spacetime is $`_C`$, i.e. the Milne space $``$ is only used as a tool for defining $`_C`$. Thus our main concern is quantization of particle dynamics in $`_C`$. It means that we do not intend to present the quantization scheme which is, e.g. boost-invariant. Quantum theories of a particle in $`_C`$ and $``$ are different because the phase spaces of both systems have different topologies<sup>7</sup><sup>7</sup>7We put the emphasis on the topology in a quantization scheme because it has basic importance..
Before we begin quantization, it is advantageous to redefine the algebra (25). It is known (see Piechocki:2003hh ; LP ; LP2 ; Brzezinski:1992gu ; Kowalski:1998hx ; Gonzalez:1998kj and references therein) that in case canonical variables $`(\pi ,\beta )`$ have the topology $`^1\times 𝕊^1`$, it is necessary to replace $`\beta `$ by $`U:=\mathrm{exp}(i\beta )`$, and replace the Poisson bracket
$$\{,\}=\frac{}{\pi }\frac{}{\beta }\frac{}{\beta }\frac{}{\pi }$$
(29)
by the bracket
$$<,>:=(\frac{}{\pi }\frac{}{U}\frac{}{U}\frac{}{\pi })U=\{,\}U.$$
(30)
So, in particular one gets $`<\pi ,U>=U`$, instead of $`\{\pi ,\beta \}=1`$.
### IV.1 Quantization corresponding to the continuous symmetry case
Applying the redefinition (30) to the algebra (25) leads to
$$c_1,U_2=U_2,c_1,U_3=U_3,U_2,U_3=0,$$
(31)
where $`U_2:=\mathrm{exp}(ic_2)`$ and $`U_3:=\mathrm{exp}(ic_3)`$, and where the algebra multiplication reads
$$,:=\left(\frac{}{c_1}\frac{}{U_2}\frac{}{U_2}\frac{}{c_1}\right)U_2+\left(\frac{}{c_1}\frac{}{U_3}\frac{}{U_3}\frac{}{c_1}\right)U_3.$$
(32)
One may verify that (32) defines the Lie multiplication.
Now, let us quantize the algebra (31). To begin with, we define the mappings
$$c_1\widehat{c_1}\psi (\beta )\phi (\alpha ):=i\frac{d}{d\beta }\psi (\beta )\phi (\alpha ),$$
(33)
$$U_2\widehat{U_2}\psi (\beta )\phi (\alpha ):=e^{i\beta }\psi (\beta )\phi (\alpha ),U_3\widehat{U_3}\psi (\beta )\phi (\alpha ):=e^{i\beta }\psi (\beta )e^{i\alpha }\phi (\alpha ),$$
(34)
where $`0\beta ,\alpha <2\pi `$. The operators $`\widehat{c_1},\widehat{U_2}`$ and $`\widehat{U_3}`$ act on the space $`\mathrm{\Omega }_\lambda \mathrm{\Omega }_{\stackrel{ˇ}{\lambda }}`$, where $`\mathrm{\Omega }_\lambda ,0\lambda <2\pi ,`$ is a dense subspace of $`L^2(𝕊^1)`$ defined as follows
$$\mathrm{\Omega }_\lambda =\{\psi L^2(𝕊^1)|\psi C^{\mathrm{}}[0,2\pi ],\psi ^{(n)}(2\pi )=e^{i\lambda }\psi ^{(n)}(0),n=0,1,2,\mathrm{}\}.$$
(35)
The space $`\mathrm{\Omega }_{\stackrel{ˇ}{\lambda }}`$ may be chosen to have more general form than $`\mathrm{\Omega }_\lambda `$. For simplicity, we assume that it is defined by (35) as well. However, we do not require that $`\stackrel{ˇ}{\lambda }=\lambda `$, which means that the resulting representation may be labelled by $`\stackrel{ˇ}{\lambda }`$ and $`\lambda `$ independently.
The space $`\mathrm{\Omega }_\lambda \mathrm{\Omega }_{\stackrel{ˇ}{\lambda }}`$ is dense in $`L^2(𝕊^1𝕊^1)`$, so the unbounded operator $`\widehat{c_1}`$ is well defined. The operators $`\widehat{U_2}`$ and $`\widehat{U_3}`$ are well defined on the entire Hilbert space $`L^2(𝕊^1𝕊^1)`$, since they are unitary, hence bounded. It is clear that $`\mathrm{\Omega }_\lambda \mathrm{\Omega }_{\stackrel{ˇ}{\lambda }}`$ is a common invariant domain for all three operators (33) and their products.
One may easily verify that
$$[\widehat{c_1},\widehat{U_2}]=\widehat{<c_1,U_2>},[\widehat{c_1},\widehat{U_3}]=\widehat{<c_1,U_3>},[\widehat{U_2},\widehat{U_3}]=\widehat{<U_2,U_3>},$$
(36)
($`[,]`$ denotes commutator), which shows that the mapping defined by (33) and (34) is a homomorphism.
The operator $`\widehat{c_1}`$ is symmetric on $`\mathrm{\Omega }_\lambda \mathrm{\Omega }_{\stackrel{ˇ}{\lambda }}`$, due to the boundary properties of $`\psi \mathrm{\Omega }_\lambda `$. It is straightforward to show that $`\widehat{c_1}`$ is self-adjoint by solving the deficiency indices equation RS for the adjoint $`\widehat{c_1}^{}`$ of $`\widehat{c_1}`$ (for more details see Appendix A of WP ).
The space $`\mathrm{\Omega }_\lambda `$ may be spanned by the set of orthonormal eigenfunctions of the operator $`\widehat{c_1}`$ with reduced domain from $`\mathrm{\Omega }_\lambda \mathrm{\Omega }_{\stackrel{ˇ}{\lambda }}`$ to $`\mathrm{\Omega }_\lambda `$, which are easily found to be
$$f_{m,\lambda }(\beta ):=(2\pi )^{1/2}\mathrm{exp}i\beta (m+\lambda /2\pi ),m=0,\pm 1,\pm 2,\mathrm{}$$
(37)
The space $`\mathrm{\Omega }_{\stackrel{ˇ}{\lambda }}`$ may be also spanned by the set of functions of the form (37).
We conclude that the mapping defined by (33) and (34) leads to the self-adjoint representation of (31).
### IV.2 Quantization corresponding to the continuous and discrete symmetries case
Making use of the method presented in preceding subsection we redefine the algebra (27) to the form
$$c_1,U_2=U_2,c_1,U_3=\epsilon U_3,U_2,U_3=0,$$
(38)
where $`\epsilon =\pm 1`$. We quantize the algebra (38) by the mapping
$$c_1\widehat{c_1}\psi (\beta )f_\epsilon \phi (\alpha ):=i\frac{d}{d\beta }\psi (\beta )f_\epsilon \phi (\alpha ),U_2\widehat{U_2}\psi (\beta )f_\epsilon \phi (\alpha ):=e^{i\beta }\psi (\beta )f_\epsilon \phi (\alpha ),$$
(39)
$$U_3\widehat{U_3}\psi (\beta )f_\epsilon \phi (\alpha ):=e^{i\beta \widehat{\epsilon }}e^{i\alpha }\psi (\beta )f_\epsilon \phi (\alpha ):=e^{i\beta \epsilon }\psi (\beta )f_\epsilon e^{i\alpha }\phi (\alpha ),$$
(40)
where $`\widehat{\epsilon }`$ is the operator acting on the two-dimensional Hilbert space $`E`$ spanned by the eigenstates $`f_\epsilon `$ defined by
$$\widehat{\epsilon }f_\epsilon =\epsilon f_\epsilon .$$
(41)
It is easy to check that
$$[\widehat{c_1},\widehat{U_2}]=\widehat{U_2},[\widehat{c_1},\widehat{U_3}]=\widehat{\epsilon }\widehat{U_3},[\widehat{U_2},\widehat{U_3}]=0.$$
(42)
The domain space of operators (39) and (40) is defined to be the space $`\mathrm{\Omega }_\lambda E\mathrm{\Omega }_{\stackrel{ˇ}{\lambda }}`$. It is evident that $`\widehat{\epsilon }`$ commutes with all operators, so the algebra (42) is well defined. It is easy to check (applying results of preceding subsection) that the representation is self-adjoint.
### IV.3 Quantization in case the system consists of two almost independent parts
In the last case, the only connection between dynamics in $`𝒮_{}`$ and $`𝒮_{}`$ is that a particle assumed to exist in $`𝒮_{}`$, can propagate through the singularity into $`𝒮_{}`$. It is clear that now quantization of the system may be expressed in terms of quantizations done separately in $`𝒮_{}`$ and $`𝒮_{}`$. To be specific, we carry out the reasoning for $`𝒮_{}`$:
The phase space has topology $`\mathrm{\Gamma }_{}=^1\times 𝕊^1`$ and the algebra of observables read
$$c_1,U_2=U_2.$$
(43)
Quantization of (43) immediately gives
$$c_1\widehat{c_1}\psi (\beta ):=i\frac{d}{d\beta }\psi (\beta ),U_2\widehat{U_2}\psi (\beta ):=e^{i\beta }\psi (\beta ),\psi \mathrm{\Omega }_\lambda ,$$
(44)
which leads to
$$[\widehat{c_1},\widehat{U_2}]=\widehat{<c_1,U_2>}=\widehat{U_2}.$$
(45)
It is obvious that the same reasoning applies to a particle in $`𝒮_{}`$.
At this stage we can present quantization of the entire system having phase space with topology $`\mathrm{\Gamma }_C:=\mathrm{\Gamma }_{}\times \mathrm{\Gamma }_{}`$. The algebra of classical observables reads
$$c_1,U_2=U_2,c_4,U_3=U_3,$$
(46)
with all other possible Lie brackets equal to zero.
Quantization of the algebra (46) is defined by
$$c_1\widehat{c_1}\psi (\beta )\phi (\alpha ):=i\frac{d}{d\beta }\psi (\beta )\phi (\alpha ),U_2\widehat{U_2}\psi (\beta )\phi (\alpha ):=e^{i\beta }\psi (\beta )\phi (\alpha ),$$
(47)
$$c_4\widehat{c_4}\psi (\beta )\phi (\alpha ):=\psi (\beta )\left(i\frac{d}{d\alpha }\phi (\alpha )\right),U_3\widehat{U_3}\psi (\beta )\phi (\alpha ):=\psi (\beta )e^{i\alpha }\phi (\alpha ),$$
(48)
where the domain of the operators $`\widehat{c_1},\widehat{c_4},\widehat{U_2}`$ and $`\widehat{U_3}`$ is $`\mathrm{\Omega }_\lambda \mathrm{\Omega }_{\stackrel{ˇ}{\lambda }}`$.
It is evident that presented representation is self-adjoint.
### IV.4 Time-reversal invariance
The system of a test particle in the Milne space is a non-dissipative one. Thus, its theory should be invariant with respect to time-reversal transformation $`T`$. The imposition of this symmetry upon the quantum system, corresponding to the classical one enjoying such an invariance, may reduce the ambiguity of quantization procedure commonly associated with any quantization method AST .
In our case the ambiguity is connected with the freedom in the choice of $`\lambda `$. Since $`0\lambda <2\pi `$, there are infinite number of unitarily non-equivalent representations for the algebras of observables considered in the preceding subsections. One may reduce this ambiguity following the method of the imposition of $`T`$-invariance used for particle dynamics in de Sitter’s space. However, imposition of the rotational invariance on the space of trajectories makes the definition of time-reversal invariance meaningless in cases considered in subsections A, B and C of section III. The $`T`$-invariance may be imposed only on the dynamics considered in the subsection D. The first step of quantization for this case is specified by Eqs. (43) and (44). The imposition of the $`T`$-invariance upon the system may be achieved by the requirement of the time-reversal invariance of the algebra (45). Formally, the algebra is $`\widehat{T}`$-invariant since
$$\widehat{T}\widehat{c_1}\widehat{T}^1=\widehat{c_1},\widehat{T}\widehat{U_2}\widehat{T}^1=\widehat{U_2}^1,$$
(49)
where $`\widehat{T}`$ denotes an anti-unitary operator corresponding to the transformation $`T`$. The first equation in (49) results from the correspondence principle between classical and quantum physics, because $`c_1`$ has interpretation of momentum of a particle. The assumed form of $`\widehat{U_2}`$ and anti-unitarity of $`\widehat{T}`$ lead to the second equation in (49). The formal reasoning at the level of operators should be completed by the corresponding one at the level of the domain space $`\mathrm{\Omega }_\lambda `$ of the algebra (45). Following step-by-step the method of the imposition of the $`T`$-invariance upon dynamics of a test particle in de Sitter’s space, presented in Sec.(4.3) of Piechocki:2003hh , leads to the result that the range of the parameter $`\lambda `$ must be restricted to the two values: $`\lambda =0`$ and $`\lambda =\pi `$.
Now, let us take into account that quantum theory is expected to be more fundamental than its classical counterpart (if the latter exists). In the context of the time-reversal invariance it means that $`\widehat{T}`$-invariance may be treated to be more fundamental than $`T`$-invariance. Applying this idea to quantum particle in the Milne space, we may ignore the lack of $`T`$-invariance of classical dynamics considered in subsections A, B and C. In these cases we propose to mean by the time-reversal invariance the $`\widehat{T}`$-invariance only. It may be realized by the requirement of $`\widehat{T}`$-invariance of the corresponding algebras. For instance, the algebra (42) is formally $`\widehat{T}`$-invariant if the observables transform as follows
$$\widehat{T}\widehat{c_1}\widehat{T}^1=\widehat{c_1},\widehat{T}\widehat{U_2}\widehat{T}^1=\widehat{U_2}^1,\widehat{T}\widehat{U_3}\widehat{T}^1=\widehat{U_3}^1,\widehat{T}\widehat{\epsilon }\widehat{T}^1=\widehat{\epsilon }.$$
(50)
We require the first equation of (50) to hold. All other equations in (50) result from the functional forms of $`\widehat{U_2}`$, $`\widehat{U_3}`$ and $`\widehat{\epsilon }`$, and the anti-unitarity of $`\widehat{T}`$. These analysis should be completed by the corresponding one at the level of the the domain space $`\mathrm{\Omega }_\lambda E\mathrm{\Omega }_{\stackrel{ˇ}{\lambda }}`$ of the algebra (42), but we do not enter into such details.
The imposition of $`\widehat{T}`$-invariance not only meets the expectation that a system with no dissipation of energy should have this property, but also helps to reduce the quantization ambiguity as it was demonstrated in the simplest case (It is clear that three other cases enjoy this reduction too.).
## V Relation to the FLRW comologies
It appears that our model of a particle in compactified 2D Milne space is only a toy model. In fact it may be linked to some cosmology models obtained by compactification to hyperbolic scalar target spaces of some higher-dimensional string/M theories Russo:2004am ; Bergshoeff:2005cp ; Bergshoeff:2005bt . It is very interesting that some cosmological solutions of these effective field theories can be interpreted<sup>8</sup><sup>8</sup>8Due to the construction based on the Maupertuis-Jacobi principle of classical mechanics Russo:2004am . in terms of time-like straight-line geodesics in higher-dimensional noncompactified Milne space $``$. It turns out that points at which these straight lines meet the Milne horizon may correspond to cosmological singularities of the original effective field theories.
To be specific, let us take a straight-line time-like geodesic in the 3D moduli space of flat FLRW universe, $``$, of the above field theory presented in Fig. 1 of Russo:2004am , and let us make the mapping of this straight-line into our 2D $`_C`$ space. It is not difficult to see that the image of its part which belongs to the past-Milne, is a spiral geodesic in the lover cone of $`_C`$. The part which lies in the future-Milne, maps onto a spiral geodesic of the upper cone of $`_C`$. The resulting spiral geodesic in $`_C`$ belong to the type of geodesics considered in subsection D. The big-crunch and big-bang represented in $``$ by two intersection points of the straight-line with the Milne horizon are mapped into the vertex of $`_C`$, i.e. into the big-crunch/big-bang singularity of our model spacetime. Therefore, at the classical level the big-crunch/big-bang singularity of $`_C`$ space may correspond to big-crunch and big-bang singularities of some effective higher-dimensional field theories.
Next step is investigation of this analogy at the quantum level. In paper Russo:2004am the part of time-like straight-line trajectory which is in the Rindler space, i.e. ‘non-Milne’ region of space, is interpreted as forbidden, because the scale factor, $`\eta `$, of the corresponding FLRW becomes complex there. The interpretation is that in that ‘hidden’ region quantum effects should be taken into account. The quantization is carried out by imposition of some suitable phase space constraint<sup>9</sup><sup>9</sup>9The constraint is connected with the Friedmann constraint of FLRW dynamics which translates into the mass-shell constraint of particle dynamics in the moduli space of FLRW universe. as an operator constraint on a Hilbert space. Solution to this equation defines the wave function of the universe which is analytic in $`\eta `$. The final conclusion is that a collapsing universe can pass the classically forbidden region, owing to the quantum mechanical tunnelling, into a region where it becomes an expanding universe Russo:2004am . In our case quantum description of a particle in $`_C`$, corresponding to the dynamics presented in the subsection D, is mathematically well defined. It is clear that there must exist some relation between both results. The problem is that our quantization method is quite different from the method applied in Russo:2004am . Finding explicit relation between both quantum models needs further analysis PWN .
Comparison of our results with the results of Bergshoeff:2005cp ; Bergshoeff:2005bt will become possible after we generalize the analysis of particle dynamics from compactified Milne to compactified Misner space. The latter includes the compactified Rindler space. By extending reasoning of the previous paragraph, we expect that particle dynamics in the compactified Rindler space may be related with the instanton phase of particle propagation in the moduli space of FLRW universes . As the result, the full cosmology/instanton solutions presented in Bergshoeff:2005cp ; Bergshoeff:2005bt may be linked to the dynamics of a particle in the compactified 2D Misner space. We postpone investigation of this possibility to our next papers.
## VI Conclusions
Finding specific model(s) of a quantum particle in the compactified Milne space is our main result. We have analysed four ways of particle’s transition across the singularity, but more cases are possible. It is so because at the singularity the equations defining classical dynamics are not well defined. It is the direct consequence of the fact that at the singularity the space dimension disappears, which causes that the Cauchy problem for time-like geodesics is not well defined. In case there is no clear reason to choose specific transition across the singularity, it acts as ‘generator’ of uncertainty in the propagation of a particle from the pre- to post-singularity era.
Extension of our model to higher dimensional compactified Milne space may be carried out (as it was done in case of de Sitter space Jorjadze:1999wb ) to make it more realistic, but we do not expect that the main conclusion would be changed owing to the Cauchy problem at the singularity which could not be avoided. However, such generalization should be done due to the connection of our results to cosmology models of higher dimensional effective field theories<sup>10</sup><sup>10</sup>10In general, moduli spaces of these theories are higher dimensional. considered in Russo:2004am ; Bergshoeff:2005cp ; Bergshoeff:2005bt .
The quantum theory depends on the assumptions one makes for the passage of a particle through the singularity. This way, however, one may put forward some hypothesis concerning its nature. Such flexibility does not occur in case of de Sitter space, owing to the uniqueness of particle dynamics WP ; Piechocki:2003hh .
It is amazing that time-like geodesics in $`_C`$ may have interpretation in terms of cosmological solutions of some sophisticated higher dimensional field theories. This connection deserves further investigation especially at the quantum level to reveal the nature of the cosmic singularity PWN .
Our analysis are based on the assumption that a classical particle is able to pass the singularity. Justification for such assumption is that we consider a test particle. Physical particle might collapse into a black hole at the singularity, modify the spacetime there, or both. We have also ignored the effect of particle’s own gravitational field on its motion Poisson:2004gg . Some modelling of these effects PWN may be carried out by considering particle dynamics in a spacetime which regularizes the space $`_C`$.
Our model concerns point-like objects. Next natural step would be examination of dynamics of extended objects like strings or membranes. According to string/M theory (see, e.g. JP ), they are more elementary than point particles. It was recently shown (see Durin:2005ix ; Pioline:2003bs ; Turok:2004gb ; Nekrasov:2002kf and references therein) that a test string in the zero-mode state twisted around the shrinking dimension propagates smoothly and uniquely across the Milne space singularity. It is interesting that strings in such states do not suffer from the blue-shift effect specific for a point particle. However, as it was pointed out in Pawlowski:2005bs , understanding of propagation of a string in the zero-mode state is not the end of the story. For drawing firm conclusions about the physics of the problem one should also examine the non-zero string modes and go beyond the semi-classical approximation.
## Appendix A
In this section we present solutions to the equations (12). For the Lagrangian (10) the equations read
$$\frac{d}{d\tau }\left(\frac{mt^2\dot{\theta }}{e}\right)=0,\ddot{t}\left(\frac{\dot{e}}{e}\right)\dot{t}+\dot{\theta }^2t=0,e^2=\dot{t}^2t^2\dot{\theta }^2.$$
(51)
From the first and third equations of (51) it is easily seen that
$$mt^2\dot{\theta }=ec_1\text{and}t^2\dot{\theta }^2=\dot{t}^2e^2,$$
(52)
respectively, where $`c_1`$. Combining equations of (52) we see at once that
$$e^2=t^2\dot{t}^2/((c_1/m)^2+t^2).$$
(53)
Making use of (53) to rewrite the identity $`\dot{e}/e=\frac{d}{d\tau }e^2/2e^2`$ ($`e0`$ for time-like geodesics) and substituting the resulting expression into (51) yields
$$\text{sgn}(\dot{\theta })=\text{sgn}(e)\text{sgn}(c_1),\dot{\theta }^2=\frac{c_1^2\dot{t}^2}{m^2t^4+c_1^2t^2},e^2=\dot{t}^2t^2\dot{\theta }^2.$$
(54)
(In what follows we choose $`\text{sgn}(e)=1`$ to be specific.)
Now, we will find solution to (54). Since the action (9) is reparametrization invariant, the equations (54) have this property too. The mapping $`\tau \overline{\tau }`$ leads to
$$e^2\left(\frac{d\tau }{d\overline{\tau }}\right)e^2\left(\frac{d\tau }{d\overline{\tau }}\right),\dot{\theta }\left(\frac{d\tau }{d\overline{\tau }}\right)\dot{\theta },\dot{t}\left(\frac{d\tau }{d\overline{\tau }}\right)\dot{t}$$
(55)
It means that we can arbitrarily choose either $`e`$ or $`\dot{t}`$. Since we consider time-like geodesics, we cannot choose an arbitrary $`\dot{\theta }`$ (there exist solutions to (54) with $`\dot{\theta }=0`$ and we are unable to assign to them other values). Let us choose
$$\dot{t}=1,\text{or equivalently}\tau :=t+C.$$
(56)
(For simplicity we put $`C=0`$.) The gauge (56) leads to
$$\left(\frac{d\theta }{dt}\right)^2=\frac{c_1^2}{m^2t^4+c_1^2t^2},e^2=\frac{t^2}{t^2+(c_1/m)^2}.$$
(57)
The solution of (57) reads
$$\theta (t)=\frac{d(\frac{c_1}{mt})}{\sqrt{1+(\frac{c_1}{mt})^2}}=\text{arsinh}\left(\frac{c_1}{mt}\right)+c_2,0c_2<2\pi .$$
(58)
(Note that $`\text{sgn}(c_1)=1`$, due to the choice $`\text{sgn}(e)=1`$ done after (54)).
Now we determine $`p_t`$ and $`p_\theta `$. Applying (52) and (53) we get
$$p_t=L/\dot{t}=m\frac{\dot{t}}{e}=m\sqrt{1+\left(\frac{c_1}{mt}\right)^2},p_\theta =L/\dot{\theta }=m\frac{\dot{\theta }t^2}{e}=c_1.$$
(59)
Finally, we rewrite the solution (58) in terms of the dynamical integrals (13)-(15). Let us notice that for $`c_1=0`$ we have
$$\theta =c_2\text{and}I_2/I_1=\mathrm{tanh}\theta .$$
(60)
Hence for $`c_1=0`$ we get $`\theta =\mathrm{tanh}^1(I_2/I_1)`$. Thus, the solution reads
$$\theta (t)=\mathrm{sinh}^1\left(\frac{I_3}{mt}\right)+\mathrm{tanh}^1\left(\frac{I_2}{I_1}\right).$$
(61)
We conclude that $`c_1`$ and $`c_2`$ parametrizing geodesics have the interpretation of particle observables.
Using the solution (58), we can rewrite the dynamical integrals in terms of $`c_1`$ and $`c_2`$ as follows
$$I_1=m\mathrm{cosh}(c_2),I_2=m\mathrm{sinh}(c_2),I_3=c_1.$$
(62)
Introducing the Poisson bracket by
$$\{,\}:=\frac{}{c_1}\frac{}{c_2}\frac{}{c_2}\frac{}{c_1},$$
(63)
one may easily verify that
$$\{I_1,I_2\}=0,\{I_3,I_2\}=I_1,\{I_3,I_1\}=I_2.$$
(64)
Thus, the algebras (64), (19) and (8) are isomorphic. Eqs. (64) characterize the local symmetry of the system for either $`t<0`$ or $`t>0`$. Owing to the obvious relation $`\{c_1,c_2\}=1`$, the variables $`c_1`$ and $`c_2`$ are canonical.
###### Acknowledgements.
We would like to thank Jean-Pierre Gazeau, Marek Pawłowski, Boris Pioline, Michał Spaliński and Neil Turok for helpful discussions, and to the anonymous referees for constructive criticisms. One of us (WP) would like to thank the European Network of Theoretical Astroparticle Physics ILIAS/N6 under contract number RII3-CT-2004-506222 for partial financial support. |
warning/0507/hep-th0507212.html | ar5iv | text | # 1 Introduction: string propagation in a chirally broken background
## 1 Introduction: string propagation in a chirally broken background
The history of attempts to describe the hadrons in the framework of a string theory derived from, or at least inspired by, QCD encompasses already more than 30 years (see, - as well as the reviews - and an incomplete list of references therein). The commonly cited arguments to justify a stringy description of QCD are the dominance of planar gluon diagrams in the large $`N`$ limit ‘filling in’ a surface (interpreted as the world-sheet of a string), the expansion in terms of surfaces built out of plaquettes in strong-coupling lattice QCD , and to some extent the incarnation of Regge phenomenology within QCD . Recently, the developments based on the Maldacena conjecture and holographic duality have added further strength to these arguments. The last but not the least is an advent of Nambu-Goto string in lattice QCD of static heavy quark and antiquark .
Clearly the simplest string models (bosonic string, supersymmetric string, …) do not lead to realistic amplitudes. The paradigmatic example is the Veneziano amplitude; expanding it in powers of the Mandelstam variables $`s`$ and $`t`$ one does not find the right Adler zeroes and, of course, reveals a tachyon in the spectrum. The supersymmetric version partially solves one of the problems by projecting out the tachyon, but the wrong chiral behavior persists. Both difficulties are absent in the phenomenologically inspired Lovelace-Shapiro amplitude , but this amplitude does not seem to derive from any known string theory and there are good reasons to believe that it is anyway incompatible with QCD asymptotics (see below).
Thus so far it is not yet clear what is a phenomenologically acceptable QCD string action, even though there is a motivated agreement based on universality considerations that in a certain kinematic regime the Nambu-Goto (or the Polyakov ) string action should be basically correct or, at least, provide the basic description. A general characteristic of all the above amplitudes (including, incidentally the Lovelace-Shapiro one) is that they lead to linearly rising trajectories. General arguments and recent work indicate that while this behavior corresponds to a confining theory with an infinite number of narrow resonances (in the large $`N_c`$ limit) it does not reproduce the chiral properties of the QCD correlators. In fact, it can be proven that any strictly linearly rising Regge trajectory leads to complete degeneracy between the vector and the axial-vector channels — not the way chiral symmetry is realized in QCD. Exponentially small (of the form $`\mathrm{exp}(an)`$, $`n`$ being the principal quantum number) deviations are required and that means that none of the existing amplitudes can reproduce the chiral properties of QCD.
It is quite plausible that the main reason for the presence of a tachyon in the spectrum and the wrong chiral properties is a wrong choice of the vacuum . One can make a parallel with scalar field theory with the potential $`V(\varphi )=\mu ^2\varphi ^2+\lambda \varphi ^4`$, that generates spontaneous symmetry breaking with a sensible ground state, but where perturbing around $`\varphi =0`$ gives negative $`m^2`$ values for all components. Thus we assume that the string amplitudes obtained through the use of the canonical vertex operators correspond to amplitudes for excitations perturbed around the wrong, unphysical vacuum.
A possible way to take into account the non-trivial nature of the QCD vacuum was suggested in and developed in . Namely, one can assume that in QCD chiral symmetry breaking takes place and the light (massless in the chiral limit) pseudoscalar mesons form the background of the QCD vacuum, whereas other massive excitations are assembled into a string. The massless pseudoscalars can be collected in a unitary matrix $`U(x)`$. This matrix transforms as $`U(x)U^{}(x)=LU(x)R^{}`$ under chiral transformations belonging to $`SU(3)_L\times SU(3)_R`$ and describe excitations around the non-perturbative vacuum. From the string point of view $`U(x)`$ is nothing but a bunch of couplings involving the string variable $`x_\mu (\tau ,\sigma )`$. The unitary matrix $`U(x)`$ has to be somehow coupled to the boundary of the string, which is where flavor ‘lives’. Our goal is to find eventually a consistent string propagation in this non-perturbative background.
An essential property of string theory is, certainly, conformal invariance. In the limit of large $`N_c`$ at least, the hadronic string action should obey re-parameterization and conformal invariance as describing zero-width, point-like resonance states<sup>1</sup><sup>1</sup>1Perhaps only in a dual manner – after all there is a natural scale in QCD and as we get to shorter and shorter distances the partonic picture eventually sets in.. Since conformal invariance must hold when perturbing the string around any vacuum we demand the new coupling to chiral fields, living on the boundary, to preserve conformal invariance too (compare with ). The requirement of conformal invariance will provide the equations of motion of the background fields and, indirectly, their Lagrangian.
In the present paper we begin by describing the basic characteristics of our approach. We start by elucidating of how to incorporate ’quarks’ at the end of the bosonic string, in a manner respectful with conformal and the chiral symmetry properties of QCD and its vacuum, by adding a suitable set of Grassmann variables , and, further on, establish the general setting of the formalism . We briefly review the results obtained in this way, most notably the phenomenologically successful prediction of the $`𝒪(p^4)`$ equations of motion and the related low-energy constants $`L_1`$, $`L_2`$ and $`L_3`$ of the effective chiral Lagrangian . We derive next a covariant version of the results at order $`p^2`$ by coupling external gauge fields. Then we proceed to the issue of deriving the odd intrinsic parity part of the action from this approach and we immediately deduce the need of including the spin degrees of freedom of the quarks (absent in the usual effective string treatment). By doing so we obtain rather easily the anomalous part of the effective action in two dimensions . Finally we turn to the four dimensional case that happens to be more involved. We discuss the general formalism and introduce a set of operators (that eventually turn out to be embedded into an algebra) that implement the spin-flavor coupling. In the subsequent sections we derive the counterterms at the one and two loop level that are subjected to vanish in order to guarantee conformal invariance. We see at once that the ’quarks’ represented by the Grassmann variables at the end of the string cannot be in a $`s=1/2`$ angular momentum state if one requires those counterterms to vanish and that they can be interpreted as parts of equations of motion of local non-linear sigma model. General considerations regarding the algebra satisfied by the spin-flavor coupling operators are presented.
### 1.1 Basic concepts
The hadronic string is described by the following conformal field theory action which has four dimensional Euclidean space-time as target space
$$𝒲_{str}=\frac{1}{4\pi \alpha ^{}}d^{2+ϵ}\sigma \left(\frac{\phi }{\mu }\right)^ϵ\sqrt{|g|}g_{ik}_ix_\mu _kx_\mu ,$$
(1)
where, for $`ϵ=0`$, in the conformal gauge $`\sqrt{|g|}g_{ik}=\delta _{ik}`$ and one takes
$$x_\mu =x_\mu (\tau ,\sigma );\mathrm{}<\tau <\mathrm{},0<\sigma <\mathrm{};i=\tau ,\sigma \mu =1,\mathrm{},4.$$
The conformal factor $`\phi (\tau ,\sigma )`$ is introduced to restore the conformal invariance in $`2+ϵ`$ dimensions<sup>2</sup><sup>2</sup>2Finally this factor becomes a dilaton degree of freedom extending the four-component hadronic string to a five-component one that however is beyond the scope of the present paper.. The Regge trajectory slope (related to the inverse string tension) is known to be universal $`\alpha ^{}0.9`$ GeV<sup>-2</sup> from the meson phenomenology .
We would like to couple the matrix in flavor space $`U(x)`$ containing the meson fields in a chiral invariant manner to the string degrees of freedom while preserving general covariance in the two dimensional coordinates and conformal invariance under local scale transformations of the two-dimensional metric tensor. Since the string variable $`x`$ does not contain any flavor dependence, we introduce two dimensionless Grassmann variables (‘quarks’) living on the boundary of the string sheet. The boundary quark fields $`\psi _L(\tau ),\psi _R(\tau )`$ transform in the fundamental representation of the light-flavor group $`SU(3)`$. The subscripts $`L,R`$ are related to the chiral spinors produced by the projectors (in what follows we use Euclidean space-time),
$$P_L=\frac{1}{2}(1+\gamma _5),P_R=\frac{1}{2}(1\gamma _5),\gamma _5=\gamma _0\gamma _1\gamma _2\gamma _3.$$
(2)
A local hermitian action $`S_b=𝑑\tau L^{(f)}`$ is introduced on the boundary $`\sigma =0,\mathrm{}<\tau <\mathrm{}`$ to realize the interaction with background chiral fields $`U(x(\tau ))=\mathrm{exp}(i\mathrm{\Pi }(x)/f_\pi )`$ where $`f_\pi 90MeV`$, the weak pion decay constant, relates the matrix field $`\mathrm{\Pi }(x)`$ to a bunch of light pseudoscalar mesons.
The boundary Lagrangian is chosen to be reparameterization invariant and in its minimal form reads
$`L_{min}^{(f)}`$ $`=`$ $`{\displaystyle \frac{1}{2}}i\left(\overline{\psi }_LU(1z)\dot{\psi }_R\dot{\overline{\psi }}_LU(1+z)\psi _R+\overline{\psi }_RU^+(1+z^{})\dot{\psi }_L\dot{\overline{\psi }}_RU^+(1z^{})\psi _L\right),`$ (3)
where a dot implies a $`\tau `$ derivative. The CP symmetry under transformation,
$$\psi \gamma _0\psi ;\psi _R\gamma _0\psi _L;\psi _L\gamma _0\psi _R;UU^{},$$
(4)
requires purely imaginary constants $`z=z^{}=\pm i|z|`$.
It is easy to see that the string action (1) is classically invariant under general coordinate transformations of the two dimensional world sheet. The fermion action is also automatically conformally invariant, because it does not contain the two dimensional world sheet metric tensor since it can be written as a line integral.
Upon obeying conformal invariance at the quantum level, one obtains the requirement of a vanishing beta-function; in this case a beta-functional of chiral fields and their derivatives as being coupling constants of boundary string interaction. This beta-functional constraints the chiral field $`U(x)`$ in order to have a consistent (i.e. conformally invariant) string propagation. They have to be interpreted as the equations of motion for the collective field $`U(x)`$. Adding the requirement of locality, the corresponding effective Lagrangian is uniquely reconstructed. In what concerns the parity-even part, this procedure will be explained in some more details in section 2.
It is well known that the bosonic string is inconsistent at $`d=4`$ and that a dependence on the conformal factor appears for non-critical dimensions. We regard this issue as collateral here, the reason being that in a covariant treatment inconsistencies appear only at the one-loop level in string perturbation theory. For an effective hadronic string of the type discussed here, this would involve going beyond the $`1/N_c`$ limit and then the exact correspondence of QCD with an string theory can be called into question anyway. In fact, nowhere in the calculation there appears any interference between the requirement of conformal invariance for the $`\mathrm{\Pi }`$-onic background and the string trace anomaly. The matter deserves further study though.
### 1.2 Feynman rules and perturbation theory
In order to develop a perturbative expansion we expand the function $`U(x(\tau ))`$ in powers of the string coordinate field $`x_\mu (\tau )=x_{0,\mu }+\stackrel{~}{x}_\mu (\tau )`$ around a constant $`x_0`$,
$$U(x(\tau ))=U(x_0)+\stackrel{~}{x}_\mu (\tau )_\mu U(x_0)+\frac{1}{2}\stackrel{~}{x}_\mu (\tau )\stackrel{~}{x}_\nu (\tau )_\mu _\nu U(x_0)+\mathrm{}U(x_0)+𝒱(\stackrel{~}{x}).$$
(5)
and look for the potentially divergent one particle irreducible diagrams. We classify them according to the number of loops. Each additional loop comes with a power of $`\alpha ^{}`$. One can find a resemblance to the familiar derivative expansion of chiral perturbation theory .
The free fermion propagator is
$$\psi _R(\tau )\overline{\psi }_L(\tau ^{})=U^{}(x_0)\theta (\tau \tau ^{}).$$
(6)
If we impose $`CP`$ symmetry then
$$\psi _L(\tau )\overline{\psi }_R(\tau ^{})=\psi _R(\tau )\overline{\psi }_L(\tau ^{})^{}=U(x_0)\theta (\tau \tau ^{}),$$
(7)
for unitary chiral fields $`U(x)`$.
The free boson propagator projected on the boundary is
$$\stackrel{~}{x}_\mu (\tau )\stackrel{~}{x}_\nu (\tau ^{})=\delta _{\mu \nu }\mathrm{\Delta }(\tau \tau ^{}),\mathrm{\Delta }(\tau \tau ^{})=\mathrm{\Delta }(0)\frac{\alpha ^{}}{ϵ},_\tau \mathrm{\Delta }(\tau \tau ^{})=0,$$
(8)
the latter results hold in dimensional regularization (see below).
The normalization of the string propagator can be inferred from the definition of the kernel of the N-point tachyon amplitude for the open string.
$$\mathrm{\Delta }(\tau _j\tau _l)=2\alpha ^{}\mathrm{ln}(|\tau _j\tau _l|\mu ).$$
(9)
Keeping in mind this definition let us determine the string propagator in dimensional regularization, restricted on the boundary. First we calculate the momentum integral in $`2+ϵ`$ dimensions,
$$\mathrm{\Delta }_ϵ(\tau )=\alpha ^{}\mathrm{\Gamma }\left(\frac{ϵ}{2}\right)\left|\frac{\tau \mu \sqrt{\pi }}{\phi }\right|^ϵ.$$
(10)
This dimensionally regularized propagator should be properly normalized to reproduce (9). It can be done by subtracting from (10) its value at $`\tau \mu =1`$
$$\mathrm{\Delta }_ϵ(\tau )|_{reg}=\alpha ^{}\mathrm{\Gamma }\left(\frac{ϵ}{2}\right)\left\{\left|\frac{\tau \mu \sqrt{\pi }}{\phi }\right|^ϵ\left|\frac{\sqrt{\pi }}{\phi }\right|^ϵ\right\}.$$
(11)
Therefrom one unambiguously finds the relation
$$\mathrm{\Delta }(0)=\alpha ^{}\mathrm{\Gamma }\left(\frac{ϵ}{2}\right)\left|\frac{\sqrt{\pi }}{\phi }\right|^ϵ\stackrel{ϵ0}{=}2\alpha ^{}\left[\frac{1}{ϵ}+C+\mathrm{ln}\phi \right]+𝒪(ϵ)\mathrm{\Delta }_ϵ2\alpha ^{}\mathrm{ln}\phi ,$$
(12)
where following the recipe of dimensional regularization we have taken $`ϵ<0`$ and hence the first term in (11) vanishes at $`\tau =0`$.
The two-fermion, $`N`$-boson vertex operators are generated by the expansion (5), from the generating functional $`Z_b=\mathrm{exp}(iS_b)`$ and Eq.(3). In particular, for the $`LR`$ transition one has vertices containing $`N`$ derivatives of the chiral field and $`N`$ bosonic coordinates $`\stackrel{~}{x}`$
$$V=\frac{1}{2}\left((1z)𝒱(\stackrel{~}{x})_\tau +(1+z)_\tau \left[𝒱(\stackrel{~}{x})\mathrm{}\right]\right),$$
(13)
and for the $`RL`$ transition the Hermitian conjugated vertex $`V^+`$ appears.
To implement the renormalization process we perform a loop (equivalent to a derivative) expansion and proceed to determine the counterterms required to make the theory finite. This will provide a beta functional for the couplings $`U(x)`$ and their derivatives, which shall be required to vanish up to the two loop level in order to implement the condition of vanishing conformal anomaly. The fact that we are working with a boundary field theory makes the required calculation quite manageable. For a detailed derivation of the different Feynman diagrams we refer the reader to and herein will report only the final expressions.
## 2 Summary of the parity-even sector results
In this section we summarize the main results of which are derived by using the previous Feynman rules.
At one-loop, the coefficient of the single pole gives the appropriate counterterms. First we determine the fermion propagator counterterm at the one loop order. Power counting indicates that this should be of $`𝒪(p^2)`$ in the chiral expansion; i.e. two derivatives acting on the $`U(x)`$ field. This gives the following counterterm
$$\delta ^{(2)}U=\mathrm{\Delta }(0)\left[\frac{1}{2}_\mu ^2U\frac{3+z^2}{4}_\mu UU^{}_\mu U\right],$$
(14)
The coupling constant must be imaginary to provide the CP symmetry. Its absolute value is determined by local integrability, i.e. by the requirement that the equation $`\delta U=0`$, derives from a local Weinberg action,
$$S_W=d^4x\frac{f_\pi ^2}{4}\text{tr}\left[_\mu U_\mu U^{}\right];\delta ^{(2)}U=\frac{\mathrm{\Delta }(0)}{f_\pi ^2}\frac{\delta S_W}{\delta U^{}(x)}.$$
(15)
The latter one constraints $`z^2=1`$. This condition also ensures the (perturbative) unitarity of the chiral field.
The next step is to consider the renormalization of the one-loop divergences in vertices with any number of “bosons” – string coordinates $`x_\mu (\tau ,\sigma =0)`$. Some of the divergences are removed by the $`U`$ redefinition we just discussed, as this automatically implies a counterterm for $`_\mu U`$, the one-boson, two-fermion tree-level vertex. This however is not sufficient to make these vertices finite and an extension of the boundary action is needed.
The relevant counterterms can be parameterized<sup>3</sup><sup>3</sup>3As compared to here we introduce the dimensionless constants $`g_i`$ factorizing out the Regge slope scale $`\alpha ^{}`$ . with three bare constants $`g_1`$ , $`g_2`$ and $`g_3`$ (which are real if CP invariance holds),
$`\mathrm{\Delta }L_{bare}`$ $`=`$ $`{\displaystyle \frac{i}{4}}\alpha ^{}(1z^2)\overline{\psi }_L((1z)g_\nu \dot{U}U^{}_\nu U(1+z)g^{}_\nu UU^{}_\nu \dot{U}`$ (16)
$`+zg_3_\nu UU^{}\dot{U}U^{}_\nu U)\psi _R+\text{h.c.},`$
where the complex constant $`g`$ is related to real constants from as follows,
$$g_1=2(\text{Re}g+\text{Im}g);g_2=2(\text{Re}g\text{Im}g).$$
(17)
This definition will be justified after generalization of boundary action in Sec. 5. Renormalization is accomplished by redefining the couplings $`g_i`$ in the following way
$$g_i=g_{i,r}\frac{\mathrm{\Delta }(0)}{\alpha ^{}}.$$
(18)
In spite of the fact that the new vertices are higher-dimensional , it turns out that their contribution into the trace of the energy-momentum tensor vanishes once the requirements of CP invariance and unitarity of $`U`$ are taken into account and therefore conformal invariance is not broken (see ). One can also prove that vertices with more boson legs are made automatically finite once the renormalization of $`U`$ and $`g_i`$ has been performed — this completes the renormalization program at the one-loop level.
At two loops calculations are certainly more involved, but still relatively simple since we are working in a boundary field theory. We need to consider here only the renormalization of the fermion propagator. At this order there are several contributions: genuine two-loop contributions, one-loop $`U`$-counterterms inserted in one-loop diagrams (both in the propagator and in the vertices), and also the counterterms we just discussed (16) inserted in the vertices of one-loop diagrams. We do not provide detailed formulae here because in section 5 we analyze in detail a generalization of these results that take into account spin effects. The expressions that are relevant for this section can be obtained from those in section 5 by simplifying to the present case.
After interpreting the vanishing beta-function condition as an equation of motion of the (parity-even) chiral Lagrangian one unambiguously obtains the low-energy constants appearing in the order $`p^4`$ chiral Lagrangian. They are expressed in terms of the product of the Regge trajectory slope $`\alpha ^{}0.9`$ GeV<sup>-2</sup>, $`f_\pi ^2`$ and certain rational numbers (equivalently they can be characterized by the ratio of $`f_\pi ^2`$ to the hadron string tension $`T=1/2\pi \alpha ^{}`$). The unique solution is
$`g_{1,r}=g_{2,r}=g_{3,r}=2;g=i;`$ (19)
$`2L_1=L_2={\displaystyle \frac{1}{2}}L_3={\displaystyle \frac{f_\pi ^2\alpha ^{}}{8}}={\displaystyle \frac{f_\pi ^2}{16\pi T}}10^3\xi .`$ (20)
This prediction fits quite well the phenomenological values . The small dimensionless parameter $`\xi `$ is an expansion parameter of Chiral Perturbation Theory and represents a natural scale for dimension-4 structural constants. Respectively in Section 5 it will be used in the parameterization of a most general local chiral Lagrangian.
Meantime the Lagrangian (3) only contains intrinsic parity-even terms and does not contain any operators which can eventually entail the anomalous P-odd part of the chiral dynamics. We turn to this interesting question next in sections 4 and 5.
## 3 Covariant equations of motion
Let us incorporate external abelian gauge fields into the boundary chiral action (3). The tree-level Lagrangian has to be translation- and time-reparameterization invariant and invariant under the gauge transformation, generated by an electric charge $`Q`$,
$`\psi (x)e^{i\mathrm{\Lambda }(x)Q}\psi (x),A_\mu (x)A_\mu (x)+Q_\mu \mathrm{\Lambda }(x),`$
$$U(x)e^{i\mathrm{\Lambda }(x)Q}U(x)e^{i\mathrm{\Lambda }(x)Q}.$$
(21)
Thus, in principle, the boundary Lagrangian can be constructed with the help of the covariant derivative projected on the boundary,
$$\dot{x}_\mu \left(_\mu iA_\mu (x)\right)=_\tau i\dot{x}_\mu A_\mu e^{i\mathrm{\Lambda }(x)Q}\left(_\tau i\dot{x}_\mu A_\mu \right)e^{i\mathrm{\Lambda }(x)Q}.$$
(22)
$$=\frac{i}{2}\overline{\psi }_L\left\{(1z)U(x)(_\tau i\dot{x}_\mu A_\mu )+(1+z)(_\tau i\dot{x}_\mu A_\mu )U(x)\right\}\psi _R+h.c.$$
(23)
However it turns out that for such a Lagrangian the corresponding fermion propagator is not gauge invariant, rather being bilocally covariant. As a consequence, the divergences do not form a gauge covariant combination and one ends up with equations of motion that do not derive from a manifestly gauge invariant Lagrangian. One has to proceed to the fermion fields dressed with the Dirac string phase factor so that the Lagrangian is written as
$$=\frac{i}{2}\overline{\mathrm{\Psi }}_L\left\{(1z)\stackrel{~}{U}(x)(_\tau i\dot{x}_\mu \mathrm{\Delta }A_\mu ^{})+(1+z)(_\tau i\dot{x}_\mu \mathrm{\Delta }A_\mu ^{})\stackrel{~}{U}(x)\right\}\mathrm{\Psi }_R+h.c.$$
(24)
Herein we redistribute the e.m. interaction between dressed fermions ($`\mathrm{\Psi }=e^{i\phi (x)_{}Q}\psi `$), chiral fields
$$U(x)\stackrel{~}{U}(x)=e^{i\phi (x)_{}Q}U(x)e^{i\phi (x)_{}Q};$$
(25)
and the covariant derivative. $`\phi _{}`$ is defined as
$$\phi _{}=\stackrel{~}{x}_\mu (\tau )A_\mu (x_0)+\underset{n=1}{\overset{\mathrm{}}{}}\frac{1}{(n+1)!}\stackrel{~}{x}_\mu (\tau )\stackrel{~}{x}_{\nu _1}(\tau )\mathrm{}\stackrel{~}{x}_{\nu _n}(\tau )_{\nu _1}\mathrm{}_{\nu _n}A_\mu (x_0);$$
(26)
whereas the remaining transversal part of the covariant derivative reads
$$\mathrm{\Delta }A_\mu ^{}=\underset{n=1}{\overset{\mathrm{}}{}}\frac{n}{(n+1)!}\stackrel{~}{x}_{\nu _1}(\tau ^{})\mathrm{}\stackrel{~}{x}_{\nu _n}(\tau ^{})_{\nu _1}\mathrm{}_{\nu _{n1}}F_{\nu _n\mu }(x_0).$$
(27)
Now in order to control the conformal symmetry we expand $`\stackrel{~}{U}(x)`$ around a constant background $`x_0`$ and look for the potentially divergent, one particle irreducible diagram:
$`\stackrel{~}{U}(x)`$ $`=(1i\phi _{}{\displaystyle \frac{1}{2}}\phi _{}^2++\mathrm{})(U(x_0)+\stackrel{~}{x}_\mu _\mu U(x_0)+{\displaystyle \frac{1}{2}}\stackrel{~}{x}_\mu \stackrel{~}{x}_\nu _\mu _\nu U(x_0)+\mathrm{})`$ (28)
$`\times (1+i\phi _{}{\displaystyle \frac{1}{2}}\phi _{}^2+\mathrm{})`$
$`=U(x_0)+\stackrel{~}{x}_\mu D_\mu U(x_0)+{\displaystyle \frac{1}{2}}\stackrel{~}{x}_\mu \stackrel{~}{x}_\nu D_\mu D_\nu U(x_0)+\mathrm{}`$
where the covariant derivative acts in the adjoint representation
$$D_\mu U[D_\mu ,U]=_\mu Ui[A_\mu ,U].$$
Using the perturbation expansion and vertex operators from Eq. (28) one arrives to the following expression:
$$\frac{1}{4}U(x_0)^{}D_\mu U(x_0)U(x_0)^{}D_\mu U(x_0)U(x_0)^{}\{(3+z^2)\mathrm{\Delta }(0)+(1z^2)\mathrm{\Delta }(\tau _A\tau _B)\}\mathrm{\Theta }(\tau _A\tau _B)$$
$$\frac{1}{2}\mathrm{\Delta }(0)\mathrm{\Theta }(\tau _A\tau _B)U(x_0)^{}D_\mu D_\mu U(x_0)U(x_0)^{}.$$
(29)
The divergence is eliminated by introducing an appropriate counterterm
$$\delta U=\frac{1}{2}\mathrm{\Delta }(0)\left\{D_\mu ^2U\left(\frac{3+z^2}{2}\right)D_\mu UU^{}D_\mu U\right\}=0,$$
(30)
and conformal symmetry is restored if $`\delta U`$ vanishes as before. The value $`z^2=1`$ provides the integrability of this chiral dynamics, i.e. its origin from the local gauged Weinberg Lagrangian. We see that the dressed field Lagrangian produces the gauge invariant chiral dynamics which is determined unambiguously at one-loop level. The extension of the covariantization to the $`p^4`$ terms is in progress.
## 4 Two-dimensional QCD and the WZW term
The chiral bosonization of hadronic string presented in previous sections is certainly incomplete as it does not include any quark spin degrees of freedom and therefore does not generate parity-odd chiral dynamics in the form of the chiral anomaly in the equations of motion and the Wess-Zumino-Witten chiral Lagrangian. To understand the way parity-odd terms could emerge from the hadronic string built over the chirally broken QCD vacuum we investigate the toy model of two-dimensional QCD.
While a parity-even chiral-field interaction on the line may be qualitatively associated with vector quark currents a parity-odd interaction must have relation to axial-vector currents. However in QCD<sub>2</sub>, in fact, vector and axial-vector fields couple to quarks with the same matrix vertex. Indeed, in two (Euclidean) dimensions the structure of Dirac $`\gamma `$ matrices (in terms of the Pauli matrices $`\sigma _a`$),
$$\gamma _0=\sigma _1;\gamma _1=\sigma _2,\gamma _2(\text{analog of}\mathrm{"}\gamma _5\mathrm{"})=\sigma _3=i\gamma _0\gamma _1,$$
allows to relate axial-vector and vector vertices as follows,
$$\gamma _\mu \gamma _2=iϵ_{\mu \nu }\gamma _\nu ,$$
(31)
in terms of antisymmetric tensor $`ϵ_{\mu \nu }=ϵ_{\nu \mu },ϵ_{01}=1`$. Meantime the O(2) algebra is generated by $`\sigma _{\mu \nu }\frac{1}{2}i[\gamma _\mu ,\gamma _\nu ]=ϵ_{\mu \nu }\gamma _2`$.
Accordingly the boundary Lagrangian may equally well include two types of couplings,
$`L^{(f)}`$ $``$ $`{\displaystyle \frac{1}{2}}i\left\{\overline{\psi }_L\left[\{_\tau ,U\}+\widehat{F}_{\mu \nu }\dot{x}_\mu _\nu U\right]\psi _R+\overline{\psi }_R\left[\{_\tau ,U^{}\}\widehat{F}_{\mu \nu }^{}\dot{x}_\mu _\nu U^{}\right]\psi _L\right\};`$
$`\widehat{F}_{\mu \nu }`$ $``$ $`z\delta _{\mu \nu }+ig_Aϵ_{\mu \nu }.`$ (32)
The CP symmetry (4) of the Lagrangian (32) holds only if
$$z=z^{};g_A=g_A^{};\widehat{F}_{\mu \nu }=\widehat{F}_{\mu \nu }^{}.$$
(33)
Now we develop string perturbation theory expanding the function $`U(x)`$ in powers of the string coordinate field $`x_\mu (\tau )=x_{0,\mu }+\stackrel{~}{x}_\mu (\tau )`$, then expanding the boundary action in powers of $`\stackrel{~}{x}_\mu (\tau )`$ and finally looking for divergences, i.e. violations of conformal symmetry. At one loop one obtains the following condition to preserve conformal symmetry ,
$$_\mu ^2U+\frac{1}{2}(3+z^2g_A^2)_\mu UU^{}_\mu Uig_Aϵ_{\mu \nu }_\mu UU^{}_\nu U=0.$$
(34)
Unitarity of chiral fields and local integrability of Eqs. of Motion constrains the coupling constants to fulfill the relation $`g_A^2z^2=1`$. The naive QCD value (if we trust the arguments presented in Appendix A) is $`g_A=1`$. This choice ($`z=0,g_A=1`$) corresponds to the correct value of the dim-2 anomaly (last term in (34)). Thus in QCD<sub>2</sub> the hadron string induces the WZW action from the vanishing the boundary $`\beta `$ function already at one-loop level.
In turn, in QCD<sub>4</sub> the anomaly and the WZW action have dimension 4 and 5 respectively and therefore they are generated by cancellation of two-loop divergences. Therefore the antisymmetric tensor $`ϵ_{\mu \nu \rho \lambda }`$ in anomalies must arise from the algebra of $`\widehat{F}_{\mu \nu }`$ matrices.
On the other hand, the boundary quark fields $`\psi _L(\tau ),\psi _R(\tau )`$ are, in fact, one-dimensional and one should not expect that they realize the fundamental, spin-1/2 representation of the Poincare group. This is because the projection on a line is not uniquely defined (see Appendix A) and to correct this projection consistently with the conformal symmetry and integrability we eventually have to introduce a more complicated algebra than the conventional Clifford one. Certain arguments in favor of this extension will be given in the next Section.
## 5 General formalism: renormalization at the one- and two-loop order
In the next sections we are going to translate the ideas above to the four dimensional case, develop the equations up to two loops, and try to set a general framework for the search of solutions satisfying unitarity of the $`U`$ matrices and CP invariance. The goal of this section is to ensure the renormalizability at one- and two-loops of our model.
### 5.1 One loop fermion propagator: A first guess.
The starting point in this section is the following Lagrangian on the boundary for the fermions $`\psi _L`$ and $`\psi _R`$, analogous to Eq. (32).
$`L^{(f)}`$ $`=`$ $`{\displaystyle \frac{1}{2}}i\left\{\overline{\psi }_L\left[\{_\tau ,U\}+F_{\mu \nu }\dot{x}_\mu _\nu U\right]\psi _R+\overline{\psi }_R\left[\{_\tau ,U^{}\}F_{\mu \nu }^{}\dot{x}_\mu _\nu U^{}\right]\psi _L\right\};`$
$`F_{\mu \nu }`$ $`=`$ $`z\delta _{\mu \nu }+g_\sigma \sigma _{\mu \nu }.`$ (35)
The interaction term proportional to the $`F_{\mu \nu }`$ encodes the spin degrees of freedom of the fermion variables once projected to the line (boundary of the string) where the fermions live. This is why, as a first choice, we have included $`\sigma _{\mu \nu }`$ in analogy to its two-dimensional partner $`\widehat{F}_{\mu \nu }`$.
Following the procedure explained in Sec.2 for the parity even part, and following the rules of Sec. 1.1 and 1.2 we will expand the $`U(x(\tau ))`$ in powers of the coordinate fields $`x_\mu (\tau )=x_{0,\mu }+\stackrel{~}{x}_\mu (\tau )`$. This will bring us a variety of operators; some vertices will contain $`F_{\mu \nu }`$ (those coming from the expansion of the second term in the Lagrangian) and some will not (those coming from the first term in the Lagrangian). We will perform all computations with this expanded Lagrangian.
After computation of the diagrams we see that the divergence in the $`RL`$ part of the fermion propagator<sup>4</sup><sup>4</sup>4From now on we focus our analysis on the divergences in the $`RL`$ part of the fermion propagator having in mind that the $`LR`$ part is reproduced by means of hermitian conjugation. at one loop takes the form
$$O_1=\frac{1}{4}\theta (AB)\mathrm{\Delta }(0)[2^2U+(3\delta _{\alpha \beta }+F_{\beta \alpha }F_{\alpha \beta }+F_{\alpha \mu }F_{\beta \mu })_\alpha UU^{}_\beta U].$$
(36)
We can compare this equation with Eq.(14). Borrowing the ideas from the two-dimensional case and motivated by the discussion in Appendix A we write (35).
Applying this definition to the one loop propagator we see that there appear two different channels. One channel related with the trace of the $`O_1`$, and another channel defined by its traceless part. Let us compute them separately.
On one side we can perform a trace in spinor space and find
$$\frac{1}{2}\text{tr}\left[O_1\right]=\frac{1}{4}\theta (AB)[2^2U+(3+z^2+3g_\sigma ^2)_\mu UU^{}_\mu U]$$
(37)
where the identity
$$\sigma _{\mu \rho }\sigma _{\nu \rho }=3𝕀\delta _{\mu \nu }2i\sigma _{\mu \nu }$$
(38)
has been used. We can recover the already known result of by making $`g_\sigma =0`$.
The divergence in Eq.(37) is eliminated by introducing an appropriate counterterm $`UU+\delta U`$
$$\delta U=\mathrm{\Delta }(0)\left[\frac{1}{2}_\mu ^2U\frac{3+z^2+3g_\sigma ^2}{4}_\mu UU^{}_\nu U\right],$$
(39)
Conformal symmetry is restored (the $`\beta `$-function is zero) if the above contribution vanishes. The unitarity of $`U`$ is compatible with the conformal symmetry saturation condition only if we demand
$$z^2+3g_\sigma ^2=1.$$
(40)
When taking into account the CP symmetry condition $`z=z^{}`$, $`g_\sigma =g_\sigma ^{}`$ one find the following bounds on these couplings constants
$`0|z|1,`$
$`{\displaystyle \frac{1}{\sqrt{3}}}|g_\sigma |0,`$ (41)
and if the ein-bein projector gives a correct hint (see Appendix A) then $`|z|=|g_\sigma |=1/2`$.
Let’s explore now the other channel, i.e. the traceless part of $`\delta U`$. The latter is, in this case, the part proportional to $`\sigma _{\mu \nu }`$, thus
$$\overline{O}_1=\theta (AB)i\frac{1}{2}\mathrm{\Delta }(0)\left\{ig_\sigma g_\sigma ^2\right\}\sigma ^{\mu \nu }_\mu UU^{}_\nu U$$
(42)
This is a completely new term not observed before which comes directly from the inner space in $`F_{\mu \nu }`$.
We remark that for the $`LR`$ propagation the divergence is just complex conjugated, i.e. has the coefficient $`\left\{ig_\sigma ^{}+(g_\sigma ^{})^2\right\}`$. This fixes $`g_\sigma =0,i`$ in order to make zero the non-scalar part. Both choices seem to be unacceptable because for $`g_\sigma =0`$ one does not reproduce the Wess-Zumino action and for $`g_\sigma =i`$ one cannot provide the vanishing $`\beta `$-function in the scalar channel for the CP invariant choice (33). After this negative result we must accept that this, most intuitive, choice of $`F_{\mu \nu }`$ is not convenient for our purposes.
### 5.2 One loop fermion propagator: General form.
At one loop we have been already able to see the incompatibility of the guess (35) with the unitarity and CP conditions for the model. At this point we must generalize our strategy allowing for more general forms of $`F_{\mu \nu }`$. This will make the spin content not well defined since general $`F_{\mu \nu }`$ can follow a more complicated algebra than the Clifford one. Exactly as in the previous guess, here we must consider that the $`F_{\mu \nu }`$ acts on an internal spinor space. This requires some care in the computations in order to keep the right ordering of the $`F_{\mu \nu }`$’s. This will be crucial in the renormalization process.
In this framework we recover the original Lagrangian (35) leaving $`F_{\mu \nu }`$ unspecified for the time being. The complete one loop contribution to the fermion propagator in its general form reads.
$`{\displaystyle \frac{1}{2}}\theta (AB)\mathrm{\Delta }(0)\{U^{}^2UU^{}+{\displaystyle \frac{1}{2}}U^{}_\sigma UU^{}_\lambda UU^{}[3\delta _{\sigma \lambda }(F_{\sigma \lambda }F_{\lambda \sigma })+F_{\sigma \gamma }F_{\lambda \gamma }]\}`$
$`+{\displaystyle \frac{1}{4}}\theta (AB)\mathrm{\Delta }(AB)U^{}_\sigma UU^{}_\lambda UU^{}[\delta _{\sigma \lambda }+(F_{\sigma \lambda }F_{\lambda \sigma })F_{\sigma \gamma }F_{\lambda \gamma }]`$ (43)
From this we can identify the general condition that unitarity of the $`U`$ imposes,
$$\delta _{\sigma \lambda }F_{\sigma \lambda }+F_{\lambda \sigma }+F_{\sigma \gamma }F_{\lambda \gamma }=0.$$
(44)
This relation is a first constraint for the algebra we have alluded to.
The next step in the program is to compute the divergent part of one loop vertex with one boson and two fermions legs, compute the counterterms needed and try to see whether they are sufficient to renormalize all $`n`$-boson two fermion vertices.
### 5.3 One loop contribution to two-fermion one-boson vertex
In what follows we are going to compute the one loop contribution to the one boson two fermions vertex following the same lines as in Sec.2. The one-loop diagrams considered are the same considered in , with the additional $`F_{\mu \nu }`$ structure.
All contributions to this vertex have been summarized in a table contained in Appendix B. In this table we separate the different structures in derivatives of the $`U`$ matrices, and we focus on their $`F`$ structure.
The next step will be to use the relations
$`\overline{x}_\rho (A)\theta (AB)`$ $`=`$ $`{\displaystyle d\tau _\tau \theta (A\tau )\overline{x}_\rho (\tau )\theta (\tau B)\mathrm{}\mathrm{}\mathrm{}},`$
$`\overline{x}_\rho (B)\theta (AB)`$ $`=`$ $`{\displaystyle d\tau \theta (A\tau )\overline{x}_\rho (\tau )_\tau \theta (\tau B)},`$
to convert the divergence in the one-boson vertex into a tree level contribution in the Lagrangian. The vertex operators extracted from this tree level expression are, of course, directly related to the counterterms we are looking for and they read,
$$d\tau \frac{i}{2}\{\overline{\psi }_L\overline{x}_\rho (\tau )\mathrm{\Phi }_\rho ^{(1)}\dot{\psi }_R\dot{\overline{\psi }}_L\overline{x}_\rho (\tau )\mathrm{\Phi }_\rho ^{(2)}\psi _R\},$$
(45)
where,
$`\mathrm{\Phi }_\rho ^{(1)}`$ $`=`$ $`(\delta _{\eta \rho }F_{\eta \rho })_\eta \delta U`$
$`\mathrm{\Delta }(0)\{{\displaystyle \frac{1}{4}}_\eta _\sigma UU^{}_\lambda U((\delta _{\sigma \gamma }F_{\sigma \gamma })(\delta _{\eta \rho }+F_{\eta \rho })(\delta _{\lambda \gamma }F_{\lambda \gamma })[F_{\eta \rho },F_{\sigma \lambda }](\delta _{\lambda \gamma }F_{\lambda \gamma }))`$
$`{\displaystyle \frac{1}{4}}_\sigma UU^{}_\lambda _\eta U((\delta _{\sigma \gamma }+F_{\sigma \gamma })(\delta _{\eta \rho }F_{\eta \rho })(\delta _{\lambda \gamma }+F_{\lambda \gamma })(\delta _{\sigma \gamma }+F_{\sigma \gamma })[F_{\lambda \gamma },F_{\eta \rho }])`$
$`+{\displaystyle \frac{1}{2}}_\sigma UU^{}_\eta UU^{}_\lambda U(\delta _{\eta \rho }(F_{\sigma \lambda }+F_{\lambda \sigma })\delta _{\sigma \lambda }F_{\eta \rho }F_{\sigma \gamma }F_{\eta \rho }F_{\lambda \gamma }+{\displaystyle \frac{1}{2}}[F_{\eta \rho },F_{\sigma \lambda }](\delta _{\lambda \gamma }F_{\lambda \gamma }))\}`$
$``$ $`(\delta _{\eta \rho }F_{\eta \rho })_\eta \delta U\varphi _\rho ,`$
$`\mathrm{\Phi }_\rho ^{(2)}`$ $`=`$ $`(\delta _{\eta \rho }+F_{\eta \rho })_\eta \delta U`$
$`+\mathrm{\Delta }(0)\{{\displaystyle \frac{1}{4}}_\eta _\sigma UU^{}_\lambda U((\delta _{\sigma \gamma }F_{\sigma \gamma })(\delta _{\eta \rho }+F_{\eta \rho })(\delta _{\lambda \gamma }F_{\lambda \gamma })[F_{\eta \rho },F_{\sigma \gamma }](\delta _{\lambda \gamma }F_{\lambda \gamma }))`$
$`{\displaystyle \frac{1}{4}}_\sigma UU^{}_\lambda _\eta U((\delta _{\sigma \gamma }+F_{\sigma \gamma })(\delta _{\eta \rho }F_{\eta \rho })(\delta _{\lambda \gamma }+F_{\lambda \gamma })(\delta _{\sigma \gamma }+F_{\sigma \gamma })[F_{\lambda \gamma },F_{\eta \rho }])`$
$`+{\displaystyle \frac{1}{2}}_\sigma UU^{}_\eta UU^{}_\lambda U(\delta _{\eta \rho }(F_{\sigma \lambda }+F_{\lambda \sigma })\delta _{\sigma \lambda }F_{\eta \rho }F_{\sigma \gamma }F_{\eta \rho }F_{\lambda \gamma }+{\displaystyle \frac{1}{2}}[F_{\eta \rho },F_{\sigma \gamma }](\delta _{\lambda \gamma }F_{\lambda \gamma }))\}`$
$`=`$ $`(\delta _{\eta \rho }+F_{\eta \rho })_\eta \delta U+\varphi _\rho .`$
Herein the following relation (induced from Eq.(44)),
$$[F_{\eta \rho },F_{\sigma \gamma }](\delta _{\lambda \gamma }F_{\lambda \gamma })=(\delta _{\sigma \gamma }+F_{\sigma \gamma })[F_{\eta \rho },F_{\lambda \gamma }],$$
(46)
has been used in order to make CP invariance manifest.
In these equations we have already separated two important parts. The first part is a first variation of $`U`$ in the Lagrangian’s interaction part, so it is already under control. The remainder of $`\mathrm{\Phi }_\rho ^{(i)}`$ is in fact the same in both $`i=1`$ and $`2`$ (only a sign makes a difference). Denoting the remainder as $`\varphi _\rho `$ and putting all together, one finds that the divergence is generated by the operators
$`{\displaystyle }\mathrm{d}\tau [{\displaystyle \frac{i}{2}}\{\overline{\psi }_L\{\overline{x}_\rho (\tau )_\rho (\delta U),_\tau \}\psi _R+{\displaystyle \frac{i}{2}}\overline{\psi }_L\dot{\overline{x}}_\rho (\tau )F_{\eta \rho }_\eta (\delta U)\psi _R]`$ (47)
$`+{\displaystyle d\tau \frac{i}{2}\overline{\psi }_L\dot{\overline{x}}_\rho (\tau )\varphi _\rho \psi _R}.`$ (48)
The two first terms are already taken care of by the $`\delta U`$ counterterm, while the last one is dictating us the counterterms to introduce to guarantee the finiteness of this vertex. Obviously, if we set $`F_{\mu \nu }=0`$ we recover the results of the spinless case.
### 5.4 Counterterms
In order to compensate the divergence in $`\varphi _\rho `$ we have to employ the counterterm
$`\alpha ^{}{\displaystyle d\tau \frac{i}{2}\overline{\psi }_L\dot{\overline{x}}_\rho \stackrel{~}{\varphi }_\rho \psi _R},`$ (49)
where the rescaling on the dimensional constant $`\alpha ^{}`$ has been introduced to simplify some algebraic relations that follow. The structure of $`\stackrel{~}{\varphi }_\rho `$ is essentially determined by $`\varphi _\rho `$ and can be codified as follows
$`\stackrel{~}{\varphi }_\rho `$ $`=`$ $`_\sigma _\eta UU^{}_\lambda UA_{\sigma \eta \lambda \rho }^{(1)}+_\sigma UU^{}_\eta _\lambda UA_{\sigma \eta \lambda \rho }^{(2)}+_\sigma UU^{}_\eta UU^{}_\lambda UA_{\sigma \eta \lambda \rho }^{(3)}.`$ (50)
Evidently the operator coefficients $`A^{(1)}`$ and $`A^{(2)}`$ are symmetric in a pair of indices,
$$A_{\sigma \eta \lambda \rho }^{(1)}=A_{\eta \sigma \lambda \rho }^{(1)};A_{\sigma \eta \lambda \rho }^{(2)}=A_{\sigma \lambda \eta \rho }^{(2)},$$
(51)
being contracted with symmetric chiral field tensors. One can find the similarities of the counterterm (50) with its counterpart of Sec.2 when $`F_{\mu \nu }`$ reduces to $`z\delta _{\mu \nu }`$. In the present case the terms attached to each chiral field $`U`$ structure are considerably more complex. The operator nature of $`F_{\mu \nu }`$ is a reason for a larger set of coupling constants in the operator coefficients $`A_{\sigma \eta \lambda \rho }^{(i)}`$’s. As before the finite, renormalized part of this larger set of constants is to be determined by the consistency equations of vanishing beta functions as well as of local integrability of dimension-4 components of Eqs. of motion. Let us present the actual form of $`A_{\sigma \eta \lambda \rho }^{(i)}`$ more explicitly
$`A_{\sigma \eta \lambda \rho }^{(1)}`$ $`=`$ $`A_{\sigma \eta \lambda \rho }^{(1,r)}{\displaystyle \frac{\mathrm{\Delta }(0)}{8\alpha ^{}}}((\delta _{\sigma \gamma }F_{\sigma \gamma })(\delta _{\eta \rho }+F_{\eta \rho })(\delta _{\lambda \gamma }F_{\lambda \gamma })[F_{\eta \rho },F_{\sigma \gamma }](\delta _{\lambda \gamma }F_{\lambda \gamma })+\{\sigma \eta \})`$
$`=`$ $`{\displaystyle \frac{1}{8}}(g^{(r)}{\displaystyle \frac{\mathrm{\Delta }(0)}{\alpha ^{}}})((\delta _{\sigma \gamma }F_{\sigma \gamma })(\delta _{\eta \rho }+F_{\eta \rho })(\delta _{\lambda \gamma }F_{\lambda \gamma })[F_{\eta \rho },F_{\sigma \gamma }](\delta _{\lambda \gamma }F_{\lambda \gamma })+\{\sigma \eta \})`$
$`A_{\sigma \eta \lambda \rho }^{(2)}`$ $`=`$ $`A_{\sigma \eta \lambda \rho }^{(2,r)}+{\displaystyle \frac{\mathrm{\Delta }(0)}{8\alpha ^{}}}((\delta _{\sigma \gamma }+F_{\sigma \gamma })(\delta _{\eta \rho }F_{\eta \rho })(\delta _{\lambda \gamma }+F_{\lambda \gamma })(\delta _{\sigma \gamma }+F_{\sigma \gamma })[F_{\lambda \gamma },F_{\eta \rho }]+\{\eta \lambda \})`$
$`=`$ $`{\displaystyle \frac{1}{8}}(\overline{g}^{(r)}{\displaystyle \frac{\mathrm{\Delta }(0)}{\alpha ^{}}})((\delta _{\sigma \gamma }+F_{\sigma \gamma })(\delta _{\eta \rho }F_{\eta \rho })(\delta _{\lambda \gamma }+F_{\lambda \gamma })(\delta _{\sigma \gamma }+F_{\sigma \gamma })[F_{\lambda \gamma },F_{\eta \rho }]+\{\eta \lambda \})`$
$`A_{\sigma \eta \lambda \rho }^{(3)}`$ $`=`$ $`A_{\sigma \eta \lambda \rho }^{(3,r)}{\displaystyle \frac{\mathrm{\Delta }(0)}{4\alpha ^{}}}\left(\delta _{\eta \rho }(F_{\sigma \lambda }+F_{\lambda \sigma })\delta _{\sigma \lambda }F_{\eta \rho }F_{\sigma \gamma }F_{\eta \rho }F_{\lambda \gamma }+{\displaystyle \frac{1}{2}}[F_{\eta \rho },F_{\sigma \gamma }](\delta _{\lambda \gamma }F_{\lambda \gamma })\right)`$ (52)
$`=`$ $`{\displaystyle \frac{1}{4}}\left(g_3^{(r)}{\displaystyle \frac{\mathrm{\Delta }(0)}{\alpha ^{}}}\right)\left(\delta _{\eta \rho }(F_{\sigma \lambda }+F_{\lambda \sigma })\delta _{\sigma \lambda }F_{\eta \rho }F_{\sigma \gamma }F_{\eta \rho }F_{\lambda \gamma }+{\displaystyle \frac{1}{2}}[F_{\eta \rho },F_{\sigma \gamma }](\delta _{\lambda \gamma }F_{\lambda \gamma })\right),`$
where $`A_{\sigma \eta \lambda \rho }^{(i,r)}`$ are renormalized operators which depend on all finite parameters we referred above. As compared to Sec.2 these expressions contain the three similar constants $`g^{(r)},\overline{g}^{(r)},g_3^{(r)}`$ but a more complicated algebraic structure.
The actual composition of the $`A_{\sigma \eta \lambda \rho }^{(i,r)}`$ is just a sum of products of the algebra elements $`F_{\mu \nu }`$ with independent finite constants. We follow a minimal renormalization scheme and restrict the form of the $`A_{\sigma \eta \lambda \rho }^{(i,r)}`$ by adopting only the same $`F`$ combinations which appear in the corresponding infinite part.
We notice also that CP invariance of the Lagrangian imposes the relations
$`A_{\sigma \eta \lambda \rho }^{(1)}=A_{\lambda \eta \sigma \rho }^{(2)},A_{\sigma \eta \lambda \rho }^{(3)}=A_{\lambda \eta \sigma \rho }^{(3)},`$ (53)
While the CP invariance of the divergent part holds manifestly due to the Eq.(44), it is the CP invariance of the renormalized part that we are interested in. This condition applied to the parameterization (52) dictates that,
$`\overline{g}^{(r)}=(g^{(r)})^{},g_3^{(r)}=(g_3^{(r)})^{},`$ (54)
hence we end up with three real variables Re $`g^{(r)}`$, Im $`g^{(r)}`$ and $`g_3^{(r)}`$ as in the scalar Lagrangian (16).
Now one must examine the two-boson two-fermion vertex in order to prepare the two-loop renormalization of the fermion propagator. We do not display this part of the computation since it does not bring new counterterms and the algebraic expressions are rather cumbersome. All divergences in the one-loop two-boson two-fermion vertex are proven to be renormalized with the one-boson two-fermion vertex counterterms. Thereby by translational invariance all one-loop divergences in all n-boson two-fermion vertex are also entirely renormalized. Thus the renormalization program at one loop is completed. The inclusion of one-boson two-fermion counterterms (50), (52) is sufficient to ensure the complete renormalization at one loop.
### 5.5 Dimension-4 divergences from one-loop counterterms and from two-loop contributions
There are ten two-loop one-particle irreducible diagrams which are listed in . The divergences in the propagator at two-loops can be separated into five separate pieces
$$\theta (AB)[d_I+d_{II}+d_{III}+d_{IV}+d_V].$$
(55)
The first and second piece contain the double pole divergence $`\mathrm{\Delta }^2(0)`$, the third, fourth and fifth pieces contain the single pole divergence $`\mathrm{\Delta }(0)`$.
The piece $`d_I`$ represents ‘the second variation’, or one-loop divergence in the one-loop divergence and it is removed by the one-loop renormalization, hence it vanishes together with the one-loop $`\beta `$-function, i.e. when the equations of motion are imposed.
The second part represents the remaining terms of order $`\mathrm{\Delta }^2(0)`$ in two loop diagrams after subtraction of $`d_I`$. This part is made of the contributions generated by the one-loop counterterm in the vertex with two fermions and one boson line, after its insertion in a one-loop diagram.
$`d_{III}`$ contains those single-pole divergences, proportional to $`\mathrm{\Delta }(0)`$, which are removed once the one-loop renormalization of $`U`$ in the finite nonlocal part of the fermion propagator at one loop is taken into account.
The inclusion of the counterterms (50),(52) modifies in fact the fermion propagator adding terms of higher order in derivatives (of dimension 4 in the count of Chiral Perturbation Theory). Eventually the following divergent contributions to the propagator $`\mathrm{\Delta }(0)`$ arise from the finite, renormalized part (52) of the counterterms (50) when introduced in one-loop diagrams
| Coefficient | U structure | F structure |
| --- | --- | --- |
| $`+\frac{1}{8}\alpha ^{}\theta (AB)\mathrm{\Delta }(0)`$ | $`U^{}_\alpha _\sigma UU^{}_\lambda _\beta UU^{}`$ | 0 |
| $`+\frac{1}{8}\alpha ^{}\theta (AB)\mathrm{\Delta }(0)`$ | $`U^{}_\alpha _\sigma UU^{}_\lambda UU^{}_\beta UU^{}`$ | $`2A_{\sigma \alpha \lambda \mu }^{(1,r)}(\delta _{\beta \mu }F_{\beta \mu })`$ |
| $`+\frac{1}{8}\alpha ^{}\theta (AB)\mathrm{\Delta }(0)`$ | $`U^{}_\alpha UU^{}_\sigma UU^{}_\lambda _\beta UU^{}`$ | $`+2(\delta _{\alpha \mu }+F_{\alpha \mu })A_{\sigma \beta \lambda \mu }^{(2,r)}`$ |
| $`+\frac{1}{8}\alpha ^{}\theta (AB)\mathrm{\Delta }(0)`$ | $`U^{}_\alpha UU^{}_\sigma _\beta UU^{}_\lambda UU^{}`$ | $`2((\delta _{\alpha \mu }+F_{\alpha \mu })A_{\sigma \beta \lambda \mu }^{(1,r)}A_{\alpha \sigma \beta \mu }^{(2,r)}(\delta _{\lambda \mu }F_{\lambda \mu }))`$ |
| $`+\frac{1}{8}\alpha ^{}\theta (AB)\mathrm{\Delta }(0)`$ | $`U^{}_\alpha UU^{}_\sigma UU^{}_\beta UU^{}_\lambda UU^{}`$ | $`2((\delta _{\alpha \mu }+F_{\alpha \mu })A_{\sigma \beta \lambda \mu }^{(3,r)}A_{\alpha \sigma \beta \mu }^{(3,r)}(\delta _{\lambda \mu }F_{\lambda \mu }))`$ |
Table 1. Divergent contribution to the propagator from the vertex counterterms.
In $`d_{IV}`$ we include the divergences that are eliminated when the additional counterterms in the one-boson vertices (those proportional to $`A^{(i)}`$) are included in the finite part of the one-loop fermion propagator. One can check (in a way similar to ) that all terms in the two loop fermion propagator linear in $`\mathrm{\Delta }(0)`$ and in $`\mathrm{\Delta }(A,B)`$ belong either to $`d_{III}`$ or to $`d_{IV}`$.
Finally, some single-pole divergences remain and they are gathered in $`d_V`$. Namely, there are divergences linear in $`\mathrm{\Delta }(0)`$ which come from the double integral in irreducible two-loop diagrams with maximal number of vertices (overlapping divergences),
$$J(AB)=_B^A𝑑\tau _1_B^{\tau _1}𝑑\tau _2_{\tau _1}\mathrm{\Delta }(\tau _1\tau _2)_{\tau _2}\mathrm{\Delta }(\tau _1\tau _2)=2\alpha ^{}\mathrm{\Delta }(0)+\text{finite part}.$$
(56)
These operators are described in the table
| Coefficient | U structure | F structure |
| --- | --- | --- |
| $`+\frac{1}{8}\alpha ^{}\theta (AB)\mathrm{\Delta }(0)`$ | $`U^{}_\alpha _\sigma UU^{}_\lambda _\beta UU^{}`$ | $`(F_{\sigma \lambda }F_{\beta \alpha }\delta _{\alpha \beta }F_{\sigma \gamma }F_{\lambda \gamma }+\{\alpha \sigma ,\lambda \beta \})`$ |
| $`+\frac{1}{8}\alpha ^{}\theta (AB)\mathrm{\Delta }(0)`$ | $`U^{}_\alpha _\sigma UU^{}_\lambda UU^{}_\beta UU^{}`$ | $`(F_{\alpha \rho }(\delta _{\sigma \lambda }F_{\lambda \sigma })(\delta _{\beta \rho }+F_{\beta \rho })`$ |
| | | $`F_{\sigma \gamma }(\delta _{\lambda \gamma }F_{\lambda \gamma })(\delta _{\alpha \beta }+F_{\beta \alpha })+\{\alpha \sigma \})`$ |
| $`+\frac{1}{8}\alpha ^{}\theta (AB)\mathrm{\Delta }(0)`$ | $`U^{}_\alpha UU^{}_\sigma UU^{}_\lambda _\beta UU^{}`$ | $`((\delta _{\alpha \beta }F_{\alpha \beta })(\delta _{\sigma \gamma }+F_{\sigma \gamma })F_{\lambda \gamma }`$ |
| | | $`(\delta _{\alpha \rho }F_{\alpha \rho })(\delta _{\sigma \lambda }+F_{\sigma \lambda })F_{\beta \rho }+\{\lambda \beta \})`$ |
| $`+\frac{1}{8}\alpha ^{}\theta (AB)\mathrm{\Delta }(0)`$ | $`U^{}_\alpha UU^{}_\sigma UU^{}_\lambda UU^{}_\beta UU^{}`$ | $`((\delta _{\alpha \rho }+F_{\alpha \rho })(\delta _{\sigma \gamma }+F_{\sigma \gamma })(\delta _{\lambda \gamma }F_{\lambda \gamma })`$ |
| | | $`\times (\delta _{\beta \rho }F_{\beta \rho })(\delta _{\alpha \rho }F_{\alpha \rho })`$ |
| | | $`\times (\delta _{\sigma \gamma }+F_{\sigma \gamma })(\delta _{\lambda \rho }F_{\lambda \rho })(\delta _{\beta \gamma }+F_{\beta \gamma }))`$ |
Table 2. Relevant part of arising overlapping divergences.
The terms in $`d_V`$ survive after adding all the counterterms and together with Table 1 are the only new genuine divergences that can contribute to the beta function (single poles). It must therefore be added to the equation of motion at the next order in the $`\alpha ^{}`$ expansion. Thus two sets of operators listed in Tables 1 and 2 form the genuine contribution $`U^{}\delta ^{(4)}UU^{}`$ of chiral dimension 4 into the beta-functional of the chiral field renormalization. To this order the condition of conformal invariance reads
$$\delta ^{(2)}U+\delta ^{(4)}U=0,$$
(57)
and this equation must be identified with an equation of local chiral dynamics if we deal with the Goldstone boson physics of pseudoscalar mesons. However such an identification is not unique as one may have certain terms in $`\delta ^{(4)}U`$ vanishing on the mass-shell $`\delta ^{(2)}U=0`$ . This is a logic of Chiral Perturbation Theory. Therefore in the comparison of the beta-functional $`\delta ^{(4)}U`$ and a relevant functional of local chiral dynamics one must include all possible operators vanishing on-shell.
## 6 Local integrability of dimension-4 part of Eqs. of motion
If the corresponding terms with four derivatives that we have found in the previous section originate from a dimension-four operators in a quasi-local effective Lagrangian then certain constraints are to be imposed on the constants $`A_\mu ^{(i,r)}`$.
On mass-shell such a Lagrangian has only three terms compatible with the chiral symmetry if we employ the dimension-two equations of motion (14),
$`S^{(4)}`$ $`=`$ $`{\displaystyle \frac{f_\pi ^2\alpha ^{}}{8}}{\displaystyle }d^4x\text{tr}(K_1_\mu U_\rho U^{}_\mu U_\rho U^{}+K_2_\mu U_\mu U^{}_\rho U_\rho U^{}`$ (58)
$`{\displaystyle \frac{1}{5}}{\displaystyle _0^1}dx_5K_3ϵ_{ABCDE}_AUU^{}_BUU^{}_CUU^{}_DUU^{}_EUU^{}),`$
the last operator being the celebrated Wess-Zumino-Witten term . The capital Latin indices
$`A,B,C,D,E=1,\mathrm{}5`$ mark tensors in the five-dimensional space with a compact fifth coordinate $`0x_51`$ whereas the Greek indices mark the four dimensional Euclidean coordinates. The fully antisymmetric tensor $`ϵ_{ABCDE}`$ is conventionally normalized to $`ϵ_{12345}=1`$. It is assumed also that $`U(x_5=0)=1`$ and $`U(x_5=1)U(x_\mu )`$. The normalization in the front of the integral of Eq.(58) is chosen to simplify the forthcoming consistency conditions.
The terms
$$_\mu ^2U_\rho U^{}_\rho UU^{},_\mu ^2U_\rho ^2U^{},(_\mu ^2)^2UU^{},_\mu _\rho U_\mu _\rho U^{}$$
which are in principle acceptable are reduced to the set (58) with the help of integration by parts in the action and of the dimension-two equations of motion (14) (on-shell conditions).
Variation of the previous Lagrangian gives the following addition to the equations of motion,
$`{\displaystyle \frac{1}{f_\pi ^2}}{\displaystyle \frac{\delta S^{(4)}}{\delta U}}`$ $`=`$ $`{\displaystyle \frac{\alpha ^{}}{8}}U^{}\{4K_1[_\mu _\rho UU^{}_\mu UU^{}_\rho U+_\mu UU^{}_\rho UU^{}_\mu _\rho U`$ (59)
$`_\mu UU^{}_\rho UU^{}_\mu UU^{}_\rho U2_\mu UU^{}_\rho UU^{}_\rho UU^{}_\mu U`$
$`+_\mu UU^{}_\rho ^2UU^{}_\mu U]`$
$`+2K_2[_\mu _\rho UU^{}_\mu UU^{}_\rho U+_\mu UU^{}_\rho UU^{}_\mu _\rho U`$
$`+2_\mu UU^{}_\mu _\rho UU^{}_\rho +_\mu ^2UU^{}_\rho UU^{}_\rho U+_\mu UU^{}_\mu U^{}_\rho ^2U`$
$`2_\mu UU^{}_\rho UU^{}_\mu UU^{}_\rho U_\mu UU^{}_\rho UU^{}_\rho UU^{}_\mu U`$
$`3_\mu UU^{}_\mu UU^{}_\rho UU^{}_\rho U]+K_3ϵ_{\alpha \sigma \lambda \beta }_\alpha UU^{}_\sigma UU^{}_\lambda UU^{}_\beta U\}U^{}.`$
Now we proceed to the comparison of the beta-functional $`\delta ^{(4)}U`$ and the four-derivative part of Eqs. of chiral dynamics (59). As it has been elucidated in the previous section one expects the entire identification on the dimension 2 mass-shell (i.e. after applying of the $`O(p^2)`$ Eqs. of motion). Off-shell one must extend (59) to the following general set of operators and coefficients for the various chiral field structures,
| U structure | F structure |
| --- | --- |
| $`\frac{\alpha ^{}}{8}U^{}_\alpha _\sigma UU^{}_\lambda _\beta UU^{}`$ | $`C_0\delta _{\alpha \sigma }\delta _{\beta \lambda }+B_{\alpha \sigma }^{(0)}\delta _{\beta \lambda }+\delta _{\alpha \sigma }B_{\beta \lambda }^{(1)}`$ |
| $`\frac{\alpha ^{}}{8}U^{}_\alpha _\sigma UU^{}_\lambda UU^{}_\beta UU^{}`$ | $`(2K_1+K_2)(\delta _{\alpha \beta }\delta _{\sigma \lambda }+\delta _{\sigma \beta }\delta _{\alpha \lambda })B_{\alpha \sigma }^{(0)}\delta _{\beta \lambda }`$ |
| | $`+\delta _{\alpha \sigma }(C_1\delta _{\beta \lambda }+B_{\beta \lambda }^{(2)})`$ |
| $`\frac{\alpha ^{}}{8}U^{}_\alpha UU^{}_\sigma UU^{}_\lambda _\beta UU^{}`$ | $`(2K_1+K_2)(\delta _{\alpha \beta }\delta _{\sigma \lambda }+\delta _{\alpha \lambda }\delta _{\sigma \beta })\delta _{\alpha \sigma }B_{\beta \lambda }^{(1)}`$ |
| | $`+\left(C_2\delta _{\alpha \sigma }+B_{\alpha \sigma }^{(3)}\right)\delta _{\beta \lambda }`$ |
| $`\frac{\alpha ^{}}{8}U^{}_\alpha UU^{}_\sigma _\lambda UU^{}_\beta UU^{}`$ | $`2K_2(\delta _{\alpha \sigma }\delta _{\beta \lambda }+\delta _{\alpha \lambda }\delta _{\beta \sigma })+\delta _{\sigma \lambda }(C_3\delta _{\alpha \beta }+B_{\alpha \beta }^{(4)})`$ |
| $`\frac{\alpha ^{}}{8}U^{}_\alpha UU^{}_\sigma UU^{}_\lambda UU^{}_\beta UU^{}`$ | $`(2(2K_1+K_2)C_3)\delta _{\alpha \beta }\delta _{\sigma \lambda }+(2K_2C_0C_1C_2)\delta _{\alpha \sigma }\delta _{\beta \lambda }`$ |
| | $`+4(K_1+K_2)\delta _{\alpha \lambda }\delta _{\sigma \beta }+K_3ϵ_{\alpha \sigma \lambda \beta }\delta _{\alpha \sigma }B_{\beta \lambda }^{(2)}`$ |
| | $`B_{\alpha \sigma }^{(3)}\delta _{\beta \lambda }\delta _{\sigma \lambda }B_{\alpha \beta }^{(4)}`$ |
Table 3. Dimension 4 operators from a local chiral Lagrangian supplemented by the ’additional’ off-shell contributions . See the text for an explanation of their necessity .
In this Table the numeric, $`C_i`$ and operator, $`B_i`$ coefficients have been inserted so that they compensate each other in the total sum on the mass-shell. Evidently the two operators $`B^{(0)},B^{(1)}`$ are symmetric tensors,
$$B_{\alpha \beta }^{(0)}=B_{\beta \alpha }^{(0)},B_{\alpha \beta }^{(1)}=B_{\beta \alpha }^{(1)}.$$
(60)
Let us identify this parameterization of local Lagrangian descendants with the coupling constants and operator coefficients arising from the vertices in Tables 1 and 2. In particular the coefficients $`C_i`$ and $`B_i`$ admit weaker algebraic restrictions on the operators $`F_{\mu \nu }`$. The pertinent consistency equations read, in the same order as in the preceding table,
$`(F_{\sigma \lambda }F_{\beta \alpha }\delta _{\alpha \beta }F_{\sigma \gamma }F_{\lambda \gamma })+(F_{\alpha \lambda }F_{\beta \sigma }\delta _{\sigma \beta }F_{\alpha \gamma }F_{\lambda \gamma })`$
$`+(F_{\sigma \beta }F_{\lambda \alpha }\delta _{\alpha \lambda }F_{\sigma \gamma }F_{\beta \gamma })+(F_{\alpha \beta }F_{\lambda \sigma }\delta _{\sigma \lambda }F_{\alpha \gamma }F_{\beta \gamma })`$ $`=`$ $`C_0\delta _{\alpha \sigma }\delta _{\beta \lambda }+B_{\alpha \sigma }^{(0)}\delta _{\beta \lambda }+\delta _{\alpha \sigma }B_{\beta \lambda }^{(1)};`$ (61)
$`(F_{\alpha \rho }(\delta _{\sigma \lambda }F_{\lambda \sigma })(\delta _{\beta \rho }+F_{\beta \rho })F_{\sigma \gamma }(\delta _{\lambda \gamma }F_{\lambda \gamma })(\delta _{\alpha \beta }+F_{\beta \alpha }))`$
$`+(F_{\sigma \rho }(\delta _{\alpha \lambda }F_{\lambda \alpha })(\delta _{\beta \rho }+F_{\beta \rho })F_{\alpha \gamma }(\delta _{\lambda \gamma }F_{\lambda \gamma })(\delta _{\sigma \beta }+F_{\beta \sigma }))`$ (62)
$`2A_{\sigma \alpha \lambda \mu }^{(1,r)}(\delta _{\beta \mu }F_{\beta \mu })`$ $`=`$ $`(2K_1+K_2)(\delta _{\alpha \beta }\delta _{\sigma \lambda }+\delta _{\sigma \beta }\delta _{\alpha \lambda })`$
$`B_{\alpha \sigma }^{(0)}\delta _{\beta \lambda }+\delta _{\alpha \sigma }(C_1\delta _{\beta \lambda }+B_{\beta \lambda }^{(2)});`$
$`((\delta _{\alpha \beta }F_{\alpha \beta })(\delta _{\sigma \gamma }+F_{\sigma \gamma })F_{\lambda \gamma }(\delta _{\alpha \rho }F_{\alpha \rho })(\delta _{\sigma \lambda }+F_{\sigma \lambda })F_{\beta \rho })`$
$`+((\delta _{\alpha \lambda }F_{\alpha \lambda })(\delta _{\sigma \gamma }+F_{\sigma \gamma })F_{\beta \gamma }(\delta _{\alpha \rho }F_{\alpha \rho })(\delta _{\sigma \beta }+F_{\sigma \beta })F_{\lambda \rho })`$ (63)
$`+2(\delta _{\alpha \mu }+F_{\alpha \mu })A_{\sigma \beta \lambda \mu }^{(2,r)}`$ $`=`$ $`(2K_1+K_2)(\delta _{\alpha \beta }\delta _{\sigma \lambda }+\delta _{\alpha \lambda }\delta _{\sigma \beta })`$
$`\delta _{\alpha \sigma }B_{\beta \lambda }^{(1)}+\left(C_2\delta _{\alpha \sigma }+B_{\alpha \sigma }^{(3)}\right)\delta _{\beta \lambda };`$
$`2((\delta _{\alpha \mu }+F_{\alpha \mu })A_{\sigma \beta \lambda \mu }^{(1,r)}A_{\alpha \sigma \beta \mu }^{(2,r)}(\delta _{\lambda \mu }F_{\lambda \mu }))`$ $`=`$ $`2K_2(\delta _{\alpha \sigma }\delta _{\beta \lambda }+\delta _{\alpha \lambda }\delta _{\beta \sigma })`$
$`+\delta _{\sigma \lambda }(C_3\delta _{\alpha \beta }+B_{\alpha \beta }^{(4)});`$
$`(\delta _{\alpha \rho }+F_{\alpha \rho })(\delta _{\sigma \gamma }+F_{\sigma \gamma })(\delta _{\beta \gamma }F_{\beta \gamma })(\delta _{\lambda \rho }F_{\lambda \rho })`$
$`+(\delta _{\alpha \rho }F_{\alpha \rho })(\delta _{\sigma \gamma }+F_{\sigma \gamma })(\delta _{\beta \rho }F_{\beta \rho })(\delta _{\lambda \gamma }+F_{\lambda \gamma })`$
$`+2((\delta _{\alpha \mu }+F_{\alpha \mu })A_{\sigma \beta \lambda \mu }^{(3,r)}A_{\alpha \sigma \beta \mu }^{(3,r)}(\delta _{\lambda \mu }F_{\lambda \mu }))`$ $`=`$ $`(2(2K_1+K_2)C_3)\delta _{\alpha \beta }\delta _{\sigma \lambda }`$
$`+(2K_2C_0C_1C_2)\delta _{\alpha \sigma }\delta _{\beta \lambda }`$
$`+4(K_1+K_2)\delta _{\alpha \lambda }\delta _{\sigma \beta }+K_3ϵ_{\alpha \sigma \lambda \beta }`$
$`\delta _{\alpha \sigma }B_{\beta \lambda }^{(2)}B_{\alpha \sigma }^{(3)}\delta _{\beta \lambda }\delta _{\sigma \lambda }B_{\alpha \beta }^{(4)}.`$
These equations dictate the consistency conditions for the algebra of the operators $`F_{\mu \nu }`$ and bound the values of the low-energy constants $`K_1,K_2,K_3`$ in the chiral Lagrangian. The simplest solution for $`F_{\mu \nu }=z\delta _{\mu \nu }`$ of this set of conditions was obtained in and briefly described in Section 2. This solution, however, does not describe the spin degrees of freedom as implicitly assumes that quarks are scalar objects under rotations. The next simplest hypothesis is that $`F_{\mu \nu }`$ has an antisymmetric part proportional to $`\sigma _{\mu \nu }`$. Indeed, the coupling $`\overline{\psi }_L\sigma _{\mu \nu }\psi _R\times \dot{x}_\mu _\nu U`$ intuitively reflects the coupling between the string angular momentum (at the boundary) assuming that ’quarks’ have $`s=1/2`$ and the angular momentum of the $`U`$-field. As we have seen at the beginning of section 5 this not compatible with the one-loop renormalization properties of the model. We are then forced to conclude that the Grassmann variables are not in a state of well defined spin $`s=1/2`$.
It is possible to use the previous set of equations (61)–(6) to further constrain the operators $`F_{\mu \nu }`$. This is a rather non-trivial task. In next Section we explore some identities for operators $`F_{\mu \nu }`$ and discuss possible realizations of this algebra. Even though we do not have a final answer and, in a sense, Eqs.(44), (61)–(6) are our final result, the problem is interesting enough, deserving more detailed considerations.
## 7 Algebra considerations
It is probably worth to recapitulate where we stand.
The elimination of all divergences at the one-loop order requires, in addition to redefining the unitary matrix $`U`$, additional counterterms that are given in Eqs.(50), (52) in terms of a certain number of constants $`g^{(r)},\overline{g}^{(r)}g_3^{(r)}`$. In spite of the rather large number of structures involving $`F_{\mu \nu }`$, only three independent combinations appear in the counterterms described in Eqs.(50), (52). This is somewhat reminiscent of the situation without the spin structures, where three additional constants $`g_{i,r}`$, each one accompanying a different chiral structure, are engaged. The complication here lies of course in the fact that the $`F_{\mu \nu }`$ are operators taking values in some algebra yet to be specified.
Renormalizing the two-loop order propagator (i.e. $`U`$) needs taking all these one-loop counterterms. Adding all the contributions up leads to the conditions listed in Tables 1 – 3 and to the set of Eqs. (61)– (6).
As explained, these relations equate the single pole divergent part of the fermion propagator (a combination of chiral fields $`U`$, their derivatives, and operators $`F_{\mu \nu }`$) with the equivalent terms arising in equations of motion derived from the local Chiral Lagrangian (58). These equations of motion involve chiral fields and their derivatives, but not $`F_{\mu \nu }`$. If we insist, as we should, in making the two set of expressions equivalent this naturally brings about new relations involving the $`F_{\mu \nu }`$.
Through these equations we can learn more about the form of the $`F_{\mu \nu }`$ operator matrix and thus the way the spinor interaction degrees of freedom are implemented into this $`F_{\mu \nu }`$ operator, and of course, when possible, fix as much as we can the value of $`K_1`$, $`K_2`$ and $`K_3`$. These relations stem from the requirements of chiral invariance and locality of the effective action and they should be understood as restrictions that these conditions place on the algebra that the $`F_{\mu \nu }`$ satisfy.
To be specific, from Subsec. 5.2 , Eq. (44) we have
$$F_{\sigma \gamma }F_{\lambda \gamma }=\delta _{\sigma \lambda }+F_{\sigma \lambda }F_{\lambda \sigma };F_{\lambda \gamma }F_{\lambda \gamma }=4.$$
(66)
Next the fulfillment of Eq. (61) turns out to be very crucial as it removes the chiral field structure which is a serious obstruction for local integrability of Eqs. of motion. Therefrom, after contracting two of the indices with $`\delta _{\alpha \beta }`$, we obtain
$$F_{\gamma \sigma }F_{\gamma \lambda }=(9+C_0)\delta _{\sigma \lambda }5(F_{\sigma \lambda }F_{\lambda \sigma })\text{tr}\left[F\right]F_{\sigma \lambda }F_{\lambda \sigma }\text{tr}\left[F\right]+B_{\sigma \lambda }^{(0)}+B_{\sigma \lambda }^{(1)}.$$
(67)
As the components of the operator $`F_{\sigma \lambda }`$ are antihermitian it comes out from (67) that,
$$C_0=(C_0)^{};\left(B_{\sigma \lambda }^{(0)}+B_{\sigma \lambda }^{(1)}\right)^{}=B_{\sigma \lambda }^{(0)}+B_{\sigma \lambda }^{(1)}.$$
(68)
As well from the further contraction of indices $`\sigma =\lambda `$ one determines the trace of the operator $`F_{\sigma \lambda }`$,
$$(\text{tr}\left[F\right])^2=2C_016+\text{tr}\left[B^{(0)}+B^{(1)}\right].$$
(69)
We however stress that in general it represents an operator relation when one of the traces is not a c-number.
Finally, a non-equivalent contraction allows us to fix the symmetric part of twist-contracted products of $`F_{\gamma \sigma }`$,
$$F_{\gamma \sigma }F_{\lambda \gamma }+F_{\gamma \lambda }F_{\sigma \gamma }=2(C_01+\frac{1}{4}\text{tr}\left[B^{(0)}\right])\delta _{\sigma \lambda }+2B_{\sigma \lambda }^{(1)},$$
(70)
that allows for the determination of twisted normalization of the operator $`F_{\sigma \lambda }`$,
$$F_{\gamma \lambda }F_{\lambda \gamma }=4(C_01)+\text{tr}\left[B^{(0)}+B^{(1)}\right]=28+2(\text{tr}\left[F\right])^2.$$
(71)
All these algebraic relations originate from the requirement of local integrability of the would-be equations of motion. Notice that the last one (70) does not give us an explicit algebraic expression for the antisymmetric part. An ansatz admitting lineal in $`F`$ right-hand parts of Eqs. (67), (70) would close the algebra. However it happens to lead to a definite contradiction when the associativity of the algebra of contracted and twist-contracted products of three $`F_{\gamma \sigma }`$ is examined. Hence the ansatz is not correct and the algebra does not close.
Unfortunately, at the end of the day, we shall not have an explicit realization of the $`F_{\mu \nu }`$ satisfying all the previous requirements. Some obvious possibilities are however ruled out. We have already mentioned that the attempt of identifying the antisymmetric part of $`F_{\mu \nu }`$ with $`\sigma _{\mu \nu }`$ fails (see Subsection 5.1). It is somewhat more surprising that if Eqs. (44), (61) – (6) are to be imposed, the described algebra spanned by the $`F_{\mu \nu }`$ does not close, so it must necessarily be embedded in a larger algebra.
We are then forced to somewhat loosen the requirement of closure of the algebra. At this point, we discontinue the analysis of the implications of the algebraic relations (44), (67), (70). We regard these equations as constraints that the algebra of the $`F`$’s must satisfy in order to provide consistent propagation of the hadronic string in a chirally non-invariant vacuum when the spin degrees of freedom are taken into account. The remaining relations (62) – (6) are rather tools for the estimation of all coupling constants introduced on the boundary as well as the chiral constants $`K_1,K_2,K_3`$. This program nevertheless requires first to discover the algebra of $`F_{\mu \nu }`$ to be predictive .
## 8 Conclusions
In this work we have analyzed in detail the conditions that the effective string conceived to describe the interactions between quarks at long distances in QCD must meet. An essential ingredient for this string is the assumption that in the real QCD vacuum chiral symmetry is broken and the propagation takes place in a background of $`\mathrm{\Pi }`$-on fields (not states on the Regge trajectories). The condition of locality, chiral symmetry and conformal invariance place strong constraints on this background, eventually leading to vanishing beta functionals to be interpreted as equations of motion of the non-linear sigma model describing $`\mathrm{\Pi }`$-on interactions.
The work reported here dwells on a previous analysis where quarks (represented by Grassmann variables living on a line) were consider to be scalars. But spin is indeed an important variable in Regge analysis (let us recall here the existence of the so-called $`S`$ and $`D`$ Regge trajectories). More importantly, it is not difficult to see that without considering angular momentum, the odd (internal) parity of the $`\mathrm{\Pi }`$-on Lagrangian (i.e. the Wess-Zumino-Witten action) will never be obtained as one of the byproducts of requiring conformal invariance.
In the preceding pages a number of new results have been obtained. We have managed to couple an external gauge field and in this way to derive the covariant $`O(p^2)`$ equations of motion. The analysis of the Wess-Zumino-Witten action in dimension 2 turns out to be rather straightforward and it reproduces well the expected results.
Angular momentum in two dimensions is somewhat special and this is reflected in its realization in terms of gamma matrices. In fact the calculation can be fully reformulated using scalar variables. When proceeding to the four-dimensional case, things become rather more involved. We construct the general coupling that involves some operator coupling $`F_{\mu \nu }`$ (acting on the angular momentum degrees of freedom of the quarks). Consistency conditions of the string propagation indeed remarkably seem to suggest that the quarks are not in a definite state of angular momentum. A deeper reason may be in that hadron string realizes the Reggeization of meson states which, in the spirit of quark-hadron duality, presumably follows from a Reggeization of quarks and gluons as it happens in the semi-hard high-energy scattering in QCD . If such a quark-hadron duality holds then one cannot expect the boundary quarks to carry a definite spin. Rather they may be thought of in terms of an infinite-dimensional reducible representation of the Poincare group with any half-integer spin incorporated. Of course when $`F_{\mu \nu }`$ reduces to the scalar case, the results of are fully reproduced. These results are in excellent agreement with phenomenology.
We finally spelled out the restrictions that locality, chiral symmetry and conformal invariance place on the couplings $`F_{\mu \nu }`$ and formulated the way to search for the consistent realization of the $`F_{\mu \nu }`$ algebra.
## Acknowledgments
We are grateful to J. Alfaro for useful discussions and to P. Labraña for collaboration at the earlier stage of this work. A.A. was supported by Grant RFBR 05-02-17477 and Grant UR 02.01.299. The work of D.E. and A.P. was supported by the EURIDICE Network, by grant FPA2004-04582 and grant 2001SGR-00065. Exchange visits were supported by the CICYT-INFN bilateral agreements. A.P. acknowledges the support from a graduate fellowship from Generalitat de Catalunya.
## Appendix A. Ein-bein projection of the Dirac operator on the string boundary
The Lagrangian (3) does not contain any operators that could give rise to the anomalous P-odd part of the Chiral Dynamics. To approach the required modification let us guess on what might be the form of boundary Lagrangian if one derives it, say, from the essential part of the Chiral Quark Model projecting it on the string boundary. In what follows the Minkowski space-time is employed to keep the axial-vector vertex to be Hermitian.
Let us introduce the constituent quark fields to control properly the chiral symmetry during the ”ein-bein” projection,
$$Q_L\xi ^{}\psi _L,Q_R\xi \psi _R,\xi ^2U.$$
(72)
Under chiral rotations $`U\mathrm{\Omega }_RU\mathrm{\Omega }_L^+`$ the fields $`\xi `$ transform as follows
$$\xi h_\xi \xi \mathrm{\Omega }_L^+=\mathrm{\Omega }_R\xi h_\xi ^+,$$
(73)
with $`h_\xi `$ being a nonlinear functional of fields $`\xi `$. As a consequence the hidden vector symmetry of the constituent field action replaces the original chiral invariance.
In these variables the CQM Lagrangian density and the pertinent E.o.M. read
$$_{CQM}=i\overline{Q}\left(\overline{)}+\overline{)}v+g_A\overline{)}a\gamma _5\right)Q+\text{mass terms};i\left(\overline{)}+\overline{)}v+g_A\overline{)}a\gamma _5\right)Q+\text{mass terms}=0,$$
(74)
where
$`QQ_L+Q_R,\overline{)}A\gamma ^\mu A_\mu ,`$
$`v_\mu {\displaystyle \frac{1}{2}}(\xi ^{}(_\mu \xi )(_\mu \xi )\xi ^{}),a_\mu {\displaystyle \frac{1}{2}}(\xi ^{}(_\mu \xi )+(_\mu \xi )\xi ^{}),`$ (75)
and $`g_A1\delta g_A`$ is an axial coupling constant of quarks to $`\mathrm{\Pi }`$-ons. We skip all mass effects of the CQM, thereby neglecting the current quark mass in the chiral limit whereas relegating the effects of constituent quark mass to the gluodynamics encoded in the string interaction. Then one can decouple the left and right components of boundary fields in the process of dim-1 projection.
We assume the quark fields be located on the dim-1 boundary with coordinates $`x_\mu x_\mu (\tau )`$. The first step in projection of the E.o.M. (74) can be performed by their multiplication on $`\gamma ^\mu \dot{x}_\mu `$ which leads to the following boundary equations,
$$\left\{i\left(_\tau +\dot{x}_\mu v^\mu +g_A\gamma _5\dot{x}_\mu a^\mu \right)+\sigma ^{\mu \nu }\dot{x}_\mu \left(_\nu +v_\nu +g_A\gamma _5a_\nu \right)\right\}Q=0;\sigma ^{\mu \nu }\frac{1}{2}i[\gamma ^\mu \gamma ^\nu ].$$
(76)
Notice that this projected Dirac-type equation seems to be associated to the boundary action with a Lagrangian of type (3). But in order to provide the correct Hermitian properties of the Lorentz symmetry generators $`\sigma _{\mu \nu }`$ one must involve the Dirac conjugated spinors, $`\overline{\psi }\psi ^{}\gamma _0`$. As a consequence, the axial-vector part in the first, scalar contribution becomes anti-Hermitian as $`\gamma `$ matrices anticommute. It can be cured by the prescription of analytic continuation $`g_Az=ig_A`$ in the scalar part (only). At this place we must adopt an arbitrary constant $`z`$ subject to the consistency conditions from the string with boundary..
Let us restore the current quark basis of fields $`\psi _L`$ thereby going back to the original chiral fields $`U`$. We use Eqs.(72) and multiply the left and right component of Eq.(76) by $`\xi `$ and $`\xi ^{}`$ respectively. The result is that,
$`{\displaystyle \frac{1}{2}}\left\{i\left(\{_\tau ,U^{}\}+z\dot{U}^{}\right)+\sigma ^{\mu \nu }\dot{x}_\mu \left(\{_\nu ,U^{}\}+g_A_\nu U^{}\right)\right\}\psi _L=0;`$
$`{\displaystyle \frac{1}{2}}\left\{i\left(\{_\tau ,U\}+z\dot{U}\right)+\sigma ^{\mu \nu }\dot{x}_\mu \left(\{_\nu ,U\}+g_A_\nu U\right)\right\}\psi _R=0.`$ (77)
Now the culminating point of the ”ein-bein” projection consists of making the quark fields $`\psi `$ truly one-dimensional. Namely we define their gradient in terms of the tangent vector $`\dot{x}_\mu `$ and arbitrary matrix functions $`f(x_\mu ),b(x_\mu )`$ of $`x_\mu (\tau )`$,
$`_\mu (f_L\psi _L)+_\mu (b_L)\psi _L{\displaystyle \frac{\dot{x}_\mu }{\dot{x}_\nu \dot{x}^\nu }}\left[_\tau (f_L\psi _L)+_\tau (b_L)\psi _L\right];`$
$`_\mu (f_R\psi _R)+_\mu (b_R)\psi _R{\displaystyle \frac{\dot{x}_\mu }{\dot{x}_\nu \dot{x}^\nu }}\left[_\tau (f_R\psi _R)+_\tau (b_R)\psi _R\right].`$ (78)
Keeping in mind our program we choose the functions $`f(x_\mu ),b(x_\mu )`$ to provide the correct chiral properties, translational and reparameterization invariance (in terms of chiral fields $`U`$). As well the operator appeared in projection must be anti-self-adjoint in respect to the dim-4 Dirac scalar product. All these requirements are satisfied by the choice,
$$\{_\mu ,U^{}\}\psi _L\frac{\dot{x}_\mu }{\dot{x}_\nu \dot{x}^\nu }\{_\tau ,U^{}\}\psi _L;\{_\mu ,U\}\psi _R\frac{\dot{x}_\mu }{\dot{x}_\nu \dot{x}^\nu }\{_\tau ,U\}\psi _R.$$
(79)
Finally, the projected equations are originated from the boundary Lagrangian,
$`L^{(f)}`$ $`=`$ $`{\displaystyle \frac{1}{2}}i\left\{\overline{\psi }_L\left[\{_\tau ,U\}+z\dot{U}+g_\sigma \sigma ^{\mu \nu }\dot{x}_\mu _\nu U\right]\psi _R+\overline{\psi }_R\left[\{_\tau ,U^{}\}z^{}\dot{U}^{}g_\sigma ^{}\sigma ^{\mu \nu }\dot{x}_\mu _\nu U^{}\right]\psi _L\right\};`$
$``$ $`{\displaystyle \frac{1}{2}}i\left\{\overline{\psi }_L\left[\{_\tau ,U\}+\widehat{F}^{\mu \nu }\dot{x}_\mu _\nu U\right]\psi _R+\overline{\psi }_R\left[\{_\tau ,U^{}\}\widehat{F}_{\mathrm{}}^{\mu \nu }\dot{x}_\mu _\nu U^{}\right]\psi _L\right\};`$
$`\widehat{F}^{\mu \nu }`$ $``$ $`zg^{\mu \nu }+g_\sigma \sigma ^{\mu \nu };\widehat{F}_{\mathrm{}}^{\mu \nu }\gamma _0\left(\widehat{F}^{\mu \nu }\right)^{}\gamma _0,`$ (80)
where we have obtained the indications that $`g_\sigma =ig_A`$. Still keeping in mind a certain ambiguity in the projection procedure we must consider both constants $`z`$ and $`g_\sigma `$ as arbitrary ones and search for their values from the consistency of the hadron string with chiral fields on its boundary.
The meaning of purely imaginary $`z`$ and $`g_\sigma `$ is clarified by the CP symmetry (4) of the Lagrangian (80). Indeed it is CP symmetric only if
$$z=z^{};g_\sigma =g_\sigma ^{}.$$
(81)
## Appendix B. One-loop two-fermion one-boson vertex
In this Appendix we present the calculation of 1-boson vertex for the boundary Lagrangian (35) including a more general spin structure $`F_{\mu \nu }`$.
| Coefficient | U structure | F structure |
| --- | --- | --- |
| Divergent part | | |
| $`\frac{1}{4}\theta (AB)\mathrm{\Delta }(0)`$ | $`U^{}_\eta _\sigma _\lambda UU^{}`$ | $`\delta _{\sigma \lambda }[\overline{X}_\rho (A)(\delta _{\eta \rho }+F_{\eta \rho })+\overline{X}_\rho (B)(\delta _{\eta \rho }F_{\eta \rho })]`$ |
| $`\frac{1}{4}\theta (AB)\mathrm{\Delta }(0)`$ | $`U^{}_\eta _\sigma UU^{}_\lambda UU^{}`$ | $`[\overline{X}_\rho (A)(\delta _{\sigma \lambda }+F_{\eta \rho })(\delta _{\sigma \lambda }+F_{\lambda \sigma })+\overline{X}_\rho (B)(2\delta _{\eta \rho }\delta _{\sigma \lambda }`$ |
| | | $`F_{\eta \rho }\delta _{\sigma \lambda }\delta _{\eta \rho }F_{\lambda \sigma }+\delta _{\eta \rho }F_{\sigma \gamma }F_{\lambda \gamma }F_{\eta \rho }F_{\lambda \sigma })]`$ |
| $`\frac{1}{4}\theta (AB)\mathrm{\Delta }(0)`$ | $`U^{}_\sigma UU^{}_\lambda _\eta UU^{}`$ | $`[\overline{X}_\rho (B)(\delta _{\sigma \lambda }F_{\sigma \lambda })(\delta _{\sigma \lambda }F_{\eta \rho })+\overline{X}_\rho (A)(2\delta _{\eta \rho }\delta _{\sigma \lambda }`$ |
| | | $`F_{\eta \rho }\delta _{\sigma \lambda }\delta _{\eta \rho }F_{\sigma \lambda }+\delta _{\eta \rho }F_{\sigma \gamma }F_{\lambda \gamma }F_{\sigma \lambda }F_{\eta \rho })]`$ |
| $`\frac{1}{8}\theta (AB)\mathrm{\Delta }(0)`$ | $`U^{}_\sigma UU^{}_\eta UU^{}_\lambda UU^{}`$ | $`[\overline{X}_\rho (A)((3\delta _{\eta \rho }+F_{\eta \rho })(\delta _{\sigma \lambda }+F_{\lambda \sigma })`$ |
| | | $`+F_{\sigma \gamma }(\delta _{\eta \rho }F_{\eta \rho })(\delta _{\lambda \gamma }+F_{\lambda \gamma }))`$ |
| | | $`+\overline{X}_\rho (B)((\delta _{\sigma \lambda }F_{\sigma \lambda })(3\delta _{\eta \rho }F_{\eta \rho })`$ |
| | | $`(\delta _{\sigma \gamma }F_{\sigma \gamma })(\delta _{\eta \rho }+F_{\eta \rho })F_{\lambda \gamma })]`$ |
| Finite part $`\mathrm{\Delta }(AB)`$ | | |
| $`\frac{1}{4}\theta (AB)\mathrm{\Delta }(AB)`$ | $`U^{}_\eta _\sigma UU^{}_\lambda UU^{}`$ | $`\overline{X}_\rho (A)\delta _{\eta \rho }(\frac{1}{2}\delta _{\sigma \lambda }+F_{\sigma \lambda }F_{\lambda \sigma }F_{\sigma \gamma }F_{\lambda \gamma })`$ |
| $`\frac{1}{4}\theta (AB)\mathrm{\Delta }(AB)`$ | $`U^{}_\sigma UU^{}_\lambda _\eta UU^{}`$ | $`\overline{X}_\rho (B)\delta _{\eta \rho }(\frac{1}{2}\delta _{\sigma \lambda }F_{\sigma \lambda }+F_{\lambda \sigma }F_{\sigma \gamma }F_{\lambda \gamma })`$ |
| $`\frac{1}{8}\theta (AB)\mathrm{\Delta }(AB)`$ | $`U^{}_\sigma UU^{}_\eta UU^{}_\lambda UU^{}`$ | $`[\overline{X}_\rho (A)(\delta _{\sigma \gamma }+F_{\sigma \gamma })(\delta _{\eta \rho }+F_{\eta \rho })(\delta _{\lambda \gamma }F_{\lambda \gamma })`$ |
| | | $`+\overline{X}_\rho (B)(\delta _{\sigma \gamma }+F_{\sigma \gamma })(\delta _{\eta \rho }F_{\eta \rho })(\delta _{\lambda \gamma }F_{\lambda \gamma })]`$ |
| Finite part $`\underset{B}{\overset{A}{}}d\tau \dot{\overline{X}}_\rho (\tau )\mathrm{\Delta }(A\tau )\mathrm{Int}_A`$ | | |
| $`\frac{1}{4}\theta (AB)\mathrm{Int}_A`$ | $`U^{}_\sigma UU^{}_\lambda _\eta UU^{}`$ | $`(\delta _{\sigma \gamma }+F_{\sigma \gamma })(\delta _{\lambda \gamma }F_{\eta \rho }\delta _{\eta \rho }F_{\lambda \gamma })`$ |
| $`\frac{1}{8}\theta (AB)\mathrm{Int}_A`$ | $`U^{}_\sigma UU^{}_\eta UU^{}_\lambda UU^{}`$ | $`(\delta _{\sigma \gamma }+F_{\sigma \gamma })(\delta _{\eta \rho }F_{\eta \rho })(\delta _{\lambda \gamma }+F_{\lambda \gamma })`$ |
| Finite part $`\underset{B}{\overset{A}{}}d\tau \dot{\overline{X}}_\rho (\tau )\mathrm{\Delta }(\tau B)\mathrm{Int}_B`$ | | |
| $`\frac{1}{4}\theta (AB)\mathrm{Int}_B`$ | $`U^{}_\eta _\sigma UU^{}_\lambda UU^{}`$ | $`(\delta _{\sigma \gamma }F_{\eta \rho }F_{\sigma \gamma }\delta _{\eta \rho })(\delta _{\lambda \gamma }F_{\lambda \gamma })`$ |
| $`\frac{1}{8}\theta (AB)\mathrm{Int}_B`$ | $`U^{}_\sigma UU^{}_\eta UU^{}_\lambda UU^{}`$ | $`(\delta _{\sigma \gamma }F_{\sigma \gamma })(\delta _{\eta \rho }+F_{\eta \rho })(\delta _{\lambda \gamma }F_{\lambda \gamma })`$ | |
warning/0507/hep-th0507018.html | ar5iv | text | # 1 Introduction
## 1 Introduction
The equations of motion of four-dimensional $`N=1`$ supergravity can be obtained using the Euler-Lagrange equations applied to the Lagrangian <sup>3</sup><sup>3</sup>3 Here $`\mu ,\nu ,\mathrm{}=0,1,2,3`$ are world indices and an underline denotes the tangent frame. We use the $`(,+,+,+)`$ signature with $`\{\mathrm{\Gamma }_\mu ,\mathrm{\Gamma }_\nu \}=2g_{\mu \nu }`$, $`\mathrm{\Gamma }_5=\frac{1}{24}ϵ^{\mu \nu \lambda \rho }\mathrm{\Gamma }_{\mu \nu \lambda \rho }`$ and $`ϵ^{0123}=1`$.
$$_{onshell}=eR4i\overline{\psi }_\mu \mathrm{\Gamma }^{\mu \nu \rho }D_\nu \psi _\rho $$
(1)
While this Lagrangian is invariant, up to a total derivative, under the transformation
$$\delta _ϵe_\mu ^{\underset{¯}{\nu }}=2i\overline{ϵ}\mathrm{\Gamma }^{\underset{¯}{\nu }}\psi _\mu ,\delta _ϵ\psi _\mu =D_\mu ϵ$$
(2)
these transformations do not close to the supersymmetry algebra unless the fields are taken to be on-shell. Indeed off-shell there are only six Bosonic degrees of freedom whereas there are twelve Fermionic degrees of freedom. Thus it is of interest to construct off-shell extensions of the supergravity Lagrangian.
The original motivations for studying off-shell completions of supergravity were to ensure that supersymmetry remains a valid symmetry at the quantum level as well as to facilitate the proof of non-renormalization theorems. A natural starting point for supergravity is the geometrical analysis of four-dimensional $`N=1`$ superspace. Four-dimensional $`N=1`$ superfields carry reducible supersymmetry multiplets. Therefore additional constraints need to be imposed to truncate the superfields. These are then combined with the torsion and Bianchi identities to solve for the independent fields. It turns out that there are various ways to do this and hence there are several different off-shell formulations. The predominant view of these various off-shell completions is that they are all equivalent on-shell. The purpose of this paper is to show that while this is locally true, it is globally false.
In this paper we will explore some novel aspects of an off-shell formulation of four-dimensional $`N=1`$ supergravity known as “new minimal supergravity” (NMS) . We show that NMS admits Killing spinors on manifolds which are not supersymmetric in other formulations such as the “old-minimal” formulation . Moreover, when formulated on a manifold with boundary certain gauge modes of the auxiliary fields of NMS can become dynamical (depending on boundary conditions).
From a string theory perspective off-shell formulations are often viewed as unnecessary luxuries since one is simply viewing supergravity as a low energy effective theory which reproduces the correct on-shell physics. Our results call that point of view into question. A key motivation for this study was the desire to formulate 11-dimensional supergravity on Spin<sup>c</sup> manifolds (see the discussion section below). Indeed NMS contains an auxiliary gauge field which allows one to define the theory (off-shell) on Spin<sup>c</sup> manifolds. Unfortunately very little is known about off-shell completions of 11-dimensional supergravity. Indeed it is generally believed that an infinite number of auxiliary fields are required (although in some circumstances one could consider a finite collection of auxiliary fields which do not entirely close the algebra ). Our results raise the important question of whether different off-shell formulations of M-theory could be physically inequivalent.
## 2 Old and New Minimal Supergravity
The most familiar off-shell completion of $`N=1`$ four-dimensional supergravity is the so-called old-minimal formulation which includes two real scalar fields $`M`$ and $`N`$ along with a one-form $`b`$ . The action is simply
$$_{oldminimal}=_{onshell}\frac{1}{2}eM^2\frac{1}{2}eN^2+\frac{1}{2}bb$$
(3)
and there are twelve Bosonic and twelve Fermionic degrees of freedom off-shell. Clearly these new fields do not alter the theory in any non-trivial way. However one does find that the supersymmetry algebra closes off-shell (along with appropriate modifications to the supertransformation rules to include the auxiliary fields).
In Sohnius and West gave an off-shell formulation of four-dimensional $`N=1`$ supergravity which, in addition to the graviton and gravitini, includes an auxiliary 1-form $`A=A_\mu dx^\mu `$ and a 2-form $`B=\frac{1}{2}B_{\mu \nu }dx^\mu dx^\nu `$. The Lagrangian is
$$_{NMS}=eR4ie\overline{\psi }_\mu \mathrm{\Gamma }^{\mu \nu \rho }𝒟_\nu ^+\psi _\rho 6VV4AdB$$
(4)
where
$`V`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(dBi\overline{\psi }_\nu \mathrm{\Gamma }_\lambda \psi _\rho dx^\nu dx^\lambda dx^\rho \right)`$
$`𝒟_\mu ^+\psi _\nu `$ $`=`$ $`D_\mu \psi _\nu +A_\mu \mathrm{\Gamma }_5\psi _\nu `$
$`\omega _{\mu \underset{¯}{\lambda \rho }}`$ $`=`$ $`\omega _{\mu \underset{¯}{\lambda \rho }}^{\mathrm{Levi}\mathrm{Civita}}i(\overline{\psi }_{\underset{¯}{\lambda }}\mathrm{\Gamma }_\mu \psi _{\underset{¯}{\rho }}+\overline{\psi }_\mu \mathrm{\Gamma }_{\underset{¯}{\lambda }}\psi _{\underset{¯}{\rho }}\overline{\psi }_\mu \mathrm{\Gamma }_{\underset{¯}{\rho }}\psi _{\underset{¯}{\lambda }})`$ (5)
The equation of motion for $`A`$ sets
$$V=0$$
(6)
and then the equation of motion for $`B`$ determines that $`A`$ is a flat connection
$$dA=0$$
(7)
The particular choice $`A=0`$ then leads to the usual equations of motion for the graviton and gravitini.
In addition to diffeomorphisms the theory is invariant under the local supersymmetry transformation
$`\delta _ϵe_\mu ^{\underset{¯}{\nu }}`$ $`=`$ $`2i\overline{ϵ}\mathrm{\Gamma }^{\underset{¯}{\nu }}\psi _\mu `$
$`\delta _ϵ\psi _\mu `$ $`=`$ $`𝒟_\mu ^+ϵV_\mu \mathrm{\Gamma }_5ϵ+{\displaystyle \frac{1}{2}}\mathrm{\Gamma }_\mu ^\nu \mathrm{\Gamma }_5V_\nu ϵ`$
$`\delta _ϵB_{\mu \nu }`$ $`=`$ $`4i\overline{ϵ}\mathrm{\Gamma }_{[\mu }\psi _{\nu ]}`$
$`\delta _ϵA_\mu `$ $`=`$ $`2i\overline{ϵ}\mathrm{\Gamma }_5\mathrm{\Gamma }_\mu \mathrm{\Gamma }^{\nu \lambda }\left(𝒟_\nu ^+\psi _\lambda +3\mathrm{\Gamma }_5\psi _\nu V_\lambda +{\displaystyle \frac{3}{2}}\mathrm{\Gamma }_\nu ^\rho \psi _\lambda V_\rho \right)`$
However the crucial difference between old and new minimal supergravity is that NMS is also invariant under a local chiral rotation
$`\delta _\chi e_\mu ^{\underset{¯}{\nu }}`$ $`=`$ $`0`$
$`\delta _\chi \psi _\mu `$ $`=`$ $`\chi \mathrm{\Gamma }_5\psi _\mu `$
$`\delta _\chi B`$ $`=`$ $`0`$
$`\delta _\chi A`$ $`=`$ $`d\chi .`$
which is broken in old minimal by the supersymmetry transformation rules. The Lagrangian also has a trivial gauge transformation which only acts on $`B`$
$$\delta _\lambda B=d\lambda .$$
(10)
It is easy to check that all these symmetries commute with each other
$$[\delta _ϵ,\delta _\chi ]=[\delta _ϵ,\delta _\lambda ]=[\delta _\chi ,\delta _\lambda ]=0$$
(11)
provided that the supersymmetry generator is also taken to transform under local chiral rotations.
Under supersymmetry the action is not invariant but rather transforms into a boundary term. To quadratic order in the Fermions one finds
$`\delta _ϵS`$ $`=`$ $`{\displaystyle _{}}\delta _ϵ_{NMS}`$
$`=`$ $`2i{\displaystyle _{}}\sqrt{h}(h^{\rho \nu }n^\mu +h^{\rho \mu }n^\nu 2h^{\mu \nu }n^\rho )D_\rho (\overline{ϵ}\mathrm{\Gamma }_\mu \psi _\nu )`$
$`+4i{\displaystyle _{}}\overline{ϵ}\left(\mathrm{\Gamma }_5\mathrm{\Gamma }_\lambda 𝒟_\nu ^{}\psi _\rho +\mathrm{\Gamma }_\lambda V_\nu \psi _\rho +{\displaystyle \frac{1}{2}}\mathrm{\Gamma }_\nu ^\tau \mathrm{\Gamma }_\lambda V_\tau \psi _\rho \right)𝑑x^\lambda dx^\nu dx^\rho `$
where $`n^\mu `$ is the unit inward pointing normal vector to the boundary, $`h_{\mu \nu }=g_{\mu \nu }n_\mu n_\nu `$ is the induced metric and
$$𝒟_\nu ^{}\psi _\rho =D_\nu \psi _\rho \mathrm{\Gamma }_5A_\nu \psi _\rho $$
(13)
is the anti-chiral covariant derivative, i.e. it corresponds to gauging chiral rotations with the opposite choice of $`\mathrm{\Gamma }_5`$. The fact that $`\delta _ϵ`$ is no longer chirally invariant seems odd but can be verified by noting that, due to the final Chern-Simons term, the Lagranian is not exactly chirally invariant either
$$\delta _\chi _{NMS}=4d\chi dB=4d(\chi dB)$$
(14)
The failure of chiral symmetry in $`\delta _ϵ`$ is then necessary to account for the variation of $`\delta _\chi `$ under supersymmetry since it must be true that
$$\delta _\chi \delta _ϵ_{NMS}=\delta _ϵ\delta _\chi _{NMS}$$
(15)
which one can readily check is indeed the case with these boundary terms.
However we can correct for this by adding the total derivative term
$$_{bdry}=4d\left(AB\right)$$
(16)
to the Lagrangian. This has the effect of changing the Lagrangian to
$$_{NMS}^{}=eR4ie\overline{\psi }_\mu \mathrm{\Gamma }^{\mu \nu \rho }𝒟_\nu ^+\psi _\rho 6VV4dAB$$
(17)
Clearly $`_{NMS}^{}`$ is invariant under chiral rotations, with no boundary terms. This implies that the boundary terms must be invariant under chiral rotations. The variation of $`_{bdry}`$ under supersymmetry is
$`\delta _ϵ_{bdry}`$ $`=`$ $`4d\left(\delta _ϵAB2i\overline{ϵ}A_\lambda \mathrm{\Gamma }_\nu \psi _\rho dx^\lambda dx^\nu dx^\rho \right)`$
The second term converts the $`𝒟_\mu ^{}`$ into a $`𝒟_\mu ^+`$ derivative. Thus the variation of the improved action under supersymmetry is
$`\delta _ϵS^{}`$ $`=`$ $`{\displaystyle _{}}\delta _ϵ_{NMS}+{\displaystyle _{}}\delta _ϵ_{bdry}`$
$`=`$ $`2i{\displaystyle _{}}\sqrt{h}(h^{\rho \nu }n^\mu +h^{\rho \mu }n^\nu 2h^{\mu \nu }n^\rho )D_\rho (\overline{ϵ}\mathrm{\Gamma }_\mu \psi _\nu )`$
$`+4i{\displaystyle _{}}\overline{ϵ}\left(\mathrm{\Gamma }_5\mathrm{\Gamma }_\lambda 𝒟_\nu ^+\psi _\rho +\mathrm{\Gamma }_\lambda V_\nu \psi _\rho +{\displaystyle \frac{1}{2}}\mathrm{\Gamma }_\nu ^\tau \mathrm{\Gamma }_\lambda V_\tau \psi _\rho \right)𝑑x^\nu dx^\rho dx^\lambda `$
$`+4{\displaystyle _{}}\delta _ϵAB`$
which is indeed invariant under chiral rotations.
### 2.1 Supersymmetry in the presence of boundaries
It is well known that if supergravity is placed on a manifold with a boundary then at least half of the supersymmetries will be broken. If $`0`$ then the action obtained from $`^{}`$ is not invariant under supersymmetry. However by adding suitable boundary terms this can be corrected. These boundary terms are used to set-up a well-posed boundary value problem and also to preserve half of the supersymmetries. In the case of eleven-dimensional supergravity this has been done in and we wish to follow a similar analysis for NMS, although we will restrict our attention to quadratic terms in the Fermions.
The first step to including boundaries is to add the Gibbons-Hawking term to make the pure gravitational variational problem well posed
$$S_{GH}=2_{}\sqrt{h}K$$
(20)
where $`K_{\mu \nu }=h_{\mu \lambda }h_{\nu \rho }D^\lambda n^\rho `$ is the extrinsic curvature. With this term in place one finds that the variation of the standard Einstein-Hilbert plus Gibbons-Hawking term results in the boundary term
$$\delta _g(S_{EH}+S_{GH})=_{}\sqrt{h}(K_{\mu \nu }g_{\mu \nu }K)\delta g^{\mu \nu }$$
(21)
which is required to cancel with any additional stress-energy tensor that is localized to the boundary (which in our case vanishes).
Following we need to add a Fermionic boundary term. Let us first recall the case of the familiar on-shell supergravity. Varying the Fermionic term gives the equations of motion plus the boundary term
$$\delta _\psi S=4i_{}\sqrt{h}\overline{\psi }_\mu ^{}\mathrm{\Gamma }^{\mu ^{}\nu ^{}}\mathrm{\Gamma }^n\delta \psi _\nu ^{}$$
(22)
where $`\mathrm{\Gamma }^n=n^\mu \mathrm{\Gamma }_\mu `$ and $`\mu ^{},\nu ^{}`$ are the coordinates tangential to the boundary. To cancel this one adds the term
$$S_{LM}=2i\xi _{}\sqrt{h}\overline{\psi }_\mu ^{}\mathrm{\Gamma }^{\mu ^{}\nu ^{}}\psi _\nu ^{}$$
(23)
where $`\xi =\pm 1`$. Variation of this term gives
$$\delta _\psi S_{LM}=4i\xi _{}\sqrt{h}\overline{\psi }_\mu ^{}\mathrm{\Gamma }^{\mu ^{}\nu ^{}}\delta \psi _\nu ^{}$$
(24)
Thus a suitable boundary condition is $`\psi _\mu ^{}=\xi \mathrm{\Gamma }^n\psi _\mu ^{}`$.
In NMS we again encounter the boundary term (22) however we cannot simply add (23) since this term is not invariant under chiral rotations. More properly it doesn’t even make sense as $`\psi _\mu `$ is a section of a chiral spinor bundle whereas $`C\mathrm{\Gamma }^{\mu ^{}\nu ^{}}`$ is a map between the chiral and anti-chiral spinor bundles.
If we choose a boundary condition where $`A=d\mathrm{\Phi }_A`$ on $``$ then we can add the boundary term
$$S_\psi =2i\xi _{}\sqrt{h}\overline{\psi }_\mu ^{}\mathrm{\Gamma }^{\mu ^{}\nu ^{}}e^{2\mathrm{\Phi }_A\mathrm{\Gamma }_5}\psi _\nu ^{}$$
(25)
i.e. we can map NMS to on-shell supergravity by a chiral gauge rotation. (Note that this boundary condition implies that the gauge field $`A`$ is trivial on the boundary and this is not the case in general.) We now find the Fermionic boundary condition
$$\mathrm{\Gamma }^ne^{\mathrm{\Phi }_A\mathrm{\Gamma }_5}\psi _\nu ^{}=\xi e^{\mathrm{\Phi }_A\mathrm{\Gamma }_5}\psi _\nu ^{}$$
(26)
Our next task is to show that
$$S=S_{GH}+S_\psi +_{}^{}$$
(27)
is indeed supersymmetric with a well posed variational problem when we impose the boundary conditions
$$\mathrm{\Gamma }^ne^{2\mathrm{\Phi }_A\mathrm{\Gamma }_5}\psi _\nu ^{}=\xi \psi _\nu ^{},A=d\mathrm{\Phi }_A,V=0$$
(28)
on $``$. We derived the boundary terms by ensuring a well-posed boundary value problem for the metric and $`\psi _\mu `$. Thus it remains to check that variations of the form $`\delta A=d\delta \mathrm{\Phi }_A`$ on the boundary are well posed, i.e. that they do not over constrain the system. There are two sources for these variations, boundary terms from the bulk $`d\delta AB`$ term and also terms that arise directly from varying $`S_\psi `$. Putting these together we find
$`\delta _{\mathrm{\Phi }_A}S`$ $`=`$ $`4{\displaystyle _{}}𝑑\delta \mathrm{\Phi }_AB+4i\xi {\displaystyle _{}}\sqrt{h}\delta \mathrm{\Phi }_A\overline{\psi }_\mu ^{}\mathrm{\Gamma }^{\mu ^{}\nu ^{}}\mathrm{\Gamma }_5e^{2\mathrm{\Phi }_A\mathrm{\Gamma }_5}\psi _\nu ^{}`$
$`=`$ $`4{\displaystyle _{}}\delta \mathrm{\Phi }_AdBi\sqrt{h}\delta \mathrm{\Phi }_A\overline{\psi }_\mu ^{}\mathrm{\Gamma }^{\mu ^{}\nu ^{}}\mathrm{\Gamma }_5\mathrm{\Gamma }^n\psi _\nu ^{}`$
$`=`$ $`4{\displaystyle _{}}\delta \mathrm{\Phi }_AV`$
$`=`$ $`0`$
Next we examine the variation of the action under supersymmetry. We already know that the bulk is supersymmetric. We would like to show that the variation of the additional boundary terms $`S_{GH}+S_\psi `$ cancels (LABEL:newsuperJ). Of course the boundary condition on the Fermions breaks half of the supersymmetries, leaving only those with $`\mathrm{\Gamma }^ne^{\mathrm{\Phi }_A\mathrm{\Gamma }_5}ϵ=\xi e^{\mathrm{\Phi }_A\mathrm{\Gamma }_5}ϵ`$. Therefore we compute
$`\delta _ϵ(S_{EH}+S_{GH})`$ $`=`$ $`4i{\displaystyle _{}}\sqrt{h}\overline{ϵ}(K^{\mu ^{}\nu ^{}}h^{\mu ^{}\nu ^{}}K)\mathrm{\Gamma }_\mu ^{}\psi _\nu ^{}`$
and (using the boundary conditions)
$`\delta _ϵS_\psi `$ $`=`$ $`4i\xi {\displaystyle _{}}\sqrt{h}\left(𝒟_\mu ^{}^+\overline{ϵ}\mathrm{\Gamma }^{\mu ^{}\nu ^{}}e^{2\mathrm{\Phi }_A\mathrm{\Gamma }_5}\psi _\nu ^{}+\delta _ϵ\mathrm{\Phi }_A\overline{\psi }_\mu ^{}\mathrm{\Gamma }^{\mu ^{}\nu ^{}}e^{2\mathrm{\Phi }_A\mathrm{\Gamma }_5}\mathrm{\Gamma }_5\psi _\nu ^{}\right)`$
$`=`$ $`4i{\displaystyle _{}}\sqrt{h}\left(𝒟_\mu ^{}^+\overline{ϵ}\mathrm{\Gamma }^{\mu ^{}\nu ^{}}\mathrm{\Gamma }^n\psi _\nu ^{}+\delta _ϵ\mathrm{\Phi }_A\overline{\psi }_\mu ^{}\mathrm{\Gamma }^{\mu ^{}\nu ^{}}\mathrm{\Gamma }_5\mathrm{\Gamma }^n\psi _\nu ^{}\right)`$
$`=`$ $`4i{\displaystyle _{}}\sqrt{h}\left(\overline{ϵ}\mathrm{\Gamma }^{\mu ^{}\nu ^{}}\mathrm{\Gamma }^n𝒟_\mu ^{}^+\psi _\nu ^{}+\overline{ϵ}\mathrm{\Gamma }^{\mu ^{}\nu ^{}}D_\mu ^{}n_\lambda \mathrm{\Gamma }^\lambda \psi _\nu ^{}\right)`$
$`+4{\displaystyle _{}}\delta _ϵ\mathrm{\Phi }_AdB`$
$`=`$ $`4i{\displaystyle _{}}\sqrt{h}\left(\overline{ϵ}\mathrm{\Gamma }^{\mu ^{}\nu ^{}}\mathrm{\Gamma }^nD_\mu ^{}^+\psi _\nu ^{}+\overline{ϵ}(K^{\mu ^{}\nu ^{}}Kh^{\mu ^{}\nu ^{}})\mathrm{\Gamma }_\mu ^{}\psi _\nu ^{}\right)`$
$`4{\displaystyle _{}}\delta _ϵAB`$
Here we have used the identity
$$\mathrm{\Gamma }^{\mu ^{}\nu ^{}}D_\mu ^{}n_\lambda \mathrm{\Gamma }^\lambda =K_{\mu ^{}\lambda ^{}}\mathrm{\Gamma }^{\mu ^{}\nu ^{}}\mathrm{\Gamma }^\lambda ^{}=(K^{\mu ^{}\nu ^{}}Kh^{\mu ^{}\nu ^{}})\mathrm{\Gamma }_\mu ^{}$$
(32)
One can now see that these terms precisely cancel the terms in (LABEL:newsuperJ).
Note that we never had to deduce what $`\delta _ϵ\mathrm{\Phi }_A`$ was from $`\delta _ϵA`$. However we have assumed that $`A=d\mathrm{\Phi }_A`$ and $`V=0`$ on the boundary and these conditions impose constraints on the Fermions on the boundary (which are certainly satisfied if the Fermions are on-shell).
Returning to the case of a general $`A`$ we see that if it is non-trivial on $``$ then there is no boundary term that we can write down that will cancel the Fermion boundary variation. Thus in these cases supersymmetry is broken by the presence of a boundary.
## 3 Higher-Dimensional Supergravity and an On-Shell Variant of New-Minimal
Ultimately we are interested in extending our analysis to the ten and eleven-dimensional supergravities associated with string theory and M-theory. As mentioned above, no off-shell completions are known for these theories and it is generally believed that if any such formulations exist then they must have an infinite number of auxiliary fields.
However one can see that the type of physics explored here can be extended in part to other supergravities. Suppose that there is a supergravity Lagrangian $`_{sugra}`$ which is also invariant under a global symmetry (up to a boundary term). Then we can make this symmetry local in the usual manner by replacing covariant derivatives with gauge-covariant derivatives
$$D_\mu 𝒟_\mu =D_\mu +𝒜_\mu $$
(33)
where $`𝒜_\mu `$ is the appropriate gauge connection. In this way we obtain the new Lagrangian
$$_𝒜=_{sugra}(D_\mu 𝒟_\mu )$$
(34)
Next we must arrange for $`_𝒜`$ to be supersymmetric. One sees that if we choose
$$\delta _ϵ𝒜_\mu =0$$
(35)
then the variation of the action, ignoring boundary terms, must be of the form
$$\delta _ϵ_𝒜=\mathrm{Tr}\left(\mathrm{\Omega }\right)$$
(36)
where $``$ is the gauge-invariant field strength of $`𝒜`$. To cancel such a term we need only invent a new form field $``$ with
$$\delta _ϵ=\mathrm{\Omega }$$
(37)
so that
$$_𝒜^{}=_𝒜\mathrm{Tr}\left(\right)$$
(38)
is supersymmetric, up to possible boundary terms.
In this way we have arrived a form of supergravity that is similar to NMS. Of course the supersymmetry algebra is not closed off-shell, indeed we have merely added a supersymmetry singlet $`𝒜`$, along with a non-singlet field $``$. Thus these modifications may lead to problems in the quantum theory.
For example we could consider $`N=1`$ supergravity in four dimensions and gauge the chiral $`U(1)`$ symmetry that new-minimal exploits. Following the above procedure we arrive at the Lagrangian
$$\stackrel{~}{}=eR4ie\overline{\psi }_\mu \mathrm{\Gamma }^{\mu \nu \rho }𝒟_\nu ^+\psi _\rho 4FB$$
(39)
which is invariant under
$`\delta _ϵe_\mu ^{\underset{¯}{\nu }}`$ $`=`$ $`2i\overline{ϵ}\mathrm{\Gamma }^{\underset{¯}{\nu }}\psi _\mu `$
$`\delta _ϵ\psi _\mu `$ $`=`$ $`𝒟_\mu ^+ϵ`$
$`\delta _ϵB_{\mu \nu }`$ $`=`$ $`4i\overline{ϵ}\mathrm{\Gamma }_{[\mu }\psi _{\nu ]}`$
$`\delta _ϵA_\mu `$ $`=`$ $`0`$
up to boundary terms. This is similar to, but not identical to, new-minimal with $`V=0`$. In addition it is easy to see that this theory can be made supersymmetric on a manifold with boundary using the same boundary terms and conditions that we used for new-minimal.
## 4 Supersymmetric Compactifications
We can classify all supersymmetric compactifications of NMS supergravity (and also its variant (39)) of the form $`𝐑^{4d}\times _d`$ with $`d=1,2,3`$ and $`_d`$ compact without boundary. Our first condition on the manifold $`_d`$, apart from compactness, is that it admits some kind of spinor and hence must be orientable (we will not consider the possibility of pinors here). The existence of a Killing spinor $`ϵ`$ such that $`𝒟_\mu ^+ϵ=0`$ also implies that
$$0=[𝒟_\mu ^+,𝒟_\nu ^+]ϵ=\frac{1}{4}R_{\mu \nu \lambda \rho }\mathrm{\Gamma }^{\lambda \rho }ϵ+F_{\mu \nu }\mathrm{\Gamma }_5ϵ$$
(41)
Since $`F=dA=0`$ on-shell this implies that $`_d`$ is Ricci-flat and hence also Riemann flat since $`d=1,2,3`$.
For $`d=1,2`$ the only possible internal manifolds are tori. These can clearly be made supersymmetric. On the other hand we could also turn on a non-trivial $`A=\frac{\alpha }{R_3}dx^3`$ where $`x^3`$ is taken to be periodic with period $`2\pi R_3`$. The Killing spinors take the form
$$ϵ=e^{\alpha \frac{x^3}{R_3}\mathrm{\Gamma }_5}ϵ_0$$
(42)
with $`ϵ_0`$ a constant spinor. Only if $`\alpha 𝐙`$ do we find a single valued spinor. In this case the supercurrent will not be single valued so that the variation of the Lagrangian is not exact and hence the action not invariant. Thus a generic $`A`$ will apparently break all the supersymmetries. However we can undo the damage if we simply change the boundary conditions of the gravitino to $`\psi _\mu (x^3+2\pi R_3)=e^{2\pi \alpha \mathrm{\Gamma }_5}\psi _\mu (x^3)`$. In this case the supercurrent will be single valued and hence the variation of the Lagrangian is an exact form and the action invariant.
### 4.1 Compactification to one-dimension
The case of $`d=3`$ is more interesting. There are in fact six compact orientable Riemann flat three-manifolds called Bierberbach manifolds (for example see ). They are all obtained as quotients of $`𝐑^3`$ by some freely acting group $`G`$ and can be identified by their holonomies
$$(_3)=\mathrm{𝟏},𝐙_2,𝐙_3,𝐙_4,𝐙_6,𝐙_2\times 𝐙_2$$
(43)
The first case is of course that of the torus $`𝐓^3=𝐑^\mathrm{𝟑}/G`$ with $`G`$ generated by the three elements
$`\alpha _1:(x^1,x^2,x^3)`$ $``$ $`(x^1+2\pi R_1,x^2,x^3)`$
$`\alpha _2:(x^1,x^2,x^3)`$ $``$ $`(x^1,x^2+2\pi R_2,x^3)`$
$`\alpha _3:(x^1,x^2,x^3)`$ $``$ $`(x^1,x^2,x^3+2\pi R_3)`$
The first Bieberbach manifold with nontrivial holonomy is a quotient of $`𝐑^3`$ generated by $`\alpha ^i`$ along with the element $`\beta `$;
$$\beta :(x^1,x^2,x^3)(x^1,x^2,x^3+\pi R_3).$$
(45)
so that $`\beta ^2=\alpha _3`$ and $`\beta \alpha _i\beta ^1=\alpha _i^1`$ if $`i3`$. This leads to a space with holonomy $`𝐙_2`$. (We have not written the most general such manifold: the lattice in the $`12`$ plane can be arbitrary and need not be rectangular.)
The $`𝐙_4`$ example is similar to the $`𝐙_2`$ case only now we take $`R_1=R_2`$ and $`\beta `$ acts as
$$\beta :(x^1,x^2,x^3)(x^2,x^1,x^3+\frac{\pi }{2}R_3).$$
(46)
and hence we have $`\beta ^4=\alpha _3`$, $`\beta \alpha _1\beta ^1=\alpha _2`$ and $`\beta \alpha _2\beta ^1=\alpha _1^1`$.
Next we consider the $`𝐙_3`$ case. Here one must fix $`R_1=R_2`$ and start with a hexagonal lattice, so that the $`\alpha _1`$ generator is modified to
$$\alpha _1:(x^1,x^2,x^3)(x^1+\sqrt{3}\pi R_1,x^2+\pi R_1,x^3)$$
(47)
while $`\alpha _2,\alpha _3`$ are unchanged. The generator $`\beta `$ is now
$$\beta :(x^1,x^2,x^3)(\frac{1}{2}x^1+\frac{\sqrt{3}}{2}x^2,\frac{\sqrt{3}}{2}x^1\frac{1}{2}x^2,x^3+\frac{2\pi }{3}R_3)$$
(48)
which satisfies $`\beta ^3=\alpha _3`$, $`\beta \alpha _1\beta ^1=\alpha _2^1`$ and $`\beta \alpha _2\beta ^1=\alpha _1\alpha _2^1`$.
Next we consider the $`𝐙_6`$ case. Again we must fix $`R_1=R_2`$ and start with a hexagonal lattice, so that the $`\alpha _1`$ generator is modified to
$$\alpha _1:(x^1,x^2,x^3)(x^1+\sqrt{3}\pi R_1,x^2\pi R_1,x^3)$$
(49)
The generator $`\beta `$ is now
$$\beta :(x^1,x^2,x^3)(\frac{1}{2}x^1+\frac{\sqrt{3}}{2}x^2,\frac{\sqrt{3}}{2}x^1+\frac{1}{2}x^2,x^3+\frac{\pi }{3}R_3).$$
(50)
and satisfies $`\beta ^6=\alpha _3`$, $`\beta \alpha _1\beta ^1=\alpha _2^1`$ and $`\beta \alpha _2\beta ^1=\alpha _1\alpha _2`$.
The final case has holonomy $`𝐙_2\times 𝐙_2`$. In addition to the $`\alpha _i`$ (defined to generate a rectangular lattice) we introduce three additional generators
$`\beta _1:(x^1,x^2,x^3)`$ $``$ $`(x^1+\pi R_1,x^2+\pi R_2,x^3)`$
$`\beta _2:(x^1,x^2,x^3)`$ $``$ $`(x^1+\pi R_1,x^2+\pi R_2,x^3+\pi R_3)`$
$`\beta _3:(x^1,x^2,x^3)`$ $``$ $`(x^1,x^2,x^3+\pi R_3)`$
which satisfy $`\beta _i^2=\alpha _i`$, $`\beta _i\alpha _j\beta _i^1=\alpha _j^1`$ if $`ij`$ and $`\beta _1\beta _2\beta _3=\alpha _1`$.
It has been shown in that there are no Killing spinors on a Bieberbach manifold with nontrivial holonomy. To see this one notes that a Killing spinor on a Bieberbach manifold will lift to a Killing spinor on the covering space $`𝐑^3`$. However it must lift to a Killing spinor which is invariant under the group $`G`$.
In order to proceed we need to define a lift of the group $`G`$ to a group $`\stackrel{~}{G}Spin(3)`$ acting on the spinor bundle of $`𝐑^3`$, i.e. for each generator $`g`$ of $`G`$ we must find an element $`\stackrel{~}{g}\stackrel{~}{G}`$ such that $`\pi (\stackrel{~}{g})=g`$ and which preserves the relations of the group $`G`$. Here $`\pi :Spin(3)SO(3)`$ is the usual 2-1 map. As detailed in for each group $`G`$ there will generically be several choices for $`\stackrel{~}{G}`$ and these correspond to different spin structures on the Bieberbach manifold.
Next we must ask that the Killing spinor is invariant. This leads to a condition
$$\stackrel{~}{g}ϵ(gx)=ϵ(x)$$
(52)
Since there is a unique spin bundle on the covering space we may choose a frame on $`𝐑^3`$ so that the Killing spinors are just constant spinors $`ϵ=ϵ_0`$. The condition (52) is simply that $`\stackrel{~}{g}ϵ_0=ϵ_0`$. We now note that all of the non-trivial Bieberbach manifolds contain a generator $`\beta `$ which includes a rotation in some plane by an amount different from $`2\pi `$. The lift of such a generator is an element $`\stackrel{~}{\beta }Spin(3)`$ such that $`\stackrel{~}{\beta }1`$. Hence it is impossible to find a constant spinor $`ϵ_0`$ such that $`\stackrel{~}{\beta }ϵ_0=ϵ_0`$.
However if we turn on the flat gauge connection
$$A=\frac{1}{2R_3}dx^3$$
(53)
then we can construct invariant spinors. To see this note that the Killing spinors on $`𝐑^3`$, i.e. spinors which satisfy $`𝒟_\mu ^+ϵ=0`$, are now
$$ϵ=e^{\frac{x^3}{2R_3}\mathrm{\Gamma }_5}ϵ_0$$
(54)
where $`ϵ_0`$ is a constant spinor. The invariance condition (52) is now
$$e^{\frac{x^3}{2R_3}\mathrm{\Gamma }_5}\stackrel{~}{g}e^{\frac{(gx)^3}{2R_3}\mathrm{\Gamma }_5}ϵ_0=ϵ_0$$
(55)
For the first four non-trivial Bieberbach manifolds the only non-trivial generator is $`\beta `$ which acts as
$$x^3x^3+\theta R^3,\left(\begin{array}{c}x^1\\ x^2\end{array}\right)\left(\begin{array}{cc}\mathrm{cos}\theta & \mathrm{sin}\theta \\ \mathrm{sin}\theta & \mathrm{cos}\theta \end{array}\right)\left(\begin{array}{c}x^1\\ x^2\end{array}\right)$$
(56)
where $`\theta =\pi ,2\pi /3,\pi /4,\pi /3`$ for the holonomies $`=𝐙_2,𝐙_3,𝐙_4,𝐙_6`$ respectively. The corresponding lift to $`Spin(3)`$ of $`\beta `$ is
$$\stackrel{~}{\beta }=\pm e^{\frac{\theta }{2}\mathrm{\Gamma }_{12}}$$
(57)
where the choice of sign reflects a choice of spin structure on $`_3`$. The invariance condition is now simply
$$\pm e^{\frac{\theta }{2}\mathrm{\Gamma }_{12}}e^{\frac{\theta }{2}\mathrm{\Gamma }_5}ϵ_0=ϵ_0$$
(58)
This can be solved by choosing the spin structure corresponding to the plus sign and projecting onto constant spinors that satisfy
$$\mathrm{\Gamma }_{03}ϵ_0=ϵ_0$$
(59)
For the $`𝐙_2`$ case one can also find a Killing spinor with the spin structure corresponding to the the minus sign by taking $`\mathrm{\Gamma }_{03}ϵ_0=ϵ_0`$.
A key property of the first four non-trivial Bieberbach manifolds that enables these Killing spinors to exist is that the lift of $`G`$ to $`\stackrel{~}{G}`$ is $`\stackrel{~}{G}=U(1)Spin(3)`$. This allows the holonomy of the spinor induced by each generator to be canceled by the phase shift induced by the $`U(1)`$ gauge connection $`A`$. For the final Bieberbach manifold, with holonomy $`𝐙_2\times 𝐙_2`$, $`\stackrel{~}{G}`$ is not contained in a $`U(1)`$ subgroup of $`Spin(3)`$ and hence the holonomies cannot be canceled. Thus there are no Killing spinors.
How did this work? In ordinary supergravity the gravitinos are sections of $`T^{}()S()`$, where $`S()`$ is a spinor bundle and $`T^{}()`$ is the cotangent bundle. For the manifolds constructed above there are no Killing spinors, i.e. covariantly constant sections of $`S()`$. In the NMS the gravitinos are sections of $`T^{}()S()L()`$ where $`L()`$ is an additional flat line bundle. The point is that there are covariantly constant sections of $`S()L()`$.
Note that one might try to make the Beiberbach manifolds supersymmetric in old minimal supergravity by changing the boundary conditions to $`\psi _\mu (x^3+\theta R_3)=\mathrm{\Gamma }_5\psi _\mu (x^3)`$ <sup>4</sup><sup>4</sup>4We thank J. Maldacena for discussion on this point.. Such a boundary condition is not compatible with the possible spin structures of spacetime but in principle this could be rectified by taking the Fermions to be sections of a line bundle associated to chiral rotations, as is the case in NMS, although without including a connection. However this approach is problematic as the supersymmetry variations of the auxiliary fields in old minimal supergravity are not chirally covariant.
This is reminiscent of the situation with spin<sup>c</sup> structures. In these cases there are manifolds, for example $`\mathrm{𝐂𝐏}^2`$, which don’t admit any spinors at all, let alone covariantly constant ones. However they do admit sections of the spin bundle tensored with a complex line bundle; $`S()L()`$ (e.g. see ). Indeed this situation can arise in string theory and M-theory . Typically the complex line bundle is not flat and so cannot be a solution of NMS, at least without coupling to additional fields. However in NMS it is possible to include Spin<sup>c</sup> manifolds in the off-shell formulation of the theory by taking the gauge field $`F=dA`$ to be non-vanishing and (cohomologically) non-trivial. In this sense $`\mathrm{𝐂𝐏}^2`$ is no more problematic than $`S^4`$, i.e. the theory is defined for such manifolds but they do not satisfy the equations of motion.
## 5 Cylindrical Spacetimes and Singletons
Gauge degrees of freedom are typically thought of as unphysical. However this is not necessarily the case if spacetime has a boundary. Quite generally, putting a gauge theory on a spacetime with a spatial boundary can lead to physical gauge modes that live on the boundary. This happens, for example, in three-dimensional Chern-Simons gauge theory and in the theory of the fractional quantum Hall effect, where the boundary degrees of freedom are known as “edge states.” In the context of supergravity one can find some discussion of these “singleton modes” in . Since NMS (and the variant (39)) has additional, but auxiliary, gauge degrees of freedom as compared with old minimal supergravity it is possible that one could in principle distinguish between them by considering spacetimes with a boundary. In some cases we may therefore hope to see residual gauge degrees of freedom propagating on the boundary. A particular class of spacetimes with a boundary are the so-called cylindrical spacetimes $`=^{}\times 𝐑`$, where $`𝐑`$ is the time dimension and $`^{}=\mathrm{\Sigma }`$. Thus the boundary of spacetime is $`\mathrm{\Sigma }\times 𝐑`$. Note that $`\mathrm{\Sigma }`$ could have several disconnected pieces.
We wish to show that, in NMS, there is a consistent choice of boundary conditions so that the theory on a boundary contains additional physical modes that propagate on the boundary due to the auxiliary fields. Hence one can, in principle, physically distinguish between different off-shell formulations of supergravity and, for example, determine the existence or non-existence of a given set of auxiliary fields.
### 5.1 An example
First it is helpful to review the discussion of appendix A in . Consider a Bosonic action
$$S=_{}𝑑AB$$
(60)
where $`=^{}\times 𝐑`$ with $`^{}=\mathrm{\Sigma }`$.
To exhibit the singleton modes on the boundary we must carefully consider the boundary conditions to ensure a well posed variational problem. Expanding in terms of the temporal and spatial components of $`A`$ and $`B`$ this action is
$$S=𝑑t_{^{}}d^{}A^{}B_0+A_0d^{}B^{}+\dot{A}^{}B^{}𝑑t_\mathrm{\Sigma }A_0B^{}$$
(61)
We can proceed in two ways. We could invoke the boundary condition $`A_0=0`$ on $`\mathrm{\Sigma }`$. Alternatively we could simply add an additional boundary term to the theory
$$S_{}=𝑑t_\mathrm{\Sigma }A_0B^{}$$
(62)
to cancel the existing boundary term in (61). No boundary condition is now required on $`A_0`$ or $`B_0`$. Presumably these two appoaches are equivalent and in either case the gauge symmetry is broken on the boundary.
Continuing we can integrate out $`A_0`$ and $`B_0`$ since they are non-dynamical to find
$$d^{}A^{}=0d^{}B^{}=0$$
(63)
which we solve by
$$A^{}=d^{}\mathrm{\Phi }_AB^{}=d^{}\mathrm{\Phi }_B$$
(64)
where $`\mathrm{\Phi }_A`$ and $`\mathrm{\Phi }_B`$ are arbitrary. Substituting this back into the action leads to
$`S`$ $`=`$ $`{\displaystyle 𝑑t_{^{}}d^{}\dot{\mathrm{\Phi }}_A}d^{}\mathrm{\Phi }_B`$
$`=`$ $`{\displaystyle 𝑑t_\mathrm{\Sigma }\dot{\mathrm{\Phi }}_A}d^{}\mathrm{\Phi }_B`$
Here we see the propagating singleton modes on the boundary.
One can think of these singleton modes as arising from pure gauge modes which violate the boundary condition $`A_0=0`$. To illustrate this point we note that in order to obtain a well-defined boundary value problem we can also choose the boundary condition
$$A=d\mathrm{\Phi }_A,dB=0$$
(66)
with $`\mathrm{\Phi }_A`$ arbitrary, i.e. $`A`$ is exact and $`B`$ closed on $``$. Although it is important to note that such a boundary condition removes topologically non-trivial gauge configurations.
In this case no gauge symmetries are broken by the boundary. Let us proceed as above and integrate over the bulk $`A_0`$ and $`B_0`$ fields. By this we mean that we split $`A_0=a_0+\stackrel{~}{A}_0`$, where $`\stackrel{~}{A}_0`$ vanishes on $``$ and $`a_0`$ has support on $``$, and then integrate over $`\stackrel{~}{A}_0`$. In this way we find
$$S=𝑑t_\mathrm{\Sigma }\dot{\mathrm{\Phi }}_Ad^{}\mathrm{\Phi }_Ba_0d^{}\mathrm{\Phi }_B$$
(67)
Finally we observe that the boundary conditions imply that $`a_0=\dot{\mathrm{\Phi }}_A`$ and hence the boundary action vanishes. Thus, in this case, there are no boundary modes.
Finally we can consider what happens in the case where the theory is not quite topological but includes a standard kinetic term for $`B`$
$$S=_{}dAB+\frac{1}{2}dBdB$$
(68)
It is not longer so simple to integrate out $`B_0`$. However we can see that if we choose the boundary conditions which break the gauge symmetries then there will be massless gauge modes that propagate along the boundary. These can also be thought of as Goldstone modes for the global symmetry resulting from gauge transformations which do not vanish on the boundary. For further details on singleton modes in such theories see .
### 5.2 Singletons in NMS
NMS and its variant (39) contain the same $`dAB`$ coupling that we have just discussed. (In NMS there is also a kinetic term for $`B`$ which is absent in the latter case.) Therefore we expect that if we choose gauge symmetry violating boundary conditions then singleton modes will propagate along the boundary.
We saw that by adding a suitable boundary term we could ensure that both these actions were supersymmetric on a manifold with boundary provided that we imposed the correct boundary conditions and terms. In particular we required that $`A=d\mathrm{\Phi }_A`$ and $`V=0`$ on $``$. These boundary conditions restrict the topology of the connection $`A`$ but preserve the gauge symmetry in the presence of the boundary. Thus there will not be any singleton modes in this case.
In the more interesting case that we do not want to, or cannot, restrict the gauge field $`A`$ to be exact then supersymmetry will be broken by the boundary. Furthermore given the previous discussion we expect to see singleton modes. In the case of NMS there is a kinetic term for $`B`$, just as in the action (68). However from the discussion of (68) it is clear that there will be singleton modes from the gauge symmetry if we impose the boundary condition $`A_0=0`$ on $``$.
In the case of the variant theory (39) there is no kinetic term for $`B`$ and we can be more explicit. In particular we choose the boundary condition $`A_0=0`$. Proceeding as before we can integrate out $`A_0`$ which leads to the constaint
$$d^{}B^{}i\overline{\psi }_i\mathrm{\Gamma }_j\psi _kdx^idx^jdx^k=0$$
(69)
Next we integrate out $`B_0`$ to find
$$d^{}A^{}=0$$
(70)
Thus we can set $`A^{}=d^{}\mathrm{\Phi }_A`$. Substituting all this back into the action we find
$$S=𝑑t_{^{}}\sqrt{g}R4i\overline{\psi }_i\mathrm{\Gamma }^{i0k}D_0\psi _k4i\overline{\psi }_\mu \mathrm{\Gamma }^{\mu i\nu }𝒟_i^+\psi _\nu 4d^{}\dot{\mathrm{\Phi }}_AB^{}$$
(71)
Note that $`D_0`$ appears instead of $`𝒟_0^+`$. Next we integrate the last term by parts and use the constraint (69)
$$S=𝑑t_{^{}}\sqrt{g}R4i\overline{\psi }_\mu \mathrm{\Gamma }^{\mu \nu \lambda }e^{\mathrm{\Phi }_A\mathrm{\Gamma }_5}D_\nu (e^{\mathrm{\Phi }_A\mathrm{\Gamma }_5}\psi _\lambda )4𝑑t_\mathrm{\Sigma }\dot{\mathrm{\Phi }}_AB^{}$$
(72)
Lastly we perform the field redefinition $`\psi _\mu =e^{\mathrm{\Phi }_A\mathrm{\Gamma }_5}\mathrm{\Psi }_\mu `$ and arrive at the familar on-shell supergravity but with an additional boundary term
$`S={\displaystyle 𝑑t_{^{}}\sqrt{g}R}4i\overline{\mathrm{\Psi }}_\mu \mathrm{\Gamma }^{\mu \nu \lambda }D_\nu \mathrm{\Psi }_\lambda 4{\displaystyle 𝑑t_\mathrm{\Sigma }\dot{\mathrm{\Phi }}_A}B^{}`$ (73)
We must still solve for the contraint (69), be precise about the Fermionic boundary conditions and include any appropriate boundary terms, such as the Gibbons-Hawking term. However regardless of how we do this it is clear that we will always have $`B^{}=d\mathrm{\Phi }_B+\mathrm{}`$ where $`\mathrm{\Phi }_B`$ is a Bosonic boundary mode and the ellipsis denotes Fermionic terms. For example if we assume that $`A_0=B_0=0`$, $`\mathrm{\Psi }_\mu =D_\mu \eta `$ is pure gauge and $`R_{\mu \nu \lambda \rho }=0`$ on the boundary then we have a well posed boundary value problem and we find that
$$B^{}=d^{}\mathrm{\Phi }_B+i\overline{\eta }\mathrm{\Gamma }_iD_j\eta dx^idx^j$$
(74)
on $``$ so that the Fermionic gauge modes also propagate along the boundary. Note that as a consequence of their topological origin the singleton modes do not come with a factor of $`\sqrt{g}`$ and hence do not contribute to Einstein’s equation.
### 5.3 Supersymmetry transformation of $`\mathrm{\Phi }_B`$
It is helpful to consider the on-shell supersymmetries of NMS. These are
$`\delta _ϵe_\mu ^{\underset{¯}{\nu }}`$ $`=`$ $`2i\overline{ϵ}\mathrm{\Gamma }^{\underset{¯}{\nu }}\psi _\mu `$
$`\delta _ϵ\psi _\mu `$ $`=`$ $`𝒟_\mu ^+ϵ`$
$`\delta _ϵB_{\mu \nu }`$ $`=`$ $`4i\overline{ϵ}\mathrm{\Gamma }_{[\mu }\psi _{\nu ]}`$
$`\delta _ϵA_\mu `$ $`=`$ $`0`$
and one can check that they preserve the on-shell conditions $`dA=V=0`$, as they should. Note that from the condition $`V=0`$ we must have
$$dBi\overline{\psi }_\nu \mathrm{\Gamma }_\lambda \psi _\rho dx^\nu dx^\lambda dx^\rho =0$$
(76)
If $`\psi _\mu =𝒟_\mu ^+\eta `$ is pure gauge then
$$dB=id\left(\overline{\eta }\mathrm{\Gamma }_\lambda 𝒟_\rho ^+\eta dx^\lambda dx^\rho \right)i\overline{\eta }\mathrm{\Gamma }_\lambda 𝒟_\nu ^+𝒟_\rho ^+\eta dx^\nu dx^\lambda dx^\rho $$
(77)
The second term will vanish on-shell so that
$$B=d\mathrm{\Phi }_B+i\overline{\eta }\mathrm{\Gamma }_\lambda 𝒟_\rho ^+\eta dx^\lambda dx^\rho $$
(78)
for an arbitrary one-form $`\mathrm{\Phi }_B`$. Under a supersymmetry generated by $`ϵ`$ we clearly have that
$$\delta _ϵ\eta =ϵ$$
(79)
Using the expression above for $`\delta _ϵB_{\mu \nu }`$ we see that
$`2i\overline{ϵ}\mathrm{\Gamma }_\mu 𝒟_\nu ^+\eta dx^\mu dx^\nu `$ $`=`$ $`d\delta _ϵ\mathrm{\Phi }_B+(i\overline{ϵ}\mathrm{\Gamma }_\lambda 𝒟_\rho ^+\eta +i\overline{\eta }\mathrm{\Gamma }_\lambda 𝒟_\rho ^+ϵ)dx^\lambda dx^\rho `$
$`=`$ $`d\left(\delta _ϵ\mathrm{\Phi }_B+i\overline{ϵ}\mathrm{\Gamma }_\lambda \eta dx^\lambda \right)+2i\overline{ϵ}\mathrm{\Gamma }_\mu 𝒟_\nu ^+\eta dx^\mu dx^\nu `$
Thus
$$\delta _ϵ\mathrm{\Phi }_B=i\overline{ϵ}\mathrm{\Gamma }_\mu \eta dx^\mu $$
(81)
Hence we see that the gauge zero modes $`\eta `$ and $`\mathrm{\Phi }_B`$ are related by supersymmetry.
If we are on a manifold with boundary and use the boundary conditions (26) then the preserved supersymmetry is $`ϵ_+`$ and we must set $`\eta _{}=0`$ on $``$, where the signs denote the eigenvalue of $`\xi \mathrm{\Gamma }^ne^{2\mathrm{\Phi }_A\mathrm{\Gamma }_5}`$. In this case we see that
$$\delta _ϵ\mathrm{\Phi }_B=i\overline{ϵ}_+\mathrm{\Gamma }_\mu ^{}\eta _+dx^\mu ^{}$$
(82)
Thus only the component of $`\mathrm{\Phi }_B`$ that is tangential to the boundary is related to $`\eta _+`$ by supersymmetry.
## 6 Discussion
In this paper we have discussed various aspects of the auxiliary fields that arise in new minimal supergravity (NMS). In particular we showed that there are compact three-manifolds with well-defined Killing spinors which are not well-defined in old minimal supergravity or simple off-shell supergravity. We also showed that, subject to suitable boundary conditions, the auxiliary fields actually give rise to physical on-shell degrees of freedom that reside on the boundary of spacetime. Thus one can in principle distinguish between different off-shell forms of supergravity using on-shell physics. We also demonstrated how half of the supersymmetry could be preserved in NMS on a manifold with boundary, provided the gauge field is trivial on the boundary. This suggests that there might be interesting applications to brane world senarios where a topologically non-trivial auxiliary gauge field would lead to supersymmetry breaking. Finally we would like to address some related issues.
It would be worthwhile extending the discussion of the present paper to other off-shell formulations of supergravity. Apart from old and new minmal supergravity there is also the so-called $`\beta `$FFC formulation . It was observed in that the $`\beta `$FFC formulation can be understood as the coupling of NMS supergravity to a compensating chiral multiplet whose Bosonic content is a complex scalar (along with an auxiliary field). The logarithm of the absolute value of the scalar field is identified with the dilaton $`\varphi `$ whereas its phase is eaten by the two-form $`B`$ to produce a dynamical two-form which is dualized to the axion $`a`$.
Let us describe the $`\beta `$FFC formulation in more detail. The complex scalar of the compensating chiral multiplet is given a non-vanishing chiral weight. In particular, under a chiral transformation, its phase $`\phi `$ is shifted; $`\phi \phi \chi `$ while its absolute value is invariant. The chiral covariant derivative of $`\phi `$ is therefore
$$𝒟_\mu ^+\phi =_\mu \phi +A_\mu $$
(83)
Hence the kinetic term for $`\phi `$ introduces a quadratic term for the chiral gauge field $`A`$ in the Lagrangian. The resulting $`A`$ equation of motion now algebraically determines $`A`$ in terms of $`B`$ and $`\phi `$ to be
$$A=d\phi 4dB$$
(84)
(Recall that without coupling to the compensating chiral multiplet the $`A`$ equation of motion ensured that $`B`$ was non-dynamical: $`dB=0`$.) Thus in the $`\beta `$FFC formulation there is still a chiral gauge field that couples minimally to the Fermions, only now it is determined by $`B`$ and $`\phi `$. Note that the equation of motion for $`B`$ is $`d(e^{2\varphi }dB)=0`$ and hence it is possible to have $`dA0`$ on-shell.
However we cannot make the Bieberbach manifolds supersymmetric as we did for NMS since if $`A=dx^3/2R`$ then we must have $`\phi =x^3/2R`$ or $`dB0`$. The former case is forbidden as there are couplings of $`\phi `$ to the Fermions in the Lagrangian which require that $`\phi `$ be single valued. In the latter case one sees that a non-zero $`dB`$ will lead to a non-vanishing energy-momentum tensor so that the Bieberbach manifolds will no longer satisfy the Einstein equations (although this raises the possibility of interesting new supersymmetric “flux compactifications”).
It is important to observe that the chiral symmetry that NMS supergravity gauges is anomalous. This has been shown to lead to supersymmetry anomalies in the quantum theory. Happily all is not lost as a Green-Schwarz anomaly cancelation for NMS supergravity has been found in and one can show that the Bieberbach manifolds remain supersymmetric.
There has been some debate in the literature as to whether or not old minimal, NMS or the $`\beta `$FFC formulation results from four-dimensional string theory (see also for related discussions on the appearance of new minimal and the $`\beta `$FFC formulations). However the main message of this paper has been to show that there can be hidden on-shell physics in the auxiliary fields and these remain largely unknown in higher dimensions. Therefore to make further contact with string theory it is important to develop NMS and other off-shell formulations further. In particular it is not clear how to couple NMS to chiral multiplets with a potential, as is needed in string theory. The problem is that the superpotential must transform under chiral symmetries. One way to achieve this might be to postulate a chiral multiplet with a complex scalar $`\varphi _0`$ that shifts under the chiral symmetry
$$\delta _\chi \varphi _0=\chi $$
(85)
in addition to the other scalars $`\varphi _I`$ which are chirally invariant. Therefore the covariant derivative acts on $`\varphi _0`$ as
$$𝒟_\mu ^+\varphi _0=D_\mu \varphi _0+A_\mu $$
(86)
and is the ordinary derivative on $`\varphi _I`$. If this could be incorporated into NMS then one could attempt to include couplings to a superpotential of the form
$$W=e^{2i\varphi _0}\stackrel{~}{W}(\varphi _I)$$
(87)
where $`\stackrel{~}{W}`$ depends holomorphically on $`\varphi _I`$. This suggestion is reminicient of the $`\beta `$FFC formulation, coupled to a superpotential. Therefore one expects similar effects whereby $`B`$ eats the real part of $`\varphi _0`$ and becomes the dynamical axion and the auxiliary gauge field $`A_\mu `$ is algebraically determined in terms of the other fields.
Finally, let us return to our motivation of formulating $`M`$-theory on Spin<sup>c</sup> manifolds. Of course, we do not want to introduce a new propagating degree of freedom through the Spin<sup>c</sup> connection. In this degree of freedom is part of the $`B`$-field, but it is not evident how to implement such a relation in general. Another problem one must face is reconciling the Spin<sup>c</sup> structure with the standard reality properties of the gravitino. Finally, anomaly cancellation arguments would need to be modified. For example, the quantization of $`G`$-flux of now becomes $`[G]_{DR}=\left(\frac{1}{4}\overline{p}_1+\frac{1}{2}(\overline{c}_1)^2\right)\mathrm{mod}\overline{H}^4(Y,Z)`$, where the overline denotes reduction modulo torsion, and $`c_1`$ is the Chern class of the Spin<sup>c</sup> structure. (See for related discussion.) Thus, finding such a generalization of $`M`$-theory - if it exists - seems quite challenging. The results of this paper make it clear that in such a search, one must first decide on some choice of off-shell formulation of the theory.
## Acknowledgements
We would like to thank B. Acharya for discussions on Killing spinors for Bieberbach manifolds. N.L. was supported by a PPARC Advanced Fellowship and in part by the grant PPA/G/O/2002/00475. He would also like to thank Rutgers University for its hospitality while some of this work was completed. The work of GM is supported in part by DOE grant DE-FG02-96ER40949. |
warning/0507/hep-lat0507025.html | ar5iv | text | # 1 Introduction
## 1 Introduction
Gauge invariance of the chiral determinant on the lattice has been considered a major problem . Particular developments aiming at gauge invariance have been presented in Ref. and have been the basis of various further works. Motivated by our observation that actually more information is available on the transformation properties of the bases involved, we here reinvestigate the subject. We also extend the considerations beyond the vacuum sector admitting zero modes and any value of the index.
We first show how the expressions for the chiral determinant in Ref. , which are based on gauge variations, can be evaluated further and find that with the covariance requirement for the current introduced there everything is fixed without any further assumptions. It thus becomes obvious that the developments in Ref. are not relevant for the question of gauge invariance, which we discuss in detail.
In our more general analysis we then reveal that not allowing arbitrary switching to different equivalence classes of pairs of bases is the general principle which also implies covariance of the current mentioned above. To admit zero modes and any value of the index we consider appropriate forms of fermionic correlation functions and investigate their behavior under finite gauge transformations, finding gauge covariance up to constant phase factors.
In Section 2 we collect general relations. In Section 3 we consider variations of the effective action and the special case of Ref. . In Section 4 we use finite transformations to analyze general correlation functions. Section 5 contains our conclusions.
## 2 General relations
### 2.1 Basic quantities
The chiral projections $`\overline{P}_+`$ and $`P_{}`$ are subject to
$$\overline{P}_+D=DP_{},$$
(2.1)
where $`D`$ is the Dirac operator. They can be expressed as
$$P_{}=\frac{1}{2}(\text{1l}\gamma _5G),\overline{P}_+=\frac{1}{2}(\text{1l}+\overline{G}\gamma _5),$$
(2.2)
which because of $`P_{}^{}=P_{}=P_{}^2`$ and $`\overline{P}_+^{}=\overline{P}_+=\overline{P}_+^2`$ implies $`G^1=G^{}=\gamma _5G\gamma _5`$ and $`\overline{G}^1=\overline{G}^{}=\gamma _5\overline{G}\gamma _5`$. Requiring $`D`$ to be $`\gamma _5`$-Hermitian and normal and $`G`$ and $`\overline{G}`$ to be functions of $`D`$ we get $`\overline{N}N=I`$ for the numbers of anti-Weyl and Weyl degrees of freedom $`\overline{N}=\text{Tr}\overline{P}_+`$ and $`N=\text{Tr}P_{}`$ and the index $`I`$ of $`D`$. For more details on the operator properties we refer to the recent analysis in Ref. .
Integrating out the Grassmann variables basic fermionic correlation functions for the Weyl degrees of freedom are given by
$$\psi _{\sigma _{r+1}}\mathrm{}\psi _{\sigma _N}\overline{\psi }_{\overline{\sigma }_{r+1}}\mathrm{}\overline{\psi }_{\overline{\sigma }_{\overline{N}}}_\text{f}=\frac{1}{r!}\underset{\overline{\sigma }_1\mathrm{}\overline{\sigma }_r}{}\underset{\sigma _1,\mathrm{},\sigma _r}{}\overline{\mathrm{{\rm Y}}}_{\overline{\sigma }_1\mathrm{}\overline{\sigma }_{\overline{N}}}^{}\mathrm{{\rm Y}}_{\sigma _1\mathrm{}\sigma _N}D_{\overline{\sigma }_1\sigma _1}\mathrm{}D_{\overline{\sigma }_r\sigma _r}$$
(2.3)
with the alternating multilinear forms
$$\mathrm{{\rm Y}}_{\sigma _1\mathrm{}\sigma _N}=\underset{i_1,\mathrm{},i_N=1}{\overset{N}{}}ϵ_{i_1,\mathrm{},i_N}u_{\sigma _1i_1}\mathrm{}u_{\sigma _Ni_N},\overline{\mathrm{{\rm Y}}}_{\overline{\sigma }_1\mathrm{}\overline{\sigma }_{\overline{N}}}=\underset{j_1,\mathrm{},j_{\overline{N}}=1}{\overset{\overline{N}}{}}ϵ_{j_1,\mathrm{},j_{\overline{N}}}\overline{u}_{\overline{\sigma }_1j_1}\mathrm{}\overline{u}_{\overline{\sigma }_{\overline{N}}j_{\overline{N}}}.$$
(2.4)
The bases $`\overline{u}_{\overline{\sigma }j}`$ and $`u_{\sigma i}`$ in (2.4) satisfy
$$P_{}=uu^{},u^{}u=\text{1l}_\mathrm{w},\overline{P}_+=\overline{u}\overline{u}^{},\overline{u}^{}\overline{u}=\text{1l}_{\overline{\mathrm{w}}}.$$
(2.5)
General fermionic functions are linear combinations of the basic ones (2.3).
### 2.2 Subsets of bases
By (2.5) the bases are only fixed up to unitary transformations, $`u^{[S]}=uS`$, $`\overline{u}^{[\overline{S}]}=\overline{u}\overline{S}`$, under which the forms (2.4) get multiplied by factors $`det_\mathrm{w}S`$ and $`det_{\overline{\mathrm{w}}}\overline{S}`$, respectively, and therefore the correlation functions (2.3) by a factor<sup>1</sup><sup>1</sup>1To compare with vector theory we can consider its formulation analogous to (2.3) with Tr1l instead of $`\overline{N}`$ and $`N`$ and see that because of $`\overline{u}=u`$ we there get $`detSdetS^{}=1`$ instead of (2.6).
$$\underset{\mathrm{w}}{det}S\underset{\overline{\mathrm{w}}}{det}\overline{S}^{}=e^{i\vartheta }.$$
(2.6)
Therefore, in order that full correlation functions remain invariant, we have to require
$$\vartheta =\text{ const},$$
(2.7)
i.e. that the phase $`\vartheta `$ is independent of the gauge field. While without condition (2.7) all bases related to a chiral projection are connected by unitary transformations, with it the total set of pairs of bases $`u`$ and $`\overline{u}`$ is decomposed into inequivalent subsets, beyond which legitimate transformations do not connect. This has the important consequence that for the formulation of the theory one has to restrict to one of such subsets.
Different ones of the indicated subsets, which obviously are equivalence classes, are related by pairs of of basis transformations for which $`\vartheta `$ in (2.6) depends on the gauge field. The phase factor $`e^{i\vartheta (U)}`$ then determines how the results of the theory differ for the respective classes. In view of such differences one has to decide which class is appropriate for the description of physics, for which there is, however, so far no criterion.
### 2.3 Gauge transformations
Since $`G`$ and $`\overline{G}`$ are functions of $`D`$ the gauge-transformation behavior $`D^{}=𝒯D𝒯^{}`$ is inherited by them and then also by the chiral projections, which thus satisfy $`P_{}^{}=𝒯P_{}𝒯^{}`$ and $`\overline{P}_+^{}=𝒯\overline{P}_+𝒯^{}`$ in accordance with (2.1). In addition to the case where none of the chiral projections commutes with $`𝒯`$ the case where one of them is constant and thus commutes is of interest (examples of which are the particular choices in Ref. and in Ref. , respectively).
Considering $`[𝒯,P_{}]0`$ we note that given a solution $`u`$ of the conditions (2.5), then $`𝒯u`$ is a solution of the transformed conditions (2.5). To account for the fact that $`u`$ and $`u^{}`$ are only fixed up to unitary transformations we introduce the unitary transformation $`𝒮`$ getting $`u^{}=𝒯u𝒮`$ for all solutions of the transformed conditions. Analogous considerations apply to $`[𝒯,\overline{P}_+]0`$. In the case where $`[𝒯,P_{}]0`$ and $`[𝒯,\overline{P}_+]0`$ we thus have the general relations
$$u^{}=𝒯u𝒮,\overline{u}^{}=𝒯\overline{u}\overline{𝒮}.$$
(2.8)
For the phase $`\mathrm{\Theta }`$ in
$$\underset{\mathrm{w}}{det}𝒮\underset{\overline{\mathrm{w}}}{det}\overline{𝒮}^{}=e^{i\mathrm{\Theta }(𝒯)}$$
(2.9)
using (2.6) with (2.7) we then immediately get
$$\mathrm{\Theta }(\text{1l})=\text{const},$$
(2.10)
i.e. independence of the gauge field at least for $`𝒯=\text{1l}`$.
In the case where $`[𝒯,P_{}]0`$ and $`\overline{P}_+=`$ const the equivalence class of pairs of bases always contains constant $`\overline{u}_\mathrm{c}`$. This follows since given a pair $`u`$, $`\overline{u}`$ the basis $`\overline{u}`$ is generally related to $`\overline{u}_\mathrm{c}`$ by a unitary transformation $`\overline{u}=\overline{u}_\mathrm{c}\overline{S}_\mathrm{e}`$. Then transforming $`u`$ as $`u=u_\mathrm{e}S_\mathrm{e}`$, where the unitary $`S_\mathrm{e}`$ is subject to $`det_\mathrm{w}S_\mathrm{e}det_{\overline{\mathrm{w}}}\overline{S}_\mathrm{e}^{}=`$ const, according to (2.6) with (2.7) the pair $`u_\mathrm{e}`$, $`\overline{u}_\mathrm{c}`$ is in the same equivalence class as the pair $`u`$, $`\overline{u}`$. Analogously for a transformed pair $`u^{}`$, $`\overline{u}^{}`$ we get the equivalent one $`u_\mathrm{e}^{}`$, $`\overline{u}_\mathrm{c}`$. Instead of (2.8) we then have
$$u_\mathrm{e}^{}=𝒯u_\mathrm{e}\stackrel{~}{𝒮},\overline{u}_\mathrm{c}=\text{const},$$
(2.11)
with unitary $`\stackrel{~}{𝒮}`$, and instead of (2.9) obtain
$$\underset{\mathrm{w}}{det}\stackrel{~}{𝒮}=e^{i\stackrel{~}{\mathrm{\Theta }}(𝒯)}.$$
(2.12)
Using (2.6) with (2.7) we thus get the analogon to (2.10),
$$\stackrel{~}{\mathrm{\Theta }}(\text{1l})=\text{const},$$
(2.13)
i.e. again independence of the gauge field at least for $`𝒯=\text{1l}`$.
## 3 Variational approach
### 3.1 General relations
We define general gauge-field variations of a function $`\varphi (𝒰)`$ by
$$\delta \varphi (𝒰)=\frac{\text{d}\varphi \left(𝒰(t)\right)}{\text{d}t}|_{t=0},𝒰_\mu (t)=e^{t_\mu ^{\mathrm{left}}}𝒰_\mu e^{t_\mu ^{\mathrm{right}}},$$
(3.1)
where $`(𝒰_\mu )_{n^{}n}=U_{\mu n}\delta _{n^{},n+\widehat{\mu }}^4`$ and $`(_\mu ^{\mathrm{left}/\mathrm{right}})_{n^{}n}=B_{\mu n}^{\mathrm{left}/\mathrm{right}}\delta _{n^{},n}^4`$. The special case of gauge transformations then is straightforwardly described by
$$_\mu ^{\mathrm{left}}=_\mu ^{\mathrm{right}}=.$$
(3.2)
In the case of gauge transformations we can use the definition (3.1) and the finite transformation relations to obtain the related variations explicitly. For operators with $`𝒪(t)=𝒯(t)𝒪𝒯^{}(t)`$ and $`𝒯(t)=e^t`$ this gives
$$\delta ^\text{G}𝒪=[,𝒪].$$
(3.3)
In the case $`[𝒯,P_{}]0`$, $`[𝒯,\overline{P}_+]0`$ according to (2.8) we have for the bases $`u(t)=𝒯(t)u𝒮(t)`$, $`\overline{u}(t)=𝒯(t)\overline{u}\overline{𝒮}(t)`$ and obtain
$$\delta ^\text{G}u=u+u𝒮^{}\delta ^\text{G}𝒮,\delta ^\text{G}\overline{u}=\overline{u}+\overline{u}\overline{𝒮}^{}\delta ^\text{G}\overline{𝒮}.$$
(3.4)
In the case $`[𝒯,P_{}]0`$, $`\overline{P}_+=`$ const according to (2.11) we get
$$\delta ^\text{G}u_\mathrm{e}=u_\mathrm{e}+u_\mathrm{e}\stackrel{~}{𝒮}^{}\delta ^\text{G}\stackrel{~}{𝒮},\delta ^\text{G}\overline{u}_\mathrm{c}=0.$$
(3.5)
### 3.2 Effective action
Requiring absence of zero modes of $`D`$ (and thus also restricting to the vacuum sector) the effective action can be considered, for the variation of which one gets
$$\delta \mathrm{ln}\underset{\overline{\mathrm{w}}\mathrm{w}}{det}(\overline{u}^{}Du)=\text{Tr}(P_{}D^1\delta D)+\text{Tr}(\delta uu^{})\text{Tr}(\delta \overline{u}\overline{u}^{}),$$
(3.6)
in which due to (3.3)
$$\text{Tr}(P_{}D^1\delta ^\text{G}D)=\text{Tr}(\overline{P}_+)\text{Tr}(P_{}).$$
(3.7)
In the case $`[𝒯,P_{}]0`$, $`[𝒯,\overline{P}_+]0`$ we obtain with (3.4)
$$\text{Tr}(\delta ^\text{G}uu^{})=\text{Tr}(P_{})+\text{Tr}_\mathrm{w}(𝒮^{}\delta ^\text{G}𝒮),$$
(3.8)
$$\text{Tr}(\delta ^\text{G}\overline{u}\overline{u}^{})=\text{Tr}(\overline{P}_+)+\text{Tr}_{\overline{\mathrm{w}}}(\overline{𝒮}^{}\delta ^\text{G}\overline{𝒮}),$$
(3.9)
and therefore
$$\delta ^\text{G}\mathrm{ln}\underset{\overline{\mathrm{w}}\mathrm{w}}{det}(\overline{u}^{}Du)=\text{Tr}_\mathrm{w}(𝒮^{}\delta ^\text{G}𝒮)\text{Tr}_{\overline{\mathrm{w}}}(\overline{𝒮}^{}\delta ^\text{G}\overline{𝒮}).$$
(3.10)
In the case $`[𝒯,P_{}]0`$, $`\overline{P}_+=`$ const we get with (3.5)
$$\text{Tr}(\delta ^\text{G}u_\mathrm{e}u_\mathrm{e}^{})=\text{Tr}_\mathrm{w}(\stackrel{~}{𝒮}^{}\delta ^\text{G}\stackrel{~}{𝒮})+\text{Tr}(P_{}),$$
(3.11)
$$\text{Tr}(\delta ^\text{G}\overline{u}_\mathrm{c}\overline{u}_\mathrm{c}^{})=0,$$
(3.12)
and remembering that $`u`$, $`\overline{u}`$ and $`u_\mathrm{e}`$, $`\overline{u}_\mathrm{c}`$ are in the same equivalence class
$$\delta ^\text{G}\mathrm{ln}\underset{\overline{\mathrm{w}}\mathrm{w}}{det}(\overline{u}^{}Du)=\delta ^\text{G}\mathrm{ln}\underset{\overline{\mathrm{w}}\mathrm{w}}{det}(\overline{u}_\mathrm{c}^{}Du_\mathrm{e})=\text{Tr}_\mathrm{w}(\stackrel{~}{𝒮}^{}\delta ^\text{G}\stackrel{~}{𝒮})+\text{Tr}(\overline{P}_+),$$
(3.13)
where now $`\text{Tr}(\overline{P}_+)`$ is constant.
### 3.3 Special case of Lüscher
Lüscher considers the variation of the effective action imposing the Ginsparg-Wilson relation $`\{\gamma _5,D\}=D\gamma _5D`$ and using chiral projections which correspond to the choice $`\overline{G}=\text{1l}`$ and $`G=\text{1l}D`$ in (2.2). He assumes $`\overline{P}_+`$ to be represented by constant bases so that he is effectively starting from the pair $`u_\mathrm{e}`$, $`\overline{u}_\mathrm{c}`$ of our formulation.
An important point in Lüscher’s work is the definition of a current $`j_{\mu n}`$ by
$$\text{Tr}(\delta u_\mathrm{e}u_\mathrm{e}^{})=i\underset{\mu ,n}{}\text{tr}_\text{g}(\eta _{\mu n}j_{\mu n}),\delta U_{\mu n}=\eta _{\mu n}U_{\mu n},$$
(3.14)
which he requires to transform gauge-covariantly.
His generator is given by $`\eta _{\mu n}=B_{\mu ,n+\widehat{\mu }}^{\mathrm{left}}U_{\mu n}B_{\mu n}^{\mathrm{right}}U_{\mu n}^{}`$ in terms of our left and right generators. We get explicitly
$$j_{\mu n}=i(U_{\mu n}\rho _{\mu n}+\rho _{\mu n}^{}U_{\mu n}^{}),\rho _{\mu n,\alpha ^{}\alpha }=\underset{j,\sigma }{}u_{j\sigma }^{}\frac{u_{\sigma j}}{U_{\mu n,\alpha \alpha ^{}}}.$$
(3.15)
The requirement of gauge-covariance $`j_{\mu n}^{}=e^{B_{n+\widehat{\mu }}}j_{\mu n}e^{B_{n+\widehat{\mu }}}`$ because of $`U_{\mu n}^{}=e^{B_{n+\widehat{\mu }}}U_{\mu n}e^{B_n}`$ implies that one must have
$$\rho _{\mu n}^{}=e^{B_n}\rho _{\mu n}e^{B_{n+\widehat{\mu }}},$$
(3.16)
which with (2.11) leads to the condition
$$\underset{j,k}{}\stackrel{~}{𝒮}_{kj}^{}\frac{\stackrel{~}{𝒮}_{jk}}{U_{\mu n,\alpha \alpha ^{}}}=0.$$
(3.17)
Using (3.17) it follows that
$$\text{Tr}_\mathrm{w}(\stackrel{~}{𝒮}^{}\delta \stackrel{~}{𝒮})=0.$$
(3.18)
Because of $`\text{Tr}_\mathrm{w}(\stackrel{~}{𝒮}^{}\delta \stackrel{~}{𝒮})=\delta \mathrm{ln}det_\mathrm{w}\stackrel{~}{𝒮}`$ it is seen that (3.18) requires $`det_\mathrm{w}\stackrel{~}{𝒮}`$ to be independent of the gauge field. With (2.12) we thus obtain
$$\stackrel{~}{\mathrm{\Theta }}(𝒯)=\text{const},$$
(3.19)
i.e. that (2.13) extends to all $`𝒯`$.
Since (3.18) implies $`\text{Tr}_\mathrm{w}(\stackrel{~}{𝒮}^{}\delta ^\text{G}\stackrel{~}{𝒮})=0`$ and because of the particular form $`\overline{P}_+=\frac{1}{2}(1+\gamma _5)\text{1l}`$ we now get from (3.13)
$$\delta ^\text{G}\mathrm{ln}\underset{\overline{\mathrm{w}}\mathrm{w}}{det}(\overline{u}^{}Du)=\frac{1}{2}\text{Tr},$$
(3.20)
i.e. a definite result following without any further assumptions.
Because $`\mathrm{exp}(\frac{1}{2}\text{Tr})`$ in (3.20) depends only on the gauge transformation but not on the gauge field the chiral determinant is gauge invariant up to a constant (gauge-field independent) phase factor, which is $`\mathrm{exp}(i\stackrel{~}{\mathrm{\Theta }}+\frac{1}{2}\text{Tr})`$ as will be confirmed by the general analysis in Section 4.3.
### 3.4 Both chiral projections non-commuting
In the case $`[𝒯,P_{}]0`$, $`[𝒯,\overline{P}_+]0`$, starting analogously from $`\text{Tr}(\delta uu^{})\text{Tr}(\delta \overline{u}\overline{u}^{})`$ as in (3.14) from $`\text{Tr}(\delta u_\mathrm{e}u_\mathrm{e}^{})`$, one straightforwardly arrives at the analogon of (3.18),
$$\text{Tr}_\mathrm{w}(𝒮^{}\delta 𝒮)\text{Tr}_{\overline{\mathrm{w}}}(\overline{𝒮}^{}\delta \overline{𝒮})=0.$$
(3.21)
This with (2.9) leads to
$$\mathrm{\Theta }(𝒯)=\text{const},$$
(3.22)
generalizing (2.10) to all $`𝒯`$. With (3.21) we get from (3.10)
$$\delta ^\text{G}\mathrm{ln}\underset{\overline{\mathrm{w}}\mathrm{w}}{det}(\overline{u}^{}Du)=0,$$
(3.23)
i.e. again a definite result following without any further assumptions. The chiral determinant thus is gauge invariant up to a constant phase factor, i.e. up to the factor $`\mathrm{exp}(i\mathrm{\Theta })`$ as will be confirmed by the general analysis in Section 4.2.
### 3.5 Discussion
The definite result (3.20) shows that the developments presented in Ref. are not relevant for the question of gauge invariance. It reveals that whatever the gauge-field dependences of the bases might be<sup>2</sup><sup>2</sup>2For an observation with respect to general gauge-field dependences see the end of Section 4.1. their gauge-transformation properties are such that gauge variations of the effective action in the special case there are equal to $`\frac{1}{2}\text{Tr}`$. (Furthermore, also the aim $`\delta ^\text{G}\mathrm{ln}det_{\overline{\mathrm{w}}\mathrm{w}}(\overline{u}^{}Du)=0`$ in Ref. disagrees with the precise result (3.20)).
In Ref. a main argument was that without the anomaly cancelation condition one would be unable to cancel the anomaly term. However, as has become explicit here this is not true. Indeed, in the case considered there the basis term being of form $`\text{Tr}(P_{})`$ just compensates the respective contribution in the anomaly term (3.7) so that one gets cancelation up to the irrelevant quantity $`\text{Tr}(\overline{P}_+)=\frac{1}{2}\text{Tr}`$. In the case where both chiral projections do not commute with $`𝒯`$ the contribution $`\text{Tr}(P_{})\text{Tr}(\overline{P}_+)`$ of the bases even fully compensates the anomaly term.
Thus in detail the considerations of topological fields (anyway only feasible in the Abelian case) and the substantial additional assumptions in Ref. turn out to be irrelevant for gauge invariance.
It is to be emphasized in the present context that here as well as in Ref. and in the related discussion one is concerned with the formulation on the finite lattice and the non-perturbative description. Thus one cannot a priori expect to find the same situation as in continuum perturbation theory.<sup>3</sup><sup>3</sup>3In the limit of a perturbation expansion the compensating basis terms vanish so that one gets the usual setting of continuum perturbation theory where the anomaly cancelation condition is needed .
## 4 General analysis
### 4.1 Equivalence-class requirements
So far the conditions (3.19) and (3.22) have emerged as consequences of the covariance requirement for Lüscher’s current and have been seen to prevent the addition of completely arbitrary terms to the gauge variation of the effective action. We now turn to the general principle from which these conditions follow.
In the case where $`[𝒯,P_{}]0`$ and $`[𝒯,\overline{P}_+]0`$ from (2.8) with (2.9),
$$u^{}=𝒯u𝒮,\overline{u}^{}=𝒯\overline{u}\overline{𝒮},\underset{\mathrm{w}}{det}𝒮\underset{\overline{\mathrm{w}}}{det}\overline{𝒮}^{}=e^{i\mathrm{\Theta }},$$
(4.1)
and in the case where $`[𝒯,P_{}]0`$ and $`\overline{P}_+=`$ const from and (2.11) with (2.12),
$$u_\mathrm{e}^{}=𝒯u_\mathrm{e}\stackrel{~}{𝒮},\overline{u}_\mathrm{c}=\text{const},\underset{\mathrm{w}}{det}\stackrel{~}{𝒮}=e^{i\stackrel{~}{\mathrm{\Theta }}},$$
(4.2)
it is seen that admitting gauge-field dependence of $`\mathrm{\Theta }`$ and of $`\stackrel{~}{\mathrm{\Theta }}`$, respectively, means to allow switching to arbitrary inequivalent subsets of pairs of bases. Such combinations of gauge transformations with transformations to arbitrary different equivalence classes of pairs of bases would obviously introduce severe ambiguities. Thus to avoid these ambiguities by requiring (3.22),
$$\mathrm{\Theta }=\text{const},$$
(4.3)
and (3.19),
$$\stackrel{~}{\mathrm{\Theta }}=\text{const},$$
(4.4)
respectively, turns out to be appropriate, which also accounts for the fact that to describe physics one must restrict to one of the equivalence classes.
The important observation here is that given an equivalence class of pairs of bases in this way the equivalence class after the transformation remains uniquely determined, which is possible because in the case of gauge transformations we have the explicit relations (2.8) and (2.11), respectively. In contrast to this, considering general gauge-field dependences, given an equivalence class for one set of fields there is so far no criterion determining the equivalence class after a general change of the fields.
### 4.2 Non-commuting chiral projections
Considering the transformation of correlation functions in the case where $`[𝒯,P_{}]0`$ and $`[𝒯,\overline{P}_+]0`$ using (4.1) we obtain
$`\psi _{\sigma _1^{}}^{}\mathrm{}\psi _{\sigma _R^{}}^{}\overline{\psi }_{\overline{\sigma }_1^{}}^{}\mathrm{}\overline{\psi }_{\overline{\sigma }_{\overline{R}}^{}}^{}_\text{f}^{}=`$
$`e^{i\mathrm{\Theta }}{\displaystyle \underset{\sigma _1,\mathrm{},\sigma _R}{}}{\displaystyle \underset{\overline{\sigma }_1,\mathrm{},\overline{\sigma }_{\overline{R}}}{}}𝒯_{\sigma _1^{}\sigma _1}\mathrm{}𝒯_{\sigma _R^{}\sigma _R}\psi _{\sigma _1}\mathrm{}\psi _{\sigma _R}\overline{\psi }_{\overline{\sigma }_1}\mathrm{}\overline{\psi }_{\overline{\sigma }_{\overline{R}}}_\text{f}𝒯_{\overline{\sigma }_1\overline{\sigma }_1^{}}^{}\mathrm{}𝒯_{\overline{\sigma }_{\overline{R}}\overline{\sigma }_{\overline{R}}^{}}^{}.`$ (4.5)
Due to (4.3) the correlation functions thus turn out to transform gauge-covariantly up to a constant phase factor, i.e. up to the factor $`e^{i\mathrm{\Theta }}`$.
### 4.3 One constant chiral projection
In the case where $`[𝒯,P_{}]0`$ and $`\overline{P}_+=`$ const we can rewrite $`\overline{u}_\mathrm{c}`$ as
$$\overline{u}_\mathrm{c}=𝒯\overline{u}_\mathrm{c}S_𝒯$$
(4.6)
where $`S_𝒯`$ because of $`[𝒯,\overline{P}_+]=0`$ is unitary. Using this and (4.2) we get for the transformation of the correlation functions again the form (4.5) but with $`\mathrm{\Theta }`$ being replaced by $`\stackrel{~}{\mathrm{\Theta }}+\theta _𝒯`$ where $`\theta _𝒯`$ is given by
$$e^{i\theta _𝒯}=\underset{\overline{\mathrm{w}}}{det}S_𝒯^{}=\underset{\overline{\mathrm{w}}}{det}(\overline{u}_\mathrm{c}^{}𝒯\overline{u}_\mathrm{c}).$$
(4.7)
Since (4.7) does not depend on in the gauge field and because $`\stackrel{~}{\mathrm{\Theta }}`$ according to (4.4) is also constant the correlation functions thus are again seen to show gauge-covariant behavior up to a constant phase factor, i.e. up to the factor $`e^{i(\stackrel{~}{\mathrm{\Theta }}+\theta _𝒯)}`$.
To calculate $`i\theta _𝒯`$ we note that with $`[𝒯,\overline{P}_+]=0`$ and $`𝒯=e^{}`$ we get $`\overline{u}_\mathrm{c}^{}𝒯\overline{u}_\mathrm{c}=\overline{u}_\mathrm{c}^{}e^{\overline{P}_+}\overline{u}_\mathrm{c}`$ and the eigenequations $`\overline{P}_+\overline{u}_j^\mathrm{d}=\omega _j\overline{u}_j^\mathrm{d}`$ and $`\overline{P}_+\overline{u}_j^\mathrm{d}=\overline{u}_j^\mathrm{d}`$. With this we obtain $`det_{\overline{\mathrm{w}}}(\overline{u}_\mathrm{c}^{}e^{\overline{P}_+}\overline{u}_\mathrm{c})=_je^{\omega _j}=\mathrm{exp}(\text{Tr}(\overline{P}_+))`$, so that we find $`i\theta _𝒯=\text{Tr}(\overline{P}_+)`$. For $`\overline{P}_+=\frac{1}{2}(1+\gamma _5)\text{1l}`$ we then have i$`\theta _𝒯=\frac{1}{2}\text{Tr}`$.
## 5 Conclusions
We have given an unambiguous derivation of the gauge-transformation properties in chiral gauge theories on the finite lattices observing that there are more informations on the bases available which must not be ignored.
We have first considered the subject in terms of variations of the effective action In this context we have shown that satisfying the covariance requirement for Lüscher’s current the gauge variation leads to a definite field-independent quantity without any further assumptions. This means that the developments presented in Ref. are irrelevant for the question of gauge invariance.
In detail it has become explicit that on the lattice the anomaly term is canceled without imposing a respective condition. Thus the considerations of topological fields and the substantial additional assumptions in Ref. have turned out to be not relevant for gauge invariance.
In our more general analysis we then have pointed out that not allowing to combine gauge transformations with arbitrary switching to different equivalence classes of pairs of bases is the general principle which also implies covariance for Lüscher’s current.
In order to extend the considerations beyond the vacuum sector we have investigated the behavior of correlation functions also in the presence of zero modes and for any value of the index using finite gauge transformations. We have found that fermionic correlation functions transform gauge-covariantly up to constant phase factors.
## Acknowledgement
I wish to thank Michael Müller-Preussker and his group for their kind hospitality. |
warning/0507/hep-ph0507271.html | ar5iv | text | # Color Neutral Ground State of 2SC Quark Matter
## Abstract
We construct a new color neutral ground state of two-flavor color superconducting quark matter. It is shown that, in contrast with the conventionally considered ground state with diquark pairing in only one color direction, this new state is stable against arbitrary diquark fluctuations. In addition, the thermodynamical potential is found to be lower for this new state than for the conventional one.
Recent investigations on the QCD phase diagram have discovered a rich diversity of color superconducting quark matter phases at low temperatures and intermediate densities cscreview . Particularly interesting are possible implications of these results for the physics of compact stars cscstars as well as heavy ion collision experiments cschic addressing the domain of densities and temperatures where the strange quarks are still heavy and confined Ruster:2005jc ; Blaschke:2005uj . In both systems, the constraint of global color neutrality has to be fulfilled. In addition, the constraint of global electric neutrality has to be satisfied when macroscopic objects like compact stars are considered and flavor changing processes have enough time to adjust $`u`$ and $`d`$ quark chemical potentials according to $`\beta `$-equilibrium.
In view of the nonperturbative character of QCD, the theoretical treatment of hadronic matter at the vicinity of the phase transitions for low temperatures and finite densities is a problem of highest complexity, where rigorous theoretical approaches are not yet available and lattice QCD simulations are up to now not applicable. Therefore one has to rely on effective field-theoretical models of interacting quark matter, which are built taking into account the symmetry requirements of the QCD lagrangian and offer the possibility of dealing with the yet simplified interactions in a systematic way. Chiral quark models of QCD that adopt a current-current form of the interaction with mesonic and diquark components have been particularly useful, since the theories can be bosonized in a straightforward way. In the meson-diquark representation of these models an effective quantum hadrodynamics can be derived cahill , but it has not yet been used to explore the QCD phase diagram. A preparatory step for this formidable task is the investigation of the mean field approximation (MFA), where important progress has been recently made within a nonlocal, covariant formuation that has been extended even to the study of color superconductivity in the QCD phase diagram Blaschke:2004cc .
The two-flavor color superconductivity (2SC) phase of quark matter has been first considered in instanton-motivated QCD models in Rapp:1997zu , where the question of color neutrality was disregarded. However, it was soon realized that the 2SC phase in which color symmetry is broken by the orientation of the diquark field in one of the color directions (2SC-$`b`$) entails a mismatch in the quark densities of paired and unpaired colors provided that color chemical potentials are all equal to each other, $`\mu _r=\mu _g=\mu _b`$. Therefore, in the 2SC-$`b`$ state color neutrality requires color charge chemical potentials to be readjusted so that $`\mu _8=\frac{1}{2}(\mu _r+\mu _g2\mu _b)`$ acquires a nonvanishing value while $`\mu _3=\frac{3}{2}(\mu _r\mu _g)`$ remains zero due to the degeneracy of the red and green colors. While this adjustment of $`\mu _80`$ has long been considered a proper solution of the color neutrality constraint Huang:2002zd , a recent investigation of fluctuations around the mean field oriented in the blue (0,0,1) direction has revealed the instability of this state once color neutrality is required hjz . In the present paper we investigate the entire space of mean-field orientations in order to look for color neutral states which are stable against fluctuations. We find that these correspond to color neutral symmetric states (2SC-s) for which the condensates are equal in modulus in all the three directions of the color space.
As in Ref. hjz , we consider the simplest version of the flavor SU(2) Nambu$``$Jona-Lasinio model njl ; njlquark ; klev , extended so as to include the quark-quark interaction sector and finite chemical potentials
$``$ $`=`$ $`\overline{\psi }\left(i\gamma ^\mu _\mu +\mu \gamma _0\widehat{m}\right)\psi +G_S[\left(\overline{\psi }\psi \right)^2+\left(\overline{\psi }i\gamma _5\stackrel{}{\tau }\psi \right)^2]`$ (1)
$`+`$ $`G_D\left(\overline{\psi }_{i\alpha }^ci\gamma ^5ϵ^{ij}ϵ^{\alpha \beta \gamma }\psi _{j\beta }\right)\left(\overline{\psi }_{i\alpha }i\gamma ^5ϵ^{ij}ϵ^{\alpha \beta \gamma }\psi _{j\beta }^c\right).`$
Here $`\widehat{m}`$ is the diagonal current mass matrix for light quarks, $`G_S`$ and $`G_D`$ are coupling constants in color singlet and anti-triplet channels respectively, and $`\psi _{i\alpha }`$ stands for quark fields with flavor index $`i=u,d`$ and color index $`\alpha =r,g,b`$ (charge conjugated fields are given by $`\psi _{i\alpha }^c=i\gamma ^2\gamma ^0\overline{\psi }_{i\alpha }^T`$). The Pauli matrices $`\stackrel{}{\tau }=(\tau _1,\tau _2,\tau _3)`$ act in flavor space, while and $`ϵ^{ij}`$ and $`ϵ^{\alpha \beta \gamma }`$ are totally antisymmetric tensors in flavor and color spaces, respectively. In the present letter we are mainly concerned with the effect of color neutrality on the ground state of a two-flavor color superconductor, therefore we will restrict here the discussion to the flavor symmetric case and consider the extension to electrically neutral matter elsewhere. Regarding quark chemical potentials, the elements of the matrix
$$\mu =\mathrm{diag}(\mu _r,\mu _g,\mu _b,\mu _r,\mu _g,\mu _b)$$
(2)
can be written as
$`\mu _r=\mu _B/3+\mu _8/3+\mu _3/3,`$
$`\mu _g=\mu _B/3+\mu _8/3\mu _3/3,`$
$`\mu _b=\mu _B/32\mu _8/3,`$ (3)
where $`\mu _B`$ is the baryon chemical potential, while $`\mu _8`$ and $`\mu _3`$ are introduced to ensure color charge neutrality. Our aim is to discuss the color superconducting phase of the model in the mean-field approximation, which in general is characterized by nonvanishing diquark condensates
$$\mathrm{\Delta }_\alpha =\mathrm{\hspace{0.17em}2}G_D\overline{\psi }_{i\beta }^ci\gamma ^5ϵ^{ij}ϵ^{\beta \gamma \alpha }\psi _{j\gamma },\alpha =r,g,b.$$
(4)
For standard values of the diquark coupling, $`G_D=\frac{3}{4}G_S`$, there is no simultaneous chiral symmetry breaking in this phase. In addition, since for light quarks the current quark masses are significantly smaller than the typical values of $`\mu _B`$ and $`\mathrm{\Delta }_\alpha `$, for the purpose of the present study we can safely neglect both these small masses and the corresponding mesonic mean fields. Within this limit, we proceed to calculate the thermodynamical potential per unit volume at zero temperature. For convenience we perform our calculations in Euclidean space, where the thermodynamical potential in MFA is given by
$$\mathrm{\Omega }^{\text{MFA}}=\frac{\mathrm{\Delta }^2}{4G_D}\frac{1}{2}\frac{d^4p}{(2\pi )^4}\mathrm{ln}\mathrm{det}M,$$
(5)
and the constraint of global color neutrality has to be obeyed, i.e. color charge densities should vanish
$$Q_\alpha =\mathrm{\Omega }/\mu _\alpha =0,\alpha =3,8.$$
(6)
Here we have defined $`\mathrm{\Delta }^2=_{\alpha =r,g,b}\mathrm{\Delta }_\alpha ^2`$, and the space integral is regulated as usual by introducing a sharp three-momentum cutoff $`\mathrm{\Lambda }`$. The inverse fermion propagator $`M`$ is a $`48\times 48`$ matrix in Nambu-Gorkov, Dirac, color and flavor spaces, and can be conveniently written as
$$M=\left(\begin{array}{cc}M^+& 0\\ 0& M^{}\end{array}\right),M^{}=M_{}^{+}{}_{}{}^{},$$
(7)
with
$$M^+=\left(\begin{array}{cccccc}(G_{0r}^+)^1& 0& 0& 0& \mathrm{\Delta }_b^{}& \mathrm{\Delta }_g^{}\\ 0& (G_{0g}^+)^1& 0& \mathrm{\Delta }_b^{}& 0& \mathrm{\Delta }_r^{}\\ 0& 0& (G_{0b}^+)^1& \mathrm{\Delta }_g^{}& \mathrm{\Delta }_r^{}& 0\\ 0& \mathrm{\Delta }_b^{}& \mathrm{\Delta }_g^{}& (G_{0r}^{})^1& 0& 0\\ \mathrm{\Delta }_b^{}& 0& \mathrm{\Delta }_r^{}& 0& (G_{0g}^{})^1& 0\\ \mathrm{\Delta }_g^{}& \mathrm{\Delta }_r^{}& 0& 0& 0& (G_{0b}^{})^1\end{array}\right),$$
(8)
where $`\mathrm{\Delta }_\alpha ^{}=i\gamma _5\mathrm{\Delta }_\alpha `$ and $`(G_{0\alpha }^\pm )^1=[(p_4i\mu _\alpha )\gamma _4+\stackrel{}{p}\stackrel{}{\gamma }]`$ are $`4\times 4`$ matrices in Dirac space. After some algebra, it is seen that the determinant can be cast into the form
$$\mathrm{det}M=(S^+S^{})^4,$$
(9)
where
$`S^\pm `$ $`=`$ $`|C_r^\pm C_g^\pm C_b^\pm |^2+|C_r^\pm \mathrm{\Delta }_r^2+C_g^\pm \mathrm{\Delta }_g^2+C_b^\pm \mathrm{\Delta }_b^2|^2`$ (10)
$`+2[|C_r^\pm |^2\mathrm{Re}(C_{g}^{\pm }{}_{}{}^{}C_b^\pm )\mathrm{\Delta }_r^2+\mathrm{cycl}.\mathrm{perm}.\{\mathrm{𝑟𝑔𝑏}\}],`$
with
$$C_\alpha ^\pm =\mu _\alpha \pm |\stackrel{}{p}|+ip_4.$$
With this general expression at hand we can investigate the problem of finding the most favored state under the constraint of the color neutrality, i.e. $`Q_3=Q_8=0`$.
Since the expression in Eq.(10) is totally symmetric under cyclic permutations of the three color indices, a cubic symmetry is expected in the three-dimensional space spanned by the three color directions along which the magnitudes of the colored diquark condensates are the coordinates. In the standard Cartesian representation, the condensate vector in color space is given by
$$\stackrel{}{\mathrm{\Delta }}=\mathrm{\Delta }_r\stackrel{}{\mathrm{e}}_r+\mathrm{\Delta }_g\stackrel{}{\mathrm{e}}_g+\mathrm{\Delta }_b\stackrel{}{\mathrm{e}}_b.$$
(11)
Due to the above mentioned symmetry, in what follows we will restrict our discussion to the sector defined by nonnegative values of the Cartesian coordinates $`\mathrm{\Delta }_\alpha `$. In this representation, the 2SC-$`b`$ state is given by $`\stackrel{}{\mathrm{\Delta }}_{2\mathrm{S}\mathrm{C}b}=(0,0,\mathrm{\Delta }_b)`$, whereas the color symmetric state is $`\stackrel{}{\mathrm{\Delta }}_{2\mathrm{S}\mathrm{C}\mathrm{s}}=(\mathrm{\Delta }_s,\mathrm{\Delta }_s,\mathrm{\Delta }_s)`$, i.e. a vector pointing from the center to one of the edges of a cube in color space. For the 2SC-s state color neutrality is achieved with $`\mu _8=\mu _3=0`$, while everywhere else color symmetry is broken, entailing that either $`\mu _8`$, $`\mu _3`$ or both have to be different from zero in order to fulfill the color neutrality constraint.
The observation made in hjz stating that a 2SC-$`b`$ state defines a saddle point of the thermodynamical potential (5) in the order parameter space, being a minimum in the blue direction but a maximum in the red and green ones, leads to the important fact that the 2SC-$`b`$ state widely considered in the literature is not the true ground state of quark matter in the 2SC phase. Now the problem is to find the true 2SC ground state, which should be thermodynamically more favorable not only by a lower energy but also by its stability with respect to fluctuations in the amplitude and the orientation of the condensate.
We show here that the 2SC-s state proposed in this paper fulfills these conditions. To do this, we perform a numerical analysis taking a phenomenologically acceptable set of model parameters, namely $`\mathrm{\Lambda }=653`$ MeV, $`G_S\mathrm{\Lambda }^2=2.14`$ klev , together with $`G_D=\frac{3}{4}G_S`$. First, we show that the difference between the thermodynamical potentials of the 2SC-$`b`$ and 2SC-s states is positive along the relevant range of baryochemical potentials, say $`350\mathrm{MeV}\mu _B/3600\mathrm{MeV}`$. The corresponding curve is plotted in the upper panel of Fig. 1. Second, in the lower panel of Fig. 1 we prove the stability of the 2SC-s solution by showing the strict positivity of the eigenvalues $`k_1`$, $`k_2`$, $`k_3`$ of the curvature tensor
$$K_{ij}=\frac{1}{2}\frac{\mathrm{\Omega }^{\text{MFA}}}{\mathrm{\Delta }_i\mathrm{\Delta }_j}|_{2\mathrm{S}\mathrm{C}\mathrm{s}}$$
(12)
derived from the thermodynamical potential in the 2SC-s state. Interestingly, the largest eigenvalue $`k_1`$ corresponds to fluctuations in the “radial” direction $`(1,1,1)`$, while the two eigenvalues denoting the curvature in the orthogonal “angular” directions are found to be degenerate ($`k_2=k_3`$).
Note that a special situation occurs at a chemical potential $`\mu _B/3550`$ MeV, where the 2SC-s state becomes energetically degenerate with the 2SC-$`b`$ state and simultaneously the curvature in the angular directions vanishes. This implies that all orientations of the condensate vector $`\stackrel{}{\mathrm{\Delta }}`$ are equivalent to each other, i.e. the cubic symmetry degenerates to a spherical one and the preference of the 2SC-s state gets lost (one has in this case $`\mathrm{\Delta }_b=\sqrt{3}\mathrm{\Delta }_s`$). In contrast, for all other values of $`\mu _B`$ the 2SC-s state is preferred, owing to the penalty introduced for all other states which need finite color chemical potentials to achieve color neutrality.
The behavior of this penalty $`\delta \mathrm{\Omega }`$ in the condensate state space is illustrated by the contour plot shown in Fig. 2 for a baryochemical potential $`\mu _B=1200`$ MeV chosen such that a maximal effect can be demonstrated, see Fig. 1. In this plot we consider for simplicity the plane spanned by the axes $`\mathrm{\Delta }_r=\mathrm{\Delta }_g`$ and $`\mathrm{\Delta }_b`$, so that red and green colors are degenerate and one has $`\mu _3=0`$. The penalty, arising from color symmetry breaking, is a function of $`\mu _8`$ and vanish in the 2SC-s state for which $`\mathrm{\Delta }_r=\mathrm{\Delta }_g=\mathrm{\Delta }_b`$ and $`\mu _8=0`$. This figure strongly suggests that the 2SC-s state is, in fact, the absolute minimum of the mean field thermodynamical potential.
Another important feature of the 2SC-s state concerns the corresponding 12 quasiparticle modes. It turns out that for this state the dispersion relations and degeneracy factors are given by
$`E_0^\pm `$ $`=`$ $`|\stackrel{}{p}|\pm \mu _B/3[\times 2],`$ (13)
$`E_\mathrm{\Delta }^\pm `$ $`=`$ $`\sqrt{(|\stackrel{}{p}|\pm \mu _B/3)^2+\mathrm{\Delta }^2}[\times 4],`$ (14)
where $`\mathrm{\Delta }=\sqrt{3}\mathrm{\Delta }_s`$. This means that the system contains two gapless modes, just like in the case of the conventional 2SC-$`b`$ state. However, in the present case the gapless modes cannot be identified with the original $`u`$ and $`d`$ quarks of “blue” (unpaired) color in the original $`\{r,g,b\}`$ color basis but arise as a combination of all three color states.
In summary, we have shown in this letter that in the case of color neutrality the ground state of 2SC quark matter should be constructed in a “democratic” way, so that color symmetry is not broken by the choice of the orientation of the condensate vector in color space. For this state, the condition of color neutrality is fulfilled in a trivial way, since the penalty induced by otherwise necessary color chemical potentials is avoided. We have shown that this 2SC-s ground state is stable against fluctuations, thus the problem observed in hjz is solved. The 2SC-s state can serve as a starting point for considering hadronic correlations on the superconducting QCD vacuum, where due to the entanglement of the quark color states in the new basis color neutral quark and diquark excitations arise along the radial direction besides colored excitations in the tangential plane. The conclusions drawn so far should not be qualitatively altered when the additional constraint of electrical neutrality is imposed, which is important for applications in compact stars. Flavor asymmetry induced by this constraint could possibly inhibit the formation of the 2SC state in regions of the neutron star matter phase diagram. However, provided that a phase transition to 2SC quark matter is accomplished, it should be described by the 2SC-s state introduced in this paper. This issue, together with other extensions of the present work, will be explicitly discussed in forthcoming publications.
Acknowledgements. D.B. acknowledges discussions with Michael Buballa and the inspiring Color Superconductivity seminar of the Virtual Institute held at Frankfurt and Darmstadt. This work has been supported in part by CONICET and ANPCyT (Argentina), under grants PIP 02368, PICT00-03-08580 and PICT02-03-10718, and by a scientist exchange program between Germany and Argentina funded jointly by DAAD under grant No. DE/04/27956 and ANTORCHAS under grant No. 4248-6. |
warning/0507/nucl-th0507025.html | ar5iv | text | # Kaon-nucleon interaction in the extended chiral SU(3) quark model
## I Introduction
The kaon-nucleon ($`KN`$) scattering process has aroused particular interest in the past and many works have been devoted to this issue rbu90 ; dha02 ; barnes94 ; nbl02 ; bsi97 ; sle02 ; sle03 ; hjw03 . In Ref. rbu90 , the Jülich group presented a meson-exchange model on hadronic degrees of freedom to study the $`KN`$ phase shifts. Considering single boson exchanges ($`\sigma `$, $`\rho `$, and $`\omega `$) together with contributions from higher-order diagrams involving $`N`$, $`\mathrm{\Delta }`$, $`K`$, and $`K^{}`$ intermediate states, the authors can give a good description of $`KN`$ interaction, but the exchange of a short-range ($``$ 0.2 fm) phenomenological repulsive scalar meson $`\sigma _{rep}`$ had to be added in order to reproduce the $`S`$-wave phase shifts in the isospin $`I=0`$ channel. The range of this repulsion is much smaller than the nucleon size, which clearly shows that the quark substructure of the kaon and nucleon cannot be neglected. Further in Ref. dha02 the authors refined this model by replacing the phenomenological $`\sigma _{rep}`$ by one-gluon-exchange (OGE), and a satisfactory description of the $`KN`$ experimental data was gotten. However, in this hybrid model the one-pion exchange is supposed to be absent, which is true on the hadron level, but is not the case in a genuine quark model study, because the quark exchange effect in the single boson exchanges has to be considered. In Ref. barnes94 , Barnes and Swanson used the quark-Born-diagram (QBD) method to derive the $`KN`$ scattering amplitudes, and obtained reasonable results for the $`KN`$ phase shifts, but it is limited to $`S`$-wave. Subsequently, the Born approximation was applied to investigate the $`KN`$ scattering more extensively in Ref. nbl02 . Nevertheless, the magnitudes of most calculated phase shifts are too small. In Ref. bsi97 , taking the $`\pi `$ and $`\sigma `$ boson exchanges as well as the OGE and confining potential as the quark-quark interactions, the authors calculated the $`S`$-wave $`KN`$ phase shifts in a constituent quark model by using the resonating group method (RGM). The results are too attraction for $`I=0`$ channel and too repulsion for $`I=1`$ channel, and thus the authors concluded that a consistent description of $`S`$-wave $`KN`$ phase shifts in both isospin $`I=0`$ and $`I=1`$ channels simultaneously is not possible. In Ref. sle02 , Lemaire et al. studied the $`KN`$ phase shifts up to the orbit angular momentum $`L=4`$ on the quark level by using the RGM method. They only considered the OGE and confining potential as the quark-quark interaction, and their results can give a reasonably description of the $`S`$-wave phase shifts, but the $`P`$ and higher partial waves are poorly described. The authors further incorporated $`\pi `$ and $`\sigma `$ exchanges besides the OGE and confining potential in the quark-quark interaction in Ref. sle03 , but the agreement obtained with the experimental data is quite poor, especially the signs of the $`S_{01}`$, $`P_{03}`$, $`P_{11}`$, $`D_{05}`$, $`D_{13}`$, $`D_{15}`$, $`F_{07}`$, and $`F_{15}`$ waves are opposite to the experiment values. Recently, Wang et al. hjw03 gave a study on the $`KN`$ elastic scattering in a quark potential model. Their results are consistent with the experimental data, but in their model, a factor of color octet component is added arbitrarily and the size parameter of harmonic oscillator is chosen to be $`b_u=0.255`$ fm, which is too small compared with the radius of nucleon.
In spite of great successes, the constituent quark model needs to have a logical explanation, from the underlying theory of the strong interaction \[i.e., Quantum Chromodynamics (QCD)\] of the source of the constituent quark mass. Thus spontaneous vacuum breaking has to be considered, and as a consequence the coupling between the quark field and the Goldstone boson is introduced to restore the chiral symmetry. In this sense, the chiral quark model can be regarded as a quite reasonable and useful model to describe the medium-range nonperturbative QCD effect. By generalizing the SU(2) linear $`\sigma `$ model, a chiral SU(3) quark model is developed to describe the system with strangeness zyz97 . This model has been quite successful in reproducing the energies of the baryon ground states, the binding energy of deuteron, the nucleon-nucleon ($`NN`$) scattering phase shifts of different partial waves, and the hyperon-nucleon ($`YN`$) cross sections by performing the RGM calculations zyz97 ; lrd03 . Inspired by these achievements, we try to extend this model to study the baryon-meson interactions. In our previous works fhuang04nk ; fhuang04dk , we dynamically studied the $`S`$-, $`P`$-, $`D`$-, and $`F`$-wave $`KN`$ phase shifts by performing a RGM calculation. Comparing with Ref. sle03 , we obtained correct signs of the phase shifts of $`S_{01}`$, $`P_{11}`$, $`P_{03}`$, $`D_{13}`$, $`D_{05}`$, $`F_{15}`$, and $`F_{07}`$ partial waves, and for $`P_{01}`$, $`D_{03}`$, and $`D_{15}`$ channels we also got a considerable improvement in the magnitude. At the same time, the satisfactory results also show that the short-range $`KN`$ interaction dominantly originates from the quark and one-gluon exchanges.
It is a consensus that constituent quark is the dominat effective degree of freedom for low-energy hadron physics, but about what other proper effective degrees of freedom may be there still has been a debate glozman96 ; glozman00 ; isgur021 ; isgur022 ; liu99 ; liu00 . Glozman and Riska proposed that the Goldstone boson is the only other proper effective degree of freedom. In Ref. glozman96 ; glozman00 , they applied the quark-chiral field coupling model to study the baryon structure, and replaced OGE by vector-meson coupling. They pointed out the spin-flavor interaction is important in explaining the energy of the Roper resonance and got a comparatively good fit to the baryon spectrum. However Isgur gave a critique of the boson exchange model and insisted that the OGE governs the baryon structure isgur021 ; isgur022 . In Refs. liu99 ; liu00 , Liu et al. produced a valence lattice QCD result which supports the Goldstone boson exchange picture, but Isgur pointed out that this is unjustified isgur021 ; isgur022 . On the other hand, in the study of $`NN`$ interactions on the quark level, the short-range feature can be explained by OGE interaction and quark exchange effect, while in the traditional one-boson exchange (OBE) model on the baryon level it comes from vector-meson ($`\rho `$, $`K^{}`$, $`\omega `$, and $`\varphi `$) exchange. Some authors also studied the short-range interaction as stemming from the Goldstone boson exchanges on the quark level stancu97 ; shimizu00 ; lrd03 , and it has been shown that these interactions can substitute traditional OGE mechanism. Anyhow, for low-energy hadron physics, what other proper effective degrees of freedom besides constituent quarks may be, whether OGE or vector-meson exchange is the right mechanism for describing the short-range quark-quark interaction, or both of them are important, is still a controversial and challenging problem.
In this paper, we extend the chiral SU(3) quark model to include the coupling between the quark and vector chiral fields. The OGE which dominantly governs the short-range quark-quark interaction in the original chiral SU(3) quark model is now nearly replaced by the vector-meson exchange. As we did in Refs. fhuang04nk ; fhuang04dk , the mass of the $`\sigma `$ meson is taken to be $`675`$ MeV and the mixing of $`\sigma _0`$ (scalar singlet) and $`\sigma _8`$ (scalar iso-scalar) is considered. The set of parameters we used can satisfactorily reproduce the energies of the ground states of the octet and decuplet baryons. Using this model, we perform a dynamical calculation of the $`S`$, $`P`$, $`D`$, $`F`$ wave $`KN`$ phase shifts in the isospin $`I=0`$ and $`I=1`$ channels by solving a RGM equation. The calculated phase shifts for different partial waves are similar to those obtained by the original chiral SU(3) quark model. In comparison with a recent RGM study on a quark level sle03 , our investigation achieves a considerable improvement on the theoretical phase shifts, and for many channels the theoretical results are qualitatively consistent with the experimental data. Nevertheless there is no improvement in this new approach for the $`P_{13}`$ and $`D_{15}`$ partial waves, of which the calculated phase shifts are too much repulsive and attractive respectively when the laboratory momentum of the kaon meson is greater than 300 MeV, as it was the case in the past. It would be studied in future work if there are some physical ingredients missing in our quark model investigations.
The paper is organized as follows. In the next section the framework of the extended chiral SU(3) quark model is briefly introduced. The results for the $`S`$-, $`P`$-, $`D`$-, and $`F`$-wave $`KN`$ phase shifts are shown in Sec. III, where some discussion is presented as well. Finally, the summary is given in Sec. IV.
## II Formulation
### II.1 Model
The chiral SU(3) quark model has been widely described in the literature fhuang04nk ; fhuang04dk and we refer the reader to those works for details. Here we just give the salient features of the extended chiral SU(3) quark model.
In the extended chiral SU(3) quark model, besides the nonet pseudoscalar meson fields and nonet scalar meson fields, the couplings among vector meson fields with quarks is also considered. With this generalization, in the interaction Lagrangian a term of coupling between the quark and vector meson field is introduced,
$`_I^v=g_{chv}\overline{\psi }\gamma _\mu T^aA_a^\mu \psi {\displaystyle \frac{f_{chv}}{2M_P}}\overline{\psi }\sigma _{\mu \nu }T^a^\nu A_a^\mu \psi .`$ (1)
Thus the meson fields induced effective quark-quark potentials can be written as
$`V_{ij}^{ch}={\displaystyle \underset{a=0}{\overset{8}{}}}V_{\sigma _a}(𝒓_{ij})+{\displaystyle \underset{a=0}{\overset{8}{}}}V_{\pi _a}(𝒓_{ij})+{\displaystyle \underset{a=0}{\overset{8}{}}}V_{\rho _a}(𝒓_{ij}),`$ (2)
where $`\sigma _0,\mathrm{},\sigma _8`$ are the scalar nonet fields, $`\pi _0,..,\pi _8`$ the pseudoscalar nonet fields, and $`\rho _0,..,\rho _8`$ the vector nonet fields. The expressions of these potentials are
$`V_{\sigma _a}(𝒓_{ij})=C(g_{ch},m_{\sigma _a},\mathrm{\Lambda })X_1(m_{\sigma _a},\mathrm{\Lambda },r_{ij})[\lambda _a(i)\lambda _a(j)]+V_{\sigma _a}^{𝒍\mathbf{}𝒔}(𝒓_{ij}),`$ (3)
$`V_{\pi _a}(𝒓_{ij})=C(g_{ch},m_{\pi _a},\mathrm{\Lambda }){\displaystyle \frac{m_{\pi _a}^2}{12m_{q_i}m_{q_j}}}X_2(m_{\pi _a},\mathrm{\Lambda },r_{ij})(𝝈_i𝝈_j)[\lambda _a(i)\lambda _a(j)]+V_{\pi _a}^{ten}(𝒓_{ij}),`$ (4)
$`V_{\rho _a}(𝒓_{ij})`$ $`=`$ $`C(g_{chv},m_{\rho _a},\mathrm{\Lambda })\{X_1(m_{\rho _a},\mathrm{\Lambda },r_{ij})+{\displaystyle \frac{m_{\rho _a}^2}{6m_{q_i}m_{q_j}}}(1+{\displaystyle \frac{f_{chv}}{g_{chv}}}{\displaystyle \frac{m_{q_i}+m_{q_j}}{M_P}}+{\displaystyle \frac{f_{chv}^2}{g_{chv}^2}}`$ (5)
$`\times {\displaystyle \frac{m_{q_i}m_{q_j}}{M_P^2}})X_2(m_{\rho _a},\mathrm{\Lambda },r_{ij})(𝝈_i𝝈_j)\}[\lambda _a(i)\lambda _a(j)]+V_{\rho _a}^{𝒍\mathbf{}𝒔}(𝒓_{ij})+V_{\rho _a}^{ten}(𝒓_{ij}),`$
with
$`V_{\sigma _a}^{𝒍\mathbf{}𝒔}(𝒓_{ij})`$ $`=`$ $`C(g_{ch},m_{\sigma _a},\mathrm{\Lambda }){\displaystyle \frac{m_{\sigma _a}^2}{4m_{q_i}m_{q_j}}}\left\{G(m_{\sigma _a}r_{ij})\left({\displaystyle \frac{\mathrm{\Lambda }}{m_{\sigma _a}}}\right)^3G(\mathrm{\Lambda }r_{ij})\right\}`$ (6)
$`\times [𝑳(𝝈_i+𝝈_j)][\lambda _a(i)\lambda _a(j)],`$
$`V_{\rho _a}^{𝒍\mathbf{}𝒔}(𝒓_{ij})`$ $`=`$ $`C(g_{chv},m_{\rho _a},\mathrm{\Lambda }){\displaystyle \frac{3m_{\rho _a}^2}{4m_{q_i}m_{q_j}}}\left(1+{\displaystyle \frac{f_{chv}}{g_{chv}}}{\displaystyle \frac{2(m_{q_i}+m_{q_j})}{3M_P}}\right)`$ (7)
$`\times \left\{G(m_{\rho _a}r_{ij})\left({\displaystyle \frac{\mathrm{\Lambda }}{m_{\rho _a}}}\right)^3G(\mathrm{\Lambda }r_{ij})\right\}[𝑳(𝝈_i+𝝈_j)][\lambda _a(i)\lambda _a(j)],`$
and
$`V_{\pi _a}^{ten}(𝒓_{ij})=C(g_{ch},m_{\pi _a},\mathrm{\Lambda }){\displaystyle \frac{m_{\pi _a}^2}{12m_{q_i}m_{q_j}}}\left\{H(m_{\pi _a}r_{ij})\left({\displaystyle \frac{\mathrm{\Lambda }}{m_{\pi _a}}}\right)^3H(\mathrm{\Lambda }r_{ij})\right\}\widehat{S}_{ij}[\lambda _a(i)\lambda _a(j)],`$ (8)
$`V_{\rho _a}^{ten}(𝒓_{ij})`$ $`=`$ $`C(g_{chv},m_{\rho _a},\mathrm{\Lambda }){\displaystyle \frac{m_{\rho _a}^2}{12m_{q_i}m_{q_j}}}\left(1+{\displaystyle \frac{f_{chv}}{g_{chv}}}{\displaystyle \frac{m_{q_i}+m_{q_j}}{M_P}}+{\displaystyle \frac{f_{chv}^2}{g_{chv}^2}}{\displaystyle \frac{m_{q_i}m_{q_j}}{M_P^2}}\right)`$ (9)
$`\times \left\{H(m_{\pi _a}r_{ij})\left({\displaystyle \frac{\mathrm{\Lambda }}{m_{\pi _a}}}\right)^3H(\mathrm{\Lambda }r_{ij})\right\}\widehat{S}_{ij}[\lambda _a(i)\lambda _a(j)],`$
where
$`C(g_{ch},m,\mathrm{\Lambda })={\displaystyle \frac{g_{ch}^2}{4\pi }}{\displaystyle \frac{\mathrm{\Lambda }^2}{\mathrm{\Lambda }^2m^2}}m,`$ (10)
$`X_1(m,\mathrm{\Lambda },r)=Y(mr){\displaystyle \frac{\mathrm{\Lambda }}{m}}Y(\mathrm{\Lambda }r),`$ (11)
$`X_2(m,\mathrm{\Lambda },r)=Y(mr)\left({\displaystyle \frac{\mathrm{\Lambda }}{m}}\right)^3Y(\mathrm{\Lambda }r),`$ (12)
$`Y(x)={\displaystyle \frac{1}{x}}e^x,`$ (13)
$`G(x)={\displaystyle \frac{1}{x}}\left(1+{\displaystyle \frac{1}{x}}\right)Y(x),`$ (14)
$`H(x)=\left(1+{\displaystyle \frac{3}{x}}+{\displaystyle \frac{3}{x^2}}\right)Y(x),`$ (15)
$`\widehat{S}_{ij}=\left[3(𝝈_i\widehat{r}_{ij})(𝝈_j\widehat{r}_{ij})𝝈_i𝝈_j\right],`$ (16)
and $`M_P`$ being a mass scale, taken as proton mass. $`m_{\sigma _a}`$ is the mass of the scalar meson, $`m_{\pi _a}`$ the mass of the pseudoscalar meson, and $`m_{\rho _a}`$ the mass of the vector meson.
For the systems with an antiquark $`\overline{s}`$, the total Hamiltonian can be written as
$`H={\displaystyle \underset{i=1}{\overset{5}{}}}T_iT_G+{\displaystyle \underset{i<j=1}{\overset{4}{}}}V_{ij}+{\displaystyle \underset{i=1}{\overset{4}{}}}V_{i\overline{5}},`$ (17)
where $`T_G`$ is the kinetic energy operator for the center-of-mass motion, and $`V_{ij}`$ and $`V_{i\overline{5}}`$ represent the quark-quark ($`qq`$) and quark-antiquark ($`q\overline{q}`$) interactions, respectively,
$`V_{ij}=V_{ij}^{OGE}+V_{ij}^{conf}+V_{ij}^{ch},`$ (18)
where $`V_{ij}^{OGE}`$ is the one-gluon-exchange interaction,
$`V_{ij}^{OGE}={\displaystyle \frac{1}{4}}g_ig_j\left(\lambda _i^c\lambda _j^c\right)\left\{{\displaystyle \frac{1}{r_{ij}}}{\displaystyle \frac{\pi }{2}}\delta (𝒓_{ij})\left({\displaystyle \frac{1}{m_{q_i}^2}}+{\displaystyle \frac{1}{m_{q_j}^2}}+{\displaystyle \frac{4}{3}}{\displaystyle \frac{1}{m_{q_i}m_{q_j}}}(𝝈_i𝝈_j)\right)\right\}+V_{OGE}^{𝒍𝒔},`$ (19)
with
$`V_{OGE}^{𝒍𝒔}={\displaystyle \frac{1}{16}}g_ig_j\left(\lambda _i^c\lambda _j^c\right){\displaystyle \frac{3}{m_{q_i}m_{q_j}}}{\displaystyle \frac{1}{r_{ij}^3}}𝑳(𝝈_i+𝝈_j),`$ (20)
and $`V_{ij}^{conf}`$ is the confinement potential, taken as the quadratic form,
$`V_{ij}^{conf}=a_{ij}^c(\lambda _i^c\lambda _j^c)r_{ij}^2a_{ij}^{c0}(\lambda _i^c\lambda _j^c).`$ (21)
$`V_{i\overline{5}}`$ in Eq. (17) includes two parts: direct interaction and annihilation parts,
$`V_{i\overline{5}}=V_{i\overline{5}}^{dir}+V_{i\overline{5}}^{ann},`$ (22)
with
$`V_{i\overline{5}}^{dir}=V_{i\overline{5}}^{conf}+V_{i\overline{5}}^{OGE}+V_{i\overline{5}}^{ch},`$ (23)
where
$`V_{i\overline{5}}^{conf}=a_{i5}^c\left(\lambda _i^c\lambda _{5}^{c}{}_{}{}^{}\right)r_{i\overline{5}}^2a_{i5}^{c0}\left(\lambda _i^c\lambda _{5}^{c}{}_{}{}^{}\right),`$ (24)
$`V_{i\overline{5}}^{OGE}`$ $`=`$ $`{\displaystyle \frac{1}{4}}g_ig_s\left(\lambda _i^c\lambda _{5}^{c}{}_{}{}^{}\right)\left\{{\displaystyle \frac{1}{r_{i\overline{5}}}}{\displaystyle \frac{\pi }{2}}\delta (𝒓_{i\overline{5}})\left({\displaystyle \frac{1}{m_{q_i}^2}}+{\displaystyle \frac{1}{m_s^2}}+{\displaystyle \frac{4}{3}}{\displaystyle \frac{1}{m_{q_i}m_s}}(𝝈_i𝝈_5)\right)\right\}`$ (25)
$`{\displaystyle \frac{1}{16}}g_ig_s\left(\lambda _i^c\lambda _{5}^{c}{}_{}{}^{}\right){\displaystyle \frac{3}{m_{q_i}m_{q_5}}}{\displaystyle \frac{1}{r_{i\overline{5}}^3}}𝑳(𝝈_i+𝝈_5),`$
and
$`V_{i\overline{5}}^{ch}={\displaystyle \underset{j}{}}(1)^{G_j}V_{i5}^{ch,j}.`$ (26)
Here $`(1)^{G_j}`$ represents the G parity of the $`j`$th meson. For the $`NK`$ system, $`u(d)\overline{s}`$ can only annihilate into $`K`$ and $`K^{}`$ mesons—i.e.,
$`V_{i\overline{5}}^{ann}=V_{ann}^K+V_{ann}^K^{},`$ (27)
with
$`V_{ann}^K=C^K\left({\displaystyle \frac{1𝝈_q𝝈_{\overline{q}}}{2}}\right)_s\left({\displaystyle \frac{2+3\lambda _q\lambda _{\overline{q}}^{}}{6}}\right)_c\left({\displaystyle \frac{38+3\lambda _q\lambda _{\overline{q}}^{}}{18}}\right)_f\delta (𝒓),`$ (28)
and
$`V_{ann}^K^{}=C^K^{}\left({\displaystyle \frac{3+𝝈_q𝝈_{\overline{q}}}{2}}\right)_s\left({\displaystyle \frac{2+3\lambda _q\lambda _{\overline{q}}^{}}{6}}\right)_c\left({\displaystyle \frac{38+3\lambda _q\lambda _{\overline{q}}^{}}{18}}\right)_f\delta (𝒓),`$ (29)
where $`C^K`$ and $`C^K^{}`$ are treated as parameters and we adjust them to fit the mass of $`K`$ and $`K^{}`$ mesons.
### II.2 Determination of the parameters
The harmonic-oscillator width parameter $`b_u`$ is taken with different values for the two models: $`b_u=0.50`$ fm in the chiral SU(3) quark model and $`b_u=0.45`$ fm in the extended chiral SU(3) quark model. This means that the bare radius of baryon becomes smaller when more meson clouds are included in the model, which sounds reasonable in the sense of the physical picture. The up (down) quark mass $`m_{u(d)}`$ and the strange quark mass $`m_s`$ are taken to be the usual values: $`m_{u(d)}=313`$ MeV and $`m_s=470`$ MeV. The coupling constant for scalar and pseudoscalar chiral field coupling, $`g_{ch}`$, is determined according to the relation
$`{\displaystyle \frac{g_{ch}^2}{4\pi }}=\left({\displaystyle \frac{3}{5}}\right)^2{\displaystyle \frac{g_{NN\pi }^2}{4\pi }}{\displaystyle \frac{m_u^2}{M_N^2}},`$ (30)
with empirical value $`g_{NN\pi }^2/4\pi =13.67`$. $`g_{chv}`$ and $`f_{chv}`$ are the coupling constants for vector coupling and tensor coupling of the vector meson field, respectively. In the study of nucleon resonance transition coupling to vector meson, Riska and Brown took $`g_{chv}=3.0`$ and neglected the tensor coupling part riska01 . From the one-boson exchange theory on the baryon level, we can also obtain these two values according to the SU(3) relation between quark and baryon levels. For example,
$`g_{chv}=g_{NN\rho },`$ (31)
$`f_{chv}={\displaystyle \frac{3}{5}}(f_{NN\rho }4g_{NN\rho }).`$ (32)
In the Nijmegen model D, $`g_{NN\rho }=2.09`$ and $`f_{NN\rho }=17.12`$ nagels75 . From the two equations above, we get $`g_{chv}=2.09`$ and $`f_{chv}=5.26`$. In this work, we neglect the tensor coupling part of the vector meson field as did by Riska and Brown riska01 , and take the coupling constant for vector coupling of the vector-meson field to be $`g_{chv}=2.351`$ as the same we used in the $`NN`$ scattering calculation lrd03 , which is a little bit smaller than the value used in Ref. riska01 , but slightly larger than the value obtained from the $`NN\rho `$ coupling constant of Nijmegen model D nagels75 . The masses of all the mesons are taken to be the experimental values, except for the $`\sigma `$ meson, whose mass is treated as an adjustable parameter. We chose $`m_\sigma =675`$ MeV as the same in the original chiral SU(3) quark model fhuang04nk , where it is fixed by the $`S`$-wave $`KN`$ phase shifts. The cutoff radius $`\mathrm{\Lambda }^1`$ is taken to be the value close to the chiral symmetry breaking scale ito90 ; amk91 ; abu91 ; emh91 . After the parameters of chiral fields are fixed, the coupling constants of OGE, $`g_u`$ and $`g_s`$, can be determined by the mass splits between $`N`$, $`\mathrm{\Delta }`$ and $`\mathrm{\Sigma }`$, $`\mathrm{\Lambda }`$, respectively. The confinement strengths $`a_{uu}^c`$, $`a_{us}^c`$, and $`a_{ss}^c`$ are fixed by the stability conditions of $`N`$, $`\mathrm{\Lambda }`$, and $`\mathrm{\Xi }`$ and the zero-point energies $`a_{uu}^{c0}`$, $`a_{us}^{c0}`$, and $`a_{ss}^{c0}`$ by fitting the masses of $`N`$, $`\mathrm{\Sigma }`$, and $`\overline{\mathrm{\Xi }+\mathrm{\Omega }}`$, respectively.
In the calculation, $`\eta `$ and $`\eta ^{}`$ mesons are mixed by $`\eta _1`$ and $`\eta _8`$ with the mixing angle $`\theta ^{PS}`$ taken to be the usual value $`23^{}`$. $`\omega `$ and $`\varphi `$ mesons consist of $`\sqrt{1/2}(u\overline{u}+d\overline{d})`$ and $`(s\overline{s})`$, respectively, i.e., they are ideally mixed by $`\omega _1`$ and $`\omega _8`$ with the mixing angle $`\theta ^V=35.264^{}`$. For the $`KN`$ case, we also consider the mixing between $`\sigma _0`$ and $`\sigma _8`$. The mixing angle $`\theta ^S`$ is an open issue because the structure of the $`\sigma `$ meson is still unclear and controversial. We adopt two possible values as in our previous works fhuang04nk ; fhuang04dk , one is $`35.264^{}`$ which means that $`\sigma `$ and $`f_0`$ \[In our previous works $`f_0`$ was named $`ϵ`$ and $`a_0`$ was named $`\sigma ^{}`$\] are ideally mixed by $`\sigma _0`$ and $`\sigma _8`$, and the other is $`18^{}`$ which is provided by Dai and Wu based on their recent investigation of a dynamically spontaneous symmetry breaking mechanism ybd03 . In both of these two cases, the attraction of the $`\sigma `$ meson can be reduced a lot, and thus we can get reasonable $`S`$-wave $`KN`$ phase shifts.
The model parameters are summarized in Table 1. The masses of octet and decuplet baryons obtained from the extended chiral SU(3) quark model are listed in Table 2.
### II.3 Dynamical study of the $`KN`$ phase shifts
With all parameters determined in the extended chiral SU(3) quark model, the $`KN`$ phase shifts can be dynamically studied in the frame work of the RGM. The wave function of the five quark system is of the following form:
$`\mathrm{\Psi }=𝒜[\widehat{\varphi }_N(𝝃_1,𝝃_2)\widehat{\varphi }_K(𝝃_3)\chi (𝑹_{NK})],`$ (33)
where $`𝝃_1`$ and $`𝝃_2`$ are the internal coordinates for the cluster $`N`$, and $`𝝃_3`$ the internal coordinate for the cluster $`K`$. $`𝑹_{NK}𝑹_N𝑹_K`$ is the relative coordinate between the two clusters, $`N`$ and $`K`$. The $`\widehat{\varphi }_N`$ is the antisymmetrized internal cluster wave function of $`N`$, and $`\chi (𝑹_{NK})`$ the relative wave function of the two clusters. The symbol $`𝒜`$ is the antisymmetrizing operator defined as
$$𝒜1\underset{iN}{}P_{i4}13P_{34}.$$
(34)
Substituting $`\mathrm{\Psi }`$ into the projection equation
$$\delta \mathrm{\Psi }|(HE)|\mathrm{\Psi }=0,$$
(35)
we obtain the coupled integro-differential equation for the relative function $`\chi `$ as
$`{\displaystyle \left[(𝑹,𝑹^{})E𝒩(𝑹,𝑹^{})\right]\chi (𝑹^{})𝑑𝑹^{}}=0,`$ (36)
where the Hamiltonian kernel $``$ and normalization kernel $`𝒩`$ can, respectively, be calculated by
$`\left\{\begin{array}{c}(𝑹,𝑹^{})\\ 𝒩(𝑹,𝑹^{})\end{array}\right\}=[\widehat{\varphi }_N(𝝃_1,𝝃_2)\widehat{\varphi }_K(𝝃_3)]\delta (𝑹𝑹_{NK})\left|\left\{\begin{array}{c}H\\ 1\end{array}\right\}\right|`$ (41)
$`𝒜\left[[\widehat{\varphi }_N(𝝃_1,𝝃_2)\widehat{\varphi }_K(𝝃_3)]\delta (𝑹^{}𝑹_{NK})\right].`$ (42)
Eq. $`(\text{36})`$ is the so-called coupled-channel RGM equation. Expanding unknown $`\chi (𝑹_{NK})`$ by employing well-defined basis wave functions, such as Gaussian functions, one can solve the coupled-channel RGM equation for a bound-state problem or a scattering one to obtain the binding energy or scattering phase shifts for the two-cluster systems. The details of solving the RGM equation can be found in Refs. kwi77 ; mka77 ; mok81 ; fhuang04nk .
## III Results and discussions
In the extended chiral SU(3) quark model, the coupling of quarks and vector-meson field is considered, and thus the coupling constants of OGE are greatly reduced by fitting the mass difference between $`N`$, $`\mathrm{\Delta }`$ and $`\mathrm{\Lambda }`$, $`\mathrm{\Sigma }`$. From Table 1, one can see that for both set I and set II, $`g_u^2=0.0748`$ and $`g_s^2=0.0001`$, which are much smaller than the values of the original chiral SU(3) quark model ($`g_u^2=0.7704`$ and $`g_s^2=0.5525`$). This means that the OGE, which plays an important role of the $`KN`$ short-range interaction in the original chiral SU(3) quark model, is now nearly replaced by the vector-meson exchanges. In other words, in the $`KN`$ system the mechanisms of the quark-quark short-range interactions of these two models are totally different.
A RGM dynamical calculation of the $`S`$-, $`P`$-, $`D`$-, and $`F`$-wave $`KN`$ phase shifts with isospin $`I=0`$ and $`I=1`$ is performed, and the numerical results are shown in Figs. 1–4. Here we use the conventional partial wave notation: the first subscript denotes the isospin quantum number and the second one twice of the total angular momentum of the $`KN`$ system. For comparison the phase shifts calculated in the original chiral SU(3) quark model are also shown in these figures.
Let’s first concentrate on the $`S`$-wave results (Fig. 1). In a previous quark model study bsi97 where the $`\pi `$ and $`\sigma `$ boson exchanges as well as the OGE and confining potential are taken as the quark-quark interaction, the authors concluded that a consistent description of the $`S`$-wave $`KN`$ phase shifts in both isospin $`I=0`$ and $`I=1`$ channels simultaneously is not possible. Another recent work in a constituent quark model based on the RGM calculation gave an opposite sign of the $`S_{01}`$ channel phase shifts sle03 . From Fig. 1 one can see that we obtain a successful description of the $`S_{01}`$ channel phase shifts, and for the $`S_{11}`$ partial wave, similar to that obtained in the original chiral SU(3) model, the trend of the theoretical phase shifts is also in agreement with the experiment. Since there is no contribution coming from the spin-orbit coupling in the $`S`$-wave, only the central force of the quark-quark interaction can enter in the scattering process, thus it plays a dominantly important role. To understand the contributions of various chiral fields to the $`KN`$ interaction, in Fig. 5 we show the central force diagonal matrix elements of the generator coordinate method (GCM) calculation kwi77 of the $`\sigma `$, $`a_0`$, $`\pi `$, $`\rho `$, and $`\omega `$ boson exchanges in the extended chiral SU(3) quark model, which can describe the interaction between two clusters $`N`$ and $`K`$ qualitatively. In Fig. 5, $`s`$ denotes the generator coordinate and $`V(s)`$ is the effective boson-exchange potential between the two clusters. Form this figure we can see that the $`\sigma `$ exchange always offers attraction and $`\omega `$ exchange offers repulsion in both isospin $`I=0`$ and $`I=1`$ channels. This is reasonable since the $`\sigma `$ and $`\omega `$ exchanges are isospin independent. Contrarily, the $`a_0`$, $`\pi `$, and $`\rho `$ exchanges are isospin dependent. In the $`S_{01}`$ partial wave the $`a_0`$ exchange offers repulsive and $`\rho `$ exchange offers a little attractive, while in the $`S_{11}`$ partial wave the $`a_0`$ exchange offers a little attraction and $`\rho `$ exchange offers repulsion. In both of these two channels the $`\pi `$ exchange, existing due to the quark exchange required by the Pauli principle, always offers much strong repulsion though the repulsion strength is different. This means that the one-pion exchange is important and cannot be neglected on the quark level, which is quite different from the works on the hadron level where the one-pion exchange is absent in the $`KN`$ interaction.
Now look at the $`P`$-wave $`KN`$ phase shifts (Fig. 2). The results for the $`P_{13}`$ channel, which are too repulsive in the original chiral SU(3) quark model when the laboratory momentum of the kaon meson is greater than 300 MeV, the same case as in Black’s previous work nbl02 , are now much more repulsive in the extended chiral SU(3) quark model. For the other channels the results in both these two models are similar to each other. Comparing with Ref. sle03 , we get correct signs and proper magnitudes of $`P_{11}`$ and $`P_{03}`$ waves in both the extended chiral SU(3) quark model and the original chiral SU(3) quark model.
For higher-angular-momentum partial waves (Figs. 34), the theoretical phase shifts of $`D_{15}`$ and $`F_{17}`$ in the extended chiral SU(3) quark model are improved in comparison with those obtained from the original chiral SU(3) quark model, while in the case of the $`D_{13}`$ the situation is somewhat less satisfying. For the other channels, the trends of the calculated phase shifts in both these two models are all in qualitative agreement with the experiment. Comparing with Ref. sle03 , in both these two models, we can get correct signs of $`D_{13}`$, $`D_{05}`$, $`F_{15}`$, and $`F_{07}`$ waves, and for $`D_{03}`$ and $`D_{15}`$ channels we also obtain a considerable improvement on the theoretical phase shifts in the magnitude.
As discussed in Refs. fhuang04nk ; fhuang04dk , the annihilation interaction is not clear and its influence on the phase shifts should be examined. We omit the annihilation part entirely to see its effect and find that the numerical phase shifts only have very small changes. This is because in the $`KN`$ system the annihilations to gluons and vacuum are forbidden and $`u(d)\overline{s}`$ can only annihilate to $`K`$ and $`K^{}`$ mesons. This annihilation part originating from the $`S`$-channel acts in the very short range, so that it plays a negligible role in the $`KN`$ scattering process.
The other thing we would like to mention is that our results of $`KN`$ phase shifts are independent of the confinement potential in the present one-channel two-color-singlet-cluster calculation. Thus the numerical results will almost remain unchanged even the color quadratic confinement is replaced by the color linear one.
From the above discussion, one sees that though the mechanisms of the quark-quark short-range interactions are totally different in the original chiral SU(3) quark model and the extended chiral SU(3) quark model, the theoretical $`KN`$ phase shifts of $`S`$, $`P`$, $`D`$, and $`F`$ waves in these two models are very similar to each other. Comparing with others’ previous quark model studies, we can obtain a considerable improvement for many channels. However, in the present work the $`P_{13}`$ and $`D_{15}`$ partial waves have not yet been satisfactorily described. In this sense, one can say that the present quark model still has some difficulties to describe the $`KN`$ scattering well enough for all of the partial waves. It should be studied in future work the possibility of that if there are some physical ingredients missing in our quark model investigations, as well as the relativistic effects and the nonelastic channel effects on the $`KN`$ phase shifts.
By the way, to study the short-range quark-quark interaction more extensively, or on the other words, to examine whether the OGE or the vector meson exchange governs the short range interaction between quarks, besides the $`KN`$ systems the $`\overline{K}N`$ is also an interesting case, since there is a close connection of the vector-meson exchanges between the $`KN`$ and $`\overline{K}N`$ interactions due to $`G`$-parity transition. Specially, the repulsive $`\omega `$ exchange changes sign for $`\overline{K}N`$, because of the negative $`G`$ parity of the $`\omega `$ meson, and becomes attractive. However, one should note that the treatment of the $`\overline{K}N`$ channel is more complicated than the $`KN`$ system since it involves $`s`$-channel gluon and vacuum contributions. Still the extension of our chiral quark model to incorporate the gluon and vacuum annihilations in the $`\overline{K}N`$ system would be a very interesting new development. Investigations along this line are planned for the future.
## IV Summary
In this paper, we extend the chiral SU(3) quark model to include the coupling between quarks and vector chiral field. The OGE which dominantly governed the short-range quark-quark interactions in the original chiral SU(3) quark model is now nearly replaced by the vector-meson exchange. Using this model, a dynamical calculation of the $`S`$-, $`P`$-, $`D`$-, and $`F`$-wave $`KN`$ phase shifts is performed in the isospin $`I=0`$ and $`I=1`$ channels by solving a RGM equation. The calculated phase shifts of different partial waves are similar to those given by the original chiral SU(3) quark model. Comparing with Ref. sle03 , a recent RGM calculation in a constituent quark model, we can obtain correct signs of several partial waves and a considerable improvement in the magnitude for many channels. Nevertheless, in the present work we do not obtain a satisfactory improvement for the $`P_{13}`$ and $`D_{15}`$ partial waves, of which the theoretical phase shifts are too much repulsive and attractive respectively when the laboratory momentum of the kaon meson is greater than 300 MeV. Further the effects of the coupling to the inelastic channels and hidden color channels will be considered and the interesting and more complicated $`\overline{K}N`$ system will be investigated in future work.
###### Acknowledgements.
This work was supported in part by the National Natural Science Foundation of China, Grant No. 10475087. |
warning/0507/hep-ph0507285.html | ar5iv | text | # 1 Introduction and summary of the results
## 1 Introduction and summary of the results
A long-standing problem in particle physics is the understanding of strong interactions at low energies. While at very low energies, of the order of the hadronic scale $`\mathrm{\Lambda }300`$ MeV, perturbative QCD is of no use and alternative methods have been developed in decades (such as quark models, chiral lagrangians, lattice QCD, etc.), at intermediate energies, of the order of a few GeV, perturbative computations can be combined with non-perturbative models to predict a variety of cross sections and decay rates. Among these moderate hard scale phenomena is beauty physics, which is indeed characterized by a hard scale of a few GeV. The measured decay spectra often receive large contributions at the endpoints — in the case of the hadron energy spectrum, in the middle of the domain — from long-distance effects related to soft interactions between the heavy quark and the light degrees of freedom.
The main non perturbative effect is the well-known Fermi motion, which classically can be described as a small vibration of the heavy quark inside the $`B`$ meson because of the momentum exchange with the valence quark; in the quantum theory it is also the virtuality of the heavy quark that matters. This effect is important in the end-point region, because it produces some smearing of the partonic spectra.
These long distance effects manifest themselves in perturbation theory in the form of series of large infrared logarithms, coming from an “incomplete” cancellation of infrared divergencies in real and virtual diagrams. The probability for instance for a light quark produced in a process with a hard scale $`Q`$ to evolve into a jet with an invariant mass smaller than $`m`$ is written in leading order as :
$`J(m)`$ $`=`$ $`1+A_1\alpha _S{\displaystyle _0^1}{\displaystyle \frac{d\omega }{\omega }}{\displaystyle _0^1}{\displaystyle \frac{d\theta ^2}{\theta ^2}}\mathrm{\Theta }\left({\displaystyle \frac{m^2}{Q^2}}\omega \theta ^2\right)A_1\alpha _S{\displaystyle _0^1}{\displaystyle \frac{d\omega }{\omega }}{\displaystyle _0^1}{\displaystyle \frac{d\theta ^2}{\theta ^2}}`$ (1)
$`=`$ $`1{\displaystyle \frac{A_1}{2}}\alpha _S\mathrm{log}^2\left({\displaystyle \frac{Q^2}{m^2}}\right),`$
where $`\omega `$ is the energy of a gluon emitted by the light quark normalized to the hard scale, $`\theta `$ is its emission angle and $`A_1`$ is a positive constant (see sec. 3). The first integral on the r.h.s. is the real contribution while the second integral is the virtual one. Both integrals are separately divergent for $`\omega =0`$ — soft singularity — as well as for $`\theta =0`$ — collinear singularity, but their sum is finite. “Complete” real-virtual cancellation occurs only for $`m=Q`$, i.e. in the completely inclusive evolution of the quark line, while for $`m<Q`$ there is a left-over double logarithm because of the smaller integration region of the real diagrams. Multiple gluon emission occurs in higher orders of perturbation theory; it can be described as a classical branching process and gives rise to the double logarithmic series, i.e. to powers of the last term $`\alpha _S\mathrm{log}^2\left(Q^2/m^2\right)`$ on the r.h.s. of eq. (1) .
We may say that perturbation theory “signals” long-distance effects in a specific way — even though a quantitative description of the latter has to include also some truly non-perturbative component. A theoretical study of the universality of these long-distance effects can therefore be done inside perturbation theory, by comparing the logarithmic structure of different distributions. In other words, if these long-distance effects are universal, this has certainly to show up in perturbation theory: things have to work in perturbation theory first. The aim of this work is to study the relation of long-distance effects between different distributions by means of resummed perturbation theory.
In general, let us consider the semi-inclusive decays
$$BX_q+(\mathrm{non}\mathrm{QCD}\mathrm{partons}),$$
(2)
where $`X_q`$ is any hadronic final state coming from the fragmentation of the light quark $`q=u,d,s`$ and the non QCD partons are typically a photon, a lepton-neutrino pair, a lepton-antilepton pair, etc. This system of particle(s), with total four-momentum $`q_\mu `$, constitutes a “probe” for the hadronic process, as in the case of deep-inelastic-scattering (DIS) of leptons off hadrons. Without any generality loss, we can work in the $`b`$ rest frame, where $`p_b^\mu =m_bv^\mu `$, with $`m_b`$ being the beauty mass and $`v^\mu =(1;0,0,0)`$ being the classical 4-velocity. The hadronic subprocess in (2) is characterized by the following three scales:
$$m_b,E_X\mathrm{and}m_X(m_bE_X),$$
(3)
where $`m_X`$ and $`E_X`$ are the invariant mass and the total energy of the final hadronic state $`X_q`$, respectively. We are interested in the so-called threshold region, which can be defined in all generality as the one having
$$m_XE_X.$$
(4)
The region (4) is sometimes called radiation-inhibited, because the emitted radiation naturally produces final states with an invariant mass of the order of the hard scale: $`m_XO(E_X)`$. It is also called semi-inclusive because experimentally, to satisfy the constraint (4), most hadronic final states have to be discarded.
The processes we are going to consider are the well-known radiative decay with a real photon in the final state,
$$BX_s+\gamma $$
(5)
and the semi-leptonic decay, <sup>4</sup><sup>4</sup>4 The results for the semileptonic decay are easily extended to the radiative decay with the photon converting into a lepton pair,
$$BX_s+l+\overline{l}.$$
(6)
$$BX_u+l+\nu .$$
(7)
In perturbative QCD, the hadronic subprocess in (2) consists of a heavy quark decaying into a light quark which evolves later into a jet of soft and collinear partons because of infrared divergencies. In leading order, one only considers the emission of soft gluons at small angle by the light quark (see eq. (1)); the final state $`X_q`$ consists of a jet with the leading (i.e. most energetic) quark $`q`$ originating the jet itself. In next-to-leading order one has to take into account two different single-logarithmic effects: $`(a)`$ hard emission at small angle by the light quark $`q`$ and $`(b)`$ soft emission at large angle by the heavy quark. Because of $`(a)`$, the final state consists of a jet with many hard partons and, in general, the leading parton is no longer the quark $`q`$ which originated the jet itself. Because of $`(b)`$, the final state does not contain only an isolated jet, but also soft partons in any space direction. The main result of is that the large threshold logarithms appearing in (2) are conveniently organized as a series of the form:
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \underset{k=1}{\overset{2n}{}}}c_{nk}\alpha ^n(Q)\mathrm{log}^k{\displaystyle \frac{Q^2}{m_X^2}}`$ (8)
$`=`$ $`c_{12}\alpha (Q)\mathrm{log}^2{\displaystyle \frac{Q^2}{m_X^2}}+c_{11}\alpha (Q)\mathrm{log}{\displaystyle \frac{Q^2}{m_X^2}}+c_{24}\alpha ^2(Q)\mathrm{log}^4{\displaystyle \frac{Q^2}{m_X^2}}+c_{23}\alpha ^2(Q)\mathrm{log}^3{\displaystyle \frac{Q^2}{m_X^2}}+\mathrm{}\mathrm{},`$
where $`\alpha (Q)=\alpha _S(Q)`$ is the QCD coupling and the hard scale $`Q`$ is determined by the final hadronic energy $`E_X`$ <sup>5</sup><sup>5</sup>5 The factor two is inserted in such a way that the hard scale coincides with $`m_b`$ in the radiative decay (see later). The essential point however is that $`Q`$ is proportional to $`E_X`$ via a proportionality constant of order one, whose precise value is irrelevant.:
$$Q=\mathrm{\hspace{0.17em}2}E_X.$$
(9)
These large logarithms are factorized into a universal QCD form factor. Let us summarize the derivation of (8) and (9). We take the infinite mass limit for the beauty quark while keeping the hadronic energy and the hadronic mass fixed <sup>6</sup><sup>6</sup>6This limit has not to be confused with that one relevant for the shape function, also called structure function of the heavy flavors, which is $`E_X\mathrm{}`$, $`m_X\mathrm{}`$ with $`m_X^2/E_X\mathrm{const}`$ (the latter implies $`m_b\mathrm{}`$, but the converse is not true). The shape function describes soft interactions only and therefore does not factorize the whole logarithmic structure, missing the large logarithms coming from hard collinear emission off the light quark .:
$$m_b\mathrm{},\mathrm{with}E_X\mathrm{and}m_X\mathrm{const}.$$
(10)
This takes us into an effective theory in which the beauty quark is replaced by a static quark, as recoil effects are neglected in the limit (10). If we write the beauty quark momentum as $`p_b=m_bv+k`$, where $`k`$ is a soft momentum, the infinite mass limit of the propagator is easily obtained as:
$$S_F(p)=\left(\frac{1+\widehat{v}}{2}+\frac{\widehat{k}}{2m}\right)\frac{1}{vk+k^2/(2m)+iϵ}\frac{1+\widehat{v}}{2}\frac{1}{vk+iϵ}(\mathrm{static}\mathrm{limit}),$$
(11)
where $`\widehat{a}\gamma _\mu a^\mu `$. As discussed above, the beauty quark contributes to the QCD form factor via large logarithms coming from soft emissions, which are correctly described by a static quark. Since the light quark propagator is not touched by the limit (10), we conclude that all soft and/or collinear emissions are correctly described by this limit. Since the heavy flavor mass has disappeared with the limit (10), the only remaining scales in the hadronic subprocess are $`m_X`$ and $`E_X`$. Only one adimensional quantity can be constructed out of them, for example the ratio $`E_X/m_X`$, which is therefore the only possible argument for the large logarithms, in agreement with (8). Furthermore, the hard scale $`Q`$ is given by the greatest scale in the game, i.e. by the hadronic energy $`E_X`$, in agreement with (9).
The argument given above, however, is not rigorous: let us refine it. The limit (10) is indeed singular in quantum field theory: one cannot remove degrees of freedom without paying some price. Let us consider for simplicity’s sake the semileptonic decay (7), even though the conclusions are general. The vector and axial-vector currents responsible for the $`bu`$ transition are conserved or partially conserved in QCD, implying that the $`O(\alpha )`$ virtual corrections are ultraviolet finite. These corrections contain however terms of the form
$$\gamma _0\alpha \mathrm{log}\frac{m_b}{E_X}(\mathrm{ordinary}\mathrm{QCD}),$$
(12)
which diverge in the limit (10) ($`\gamma _0`$ is a constant). If one takes the limit (10) ab initio, i.e. before integrating the loop, some divergence is expected in the loop integrals, as it is indeed the case. Technically, that occurs because the static propagator is of the form $`1/(k_0+iϵ)`$ (see eq. (11)) and, unlike the ordinary propagator, has no damping for $`|\stackrel{}{k}|\mathrm{}`$. It can be shown that the $`bu`$ vector and axial-vector currents are no more conserved or partially conserved in the static theory. Therefore, unlike the QCD case, the $`O(\alpha )`$ virtual corrections are ultraviolet divergent in the static theory and produce, after renormalization, terms corresponding to (12) of the form
$$\gamma _0\alpha \mathrm{log}\frac{\mu }{E_X}(\mathrm{effective}\mathrm{theory}),$$
(13)
in which basically the heavy flavor mass $`m_b`$ is replaced by the renormalization point $`\mu `$ — the coefficient $`\gamma _0`$ being the same. The hadronic subprocess in the static theory therefore has amplitudes depending on the physical scales $`E_X`$ and $`m_X`$ as well as on the renormalization scale $`\mu `$. If we neglect terms suppressed by inverse powers of the beauty mass $`1/m_b^n`$, we have that the physical scale $`m_b`$ is replaced by the renormalization point $`\mu `$ in the effective theory: $`m_b\mu `$. The effective currents $`\stackrel{~}{J}_\nu `$ and the coupling constant $`\alpha `$ are renormalized at the scale $`\mu `$: $`\stackrel{~}{J}_\nu =\stackrel{~}{J}_\nu (\mu )`$ and $`\alpha =\alpha (\mu )`$. The effective amplitudes contain terms of the form $`\alpha ^n\mathrm{log}^k\mu /E_X`$ $`(kn)`$, which are large logarithms for $`\mu E_X`$ or $`\mu E_X`$. To have convergence of the perturbative series, the large logarithms above must be resummed by taking $`\mu =O(E_X)`$, i.e. $`\mu =kE_X`$ with $`k=O(1)`$. This implies that the effective currents and the coupling are evaluated at a scale of the order of the hadronic energy: $`\stackrel{~}{J}_\nu =\stackrel{~}{J}_\nu (kE_X)`$ and $`\alpha =\alpha (kE_X)`$. We have therefore proved that the hard scale $`Q`$ is fixed by the final hadronic energy $`E_X`$ and not by the beauty mass $`m_b`$:
$$Q=\mu =kE_X\mathrm{with}k=O(1).$$
(14)
Let us go back to the general process (2). Kinematics gives:
$$2E_X=m_b\left(1\frac{q^2}{m_b^2}+\frac{m_X^2}{m_b^2}\right).$$
(15)
The simplest processes are those with a light-like probe, i.e. with $`q^2=0`$, where
$$2E_X=m_b\left(1+\frac{m_X^2}{m_b^2}\right)m_b.$$
(16)
This case corresponds to the radiative decay (5). In this case, the final hadronic energy is always large and of the order of the heavy-flavor mass:
$$Qm_b(\mathrm{radiative}\mathrm{decay}).$$
(17)
On the other hand, in the semi-leptonic decay (7) <sup>7</sup><sup>7</sup>7 The same is also true for the radiative decay with the photon converting into a lepton pair (6)., the lepton pair can have a large invariant mass,
$$q^2O\left(m_b^2\right),$$
(18)
implying a substantial reduction of the hard scale:
$$Qm_b.$$
(19)
This fact is one of the complications in the threshold resummation of the semileptonic decay spectra: while in the radiative decay (5), the hard scale $`Q`$ is always large in the threshold region, and of the order of $`m_b`$, this is no longer true in the semileptonic decay. The hadronic subprocesses have in general different hard scales in the two decays. If one integrates over $`q^2`$, for example because of undetected neutrino momentum, there is a mixing of hadronic contributions with different hard scales in the semileptonic case. However, it turns out by explicit computation that the contributions from a large $`q^2`$, i.e. with a small hard scale in the hadronic subprocess, are rather suppressed (see sec. 4).
At fixed $`Q`$, the large logarithms in (2) can be factorized into a QCD form factor, which is universal in the sense that it depends only on the hadronic subprocess. The differences between, let us say, the radiative decay (5) and the semileptonic decay (7) only enter in the specific form of a short-distance coefficient function multiplying the QCD form factor (and in the form of a remainder function collecting non factorized, small contributions, see next section).
The discussion above can be summarized as follows. The hard scale $`Q=2E_X`$ in (2) appears in the argument in the infrared logarithms as well as in the argument of the running coupling. In the radiative decay, because of kinematics, the hard scale is always large and of the order of the beauty mass: $`Qm_b`$, while in the semileptonic case kinematical configurations are possible with $`Qm_b`$ as well as with $`Qm_b`$. The main complication in semileptonic decays is that by performing kinematical integrations (for example over the neutrino energy), one may integrate over the hard scale of the hadronic subprocess. While in radiative decays the hard scale is fixed, in the semileptonic decays there can be a mixing of different hadronic subprocesses. A non-trivial picture of some semileptonic decay spectra emerges: there are long-distance effects which cannot be extracted by the radiative decay, related to a small final hadronic energy, but their effect turns out to be small at the end because of a kinematical suppression of the states with a small hard scale. The decay spectra in (7) can therefore be divided into two classes:
1. distributions in which the hadronic energy $`E_X`$ is not integrated over. These distributions can be related via short-distance coefficients to the photon spectrum in the radiative decay (5). In particular, the structure of the threshold logarithms is the same as in decay (5). In this paper we restrict ourselves to these simpler distributions;
2. distributions in which the hadronic energy is integrated over and therefore all the hadronic energies contribute. These are for instance the hadron mass distribution or the charged lepton energy distribution. In all these cases, the structure of the threshold logarithms is different from that one in (5) and by far more complicated. The analysis of some of these distributions, which present novel features with respect to $`BX_s\gamma `$, is given in .
Let us make a simple analogy with $`e^+e^{}`$ annihilation into hadrons. In the center-of-mass (c.o.m.) frame, the final state consists of a $`q\overline{q}`$ pair, which are emitted back to back with a high virtuality and evolve later into two jets:
$$e^++e^{}q+\overline{q}J_q+J_{\overline{q}}.$$
(20)
Roughly speaking, the final state $`X_q`$ in (2), consisting in a single jet, is “half” of that in (20), consisting of the two jets $`J_q`$ and $`J_{\overline{q}}`$. Deviations from this independent fragmentation picture arise in next-to-leading order because of large-angle soft emission by the heavy quark in (2), which has no analogue in (20). The structure of $`e^+e^{}`$ hadronic final states is conveniently analyzed by means of so-called shape variables, one of the most studied being the heavy jet mass $`m_H^2`$, defined as
$$m_H^2=\mathrm{max}\{m_R^2,m_L^2\},$$
(21)
where $`m_R`$ and $`m_L`$ are the invariant masses of the particles in the right and left hemispheres of the event respectively. The hemispheres are defined cutting the space with a plane orthogonal to the thrust axis $`\stackrel{}{n}`$, the latter defined as the direction maximizing
$$\underset{i}{}|\stackrel{}{p}_i\widehat{n}|,$$
(22)
i.e. basically the sum of length of longitudinal momenta. The sum extends over all hadrons — partons in the perturbative computation. For $`m_HQ`$, where $`Q`$ is the hard scale to be identified here with the c.o.m. energy, hard emission at large angle by the $`q\overline{q}`$ pair cannot occur and the final state consists of two narrow jets around the original $`q\overline{q}`$ direction, which can be identified with $`\stackrel{}{n}`$. The $`O(\alpha )`$ computation gives large logarithms of similar form to those in (8) :
$$\alpha (Q)\mathrm{log}^2\frac{Q^2}{m_H^2}\mathrm{and}\alpha (Q)\mathrm{log}\frac{Q^2}{m_H^2}.$$
(23)
There is not a simple relation between, let us say, the heavy jet mass distribution at the $`Z^0`$ peak,
$$\frac{d\sigma }{dm_H^2}\left(Q=m_Z\right)$$
(24)
and the integral of this quantity over $`Q`$ from a small energy $`ϵm_H`$ up to $`m_Z`$ with some weight function $`\varphi (Q)`$:
$$\frac{d\widehat{\sigma }}{dm_H^2}=_ϵ^{m_Z}𝑑Q\varphi (Q)\frac{d\sigma }{dm_H^2}\left(Q\right).$$
(25)
Radiative $`B`$ decays (5) and semileptonic spectra (7) in class 1. are the analog of the former distribution (24), while semileptonic spectra in class 2. are the analog of the latter case (25). The analog of the suppression in the semileptonic spectra 2. of the contributions from large $`q^2`$ is the suppression of the weight function $`\varphi (Q)`$ for $`Qm_Z`$.
Many properties of the distributions we are going to derive in this work can be understood with a qualitative discussion on the hadron energy spectrum,
$$\frac{d\mathrm{\Gamma }}{dE_X},$$
(26)
which exhibits a remarkable phenomenon related to the occurrence of infrared singularities inside the physical domain, instead than at the boundary as it is usually the case. This phenomenon has been studied in the framework of jet physics and is known as the “Sudakov shoulder” . Let us discuss it in the present case in physical terms. In lowest order, the semileptonic decay (7) involves three massless partons in the final state:
$$bu+l+\nu .$$
(27)
According to kinematics, any final state parton can take at most half of the initial energy, implying that
$$E_X^{(0)}=E_u\frac{m_b}{2}.$$
(28)
In lowest order, the final hadronic state consists indeed of the $`up`$ quark only: $`X_u=u`$. To order $`\alpha `$, a real gluon is radiated and the final hadronic state is a two-particle system: $`X_u=u+g`$. The final hadronic energy is not restricted anymore to half the beauty mass but can go up to the whole beauty mass:
$$E_X^{(1)}=E_u+E_gm_b.$$
(29)
For example, just consider an energetic up quark recoiling against the gluon, with a soft electron and a soft neutrino. The relevant case for us is a final state with the $`up`$ quark of energy $`m_b/2`$ and a soft and/or a collinear gluon. Such a state has a total energy slightly above $`m_b/2`$ and the matrix element is logarithmically enhanced because of the well-known infrared singularities. Such logarithmic enhancement cannot be cancelled by the $`O(\alpha )`$ virtual corrections, because of their tree-level kinematical limitation (28). We are left therefore with large infrared logarithms, of the form
$$\alpha \mathrm{log}^2\left(E_X\frac{m_b}{2}\right),\alpha \mathrm{log}\left(E_X\frac{m_b}{2}\right)\left(E_X\frac{m_b}{2}\right),$$
(30)
which are final and produce an infrared divergence for $`E_X+m_b/2`$. On the other hand, for $`E_X<m_b/2`$ there are no large logarithms of the form $`\alpha \mathrm{log}^k(m_b/2E_X)`$ $`(k=1,2)`$, because in this case real-virtual cancellation may occur, and it actually does. Let us summarize: the $`O(\alpha )`$ spectrum has an infrared singularity right in the middle of the domain, for $`\overline{E}_X=m_b/2`$, because the lowest order spectrum has a discontinuity in this point, above which it vanishes identically because of kinematics.
This infrared singularity is integrable, as
$$_{m_b/2}^{m_b/2+\delta }𝑑E_X\alpha \mathrm{log}^k\left(E_X\frac{m_b}{2}\right)<\mathrm{},$$
(31)
where $`\delta >0`$ is some energy-resolution parameter. The infrared divergence is therefore eliminated with some smearing over the hadronic energy, which experimentally is always the case. Furthermore, hadronization corrections certainly produce some smearing on the partonic final states because of parton recombination. In other words, non-perturbative mechanisms wash out this infrared divergence, which therefore does not present any problem of principle. As we are going to show, however, perturbation theory “saves itself” and no mechanism outside perturbation theory is needed to have a consistent prediction: resummation of the infrared logarithms in (30) to all orders completely eliminates the singularity, as in the cases of the usual infrared divergencies . Since large logarithms occur for $`E_Xm_b/2`$, we have that the hard scale is given for this spectrum by the beauty mass,
$$Q=m_b(\mathrm{hadron}\mathrm{energy}\mathrm{spectrum}),$$
(32)
just like in radiative decays. This equality is noticeable, as it comes from completely independent kinematics with respect to the one in (5). There is therefore a pure short-distance relation between the hadron mass distribution in (5) and the hadron energy distribution in (7). This property remains true when we consider non-perturbative Fermi motion effects, which are factorized by the well-known structure function of the heavy flavors, also called the shape function.
This paper is organized as follows:
In sec. (2) we presents the results for the resummed triple-differential distribution, which is the most general distribution and the starting point of our analysis;
In sec. (3) we review the theory of threshold resummation in heavy flavor decays, giving explicit formulas for the QCD form factor in next-to-next-to-leading order (NNLO). The transformation to Mellin space in order to solve the kinematical constraints for multiple soft emission is discussed, together with the inverse transform to the original momentum space;
In sec. (4) we derive the double distribution in the hadronic energy and in the ratio (hadronic mass)/(hadronic energy), which are the most convenient variables for threshold resummation (these are the variables $`w`$ and $`u`$ defined there). The distribution in any hadronic variable can be obtained from this distribution by integration;
In sec. (5) we present the results for the resummed hadron energy spectrum in next-to-leading order, whose main physical properties have already been anticipated here. We also compute the average hadronic energy to first order and compare with the radiative decay. The hadron energy spectrum with an upper cutoff on the hadron mass, which is the easiest thing to measure in experiments, is derived in leading order;
In sec. (6) we derive the double distribution in the hadron and electron energies, i.e. in the two independent energies. A peculiarity of this spectrum is that it is characterized by the presence of two different series of large logarithms, which are factorized by two different QCD form factors. Another peculiarity is that this double differential distribution contains partially-integrated QCD form factors instead of differential ones. That implies that the infrared singularities occurring in this distribution are integrable, as in the case of the Sudakov shoulder which we have discussed before;
Finally, in sec. (7) we present our conclusions together with a discussion about natural developments.
## 2 Triple differential distribution
The triple differential distribution in the decay (7) is the starting point of our analysis. It has a resummed expression of the form :<sup>8</sup><sup>8</sup>8 We have normalized the distribution to the radiatively-corrected total semileptonic width $`\mathrm{\Gamma }=\mathrm{\Gamma }_0\left[1+\alpha C_F/\pi \left(25/8\pi ^2/2\right)+O(\alpha ^2)\right]`$ and not to the Born width $`\mathrm{\Gamma }^{(0)}`$, as originally done in . We consider it to be a better choice because $`\mathrm{\Gamma }`$, unlike $`\mathrm{\Gamma }^{(0)}`$, is a physical quantity, directly measurable in the experiments and we are not interested in the prediction of total rates, but only in how a given rate distributes among different hadronic channels.
$$\frac{1}{\mathrm{\Gamma }}\frac{d^3\mathrm{\Gamma }}{dxdudw}=C[x,w;\alpha (wm_b)]\sigma [u;\alpha (wm_b)]+d[x,u,w;\alpha (wm_b)],$$
(33)
where we have defined the following kinematical variables:
$$w=\frac{2E_X}{m_b}(0w2),x=\frac{2E_l}{m_b}(0x1)$$
(34)
and <sup>9</sup><sup>9</sup>9Note that a similar variable simplifies two-loop computations with heavy quarks .
$$u=\frac{E_X\sqrt{E_X^2m_X^2}}{E_X+\sqrt{E_X^2m_X^2}}(0u1).$$
(35)
It is convenient to express $`u`$ as:
$$u=\frac{1\sqrt{14y}}{1+\sqrt{14y}},$$
(36)
with
$$y=\frac{m_X^2}{4E_X^2}(0y1/4).$$
(37)
The inverse formula of (36) reads:
$$y=\frac{u}{(1+u)^2}.$$
(38)
The functions entering the r.h.s. of eq. (33) are:
* $`C[x,w;\alpha (wm_b)]`$, a short-distance, process-dependent coefficient function, whose explicit expression will be given later. It depends on two independent energies $`x`$ and $`w`$ and on the QCD coupling $`\alpha `$;
* $`\sigma [u;\alpha (wm_b)]`$, a process-independent, long-distance dominated, QCD form factor. It factorizes the threshold logarithms appearing in the perturbative expansion. At order $`\alpha `$:
$$\sigma (u;\alpha )=\delta (u)\frac{C_F\alpha }{\pi }\left(\frac{\mathrm{log}u}{u}\right)_+\frac{7C_F\alpha }{4\pi }\left(\frac{1}{u}\right)_++O(\alpha ^2),$$
(39)
where $`C_F`$ is the Casimir of the fundamental representation of $`SU(3)_c`$, $`C_F=(N_c^21)/(2N_c)`$ with $`N_c=3`$ (the number of colors) and the plus distributions are defined as usual as:
$$P(u)_+=P(u)\delta (u)_0^1𝑑u^{}P(u^{}).$$
(40)
The action on a test function $`f(u)`$ is therefore:
$$_0^1𝑑uP(u)_+f(u)=_0^1𝑑uP(u)[f(u)f(0)].$$
(41)
The plus-distributions are sometimes called star-distributions and can also be defined as limits of ordinary functions as:
$`P(u)_+`$ $`=`$ $`\underset{ϵ0^+}{lim}\left[\theta (uϵ)P(u)\delta (u){\displaystyle _ϵ^1}𝑑u^{}P(u^{})\right]`$ (42)
$`=`$ $`\underset{ϵ0^+}{lim}\left[\theta (uϵ)P(u)\delta (uϵ){\displaystyle _ϵ^1}𝑑u^{}P(u^{})\right]`$
$`=`$ $`\underset{ϵ0^+}{lim}{\displaystyle \frac{d}{du}}\left[\theta (uϵ){\displaystyle _u^1}𝑑u^{}P(u^{})\right].`$
An important property of the plus-distributions is that their integral on the unit interval vanishes:
$$_0^1P(u)_+𝑑u=\mathrm{\hspace{0.17em}0}.$$
(43)
We have assumed a minimal factorization scheme in eq. (39), in which only terms containing plus-distributions are included in the form factor. The resummation of the logarithmically enhanced terms in $`\sigma `$ to all orders in perturbation theory will be discussed in the next section;
* $`d(x,u,w;\alpha )`$ is a short-distance, process-dependent, remainder function, not containing large logarithms. Formally, it can have at most an integrable singularity for $`u+\mathrm{\hspace{0.17em}0}`$, i.e. we require that:
$$\underset{u+0}{lim}_0^u𝑑u^{}d(x,w,u^{};\alpha )=\mathrm{\hspace{0.17em}0}.$$
(44)
This term is added to $`C\sigma `$ in order to correctly describe the region $`uO(1)`$ and to reproduce the total rate. It depends on all the kinematical variables $`x,w`$ and $`u`$ and the explicit expression will be given later.
Eq. (33) is a generalization of the threshold resummation formula for the radiative decay in (5) :
$$\frac{1}{\mathrm{\Gamma }_R}\frac{d\mathrm{\Gamma }_R}{dt_s}=C_R\left[\alpha (wm_b)\right]\sigma [t_s;\alpha (wm_b)]+d_R[t_s;\alpha (wm_b)],$$
(45)
where<sup>10</sup><sup>10</sup>10 The relation with the photon energy $`x_\gamma =2E_\gamma /m_b`$ is $`t_s=1x_\gamma `$.
$$t_s=\frac{m_{X_s}^2}{m_b^2}.$$
(46)
In this simpler case, the coefficient function $`C_R(\alpha )`$ does not depend on any kinematical variable but only on the QCD coupling $`\alpha `$ and has an expansion of the form<sup>11</sup><sup>11</sup>11 We perform expansions in powers of $`\alpha `$, while the traditional expansion is in powers of $`\alpha /(2\pi )`$.:
$$C_R(\alpha )=\mathrm{\hspace{0.17em}1}+\alpha C_R^{(1)}+\alpha ^2C_R^{(2)}+O(\alpha ^3),$$
(47)
where $`C_R^{(i)}`$ are numerical coefficients. Basically, going from the 2-body decay (5) to the 3-body decay (7), the coefficient function acquires a dependence on the additional kinematical variables, namely two energies. The remainder function in eq. (45) depends on the (unique) variable $`t_s`$ and has an expansion of the form:
$$d_R(t_s;\alpha )=\alpha d_R^{(1)}(t_s)+\alpha ^2d_R^{(2)}(t_s)+O(\alpha ^3).$$
(48)
The main point is that the QCD form factor $`\sigma `$ in the same in both distributions (33) and (45), explicitly showing universality of long-distance effects in the two different decays. By universality we mean that we have the same function, evaluated at the argument $`u`$ in the semileptonic case and at $`t_s`$ in the radiative decay. This property is not explicit in the original formulation , in which the form factors differ in subleading order (see next section).
Since, as shown in the introduction, $`w1`$ in the radiative decay, we can make everywhere in eq. (45) the replacement
$$\alpha (wm_b)\alpha (m_b)(\mathrm{radiative}\mathrm{case}\mathrm{only}),$$
(49)
to obtain:
$$\frac{1}{\mathrm{\Gamma }_R}\frac{d\mathrm{\Gamma }_R}{dt_s}=C_R\left[\alpha (m_b)\right]\sigma [t_s;\alpha (m_b)]+d_R[t_s;\alpha (m_b)].$$
(50)
The distribution contains now a constant coupling, independent on the kinematics $`\alpha (m_b)0.22`$. The replacement (49) cannot be done in the semileptonic case.
In the triple differential distribution was originally given in terms of the variable $`y`$ instead of $`u`$, with the latter $`u=1\xi `$ being introduced in . The variables $`u`$ and $`y`$ coincide in the threshold region in leading twist, i.e. at leading order in $`u`$ in the expansion for $`u0`$, as $`y=u+O(u^2)`$. Going from the variable $`y`$ to the variable $`u`$ only modifies the remainder function. The advantages of $`u`$ over $`y`$ are both technical and physical:
* $`u`$ has, unlike $`y`$, unitary range;
* when we impose the kinematical relation between hadronic energy $`E_{X_s}`$ and hadronic mass $`m_{X_s}`$ of the radiative decay (5), $`u`$ exactly equals $`t_s`$:
$$u|_{E_{X_s}=m_b/2(1+m_{X_s}^2/m_b^2)}=t_s.$$
(51)
This property suggests that some higher-twist effects may cancel in taking proper ratios of radiative and semileptonic spectra.
Let us now give the explicit expression of the coefficient function in the semileptonic case:
$$C(\overline{x},w;\alpha )=C^{(0)}(\overline{x},w)+\alpha C^{(1)}(\overline{x},w)+\alpha ^2C^{(2)}(\overline{x},w)+O(\alpha ^3),$$
(52)
where
$`C^{(0)}(\overline{x},w)`$ $`=`$ $`12(w\overline{x})(1+\overline{x}w);`$ (53)
$`C^{(1)}(\overline{x},w)`$ $`=`$ $`12{\displaystyle \frac{C_F}{\pi }}(w\overline{x})\{(1+\overline{x}w)[\mathrm{Li}_2(w)+\mathrm{log}w\mathrm{log}(1w){\displaystyle \frac{3}{2}}\mathrm{log}w{\displaystyle \frac{w\mathrm{log}w}{2(1w)}}{\displaystyle \frac{35}{8}}]+`$ (54)
$`+{\displaystyle \frac{\overline{x}\mathrm{log}w}{2(1w)}}\}`$
with $`\overline{x}=1x`$ <sup>12</sup><sup>12</sup>12 To avoid spurious imaginary parts for $`w>1`$ one can use the relation $`\mathrm{Li}_2(w)=\mathrm{Li}_2(1w)\mathrm{log}w\mathrm{log}(1w)+\pi ^2/6`$.. Note that the coefficient function contains the overall factor $`w\overline{x}=\overline{x}_\nu `$, which vanishes linearly at the endpoint of the neutrino spectrum. We have defined $`\overline{x}_\nu =1x_\nu `$ and $`x_\nu =2E_\nu /m_b`$.
Unlike the coefficient function, the remainder function $`d(x,u,w;\alpha )`$ has an expansion starting at $`O(\alpha )`$:
$$d(x,w,u;\alpha )=\alpha d^{(1)}(x,w,u)+\alpha ^2d^{(2)}(x,w,u)+O(\alpha ^3).$$
(55)
Omitting the overall factor $`C_F/\pi `$, we obtain: <sup>13</sup><sup>13</sup>13 The $`O(\alpha )`$ function is obtained from that one given in $`d_{old}^{(1)}(x,w,y)`$ in terms of the variables $`z=1y`$ and $`\zeta =14y`$, by using a relation extending eq. (23) of :
$$d^{(1)}(x,w,u)=d_{old}^{(1)}(x,w,y(u))\frac{dy}{du}(u)+C^{(0)}(x,w)\frac{C_F}{\pi }\left[\frac{\mathrm{log}u+7/4}{u}\frac{\mathrm{log}y(u)+7/4}{y(u)}\frac{dy}{du}(u)\right].$$
(56) This function can also be obtained with a direct matching with the $`O(\alpha )`$ triple differential distribution computed in after a change of variable (see the end of this section for a discussion about matching).
$`d^{(1)}(\overline{x},w,u)`$ $`=`$ $`{\displaystyle \frac{3w^4\left(24+3w8\overline{x}\right)}{4\left(1+u\right)^5}}+{\displaystyle \frac{9w^4\left(24+3w8\overline{x}\right)}{8\left(1+u\right)^4}}+`$ (57)
$``$ $`{\displaystyle \frac{9\left(12+w\right)\left(2+w\right)^2\left(w2\overline{x}\right)^2}{16\left(1u\right)^3}}+{\displaystyle \frac{9\left(12+w\right)\left(2+w\right)^2\left(w2\overline{x}\right)^2}{32\left(1u\right)^2}}+`$
$``$ $`{\displaystyle \frac{3w^2\left(3247w8w^2+16\overline{x}+20w\overline{x}+w^2\overline{x}+8\overline{x}^23w\overline{x}^2\right)}{8\left(1+u\right)^2}}+`$
$``$ $`{\displaystyle \frac{3w^2\left(64+94w+40w^2+3w^332\overline{x}40w\overline{x}10w^2\overline{x}16\overline{x}^2+6w\overline{x}^2\right)}{8\left(1+u\right)^3}}+`$
$`+`$ $`{\displaystyle \frac{3}{64\left(1+u\right)}}(640w368w^2200w^316w^4+3w^5384\overline{x}+320w\overline{x}+528w^2\overline{x}+`$
$`+`$ $`112w^3\overline{x}16w^4\overline{x}256\overline{x}^248w\overline{x}^2224w^2\overline{x}^2+24w^3\overline{x}^2)+`$
$`+`$ $`{\displaystyle \frac{3}{64\left(1u\right)}}(256w+528w^2200w^316w^4+3w^5+512\overline{x}1472w\overline{x}+`$
$`+`$ $`528w^2\overline{x}+112w^3\overline{x}16w^4\overline{x}+640\overline{x}^248w\overline{x}^2224w^2\overline{x}^2+24w^3\overline{x}^2)+`$
$``$ $`{\displaystyle \frac{9w^5\mathrm{log}u}{4\left(1+u\right)^6}}+{\displaystyle \frac{9w^5\mathrm{log}u}{2\left(1+u\right)^5}}{\displaystyle \frac{9\left(12+w\right)\left(2+w\right)^2\left(w2\overline{x}\right)^2\mathrm{log}u}{16\left(1u\right)^4}}+`$
$`+`$ $`{\displaystyle \frac{9\left(12+w\right)\left(2+w\right)^2\left(w2\overline{x}\right)^2\mathrm{log}u}{16\left(1u\right)^3}}+`$
$`+`$ $`{\displaystyle \frac{3w^3\left(10+16w+w^2+8\overline{x}2w\overline{x}2\overline{x}^2\right)\mathrm{log}u}{8\left(1+u\right)^3}}+`$
$``$ $`{\displaystyle \frac{3w^3\left(10+16w+7w^2+8\overline{x}2w\overline{x}2\overline{x}^2\right)\mathrm{log}u}{8\left(1+u\right)^4}}+`$
$``$ $`{\displaystyle \frac{3\mathrm{log}u}{64\left(1+u\right)^2}}w(144w+208w^2+16w^3+w^464\overline{x}80w\overline{x}16w^2\overline{x}+`$
$``$ $`8w^3\overline{x}+48\overline{x}^296w\overline{x}^2+16w^2\overline{x}^2)+`$
$`+`$ $`{\displaystyle \frac{3\mathrm{log}u}{64\left(1u\right)^2}}(256w+624w^2304w^3+16w^4+w^5+512\overline{x}1856w\overline{x}+`$
$`+`$ $`944w^2\overline{x}16w^3\overline{x}8w^4\overline{x}+1024\overline{x}^2464w\overline{x}^296w^2\overline{x}^2+16w^3\overline{x}^2).`$
The remainder function is a combination of rational functions of $`u`$ multiplied in some cases by $`\mathrm{log}u`$, with coefficients given by polynomials in $`w`$ and $`\overline{x}`$.
The main point about the semileptonic decay (7) is that it has — unlike the radiative decay (5) — $`q^20`$ and consequently the form factor depends not only on $`u`$ but also on the hadronic energy $`w`$ through the running coupling:
$$\sigma =\sigma [u;\alpha (wm_b)].$$
(58)
The form factor is therefore a function of two variables.
We work in next-to-leading order (NLO), in which only the $`O(\alpha )`$ corrections to the coefficient function and remainder function are retained (see next section). Since the difference between $`\alpha (wm_b)`$ and $`\alpha (m_b)`$ is $`O(\alpha ^2)`$, we can set $`w=1`$ in the argument of the coupling entering the coefficient function and the remainder function. We then obtain the simpler expression:
$$\frac{1}{\mathrm{\Gamma }}\frac{d^3\mathrm{\Gamma }}{dxdudw}=C[x,w;\alpha (m_b)]\sigma [u;\alpha (wm_b)]+d[x,u,w;\alpha (m_b)](\mathrm{NLO}).$$
(59)
Note that we cannot set $`w=1`$ in the coupling entering the form factor, because in the latter case $`\alpha `$ is multiplied by large logarithms, which “amplify” $`O(\alpha ^2)`$ differences in the couplings (see next section).
Let us make a few remarks about the final result of this section, eq. (59):
* it describes semi-inclusive decays, in which the internal structure of the hadronic final states is not observed, but only the total mass and energy are measured. Less inclusive quantities, such as for instance the energy distribution of the final $`up`$ quark (i.e. the fragmentation function of the $`up`$ quark), cannot be computed in this framework;
* it constitutes an improvement of the fixed-order $`O(\alpha )`$ result in all the cases in which there are large threshold logarithms. In all the other cases, where there are no threshold logarithms, such as for example the dilepton mass distribution , there is not any advantage of the resummed formula over the fixed-order one.
In the next sections we integrate the resummed triple-differential distribution to obtain double and single (resummed) spectra. There are two methods to accomplish this task which are completely equivalent:
1. The first method involves the direct integration of the complete triple-differential distribution. Schematically:
$$(\mathrm{spectrum})=C\sigma +d.$$
(60)
Large logarithms come only from the first term on the r.h.s. of (60), while non-logarithmic, “small” terms come both from the first and the second term. To obtain a factorized form for the spectrum analogous to the one for the triple-distribution, in which the remainder function collects all the small terms, one rearranges the r.h.s. of (60): the small terms coming from the integration of $`C\sigma `$ are put in the remainder function;
2. In the second method, one integrates the block $`C\sigma `$ only and drops the small terms coming from the integration. The remainder function is obtained by expanding the resummed expression in powers of $`\alpha `$ and comparing with the fixed-order spectrum.
## 3 Threshold Resummation
It is convenient to define the partially integrated or cumulative form factor $`\mathrm{\Sigma }(u,\alpha )`$:
$$\mathrm{\Sigma }(u;\alpha )=_0^u𝑑u^{}\sigma (u^{};\alpha ).$$
(61)
Performing the integrations, one obtains for the $`O(\alpha )`$ form factor:
$$\mathrm{\Sigma }(u;\alpha )=\mathrm{\hspace{0.17em}1}\frac{C_F\alpha }{2\pi }L^2+\frac{7C_F\alpha }{4\pi }L+O(\alpha ^2),$$
(62)
where
$$L=\mathrm{log}\frac{1}{u}.$$
(63)
$`\mathrm{\Sigma }`$ contains a double logarithm coming from the overlap of the soft and the collinear region and a single logarithm of soft or collinear origin. The normalization condition reads:
$$\mathrm{\Sigma }(1;\alpha )=_0^1𝑑u\sigma (u;\alpha )=\mathrm{\hspace{0.17em}1}.$$
(64)
As already noted, we have assumed a minimal factorization scheme, in which only logarithms and not constants or other functions are contained in the form factor. The expression of the partially integrated form factor $`\mathrm{\Sigma }`$ is technically simpler than the one for the differential form factor $`\sigma `$, as it involves ordinary functions instead of generalized ones. Furthermore, in experiments one always measures some integral of $`\sigma `$ around a central $`u`$ value because of the binning.
In the limit $`u0^+`$, no final states are included in the distribution and therefore one expects, on physical grounds, that
$$\underset{u0^+}{lim}\mathrm{\Sigma }(u;\alpha )=\mathrm{\hspace{0.17em}0}.$$
(65)
The $`O(\alpha )`$ expression (62) does not have this limit and it is actually divergent to $`\mathrm{}`$ — a completely un-physical result. In general, a truncated expansion in powers of $`\alpha `$ is divergent for $`u0^+`$, because the coefficients diverge in this limit. Therefore, one has to resum the infrared logarithms, i.e. the terms of the form $`\alpha ^nL^k`$, to all orders in perturbation theory. In higher orders, $`\mathrm{\Sigma }`$ contains at most two logarithms for each power of $`\alpha `$, one of soft origin and another one of collinear origin. Its general expression is then:
$$\mathrm{\Sigma }(L,\alpha )=\mathrm{\hspace{0.17em}1}+\underset{n=1}{\overset{\mathrm{}}{}}\underset{k=1}{\overset{2n}{}}\mathrm{\Sigma }_{nk}\alpha ^nL^k,$$
(66)
where $`\mathrm{\Sigma }_{nk}`$ are numerical coefficients. At present, a complete resummation of all the logarithmically-enhanced terms on the r.h.s. of eq. (66) is not feasible in QCD: one has to resort to approximate schemes. The most crude approximation consists of picking up the most singular term for $`u0^+`$ for each power of $`\alpha `$, i.e. all the terms of the form:
$$\alpha ^nL^{2n}(\mathrm{double}\mathrm{logarithmic}\mathrm{approximation}).$$
(67)
In this approximation, we can neglect running coupling effects and effects related to the kinematical constraints: higher orders simply exponentiate the $`O(\alpha )`$ double logarithm and one obtains
$$\mathrm{\Sigma }(u;\alpha )=e^{C_F\alpha /(2\pi )L^2}(\mathrm{double}\mathrm{logarithmic}\mathrm{approximation}).$$
(68)
Let us note that the resummed expression (68), unlike the fixed-order one (62), does satisfy the condition (65). The exponent in the resummed form factor involves a single term, $`C_F\alpha /(2\pi )L^2`$, and has therefore a simpler form than the form factor itself. This remains true when more accurate resummation schemes are constructed, so it is convenient to define $`G`$ as:
$$\mathrm{\Sigma }=e^G.$$
(69)
It can be shown that the expansion for the function $`G`$ is of the form :
$$G(L;\alpha )=\underset{n=1}{\overset{\mathrm{}}{}}\underset{k=1}{\overset{n+1}{}}G_{nk}\alpha ^nL^k,$$
(70)
where $`G_{nk}`$ are numerical coefficients. Let us note that the sum over $`k`$ extends up to $`n+1`$ in (70), while it extends up to $`2n`$ in the form factor in eq. (66). This property is a generalization of the simple exponentiation of the $`O(\alpha )`$ logarithms which holds in QED and is called generalized exponentiation. In general, this property holds for quantities analogous to the semi-inclusive form factors, in which the gluon radiation is not directly observed. One sums therefore over all possible final states coming from the evolution of the emitted gluons (inclusive gluon decay quantities). The property expressed by eq. (70) does not hold for quantities in which gluon radiation is observed directly, as for example in parton multiplicities, where different evolutions of gluon jets give rise to different multiplicities.
### 3.1 $`N`$-space
A systematic resummation is consistently done in $`N`$-moment space or Mellin space, in which kinematical constraints are factorized in the soft limit and are easily integrated over . One considers the Mellin transform of the form factor $`\sigma (u;\alpha )`$:
$$\sigma _N(\alpha )_0^1𝑑u(1u)^{N1}\sigma (u;\alpha ).$$
(71)
The threshold region is studied in moment space by taking the limit $`N\mathrm{}`$, because for large $`N`$ the integral above takes contributions mainly from the region $`u1`$. For example, the Mellin transform of the spectrum in eq. (50) is of the form
$$_0^1(1t_s)^{N1}\frac{1}{\mathrm{\Gamma }_R}\frac{d\mathrm{\Gamma }_R}{dt_s}𝑑t_s=C_R(\alpha )\sigma _N(\alpha )+d_{R,N}(\alpha ),$$
(72)
where
$$d_{R,N}(\alpha )\mathrm{\hspace{0.17em}0}\mathrm{for}N\mathrm{}.$$
(73)
The total rate in Mellin space is obtained by taking $`N=1`$.
It can be shown that the form factor in $`N`$-space has the following exponential structure:
$$\sigma _N(\alpha )=e^{G_N(\alpha )},$$
(74)
where
$$G_N(\alpha )=_0^1𝑑z\frac{z^{N1}1}{1z}\left\{_{Q^2(1z)^2}^{Q^2(1z)}\frac{dk_t^2}{k_t^2}A\left[\alpha (k_t^2)\right]+B\left[\alpha (Q^2(1z))\right]+D\left[\alpha (Q^2(1z)^2)\right]\right\}.$$
(75)
Let us note that a prescription has to be assigned to this formula since it involves integrations over the Landau pole . The functions entering the resummation formula have a standard fixed-order expansion, with numerical coefficients:
$`A(\alpha )`$ $`=`$ $`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}A_n\alpha ^n=A_1\alpha +A_2\alpha ^2+A_3\alpha ^3+A_4\alpha ^4+\mathrm{}`$ (76)
$`B(\alpha )`$ $`=`$ $`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}B_n\alpha ^n=B_1\alpha +B_2\alpha ^2+B_3\alpha ^3+\mathrm{}`$ (77)
$`D(\alpha )`$ $`=`$ $`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}D_n\alpha ^n=D_1\alpha +D_2\alpha ^2+D_3\alpha ^3+\mathrm{}`$ (78)
The known values for the resummation constants read:
$`A_1`$ $`=`$ $`{\displaystyle \frac{C_F}{\pi }};`$ (79)
$`A_2`$ $`=`$ $`{\displaystyle \frac{C_F}{\pi ^2}}\left[C_A\left({\displaystyle \frac{67}{36}}{\displaystyle \frac{z(2)}{2}}\right){\displaystyle \frac{5}{18}}n_f\right];`$ (80)
$`A_3`$ $`=`$ $`{\displaystyle \frac{C_F}{\pi ^3}}[C_A^2({\displaystyle \frac{245}{96}}+{\displaystyle \frac{11}{24}}z(3){\displaystyle \frac{67}{36}}z(2)+{\displaystyle \frac{11}{8}}z(4))C_An_f({\displaystyle \frac{209}{432}}+{\displaystyle \frac{7}{12}}z(3){\displaystyle \frac{5}{18}}z(2))+`$ (81)
$`C_Fn_f({\displaystyle \frac{55}{96}}{\displaystyle \frac{z(3)}{2}}){\displaystyle \frac{n_f^2}{108}}];`$
$`B_1`$ $`=`$ $`{\displaystyle \frac{3}{4}}{\displaystyle \frac{C_F}{\pi }};`$ (82)
$`B_2`$ $`=`$ $`{\displaystyle \frac{C_F}{\pi ^2}}\left[C_A\left({\displaystyle \frac{3155}{864}}+{\displaystyle \frac{11}{12}}z(2)+{\displaystyle \frac{5}{2}}z(3)\right)C_F\left({\displaystyle \frac{3}{32}}+{\displaystyle \frac{3}{2}}z(3){\displaystyle \frac{3}{4}}z(2)\right)+n_f\left({\displaystyle \frac{247}{432}}{\displaystyle \frac{z(2)}{6}}\right)\right];`$ (83)
$`D_1`$ $`=`$ $`{\displaystyle \frac{C_F}{\pi }};`$ (84)
$`D_2`$ $`=`$ $`{\displaystyle \frac{C_F}{\pi ^2}}\left[C_A\left({\displaystyle \frac{55}{108}}{\displaystyle \frac{9}{4}}z(3)+{\displaystyle \frac{z(2)}{2}}\right)+{\displaystyle \frac{n_f}{54}}\right],`$ (85)
where $`C_A=N_c=3`$ is the Casimir of the adjoint representation. The coefficients $`A_1`$, $`B_1`$ and $`D_1`$ are renormalization-scheme independent, as they can be obtained from tree-level amplitudes with one-gluon emission (see later). The higher-order coefficients are instead renormalization-scheme dependent and are given in the $`\overline{MS}`$ scheme for the coupling constant <sup>14</sup><sup>14</sup>14A discussion about the scheme dependence of the higher order coefficients $`A_2,B_2,`$ etc. on the coupling constant can be found in ..
To this approximation, the first three orders of the $`\beta `$-function are also needed :
$`\beta _0`$ $`=`$ $`{\displaystyle \frac{1}{4\pi }}\left[{\displaystyle \frac{11}{3}}C_A{\displaystyle \frac{2}{3}}n_f\right];`$ (86)
$`\beta _1`$ $`=`$ $`{\displaystyle \frac{1}{24\pi ^2}}\left[17C_A^{\mathrm{\hspace{0.17em}2}}\left(5C_A+3C_F\right)n_f\right];`$ (87)
$`\beta _2`$ $`=`$ $`{\displaystyle \frac{1}{64\pi ^3}}\left[{\displaystyle \frac{2857}{54}}C_A^{\mathrm{\hspace{0.17em}3}}\left({\displaystyle \frac{1415}{54}}C_A^{\mathrm{\hspace{0.17em}2}}+{\displaystyle \frac{205}{18}}C_AC_FC_F^{\mathrm{\hspace{0.17em}2}}\right)n_f+\left({\displaystyle \frac{79}{54}}C_A+{\displaystyle \frac{11}{9}}C_F\right)n_f^2\right].`$ (88)
As is well known, $`\beta _0`$ and $`\beta _1`$ are renormalization-scheme independent, while $`\beta _2`$ is not and has been given in the $`\overline{MS}`$ scheme. We define the $`\beta `$-function with an overall minus sign:
$$\frac{d\alpha }{d\mathrm{log}\mu ^2}=\beta (\alpha )=\beta _0\alpha ^2\beta _1\alpha ^3\beta _2\alpha ^4\mathrm{}.$$
(89)
The running coupling reads:
$$\alpha (\mu )=\frac{1}{\beta _0\mathrm{log}\mu ^2/\mathrm{\Lambda }^2}\frac{\beta _1}{\beta _0^3}\frac{\mathrm{log}\left(\mathrm{log}\mu ^2/\mathrm{\Lambda }^2\right)}{\mathrm{log}^2\mu ^2/\mathrm{\Lambda }^2}+\frac{\beta _1^2}{\beta _0^5}\frac{\mathrm{log}^2\left(\mathrm{log}\mu ^2/\mathrm{\Lambda }^2\right)\mathrm{log}\left(\mathrm{log}\mu ^2/\mathrm{\Lambda }^2\right)\mathrm{\hspace{0.17em}1}}{\mathrm{log}^3\mu ^2/\mathrm{\Lambda }^2}+\frac{\beta _2}{\beta _0^4}\frac{1}{\mathrm{log}^3\mu ^2/\mathrm{\Lambda }^2}.$$
(90)
The functions $`A(\alpha )`$, $`B(\alpha )`$ and $`D(\alpha )`$ have the following physical interpretation (see for example ):
* The function $`A(\alpha )`$ involves a double integration over the transverse momentum $`k_t`$ and the energy $`\omega `$ of the emitted gluon and represents emissions at small angle and at small energy from the light quark. The leading term $`A_1`$ is the coefficient of that piece of the matrix element squared for one real gluon emission, which is singular in the small angle and small energy limit:
$$A_1\alpha \frac{d\omega }{\omega }\frac{d\theta ^2}{\theta ^2}A_1\alpha \frac{d\omega }{\omega }\frac{dk_t^2}{k_t^2},$$
(91)
where $`k_t\omega \theta `$ is the transverse momentum of the gluon. In (91) we have given the representation of the integral both in the angle $`\theta `$ and in the transverse momentum $`k_t`$. The subleading coefficients $`A_2`$, $`A_3`$, etc. represent corrections to the basic double-logarithmic emission. The function $`A(\alpha )`$ “counts” the number of light quark jets in different processes, i.e. we can write
$$A^{(P)}(\alpha )=n_qA(\alpha ),$$
(92)
where $`n_q`$ is the number of primary light quarks in the process $`P`$. For example, in $`e^+e^{}`$ annihilation into hadrons $`n_q=2`$, while in the heavy flavor decays (2) $`n_q=1`$. Since soft gluons only couple to the four-momentum of their emitters and not to their spin, the function $`A_g(\alpha )`$ for gluon jets is obtained from the quark one $`A(\alpha )`$ simply taking into account the change in the color charge, i.e. multiplying by $`C_A/C_F`$ ;
* the function $`B(\alpha )`$ represents emissions at small angle with a large energy from the light quark. $`B_1`$ is the coefficient of that piece of the matrix element squared which is singular in the small angle limit:
$$B_1\alpha d\omega \frac{d\theta ^2}{\theta ^2}B_1\alpha d\omega \frac{dk_t^2}{k_t^2}.$$
(93)
The non logarithmic integration over the gluon energy $`\omega `$ has been done and does not appear explicitly in eq. (75); the integration over the angle $`\theta `$ or the transverse momentum $`k_t`$ is rewritten as an integral over $`z`$. The function $`B(\alpha )`$ counts the number of final-quark jets, i.e.
$$B^{(P)}(\alpha )=n_lB(\alpha ),$$
(94)
where $`n_l`$ is the number of primary final quarks in the process $`P`$. For example in $`e^+e^{}`$ annihilation into hadrons $`n_l=2`$, while in DIS or in the heavy flavor decays (2) $`n_l=1`$. Since hard collinear emissions are sensitive to the spin of the emitting particles, the gluon function $`B_g(\alpha )`$ is not simply related to the quark one $`B(\alpha )`$ ;
* the function $`D(\alpha )`$ represents emissions at large angle and small energy from the heavy quark. $`D_1`$ is the coefficient of that piece of the matrix element squared which is singular in the small energy limit:
$$D_1\alpha \frac{d\omega }{\omega }d\theta ^2.$$
(95)
The non logarithmic integration over the angle $`\theta `$ or the transverse momentum $`k_t`$ has been done and does not appear explicitly in eq. (75); the integration over the energy $`\omega `$ is rewritten as an integral over $`z`$. $`D_1=0`$ in all the processes involving light partons only, as for instance DIS, Drell-Yan (DY) or $`e^+e^{}`$ annihilation into hadrons, while it is not zero in all the processes containing at least one heavy quark, such as for example the heavy flavor decays (2). Note that the effective coupling appearing in the $`D`$ terms is $`\alpha \left[Q^2(1z)^2\right]`$ and is therefore substantially larger for $`1z1`$ than the coupling entering the hard collinear terms, namely $`\alpha \left[Q^2(1z)\right]`$.
Eq. (75) is therefore a generalization of the $`O(\alpha )`$ result, possessing a double logarithm coming from the overlap of the soft and the collinear region and a single logarithm of soft or collinear origin (see eqs. (91), (93) and (95)) <sup>15</sup><sup>15</sup>15 Let us remember however that only two of the three functions appearing in eq. (75) are independent .. The functions $`A(\alpha )`$ and $`B(\alpha )`$ are believed to by universal, i.e. process independent to any order in perturbation theory, as they represent the development of a parton into a jet, i.e. one-particle properties. The function $`D(\alpha )`$ on the contrary is process-dependent, as it describes soft emission at large angle, with interference contributions from all the hard partons in the process, i.e. it describes global properties of the hadronic final states. Let us observe that $`A_2`$ and $`D_2`$, unlike $`B_2`$, do not have a $`C_F^2`$ contribution. That is a consequence of the eikonal identity, which holds in the soft limit . According to this identity, the abelian contributions simply exponentiate the lowest order $`O(\alpha C_F)`$ term, just like in QED. That means that there are no higher order terms in the exponent $`G_N`$. Because of similar reasonings, $`A_3`$ does not have a $`C_F^3`$ contribution.
Despite its supposed asymptotic nature, the numerical values of the coefficients show a rather good convergence of the perturbative series. Note that all the double-logarithmic coefficients $`A_i`$ are positive, implying an increasing suppression with the order of the expansion (up to the third one) of the rate in the threshold region. On the contrary, the single-logarithmic coefficients $`B_i`$ and $`D_i`$ – with the exception of $`B_2`$ — are all negative and therefore tend to enhance the rate in the threshold region . We have:
$`A_1`$ $`=`$ $`+\mathrm{\hspace{0.17em}0.424413};`$ (96)
$`A_2`$ $`=`$ $`+\mathrm{\hspace{0.17em}0.420947}\mathrm{\hspace{0.17em}0.0375264}n_f=\mathrm{\hspace{0.17em}0.308367};`$ (97)
$`A_3`$ $`=`$ $`+\mathrm{\hspace{0.17em}0.592067}\mathrm{\hspace{0.17em}0.0923137}n_f\mathrm{\hspace{0.17em}0.000398167}n_f^2=\mathrm{\hspace{0.17em}0.311542};`$ (98)
$`B_1`$ $`=`$ $`\mathrm{\hspace{0.17em}0.318310};`$ (99)
$`B_2`$ $`=`$ $`+\mathrm{\hspace{0.17em}0.229655}+\mathrm{\hspace{0.17em}0.04020}n_f=\mathrm{\hspace{0.17em}0.350269};`$ (100)
$`D_1`$ $`=`$ $`\mathrm{\hspace{0.17em}0.424413};`$ (101)
$`D_2`$ $`=`$ $`\mathrm{\hspace{0.17em}0.556416}+\mathrm{\hspace{0.17em}0.002502}n_f=\mathrm{\hspace{0.17em}0.548911}.`$ (102)
With our definition, the $`\beta `$-function coefficients are, as well known, all positive.
$`\beta _0`$ $`=`$ $`+\mathrm{\hspace{0.17em}0.87535}\mathrm{\hspace{0.17em}0.05305}n_f=+\mathrm{\hspace{0.17em}0.71620};`$ (103)
$`\beta _1`$ $`=`$ $`+\mathrm{\hspace{0.17em}0.64592}\mathrm{\hspace{0.17em}0.08021}n_f=+\mathrm{\hspace{0.17em}0.40529};`$ (104)
$`\beta _2`$ $`=`$ $`+\mathrm{\hspace{0.17em}0.71986}\mathrm{\hspace{0.17em}0.140904}n_f+\mathrm{\hspace{0.17em}0.003032}n_f^2=+\mathrm{\hspace{0.17em}0.324436}.`$ (105)
In the last member we have assumed 3 active flavors ($`n_f=3`$).
Let us now discuss the computation of the coefficients entering the resummation formula. The occurrence of a Sudakov form factor in semileptonic $`B`$ decays was acknowledged originally in , where a simple exponentiation involving $`A_1`$ and $`B_1+D_1`$ was performed. The coefficient $`A_2`$ was computed for the first time, as far as we know, in . It was denoted $`A_1K`$ since it was considered a kind of renormalization of the lowest-order contribution:
$$A_1\alpha A_1\alpha (1+K\alpha ).$$
(106)
The coefficient $`A_2`$ was obtained from the soft-singular part of the $`qq`$ two-loop splitting function , that is as the coefficient of the $`1/(1z)`$ term<sup>16</sup><sup>16</sup>16This is exactly the same procedure which has been followed to derive the third-order coefficient $`A_3`$ from the three-loop splitting function .. $`A_2`$ was subsequently recomputed in in the framework of Wilson line theory, where the function $`A(\alpha )`$ has a geometrical meaning: it is the anomalous dimension of a cusp operator, representing the radiation emitted because of a sudden change of velocity of a heavy quark,
$$\mathrm{\Gamma }_{cusp}(\alpha )=\underset{n=1}{\overset{\mathrm{}}{}}\mathrm{\Gamma }_{cusp}^{(n)}\alpha ^n=\mathrm{\Gamma }_{cusp}^{(1)}\alpha +\mathrm{\Gamma }_{cusp}^{(2)}\alpha ^2+\mathrm{}$$
(107)
Indeed, it has been explicitly checked up to second order that these two functions coincide:
$$A(\alpha )=\mathrm{\Gamma }_{cusp}(\alpha ).$$
(108)
Let us note that:
* the theory of Wilson lines and Wilson loops;
* the eikonal or soft approximation in perturbative QCD;
* the heavy quark effective theory (HQET) and the large energy effective theory (LEET),
all involve basically the same structure, i.e. the same propagators and vertices and the same amplitudes. Since the same structure has been studied in different frameworks, there is multiple notation and terminology for the same objects. Let us stress however that in ordinary QCD the function $`A(\alpha )`$ is not an anomalous dimension, since it is not obtained from ultraviolet $`1/ϵ`$ poles in renormalization constants but from infrared poles or from finite parts of scattering amplitudes. $`A(\alpha )`$ becomes an anomalous dimension in the effective theory because the latter has additional ultraviolet divergencies with respect to QCD. While in QCD one has to subtract only ultraviolet divergencies related to coupling constant renormalization, in the effective theory one has also to subtract additional ultraviolet divergencies related to the cusp operators. A scheme dependence is therefore introduced in the effective theory, which is not present in full QCD. It seems to us therefore that the equality (108) is not guaranteed a priori in higher orders and may require a specific scheme for the subtractions in the effective theory. At present, $`A_3`$ has only been derived in full QCD and not in the effective theory.
The coefficient $`B_2`$ has been computed by means of the second order correction to the inclusive DIS cross section, which contains the combination $`B_2+D_2^{DIS}`$ (the DIS analogue of eq. (136), see later) and by means of the third order correction, which contains the different combination $`B_2+2D_2^{DIS}`$ (the DIS analogue of eq. (139), see later). The knowledge of the fermionic contribution to the $`O(\alpha ^3)`$ DIS cross section was sufficient for a complete determination of $`B_2`$ , with later checks offered by the complete computation .
An incorrect value for the coefficient $`D_2`$ for heavy favor decays has been obtained in the original computation in , where the technique to compute real and virtual diagrams in the effective-theory in configuration space has also been developed. In the coefficient of the single logarithm in the radiative decay (5) to order $`\alpha ^2`$ has been presented, from which the correct value of $`D_2`$ can be extracted (let us note however that numerically the two values are not very different). A second order computation of heavy flavor fragmentation in ordinary QCD was presented in , which allows the determination of the sum $`B_2+D_2^{frag}`$ (the analogue of eq. (136), see later). Using an identity relating the coefficient for heavy flavor fragmentation with that one for heavy flavor decays, and subtracting the known value for the universal coefficient $`B_2`$, the correct value for $`D_2`$ was explicitly derived in (see also ). Still in , by repeating the Wilson line computation of , errors were found and the same value of $`D_2`$ extracted from heavy flavor fragmentation was re-obtained. Recently, the second order contribution of the chromomagnetic operator $`O_7`$ to the photon spectrum in the radiative decay (5) has been calculated , confirming these results (see also ).
According to the previous remarks concerning the relation between $`A(\alpha )`$ and $`\mathrm{\Gamma }_{cusp}(\alpha )`$, we believe it is a non-trivial fact that the same value of $`D_2`$ is obtained with two completely different methods:
* a direct computation in the effective theory, which describes the soft region only;
* an extraction from an ordinary QCD computation, which gives the sum of the soft and the collinear contributions $`B_2+D_2`$, by subtracting the collinear contribution $`B_2`$ obtained from second order and third order DIS computations.
The following expansion holds true for the exponent:
$$G_N(\alpha )=\underset{n=1}{\overset{\mathrm{}}{}}\underset{k=1}{\overset{n+1}{}}G_{nk}\alpha ^nl^k+O\left(\frac{1}{N}\right),$$
(109)
where
$$l=\mathrm{log}N.$$
(110)
The expansion of the logarithm of the form factor has a similar structure in physical space and in $`N`$-space; roughly speaking, going to $`N`$-space, $`\mathrm{log}1/u\mathrm{log}N`$.
As already discussed, we are interested in the large-$`N`$ limit; the $`O(1/N)`$ terms can be neglected in our leading-twist analysis. A resummation of all the logarithmically-enhanced terms in (109) is at present unfeasible in QCD even in $`N`$-space, so one has to rely on approximate schemes. Let us discuss the fixed-logarithmic accuracy scheme:
* Leading order (LO). One keeps in the exponent $`G_N(\alpha )`$ only the leading power of the logarithm for each power of $`\alpha `$, i.e. $`k=n+1`$:
$$G_N^{LO}=\underset{n=1}{\overset{\mathrm{}}{}}G_{nn+1}\alpha ^nl^{n+1}=G_{12}\alpha l^2+G_{23}\alpha ^2l^3+O(\alpha ^3).$$
(111)
The coefficient function is kept in lowest order, i.e. $`C^{LO}=1`$ and the remainder function is completely neglected, i.e. $`d^{LO}=0`$;
* Next-to-leading order (NLO). One keeps in $`G_N(\alpha )`$ also the terms with $`n=k`$, i.e.:
$`G_N^{NLO}`$ $`=`$ $`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}\left[G_{nn+1}\alpha ^nl^{n+1}+G_{nn}\alpha ^nl^n\right]`$ (112)
$`=`$ $`G_{12}\alpha l^2+G_{11}\alpha l+G_{23}\alpha ^2l^3+G_{22}\alpha ^2l^2+O(\alpha ^3).`$
To $`O(\alpha )`$ one retains both the double and the single logarithm. In general for each order in $`\alpha `$ one keeps the principal two logarithms. One also keeps the $`O(\alpha )`$ terms both in the coefficient function and in the remainder function:
$$C^{NLO}=\mathrm{\hspace{0.17em}1}+\alpha C^{(1)};d^{NLO}=\alpha d^{(1)}.$$
(113)
The one-loop coefficient function is needed because of the factorized form of the QCD form factor. One has indeed a resummed expression of the form:
$$\left[1+\alpha C^{(1)}\right]e^{G_{12}\alpha l^2+\mathrm{}}$$
(114)
By expanding the exponent in powers of $`\alpha `$, a term coupling the coefficient function and the double logarithm is obtained:
$$\alpha ^2C^{(1)}G_{12}l^2,$$
(115)
which must be included in the NLO approximation;
* Next-to-next-to-leading order (NNLO). One keeps in $`G_N`$ also the terms with $`n=k1`$, i.e.:
$`G_N^{NNLO}`$ $`=`$ $`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}\left[G_{nn+1}\alpha ^nl^{n+1}+G_{nn}\alpha ^nl^n+G_{nn1}\alpha ^nl^{n1}\right]`$ (116)
$`=`$ $`G_{12}\alpha l^2+G_{11}\alpha l+G_{23}\alpha ^2l^3+G_{22}\alpha ^2l^2+G_{21}\alpha ^2l+G_{34}\alpha ^3l^4+G_{33}\alpha ^3l^3+G_{32}\alpha ^3l^2+`$
$`+`$ $`O(\alpha ^4).`$
To $`O(\alpha ^2)`$, all the infrared logarithms are included. In general, for each order in $`\alpha `$, one keeps the principal three logarithms. The first omitted term is the single logarithm to order $`\alpha ^3`$. One has also to keep the $`O(\alpha ^2)`$ terms both in the coefficient function and in the remainder function:
$$C^{NNLO}=\mathrm{\hspace{0.17em}1}+\alpha C^{(1)}+\alpha ^2C^{(2)};d^{NNLO}=\alpha d^{(1)}+\alpha ^2d^{(2)}.$$
(117)
The classes of logarithms discussed above can be explicitly resummed by means of a function series expansion of $`G_N(\alpha )`$ :
$`G_N(\alpha )=lg_1(\lambda )+{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}\alpha ^ng_{2+n}(\lambda )=lg_1(\lambda )+g_2(\lambda )+\alpha g_3(\lambda )+\alpha ^2g_4(\lambda )+\mathrm{},`$ (118)
where
$$\lambda =\beta _0\alpha l.$$
(119)
The $`g_i(\lambda )`$ are homogeneous functions of $`\lambda `$ and have a series expansion around $`\lambda =0`$:
$$g_i(\lambda )=\underset{n=1}{\overset{\mathrm{}}{}}g_{in}\lambda ^n.$$
(120)
In LO one needs only the function $`g_1`$, in NLO one need also $`g_2`$, in NNLO also $`g_3`$ is needed and so on. The explicit expressions read:
$`g_1(\lambda )`$ $`=`$ $`{\displaystyle \frac{A_1}{2\beta _0\lambda }}\left[\left(12\lambda \right)\mathrm{log}(12\lambda )2\left(1\lambda \right)\mathrm{log}(1\lambda )\right];`$ (121)
$`g_2(\lambda )`$ $`=`$ $`{\displaystyle \frac{D_1}{2\beta _0}}\mathrm{log}(12\lambda )+{\displaystyle \frac{B_1}{\beta _0}}\mathrm{log}(1\lambda )+{\displaystyle \frac{A_2}{2\beta _{0}^{}{}_{}{}^{2}}}\left[\mathrm{log}(12\lambda )2\mathrm{log}(1\lambda )\right]+`$
$``$ $`{\displaystyle \frac{A_1\beta _1}{4\beta _{0}^{}{}_{}{}^{3}}}\left[2\mathrm{log}(12\lambda )+\mathrm{log}^2(12\lambda )4\mathrm{log}(1\lambda )2\mathrm{log}^2(1\lambda )\right]+`$
$`+`$ $`{\displaystyle \frac{A_1\gamma _E}{\beta _0}}\left[\mathrm{log}(12\lambda )\mathrm{log}(1\lambda )\right]+{\displaystyle \frac{A_1}{2\beta _0}}\left[\mathrm{log}(12\lambda )2\mathrm{log}(1\lambda )\right]\mathrm{log}{\displaystyle \frac{\mu ^2}{Q^2}}.`$
The function $`g_1(\lambda )`$ in is in agreement with that one obtained originally in . $`g_2(\lambda )`$ in differs instead from the corresponding $`g_2^{sl}(\lambda )`$ obtained in and it is equal to the corresponding function entering the $`BX_s\gamma `$ spectrum; the formalism we use makes explicit the universality of soft gluon dynamics in semileptonic and radiative decays. The NNLO function $`g_3`$ has the rather lengthy expression:
$`g_3(\lambda )`$ $`=`$ $`{\displaystyle \frac{D_2\lambda }{\beta _0\left(12\lambda \right)}}{\displaystyle \frac{2D_1\gamma _E\lambda }{12\lambda }}+{\displaystyle \frac{D_1\beta _1}{2\beta _{0}^{}{}_{}{}^{2}}}\left({\displaystyle \frac{2\lambda }{12\lambda }}+{\displaystyle \frac{\mathrm{log}(12\lambda )}{12\lambda }}\right){\displaystyle \frac{B_2\lambda }{\beta _0\left(1\lambda \right)}}{\displaystyle \frac{B_1\gamma _E\lambda }{1\lambda }}+`$ (123)
$`+`$ $`{\displaystyle \frac{B_1}{\beta _{0}^{}{}_{}{}^{2}}}\beta _1\left({\displaystyle \frac{\lambda }{1\lambda }}+{\displaystyle \frac{\mathrm{log}(1\lambda )}{1\lambda }}\right){\displaystyle \frac{A_3}{2\beta _{0}^{}{}_{}{}^{2}}}\left({\displaystyle \frac{\lambda }{12\lambda }}{\displaystyle \frac{\lambda }{1\lambda }}\right){\displaystyle \frac{A_2\gamma _E}{\beta _0}}\left({\displaystyle \frac{1}{12\lambda }}{\displaystyle \frac{1}{1\lambda }}\right)+`$
$`+`$ $`{\displaystyle \frac{A_2\beta _1}{2\beta _{0}^{}{}_{}{}^{3}}}\left({\displaystyle \frac{3\lambda }{12\lambda }}{\displaystyle \frac{3\lambda }{1\lambda }}+{\displaystyle \frac{\mathrm{log}(12\lambda )}{12\lambda }}{\displaystyle \frac{2\mathrm{log}(1\lambda )}{1\lambda }}\right)+`$
$``$ $`{\displaystyle \frac{A_1\gamma _{E}^{}{}_{}{}^{2}}{2}}\left({\displaystyle \frac{4\lambda }{12\lambda }}{\displaystyle \frac{\lambda }{1\lambda }}\right){\displaystyle \frac{A_1\pi ^2}{12}}\left({\displaystyle \frac{4\lambda }{12\lambda }}{\displaystyle \frac{\lambda }{1\lambda }}\right)+`$
$``$ $`{\displaystyle \frac{A_1\beta _2}{4\beta _{0}^{}{}_{}{}^{3}}}\left({\displaystyle \frac{2\lambda }{12\lambda }}{\displaystyle \frac{2\lambda }{1\lambda }}+2\mathrm{log}(12\lambda )4\mathrm{log}(1\lambda )\right)+`$
$`+`$ $`{\displaystyle \frac{A_1\beta _1\gamma _E}{\beta _{0}^{}{}_{}{}^{2}}}\left({\displaystyle \frac{1}{12\lambda }}{\displaystyle \frac{1}{1\lambda }}+{\displaystyle \frac{\mathrm{log}(12\lambda )}{12\lambda }}{\displaystyle \frac{\mathrm{log}(1\lambda )}{1\lambda }}\right)+`$
$``$ $`{\displaystyle \frac{A_1\beta _{1}^{}{}_{}{}^{2}}{2\beta _{0}^{}{}_{}{}^{4}}}({\displaystyle \frac{\lambda }{12\lambda }}{\displaystyle \frac{\lambda }{1\lambda }}\mathrm{log}(12\lambda )+{\displaystyle \frac{\mathrm{log}(12\lambda )}{12\lambda }}`$
$`+`$ $`{\displaystyle \frac{\mathrm{log}(12\lambda )^2}{2\left(12\lambda \right)}}+2\mathrm{log}(1\lambda ){\displaystyle \frac{2\mathrm{log}(1\lambda )}{1\lambda }}{\displaystyle \frac{\mathrm{log}(1\lambda )^2}{1\lambda }})+`$
$``$ $`{\displaystyle \frac{D_1\lambda }{\beta _0\left(12\lambda \right)}}\mathrm{log}{\displaystyle \frac{\mu ^2}{Q^2}}{\displaystyle \frac{B_1\lambda }{\beta _0\left(1\lambda \right)}}\mathrm{log}{\displaystyle \frac{\mu ^2}{Q^2}}{\displaystyle \frac{A_2}{\beta _{0}^{}{}_{}{}^{2}}}\left({\displaystyle \frac{\lambda }{12\lambda }}{\displaystyle \frac{\lambda }{1\lambda }}\right)\mathrm{log}{\displaystyle \frac{\mu ^2}{Q^2}}+`$
$``$ $`{\displaystyle \frac{A_1\gamma _E}{\beta _0}}\left({\displaystyle \frac{2\lambda }{12\lambda }}{\displaystyle \frac{\lambda }{1\lambda }}\right)\mathrm{log}{\displaystyle \frac{\mu ^2}{Q^2}}+`$
$`+`$ $`{\displaystyle \frac{A_1\beta _1}{\beta _{0}^{}{}_{}{}^{3}}}\left({\displaystyle \frac{\lambda }{12\lambda }}{\displaystyle \frac{\lambda }{1\lambda }}+{\displaystyle \frac{\mathrm{log}(12\lambda )}{2}}+{\displaystyle \frac{\lambda \mathrm{log}(12\lambda )}{12\lambda }}\mathrm{log}(1\lambda ){\displaystyle \frac{\lambda \mathrm{log}(1\lambda )}{1\lambda }}\right)\mathrm{log}{\displaystyle \frac{\mu ^2}{Q^2}}+`$
$``$ $`{\displaystyle \frac{A_1}{2\beta _0}}\left({\displaystyle \frac{2\lambda ^2}{12\lambda }}{\displaystyle \frac{\lambda ^2}{1\lambda }}\right)\mathrm{log}^2{\displaystyle \frac{\mu ^2}{Q^2}}.`$
The function $`g_3(\lambda )`$ was originally computed in , where the first NNLO resummation in heavy flavor decays was presented. At the time of that work, not all the fixed-order computations were available from which to extract the coefficients entering the resummation formula, namely $`A_3`$, $`B_2`$ and $`D_2`$. A numerical estimate of the three-loop coefficient $`A_3`$ was used, which was obtained in by fitting the known moments of the 3-loop splitting kernels and which has been later confirmed by the exact analytic evaluation . As far as $`B_2`$ is concerned, an approximation based on the $`qq`$ splitting function at two loops has been assumed, which was shown to be rather poor by the subsequent exact computation in . The coefficient $`D_2`$ was taken from its original computation in . There is a misprint in $`g_3(\lambda )`$ in in two terms proportional to $`A_1\beta _2`$: $`\mathrm{log}[1\lambda ]1/2\mathrm{log}[12\lambda ]`$ has to be multiplied by a factor 2, as found indeed in the recent recomputation of the $`\mu `$-independent terms . With the misprint, the terms proportional to $`A_1\beta _2`$ would indeed appear at $`\alpha ^3`$, while they have to appear only at order $`\alpha ^4`$, as shown correctly in the $`\alpha `$ expansion of the $`g_3`$ in eq. (42) of .
Let us note that the soft terms, i.e. the terms proportional to the coefficients $`A_i`$ and $`D_i`$, have the singularity closest to the origin in $`\lambda =1/2`$ while the collinear terms, proportional to $`B_i`$, have only a singularity in $`\lambda =1`$.
### 3.2 Inverse transform to physical space
The original form factor in $`u`$ space is recovered by an inverse Mellin transform:
$$\sigma (u;\alpha )=_{ci\mathrm{}}^{c+i\mathrm{}}\frac{dN}{2\pi i}(1u)^N\sigma _N(\alpha ),$$
(124)
where $`c`$ is a real constant chosen in such a way that all the singularities of $`\sigma _N`$ lie to the left of the integration contour. The inverse transform can be done to any given logarithmic accuracy in closed analytic form, where now the logarithmic accuracy is defined as before but in terms of powers of $`\alpha `$ and $`L=\mathrm{log}1/u`$ instead of $`l=\mathrm{log}N`$. To NNLO accuracy, one can write <sup>17</sup><sup>17</sup>17 A factor $`1u1`$ has been neglected in our leading twist accuracy. :
$$\mathrm{\Sigma }[u;\alpha ]=\frac{e^{Lg_1(\tau )+g_2(\tau )}}{\mathrm{\Gamma }\left[1h_1(\tau )\right]}\delta \mathrm{\Sigma },$$
(125)
where
$$\tau =\beta _0\alpha L$$
(126)
and we have defined
$$h_1(\tau )=\frac{d}{d\tau }\left[\tau g_1(\tau )\right]=g_1(\tau )+\tau g_1^{}(\tau ).$$
(127)
$`\delta \mathrm{\Sigma }`$ is a NNLO correction factor which can be set equal to one in NLO:
$$\delta \mathrm{\Sigma }_{NLO}=\mathrm{\hspace{0.17em}1}.$$
(128)
Its NNLO expression reads:
$$\delta \mathrm{\Sigma }=S/S|_{L0}$$
(129)
with
$$S=e^{\alpha g_3(\tau )}\left\{1+\beta _0\alpha g_2^{}(\tau )\psi \left[1h_1(\tau )\right]+\frac{1}{2}\beta _0\alpha h_1^{}(\tau )\left\{\psi ^2\left[1h_1(\tau )\right]\psi ^{}\left[1h_1(\tau )\right]\right\}\right\}.$$
(130)
In inhomogeneous terms were included in $`\delta \mathrm{\Sigma }`$, which have been subtracted here. $`\mathrm{\Gamma }(x)`$ is the Euler Gamma function and
$$\psi (x)=\frac{d}{dx}\mathrm{log}\mathrm{\Gamma }(x)$$
(131)
is the digamma function.
Expanding the r.h.s. of eq. (125) up to third order, one obtains the following relations:
$`G_{12}`$ $`=`$ $`{\displaystyle \frac{1}{2}}A_1;`$ (132)
$`G_{11}`$ $`=`$ $`(B_1+D_1);`$ (133)
$`G_{23}`$ $`=`$ $`{\displaystyle \frac{1}{2}}A_1\beta _0;`$ (134)
$`G_{22}`$ $`=`$ $`{\displaystyle \frac{1}{2}}A_2{\displaystyle \frac{1}{2}}\beta _0(B_1+2D_1){\displaystyle \frac{1}{2}}A_1^2z(2);`$ (135)
$`G_{21}`$ $`=`$ $`(B_2+D_2)A_1\left(B_1+D_1\right)z(2)A_1^2z(3);`$ (136)
$`G_{34}`$ $`=`$ $`{\displaystyle \frac{7}{12}}A_1\beta _0^2;`$ (137)
$`G_{33}`$ $`=`$ $`A_2\beta _0{\displaystyle \frac{1}{2}}A_1\beta _1{\displaystyle \frac{1}{3}}\beta _0^2\left(B_1+4D_1\right){\displaystyle \frac{3}{2}}A_1^2\beta _0z(2)+{\displaystyle \frac{1}{3}}A_1^3z(3);`$ (138)
$`G_{32}`$ $`=`$ $`{\displaystyle \frac{1}{2}}A_3\beta _0(B_2+2D_2){\displaystyle \frac{\beta _1}{2}}(B_1+2D_1)A_1A_2z(2){\displaystyle \frac{A_1\beta _0}{2}}(5B_1+7D_1)z(2)+`$ (139)
$`+`$ $`{\displaystyle \frac{A_{1}^{}{}_{}{}^{3}}{4}}z(4){\displaystyle \frac{9A_{1}^{}{}_{}{}^{2}\beta _0z(3)}{2}}+A_{1}^{}{}_{}{}^{2}(B_1+D_1)z(3),`$
where $`z(a)=_{n=1}^{\mathrm{}}1/n^a`$ is Riemann Zeta function with $`z(2)=\pi ^2/6=1.64493\mathrm{}`$, $`z(3)=1.20206\mathrm{}`$ and $`z(4)=\pi ^4/90=1.08232\mathrm{}`$. Note that the leading coefficients $`G_{23}`$ and $`G_{34}`$ involve products of the one-loop coefficients $`A_1`$ and $`\beta _0`$ only. The explicit expressions of the $`G_{ij}`$ read:
$`G_{12}`$ $`=`$ $`{\displaystyle \frac{C_F}{2\pi }};`$ (140)
$`G_{11}`$ $`=`$ $`{\displaystyle \frac{7C_F}{4\pi }};`$ (141)
$`G_{23}`$ $`=`$ $`{\displaystyle \frac{C_F}{8\pi ^2}}\left({\displaystyle \frac{11C_A}{3}}{\displaystyle \frac{2n_f}{3}}\right);`$ (142)
$`G_{22}`$ $`=`$ $`{\displaystyle \frac{C_F}{4\pi ^2}}\left[C_A\left({\displaystyle \frac{95}{72}}+z(2)\right){\displaystyle \frac{13n_f}{36}}2C_Fz(2)\right];`$ (143)
$`G_{21}`$ $`=`$ $`{\displaystyle \frac{C_F}{6\pi ^2}}\left[n_f\left({\displaystyle \frac{85}{24}}+z(2)\right)+C_A\left({\displaystyle \frac{905}{48}}{\displaystyle \frac{17}{2}}z(2){\displaystyle \frac{3z(3)}{2}}\right)+C_F\left({\displaystyle \frac{9}{16}}+6z(2)+3z(3)\right)\right];`$ (144)
$`G_{34}`$ $`=`$ $`{\displaystyle \frac{C_F}{48\pi ^3}}\left({\displaystyle \frac{847C_{A}^{}{}_{}{}^{2}}{36}}+{\displaystyle \frac{77C_An_f}{9}}{\displaystyle \frac{7n_{f}^{}{}_{}{}^{2}}{9}}\right);`$ (145)
$`G_{33}`$ $`=`$ $`{\displaystyle \frac{C_F}{4\pi ^3}}[{\displaystyle \frac{n_{f}^{}{}_{}{}^{2}}{108}}+C_An_f({\displaystyle \frac{20}{27}}{\displaystyle \frac{z(2)}{3}})+C_{A}^{}{}_{}{}^{2}({\displaystyle \frac{1261}{432}}+{\displaystyle \frac{11z(2)}{6}})+`$ (146)
$``$ $`{\displaystyle \frac{11z(2)}{2}}C_AC_F+C_Fn_f({\displaystyle \frac{1}{4}}+z(2))+{\displaystyle \frac{4C_{F}^{}{}_{}{}^{2}z(3)}{3}}];`$
$`G_{32}`$ $`=`$ $`{\displaystyle \frac{C_F}{4\pi ^3}}[n_{f}^{}{}_{}{}^{2}({\displaystyle \frac{275}{648}}{\displaystyle \frac{z(2)}{9}})+C_An_f({\displaystyle \frac{5399}{1296}}+{\displaystyle \frac{4z(2)}{3}}{\displaystyle \frac{z(3)}{6}})+`$ (147)
$`+`$ $`C_{A}^{}{}_{}{}^{2}\left({\displaystyle \frac{21893}{2592}}{\displaystyle \frac{119z(2)}{36}}+{\displaystyle \frac{77z(3)}{12}}{\displaystyle \frac{11z(4)}{4}}\right)+C_{F}^{}{}_{}{}^{2}\left(7z(3)+z(4)\right)+`$
$`+`$ $`C_Fn_f({\displaystyle \frac{19}{48}}{\displaystyle \frac{71z(2)}{36}}+z(3))+C_AC_F({\displaystyle \frac{11}{32}}+{\displaystyle \frac{685z(2)}{72}}11z(3)+5z(4))].`$
The numerical values of the coefficients show a good convergence of the perturbative series also in configuration space:
$`G_{12}`$ $`=`$ $`\mathrm{\hspace{0.17em}0.212207};`$ (148)
$`G_{11}`$ $`=`$ $`+\mathrm{\hspace{0.17em}0.742723};`$ (149)
$`G_{23}`$ $`=`$ $`\mathrm{\hspace{0.17em}0.185756}+\mathrm{\hspace{0.17em}0.011258}n_f=\mathrm{\hspace{0.17em}0.151982};`$ (150)
$`G_{22}`$ $`=`$ $`+\mathrm{\hspace{0.17em}0.152206}\mathrm{\hspace{0.17em}0.012196}n_f=+\mathrm{\hspace{0.17em}0.115618};`$ (151)
$`G_{21}`$ $`=`$ $`+\mathrm{\hspace{0.17em}0.628757}\mathrm{\hspace{0.17em}0.0427065}n_f=+\mathrm{\hspace{0.17em}0.500638};`$ (152)
$`G_{34}`$ $`=`$ $`\mathrm{\hspace{0.17em}0.189702}+\mathrm{\hspace{0.17em}0.022994}n_f\mathrm{\hspace{0.17em}0.0006968}n_f^2=\mathrm{\hspace{0.17em}0.126990};`$ (153)
$`G_{33}`$ $`=`$ $`\mathrm{\hspace{0.17em}0.349055}+\mathrm{\hspace{0.17em}0.033368}n_f\mathrm{\hspace{0.17em}0.0000995}n_f^2=\mathrm{\hspace{0.17em}0.249846};`$ (154)
$`G_{32}`$ $`=`$ $`+\mathrm{\hspace{0.17em}0.96117}\mathrm{\hspace{0.17em}0.09368}n_f+\mathrm{\hspace{0.17em}0.0025974}n_f^2=+\mathrm{\hspace{0.17em}0.703506},`$ (155)
where on the last member of the r.h.s. we have set $`n_f=3`$.
## 4 Distribution in the hadronic variables
The distribution in the hadronic variables $`u`$ and $`w`$ is obtained integrating the triple differential distribution (59) over the electron energy $`\overline{x}=1x`$. The integration range is
$$\overline{x}_1(w,u)\overline{x}\overline{x}_2(w,u),$$
(156)
where:
$$\overline{x}_1(w,u)=\frac{wu}{1+u}\mathrm{and}\overline{x}_2(w,u)=\frac{w}{1+u}.$$
(157)
Let us use the second method of integration of the triple-differential distribution discussed at the end of sec. (2), i.e. let us neglect at first the remainder function. Since the QCD form factor $`\sigma [u;\alpha (wm_b)]`$ does not depend on the electron energy $`\overline{x}`$, the integration only involves the coefficient function:
$$_{\overline{x}_1}^{\overline{x}_2}𝑑\overline{x}C(\overline{x},w;\alpha ).$$
(158)
We eliminate small terms $`O(u)`$ from the integral above by integrating over the range which is the limit $`u0`$ of (156): <sup>18</sup><sup>18</sup>18The latter are actually the integration regions for $`e^+e^{}q+\overline{q}+g`$ with massless quarks.
$$\overline{x}_1(w,0)\overline{x}\overline{x}_2(w,0).$$
(159)
In fact, these terms $`O(u)`$, when multiplied with the plus distributions of $`u`$ contained in the QCD form factor $`\sigma (u;\alpha )`$, give at worse terms of the form $`\mathrm{log}u`$, which miss the $`1/u`$ enhancement and therefore are to be considered as “small”. Let us define therefore the coefficient function of the double hadronic distribution as:
$$C_H(w;\alpha )=_0^w𝑑\overline{x}C(w,\overline{x};\alpha ),$$
(160)
having the usual $`\alpha `$ expansion
$$C_H(w;\alpha )=C_H^{(0)}(w)+\alpha C_H^{(1)}(w)+\alpha ^2C_H^{(2)}(w)+O(\alpha ^3).$$
(161)
One easily obtains:
$`C_H^{(0)}(w)`$ $`=`$ $`2w^2(32w);`$ (162)
$`C_H^{(1)}(w)`$ $`=`$ $`{\displaystyle \frac{C_F}{\pi }}w^2\left\{(94w)\mathrm{log}w+2(32w)\left[\mathrm{Li}_2(w)+\mathrm{log}w\mathrm{log}(1w){\displaystyle \frac{35}{8}}\right]\right\}.`$
The first two orders of the coefficient function vanish as $`w^2`$ for $`w0`$, implying a suppression of the states with a small hadronic energy (i.e. with a small hard scale), as anticipated in the introduction.
The resummed distribution in the hadronic variables $`u`$ and $`w`$ then reads:
$$\frac{1}{\mathrm{\Gamma }}\frac{d^2\mathrm{\Gamma }}{dudw}=C_H[w;\alpha (m_b)]\sigma [u;\alpha (wm_b)]+d_H[u,w;\alpha (m_b)],$$
(163)
where the remainder function has an expansion analogous to the one in the triple differential distribution:
$$d_H(u,w;\alpha )=\alpha d_H^{(1)}(u,w)+\alpha ^2d_H^{(2)}(u,w)+O(\alpha ^3).$$
(164)
Expanding to first order the above distribution and comparing (matching) with the known $`O(\alpha )`$ distribution, the following remainder function is obtained — an over-all factor $`C_F/\pi `$ is omitted:
$`d_H^{(1)}(w,u)=`$ $``$ $`{\displaystyle \frac{4w^6\mathrm{log}u}{\left(1+u\right)^7}}+{\displaystyle \frac{4w^2\left(32w\right)\mathrm{log}u}{1+u}}{\displaystyle \frac{32w^510w^6\mathrm{log}u}{\left(1+u\right)^6}}+`$ (165)
$`+`$ $`{\displaystyle \frac{3w^2\left(146w5w^2\right)2w^3\left(34w\right)\mathrm{log}u}{\left(1+u\right)^2}}+`$
$`+`$ $`{\displaystyle \frac{20w^3\left(2+w\right)\left(12w\right)w^4\left(918w2w^2\right)\mathrm{log}u}{\left(1+u\right)^4}}+`$
$`+`$ $`{\displaystyle \frac{64w^5+2w^4\left(36w4w^2\right)\mathrm{log}u}{\left(1+u\right)^5}}+`$
$``$ $`{\displaystyle \frac{4w^3\left(1015w2w^2\right)w^3\left(1213w6w^2\right)\mathrm{log}u}{\left(1+u\right)^3}}.`$
Eq. (163) provides a complete NLO resummation of the distribution in the two hadronic variables $`u`$ and $`w`$, from which the distribution in any other pair of hadronic variables can be obtained by a change of variables. One can insert in eq. (163) the NNLO form factor $`\sigma `$, whose properties have been discussed in sec. (3), allowing an approximate NNLO resummation. In fact, for a complete NNLO resummation, one also needs the second order corrections to the coefficient function $`C_H^{(2)}(w)`$ and the remainder function $`d_H^{(2)}(u,w)`$, which are unknown at present.
## 5 Hadron energy spectrum
The distribution in the total hadron energy $`w`$ is obtained by integrating the distribution in the hadronic variables (163). The integration range in $`u`$ is:
$$\mathrm{max}(0,w1)u\mathrm{\hspace{0.17em}1}.$$
(166)
Since the coefficient function $`C_H(w;\alpha )`$ does not depend on $`u`$, the integration only involves the QCD form factor and the remainder function:
$$\frac{1}{\mathrm{\Gamma }}\frac{d\mathrm{\Gamma }}{dw}=C_H(w;\alpha )\left\{1\theta (w1)\mathrm{\Sigma }[w1;\alpha (wm_b)]\right\}+_{\mathrm{max}(0,w1)}^1𝑑wd_H(u,w;\alpha ),$$
(167)
where $`\mathrm{\Sigma }(u;\alpha )`$ is the partially-integrated form factor defined in section (3).
Because of the $`\theta (w1)`$ multiplying $`\mathrm{\Sigma }(w1;\alpha )`$, there are large logarithms only for $`w>1`$, as anticipated in the qualitative discussion in the introduction. We may therefore consider the parts of the spectrum for $`w<1`$ and $`w>1`$ as two different spectra, merging in the point $`w=1`$. Let us consider the simpler case $`w<1`$ first. Since, as already noted, there are no large logarithms, no resummation is required and the $`O(\alpha )`$ fixed-order result coincides with the NLO one. There is no QCD form factor and therefore there is no way to distinguish between the coefficient function and the remainder function. The spectrum for $`w<1`$ can then be written as an ordinary $`\alpha `$ expansion:
$$\frac{1}{2\mathrm{\Gamma }}\frac{d\mathrm{\Gamma }}{dw}=L(w;\alpha )(w<1),$$
(168)
where:
$$L(w;\alpha )=L^{(0)}(w)+\alpha L^{(1)}(w)+\alpha ^2L^{(2)}(w)+O(\alpha ^3).$$
(169)
The first two orders read :
$`L^{(0)}(w)`$ $`=`$ $`w^2(32w);`$ (170)
$`L^{(1)}(w)`$ $`=`$ $`{\displaystyle \frac{C_F}{\pi }}\{w^2(32w)[{\displaystyle \frac{25}{8}}+\mathrm{Li}_2(1w)]+`$ (171)
$`+{\displaystyle \frac{1}{720}}w^2(4w^442w^3+585w^23720w+4860+1440w\mathrm{log}w3240\mathrm{log}w)\}.`$
Let us now consider the more interesting case $`w>1`$, where resummation is effective and one has to keep the resummed form of the distribution in (167). In a minimal scheme we have to subtract small terms from the first term on the r.h.s. of eq. (167), since the form factor must contain large logarithms only. This is done setting $`w=1`$ in the argument of the coupling entering the form factor $`\mathrm{\Sigma }`$ as well as in the coefficient function $`C_H`$, obtaining the simpler expression:
$$\frac{1}{\mathrm{\Gamma }}\frac{d\mathrm{\Gamma }}{dw}=C_H(1;\alpha )\left\{1\mathrm{\Sigma }[w1;\alpha (m_b)]\right\}+\mathrm{},$$
(172)
where the dots denote terms not containing large logs of $`w1`$. Let us prove the legitimacy of the transformation from (167) to (172). As far as the argument of the coupling is concerned, we expand the QCD form factor $`\mathrm{\Sigma }`$ in powers of $`\alpha (wm_b)`$. One obtains terms of the form
$$\alpha (wm_b)\mathrm{log}^2(w1)=\alpha (m_b)\mathrm{log}^2(w1)\mathrm{\hspace{0.17em}2}\beta _0\alpha (m_b)^2\mathrm{log}w\mathrm{log}^2(w1)+\mathrm{\hspace{0.17em}4}\beta _0^2\alpha (m_b)^3\mathrm{log}^2w\mathrm{log}^2(w1)+\mathrm{},$$
(173)
where on the r.h.s. an expansion of $`\alpha (wm_b)`$ around the point $`w=1`$ has been performed. All the terms on the r.h.s except the first one vanish for $`w1^+`$, therefore they are not large logarithms and can be dropped. The only large logarithm is the first term on the r.h.s., which is obtained by setting $`w=1`$ in the coupling in the original expression on the l.h.s. All this implies that the coupling can be evaluated in the infrared-singular point $`w=1`$. As far as the coefficient function is concerned, one just notices that the neglected terms,
$$\left[C_H(w;\alpha )C_H(1;\alpha )\right]\left\{1\mathrm{\Sigma }[w1;\alpha (m_b)]\right\},$$
(174)
are again vanishing for $`w1^+`$, because $`C_H(w;\alpha )C_H(1;\alpha )=O(w1)`$ and therefore can be neglected in this limit.
In NLO one has also to add a remainder function to be determined via a matching procedure. That, as already discussed in other cases, is in order to take into account also the region $`w1O(1)`$. One then has the resummed expression:
$$\frac{1}{2\mathrm{\Gamma }}\frac{d\mathrm{\Gamma }}{dw}=C_W\left(\alpha \right)\left\{1\mathrm{\Sigma }[w1;\alpha (m_b)]\right\}+H(w;\alpha )(w>1),$$
(175)
where we have defined:
$$C_W(\alpha )\frac{1}{2}C_H(1;\alpha _s).$$
(176)
The coefficient function and the remainder function have a standard $`\alpha `$ expansion:
$`C_W(\alpha )`$ $`=`$ $`1+\alpha C_W^{(1)}++\alpha ^2C_W^{(2)}+O(\alpha ^3),`$ (177)
$`H(w;\alpha )`$ $`=`$ $`\alpha H^{(1)}(w)+\alpha ^2H^{(2)}(w)+O(\alpha ^3).`$ (178)
The first order correction to the coefficient function reads:
$$C_W^{(1)}=\frac{C_F}{\pi }\left(\frac{\pi ^2}{6}\frac{35}{8}\right)=\mathrm{\hspace{0.17em}1.15868}.$$
(179)
Note that $`C_W^{(1)}`$ is negative and has a rather large size; for $`\alpha (m_b)=0.22`$ it gives a negative correction of $`\mathrm{\hspace{0.17em}25}\%`$. By using the matching procedure described at the end of sec. (3), we obtain:
$`H^{(1)}(w)`$ $`=`$ $`{\displaystyle \frac{C_F}{\pi }}\{{\displaystyle \frac{1}{2}}(2w+1)(w1)^2\mathrm{log}^2(w1){\displaystyle \frac{1}{3}}(w1)(2w^2w4)\mathrm{log}(w1)+`$ (180)
$`w^2(32w)\left[2\mathrm{L}\mathrm{i}_2\left(1w\right)+2\mathrm{log}(w1)\mathrm{log}w+{\displaystyle \frac{\pi ^2}{6}}\right]+`$
$`+{\displaystyle \frac{1}{720}}(2w)(4w^534w^4+517w^32946w^2+3798w+1248)\}.`$
The above function is positive in all the kinematical range $`1<w<2`$ and goes to zero for $`w2`$, as expected on the basis of the vanishing of the phase space in this point.
Let us make a few remarks about eq. (175). If we expand the r.h.s. of eq. (175) in powers of $`\alpha `$, we find that $`C_W^{(1)}`$ only appears in order $`\alpha ^2`$ — this occurs because the form factor multiplying the coefficient function is in this case $`1\mathrm{\Sigma }=O(\alpha )`$ and not $`\mathrm{\Sigma }=O(1)`$. At present, only a full $`O(\alpha )`$ computation is available, implying that $`C_W^{(1)}`$ cannot be determined by the matching: only the remainder function can be fixed by this procedure. The value of $`C_W^{(1)}`$ came out “automatically” as a consequence of our resummation formula (see. eq. (176)). There is however another method to fix $`C_W^{(1)}`$: we require that the resummed spectrum is continuous in $`w=1`$. Since $`\mathrm{\Sigma }(w1)0`$ for $`w1^+`$, we obtain the equation:
$$1+\alpha L^{(1)}(1)=\mathrm{\hspace{0.17em}1}+\alpha \left[C_W^{(1)}+H^{(1)}(1)\right],$$
(181)
to be solved in $`C_W^{(1)}`$:
$$C_W^{(1)}=L^{(1)}(1)H^{(1)}(1)$$
(182)
and giving again the value (179). The condition of continuity of the resummed spectrum in $`w=1`$ is very reasonable from the physical viewpoint and it is remarkable that the two methods give the same value for the coefficient function.
Even though we are considering a differential spectrum, its resummation involves, as we have explicitly seen, the partially integrated form factor. $`\mathrm{\Sigma }`$ usually enters event fractions in expressions of the form
$$R(y;\alpha )=C(\alpha )\mathrm{\Sigma }(y;\alpha )+D(y;\alpha ),$$
(183)
with a remainder function vanishing for $`y0`$, where $`y`$ is a general kinematical variable entering the large logarithms $`\mathrm{log}1/y`$. In the case of the hadron energy spectrum, its resummation is different from (183) because it involves the combination $`1\mathrm{\Sigma }`$ instead of $`\mathrm{\Sigma }`$: there is an additive constant, namely one, which makes the spectrum non vanishing for $`w1^+`$, as it should. It seem however reasonable to impose the vanishing of the remainder function $`H(w;\alpha )`$ for $`w1^+`$ also in this case. The previous factorization scheme does not satisfy this condition, because:
$$H^{(1)}(1)=\frac{C_F}{\pi }\left(\frac{2587}{720}\frac{\pi ^2}{6}\right).$$
(184)
We can construct an improved scheme satisfying this condition by introducing two coefficient functions instead of one:
$$\frac{1}{2\mathrm{\Gamma }}\frac{d\mathrm{\Gamma }}{dw}=C_{W1}\left(\alpha \right)\left\{1C_{W2}\left(\alpha \right)\mathrm{\Sigma }[w1;\alpha (m_b)]+\stackrel{~}{H}(w;\alpha )\right\}(\mathrm{improved}\mathrm{scheme},w>1),$$
(185)
where the new remainder function, vanishing in $`w=1`$, reads:
$$\stackrel{~}{H}(w;\alpha )=H(w;\alpha )H(1;\alpha ).$$
(186)
The coefficient functions have the usual fixed-order expansions:
$`C_{W1}\left(\alpha \right)`$ $`=`$ $`1+\alpha C_{W1}^{(1)}+\alpha ^2C_{W1}^{(2)}+O(\alpha ^3);`$ (187)
$`C_{W2}\left(\alpha \right)`$ $`=`$ $`1+\alpha C_{W2}^{(1)}+\alpha ^2C_{W2}^{(2)}+O(\alpha ^3).`$ (188)
By imposing the continuity in $`w=1`$ as in the previous scheme, we obtain for the first coefficient function at first order in $`\alpha `$:
$$C_{W1}^{(1)}=L^{(1)}(1)=\frac{C_F}{\pi }\frac{563}{720}=\mathrm{\hspace{0.17em}0.331868}.$$
(189)
The second coefficient function is obtained by imposing the usual matching with the first order computation:
$$C_{W2}^{(1)}=H^{(1)}(1)=\frac{C_F}{\pi }\left(\frac{2587}{720}\frac{\pi ^2}{6}\right)=\mathrm{\hspace{0.17em}0.826808}.$$
(190)
The improved resummed expression (185) is positive in all the kinematical range $`1<w<2`$ and vanishes for $`w2`$.
We can compare the hadron energy spectrum for $`w>1`$ given in eq. (175) or in eq. (185) with the hadron mass distribution in the radiative decay (5) given in eq. (50). The hadron energy distribution contains $`\mathrm{\Sigma }`$, i.e. just the integral of the form factor $`\sigma `$ entering the radiative decay spectrum. The hadron energy spectrum is therefore a very good quantity on the theoretical side — it is exceptional in this respect — being directly connected, via integration, to the radiative decay. By that we mean that the connection between the two spectra only involves short-distance coefficients. As show in , this is to be contrasted with the case of other single-differential spectra.
### 5.1 Average energy
As discussed in the introduction, the infrared singularity in $`w=1`$ of the $`O(\alpha )`$ spectrum is integrable, so one can calculate directly the average hadronic energy as a truncated expansion in $`\alpha `$:
$$w=\frac{7}{10}\left[1+\frac{\alpha C_F}{\pi }\frac{137}{840}\right]=\mathrm{\hspace{0.17em}0.71}.$$
(191)
The $`O(\alpha )`$ correction is very small, of the order of $`1\%`$, due to a large cancellation between the contribution for $`w<1`$, which is negative, and the one for $`w>1`$, which is positive. Setting for instance $`m_b=m_B`$ one obtains in leading order:
$$E_X=\frac{7}{10}\frac{m_B}{2}=\mathrm{\hspace{0.17em}1.843}\mathrm{GeV}$$
(192)
with a tiny first-order correction of $`+\mathrm{\hspace{0.17em}26}`$ MeV. This quantity can be directly compared with the experimental value. In the radiative decay (5) there is a larger final hadronic energy: in lowest order
$$E_X_{BX_s\gamma }=\frac{m_B}{2}=\mathrm{\hspace{0.17em}2.634}\mathrm{GeV}.$$
(193)
The average hadronic energy is $`30\%`$ larger in the radiative decay than in the semileptonic decay, in line with the qualitative discussion about the differences of the two decays given in the introduction.
### 5.2 Upper cut on hadron masses
In experimental analysis an upper cut on invariant masses
$$m_X<\overline{m}_X$$
(194)
is imposed in order to kill the large background from semileptonic $`bc`$ transitions. Let us define:
$$k=\mathrm{\hspace{0.17em}2}\frac{\overline{m}_X}{m_b}.$$
(195)
In practice, $`\overline{m}_X=1.6÷1.8`$ GeV, so we can assume $`k<1`$. A leading order evaluation of the spectrum with the above cut gives:
$$\frac{1}{2\mathrm{\Gamma }}\frac{d\mathrm{\Gamma }}{dw}=\{\begin{array}{cc}w^2(32w)\left\{\theta (kw)+\theta (wk)\mathrm{\Sigma }[\frac{1\sqrt{1(k/w)^2}}{1+\sqrt{1(k/w)^2}};\alpha (wm_b)]\theta (w1)\mathrm{\Sigma }[w1;\alpha (m_b)]\right\}\hfill & w<w_M\hfill \\ 0\hfill & w>w_M\hfill \end{array}$$
(196)
where
$$w_M=\mathrm{\hspace{0.17em}1}+\frac{k^2}{4}$$
(197)
is the maximal hadronic energy above which the spectrum vanishes; as expected on physical ground, cutting large hadron masses also acts as an upper cut on hadron energies. The spectrum is continuous in $`w=w_M`$ and it develops large logarithms for $`k0`$. Let us observe that the argument of the first QCD form factor $`\mathrm{\Sigma }`$ has a similar form to the variable $`u`$ defined in eq. (36). In fact,
$$\left(\frac{k}{w}\right)^2=\left(\frac{\overline{m}_X}{E_X}\right)^2$$
(198)
is the analogue of the variable $`4y`$ with $`y`$ defined in eq. (37).
## 6 Distribution in hadron and electron energies
In this section we derive the distribution in the hadron and electron energies $`w`$ and $`\overline{x}`$ by integrating the triple differential distribution (59) over $`u`$. In general, there are two independent energies in the semileptonic decay (7). That is because the hadronic final state $`X_u`$ is basically a pseudoparticle, i.e. a single entity possessing an energy $`E_X`$ and a (variable) mass $`m_X`$. We have therefore 3 particles/pseudoparticles in the final state and 3 energies, related by energy conservation:
$$x_e+x_\nu +w=\mathrm{\hspace{0.17em}2},$$
(199)
where
$$x_\nu =\frac{2E_\nu }{m_b}$$
(200)
and we have written $`x_e`$ instead of $`x`$ for aesthetical reasons. Since the neutrino energy is not usually measured, let us take as independent energies the electron and the hadron energies. We have to integrate over $`u`$ in the range
$$\mathrm{max}[0,w1]u\mathrm{min}[\frac{w\overline{x}}{\overline{x}},\frac{\overline{x}}{w\overline{x}}].$$
(201)
As in the previous section, let us use the second method of integration, i.e. let us omit at first the remainder function. Since the coefficient function $`C(x,w;\alpha )`$ does not depend on $`u`$, the integration only involves the form factor $`\sigma `$ — a complementary situation with respect to the one in the previous section — and we obtain:
$`{\displaystyle \frac{1}{\mathrm{\Gamma }}}{\displaystyle \frac{d^2\mathrm{\Gamma }}{dxdw}}`$ $`=`$ $`C[\overline{x},w;\alpha (m_b)]\{\theta (2\overline{x}w)\mathrm{\Sigma }[{\displaystyle \frac{w\overline{x}}{\overline{x}}};\alpha (wm_b)]+\theta (w2\overline{x})\mathrm{\Sigma }[{\displaystyle \frac{\overline{x}}{w\overline{x}}};\alpha (wm_b)]+`$ (202)
$`\theta (w1)\mathrm{\Sigma }[w1;\alpha (m_b)]\}+\mathrm{},`$
where the dots denote non logarithmic terms to be included later. The decay (7) involves an hadronic subprocess with a heavy quark decaying into a light quark evolving later into a jet. Hadron dynamics is therefore symmetric under the exchange of the electron and the neutrino momenta, since it is “blind” to $`W`$ decay. That is clearly seen by expressing $`w`$ through $`x_\nu `$ by means of eq. (199):
$`{\displaystyle \frac{1}{\mathrm{\Gamma }}}{\displaystyle \frac{d^2\mathrm{\Gamma }}{dx_edx_\nu }}`$ $`=`$ $`C[x_e,x_\nu ;\alpha (m_b)]\{\theta (x_\nu x_e)\mathrm{\Sigma }[{\displaystyle \frac{1x_\nu }{1x_e}};\alpha ((2x_ex_\nu )m_b)]+`$ (203)
$`+\theta (x_ex_\nu )\mathrm{\Sigma }[{\displaystyle \frac{1x_e}{1x_\nu }};\alpha ((2x_ex_\nu )m_b)]+`$
$`\theta (1x_ex_\nu )\mathrm{\Sigma }[1x_ex_\nu ;\alpha (m_b)]\}+\mathrm{}.`$
Soft-gluon dynamics — i.e. the expression above in curly brackets — is symmetric under exchange of $`x_e`$ with $`x_\nu `$. The coefficient function $`C[x_e,x_\nu ;\alpha (m_b)]`$ however is not symmetric under the exchange of the lepton energies because it does depend on the whole process, involving the decay of the $`W`$ boson into the lepton pair, and not only on the hadronic subprocess.
To proceed with resummation, however, let us go back to the more familiar variable $`w`$, i.e. to eq. (202). Large logarithms can in principle be obtained by sending to zero the argument of any of the QCD form factors $`\mathrm{\Sigma }`$’s entering (202), i.e. in the following three cases:
$$1.w\overline{x}\mathrm{\hspace{0.17em}0};2.\overline{x}\mathrm{\hspace{0.17em}0};3.w\mathrm{\hspace{0.17em}1}^+.$$
(204)
The coefficient function $`C[\overline{x},w;\alpha (m_b)]`$ vanishes in the first limit as $`O(w\overline{x})`$, implying that in this case there are actually no large logarithms. This limit corresponds to $`E_\nu m_b/2`$, a point where the tree-level spectrum vanishes suppressing soft-gluon effects. The only relevant limits are therefore the second and the third ones. It is therefore natural to write a factorization formula dropping the form factor not associated to large logarithms:
$$\frac{1}{\mathrm{\Gamma }}\frac{d^2\mathrm{\Gamma }}{dxdw}=C[\overline{x},w;\alpha (m_b)]\left\{\mathrm{\Sigma }[\overline{x}/w;\alpha (wm_b)]\theta (w1)\mathrm{\Sigma }[w1;\alpha (m_b)]\right\}+\mathrm{}.$$
(205)
We have taken the limit $`\overline{x}0`$ in the theta functions containing $`\overline{x}`$ in the argument.
Let us consider separately the cases $`w1`$ and $`w>1`$. In the simpler case $`w1`$ <sup>19</sup><sup>19</sup>19 Note that this case is a “complication” of the analogous case for the single distribution in $`w`$, where the integration over $`\overline{x}`$ has been made and therefore there are no large logarithms of $`\overline{x}`$. there is a single form factor and one can write a factorized expression of the form:
$$\frac{1}{\mathrm{\Gamma }}\frac{d^2\mathrm{\Gamma }}{dxdw}=C_L(\overline{x},w;\alpha )\mathrm{\Sigma }[\overline{x}/w;\alpha (wm_b)]+d_<(w,\overline{x};\alpha )(w<1).$$
(206)
We require that the remainder function vanishes for $`\overline{x}0`$:
$$\underset{\overline{x}0}{lim}d_<(w,\overline{x};\alpha )=\mathrm{\hspace{0.17em}0}.$$
(207)
The coefficient function $`C_L(\overline{x},w;\alpha )`$ can be taken as:
$`C_L^{(0)}(w,\overline{x})`$ $`=`$ $`12(w\overline{x})(1+\overline{x}w);`$ (208)
$`C_L^{(1)}(w,\overline{x})`$ $`=`$ $`{\displaystyle \frac{C_F}{\pi }}12(w\overline{x})(1+\overline{x}w)\left[\mathrm{Li}_2(w)+\mathrm{log}w\mathrm{log}(1w){\displaystyle \frac{3}{2}}\mathrm{log}w{\displaystyle \frac{w\mathrm{log}w}{2(1w)}}{\displaystyle \frac{35}{8}}\right].`$ (209)
In $`C_L^{(0)}(w,\overline{x})`$ we have put the factor $`12(w\overline{x})(1+\overline{x}w)`$, equal to the spectrum in lowest order, in order to have a vanishing remainder function in $`O(\alpha ^0)`$: this is a non minimal choice, since the minimal choice would imply to set $`\overline{x}=0`$ in the coefficient function. We have inserted a similar factor also in $`C_L^{(1)}(w,\overline{x})`$, in order to have a simple multiplicative form of the correction <sup>20</sup><sup>20</sup>20We could have taken as coefficient function the original one $`C(\overline{x},w;\alpha )`$ as well, which however does not always contain the factor $`12(w\overline{x})(1+\overline{x}w)`$.. As in previous cases, by matching with the full $`O(\alpha )`$ result , we determine the remainder function
$$d_<(w,\overline{x};\alpha )=\alpha d_<^{(1)}(w,\overline{x})+\alpha ^2d_<^{(2)}(w,\overline{x})+O(\alpha ^3).$$
(210)
Omitting the over-all factor $`C_F/\pi `$, we obtain for the leading contribution:
$`d_<^{(1)}(w,\overline{x})`$ $`=`$ $`{\displaystyle \frac{1}{10}}(w\overline{x})\overline{x}(210+280w10w^2+2w^360\overline{x}125w\overline{x}7w^2\overline{x}+15\overline{x}^2+`$ (211)
$`+`$ $`32w\overline{x}^215\overline{x}^3)+`$
$`+`$ $`{\displaystyle \frac{1}{5\left(1+w\right)}}(45w+60w^220w^3+10w^46w^5+w^615\overline{x}+135w\overline{x}255w^2\overline{x}+`$
$`+`$ $`85w^3\overline{x}+25w^4\overline{x}5w^5\overline{x}15\overline{x}^2+45w\overline{x}^2+75w^2\overline{x}^285w^3\overline{x}^2+10w^4\overline{x}^2)\mathrm{log}w+`$
$``$ $`6\left(1+w\overline{x}\right)\left(w\overline{x}\right)\mathrm{log}^2w+\mathrm{\hspace{0.17em}6}\left(1+w\overline{x}\right)\left(w\overline{x}\right)\mathrm{log}^2(w\overline{x})+`$
$``$ $`{\displaystyle \frac{1}{5}}(w\overline{x})(4515w+5w^25w^3+w^4+15\overline{x}10w\overline{x}+15w^2\overline{x}4w^3\overline{x}+5\overline{x}^2+`$
$``$ $`15w\overline{x}^2+6w^2\overline{x}^2+5\overline{x}^34w\overline{x}^3+\overline{x}^4)\mathrm{log}(w\overline{x})+`$
$``$ $`{\displaystyle \frac{1}{5}}\overline{x}(60180w+120w^2+60\overline{x}15w\overline{x}45w^2\overline{x}+5\overline{x}^220w\overline{x}^2+10w^2\overline{x}^2+`$
$`+`$ $`5\overline{x}^35w\overline{x}^3+\overline{x}^4)\mathrm{log}\overline{x}+`$
$`+`$ $`12\left(1+w\overline{x}\right)\left(w\overline{x}\right)\mathrm{log}w\mathrm{log}\overline{x}12\left(1+w\overline{x}\right)\left(w\overline{x}\right)\mathrm{log}(w\overline{x})\mathrm{log}\overline{x}.`$
Let us now consider the case $`w>1`$:
$$\frac{1}{\mathrm{\Gamma }}\frac{d^2\mathrm{\Gamma }}{dxdw}=C(\overline{x},w;\alpha )\left\{\mathrm{\Sigma }[\overline{x}/w;\alpha (wm_b)]\mathrm{\Sigma }[\mathrm{\Delta }w;\alpha (m_b)]\right\}+\mathrm{}(w>1),$$
(212)
where we have defined
$$\mathrm{\Delta }w=w\mathrm{\hspace{0.17em}1}>\mathrm{\hspace{0.17em}0}.$$
(213)
There are two form factors and large logarithms can be obtained in the following three kinematical configurations:
1. $`\overline{x}\mathrm{\Delta }w1`$: large logarithms of the form $`\alpha ^n\mathrm{log}^k\overline{x}`$ have to be resummed;
2. $`\mathrm{\Delta }w\overline{x}1`$: large logarithms of the form $`\mathrm{log}\mathrm{\Delta }w\mathrm{log}\overline{x}`$ have to be resummed;
3. $`\mathrm{\Delta }w\overline{x}1`$: large logarithms of the form $`\mathrm{log}\mathrm{\Delta }w`$ have to be resummed.
The first case is kinematically forbidden because
$$\mathrm{\Delta }w\overline{x}.$$
(214)
The second case does not give large logarithms because the coefficient function $`C(\overline{x},w;\alpha )`$ vanishes linearly in this limit:
$$C(\lambda \overline{x},1+\lambda \mathrm{\Delta }w;\alpha )=O(\lambda )\mathrm{for}\lambda 0.$$
(215)
The only relevant limit is therefore the third one, implying that one can drop the form factor $`\mathrm{\Sigma }(\overline{x}/w;\alpha )`$. We propose then a resummed form for this distribution which is a generalization of that one for the hadron energy spectrum:
$$\frac{1}{\mathrm{\Gamma }}\frac{d^2\mathrm{\Gamma }}{dxdw}=C_{XW1}(\overline{x};\alpha )\left\{1C_{XW2}(\overline{x};\alpha )\mathrm{\Sigma }[\mathrm{\Delta }w;\alpha (m_b)]\right\}+d_>(\mathrm{\Delta }w,\overline{x};\alpha )(w>1).$$
(216)
We require that the remainder function vanishes for $`\mathrm{\Delta }w0^+`$:
$$\underset{\mathrm{\Delta }w0^+}{lim}d_>(\mathrm{\Delta }w,\overline{x};\alpha )=\mathrm{\hspace{0.17em}0}.$$
(217)
The first coefficient function is obtained by imposing the continuity of the spectrum for $`w1`$ from both sides $`w<1`$ and $`w>1`$ and for any $`\overline{x}`$ <sup>21</sup><sup>21</sup>21 This continuity condition, which involves a single point $`w=1`$ for the hadron energy spectrum, involves in this more complicated case the line $`(w=1,\overline{x})`$.. We obtain:
$$C_{XW1}(\overline{x};\alpha )=C_L(\overline{x},1;\alpha )\mathrm{\Sigma }(\overline{x};\alpha )+d_<(1,\overline{x};\alpha ).$$
(218)
We can expand in the above equation $`\mathrm{\Sigma }(\overline{x};\alpha )`$ in powers of $`\alpha `$ (up to first order) because the coefficient function for $`w=1`$, $`C_L(\overline{x},1;\alpha )`$, vanishes linearly for $`\overline{x}0`$, killing the large logarithms in the form factor. We then obtain:
$$C_{XW1}(\overline{x};\alpha )=C_{XW1}^{(0)}(\overline{x})+\alpha C_{XW1}^{(1)}(\overline{x})+\alpha ^2C_{XW2}^{(2)}(\overline{x})+O(\alpha ^3)$$
(219)
where
$`C_{XW1}^{(0)}(\overline{x})`$ $`=`$ $`12(1\overline{x})\overline{x};`$ (220)
$`C_{XW1}^{(1)}(\overline{x})`$ $`=`$ $`{\displaystyle \frac{C_F}{\pi }}\{{\displaystyle \frac{1}{10}}(1\overline{x})\overline{x}(587+192\overline{x}47\overline{x}^2+15\overline{x}^3){\displaystyle \frac{1}{5}}\overline{x}(105105\overline{x}5\overline{x}^2+\overline{x}^4)\mathrm{log}\overline{x}+`$ (221)
$``$ $`{\displaystyle \frac{1}{5}}\left(1\overline{x}\right)\left(31+16\overline{x}4\overline{x}^2+\overline{x}^3+\overline{x}^4\right)\mathrm{log}(1\overline{x})6\left(1\overline{x}\right)\overline{x}\mathrm{log}^2(1\overline{x})+`$
$`+`$ $`12(1\overline{x})\overline{x}\mathrm{log}(1\overline{x})\mathrm{log}\overline{x}6(1\overline{x})\overline{x}\mathrm{log}^2\overline{x}+12(1\overline{x})\overline{x}z(2)\}.`$
The second coefficient function $`C_{XW2}(\overline{x};\alpha )`$ is obtained by matching with the fixed-order distribution in the limit $`\mathrm{\Delta }w0^+`$:
$$C_{XW2}(\overline{x};\alpha )=C_{XW2}^{(0)}(\overline{x})+\alpha C_{XW2}^{(1)}(\overline{x})+\alpha ^2C_{XW2}^{(2)}(\overline{x})+O(\alpha ^3)$$
(222)
with
$`C_{XW2}^{(0)}(\overline{x})`$ $`=`$ $`1;`$ (223)
$`C_{XW2}^{(1)}(\overline{x})`$ $`=`$ $`{\displaystyle \frac{C_F}{\pi }}\{{\displaystyle \frac{1}{120}}(62192\overline{x}+47\overline{x}^215\overline{x}^3)+{\displaystyle \frac{1}{60\overline{x}}}(31+16\overline{x}4\overline{x}^2+\overline{x}^3+\overline{x}^4)\mathrm{log}(1\overline{x})+`$ (224)
$`+`$ $`{\displaystyle \frac{1}{2}}\mathrm{log}^2(1\overline{x})+{\displaystyle \frac{1}{60\left(1\overline{x}\right)}}\left(105105\overline{x}5\overline{x}^2+\overline{x}^4\right)\mathrm{log}\overline{x}\mathrm{log}(1\overline{x})\mathrm{log}\overline{x}+`$
$`+`$ $`{\displaystyle \frac{1}{2}}\mathrm{log}^2\overline{x}\}.`$
The remainder function $`d_>(w,\overline{x};\alpha )`$ is obtained by matching with the fixed-order distribution for $`\mathrm{\Delta }wO(1)`$:
$$d_>(w,\overline{x};\alpha )=\alpha d_>^{(1)}(w,\overline{x})+\alpha ^2d_>^{(2)}(w,\overline{x})+O(\alpha ^3),$$
(225)
where, omitting the overall factor $`C_F/\pi `$:
$`d_>^{(1)}(w,\overline{x})`$ $`=`$ $`{\displaystyle \frac{1}{10}}(1+w)(1\overline{x})(75+142w7w^2+2w^3212\overline{x}105w\overline{x}9w^2\overline{x}+`$ (226)
$`+`$ $`132\overline{x}^2+39w\overline{x}^247\overline{x}^3)+{\displaystyle \frac{1}{5}}(1+w\left)\right(499w+w^24w^3+w^4+140\overline{x}+`$
$`+`$ $`120w\overline{x}+15w^2\overline{x}5w^3\overline{x}65\overline{x}^265w\overline{x}^2+10w^2\overline{x}^2)\mathrm{log}(1+w)+`$
$``$ $`6\left(1+w\right)\left(w2\overline{x}\right)\mathrm{log}^2(1+w)+{\displaystyle \frac{1}{5}}\left(1\overline{x}\right)\left(31+16\overline{x}4\overline{x}^2+\overline{x}^3+\overline{x}^4\right)\mathrm{log}(1\overline{x})+`$
$`+`$ $`6\left(1\overline{x}\right)\overline{x}\mathrm{log}^2(1\overline{x})+6\left(1+w\overline{x}\right)\left(w\overline{x}\right)\mathrm{log}^2(w\overline{x})+`$
$``$ $`\left(1+w\right)\left(21w+30\overline{x}+24w\overline{x}12\overline{x}^29w\overline{x}^22\overline{x}^3+2w\overline{x}^3\overline{x}^4\right)\mathrm{log}\overline{x}+`$
$``$ $`12\left(1\overline{x}\right)\overline{x}\mathrm{log}(1\overline{x})\mathrm{log}\overline{x}+6\left(1+w\right)\left(w2\overline{x}\right)\mathrm{log}^2\overline{x}+`$
$``$ $`{\displaystyle \frac{1}{5}}(w\overline{x})(4515w+5w^25w^3+w^4+15\overline{x}10w\overline{x}+15w^2\overline{x}4w^3\overline{x}+5\overline{x}^2+`$
$``$ $`15w\overline{x}^2+6w^2\overline{x}^2+5\overline{x}^34w\overline{x}^3+\overline{x}^4)\mathrm{log}(w\overline{x})+`$
$``$ $`12\left(1+w\overline{x}\right)\left(w\overline{x}\right)\mathrm{log}\overline{x}\mathrm{log}(w\overline{x}).`$
To summarize, we have presented a complete NLO resummation of the distribution in the hadron and electron energies $`w`$ and $`x`$, which is a generalization of the resummation of the hadron energy spectrum of the previous section. Resummation takes a different form in the cases $`w1`$ and $`w>1`$. In the first case there is a series of threshold logarithms of the form
$$\alpha ^n\mathrm{log}^k\frac{\overline{x}}{w}(w<1),$$
(227)
while in the second case the infrared logarithms are of the form
$$\alpha ^n\mathrm{log}^k(w1)(w>1).$$
(228)
Unlike the distribution in sec. 4, we have here a differential distribution involving the partially-integrated form factor $`\mathrm{\Sigma }`$.
## 7 Conclusions
It is a rather old idea that semi-inclusive $`B`$ decays can be related to each other because of some universal long-distance component . We have presented in this paper a critical analysis of this idea, based on a resummation formula for the triple differential distribution in the semileptonic decay (7). Long-distance effects manifest themselves in perturbation theory in the form of series of large infrared logarithms, coming from the multiple emission of soft and/or collinear gluons. The universality of long-distance effects has therefore to show up in perturbation theory in the form of identical series of large logarithms in different distributions. Semi-inclusive $`B`$ decays have been defined in all generality as decays of the form
$$BX_q+(\mathrm{non}\mathrm{QCD}\mathrm{partons}),$$
(229)
in the kinematical region close to the threshold $`m_X=0`$, i.e. for
$$m_XE_X.$$
(230)
We have shown that semileptonic distributions are naturally divided into two classes.
The first class contains distributions which are not integrated over the hadronic energy $`E_X`$ and consequently have a long-distance structure similar to the one in radiative decays (5). These are the (simpler) distributions to attack and have been treated in this paper. We have resummed to next-to-leading order:
1. the distribution in the hadronic energy $`E_X`$ and in the variable $`u`$ defined in sec. (3), which is basically the ratio $`m_X^2/(4E_X^2)`$, i.e. the hadron invariant mass squared in unit of the hard scale;
2. the hadron energy distribution, which is a case of the so-called Sudakov shoulder. This is the only single distribution which can be related to the radiative decay via short-distance factors only. The large logarithms which appear in this distribution are indeed equal to the ones which appear in the radiative decay (5). We have studied in detail the relation between the hadron energy spectrum and the photon spectrum in the radiative decay. It is remarkable that the large logarithms in the hadron energy spectrum occur at $`E_X=m_b/2`$, i.e. when the hard scale $`Q=2E_X`$ equals $`m_b`$, as in the radiative decays;
3. the distribution in the hadron and in the charged lepton energies, which contains two different classes of large logarithms according to the cases $`w1`$ or $`w>1`$. The resummation of this distribution is the most complicated and is a generalization of the resummation of the hadron energy spectrum.
The second class contains semileptonic distributions in which the hadronic energy is integrated over, such as for example the hadronic mass distribution or the charged lepton energy distribution. These distributions have a complicated logarithmic structure, which is not simply related to the one in the radiative decay and there is not a pure short-distance relation with the radiative decay spectrum. The resummation of these distributions to NLO is presented in .
Acknowledgments
One of us (U.A.) wishes to thank R. Faccini for discussions. |
warning/0507/math0507027.html | ar5iv | text | # On the nonexistence of fat partially hyperbolic horseshoes
## 1. Introduction
Let $`M`$ be a Riemannian manifold. We use $`\mathrm{Leb}`$ to denote a normalized volume form defined on the Borel sets of $`M`$ that we call Lebesgue measure. Given a submanifold $`\gamma M`$ we use $`\mathrm{Leb}_\gamma `$ to denote the measure on $`\gamma `$ induced by the restriction of the Riemannian structure to $`\gamma `$.
Let $`f:MM`$ be a $`C^1`$ diffeomorphism, and let $`\mathrm{\Lambda }M`$ be a compact invariant set, i.e. $`f(\mathrm{\Lambda })\mathrm{\Lambda }`$. We say that $`\mathrm{\Lambda }`$ is a hyperbolic set if there is a $`Df`$-invariant splitting $`T_\mathrm{\Lambda }M=E^sE^u`$ of the tangent bundle restricted to $`\mathrm{\Lambda }`$ and a constant $`\lambda <1`$ such that (for some choice of a Riemannian metric on $`M`$) for every $`x\mathrm{\Lambda }`$
$$DfE_x^s<\lambda \text{and}Df^1E_x^u<\lambda .$$
We say that an embedded disk $`\gamma M`$ is an unstable manifold, or an unstable disk, if $`\mathrm{dist}(f^n(x),f^n(y))0`$ exponentially fast as $`n\mathrm{}`$, for every $`x,y\gamma `$. Similarly, $`\gamma `$ is called a stable manifold, or a stable disk, if $`\mathrm{dist}(f^n(x),f^n(y))0`$ exponentially fast as $`n\mathrm{}`$, for every $`x,y\gamma `$. It is well-known that every point in a hyperbolic set possesses a local stable manifold $`W_{loc}^s(x)`$ and a local unstable manifold $`W_{loc}^u(x)`$ which are disks tangent to $`E_x^s`$ and $`E_x^u`$ at $`x`$ respectively.
A hyperbolic set $`\mathrm{\Lambda }`$ is said to be a horseshoe if local stable and local unstable manifolds through points in $`\mathrm{\Lambda }`$ intersect $`\mathrm{\Lambda }`$ in a Cantor set. Horseshoes were introduced by Smale and appear naturally when one unfolds a homoclinic tangency associated to some hyperbolic periodic point of saddle type. It follows from \[4, Theorem 4.11\] that a $`C^{1+\alpha }`$ diffeomorphism cannot have a fat hyperbolic horseshoe, i.e. a hyperbolic horseshoe $`\mathrm{\Lambda }`$ with $`\mathrm{Leb}(\mathrm{\Lambda })>0`$; actually the result in is proved for basic sets. Nevertheless, here we obtain a generalization of that result to a much more general situation. Let us remark that fat hyperbolic horseshoes exist for $`C^1`$ diffeomorphisms, as shown in .
We say that a compact invariant set $`\mathrm{\Lambda }`$ has a dominated splitting if there exists a continuous $`Df`$-invariant splitting $`T_\mathrm{\Lambda }M=E^{cs}E^{cu}`$ of the tangent bundle restricted to $`\mathrm{\Lambda }`$, and a constant $`0<\lambda <1`$ such that (for some choice of a Riemannian metric on $`M`$) for every $`x\mathrm{\Lambda }`$
$$DfE_x^{cs}Df^1E_{f(x)}^{cu}\lambda .$$
We call $`E^{cs}`$ the centre-stable bundle and $`E^{cu}`$ the centre-unstable bundle. We say that $`f`$ is non-uniformly expanding along the centre-unstable direction for $`x\mathrm{\Lambda }`$ if
$$\underset{n+\mathrm{}}{lim\; inf}\frac{1}{n}\underset{j=1}{\overset{n}{}}\mathrm{log}Df^1E_{f^j(x)}^{cu}<0.$$
(NUE)
Condition NUE means that the derivative has expanding behavior in average over the orbit of $`x`$. This implies that $`x`$ has $`dim(E^{cu})`$ positive Lypaunov exponents in the $`E_x^{cu}`$ direction. As shown in \[2, Theorem C\], if condition NUE holds for every point in a compact invariant set $`\mathrm{\Lambda }`$, then $`E^{cu}`$ is necessarily uniformly expanding in $`\mathrm{\Lambda }`$, i.e. there is $`0<\lambda <1`$ such that
$$Df^1E_{f(x)}^{cu}\lambda ,\text{for every }x\mathrm{\Lambda }\text{.}$$
A class of diffeomorphisms with a dominated splitting $`TM=E^{cs}E^{cu}`$ for which NUE holds Lebesgue almost everywhere in $`M`$ and $`E^{cu}`$ is not uniformly expanding can be found in \[1, Appendix A\].
###### Theorem A.
Let $`f:MM`$ be a $`C^{1+\alpha }`$ diffeomorphism and let $`\mathrm{\Lambda }M`$ have a dominated splitting. If there is $`H\mathrm{\Lambda }`$ with $`\mathrm{Leb}(H)>0`$ such that NUE holds for every $`xH`$, then $`\mathrm{\Lambda }`$ contains some local unstable disk.
We say that a compact invariant set $`\mathrm{\Lambda }`$ is partially hyperbolic if it has a dominated splitting $`T_\mathrm{\Lambda }M=E^{cs}E^{cu}`$ for which $`E^{cs}`$ is uniformly contracting or $`E^{cu}`$ is uniformly expanding, meaning that there is $`0<\lambda <1`$ such that $`0<\lambda <1`$ such that $`DfE_x^{cs}\lambda `$ for every $`x\mathrm{\Lambda }`$, or $`Df^1E_{f(x)}^{cu}\lambda `$ for every $`x\mathrm{\Lambda }`$.
The next result is a direct consequence of Theorem A, whenever $`E^{cu}`$ is uniformly expanding. If, on the other hand, $`E^{cs}`$ is uniformly contracting, then we just have to apply Theorem A to $`f^1`$.
###### Corollary B.
Let $`f:MM`$ be a $`C^{1+\alpha }`$ diffeomorphism and let $`\mathrm{\Lambda }M`$ be a partially hyperbolic set with $`\mathrm{Leb}(\mathrm{\Lambda })>0`$.
1. If $`E^{cs}`$ is uniformly contracting, then $`\mathrm{\Lambda }`$ contains a local stable disk.
2. If $`E^{cu}`$ is uniformly expanding, then $`\mathrm{\Lambda }`$ contains a local unstable disk.
In particular, $`C^{1+\alpha }`$ diffeomorphisms cannot have partially hyperbolic horseshoes with positive Lebesgue measure. The same holds for partially hyperbolic sets intersecting a local stable or a local unstable disk in a positive Lebesgue measure subset, as Corollary D below shows.
###### Theorem C.
Let $`f:MM`$ be a $`C^{1+\alpha }`$ diffeomorphism and let $`\mathrm{\Lambda }M`$ have a dominated splitting. Assume that there is a local unstable disk $`\gamma `$ with $`\mathrm{Leb}_\gamma (\gamma \mathrm{\Lambda })>0`$ such that NUE holds for every $`x\gamma \mathrm{\Lambda }`$. Then $`\mathrm{\Lambda }`$ contains some local unstable disk.
The next result is a direct consequence of Theorem C in the case that $`E^{cu}`$ is uniformly expanding, and a consequence of the theorem applied to $`f^1`$ in the case that $`E^{cs}`$ is uniformly contracting.
###### Corollary D.
Let $`f:MM`$ be a $`C^{1+\alpha }`$ diffeomorphism and let $`\mathrm{\Lambda }M`$ be a partially hyperbolic set.
1. If $`E^{cs}`$ is uniformly contracting and there is a local stable disk $`\gamma `$ such that $`\mathrm{Leb}_\gamma (\gamma \mathrm{\Lambda })>0`$, then $`\mathrm{\Lambda }`$ contains a local stable disk.
2. If $`E^{cu}`$ is uniformly expanding and there is a local unstable disk $`\gamma `$ such that $`\mathrm{Leb}_\gamma (\gamma \mathrm{\Lambda })>0`$, then $`\mathrm{\Lambda }`$ contains a local unstable disk.
Theorems A and C are in fact corollaries of a slightly more general result that we present at the beginning of Section 4.
### Acknowledgement
We are grateful to M. Viana for valuable references on this topic.
## 2. Hölder control of tangent direction
This section is a survey of results in \[1, Section 2\] concerning the Hölder control of the tangent direction of submanifolds. As observed in \[1, Remark 2.3\] those results are valid for diffeomorphisms of class $`C^{1+\alpha }`$. In this section we only use the existence of a dominated splitting $`T_\mathrm{\Lambda }M=E^{cs}E^{cu}`$. We fix continuous extensions of the two bundles $`E^{cs}`$ and $`E^{cu}`$ to some neighborhood $`U`$ of $`\mathrm{\Lambda }`$, that we denote by $`\stackrel{~}{E}^{cs}`$ and $`\stackrel{~}{E}^{cu}`$. We do not require these extensions to be invariant under $`Df`$. Given $`0<a<1`$, we define the centre-unstable cone field $`C_a^{cu}=\left(C_a^{cu}(x)\right)_{xU}`$ of width $`a`$ by
$$C_a^{cu}(x)=\left\{v_1+v_2\stackrel{~}{E}_x^{cs}\stackrel{~}{E}_x^{cu}\text{ such that }v_1av_2\right\}.$$
(1)
We define the centre-stable cone field $`C_a^{cs}=\left(C_a^{cs}(x)\right)_{xU}`$ of width $`a`$ in a similar way, just reversing the roles of the subbundles in (1). We fix $`a>0`$ and $`U`$ small enough so that, up to slightly increasing $`\lambda <1`$, the domination condition remains valid for any pair of vectors in the two cone fields:
$$Df(x)v^{cs}Df^1(f(x))v^{cu}\lambda v^{cs}v^{cu}$$
for every $`v^{cs}C_a^{cs}(x)`$, $`v^{cu}C_a^{cu}(f(x))`$, and any point $`xUf^1(U)`$. Note that the centre-unstable cone field is positively invariant:
$$Df(x)C_a^{cu}(x)C_a^{cu}(f(x)),\text{whenever }x,f(x)U\text{.}$$
Indeed, the domination property together with the invariance of $`E^{cu}=(\stackrel{~}{E}^{cu}\mathrm{\Lambda })`$ imply that
$$Df(x)C_a^{cu}(x)C_{\lambda a}^{cu}(f(x))C_a^{cu}(f(x)),$$
for every $`xK`$. This extends to any $`xUf^1(U)`$ just by continuity.
We say that an embedded $`C^1`$ submanifold $`NU`$ is tangent to the centre-unstable cone field if the tangent subspace to $`N`$ at each point $`xN`$ is contained in the corresponding cone $`C_a^{cu}(x)`$. Then $`f(N)`$ is also tangent to the centre-unstable cone field, if it is contained in $`U`$, by the domination property.
Our aim now is to express the notion of Hölder variation of the tangent bundle in local coordinates. We choose $`\delta _0>0`$ small enough so that the inverse of the exponential map $`\mathrm{exp}_x`$ is defined on the $`\delta _0`$ neighbourhood of every point $`x`$ in $`U`$. From now on we identify this neighbourhood of $`x`$ with the corresponding neighbourhood $`U_x`$ of the origin in $`T_xN`$, through the local chart defined by $`\mathrm{exp}_x^1`$. Reducing $`\delta _0`$, if necessary, we may suppose that $`\stackrel{~}{E}_x^{cs}`$ is contained in the centre-stable cone $`C_a^{cs}(y)`$ of every $`yU_x`$. In particular, the intersection of $`C_a^{cu}(y)`$ with $`\stackrel{~}{E}_x^{cs}`$ reduces to the zero vector. Then, the tangent space to $`N`$ at $`y`$ is parallel to the graph of a unique linear map $`A_x(y):T_xN\stackrel{~}{E}_x^{cs}`$. Given constants $`C>0`$ and $`0<\zeta 1`$, we say that the tangent bundle to $`N`$ is $`(C,\zeta )`$-Hölder if for every $`yNU_x`$ and $`xV_0`$
$$A_x(y)Cd_x(y)^\zeta ,$$
(2)
where $`d_x(y)`$ denotes the distance from $`x`$ to $`y`$ along $`NU_x`$, defined as the length of the shortest curve connecting $`x`$ to $`y`$ inside $`NU_x`$.
Recall that we have chosen the neighbourhood $`U`$ and the cone width $`a`$ sufficiently small so that the domination property remains valid for vectors in the cones $`C_a^{cs}(z)`$, $`C_a^{cu}(z)`$, and for any point $`z`$ in $`U`$. Then, there exist $`\lambda _1(\lambda ,1)`$ and $`\zeta (0,1]`$ such that
$$Df(z)v^{cs}Df^1(f(z))v^{cu}^{1+\zeta }\lambda _1<1$$
(3)
for every norm $`1`$ vectors $`v^{cs}C_a^{cs}(z)`$ and $`v^{cu}C_a^{cu}(z)`$, at any $`zU`$. Then, up to reducing $`\delta _0>0`$ and slightly increasing $`\lambda _1<1`$, condition (3) remains true if we replace $`z`$ by any $`yU_x`$, $`xU`$ (taking $``$ to mean the Riemannian metric in the corresponding local chart).
We fix $`\zeta `$ and $`\lambda _1`$ as above. Given a $`C^1`$ submanifold $`NU`$, we define
$$\kappa (N)=inf\{C>0:\text{the tangent bundle of }N\text{ is }(C,\zeta )\text{-Hölder}\}.$$
(4)
The next result appears in \[1, Corollary 2.4\].
###### Proposition 2.1.
There exists $`C_1>0`$ such that, given any $`C^1`$ submanifold $`NU`$ tangent to the centre-unstable cone field,
1. there exists $`n_01`$ such that $`\kappa (f^n(N))C_1`$ for every $`nn_0`$ such that $`f^k(N)U`$ for all $`0kn`$;
2. if $`\kappa (N)C_1`$, then the same is true for every iterate $`f^n(N)`$ such that $`f^k(N)U`$ for all $`0kn`$;
3. in particular, if $`N`$ and $`n`$ are as in (2), then the functions
$$J_k:f^k(N)x\mathrm{log}|det(DfT_xf^k(N))|,0kn,$$
are $`(L,\zeta )`$-Hölder continuous with $`L>0`$ depending only on $`C_1`$ and $`f`$.
## 3. Hyperbolic times and bounded distortion
The following notion will allow us to derive uniform behaviour (expansion, distortion) from the non-uniform expansion.
###### Definition 3.1.
Given $`\sigma <1`$, we say that $`n`$ is a $`\sigma `$-hyperbolic time for $`x\mathrm{\Lambda }`$ if
$$\underset{j=nk+1}{\overset{n}{}}Df^1E_{f^j(x)}^{cu}\sigma ^k,\text{for all }1kn\text{.}$$
In particular, if $`n`$ is a $`\sigma `$-hyperbolic time for $`x`$, then $`Df^kE_{f^n(x)}^{cu}`$ is a contraction for every $`1kn`$:
$$Df^kE_{f^n(x)}^{cu}\underset{j=nk+1}{\overset{n}{}}Df^1E_{f^j(x)}^{cu}\sigma ^k.$$
Moreover, if $`a>0`$ is taken sufficiently small in the definition of our cone fields, and we choose $`\delta _1>0`$ also small so that the $`\delta _1`$-neighborhood of $`\mathrm{\Lambda }`$ should be contained in $`U`$, then by continuity
$$Df^1(f(y))v\frac{1}{\sqrt{\sigma }}Df^1|E_{f(x)}^{cu}v,$$
(5)
whenever $`x\mathrm{\Lambda }`$, $`\mathrm{dist}(x,y)\delta _1`$, and $`vC_a^{cu}(y)`$.
Given any disk $`\mathrm{\Delta }M`$, we use $`\mathrm{dist}_\mathrm{\Delta }(x,y)`$ to denote the distance between $`x,y\mathrm{\Delta }`$, measured along $`\mathrm{\Delta }`$. The distance from a point $`x\mathrm{\Delta }`$ to the boundary of $`\mathrm{\Delta }`$ is $`\mathrm{dist}_\mathrm{\Delta }(x,\mathrm{\Delta })=inf_{y\mathrm{\Delta }}\mathrm{dist}_\mathrm{\Delta }(x,y)`$.
###### Lemma 3.2.
Take any $`C^1`$ disk $`\mathrm{\Delta }U`$ of radius $`\delta `$, with $`0<\delta <\delta _1`$, tangent to the centre-unstable cone field. There is $`n_01`$ such that if $`x\mathrm{\Delta }`$ with $`\mathrm{dist}_\mathrm{\Delta }(x,\mathrm{\Delta })\delta /2`$ and $`nn_0`$ is a $`\sigma `$-hyperbolic time for $`x`$, then there is a neighborhood $`V_n`$ of $`x`$ in $`\mathrm{\Delta }`$ such that:
1. $`f^n`$ maps $`V_n`$ diffeomorphically onto a disk of radius $`\delta _1`$ around $`f^n(x)`$ tangent to the centre-unstable cone field;
2. for every $`1kn`$ and $`y,zV_n`$,
$$\mathrm{dist}_{f^{nk}(V_n)}(f^{nk}(y),f^{nk}(z))\sigma ^{k/2}\mathrm{dist}_{f^n(V_n)}(f^n(y),f^n(z)).$$
###### Proof.
First we show that $`f^n(\mathrm{\Delta })`$ contains some disk of radius $`\delta _1`$ around $`f^n(x)`$, as long as
$$n>2\frac{\mathrm{log}(\delta /(2\delta _1))}{\mathrm{log}(\sigma )}.$$
(6)
Assume that there is $`y\mathrm{\Delta }`$ with $`\mathrm{dist}_{f^n(\mathrm{\Delta })}(f^n(x),f^n(y))<\delta _1`$. Let $`\eta _0`$ be a curve of minimal length in $`f^n(\mathrm{\Delta })`$ connecting $`f^n(x)`$ to $`f^n(y)`$. For $`0kn`$ we write $`\eta _k=f^{nk}(\eta _0)`$. We prove by induction that $`\mathrm{length}(\eta _k)<\sigma ^{k/2}\delta _1,`$ for $`0kn`$. Let $`1kn`$ and assume that
$$\mathrm{length}(\eta _j)<\sigma ^{j/2}\delta _1,\text{for }0jk1.$$
Denote by $`\dot{\eta }_0(w)`$ the tangent vector to the curve $`\eta _0`$ at the point $`w`$. Then, by the choice of $`\delta _1`$ in (5) and the definition of $`\sigma `$-hyperbolic time,
$$Df^k(w)\dot{\eta }_0(w)\sigma ^{k/2}\dot{\eta }_0(w)\underset{j=nk+1}{\overset{n}{}}Df^1|E_{f^j(x)}^{cu}\sigma ^{k/2}\dot{\eta }_0(w).$$
Hence,
$$\mathrm{length}(\eta _k)\sigma ^{k/2}\mathrm{length}(\eta _0)<\sigma ^{k/2}\delta _1.$$
This completes our induction. In particular, we have $`\mathrm{length}(\eta _n)<\sigma ^{n/2}\delta _1.`$ Note that $`\eta _n`$ is a curve in $`\mathrm{\Delta }`$ connecting $`x`$ to $`y\mathrm{\Delta }`$, and so $`\mathrm{length}(\eta _n)\delta /2`$. Thus we must have
$$n<2\frac{\mathrm{log}(\delta /(2\delta _1))}{\mathrm{log}(\sigma )}.$$
Hence $`f^n(\mathrm{\Delta })`$ contains some disk of radius $`\delta _1`$ around $`f^n(x)`$ for $`n`$ as in (6).
Let now $`\mathrm{\Delta }_1`$ be the disk of radius $`\delta _1`$ around $`f^n(x)`$ in $`f^n(\mathrm{\Delta })`$ and let $`V_n=f^n(\mathrm{\Delta }_1)`$, for $`n`$ as in (6). Take any $`y,zV_n`$ and let $`\eta _0`$ be a curve of minimal length in $`\mathrm{\Delta }_1`$ connecting $`f^n(y)`$ to $`f^n(z)`$. Defining $`\eta _k=f^{nk}(\eta _0)`$, for $`1kn`$, and arguing as before we inductively prove that for $`1kn`$
$$\mathrm{length}(\eta _k)\sigma ^{k/2}\mathrm{length}(\eta _0)=\sigma ^{k/2}\mathrm{dist}_{f^n(V_n)}(f^n(y),f^n(z)),$$
which implies that for $`1kn`$
$$\mathrm{dist}_{f^{nk}(V_n)}(f^{nk}(y),f^{nk}(z))\sigma ^{k/2}\mathrm{dist}_{f^n(V_n)}(f^n(y),f^n(z)).$$
This completes the proof of the lemma. ∎
We shall sometimes refer to the sets $`V_n`$ as *hyperbolic pre-balls* and to their images $`f^n(V_n)`$ as *hyperbolic balls*. Notice that the latter are indeed balls of radius $`\delta _1`$.
###### Corollary 3.3 (Bounded Distortion).
There exists $`C_2>1`$ such that given $`\mathrm{\Delta }`$ as in Lemma 3.2 with $`\kappa (\mathrm{\Delta })C_1`$, and given any hyperbolic pre-ball $`V_n\mathrm{\Delta }`$ with $`nn_0`$, then for all $`y,zV_n`$
$$\frac{1}{C_2}\frac{|detDf^nT_y\mathrm{\Delta }|}{|detDf^nT_z\mathrm{\Delta }|}C_2.$$
###### Proof.
For $`0i<n`$ and $`y\mathrm{\Delta }`$, we denote $`J_i(y)=|detDfT_{f^i(y)}f^i(\mathrm{\Delta })|`$. Then,
$$\mathrm{log}\frac{|detDf^nT_y\mathrm{\Delta }|}{|detDf^nT_z\mathrm{\Delta }|}=\underset{i=0}{\overset{n1}{}}\left(\mathrm{log}J_i(y)\mathrm{log}J_i(z)\right).$$
By Proposition 2.1, $`\mathrm{log}J_i`$ is $`(L,\zeta )`$-Hölder continuous, for some uniform constant $`L>0`$. Moreover, by Lemma 3.2, the sum of all $`\mathrm{dist}_\mathrm{\Delta }(f^j(y),f^j(z))^\zeta `$ over $`0jn`$ is bounded by $`\delta _1/(1\sigma ^{\zeta /2})`$. Now it suffices to take $`C_2=\mathrm{exp}(L\delta _1/(1\sigma ^{\zeta /2}))`$. ∎
## 4. A local unstable disk inside $`\mathrm{\Lambda }`$
Now we are able to prove Theorems A and C. These will be obtained as corollaries of the next result, as we shall see below.
###### Theorem 4.1.
Let $`f:MM`$ be a $`C^{1+\alpha }`$ diffeomorphism and let $`\mathrm{\Lambda }M`$ have a dominated splitting. Assume that there is a disk $`\mathrm{\Delta }`$ tangent to the centre-unstable cone field with $`\mathrm{Leb}_\mathrm{\Delta }(\mathrm{\Delta }\mathrm{\Lambda })>0`$ such that NUE holds for every $`x\mathrm{\Delta }\mathrm{\Lambda }`$. Then $`\mathrm{\Lambda }`$ contains some local unstable disk.
Assume that there is $`H\mathrm{\Lambda }`$ with $`\mathrm{Leb}(H)>0`$ such that NUE holds for every $`xH`$. Choosing a $`\mathrm{Leb}`$ density point of $`H`$, we laminate a neighborhood of that point into disks tangent to the centre-unstable cone field contained in $`U`$. Since the relative Lebesgue measure of the intersections of these disks with $`H`$ cannot be all equal to zero, we obtain some disk $`\mathrm{\Delta }`$ as in Theorem 4.1 under the assumption of Theorem A. For Theorem C, observe that local stable manifolds are tangent to the centre-unstable spaces and these vary continuously with the points in $`\mathrm{\Lambda }`$, thus being locally tangent to the centre-unstable cone field.
Let us now prove Theorem 4.1. Let $`\mathrm{\Delta }`$ be a disk tangent to the centre-unstable cone field intersecting $`\mathrm{\Lambda }`$ in a positive $`\mathrm{Leb}_\mathrm{\Delta }`$ subset such that NUE holds for every $`x\mathrm{\Delta }\mathrm{\Lambda }`$. Let $`H=\mathrm{\Delta }\mathrm{\Lambda }`$. Taking a subset of $`H`$, if necessary, still with positive $`\mathrm{Leb}_\mathrm{\Delta }`$ measure, we may assume that there is $`c>0`$ such that for every $`xH`$
$$\underset{n+\mathrm{}}{lim\; inf}\frac{1}{n}\underset{j=1}{\overset{n}{}}\mathrm{log}Df^1E_{f^j(x)}^{cu}c.$$
(7)
Since condition (7) remains valid under iteration, by Proposition 2.1 we may assume that $`\kappa (\mathrm{\Delta })<C_1`$. It is no restriction to assume that $`H`$ intersects the sub-disk of $`\mathrm{\Delta }`$ of radius $`\delta /2`$, for some $`0<\delta <\delta _1`$, in a positive $`\mathrm{Leb}_\mathrm{\Delta }`$ subset, and we do so.
The following lemma is due to Pliss , and a proof of it in the present form can be found in \[1, Lemma 3.1\].
###### Lemma 4.2 (Pliss).
Given $`Ac_2>c_1>0`$, let $`\theta =(c_2c_1)/(Ac_1)`$. Given any real numbers $`a_1,\mathrm{},a_N`$ such that $`a_jA`$ and
$$\underset{j=1}{\overset{N}{}}a_jc_2N,\text{for every }1jN,$$
then there are $`l>\theta N`$ and $`1<n_1<\mathrm{}<n_lN`$ so that
$$\underset{j=n+1}{\overset{n_i}{}}a_jc_1(n_in),\text{for every }0n<n_i\text{ and }i=1,\mathrm{},l.$$
###### Corollary 4.3.
There is $`\sigma >0`$ such that every $`xH`$ has infinitely many $`\sigma `$-hyperbolic times.
###### Proof.
Given $`xH`$, by (7) we have infinitely many values $`N`$ for which
$$\underset{j=1}{\overset{N}{}}\mathrm{log}Df^1|E_{f^j(x)}^{cu}\frac{c}{2}N.$$
Then it suffices to take $`c_1=c/2`$, $`c_2=c`$, $`A=sup|\mathrm{log}Df^1|E^{cu}|`$, and $`a_j=\mathrm{log}Df^1|E_{f^j(x)}^{cu}`$ in the previous lemma. ∎
Note that under assumption (7) we are unable to prove the existence of a positive frequency of hyperbolic times at infinity. This would be possible if we had $`lim\; sup`$ instead of $`lim\; inf`$ in (7), as shown in \[1, Corollary 3.2\]. The existence of infinitely many hyperbolic times is enough for what comes next.
###### Lemma 4.4.
There are an infinite sequence of integers $`1k_1<k_2<\mathrm{}`$ and, for each $`n`$, a disk $`\mathrm{\Delta }_n`$ of radius $`\delta _1/4`$ tangent to the centre-unstable cone field such that the relative Lebesgue measure of the set $`f^{k_n}(H)`$ in $`\mathrm{\Delta }_n`$ converges to 1 as $`n\mathrm{}`$.
###### Proof.
Let $`ϵ>0`$ be some small number. Let $`K`$ be a compact subset of $`H`$ and $`A`$ be an open neighborhood of $`H`$ in $`\mathrm{\Delta }`$ such that
$$\mathrm{Leb}_\mathrm{\Delta }(AK)<ϵ\mathrm{Leb}_\mathrm{\Delta }(K).$$
It follows from Corollary 4.3 and Lemma 3.2 that we can choose for each $`xK`$ a $`\sigma `$-hyperbolic time $`n(x)`$ and a hyperbolic pre-ball $`V_x`$ such that $`V_xA`$. Here $`V_x`$ is the neighborhood of $`x`$ associated to the hyperbolic time $`n(x)`$ constructed in Lemma 3.2, which is mapped diffeomorphically by $`f^{n(x)}`$ onto a ball $`B_{\delta _1}(f^{n(x)}(x))`$ of radius $`\delta _1`$ around $`f^{n(x)}(x)`$ tangent to the centre-unstable cone field. Let $`W_xV_x`$ be the pre-image of the ball $`B_{\delta _1/4}(f^{n(x)}(x))`$ of radius $`\delta _1/4`$ under this diffeomorphism.
By compactness we have $`KW_{x_1}\mathrm{}W_{x_s}`$ for some $`x_1,\mathrm{},x_mK`$. Writing
$$\{n_1,\mathrm{},n_s\}=\{n(x_1),\mathrm{},n(x_m)\},\text{with }n_1<n_2<\mathrm{}<n_s,$$
(8)
let $`𝒰_1`$ be a maximal set of $`\{1,\mathrm{},m\}`$ such that if $`u𝒰_1`$ then $`n(x_u)=n_1`$ and $`W_{x_u}W_{x_a}=\mathrm{}`$ for all $`a𝒰_1`$ with $`au`$. Inductively we define $`𝒰_j`$ for $`1<js`$ as follows. Suppose that $`𝒰_{j1}`$ has already been defined. Let $`𝒰_j`$ be a maximal set of $`\{1,\mathrm{},m\}`$ such that if $`u𝒰_j`$ then $`n(x_u)=n_j`$ and $`W_{x_u}W_{x_a}=\mathrm{}`$ for all $`a𝒰_j`$ with $`au`$, and also $`W_{x_u}W_{x_a}=\mathrm{}`$ for all $`a𝒰_1\mathrm{}𝒰_{j1}`$.
Let $`𝒰=𝒰_1\mathrm{}𝒰_s`$. By maximality, each $`W_{x_i}`$, $`1im`$, intersects some $`W_{x_u}`$ with $`u𝒰`$ and $`n(x_i)n(x_u)`$. Thus, given any $`1im`$ and taking $`u𝒰`$ such that $`W_{x_i}W_{x_u}\mathrm{}`$ and $`n(x_i)n(x_u)`$, we get
$$f^{n(x_u)}(W_{x_i})B_{\delta _1/4}(f^{n(x_u)}(x_u))\mathrm{}.$$
Lemma 3.2 assures that
$$\mathrm{diam}(f^{n(x_u)}(W_{x_i}))\frac{\delta _1}{2}\sigma ^{(n(x_i)n(x_u))/2}\frac{\delta _1}{2},$$
and so
$$f^{n(x_u)}(W_{x_i})B_{\delta _1}(f^{n(x_u)}(x_u)).$$
This implies that $`W_{x_i}V_{x_u}`$. Hence $`\{V_{x_u}\}_{u𝒰}`$ is a covering of $`K`$.
It follows from Corollary 3.3 that there is a uniform constant $`\gamma >0`$ such that
$$\frac{\mathrm{Leb}_\mathrm{\Delta }(W_{x_u})}{\mathrm{Leb}_\mathrm{\Delta }(V_{x_u})}\gamma ,\text{for every }u𝒰.$$
Hence
$`\mathrm{Leb}_\mathrm{\Delta }\left(_{u𝒰}W_{x_u}\right)`$ $`=`$ $`{\displaystyle \underset{u𝒰}{}}\mathrm{Leb}_\mathrm{\Delta }(W_{x_u})`$
$``$ $`{\displaystyle \underset{u𝒰}{}}\gamma \mathrm{Leb}_\mathrm{\Delta }(V_{x_u})`$
$``$ $`\gamma \mathrm{Leb}_\mathrm{\Delta }\left(_{u𝒰}V_{x_u}\right)`$
$``$ $`\gamma \mathrm{Leb}_\mathrm{\Delta }(K).`$
Setting
$$\rho =\mathrm{min}\left\{\frac{\mathrm{Leb}_\mathrm{\Delta }(W_{x_u}K)}{\mathrm{Leb}_\mathrm{\Delta }(W_{x_u})}:u𝒰\right\},$$
we have
$`\epsilon \mathrm{Leb}_\mathrm{\Delta }(K)`$ $``$ $`\mathrm{Leb}_\mathrm{\Delta }(AK)`$
$``$ $`\mathrm{Leb}_\mathrm{\Delta }\left(_{u𝒰}W_{x_u}K\right)`$
$``$ $`{\displaystyle \underset{u𝒰}{}}\mathrm{Leb}_\mathrm{\Delta }(W_{x_u}K)`$
$``$ $`\rho \mathrm{Leb}_\mathrm{\Delta }\left(_{u𝒰}W_{x_u}\right)`$
$``$ $`\rho \gamma \mathrm{Leb}_\mathrm{\Delta }(K).`$
This implies that $`\rho <\epsilon /\gamma `$. Since $`\epsilon >0`$ can be taken arbitrarily small, we may always choose $`W_{x_u}`$ such that the relative Lebesgue measure of $`K`$ in $`W_{x_u}`$ is arbitrarily close to 1. Then, by bounded distortion, the relative Lebesgue measure of $`f^{n(x_u)}(H)f^{n(x_u)}(K)`$ in $`f^{n(x_u)}(W_{x_u})`$, which is a disk of radius $`\delta _1/4`$ around $`f^{n(x_u)}(x_u)`$ tangent to centre-unstable cone field, is also arbitrarily close to $`1`$. Observe that since points in $`H`$ have infinitely many hyperbolic times, we may take the integer $`n(x_u)`$ arbitrarily large, as long as $`n_1`$ in (8) is also taken large enough. ∎
###### Proposition 4.5.
There is a local unstable disk $`\mathrm{\Delta }_{\mathrm{}}`$ of radius $`\delta _1/4`$ inside $`\mathrm{\Lambda }`$.
###### Proof.
Let $`(\mathrm{\Delta }_n)_n`$ be the sequence of disks given by Lemma 4.4, and consider $`(x_n)_n`$ the sequence of points at which these disks are centered. Up to taking subsequences, we may assume that the centers of the disks converge to some point $`x`$. Using Ascoli-Arzela, the disks converge to some disk $`\mathrm{\Delta }_{\mathrm{}}`$ centered at $`x`$. By construction, every point in $`\mathrm{\Delta }_{\mathrm{}}`$ is accumulated by some orbit of a point in $`H\mathrm{\Lambda }`$, and so $`\mathrm{\Delta }_{\mathrm{}}\mathrm{\Lambda }`$.
Note that each $`\mathrm{\Delta }_n`$ is contained in the $`k_n`$-iterate of $`\mathrm{\Delta }`$, which is a disk tangent to the centre-unstable cone field. The domination property implies that the angle between $`\mathrm{\Delta }_n`$ and $`E^{cu}`$ goes to zero as $`n\mathrm{}`$, uniformly on $`\mathrm{\Lambda }`$. In particular, $`\mathrm{\Delta }_{\mathrm{}}`$ is tangent to $`E^{cu}`$ at every point in $`\mathrm{\Delta }_{\mathrm{}}\mathrm{\Lambda }`$. By Lemma 3.2, given any $`k1`$, then $`f^k`$ is a $`\sigma ^{k/2}`$-contraction on $`\mathrm{\Delta }_n`$ for every large $`n`$. Passing to the limit, we get that $`f^k`$ is a $`\sigma ^{k/2}`$-contraction on $`\mathrm{\Delta }_{\mathrm{}}`$ for every $`k1`$.
In particular, we have shown that the subspace $`E_x^{cu}`$ is uniformly expanding for $`Df`$. The fact that the $`Df`$-invariant splitting $`T_\mathrm{\Lambda }M=E^{cs}E^{cu}`$ is a dominated splitting implies that any expansion $`Df`$ may exhibit along the complementary direction $`E_x^{cs}`$ is weaker than the expansion in the $`E_x^{cu}`$ direction. Then, by , there exists a unique unstable manifold $`W_{loc}^u(x)`$ tangent to $`E^{cu}`$ and which is contracted by the negative iterates of $`f`$. Since $`\mathrm{\Delta }_{\mathrm{}}`$ is contracted by every $`f^k`$, and all its negative iterates are tangent to centre-unstable cone field, then $`\mathrm{\Delta }_{\mathrm{}}`$ is contained in $`W_{loc}^u(x)`$. ∎ |
warning/0507/hep-th0507035.html | ar5iv | text | # Contents
## 1 Introduction
There has been a long history of interplay between differential geometry and supersymmetric non-linear sigma models starting with the observation that $`N=2`$ supersymmtery in two dimensions requires the sigma model target space to be a Kähler manifold . It was first pointed out in that one could construct conserved currents in $`(1,1)`$ sigma models given a covariantly constant form on the target space, and in it was shown that the $`(1,1)`$ model on a Calabi-Yau three-fold has an extended superconformal algebra involving precisely such a current constructed from the holomorphic three-form. In symmetries of this type were studied systematically in the classical sigma model setting; each manifold on Berger’s list of irreducible non-symmetric Riemannian manifolds has one or more covariantly constant forms which give rise to conserved currents and the corresponding Poisson bracket algebras are non-linear, i.e. they are of W-symmetry type. Subsequently the properties of these algebras were studied more abstractly in a conformal field theory framework and more recently in topological models .
In this paper we shall discuss two-dimensional (1,1) supersymmetric sigma models with boundaries with extra symmetries of the above type, focusing in particular on target spaces with special holonomy. In a series of papers - classical supersymmetric sigma models with boundaries have been discussed in detail and it has been shown how the fermionic boundary conditions involve a locally defined tensor $`R`$ which determines the geometry associated with the boundary. In particular, in the absence of torsion, one finds that there are integral submanifolds of the projector $`P=\frac{1}{2}(1+R)`$ which have the interpretation of being branes where the boundary can be located. These papers considered $`(1,1)`$ and $`(2,2)`$ models and the analysis was also extended to models of this type with torsion where the intepretation of $`R`$ is less straightforward. The main purpose of the current paper is to further extend this analysis to include symmetries associated with certain holonomy groups or $`G`$-structures. We shall discuss models both with and without torsion.
Torsion-free sigma models with boundaries on manifolds with special holonomy were first considered in where it was shown how the identification of the left and right currents on the boundary has a natural interpretation in terms of calibrations and calibrated submanifolds. Branes have also been discussed extensively in boundary CFT , including the $`G_2`$ case , and in topological string theory .
The main new results of the paper concern boundary (1,1) models with torsion or with a gauge field on the brane. There is no analogue of Berger’s list in the case of torsion but we can nevertheless consider target spaces with specific $`G`$-structures which arise due to the presence of covariantly constant forms of the same type. In order to generalise the discussion from the torsion-free case we require there to be two independent $`G`$-structures specified by two sets of covariantly constant forms $`\{\lambda ^+,\lambda ^{}\}`$ which are covariantly constant with respect to two metric connections $`\{\mathrm{\Gamma }^+,\mathrm{\Gamma }^{}\}`$ and which have closed skew-symmetric torsion tensors $`T^\pm =\pm H`$, where $`H=db`$, $`b`$ being the two-form potential which appears in the sigma model action. This sort of structure naturally generalises the notion of bi-hermitian geometry which occurs in $`N=2`$ sigma models with torsion and which has been studied in the boundary sigma model context in . We shall refer to this type of structure as a bi-$`G`$-structure. The groups $`G`$ which are of most interest from the point of view of spacetime symmetry are the groups which appear on Berger’s list and for this reason we use the term special holonomy. Bi-$`G`$-structures are closely related to the generalised structures which have appeared in the mathematical literature . These generalised geometries have been discussed in the $`N=2`$ sigma model context . In a recent paper they have been exploited in the context of branes and generalised calibrations.
We shall show that, in general, the geometrical conditions implied by equating the left and right currents on the boundary lead to further constraints by differentiation and that these constraints are the same as those which arise when one looks at the stability of the boundary conditions under symmetry transformations. It turns out, however, that these constraints are automatically satisfied by virtue of the target space geometry.
We then study the target space geometry of some examples, in particular bi-$`G_2`$, bi-$`SU(3)`$ and bi-$`Spin(7)`$ structures. Structures of this type have appeared in the supergravity literature in the context of supersymmetric solutions with flux .
The paper is organised as follows: in section two we review the basics of boundary sigma models, in section three we discuss additional symmetries associated with special holonomy groups or bi-$`G`$-structures, in section four we examine the consistency of the boundary conditions under symmetry variations, in section five we look at the target space geometry of bi-$`G`$ structures from a simple point of view and in section six we look at some examples of solutions of the boundary conditions for the currents defined by the covariantly constant forms.
## 2 Review of basics
The action for a $`(1,1)`$-supersymmetric sigma model without boundary is
$$S=𝑑ze_{ij}D_+X^iD_{}X^j,$$
(2.1)
where
$$e_{ij}:=g_{ij}+b_{ij},$$
(2.2)
$`b`$ being a two-form potential with field strength $`H=db`$ on the $`n`$-dimensional Riemannian target space $`(M,g)`$. $`X^i,i=1,\mathrm{}n`$, is the sigma model field represented in some local chart for $`M`$ and $`z`$ denotes the coordinates of $`(1,1)`$ superspace $`\mathrm{\Sigma }`$. We shall use a light-cone basis so that $`z=(x^{++},x^{},\theta ^+,\theta ^{})`$, with $`x^{++}=x^0+x^1,x^{}=x^0x^1`$. $`D_+`$ and $`D_{}`$ are the usual flat superspace covariant derivatives which obey the relations
$$D_+^2=i_{++};D_{}^2=i_{};\{D_+,D_{}\}=0.$$
(2.3)
We use the convention that $`_{++}x^{++}=1`$. We shall take the superspace measure to be
$$dz:=d^2xD_+D_{}$$
(2.4)
with the understanding that the superfield obtained after integrating over the odd variables (i.e after applying $`D_+D_{}`$ to the integrand) is to be evaluated at $`\theta =0`$.
As well as the usual Levi-Civita connection $``$ there are two natural metric connections $`^\pm `$ with torsion ,
$$\mathrm{\Gamma }^{(\pm )}{}_{ik}{}^{j}:=\mathrm{\Gamma }_{ik}^j\pm \frac{1}{2}H^j{}_{ik}{}^{}.$$
(2.5)
The torsion tensors of the two connections are given by
$$T^{(\pm )}{}_{jk}{}^{i}=\pm H^i{}_{jk}{}^{},$$
(2.6)
so that the torsion is a closed three-form in either case.
In the presence of a boundary, $`\mathrm{\Sigma }`$, it is necessary to add additional boundary terms to the action (2.2) when there is torsion . The boundary action is
$$S_{bdry}=_\mathrm{\Sigma }a_i\dot{X}^i+\frac{i}{4}b_{ij}(\psi _+^i\psi _+^j+\psi _{}^i\psi _{}^j),$$
(2.7)
where $`a_i`$ is a gauge field which is defined only on the submanifold where the boundary sigma model field maps takes its values. Note that the boundary here is purely bosonic so that the fields are component fields, $`\psi _\pm ^i:=D_\pm X^i|`$, the vertical bar denoting the evaluation of a superfield at $`\theta =0`$).<sup>1</sup><sup>1</sup>1We shall use $`X^i`$ to mean either the superfield or its leading component; it should be clear from the context which is meant. The boundary term ensures that the action is unchanged if we add $`dc`$ to $`b`$ provided that we shift $`a`$ to $`ac`$. The modified field strength $`F=f+b`$, where $`f=da`$, is invariant under this transformation. In the absence of a $`b`$-field one can still have a gauge field on the boundary.
In the following we briefly summarise the approach to boundary sigma models of references -. We impose the standard boundary conditions on the fermions,
$$\psi _{}^i=\eta R^i{}_{j}{}^{}\psi _{+}^{j},\eta =\pm 1,\mathrm{on}\mathrm{\Sigma }$$
(2.8)
We shall also suppose that there are both Dirichlet and Neumann directions for the bosons. That is, we assume that there is a projection operator $`Q`$ such that
$$Q^i{}_{j}{}^{}\delta X^j=Q^i{}_{j}{}^{}\dot{X}_{}^{j}=0,$$
(2.9)
on $`\mathrm{\Sigma }`$. If $`F=0`$, parity implies that $`R^2=1`$, so that $`Q=\frac{1}{2}(1R)`$, while $`P:=\frac{1}{2}(1+R)`$ is the complementary projector. In general, we shall still use $`P`$ to denote $`\frac{1}{2}(1+R)`$ and the complementary projector will be denoted by $`\pi ,\pi :=1Q`$. We can take $`Q`$ and $`\pi `$ to be orthogonal
$$\pi _i^kg_{kl}Q^l{}_{j}{}^{}=0.$$
(2.10)
Equation (2.9) must hold for any variation along the boundary. Making a supersymmetry transformation we find
$$QR+Q=0.$$
(2.11)
On the other hand, the cancellation of the fermionic terms in the boundary variation (of $`S+S_{bdry}`$), when the bulk equations of motion are satisfied, requires
$$g_{ij}=g_{kl}R^k{}_{i}{}^{}R_{}^{l}{}_{j}{}^{}.$$
(2.12)
Using this together with orthogonality one deduces the following algebraic relations,
$`QR`$ $`=`$ $`RQ=Q;QP=PQ=0;`$
$`\pi P`$ $`=`$ $`P\pi =P;\pi R=R\pi .`$ (2.13)
Making a supersymmetry variation of the fermionic boundary condition (2.8) and using the equation of motion for the auxiliary field, $`F^i:=_{}^{(+)}D_+X^i|`$, namely $`F^i=0`$, we find the bosonic boundary condition<sup>2</sup><sup>2</sup>2The occurrence of (combinations of) field equations as boundary conditions is discussed in .
$$i(_{}X^iR^i{}_{j}{}^{}_{++}^{}X^j)=(2\stackrel{~}{}_jR^i{}_{k}{}^{}P^i{}_{l}{}^{}H_{jm}^{l}R^m{}_{k}{}^{})\psi _+^j\psi _+^k,$$
(2.14)
where $`\stackrel{~}{}`$ is defined by
$$\stackrel{~}{}_i:=P^j{}_{i}{}^{}_{j}^{}.$$
(2.15)
Combining (2.14) with the bosonic boundary condition arising directly from the variation we find
$$\widehat{E}_{ji}=\widehat{E}_{ik}R^k{}_{j}{}^{},$$
(2.16)
where
$$E_{ij}:=g_{ij}+F_{ij}$$
(2.17)
and the hats denote a pull-back to the brane,
$$\widehat{E}_{ij}:=\pi ^k{}_{i}{}^{}\pi _{}^{l}{}_{j}{}^{}E_{kl}^{}.$$
(2.18)
From (2.18) we find an expression for $`R`$,
$$R^i{}_{j}{}^{}=(\widehat{E}^1)^{ik}\widehat{E}_{jk}Q^i{}_{j}{}^{},$$
(2.19)
where the inverse is taken in the tangent space to the brane, i.e.
$$(\widehat{E}^1)^{ik}\widehat{E}_{kj}=\pi ^i{}_{j}{}^{}.$$
(2.20)
We can multiply equation (2.14) with $`Q`$ to obtain
$$P^l{}_{[i}{}^{}P_{}^{m}{}_{j]}{}^{}_{l}^{}Q^k{}_{m}{}^{}=0.$$
(2.21)
Using (2.13) we can show that this implies the integrability condition for $`\pi `$,
$$\pi ^l{}_{[i}{}^{}\pi _{}^{m}{}_{j]}{}^{}_{l}^{}Q^k{}_{m}{}^{}=0.$$
(2.22)
This confirms that the distribution specified by $`\pi `$ in $`TM`$ is integrable and the boundary maps to a submanifold, or brane, $`B`$. However, in the Lagrangian approach adopted here, this is implicit in the assumption of Dirichlet boundary conditions. When $`F=0`$ the derivative of $`R`$ along the brane is essentially the second fundamental form, $`K`$. Explicitly,
$$K_{jk}^i=P^l{}_{j}{}^{}P_{}^{m}{}_{k}{}^{}_{l}^{}Q^i{}_{m}{}^{}=P^l{}_{j}{}^{}\stackrel{~}{}_{k}^{}Q^i{}_{l}{}^{}.$$
(2.23)
The left and right supercurrents are
$`T_{+3}:`$ $`=`$ $`g_{ij}_{++}X^iD_+X^j{\displaystyle \frac{i}{6}}H_{ijk}D_{+3}X^{ijk}`$ (2.24)
$`T_3:`$ $`=`$ $`g_{ij}_{}X^iD_{}X^j+{\displaystyle \frac{i}{6}}H_{ijk}D_3X^{ijk}`$ (2.25)
The conservation conditions are
$$D_{}T_{+3}=D_+T_3=0.$$
(2.26)
The superpartners of the supercurrents are the left and right components of the energy-momentum tensor, $`D_+T_{+3}`$ and $`D_{}T_3`$ respectively. If one demands invariance of the total action under supersymmetry one finds that, on the boundary, the currents are related by
$`T_{+3}`$ $`=`$ $`\eta T_3`$ (2.27)
$`D_+T_{+3}`$ $`=`$ $`D_{}T_3.`$ (2.28)
The supercurrent boundary condition has a three-fermion term which implies the vanishing of the totally antisymmetric part of
$$2Y_{i,jk}+P^l{}_{i}{}^{}H_{ljm}^{}R^m{}_{k}{}^{}+\frac{1}{6}(H_{ijk}+H_{lmn}R^l{}_{i}{}^{}R_{}^{m}{}_{j}{}^{}R_{}^{n}{}_{k}{}^{}),$$
(2.29)
where
$$Y_{i,jk}:=(R^1)_{jl}\stackrel{~}{}_iR^lj.$$
(2.30)
## 3 Additional symmetries
A general variation of (2.1), neglecting boundary terms, gives
$`\delta S`$ $`=`$ $`{\displaystyle 𝑑z\mathrm{\hspace{0.17em}2}g_{ij}\delta X^i_{}^{(+)}D_+X^j}`$ (3.1)
$`=`$ $`{\displaystyle 𝑑z\mathrm{\hspace{0.17em}2}g_{ij}\delta X^ig_{ij}_+^{()}D_{}X^j}.`$
The additional symmetries we shall discuss are transformations of the form,
$$\delta _\pm X^i=a^\pm \mathrm{}L^{(\pm )}{}_{}{}^{i}{}_{j_1\mathrm{}j_{\mathrm{}}}{}^{}D_{\pm \mathrm{}}^{}X^{j_1\mathrm{}j_{\mathrm{}}},D_\pm \mathrm{}X^{j_1\mathrm{}j_{\mathrm{}}}:=D_\pm X^{j_1}\mathrm{}D\pm +X^j_{\mathrm{}},$$
(3.2)
where $`L^{(\pm )}`$ are vector-valued $`\mathrm{}`$-forms such that
$$\lambda ^{(\pm )}{}_{i_1\mathrm{}i_{\mathrm{}+1}}{}^{}:=g_{i_1j}L^{(\pm )j}_{i_2\mathrm{}i_{\mathrm{}+1}}$$
(3.3)
are $`(\mathrm{}+1)`$-forms which are covariantly constant with respect to $`^{(\pm )}`$. For example, a left transformation of this type gives
$`\delta S`$ $`=`$ $`{\displaystyle 𝑑z\mathrm{\hspace{0.17em}2}a^+\mathrm{}\lambda ^{(+)}{}_{i_1\mathrm{}i_{\mathrm{}+1}}{}^{}D_{+\mathrm{}}^{}X^{i_2\mathrm{}i_{\mathrm{}+1}}_{}^{(+)}D_+X^{i_1}}`$ (3.4)
$`=`$ $`{\displaystyle }dz{\displaystyle \frac{2}{\mathrm{}+1}}a^+\mathrm{}\lambda ^{(+)}{}_{i_1\mathrm{}i_{\mathrm{}+1}}{}^{}_{}^{(+)}D_{+(\mathrm{}+1)}X^{i_1\mathrm{}i_{\mathrm{}+1}}`$
$`=`$ $`{\displaystyle 𝑑z(1)^{\mathrm{}}D_{}\left(\frac{2}{\mathrm{}+1}a^+\mathrm{}\lambda ^{(+)}{}_{i_1\mathrm{}i_{\mathrm{}+1}}{}^{}D_{+(\mathrm{}+1)}^{}X^{i_1\mathrm{}i_{\mathrm{}+1}}\right)},`$
where the last step follows from covariant constancy of $`\lambda ^{(+)}`$ and the chirality of the parameters,
$$D_{}a^+\mathrm{}=D_+a^{\mathrm{}}=0.$$
(3.5)
Hence these transformations are symmetries of the sigma model without boundary. In the torsion-free case the $`\lambda `$s will be the forms which exist on the non-symmetric Riemannian manifolds on Berger’s list. There is no such list in the presence of torsion but the same forms will define reductions of the structure group to the various special holonomy groups. In order to preserve the symmetry on the boundary we must have both left and right symmetries so there must be two independent such reductions. Thus we can say that we are interested in boundary sigma models on manifolds which have bi-$`G`$-structures.
The $`\lambda `$-forms can be used to construct currents $`L_{\pm (\mathrm{}+1)}^{(\pm )}`$,
$$L_{\pm (\mathrm{}+1)}^{(\pm )}:=\lambda ^{(\pm )}{}_{i_1\mathrm{}i_{\mathrm{}+1}}{}^{}D_{\pm (\mathrm{}+1)}^{}X^{i_1\mathrm{}i_{\mathrm{}+1}}$$
(3.6)
If we make both left and right transformations of the type (3.2) we obtain
$`\delta S`$ $`=`$ $`{\displaystyle \frac{2(1)^{\mathrm{}}}{\mathrm{}+1}}{\displaystyle d^2xD_+D_{}\left(D_{}(a^+\mathrm{}L_{\pm (\mathrm{}+1)}^{(+)})D_+(a^{\mathrm{}}L_{(\mathrm{}+1)}^{})\right)}`$ (3.7)
$`=`$ $`{\displaystyle \frac{i(1)^{\mathrm{}+1}}{\mathrm{}+1}}{\displaystyle _\mathrm{\Sigma }}\left(D_+(a^+\mathrm{}L_{\pm (\mathrm{}+1)}^{(+)})D_{}(a^{\mathrm{}}L_{(\mathrm{}+1)}^{})\right).`$
In order for a linear combination of the left and right symmetries to be preserved in the presence of a boundary the parameters should be related by
$`a^+\mathrm{}`$ $`=`$ $`\eta _La^{\mathrm{}},`$ (3.8)
$`D_+a^+\mathrm{}`$ $`=`$ $`\eta \eta _LD_{}a^{\mathrm{}},`$ (3.9)
on the boundary, where $`\eta _L=\pm 1`$.<sup>3</sup><sup>3</sup>3In the case that there is one pair of $`L`$ tensors. This implies that the currents and their superpartners should satisfy the boundary conditions
$`L_{+(\mathrm{}+1)}^{(+)}`$ $`=`$ $`\eta \eta _LL_{(\mathrm{}+1)}^{()},`$ (3.10)
$`D_+L_{+(\mathrm{}+1)}^{(+)}`$ $`=`$ $`\eta _LD_{}L_{(\mathrm{}+1)}^{()}.`$ (3.11)
The boundary condition (3.10) implies
$$\lambda ^{(+)}{}_{i_1\mathrm{}i_{\mathrm{}+1}}{}^{}=\eta _L\eta ^{\mathrm{}}\lambda ^{()}{}_{j_1\mathrm{}j_{\mathrm{}+1}}{}^{}R_{}^{j_1}{}_{i_1}{}^{}\mathrm{}R^{j_{\mathrm{}+1}}{}_{i_{\mathrm{}+1}}{}^{}.$$
(3.12)
The algebra of left (or right) transformations was computed in the torsion-free case in . The commutators involve various generalised Nijenhuis tensors and the classical algebra has a non-linear structure of $`W`$-type. In fact, the generalised Nijenhuis tensors vanish in the absence of torsion. However, this is not the case when torsion is present. The commutator of two plus transformations of the type given in (3.2) is (we drop the pluses on the tensors to simplify matters),
$$[\delta _L,d_M]=\delta _P+\delta _N+\delta _K$$
(3.13)
where $`P`$ and $`N`$ are antisymmetric tensors given by
$$P_{LM}=(LM)_{[L,M]}:=L_{p[L}M^p_{M]}$$
(3.14)
and
$$N_{iLM}=(\mathrm{}+m+1)H_{jk[i}L^j{}_{L}{}^{}M_{}^{k}{}_{M]}{}^{}+(1)^{\mathrm{}}\frac{\mathrm{}m}{6}H_{[i\mathrm{}_1\mathrm{}_2}Q_{L_3M]}.$$
(3.15)
The $`(\mathrm{}+m2)`$-form $`Q`$ is defined by
$$Q_{L_2M_2}=\frac{g^{ij}(LM)_{[iL_2,|j|M_2]}}{n(\mathrm{}+m2)}$$
(3.16)
Here $`L`$ stands for $`\mathrm{}`$ antisymmetrised indices, $`L_2`$ indicates that the first of these should be omitted and so on. Square brackets around the multi-indices indicate antisymmetrisation over all of the indices. The $`\delta _K`$ transformation is generated by the conserved current $`K:=TQ`$, where $`Q:=Q_{L_2M_2}D_{+(\mathrm{}+m2)}X^{L_2M_2}`$. Note that $`P`$ and $`Q`$ can be zero and that $`N`$ is not the Nijenhuis concomitant except in the special case that $`L=M=J`$, an almost complex structure.
The left and right symmetries commute up to the equations of motion. In the case of $`(2,2)`$ models, closure of the left and right algebras separately requires the two type $`(1,1)`$ tensors $`J^{(\pm )}`$ to be complex structures. They need not commute unless one demands off-shell closure without the introduction of further auxiliary fields. However, any two left and right symmetries of the above type commute up to a generalised commutator term as a simple argument shows. Let $`\delta _\pm `$ denote left and right variations with two $`L`$-tensors, of different rank in general. We have
$$\delta _+\delta _{}X^i=\delta _+\left(a^mL^{()i}{}_{K}{}^{}D_{m}^{}X^K\right),$$
(3.17)
where $`K`$ denotes a multi-index with $`m`$ antisymmetrised indices. Since all of the $`K`$ indices are contracted we can replace the $`\delta _+`$ variation by a covariant variation with the Levi-Civita connection provided that we take care of the remaining $`i`$ index. The explicit connection term drops out in the commutator by symmetry. In the remaining terms one can introduce either $`^{()}`$, acting on $`L^{()}`$, or $`^{(+)}`$, acting on $`\delta _+X^k`$, and then show that all of the torsion terms cancel, bar one, again coming from the $`i`$ index. However, this cancels in the commutator too, because the plus and minus connections are swapped in the other term. One thus finds
$`[\delta _+,\delta _{}]X^i`$ $`=`$ $`(1)^nmna^ma^{+n}(L^{()i}{}_{mK_2}{}^{}L_{}^{(+)m}{}_{pL_2}{}^{}L^{(+)i}{}_{mL_2}{}^{}L_{}^{()m}{}_{pL_2}{}^{})\times `$ (3.18)
$`\left(D_{+(l1)}X^{L_2}D_{(m1)}X^{K_2}\right)\times \left(_{}^{(+)}D_+X^p\right),`$
the third factor being the equation of motion. The multi-index $`L`$ associated with $`L^{(+)}`$ stands for $`n`$ antisymmetrised indices.
## 4 Consistency
In this section we shall examine the consistency of the boundary conditions, i.e we investigate the orbits of the boundary conditions under symmetry variations to see if further constraints arise. We shall show that the supersymmetry variation of the $`L`$-boundary condition (3.10) and the $`L`$-variation of the fermion boundary condition (2.8) are automatically satisfied if (3.12) is. To see this we differentiate (3.12) along $`B`$ to obtain
$$Y^{(+)}{}_{k,[i_1}{}^{}{}_{}{}^{m}\lambda _{}^{(+)}{}_{i_2\mathrm{}i_{\mathrm{}+1}]m}{}^{}=0,$$
(4.1)
where
$$Y^{(+)}{}_{i,jk}{}^{}:=(R^1)_{jl}(\stackrel{~}{}_i^{(+)}R^l{}_{k}{}^{}H^l{}_{im}{}^{}R_{}^{m}{}_{k}{}^{}).$$
(4.2)
Note that we have contracted the derivative with $`P`$ rather than $`\pi `$; this is permissible due to the fact that $`P\pi =\pi P=P`$. Equation (4.1) says that $`Y^{(+)}`$, regarded as a matrix-valued one-form, takes its values in the Lie algebra of the group which leaves the form $`\lambda ^{(+)}`$ invariant. The constraint corresponding to the superpartner of the $`L`$-current boundary condition is just the totally antisymmetric part of (4.1).
We now consider the variation of the fermionic boundary condition under $`L`$-transformations. We need to make both left and right transformations which together can be written
$`\delta X^i`$ $`=`$ $`2a^+\mathrm{}P^i{}_{k}{}^{}L_{}^{(+)}{}_{}{}^{k}{}_{j_1\mathrm{}j_{\mathrm{}}}{}^{}D_{+\mathrm{}}^{}X^{j_1\mathrm{}j_{\mathrm{}}}`$ (4.3)
$`=`$ $`2a^{\mathrm{}}P^i{}_{k}{}^{}L_{}^{()}{}_{}{}^{k}{}_{j_1\mathrm{}j_{\mathrm{}}}{}^{}D_{\mathrm{}}^{}X^{j_1\mathrm{}j_{\mathrm{}}}.`$ (4.4)
A straightforward computation yields
$$(2\stackrel{~}{}_{[k}R^i{}_{m]}{}^{}P^i{}_{n}{}^{}H_{}^{n}{}_{[k|p|}{}^{}R_{}^{p}{}_{m]}{}^{})L^{(+)}{}_{}{}^{k}{}_{j_1\mathrm{}j_{\mathrm{}}}{}^{}D_{+(\mathrm{}+1)}^{}X^{j_1\mathrm{}j_{\mathrm{}}m}=0.$$
(4.5)
We define
$$Z_{i,jk}^{(+)}=(R^1)_{il}(2\stackrel{~}{}_{[j}R^l{}_{k]}{}^{}+P^l{}_{m}{}^{}H_{}^{m}{}_{n[j}{}^{}R_{}^{n}{}_{k]}{}^{}),$$
(4.6)
which is the term in the bracket in (4.5) multiplied by $`R^1`$. We claim that
$$Y_{i,jk}^{(+)}=Z_{i,jk}^{(+)}.$$
(4.7)
This can be proved using (2.29) with the aid of a little algebra. Thus we have shown that, if the boundary conditions (3.12) are consistent, then the constraints following from supersymmetry variations of the $`L`$-constraints and from $`L`$-variations of the fermionic boundary condition are guaranteed to be satisfied.
If $`\lambda ^{(+)}=\lambda ^{()}:=\lambda `$ the boundary condition (3.12) typically implies that $`\pm R`$ is an element of the group which preserves $`\lambda `$. If this is the case, then (4.1) becomes an identity. However, it can happen that $`R`$ is not an element of the invariance group but that $`R^1dR`$ still takes its values in the corresponding Lie algebra. For example, if $`\lambda `$ is the two-form of a $`2m`$-dimensional Kähler manifold and the sign $`\eta _L\eta =1`$, $`R`$ is not an element of the unitary group but, since it must have mixed indices, it is easy to see that $`R^1dR`$ is itself $`\text{u}(m)`$-valued.
A similar argument applies in the general case, when $`\lambda ^{(+)}\lambda ^{()}`$. In the next section we discuss how the plus and minus forms are related by an element $`V`$ of the orthogonal group (see (5.5)). Thus equation (3.12) can be written
$$\lambda ^{()}{}_{i_1\mathrm{}i_{\mathrm{}+1}}{}^{}=\eta _L\eta ^{\mathrm{}}\lambda ^{()}{}_{j_1\mathrm{}j_{\mathrm{}+1}}{}^{}\widehat{R}_{}^{j_1}{}_{i_1}{}^{}\mathrm{}\widehat{R}^{j_{\mathrm{}+1}}{}_{i_{\mathrm{}+1}}{}^{},$$
(4.8)
where $`\widehat{R}:=RV^1`$. If we differentiate $`(\text{4.8})`$ along the brane with respect to the minus connection we can then use the above argument applied to $`\widehat{R}`$.
## 5 Target space geometry
In this section we discuss the geometry of the sigma model target space in the presence of torsion when the holonomy groups of the torsion-full connections $`^{(\pm )}`$ are of special type, specifically $`G_2,Spin(7)`$ and $`SU(3)`$. We use only the data given by the sigma model and use a simple approach based on the fact that there is a transformation which takes one from one structure to the other. We begin with $`G_2`$ and then derive the other two cases from this by dimensional reduction and oxidation.
### $`G_2`$
In this case we have a seven-dimensional Riemannian manifold $`(M,g)`$ with two $`G_2`$-forms $`\phi ^{(\pm )}`$ which are covariantly constant with respect to left and right metric connections $`^{(\pm )}`$ such that the torsion tensor is $`\pm H`$. $`G_2`$ manifolds with torsion have been studied in the mathematical literature and have arisen in supergravity solutions . Bi-$`G_2`$-structures have also appeared in this context and have been given an interpretation in terms of generalised $`G_2`$-structures . They can be studied in terms of a pair of covariantly constant spinors from which one can construct the $`G_2`$-forms, as well as other forms, as bilinears. We will not make use of this approach here, preferring to use the tensors given to us naturally by the sigma model. As noted in there is a common $`SU(3)`$ structure associated with the additional forms. We shall derive this from a slightly different perspective here.
In most of the literature use is made of the dilatino Killing spinor equation which restricts the form of $`H`$. The classical sigma model does not appear to require this restriction as the dilaton does not appear until the one-loop level. The dilatino equation is needed in order to check that one has supersymmetric supergravity solutions but is not essential for our current purposes.
For $`G_2`$ there are two covariantly constant forms, the three-form $`\phi `$ and its dual four-form $`\phi `$ (we shall drop the star when using indices). The metric can be written in terms of them. A convenient choice for $`\phi `$ is
$$\phi =\frac{1}{3!}\phi _{ijk}e^{ijk}=e^{123}e^1(e^{47}+e^{56})+e^2(e^{46}e^{57})e^3(e^{45}+e^{67})$$
(5.1)
This form is valid in flat space or in an orthonormal basis, the $`e^i`$s being basis forms. Another useful way of think about the $`G_2`$ three-form is to write it in a $`6+1`$ split. We then have
$`\phi _{ijk}`$ $`=`$ $`\lambda _{ijk}`$
$`\phi _{ij7}`$ $`=`$ $`\omega _{ij}`$
$`\phi _{ijk7}`$ $`=`$ $`\widehat{\lambda }_{ijk},`$ (5.2)
where $`i,j,k=1\mathrm{}6`$, and $`\{\lambda ,\widehat{\lambda },\omega \}`$ are the forms defining an $`SU(3)`$ structure in six dimensions. The three-forms $`\lambda `$ and $`\widehat{\lambda }`$ are the real and imaginary parts respectively of a complex three-form $`\mathrm{\Omega }`$ which is of type $`(3,0)`$ with respect to the almost complex structure defined by $`\omega `$.
On a $`G_2`$ manifold with skew-symmetric torsion, the latter is uniquely determined in terms of the Levi-Civita covariant derivative of $`\phi `$ . This follows from the covariant constancy of $`\phi `$ with respect to the torsion-full connection.
Now suppose we have a bi-$`G_2`$-structure. The two $`G_2`$ three-forms are related to one another by an $`SO(7)`$ transformation, $`V`$. If we start from $`\phi ^{()}`$ this will be determined up to an element of $`G_2^{()}`$. So we can choose a representative to be generated by an element $`w\text{s}\text{o}(7)`$ of the coset algebra with respect to $`\text{g}_2^{()}`$. This can be written
$$w_{ij}=\phi _{ijk}^{()}v^k$$
(5.3)
and $`V=e^w`$. The vector $`v`$ will be specified by a unit vector $`N`$ and an angle $`\alpha `$. It is straightforward to find $`V`$,
$$V^i{}_{j}{}^{}=\mathrm{cos}\alpha \delta ^i{}_{j}{}^{}+(1\mathrm{cos}\alpha )N^iN_j+\mathrm{sin}\alpha \phi ^{()}{}_{}{}^{i}{}_{jk}{}^{}N_{}^{k}.$$
(5.4)
Using
$$\phi ^{(+)}=\phi ^{()}V^3,$$
(5.5)
where one factor of $`V`$ acts on each of the three indices of $`\phi `$, we can find the relation between the two $`G_2`$ forms explicitly,
$$\phi _{ijk}^{(+)}=A\phi _{ijk}^{()}+B\phi _{ijkl}^{()}N^l+3C\phi _{[ij}^{()}{}_{}{}^{l}N_{k]}^{}N_l,$$
(5.6)
where
$$A=\mathrm{cos}3\alpha ,B=\mathrm{sin}3\alpha ,C=1\mathrm{cos}3\alpha .$$
(5.7)
The dual four-forms are related by
$$\phi _{ijkl}^{(+)}=(A+C)\phi _{ijkl}^{()}4B\phi _{[ijk}^{()}N_{l]}4C\phi _{[ijk}^{()}{}_{}{}^{m}N_{l]}^{}N_m.$$
(5.8)
The angle $`\alpha `$ is related to the angle between the two covariantly constant spinors. To simplify life a little we shall follow and choose these spinors to be orthogonal which amounts to setting $`\mathrm{cos}\frac{\alpha }{2}=0`$. We then find
$$\phi _{ijk}^{(+)}=\phi _{ijk}^{()}+6\phi _{[ij}^{()}{}_{}{}^{l}N_{k]}^{}N_l.$$
(5.9)
and
$$\phi _{ijkl}^{(+)}=\phi _{ijkl}^{()}8\phi _{[ijk}^{()}{}_{}{}^{m}N_{l]}^{}N_m.$$
(5.10)
We can use the vector $`N`$ to define an $`SU(3)`$ structure as above. We set
$$\omega =i_N\phi ^{()};\lambda =\phi ^{()}\omega N;\widehat{\lambda }=i_N\phi ^{()}.$$
(5.11)
The three-form $`\widehat{\lambda }`$ is the six-dimensional dual of $`\lambda `$ and the set of forms $`\{\omega ,\lambda ,\widehat{\lambda }\}`$ is the usual set of forms associated with an $`SU(3)`$ structure in six dimensions. For the plus forms we have
$`i_N\phi ^{(+)}`$ $`=`$ $`\omega `$
$`\phi ^{(+)}\omega N`$ $`=`$ $`\lambda `$
$`i_N\phi ^{(+)}`$ $`=`$ $`\widehat{\lambda }.`$ (5.12)
Thus a bi-$`G_2`$-structure is equivalent to a single $`G_2`$ structure together with a unit vector (and an angle to be more general). The unit vector $`N`$ then allows one to define a set of $`SU(3)`$ forms as above. In it is shown that the projector onto the six-dimensional subspace is integrable, but this presupposes that the dilatino Killing spinor equation holds. Since we make no use of this equation in this paper it need not be the case that integrability holds.
It is straightforward to construct a covariant derivative $`\widehat{}`$ which preserves both $`G_2`$ structures. This connection has torsion but this is no longer totally antisymmetric. It is enough to show that the covariant derivatives of $`N`$ and $`\phi ^{()}`$ are both zero. If we write
$$\widehat{}_iN_j=_i^{()}N_jS_{i,j}{}_{}{}^{k}N_{k}^{},$$
(5.13)
where $`S_{i,jk}=S_{i,kj}`$, then these conditions are fulfilled if
$$S_{i,jk}=\frac{1}{2}H_{ijk}+\frac{1}{4}H_i{}_{}{}^{lm}\phi _{lmjk}^{()}\frac{3}{2}H_{ilm}\mathrm{\Pi }^l{}_{j}{}^{}\mathrm{\Pi }_{}^{m}{}_{k}{}^{}\frac{3}{4}H_i{}_{}{}^{lm}\phi _{}^{m}i_{jkln}N^nN_m.$$
(5.14)
Here $`\mathrm{\Pi }^i{}_{j}{}^{}:=\delta ^i{}_{j}{}^{}N^iN_j`$ is the projector transverse to $`N`$.
### $`SU(3)`$
Manifolds with $`SU(3)\times SU(3)`$ have arisen in recent studies of supergravity solutions with flux -. They have also been discussed in a recent paper on generalised calibrations . A bi-$`SU(3)`$ structure on a six-dimensional manifold is given by a pair of a pair of forms $`\{\omega ^{(\pm )},\mathrm{\Omega }^{(\pm )}\}`$ of the above type which are compatible with the metric. If the sigma model algebra closes off-shell the complex structures will be integrable. The transformation relating the two structures can be found using a similar construction to that used in the $`G_2`$ case. However, we can instead derive the relations between the plus and minus forms by dimensional reduction from $`G_2`$. To this end we introduce a unit vector $`N^{}`$, which we can take to be in the seventh direction, and define the $`SU(3)`$ forms as in equation (5.2) above. We consider only the simplified bi-$`G_2`$-structure and we also then take the unit vector $`N`$ to lie within the six-dimensional space. The unit vector $`N`$ now defines an $`SO(6)`$ transformation. The relations between the plus and minus forms are given by
$`\omega _{ij}^{(+)}`$ $`=`$ $`\omega _{ij}^{()}+4\omega _{[i}^{()}{}_{}{}^{k}N_{j]}^{}N_k`$
$`\lambda _{ijk}^{(+)}`$ $`=`$ $`\lambda _{ijk}^{()}+6\lambda _{[ij}{}_{}{}^{l}N_{k]}^{}N_l.`$
$`\widehat{\lambda }_{ijk}^{(+)}`$ $`=`$ $`\widehat{\lambda }_{ijk}^{()}6\widehat{\lambda }_{[ij}^{()}{}_{}{}^{l}N_{k]}^{}N_l.`$ (5.15)
We can rewrite this in complex notation if we introduce the three-forms $`\mathrm{\Omega }^{(\pm )}:=\lambda ^{(\pm )}+i\widehat{\lambda }^{(\pm )}`$ and split $`N`$ into $`(1,0)`$ and $`(0,1)`$ parts, $`n,\overline{n}`$. So
$$N_i=n_i+\overline{n}_i;i\omega _{ij}N^j=n_i\overline{n}_i.$$
(5.16)
Note that $`n\overline{n}=\frac{1}{2}`$. Then equations (5.15) are equivalent to
$`\omega _{ij}^{(+)}`$ $`=`$ $`\omega _{ij}^{()}2in_{[i}\overline{n}_{j]}`$
$`\mathrm{\Omega }_{ijk}^{(+)}`$ $`=`$ $`6\overline{\mathrm{\Omega }}_{[ij}^{()}{}_{}{}^{l}n_{k]}^{}n_l.`$ (5.17)
This type of bi-$`SU(3)`$-structure is therefore equivalent to a single $`SU(3)`$ structure together with a normalised $`(1,0)`$-form.
### $`Spin(7)`$
A $`Spin(7)`$ structure on an eight-dimensional Riemannian manifold is specified by a self-dual four-form $`\mathrm{\Phi }`$ of a certain type. In a given basis its components can be constructed from those of the $`G_2`$ three-form. Thus
$$\mathrm{\Phi }_{abcd}=\phi _{abcd};\mathrm{and}\mathrm{\Phi }_{abc8}=\phi _{abc},$$
(5.18)
where, in this section, $`a,b,\mathrm{}`$ run from 1 to 7 and $`i,j,\mathrm{}`$ run from 1 to 8. $`Spin(7)`$ geometry with skew-symmetric torsion has been discussed and generalised $`Spin(7)`$ structures have also been studied . A bi-$`Spin(7)`$-structure on a Riemannian manifold consists of a pair of such forms, covariantly constant with respect to $`^{(\pm )}`$. We can again get from the minus form to the plus form by an orthogonal transformation, but since the dimension of $`SO(8)`$ minus the dimension of $`Spin(7)`$ is seven it is described by seven parameters. In the presence of a $`Spin(7)`$ structure one of the chiral spinor spaces, $`\mathrm{\Delta }_s`$, say, splits into one- and seven-dimensional subspaces, $`\mathrm{\Delta }_s=\text{}\mathrm{\Delta }_7`$. The transformation we seek will be described by a unit vector $`n^a\mathrm{\Delta }_7`$ together with an angle.
It will be useful to introduce some invariant tensors for $`Spin(7)`$ using this decomposition of the spin space. We set
$`\varphi _{ajk}`$ $`=`$ $`\{\begin{array}{cc}\varphi _{abc}=\phi _{abc}\hfill & \\ \varphi _{ab8}=\delta _{ab}\hfill & \end{array}`$ (5.19)
$`\varphi _{abkl}`$ $`=`$ $`\{\begin{array}{cc}\varphi _{ab}{}_{}{}^{cd}=\phi _{ab}{}_{}{}^{cd}2\delta _{[ab]}^{cd}\hfill & \\ \varphi _{abc8}=\phi _{abc}\hfill & \end{array},`$ (5.20)
where $`\phi _{abc}`$ is the $`G_2`$ invariant. It will also be useful to define
$$\varphi _{aijkl}:=\varphi _{ab[ij}\varphi ^b{}_{kl]}{}^{}.$$
(5.21)
The $`Spin(7)`$ form itself can be written as
$$\mathrm{\Phi }_{ijkl}=\varphi _{a[ij}\varphi ^a{}_{kl]}{}^{}.$$
(5.22)
The space of two-forms splits into $`7+21`$, and one can project onto the seven-dimensional subspace by means of $`\varphi _{ajk}`$. With these definitions we can now oxidise the $`G_2`$ equations relating the plus and minus structure forms to obtain
$$\mathrm{\Phi }_{ijkl}^{(+)}=\mathrm{\Phi }_{ijkl}^{()}6n_an_b\varphi ^{()a}{}_{[ij}{}^{}\varphi _{}^{()b}{}_{kl]}{}^{}.$$
(5.23)
Here the unit vector $`N`$ in the $`G_2`$ case becomes the unit spinor $`n`$.
## 6 Examples of solutions
In this section we look at solutions to the boundary conditions for the additional symmetries which can be identified with various types of brane. We shall go briefly through the main examples, confining ourselves to $`U(\frac{n}{2})`$, $`SU(\frac{n}{2})`$ and the exceptional cases $`G_2`$ and $`Spin(7)`$.
### $`U(\frac{n}{2})U(m)`$
This case corresponds to $`N=2`$ supersymmetry. For $`H=F=0`$ we assume that the supersymmetry algebra closes off-shell so that $`M`$ is a Kähler manifold with complex structure $`J`$, hermitian metric $`g`$ and Kähler form $`\omega `$. The Kähler form is closed and covariantly constant. The boundary conditions for the second supersymmetry, which can be viewed as an additional symmetry with $`\lambda =\omega `$, imply
$$\omega _{ij}=\pm \omega _{kl}R^k{}_{i}{}^{}R_{}^{l}{}_{j}{}^{}.$$
(6.1)
Thus there are two possibilities, type A where $`JR=RJ`$ and type B where $`JR=RJ`$ . Consider type B first. In this case the brane inherits a Kähler structure from the target space and so has dimension $`2k`$. If there is a non-vanishing gauge field $`F`$, it must be of type $`(1,1)`$ with respect to this structure. The calibration form is $`\omega ^k`$.
For type B with zero $`F`$ field, $`J`$ is off-diagonal in the orthonormal basis in which $`R`$ takes its canonical form
$$R=\left(\begin{array}{cc}1_p& 0\\ 0& 1_q\end{array}\right),$$
(6.2)
where $`p`$ and $`q`$ denote the dimensions of $`B`$ and the transverse tangent space, $`p+q=n`$, and $`1_p,\mathrm{\hspace{0.17em}1}_q`$ denote the corresponding unit matrices. The only possibility is $`p=q=m`$. The Kähler form vanishes on both the tangent and normal bundles to the brane, so that the brane is Lagrangian.
When the $`F`$ field is non-zero the situation is more complicated. We may take $`R`$ to have the same block-diagonal form as in (6.2) but with $`1_p`$ replaced by $`R_p`$. From (2.19)
$$R_p=(1+F)^1(1F)$$
(6.3)
The analysis of $`JR=RJ`$ shows that the brane is coisotropic . This means that there is a $`4k`$-dimensional subspace in each tangent space to the brane where $`J`$ is non-singular, there is an $`r`$-dimensional subspace on which it vanishes, and the dimension of the normal bundle is also $`r`$. The product $`(J_pF)`$ is an almost complex structure and both $`J_p`$ and $`F`$ are of type $`(2,0)+(0,2)`$ with respect to $`(J_pF)`$. For $`m=3`$ we can therefore only have $`p=5`$. For $`m=4`$ we can have $`p=5`$ but we can also have a space-filling brane with $`p=8`$.
$`N=2`$ sigma models with boundary and torsion have been discussed in ; the geometry associated with the boundary conditions is related to generalised complex geometry .
### $`SU(\frac{n}{2})SU(m)`$
In the Calabi-Yau case we have, in addition to the Kähler structure, a covariantly constant holomorphic $`(m,0)`$ form $`\mathrm{\Omega }`$ where $`m=\frac{n}{2}`$. There are two independent real covariantly constant forms, $`\lambda `$ and $`\widehat{\lambda }`$, which can be taken to be the real and imaginary parts of $`\mathrm{\Omega }`$. The corresponding $`L`$-tensors which define the symmetry transformations are related by
$$\widehat{L}^i{}_{j_1\mathrm{}j_{m1}}{}^{}=J^i{}_{k}{}^{}L_{}^{k}_{j_1\mathrm{}j_{m1}}$$
(6.4)
Because there are now two currents we can introduce a phase rather than a sign in the boundary condition. Thus
$$\mathrm{\Omega }_{i_1\mathrm{}i_m}=e^{i\alpha }\mathrm{\Omega }_{j_1\mathrm{}j_m}R^{j_1}{}_{i_1}{}^{}\mathrm{}R^{j_m}_{i_m}$$
(6.5)
A second possibility is that $`\mathrm{\Omega }`$ on the right-hand side is replaced by $`\overline{\mathrm{\Omega }}`$. For type B branes, the displayed equation is the correct condition. The $`R`$-matrix is the sum of holomorphic and anti-holomorphic parts, $`R=\overline{}`$, and (6.5) implies that
$$\mathrm{det}=e^{i\alpha }.$$
(6.6)
If $`F=0`$ this fixes the phase, but if $`F0`$ it imposes a constraint on $`F`$ which must in any case be a $`(1,1)`$ form (from $`JR=RJ`$) . The constraint is
$$\mathrm{det}_p=e^{i\alpha }(1)^{\frac{q}{2}},$$
(6.7)
or
$$\mathrm{det}(1+f)=e^{i\alpha }(1)^{\frac{q}{2}}\mathrm{det}(1f)$$
(6.8)
where $`f^a{}_{b}{}^{}=g^{a\overline{c}}F_{\overline{c}b}`$, in a unitary basis.
For type A branes, from $`JR=RJ`$ it follows that $`R`$ maps holomorphic vectors to anti-holomorphic ones and vice versa so that $`\overline{\mathrm{\Omega }}`$ must be used in (6.5). In the case that $`F=0`$ the brane is a SLAG with $`\text{R}\text{e}\mathrm{\Omega }`$ as the calibration form. For $`F0`$ we have coisotropic branes with an additional constraint on the gauge field .
The geometry of the bi-$`SU(m)`$ case has been studied from the point of view of generalised geometry and generalised calibrations in .
### $`G_2`$
The boundary conditions associated with the $`G_2`$ currents are
$`\phi `$ $`=`$ $`\eta _L\phi R^3`$
$`\phi `$ $`=`$ $`\eta _L\phi R^4\mathrm{det}R`$ (6.9)
We consider first $`F=H=0`$. From the first of these equations it follows that $`(\eta _LR)G_2`$. From this it follows that the sign in the boundary condition for $`\phi `$ is always positive because the sign of the determinant of $`R`$ is equal to $`\eta _L`$. Thus the second constraint reduces to $`\phi =\phi R^4`$.
There are two possibilities depending on the sign of $`\eta _L`$. If it is positive then non-zero components of $`\phi `$ must have an even number of normal indices, whereas if it is negative they must have an odd number of non-zero components. Since $`(\eta _LR)G_2`$, and is symmetric, it can be diagonalised by a $`G_2`$ matrix so that we can bring $`R`$ to its canonical form in a $`G_2`$ basis. Looking at the components of $`\phi `$ we see that the only possibilities which are compatible with the preservation of the non-linear symmetries on the boundary are either $`\eta _L=1`$, in which case $`B`$ is a three-dimensional associative cycle, or $`\eta _L=1`$ in which case $`B`$ is a four-dimensional co-associative cycle .
Now let us turn to $`F0`$, but $`H=0`$. We shall assume that the tangent bundle $`M`$, restricted to the brane, splits into three, $`TM|_B=T_1T_2N`$, of dimensions $`p_1,p_2`$ and $`q`$ respectively. $`N`$ is the normal bundle and $`R|_{T_2}=1_{p_2}`$. If there is at least one normal direction we may assume that one of these is $`7`$ in the conventions of (5.2). Thus the problem is reduced to a six-dimensional one, at least algebraically. The six-dimensional boundary conditions are (where $`R`$ is now a $`6\times 6`$ matrix),
$`\lambda `$ $`=`$ $`\eta _L\lambda R^3`$
$`\widehat{\lambda }`$ $`=`$ $`\eta _L\widehat{\lambda }R^3\mathrm{det}R`$
$`\omega `$ $`=`$ $`\eta _L\omega R^2,`$ (6.10)
in an obvious notation. If the sign is negative the brane is type B, whereas if $`\eta _L=+1`$ we have type A. These are the same conditions as we have just discussed in the preceding section, the only difference being that the phase is not arbitrary. The constraints on the $`F`$ field are therefore slightly stronger.
The last possibility is a space-filling brane in seven dimensions. Since $`F`$ is antisymmetric there must be at least one trivial direction for $`R`$ so that we can again reduce the algebra to the six-dimensional case. The only possibilty is $`\eta =+1`$ in which case we have type B. The non-trivial dimension must be even, and since $`\mathrm{det}=1`$ the case $`p=2`$ is also trivial.
Now let us consider the case with torsion. The boundary condition for the non-linear symmetries associated with the forms yield
$`\phi ^{(+)}`$ $`=`$ $`\eta _L\phi ^{()}R^3`$
$`\phi ^{(+)}`$ $`=`$ $`\eta _L\phi ^{()}R^4\mathrm{det}R,`$ (6.11)
When the brane is normal to $`N`$ we find, on the six-dimensional subspace,
$`\lambda `$ $`=`$ $`\eta \lambda R^3`$
$`\widehat{\lambda }`$ $`=`$ $`\eta \widehat{\lambda }R^3\mathrm{det}R`$
$`\omega `$ $`=`$ $`\eta \omega R^2,`$ (6.12)
The analysis is very similar to the case of zero torsion with $`F`$. One finds that $`\eta _L=1`$ corresponds to type B while $`\eta _L=+1`$ is type A. In particular, for type B there is a five-brane which corresponds to the five-brane wrapped on a three-cycle discussed in the supergravity literature .
### $`Spin(7)`$
In the absence of torsion, the boundary condition associated with the conserved current is
$$\mathrm{\Phi }=\widehat{\eta }\mathrm{\Phi }R^4,$$
(6.13)
for some sign factor $`\widehat{\eta }`$. If this is negative then $`\mathrm{det}R`$ is also negative so that the dimension of $`B`$ must be odd. Furthermore, $`\mathrm{\Phi }`$ must have an odd number of normal indices with respect to the decomposition of the tangent space induced by the brane. However, one can show that such a decomposition is not compatible with the algebraic properties of $`\mathrm{\Phi }`$. Therefore the sign $`\widehat{\eta }`$ must be positive. It is easy to see that a four-dimensional $`B`$ is compatible with this, and indeed we then have the standard Cayley calibration with $`\mathrm{\Phi }`$ pulled-back to the brane being equal to the induced volume form. On the other hand if $`B`$ has either two or six dimensions one can show that it is not compatible with the $`Spin(7)`$ structure. As one would expect, therefore, the only brane compatible with the non-linear symmetry associated with $`\mathrm{\Phi }`$ on the boundary is the Cayley cycle .
If $`F0`$, but $`H=0`$, and if we assume that there is at least one direction normal to the brane, then the $`Spin(7)`$ case reduces to $`G_2`$ (with $`F0`$). If the brane is space-filling but there is at least one trivial direction, then there must be at least two by symmetry and again we recover the $`G_2`$ case. But we can also have a space-filling brane which is non-trivial in all eight directions.
## Acknowledgements
This work was supported in part by EU grant (Superstring theory) MRTN-2004-512194, PPARC grant number PPA/G/O/2002/00475 and VR grant 621-2003-3454. PSH thanks the Wenner-Gren foundation. |
warning/0507/astro-ph0507462.html | ar5iv | text | # The Magnetic Structure of Coronal Loops Observed by TRACE
## 1 Introduction
Soft X-ray and EUV observations have revealed that the solar corona is a very hot and highly structured medium (e.g., Orrall 1981; Bray et al. 1991). It is clear that the magnetic field plays a dominant role in structuring the plasma, and it is very likely that the field also plays a fundamental role in the heating. Understanding the detailed properties of the coronal magnetic field is therefore a high priority. One particular property—the apparent uniformity in the thickness of coronal loops and their associated magnetic flux tubes—is very puzzling and could be a vital clue as to the origin of coronal heating.
Because the magnetic Reynolds number is so large in the corona, the plasma and magnetic field are “frozen” together. Observed structures such as soft X-ray and EUV loops trace out magnetic field lines that are rooted in the photosphere. Throughout much of the corona, and especially in the coronal part of active regions, the magnetic pressure dominates the plasma pressure ($`\beta <<1`$), and the field is approximately force-free such that
$$\times 𝐁=\alpha 𝐁,$$
(1)
where $`\alpha `$ is generally field-line dependent. Under these conditions the strength of the field must tend to decrease away from the solar surface, and a majority of the flux tubes that make up the field must therefore expand with height. This is not required of all flux tubes, however. The overall expansion of large-scale magnetic configurations such as active regions is well established, but thin plasma loops are observed to have a nearly constant thickness along their length (Klimchuk et al. 1992).
This surprising result has been confirmed by Klimchuk (2000) and Watko & Klimchuk (2000) who studied sizable collections of loops observed by the Soft X-ray Telescope (SXT) on Yohkoh and the Transition Region and Coronal Explorer (TRACE), respectively. The loops are only slightly wider at their midpoints than at their footpoints, implying a much smaller expansion than is present in standard magnetic field models. In addition, there is only modest variation of width along each loop, suggesting that the cross-section must be approximately circular if the field has non-zero twist.
Klimchuk, Antiochos & Norton (2000) suggested that the X-ray and EUV loops might correspond to significantly twisted magnetic flux tubes that are surrounded by relatively untwisted field and faint plasma. Parker (1977) and others had shown that the magnetic tension associated with the azimuthal component of the field would cause a constriction in the cores of straight twisted tubes. Klimchuk et al. argued that this constriction would be greater in the thicker, i.e., higher, parts of realistic curved loops. They constructed three-dimensional force-free field models which showed that twist can indeed promote thickness uniformity, but probably not to the degree implied by observations. The models also indicate that twist tends to circularize the loop cross section in the corona.
One important limitation of the observational studies cited above is that they did not compare the Yohkoh and TRACE loops with the corresponding magnetic flux tubes obtained, for example, from magnetic field extrapolation models based on magnetograms observed at the same time. Instead, the studies compared the typical expansion of observed loops with the characteristic expansion of generic magnetic field models. Since not all of the flux tubes in a realistic field configuration are expected to expand, it is possible that the observed loops correspond to the subset of non-expanding tubes. If that were the case, it would have important physical implications. For example, Longcope (1996) has suggested that coronal heating comes from reconnection at magnetic separators and that the magnetic field near separators has minimal expansion. The work we present here is the first comprehensive study involving one-to-one comparisons of observed loops with their corresponding flux tubes.
Our strategy is to compute linear force-free extrapolation models based on photospheric magnetograms from the Michelson Doppler Imager (MDI) on the Solar and Heliospheric Observatory (SOHO) and to compare them with carefully coaligned and nearly cotemporal images from TRACE. We vary the force-free parameter $`\alpha `$ in equation (1) to obtain the best possible fit between the model field and the observed loop, and we construct flux tubes by tracing field lines assuming a variety of possible cross sections at the footpoint. We then compare the expansion of the loops and corresponding flux tubes and evaluate the extent to which they agree.
The paper is organized as follows. Section 2 concerns the plasma loops observed by TRACE, henceforth referred to simply as “loops.” Section 3 concerns the magnetic flux tubes constructed from the extrapolation models, henceforth referred to simply as “flux tubes.” Section 4 presents the comparison of expansion factors measured for the loops and flux tubes. Section 5 is a discussion of the results and their significance.
## 2 Observed TRACE Loops
### 2.1 Data Description
Our TRACE dataset consist of full resolution (0.5 arcsec $`0.36Mm`$ pixel) images obtained in the Fe IX 171 Å passband (Handy et al. 1999). The dates, times, and positions of the images are given in Table 1. These particular images were chosen because clean loops (relatively unobscured by background emission or overlapping loops) could be identified and because the active regions are close to disk center, which is important for the magnetic field modeling. The data were processed using the standard SolarSoft analysis tools.
We selected 20 loops from 3 active regions for detailed study, as indicated in Table 1. We chose the time difference between consecutive images to be long enough that a different configuration of loops is observed. Therefore, the same loop is not chosen in consecutive images of the same active region. Each loop is fully contained within the active region, which is important for the extrapolation procedure. Upper panels in Figures 1 and 2 show two of the TRACE images used. In the lower panels the loops are labeled and corresponding model field lines are shown (see Section 3.3).
### 2.2 Loop Analysis
We follow the analysis procedure of Klimchuk et al. (1992). We start by making a straightened version of the loop. We visually select a set of points that we believe to lie along the loop axis and fit them with a polynomial, which then defines one axis of a new coordinate system. The other axis is orthogonal to it at each axis position. Intensities are assigned to a regular grid of pixels in the new coordinate system using a weighted average of the four nearest pixels in the original coordinate system. This results in a rectangular image with the loop running vertically down the middle. We next determine a background by subjectively outlining the edges of the straightened loop and performing linear interpolations between the intensities at the two edges in each row. This is the same procedure used on TRACE loops by Watko and Klimchuk (2000). Klimchuk et al. (1992) and Klimchuk (2000) used a polynomial surface fitting procedure to determine the background of Yohkoh loops, but this can introduce spurious results when the background is highly structured, as is often the case in TRACE images. Finally, we compute the standard deviation (i.e., the second moment) of the intensity profile along each row of the background-subtracted image:
$$\sigma =\left[\frac{\left(x_i\mu \right)^2I_i}{I_i}\right]^{1/2},$$
(2)
where the summation is taken over the $`x_i`$ positions in the profile, and $`\mu `$ is the mean position:
$$\mu =\frac{x_iI_i}{I_i}.$$
(3)
It is easy to demonstrate that the standard deviation of the intensity profile is 1/4 of the loop diameter for the case of a circular, uniformly filled cross section and observations with perfect resolution. Many authors freely use the term “diameter” in discussing loop thickness, but we are reluctant to do so, because it implies that the cross section is necessarily circular. Instead, we use the term “width.” To facilitate comparison with other published results, we define the width to be 4 times the standard deviation. The triangles in Figure 3 show the width plotted as a function of position along the loop for four example loops. The units are Mm in this and all subsequent figures. Note how the width is fairly constant along each loop. There is considerable small-scale structure, but little evidence for large-scale trends.
In a recent study of TRACE loops by Aschwanden and Nightingale (2005), the intensity profiles are fit to Gaussians with a linear background. For a Gaussian, the full width at half maximum (FWHM) is 2.35 times as large as the standard deviation, or 0.59 times the width as we have defined it. The mean FWHM of 1.42 Mm measured by Aschwanden and Nightingale corresponds to a mean width of 2.41 Mm, which is very similar to the values we have measured (e.g., triangles in Figure 3).
Our background subtraction procedure is of course not perfect, and residual non-loop emission may be present in the background-subtracted images. Since intensity is multiplied by the square of position in Equation 2, there may be concern that residual emission in the tails of the intensity profile could have an especially strong influence on $`\sigma `$. In the Appendix section we explore this possibility in some detail. There we determine the widths of the loops of Figure 3 using simultaneous measurements of the standard deviation, the FWHM, and the Equivalent Width. The results obtained using the three different methods are very similar. In particular, fluctuations along the loops, which could be an indication of errors in the background subtraction, are of the same order. This confirms the suitability of using the standard deviation of the intensity profile for width determinations.
The assumption of perfect resolution is of course not realistic. To quantitatively estimate the effects of finite resolution, we have simulated the observation of a circular cross section loop with a point spread function (PSF) appropriate to the combined telescope/detector system of TRACE. Figure 4 shows the resulting relationship between the computed standard deviation and the actual loop diameter (width). The curve rolls over from the straight line of slope 4 and intersects the abscissa at a value of 0.33 Mm, which is the standard deviation of the PSF itself.
The PSF used in Figure 4 is a Gaussian with a full width at half maximum (FWHM) of 2.25 pixels (approximately 0.82 Mm). Golub et al. (1999) applied a blind iterative technique to TRACE images and found that the PSF is in fact not azimuthally symmetric. The FWHM of the major and minor axes were estimated to be 3 and 2 pixels, respectively. The estimate of the major axis was subsequently revised to be closer to 2.5 pixels (Golub, 2003, private communication). Recent analysis by Nightingale (2003, private communication) indicates major and minor axes of only 2.0 and 1.6 pixels. We feel that our choice of 2.25 pixels for both axes is conservative and, if anything, overestimates the actual smearing of the telescope and detector. We note that the conversion curve given in Watko and Klimchuk (2000) is not quite correct. It assumes a FWHM of 2.5 pixels for the telescope PSF alone and separately treats the averaging over finite pixel area. The authors did not recognize at the time that the Golub et al. (1999) result represents a combined PSF of the telescope/detector system.
The asterisks in Figure 3 are obtained from the standard deviations using the conversion curve of Figure 4. We refer to these as resolution-corrected widths. As expected from the nonlinearity of the conversion curve, the difference between the corrected and uncorrected values (asterisks and triangles respectively in Figure 3) is greatest for the narrowest loop regions. For measured standard deviations close to the theoretical minimum, the correction can be very large. It sometimes happens that the measured value is smaller than the theoretical minimum, in which case the width is set to zero. See, for example, the upper-right panel in Figure 3. We have rejected loops for which 20% or more of the inferred widths are under the resolution limit.
We believe that the greatest source on uncertainty in the measurement is the subtraction of the background emission from the loop. To estimate this uncertainty, our analysis routine automatically repeats the width measurement after redefining the edge of the loop to be one pixel wider on each side. We find that the width obtained in this way differs from the original width by about 20% on average. Furthermore, it is wider than the original width in a large majority of cases. We conclude from this that our subjective identification of the loop tends to miss the faint outer “wings” of the profile.
Although we most often underestimate the width, it is nonetheless appropriate to consider the possibility of an overestimate when establishing the uncertainties. An overestimate uncertainty of 20% seems unreasonably large, so we adopt a value of 10%. The solid lines in Figure 3 are resolution-corrected widths obtained from uncorrected widths that are 20% higher and 10% lower than the original measured values. The lines are 5-point running averages, which makes the plots more readable, but does not affect any conclusions since large width variations over short distances are not reliable. We refer to the lines as the error bars.
### 2.3 Loop Expansion Factor
To quantify systematic variations in loop width, we define an expansion factor $`\mathrm{\Gamma }`$ to be the ratio of the average widths measured in different segments of the loop—at the ends and at the midpoint. The segments have a length that is 15% of the total loop length. We subjectively define the footpoints of the loop to be the locations where the intensity pattern can no longer be confidently identified. Klimchuk (2000) used a quantitative definition that tends to underestimate the true length. As we discuss later, the footpoints of the observed loop do not always correspond closely to the locations where the corresponding flux tube intersects the photosphere. The footpoint segments are labeled “start” and “end” for reasons related to the flux tube construction, and the “middle” segment is exactly halfway between them. The start footpoint always appears to the left in the figures. We compute average widths for the start ($`W_s`$), middle ($`W_m`$), and end ($`W_e`$) segments using the resolution-corrected measurements, but ignoring values below the resolution limit because their uncertainties are so large.
We are interested in differences between footpoints in addition to footpoint-midpoint differences. We therefore define four different expansion factors:
$$\mathrm{\Gamma }_{m/s}^{}=\frac{W_m}{W_s},$$
(4)
$$\mathrm{\Gamma }_{m/e}^{}=\frac{W_m}{W_e},$$
(5)
$$\mathrm{\Gamma }_{m/se}^{}=\frac{2W_m}{(W_s+W_e)},$$
(6)
and
$$\mathrm{\Gamma }_{e/s}^{}=\frac{W_e}{W_s},$$
(7)
where the asterisk is used to distinguish these definitions from slightly modified definitions introduced in Section 4. These four expansion factors were computed for each of the 20 loops, and Table 2 shows their means and standard deviations (not to be confused with the standard deviations of the intensity profiles used to determine the widths). On average, the loops are only about 15% wider at their middles than at their footpoints, and the start footpoints are statistically about as wide as the end footpoints. The standard deviations are not small, however. These results are entirely consistent with earlier findings on footpoint-to-midpoint expansion factors. Klimchuk (2000) reported a mean value of 1.30 for a sample of 43 loops observed by Yohkoh, and Watko & Klimchuk (2000) reported mean values of 0.99 and 1.13 for, respectively, non-flare and post-flare loops observed by TRACE. Klimchuk et al. (1992) found a mean expansion factor of 1.13 for 10 Yohkoh loops using uncorrected width measurements.
Finally, we test for any dependence of the expansion factor on the loop width. If the observation of nearly uniform width were an artifact of poor spatial resolution, then we would expect the expansion factor to be positively correlated with the width. In Figure 5 we plot $`\mathrm{\Gamma }_{m/s}^{}`$ versus the average width of the middle portion $`W_m`$. It can be clearly seen that there is no correlation, which we have verified using a nonparametric statistical analysis.
## 3 Modeled Magnetic Flux Tubes
### 3.1 Linear Force-free Extrapolation
We extrapolate the observed photospheric field into the corona using the method described in Alissandrakis (1981). It employs a Fast Fourier Transform (FFT) procedure to solve the linear force free-field equation (Equation 1) with $`\alpha `$ equal to a constant. The numerical code was developed by Démoulin et al. (1997) and has been used in a number of studies (e.g., Green et al. 2002). The computational volume is a 3D Cartesian box with the $`z=0`$ plane corresponding to the photosphere. Periodic boundary conditions are imposed on the side walls, and the field strength is required to decrease at the limit of large heights. This implies that each Fourier mode has an exponential decrease in the $`z`$ direction (with a scale height that depends on both the spatial wavelength of the mode and $`\alpha `$). The actual calculations are performed in two-dimensional Fourier space with 256$`\times `$256 horizontal modes. In order to save computer space, the results are kept on a 129$`\times `$129 nonuniform grid. The solution is discretized in 81 steps in the $`z`$-direction. Full-disk longitudinal magnetograms from SOHO/MDI (Scherrer et al. 1995) are used to specify the vertical component of the field at the lower boundary. The magnetograms have a spatial resolution of $``$ 1.44 Mm/pixel, but we interpolate onto a grid with a somewhat smaller spacing of 1 Mm in the finest part. To minimize the contribution of unknown transverse components to the observed line-of-sight field, we restrict our analysis to active regions that are close to disk center. We choose the horizontal dimensions of the box to be large enough that all of the active region is contained within it. Any flux imbalance is offset by not taking into account the Fourier mode (0,0) (this correspond to an added uniform weak field).
### 3.2 Image Coalignment
To study the magnetic properties of individual coronal loops, it is necessary to have an accurate coalignment between the magnetograms and TRACE images. According to the SolarSoft TRACE Analysis Guide the uncertainty of the TRACE pointing is 5-10 arcsec, though recent calibrations seem to have improved these numbers (Aschwanden 2005, private communications). Since we consider that this is not adequate for our purposes we use 171 Å images from the Extreme-ultraviolet Imaging Telescope (EIT) on SOHO (Delaboudiniere et al. 1995) as an intermediate step. The EIT images and MDI magnetograms are both full disk and can be accurately coaligned by forcing the solar limbs to agree. Repeated attempts to coalign a single EIT/magnetogram pair suggest an uncertainty of approximately 0.5-1.0 EIT pixel (0.9–1.9 Mm).
We then coalign the EIT and TRACE images by matching features that are visible in both. Visual inspection seems to work better than a purely quantitative cross-correlation approach. Many 171 Å features are quite stable, such as moss and the footpoints of large-scale magnetic structures (Berger et al. 1999, Martens et al. 2000), but other features evolve significantly over timescales of minutes. These changes can be identified and ignored in a visual comparison, but they have a significant influence on the cross correlation. Intensity differences resulting from differences in the TRACE and EIT bandpasses are also better treated by visual comparison.
Once the TRACE image is coaligned with the EIT image, it can be straightforwardly coaligned with the magnetogram by accounting for the small offset due to solar rotation during the time lag between the observations. The time lag is less than one hour, so there is no need to account for the distortions produced by differential rotation. We estimate an approximate 2 Mm uncertainty in the final TRACE/magnetogram coalignment.
### 3.3 Identifying Magnetic Axes of Loops
The first step in constructing a flux tube model of a loop is to identify the field line at the loop’s axis. The procedure is described in Green et al. (2002). Essentially, we compute many different linear force-free field models, each characterized by a unique $`\alpha `$, and compute many different field lines to find the one field line that best fits the TRACE loop as seen projected onto the plane-of-the-sky. The mean separation between the field line and loop axis is a quantitative measure of the fit. More precisely, for each defined point along the loop axis, we first find the closest point in the computed field line. This defines the local separation. Then we obtain the mean separation for all the loop axis points. This procedure permits to find the closest field line without the need to define the end points of the loop. For each model we trace many field lines from a square grid of starting positions that is centered on the better defined of the two loop footpoints (the “start” footpoint). We repeat the procedure with successively finer grids until we find the best fit for that particular model. We do this for many models covering a range of $`\alpha `$. The upper panel in Figure 6 shows an example of magnetic axis fitting. The green box corresponds to the initial sub-grid for the field line tracing, the blue crosses indicate the observed loop axis, and the red line is the best model field line for the axis. Figure 7 shows the mean separation of the best fit field line plotted as a function of $`\alpha `$ for one of our loops. The model with the smallest mean separation (the undimentionned $`\alpha `$ value is $`0.15`$ in this example) is used in the final analysis. We reject cases where the smallest mean separation is greater than 2 Mm. Note that the linear force-free field provides a much better fit than the potential field ($`\alpha =0`$).
As we have mentioned, the detectable footpoint of the TRACE loop is generally offset from the photospheric footpoint of the corresponding flux tube. We have performed hydrodynamic loop models which suggest that detectable TRACE emission should be present to within a short distance of the chromsphere. Some of the observed offsets are consistent with the expected thickness of the chromosphere (few thousand kilometers for an inclined flux tube), while others are much larger. In some cases the lower leg is simply obscured by bright background emission. In others it is too faint to be readily detected, apparently contradicting the hydrodynamic models. In still other cases it seems clear that the model field simply does not represent the loop accurately in the vicinity the footpoint. With some exceptions, the maximum footpoint offset that we allow at either end of the loop is 20 Mm. When the offset is greater, we choose a different field line even if the overall fit is not as good (though the mean separation must always be less then 2 Mm). In a few cases there is reason to believe that the flux tube is considerably longer than the visible loop, such as when the visible end falls in a region of incorrect polarity but points to a region of correct polarity. We make a subjective decision to keep these cases. In a small minority of cases both a short and a long field line give acceptable fits, and we keep both. In no instance, however, is the offset allowed to exceed 60 Mm. While we believe these selection criteria are very reasonable, we present separate results for the cases where the offset is less than 10 Mm. The results are similar.
Figures 1 and 2 lower panels show (in red) the best-fit field lines overlaid on TRACE images for a number of loops in two of the studied active regions. Loop 3 in Figure 1 and loop 1 in Figure 2 are shown also in panels (a) and (d), respectively, of Figures 39, and 10.
### 3.4 Constructing Flux Tubes
We remind the reader that the term “flux tube” refers explicitly to a magnetic flux tube based on an extrapolation model, and the term “loop” refers explicitly to an EUV intensity feature observed in a TRACE image. The terms are distinct and not interchangeable.
Flux tubes are constructed using the best fit field line as the tube axis. The shape of the tube cross section is unknown and must be assumed at some location along the axis. We choose to define the shape at the start footpoint. We consider four possible footpoint shapes, as indicated in Figure 8: a circle and three ellipses contained in the photospheric (x-y) plane. The major axes of the ellipses are oriented perpendicular to and at $`45\mathrm{deg}`$ angles to the projection of the tube axis in the x-y plane. Note that the footpoint shape will be different from the cross section whenever the tube axis is inclined to the vertical. For each of the footpoint shapes, we trace 24 field lines starting from points distributed systematically around the perimeter. These field lines define the shape and size of the cross section throughout the remainder of the tube, which can be highly variable. The lower panel in Figure 6 shows a flux tube constructed using a circular footpoint centered on the best-fit field line of the upper panel. We determine an “observed” width by finding the spread of the field line bundle when viewed in projection onto the plane-of-the-sky. The width is measured perpendicular to the tube axis.
Initially, we normalize the flux tubes by setting the width of start footpoint equal to the resolution-corrected width of the loop segment nearest that footpoint, $`W_s`$ (this segment is 15% of the total length of the loop, see Section 2.3). The ratios of the major to minor axes of the ellipses are chosen so that all three ellipses have the same area (though different from the circle) and therefore enclose approximately the same magnetic flux. Figure 9 shows how the width varies along the tube for the same loops shown in Figure 3. The broken thin lines represent the four different cross sections (circle and 3 ellipses). Also shown as asterisks between solid thin lines are the loop widths and their error bars from Figure 3. Note that the vertical scale in Figure 9 is greatly expanded compared to Figure 3. In general the flux tubes are much wider than the loops everywhere except the start footpoint, where the normalization forces them to be similar. The flux tubes also tend to be longer than the loops, especially on the “end” side, for the reasons we have discussed.
The thick solid lines in Figure 9 indicate the square root of the cross-sectional area, $`A^{1/2}`$, obtained from the on-axis field strength using conservation of magnetic flux: $`\mathrm{\Phi }=BA`$ (i.e., we plot $`B^{1/2}`$). This is strictly correct only for very thin flux tubes in which $`B`$ is constant over the cross section. We have normalized $`A^{1/2}`$ in the same way as the flux tubes. Roughly speaking, $`A^{1/2}`$ represents the average width that would be measured if the loop were observed over a complete $`360\mathrm{deg}`$ range of viewing angles. This is in contrast to the single viewing angle represented by the flux tube width curves in the figure. Large differences between the $`A^{1/2}`$ and flux tube width curves (e.g., panel c) indicate highly non-circular cross sections, because the widths of such cross sections are strongly dependent on the viewing angle. A good example of this is shown in Figure 11. The top panel shows the flux tube as viewed from above. This is the same flux tube that gives a reasonably good fit to the TRACE loop in panel d of Figure 9 (dot-dash curve). The bottom panel shows the same flux tube as viewed from the side. The difference is dramatic.
To better compare the flux tubes and loops, we consider a second normalization that forces the average width of the tube and loop to be similar over the entire region of overlap, i.e., over the entire loop except in the rare instances where the loop extends beyond the flux tube (e.g., panel c). To obtain the new normalization, we first determine the factor by which the original tube must be narrowed in order for its average width to equal the average width of the loop. We then shrink the cross section of the start footpoint by this factor and trace a new bundle of field lines. Since this is a new flux tube (and not simply a reduced-width version of the original), its width is slightly different from the width of the original tube reduced by the factor. The width curves of the tubes obtained with the second normalization are shown in Figure 10 using the same format as Figure 9. The agreement with the loop widths is improved, but still quite poor in many cases. In some instances there is at least one flux tube width curve that falls mostly within the loop width error bars (e.g., panels b and d). In other instances there is gross deviation for all of the curves (e.g., panels a and c). Even in the cases with reasonable agreement, there is a tendency for the flux tube widths, but not the loop widths, to vary systematically along the loop. In panel b, for example, the flux tube widths are systematically higher on the right side than on the left side, and in panel d they are systematically higher in the middle. The loop widths do not exhibit the same trends. These differences will be more apparent when we compare expansion factors for the tubes and loops in the next section.
Before proceeding, we examine the possibility that the model flux tubes we have identified with loops do not accurately represent the actual flux tubes on the Sun. In most cases the best-fit model field line used for the tube axis coincides quite well with the observed loop, but the match is never perfect. In a few cases the match seems qualitatively rather poor even though the fit criteria of Section 3.3 are satisfied. Differences arise for two primary reasons: the uncertainty in the MDI-TRACE coalignment and the assumption of a linear force-free field. The force-free approximation is probably quite good, so $`\alpha `$ will be constant on each field line, but it is unreasonable to expect $`\alpha `$ to be the same on all field lines. In more realistic nonlinear force-free fields, $`\alpha `$ varies across the active region. Even if the $`\alpha `$ of our model is the actual $`\alpha `$ of the loop axis field line, the shape of the field line will depend on the distribution of $`\alpha `$ elsewhere in the active region. We would not expect the flux tube and loop to coincide precisely. To evaluate the importance of these effects, we investigated one representative case in considerable detail. We repeated our analysis multiple times by introducing MDI-TRACE offsets of 2 Mm in all four compass directions, and by constructing flux tubes using $`\alpha `$ values that differ by 20% from the best-fit $`\alpha `$. The resulting changes are of the order of 10% for $`A^{1/2}`$ and no more than 20% for the width. We therefore believe, but cannot prove definitively, that the sizable discrepancies we have found between flux tubes and loops are real.
## 4 Comparing Flux Tube and Loop Expansion Factors
To quantify the systematic differences between flux tubes and loops, we compute expansion factors using precisely the same segments from both structures. We modify the segment definitions of Section 2.3 to take into account the flux tubes as much as possible, which we consider to be the more fundamental structures. As we have discussed, parts of many flux tubes are not easily identifiable as loops because the loop emission is weak or the background emission is strong. Panel (a) in Figures 9 and 10 is an example of the problems that arise when using the earlier definitions. The end segment of the loop is closer to the midpoint of the flux tube than it is to the right footpoint. It is more properly classified as a middle segment than an end segment. As before, all of the segments in our new definitions have a length that is 15$`\%`$ of the loop length. Also as before, the start segment is the first 15$`\%`$ of the loop—the part nearest the footpoint where we define the cross section and begin the field line traces. There is usually good correspondence between the loop and flux tube on this end. If the loop extends beyond the midpoint of the flux tube, the middle segment is now defined to be the 15$`\%`$ segment centered on the tube midpoint, rather than the loop midpoint, and the end segment is the last 15$`\%`$ of the loop. If the loop does not reach the midpoint of the flux tube, the last 15$`\%`$ of the loop is assigned to the middle segment, and there is no end segment.
We compute $`\mathrm{\Gamma }_{m/s}`$, $`\mathrm{\Gamma }_{m/e}`$, and $`\mathrm{\Gamma }_{e/s}`$ expansion factors from the average widths in these segments in the same way as before. The asterisk has been removed to indicate that the segment definitions are different from those in Section 2.3. We use flux tube widths based on the second normalization. Because the segments are sometimes far removed from the footpoints or midpoint of the flux tube, we reject cases that do not satisfy certain constraints. The constraints are based on the separation between the centers of the segments as measured along the flux tube axis. $`\mathrm{\Gamma }_{m/s}`$ and $`\mathrm{\Gamma }_{m/e}`$ are included only if the separation between the middle segment and the start or end segment, respectively, is greater than 25$`\%`$ of the flux tube length. $`\mathrm{\Gamma }_{e/s}`$ is included only if the separation between the start and middle segments and the separation between the middle and end segments are both be greater than 15$`\%`$ of the flux tube length.
Each loop is associated with four flux tubes, one for each cross section (circle and three ellipses). A small subset of loops have eight flux tubes because two different best-fit field lines are acceptable. For each loop, we average together the flux tube expansion factors of each type to obtain composite $`\mathrm{\Gamma }_{m/s}^{tube}`$, $`\mathrm{\Gamma }_{m/e}^{tube}`$, and $`\mathrm{\Gamma }_{e/s}^{tube}`$. We also compute $`\mathrm{\Gamma }_{m/s}^{A^{1/2}}`$, $`\mathrm{\Gamma }_{m/e}^{A^{1/2}}`$, and $`\mathrm{\Gamma }_{e/s}^{A^{1/2}}`$ using the square-root of the cross sectional area (we actually use $`B^{1/2}`$). Finally, we have $`\mathrm{\Gamma }_{m/s}^{loop}`$, $`\mathrm{\Gamma }_{m/e}^{loop}`$, and $`\mathrm{\Gamma }_{e/s}^{loop}`$ from the resolution-corrected TRACE measurements. We emphasize that the corresponding flux tube and loop expansion factors are computed using the same intervals of $`s`$ (see Figure 10), so that the comparison is meaningful.
It is appropriate to average the results for the different cross sections because this is a statistical study of many loops. If we were investigating a single loop and found that a particular cross section is able to reproduce the loop much better than the others, then we could be plausibly conclude that it is the actual cross section. It would not be unreasonable to suggest that coronal heating selects that particular bundle of field lines to be illuminated. However, it is beyond common sense to suggest that coronal heating always picks the right bundle of field lines for each of a large collection of loops. That would imply that there is something special about the position of the TRACE spacecraft with respect to the Sun. (This is even true of the circular footpoint, since the circle becomes distorted along the loop and the symmetry is broken.) We therefore choose to average the expansion factors over the four assumed cross sections. We note from Figures 9 and 10 that the expansion factors are not qualitatively different for the different footpoint cross sections.
Figure 12 shows scatter plots of $`\mathrm{\Gamma }_{m/s}^{tube}`$ versus $`\mathrm{\Gamma }_{m/s}^{loop}`$, $`\mathrm{\Gamma }_{m/e}^{tube}`$ versus $`\mathrm{\Gamma }_{m/e}^{loop}`$, and $`\mathrm{\Gamma }_{e/s}^{tube}`$ versus $`\mathrm{\Gamma }_{e/s}^{loop}`$. Asterisks correspond to cases in which the loop footpoint and the flux tube footpoint are separated by less than 10 Mm (pluses correspond to the rest of the cases, i.e., footpoint separation larger than 10 Mm). It can be seen that restricting the distance between loop and flux tube footpoints (see Section 3.3) does not affect the qualitative properties of the scatter distribution. Table 3 gives the mean and standard deviation statistics on the different expansion factors. The figure and table reveal a number of interesting properties. First consider the behavior of the loops. If we imagine vertical lines at $`\mathrm{\Gamma }^{loop}=1.0`$ in Figure 12, we see that the points are divided roughly in half in all three panels. Just as many loops exhibit footpoint-to-midpoint constriction as exhibit footpoint-to-midpoint expansion. Neither the constriction nor the expansion is very large (typically less than 50%). Half of the loops are wider at their left footpoint than at their right footpoint, and vice versa. Not surprisingly, the mean values of the three expansion factors are only slightly larger than 1.0 in Table 3. The results are similar to those of Table 2, based on the original loop segment definitions, and are fully consistent with previously published results.
Now consider the behavior of the flux tubes. Imagine horizontal lines at $`\mathrm{\Gamma }^{tube}=1.0`$ in Figure 12. All but two of the points in the upper panel lie above the line, indicating an overwhelming tendency for flux tubes to expand from the start footpoint to the midpoint. $`<\mathrm{\Gamma }_{m/s}^{tube}>=1.79`$, so the tubes are nearly twice as wide at their midpoints as at their start footpoints, on average. This is in sharp contrast to the behavior of loops. Interestingly, the points in middle panel are nearly evenly divided above and below the imaginary line, so there is no significant preference for expansion or constriction with respect to the end footpoint. Twice as many points in lower panel lie above the imaginary line as below, indicating a strong tendency for the end footpoint to be wider than the start footpoint. The difference can be very large, with $`<\mathrm{\Gamma }_{e/s}^{tube}>=2.62`$. The extreme asymmetry of flux tubes contrasts sharply with the symmetric nature of loops.
The flux tube asymmetry is much less pronounced in cross-sectional area. $`<\mathrm{\Gamma }_{e/s}^{A^{1/2}}>=1.24`$, not much greater than unity, and the standard deviation is only 0.50. This is a clear indication that the shape of the cross section at the end footpoint is much more irregular and much less compact than the circle and ellipses assumed at the start footpoint. Field strengths are comparable at the two ends, since they scale directly with the cross-sectional area.
Statistically, loops and flux tubes have very different expansion properties. These same differences show up clearly in a case-by-case analysis. We compute the ratios of the flux tube and loop expansion factors for each case separately: $`R_{m/s}=\mathrm{\Gamma }_{m/s}^{tube}/\mathrm{\Gamma }_{m/s}^{loop}`$, $`R_{m/e}=\mathrm{\Gamma }_{m/e}^{tube}/\mathrm{\Gamma }_{m/e}^{loop}`$, and $`R_{e/s}=\mathrm{\Gamma }_{e/s}^{tube}/\mathrm{\Gamma }_{e/s}^{loop}`$. The last column in Table 3 gives the means and standard deviations of these ratios. They confirm what we found above: (1) flux tubes expand twice as much as loops from start footpoint to midpoint; (2) flux tubes and loops expand or constrict to an equal degree from end footpoint to midpoint; and (3) flux tubes are far more asymmetric than loops. It may seem odd that the $`<R_{e/s}>`$ is noticably larger than the $`<R_{m/s}>`$ at the same time that the $`<R_{m/e}>`$ is essentially unity. This is not a real inconsistency and arises because the samples are not the same for the different ratios due to our selection criteria. There are fewer $`\mathrm{\Gamma }_{e/s}`$ cases than $`\mathrm{\Gamma }_{m/s}`$ cases, and still fewer $`\mathrm{\Gamma }_{m/e}`$ cases. The solid lines in Figure 12 have slopes equal to the mean values of the ratios, while the dotted lines have slopes equal to the mean values plus and minus the standard deviations. The dashed lines have a slope of 1. It is important to note that a large majority of the points in upper panel lie well above the dashed line, and only two points in lower panel lie significantly below the dashed line. Therefore, the trends we have identified are actual trends obeyed by a majority of loops and not simply artifacts of a few unusual (“outlier”) cases.
Table 4 gives the statistical results, in the same format as Table 3, for the cases in which the loop endpoint and flux tube footpoint are separated by less than 10 Mm (asterisks in Figure 12). Possible reasons for the separation are discussed in Section 3.3. We see that the results are very similar, indicating that our full sample is not contaminated by bad cases.
## 5 Discussion
Our study demonstrates that linear force-free extrapolation models based on MDI magnetograms are inconsistent with TRACE loop observations. The loops have a nearly uniform width, whereas the corresponding flux tubes generally do not. Most of the flux tubes expand appreciably with height from either one or both footpoints. Thus, it is not the case that loops correspond to a subset of uniform-width flux tubes that may exist in magnetic field configurations of the type we have modeled. This is an important result that had not been ruled out by previous studies.
Another very important result is that the model flux tubes are much less symmetric than observed loops. Most flux tubes appear considerably wider at one footpoint than the other when viewed from the TRACE line-of-sight. The cross-sectional areas of the two footpoints are very similar, on the other hand. This indicates that the asymmetry is one of shape rather than field strength. The simple and compact cross sections (circle or ellipse) that we have assumed at the start footpoints map to much more irregular and distended cross sections at the end footpoints. It is especially interesting that this result is independent of the side on which the field line traces are begun. When we begin with circles and ellipses at the location of the original end footpoint, we find that they map to irregular and distended cross sections at the location of the original start footpoint. The flux tubes are asymmetric in both cases, but in an opposite sense.
We can understand this result in terms of two effects. First, there is a general tendency for field strength to decrease with height and for cross-sectional area to increase with height to conserve flux. Second, there is a tendency for field lines to deviate from each other as they are traced and for the shape of the cross section to become progressively distorted. Such distortion is typical in the mapping of field lines in traditional magnetic field models and can be severe at quasi-separatrix layers (QSLs; see e.g. Titov et al. 2002). An important point is that the distortion is independent of the direction in which the field lines are traced. Therefore, when we start from a compact footpoint and trace field lines up to the tube apex, both the tendency for field strength to decrease and the tendency for the cross section to distort cause the width of the tube to increase. As we continue the trace down to the end footpoint, the two effects are competing. Continued distortion causes the width to increase, but increasing field strength causes the width to decrease. Which effect wins out varies from tube to tube, and $`\mathrm{\Gamma }_{m/e}^{tube}`$ values greater than 1 and less than 1 are equally common (see Figure 12).
We note that our decision to start the field line traces at a footpoint is merely a technical convenience. There is no physical motivation. Which field lines are illuminated to form an observed loop is determined entirely by coronal heating. If coronal heating were known to occur at the top of a loop, it would be appropriate to trace field lines starting at the top, using the cross section defined by the heating. It is interesting to conjecture that flux tubes defined by a circular cross section at the top would tend to be more symmetric than flux tubes defined by a circular cross section at one of the footpoints. In the first case, the cross sectional distortion will tend to increase while tracing down both legs. In the latter case, it will increase going up one leg and increase further going down the other.
These fundamental differences between the loops and flux tubes imply that either the plasma structures revealed by TRACE do not follow magnetic field lines or the magnetic field models we have used do not accurately represent the detailed properties of the solar magnetic field. The first possibility seems highly implausible. Thermal conduction and plasma motions are extremely efficient at transporting energy and matter along the magnetic field, but cross-field transport is greatly inhibited (e.g., Litwin & Rosner 1993). We conclude that the models are inadequate. Either the linear force-free approximation is poor or the magnetogram boundary conditions are lacking, or both.
Measurements of the plasma pressure suggest that the field is close to force free throughout most of the corona of active regions. Even if it were not, plasma pressure effects could not explain the discrepancies we have found. We believe that the fundamental source of the discrepancies is the assumption of a linear force-free field. The linear assumption is reasonable for modeling the large-scale structure of the field that is determined by large-scale currents (e.g., the shapes of loop axes, as discussed at the end of Section 3.4), but it cannot treat the small-scale currents that are critical for loop cross sections. Even nonlinear force-free field models based on today’s modest resolution magnetograms are inadequate for this purpose. The possibility considered by Klimchuk et al. (2000) that loops are locally twisted magnetic flux tubes would imply currents with a transverse scale smaller than a loop diameter. The nonlinear force-free models that they constructed suggest that twist cannot produce the degree of thickness uniformity observed in real loops; however, the idea cannot be ruled out entirely because the models have a maximum twist of $`2\pi `$. Greater twist is likely to produce greater uniformity, but it is not known how much greater the twist can be before realistic curved loops become kink unstable (see Gerrard, Hood, & Brown 2004).
The twisted loops considered by Klimchuk et al. have a well organized and relatively simple internal structure. Real loops are likely to be much more complicated. High resolution observations of the photospheric magnetic field reveal that is clumped into small and intense kiloGauss flux tubes (see Solanki 1993). The magnetic flux contained in each elemental tube is so small that a single TRACE loop must contain tens to hundreds of them (e.g., Priest, Heyvaerts, & Title 2002). The footpoints of these tubes are randomly displaced about the solar surface by the changing convective flow pattern (e.g., Schrijver et al. 1997). We therefore expect the field within a loop to be highly tangled, with the elemental strands wrapping around each other in complicated ways.
The basic picture of tangled field was first proposed by Parker (1988). He suggested that the energy contained in the magnetic stresses associated with the tangling would be liberated in the form of nanoflares. From energy balance considerations we can conclude that the nanoflares must occur when the angle between misaligned elemental flux tubes is approximately $`50\mathrm{deg}`$. (Parker stated this result in terms of a $`25\mathrm{deg}`$ tilt from vertical at the base of the corona). Recently, Dahlburg, Klimchuk, & Antiochos (2003, 2005) demonstrated that a mechanism called the secondary instability “switches on” when the misalignment angle reaches this critical value. They showed that energy is released impulsively and is adequate to heat the corona. This agrees nicely with studies showing that the density and temperature properties of coronal loops, especially TRACE loops, are best explained if loops are modeled as collections of unresolved impulsively-heated strands (Cargill 1994; Warren, Winebarger, & Mariska 2003; Klimchuk 2004).
The concept of internal tangling within loops may also explain our result that loops are much more symmetric than linear force-free field models would predict. To see how this might be, imagine a field that is initially very simple, so that flux tubes have compact (e.g., circular) cross sections at both ends. Systematic motions can rearrange the photospheric footpoints, even while maintaining the same overall flux distribution, so that the flux tubes become highly asymmetric. Suppose, however, that small-scale random displacements are superposed on the systematic flow pattern. The footpoints of any two elemental flux strands that are initially close together will then separate according to a random walk. As they do, they will become tangled with other strands. The tangling can only proceed so far before the secondary instability causes adjacent strands to reconnect, thereby decreasing the level of stress. For a random walk step size of 1 Mm corresponding to a granulation cell diameter, a loop length of 100 Mm, and a critical angle of $`50\mathrm{deg}`$, the footpoint separation would not be expected to exceed 5 Mm. This would seem to preclude the possibility of flux tubes that are highly asymmetric. Two elemental strands that are close together at one end cannot be greatly separated at the other end. We plan to investigate this interesting conclusion more thoroughly in future work. It is interesting to speculate that the small-scale structure in the measured loop width (e.g., Figure 3) may be due to irregular trajectories of the strands in the tube bundle. We are reluctant to make this claim just yet, since we cannot rule out the possibility that the “bumps” in the width curves are caused by variable errors in the background subtraction. The width variations along individual loops have a standard deviation that is 26% of the mean width, on average. This includes systematic variations as well as small-scale fluctuations. Aschwanden and Nightingale (2005, Figure 9) report a standard deviation of 23%, but their value is artificially small because they do not correct for the finite spatial resolution of the observations, which is a non-linear correction.
We end on something of a cautionary note. It was long ago suggested that loops appear to be uniformly wide simply because they are unresolved. Indeed, if a loop is everywhere narrower than a resolution element (combined point spread function and detector pixel), then its apparent width will be nearly constant even if it has a very large expansion factor. We have devoted enormous time and energy to addressing this possibility. Our careful analysis based on the best available information on the point spread functions of TRACE and Yohkoh/SXT indicates that both instruments are able to resolve the envelope of emission that we identify as a loop. Of course there could be unresolved internal structure. There are abundant examples of features in both data sets that are as small as the point spread functions measured before launch, indicating that the instruments are performing as expected. Furthermore, as discussed in Section 2.3, we find no evidence for a correlation between expansion factor and width as would be expected if the loops were poorly resolved.
It is nonetheless somewhat unnerving that the both TRACE and Yohkoh loops tend to be a few resolution elements wide, despite the significant difference in resolution. This could be because the $`1`$ MK loops detected by TRACE are physically quite different from the $`28`$ MK loops detected by Yohkoh. There may also be a selection bias at work. When choosing loops for detailed analysis, one is drawn to examples that stand out from the background and appear monolithic (i.e., that are not obviously multiple loops). This naturally favors loops that are a few resolution elements wide. Thinner loops have a reduced brightness contrast relative to the background, especially if they are more narrow than a pixel, because then the pixel brightness is an average of the intrinsic brightness of the loop and the background. Many loops may actually be small collections of thinner loops, say, 2-5 thinner loops, each thin loop being itself comprised of many kiloGauss flux strands (see also Aschwanden, 2005). The collection will appear monolithic if the loops are closely spaced, but not if the gaps are comparable to a resolution element. We therefore suggest a picture in which the Yohkoh loops of our earlier studies are actually small collections of unresolved TRACE-size loops. Winebarger & Warren (2005) have shown that at least some hot Yohkoh loops contain several thinner and cooler TRACE loops within their envelope. This may not be common (e.g., Schmieder et al., 2004), but it suggests that hot plasma may also be structured in thin loops. Whether this is actually the case must await high-resolution, high-temperature observations such as may be available in the mid term from NRL’s VERIS rocket experiment and in the long term from the Reconnection and Microscale (RAM) and Solar Orbiter missions. The AIA instrument on the Solar Dynamics Observatory may also provide useful information on this question, though it remains to be seen whether the temperature discrimination will be adequate.
We wish to thank Harry Warren and Amy Winebarger for useful discussions on TRACE data processing and the instrument pointing. We acknowledge the TRACE and SOHO teams. We also thank the referee, Markus Aschwanden, for his useful suggestions and for encouraging the comparison of width measurement techniques that is discussed in the Appendix. This work was supported by NASA and the Office of Naval Research.
## Appendix: Comparison between different methods of loop width determination
As discussed in Section 2.2 the computation of the standard deviation of the intensity profile ($`\sigma `$) implies a weighting of the intensity with the square of the position along the profile (see Equation 2). It can be argued that this weighting may artificially amplify the effect of a residual (unsubtracted) background at the “tails” of the profile. To explore this possibility, here we compare $`\sigma `$ with two other measures of loop thickness: the Full Width at Half Maximum (FWHM) and the Equivalent Width ($`W_{eq}`$), defined as
$$W_{eq}=\frac{I_i}{I_{max}},$$
where $`I_{max}`$ is the maximum intensity along the profile. These two alternative methods would seem to be less susceptible to the effects of residual background. As we have indicated in Section 2, the width (diameter) of a uniformly filled circular cross-section is is 4 times $`\sigma `$. It can be easily demonstrated that the width is also 1.41 times the FWHM and 1.27 times $`W_{eq}`$.
We perform simultaneous measurements of $`\sigma `$, FWHM, and $`W_{eq}`$ for the four loops shown in Figures 3, 9, and 10. We then convert to width using the conversion factors above. Since it is not relevant for the present comparison, we do not correct the results for finite resolution (see Section 2.2). As examples, in Figure 13 we plot the inferred widths for the loops shown in panels c and d of the figures. Clearly, the three methods give similar results, and the fluctuations along the loops are of the same order. The computed average fluctuations of $`\sigma `$ for the four analyzed cases are of the order of 15%. In comparison, the average fluctuations of the resolution corrected values (asterisks in Figures 3, 9, and 10) are of the order of 26%. The reason for this is the non-linearity of the resolution correction curve (see Figure 4) for loops near the resolution limit.
It is worth noting that since the results shown in Figure 13 are not corrected for finite resolution they correspond to the results plotted with triangles in Figure 3 (panels c and d). It can be noticed that the $`\sigma `$ measurements in Figure 13 (continuous lines) do not coincide exactly with those in Figure 3 (triangles). Very small differences occur because the two figures were made using different measurements of the same loops, and since our procedure involves two subjective steps (identifying the loop axis and tracing the loop boundary). The similarity of the measurements indicates that the subjectivity is not critical.
In Figure 14 we plot $`\sigma `$ versus FWHM and $`W_{eq}`$ (upper and lower panels, respectively) for the four loops. It can be seen that there is a strong statistical correlation between $`\sigma `$ and the other two width measures. We perform least square fits of the scatter data, and we find a slope of 0.31 for $`\sigma `$ as a function of FWHM and 0.35 for $`\sigma `$ as a function of $`W_{eq}`$. In comparison, the proportionality factors based on the approximation of a uniformly filled circular-cross section are: $`\sigma =0.35`$ FWHM and $`\sigma =0.32W_{eq}`$. The $`\chi ^2`$ test gives a correlation probability of 1.
The above results suggest that a circular cross-section loop with uniform density (except perhaps on a sub-resolution scale) is a reasonable approximation, and confirm the suitability of the standard deviation of the intensity profile for loop width determinations. |
warning/0507/physics0507049.html | ar5iv | text | # Superresolution and Corrections to the Diffusion Approximation in Optical Tomography
## Abstract
We demonstrate that the spatial resolution of images in optical tomography is not limited to the fundamental length scale of one transport mean free path. This result is facilitated by the introduction of novel corrections to the standard integral equations of scattering theory within the diffusion approximation to the radiative transport equation.
There has been considerable recent interest in the development of optical methods for tomographic imaging Gibson\_2005 . The physical problem that is considered is to recover the optical properties of the interior of an inhomogeneous medium from measurements taken on its surface. The starting point for the mathematical formulation of this inverse scattering problem (ISP) is a model for the propagation of light, typically taken to be the diffusion approximation (DA) to the radiative transport equation (RTE). The DA is valid when the energy density of the optical field varies slowly on the scale of the transport mean free path $`\mathrm{}^{}`$. The DA breaks down in optically thin layers, near boundary surfaces, or near the source. One or more of these conditions are encountered in biomedical applications such as imaging of small animals Graves\_2003 or of functional activity in the brain.
Within the accuracy of the DA, reconstruction algorithms based on both numerical Arridge\_1999 and analytic solutions Schotland\_1997 ; Markel\_2002 ; Markel\_2004\_1 to the ISP have been described. Regardless of the method of inversion, the spatial resolution of reconstructed images is expected to be limited to $`\mathrm{}^{}`$. This expectation is due to the intertwined effects of the ill-posedness of the ISP Markel\_2002 and intrinsic inaccuracies of the DA Yoo\_1990 . In this Letter, we introduce novel corrections to the integral equations of scattering theory within the DA. Using this result, we report the reconstruction of superresolved images whose spatial resolution is less than $`\mathrm{}^{}`$.
We begin by considering the propagation of multiply-scattered light in an inhomogenous medium characterized by an absorption coefficient $`\mu _a(𝐫)`$. In what follows, we will neglect the contribution of ballistic photons and consider only diffuse photons whose specific intensity $`I(𝐫,\widehat{𝐬})`$ at the point $`𝐫`$ in the direction $`\widehat{𝐬}`$ is taken to obey the time-independent RTE
$`\widehat{𝐬}I(𝐫,\widehat{𝐬})+(\mu _a+\mu _s)I(𝐫,\widehat{𝐬})\mu _s{\displaystyle d^2s^{}A(\widehat{𝐬},\widehat{𝐬}^{})I(𝐫,\widehat{𝐬}^{})}=S(𝐫,\widehat{𝐬}),`$ (1)
where $`\mu _s`$ is the scattering coefficient, $`A(\widehat{𝐬},\widehat{𝐬}^{})`$ is the scattering kernel, and $`S(𝐫,\widehat{𝐬})`$ is the source. The change in specific intensity due to spatial fluctuations in $`\mu _a(𝐫)`$ can be obtained from the integral equation
$$\varphi (𝐫_1,\widehat{𝐬}_1;𝐫_2,\widehat{𝐬}_2)=d^3rd^2sG(𝐫_1,\widehat{𝐬}_1;𝐫,\widehat{𝐬})G(𝐫,\widehat{𝐬};𝐫_2,\widehat{𝐬}_2)\delta \mu _a(𝐫).$$
(2)
Here the data function $`\varphi (𝐫_1,\widehat{𝐬}_1;𝐫_2,\widehat{𝐬}_2)`$ is proportional, to lowest order in $`\delta \mu _a`$, to the change in specific intensity relative to a reference medium with absorption $`\mu _a^0`$, $`G`$ is the Green’s function for (1) with $`\mu _a=\mu _a^0`$, $`\delta \mu _a(𝐫)=\mu _a(𝐫)\mu _a^0`$, $`𝐫_1,\widehat{𝐬}_1`$ and $`𝐫_2,\widehat{𝐬}_2`$ denote the position and direction of a unidirectional point source and detector, respectively.
We now show that the integral equation (2) may be used to obtain corrections to the usual formulation of scattering theory within the DA. To proceed, we note that, following Ref. Markel\_2004\_1 , the Green’s function $`G(𝐫,\widehat{𝐬};𝐫^{},\widehat{𝐬}^{})`$ may be expanded in angular harmonics of $`\widehat{𝐬}`$ and $`\widehat{𝐬}^{}`$:
$$G(𝐫,\widehat{𝐬};𝐫^{},\widehat{𝐬}^{})=\frac{c}{4\pi }\left(1+\mathrm{}^{}\widehat{𝐬}_𝐫\right)\left(1\mathrm{}^{}\widehat{𝐬}^{}_𝐫^{}\right)G(𝐫,𝐫^{}),$$
(3)
where $`\mathrm{}^{}=1/[\mu _a^0+\mu _s^{}]`$ with $`\mu _s^{}=(1g)\mu _s`$, $`g`$ being the anisotropy of the scattering kernel $`A`$. The Green’s function $`G(𝐫,𝐫^{})`$ satisfies the diffusion equation $`\left(D_0^2+\alpha _0\right)G(𝐫,𝐫^{})=\delta (𝐫𝐫^{})`$, where the diffusion coefficient $`D_0=1/3c\mathrm{}^{}`$ and $`\alpha _0=c\mu _a^0`$. In addition, the Green’s function must satisfy boundary conditions on the surface of the medium (or at infinity in the case of free boundaries). In general we will consider boundary conditions of the form $`G(𝐫,𝐫^{})+\mathrm{}\widehat{𝐧}G(𝐫,𝐫^{})=0`$, where $`\widehat{𝐧}`$ is the outward unit normal to the surface bounding the medium and $`\mathrm{}`$ is the extrapolation distance. Making use of (3) and performing the angular integration over $`\widehat{𝐬}`$ in (2) we obtain
$`\varphi (𝐫_1,\widehat{𝐬}_1;𝐫_2,\widehat{𝐬}_2)={\displaystyle \frac{c}{4\pi }}\mathrm{\Delta }_1\mathrm{\Delta }_2{\displaystyle d^3r\left[G(𝐫_1,𝐫)G(𝐫,𝐫_2)\frac{\mathrm{}_{}^{}{}_{}{}^{2}}{3}_𝐫G(𝐫_1,𝐫)_𝐫G(𝐫,𝐫_2)\right]\delta \alpha (𝐫)},`$ (4)
where the differential operators $`\mathrm{\Delta }_k=1(1)^k\mathrm{}^{}\widehat{𝐬}_k_{𝐫_k}`$ with $`k=1,2`$ and $`\delta \alpha =c\delta \mu _a`$. Note that if the source and detector are oriented in the inward and outward normal directions, respectively, then (4) becomes
$`\varphi (𝐫_1,\widehat{𝐧}(𝐫_1);𝐫_2,\widehat{𝐧}(𝐫_2))={\displaystyle \frac{c}{4\pi }}(1+{\displaystyle \frac{\mathrm{}^{}}{\mathrm{}}})^2{\displaystyle }d^3r[G(𝐫_1,𝐫)G(𝐫,𝐫_2)`$
$`{\displaystyle \frac{\mathrm{}_{}^{}{}_{}{}^{2}}{3}}_𝐫G(𝐫_1,𝐫)_𝐫G(𝐫,𝐫_2)]\delta \alpha (𝐫),`$ (5)
where we have used the boundary conditions on $`G`$ to evaluate the action of the $`\mathrm{\Delta }_k`$ operators. Eq. (Superresolution and Corrections to the Diffusion Approximation in Optical Tomography) is the main theoretical result of this Letter. It may be viewed as providing corrections to the DA since the first term on the right hand side of (Superresolution and Corrections to the Diffusion Approximation in Optical Tomography) corresponds to the standard DA in an inhomogeneous absorbing medium. We note that the second term may be interpreted as defining an effective diffusion coefficient $`D(𝐫)=D_0(\mathrm{}_{}^{}{}_{}{}^{2}/3)\delta \alpha (𝐫)`$ since the expression $`_𝐫G(𝐫_1,𝐫)_𝐫G(𝐫,𝐫_2)`$ defines the diffusion kernel in a medium with an inhomogeneous diffusion coefficient Arridge\_1999 .
For the remainder of this paper we will work in the planar measurement geometry, often encountered in small-animal imaging. In this case, (4) becomes
$$\varphi (𝒓ho_1,𝒓ho_2)=d^3rK(𝝆_1,𝝆_2;𝐫)\delta \alpha (𝐫),$$
(6)
where $`𝝆_1`$ denotes the transverse coordinates of a point source in the plane $`z=0`$, $`𝝆_2`$ denotes the transverse coordinates of a point detector in the plane $`z=L`$, and the dependence of $`\varphi `$ on $`\widehat{𝐬}_1`$ and $`\widehat{𝐬}_2`$ is not made explicit. Evidently, from considerations of invariance of the kernel $`K(𝝆_1,𝝆_2;𝐫)`$ under translations of its transverse arguments, it can be seen that $`K`$ may be expressed as the Fourier integral
$`K(𝝆_1,𝝆_2;𝐫)={\displaystyle \frac{1}{(2\pi )^4}}{\displaystyle d^2q_1d^2q_2\kappa (𝐪_1,𝐪_2;z)\mathrm{exp}\left[i(𝐪_1𝐪_2)𝝆i(𝐪_1𝝆_1𝐪_2𝝆_2)\right]},`$
where $`𝐫=(𝝆,z)`$. The function $`\kappa `$ may be obtained from the plane-wave expansion of the diffusion Green’s function obeying appropriate boundary conditions. In the case of free boundaries, it is readily seen that $`\kappa `$ is given by the expression
$`\kappa (𝐪_1,𝐪_2;z)={\displaystyle \frac{c}{16\pi D_0^2Q(𝐪_1)Q(𝐪_2)}}\left[1+{\displaystyle \frac{\mathrm{}_{}^{}{}_{}{}^{2}}{3}}\left(Q(𝐪_1)Q(𝐪_2)𝐪_1𝐪_2\right)\right]`$
$`\times \left[1+\mathrm{}^{}\left(Q(𝐪_1)+Q(𝐪_2)\right)+\mathrm{}_{}^{}{}_{}{}^{2}Q(𝐪_1)Q(𝐪_2)\right]\mathrm{exp}\left[Q(𝐪_1)|z|Q(𝐪_2)|zL|\right],`$ (7)
where $`Q(𝐪)=(q^2+\alpha _0/D_0)^{1/2}`$ and we have assumed that $`\widehat{𝐬}_1=\widehat{𝐬}_2=\widehat{𝐳}`$.
Inversion of the integral equation (6) may be carried out by analytic methods. These methods have been shown to be computationally efficient and may be applied to data sets consisting of a very large number of measurements Markel\_2002 ; Markel\_2004\_1 . The approach taken is to construct the singular value decomposition of the linear operator $`K`$ in the proper Hilbert space setting and then use this result to obtain the pseudoinverse solution to (6). In this manner, it is possible to account for the effects of sampling and thereby obtain the best (in the sense of minimizing the appropriate $`L^2`$ norm) bandlimited approximation to $`\delta \alpha `$. Here we use this approach to simulate the reconstruction of a point absorber located at a point $`𝐫_0`$ between the measurement planes with $`\delta \alpha (𝐫)=A\delta (𝐫𝐫_0)`$ for constant $`A`$. In this situation it is possible to calculate the data function $`\varphi `$ within radiative transport theory, thus avoiding “inverse crime.” To proceed, we require the Green’s function $`G(𝐫,\widehat{𝐬};𝐫^{},\widehat{𝐬}^{})`$ for the RTE in a homogeneous infinite medium which we obtain as described in Ref. Markel\_2004\_2 . Note that in this case, the angular integration over $`\widehat{𝐬}`$ in (2) may be carried out analytically.
The effects of corrections to the DA were studied in numerical simulations following the methods of Ref. Markel\_2002 . The simulations were performed for a medium with optical properties similar to breast tissue in the near infrared Peters\_1990 . The background absorption and reduced scattering coefficients were given by $`\mu _a^0=0.03\mathrm{cm}^1`$ and $`\mu _s^{}=10\mathrm{cm}^1`$. The scattering kernel was taken to be of Henyey-Greenstein type with $`A(\widehat{𝐬},\widehat{𝐬}^{})=_{\mathrm{}=0}^{\mathrm{}}g^{\mathrm{}}P_{\mathrm{}}(\widehat{𝐬}\widehat{𝐬}^{})`$ and $`g=0.98`$. This choice of parameters corresponds to $`\mathrm{}^{}=1\mathrm{mm}`$ and $`D_0=0.8\mathrm{cm}^2\mathrm{ns}^1`$. The separation between the measurement planes $`L`$ was varied in order to explore the effects of the corrections. A single point absorber was placed at the midpoint of the measurement planes with $`𝐫_0=(0,0,L/2)`$ and $`A=1\mathrm{cm}^3\mathrm{ns}^1`$. The sources and detectors were located on a square lattice with spacing $`h`$. The total number of source-detector pairs $`N`$ was varied, along with $`h`$, as indicated below. To demonstrate the stability of the reconstruction in the presence of noise, Gaussian noise of zero mean was added to the data at the 1% level, relative to the average signal. Note that the level of regularization was chosen to be the same for all reconstructions.
Reconstruction of $`\delta \alpha (𝐫)`$ for a point absorber defines the point spread function (PSF) of the reconstruction algorithm. The resolution $`\mathrm{\Delta }x`$ is defined as the half width at half maximum of the PSF. In Fig. 1(a) we consider the case of a thick layer with $`L=6.6\mathrm{cm}`$. The above parameters were chosen to be $`h=0.83\mathrm{mm}`$ and $`N=1.5\times 10^9`$. PSFs with and without corrections are shown. It may be seen that the effect of the corrections is negligible in the case of a thick layer and the resolution $`\mathrm{\Delta }x=3.5\mathrm{}^{}`$. For the case of a layer of intermediate thickness with $`L=1.1\mathrm{cm}`$, as shown in Fig. 1(b), the corrections have a more significant effect. In particular, with $`h=0.28\mathrm{mm}`$ and $`N=1.2\times 10^{11}`$, we found that $`\mathrm{\Delta }x=0.9\mathrm{}^{}`$ for the uncorrected reconstruction and $`\mathrm{\Delta }x=0.7\mathrm{}^{}`$ for the corrected reconstruction. The corrections are most significant for the thinnest layer considered in this study, achieving a factor of two improvement in resolution when $`L=0.55\mathrm{cm}`$. In this case, with $`h=0.14\mathrm{mm}`$ and $`N=1.9\times 10^{12}`$, we found that $`\mathrm{\Delta }x=0.4\mathrm{}^{}`$ for the uncorrected case and $`\mathrm{\Delta }x=0.2\mathrm{}^{}`$ for the corrected case as shown in Fig. 1(c). Note that the number of source-detector pairs $`N10^910^{12}`$ may be achieved in modern non-contact optical tomography systems Ripoll\_2004 .
In conclusion, we have described a series of corrections to the usual formulation of the DA in optical tomography. We have found that these corrections give rise to superresolved images with resolution below $`\mathrm{}^{}`$. Several comments on these results are necessary. First, the effects of corrections were demonstrated to be most significant in optically thin layers. However, corrections to the DA may also be expected to be important for thick layers when inhomogeneities in the absorption are located near the surface. Second, the results of this study were obtained without resorting to so-called inverse crime. That is, forward scattering data was obtained from the full RTE under conditions when the DA is known to break down. Third, the use of analytic reconstruction algorithms was essential for handling the extremely large data employed in this study. Finally, we note that higher order corrections to the DA are also expected to be important for the nonlinear ISP.
This research was supported by the NIH under the grant P41RR02305 and by the AFOSR under the grant F41624-02-1-7001.
Fig. 1. Reconstructions of a point absorber for different thicknesses of the slab using the corrected (solid curve) and uncorrected (dashed curve) DA. |
warning/0507/astro-ph0507545.html | ar5iv | text | # PPN-limit of Fourth Order Gravity inspired by Scalar-Tensor Gravity
## I Introduction
The recent debate about the origin of the cosmic acceleration, induced by the results of several astrophysical observations sneia ; cmbr ; lss , led to investigate several theoretical approaches capable of providing viable physical mechanisms to the dark energy problem. In this wide discussion, no scheme seems, up to now, to furnish a final answer to this puzzling conundrum. Nevertheless among the different models, ranging from quintessential scenarios steinhardt , which generalize the cosmological constant approach starobinsky-shani , to higher dimensional scenarios braneworlds ; dgp or the resort to cosmological fluids with exotic equation of state chaplygin ; vdw and unified approaches considering even dark matter hobbit ; unified , an interesting scheme which seems to deserve a major attention is represented by higher order theories of gravity. This approach obtained by the generalization of the Einstein gravity, has led to interesting results both in the metric formulation curv-quint ; noi-ijmpd ; noi-review ; carroll ; odintsov-m and in the Palatini one palatini ; francaviglia .
Recently some authors have analyzed the PPN-limit of such theories both in the metric and in the Palatini approach olmo ; allemandi-ruggiero with contrasting results. Thus, it seems interesting to deepen the discussion about the Post Parametrized Newtonian (PPN) behaviour of this theory. The purpose is to verify if the cosmological reliability of such a scheme can be drawn even on the Solar System scales and to understand if the hypothesis of a unique fluid working as a two “faces” component (matter and geometry) can be a workable one.
In this paper we exploit the strict analogy between the higher order gravity and the scalar-tensor theories to develop a PPN-formalism for a general fourth order gravity model in the metric framework, working, in general, for extended theories of gravity. There are strong analogies between these two approaches. The similarity between the non-minimally coupled scalar models and the higher order gravity ones is known since 1983 teyssandier83 , when it was demonstrated the similarity between a scalar-tensor Lagrangian of Brans-Dicke type and fourth order gravity. Actually, such an interpretation goes well beyond conformal transformations, since it is a formal analogy without any physical change in the dynamical variables of the system.
In this paper, we further discuss the analogy between fourth order gravity and scalar-tensor gravity considering the PPN-parametrization descending from such a similarity. As main result, we show, despite some recent studies olmo , that Solar System experiments do not exclude the possibility that higher order gravity theories can represent a viable approach even at scales shorter than the cosmological ones. In other words, standard General Relativity should be revised both at cosmological and Solar System distances in order to solve several mismatches between the theoretical predictions and the observational results.
## II Fourth Order Gravity vs. Scalar Tensor Gravity
Let us recall how the analogies between the two schemes arise. As it is well known, scalar-tensor gravity is obtained if a scalar-field-matter Lagrangian is non - minimally coupled with the Hilbert-Einstein Lagrangian. The general action for a such theory is capoz-scal-tens :
$$𝒜=d^4x\sqrt{g}\left[F(\varphi )R+\frac{1}{2}g^{\mu \nu }\varphi _{;\mu }\varphi _{;\nu }V(\varphi )+\kappa _m\right],$$
(1)
where $`F(\varphi )`$ is the coupling function, $`V(\varphi )`$ the self-interaction potential, $`\varphi `$ a scalar field, $`_m`$ the ordinary matter Lagrangian and $`\kappa `$ the dimensional coupling. This relation naturally provides Brans-Dicke gravity brans-dicke if it is rearranged through the substitutions : $`\phi =F(\varphi ),\omega (\phi )={\displaystyle \frac{F(\varphi )}{2F^{}(\varphi )^2}}`$ capoz-brans ; its peculiarity is to account for Mach principle which leads back inertial forces within the background of gravitational interactions.
The $`f(R)`$ gravity action in the metric formalism is the following curv-quint ; noi-review
$$𝒜=d^4x\sqrt{g}\left[f(R)+\kappa _m\right],$$
(2)
which depends on the metric $`g_{\mu \nu }`$ and the matter fields. Again $`\kappa `$ defines the dimensional coupling. The energy-momentum tensor of matter is given by the relation $`T_{\mu \nu }^m={\displaystyle \frac{2}{\sqrt{g}}}{\displaystyle \frac{\delta _m}{\delta g^{\mu \nu }}}`$.
From the action (2), we obtain the fourth order field equations :
$$f^{}(R)R_{\mu \nu }\frac{1}{2}f(R)g_{\mu \nu }=f^{}(R)^{;\alpha \beta }\left(g_{\mu \alpha }g_{\nu \beta }g_{\mu \alpha }g_{\alpha \beta }\right)+\kappa T_{\mu \nu }^m,$$
(3)
which can be recast in a more expressive form as:
$$G_{\mu \nu }=\frac{1}{f^{}(R)}\{\frac{1}{2}g_{\mu \nu }[f(R)f^{}(R)R]+f^{}(R)_{;\mu \nu }+.$$
$$.g_{\mu \nu }\mathrm{}f^{}(R)\}+\frac{\kappa }{f^{}(R)}T^m_{\mu \nu },$$
(4)
where $`G_{\mu \nu }`$ is the Einstein tensor and $`f^{}(R)df/dR`$; the two terms $`f^{}(R)_{;\mu \nu }`$ and $`\mathrm{}f^{}(R)`$ imply fourth order derivatives of the metric $`g_{\mu \nu }`$. On the other side, if $`f(R)`$ is a linear function of the scalar curvature, $`f(R)=a+bR`$, the field equations become the ordinary second-order ones.
Considering the trace of Eq.(4),
$$3\mathrm{}f^{}(R)+f^{}(R)R2f(R)=\kappa T.$$
(5)
Such an equation can be interpreted as the equation of motion of a self-interacting scalar field, where the self-interaction potential role is played by the quantity $`V(R)=f^{}(R)R2f(R)`$. This analogy can be developed each time one considers an analytic function of $`R`$ which can be algebraically inverted so that $`R`$ reads as $`R=R(f^{})`$, in other words it has to be $`f^{\prime \prime }(R)0`$. In fact, defining
$`\varphi `$ $``$ $`f^{}(R)`$ (6)
$`V(\varphi )`$ $``$ $`R(\varphi )f^{}(R)f(\varphi )`$ (7)
we can write Eqs.(4) and (5) as
$$G_{\mu \nu }=\frac{\kappa }{\varphi }T_{\mu \nu }\frac{V(\varphi )}{2\varphi }g_{\mu \nu }+\frac{1}{\varphi }\left(\varphi _{;\mu \nu }g_{\mu \nu }\mathrm{}\varphi \right)$$
(8)
$$3\mathrm{}\varphi +2V(\varphi )\varphi \frac{dV}{d\varphi }=\kappa T,$$
(9)
which can also be obtained from a Brans-Dicke action of the form
$$𝒜_\varphi =d^4x\sqrt{g}\left[\varphi RV(\varphi )+\kappa _m\right].$$
(10)
This expression is related to the so called O’Hanlon Lagrangian, which belongs to a class of Lagrangians introduced in order to achieve a covariant model for a massive dilaton theory o'hanlon .
It is evident that the Lagrangian (10) is very similar to a Brans-Dicke theory, but is lacking of the kinetic term. The formal analogy between the Brans-Dicke scheme and fourth order gravity schemes is obtained in the particular case $`\omega _{BD}=0`$.
If we consider the matter term vanishing, Eq.(5) becomes an ordinary Klein-Gordon equation, where $`f^{}(R)`$ plays the role of an effective scalar field whose mass is determined by the self-interaction potential.
## III PPN-Formalism in Scalar Tensor Gravity
Along this paper, we base our discussion on the analogy between scalar-tensor theories of gravity and the higher order ones to analyze the problem of the PPN-limit for the fourth order gravity model. Recently the cosmological relevance of higher order gravity has been widely demonstrated. On the other side, the low energy limit of such theories is still not satisfactory investigated, although some results on the galactic scales have been already achieved newtlim . A fundamental test to understand the relevance of such a scheme is to check if there is even an accord with Solar System experiments. As outlined in the introduction, some controversial results have been recently proposed olmo ; allemandi-ruggiero . To better develop this analysis, we can refer again to the scalar-tensor - higher order gravity analogy, exploiting the PPN results obtained in the scalar-tensor scheme esposito-farese .
A satisfactory description of PPN limit for this kind of theories has been developed in esposito-farese ; damour1 ; damour2 . In these works, the problem has been treated providing interesting results even in the case of strong gravitational sources like pulsars and neutron stars where the deviations from General Relativity are obtained in a non-perturbative regime damour2 . A clear summary of this formalism can be found in the papers esposito-farese and schimd05 .
The action to describe a scalar-tensor theory can be assumed, in natural units, of the form (1). The matter Lagrangian density is again considered depending only on the metric $`g_{\mu \nu }`$ and the matter fields. This action can be easily redefined in term of a minimally coupled scalar field model via a conformal transformation of the form $`g_{\mu \nu }^{}=F(\varphi )g_{\mu \nu }`$. In fact, assuming the transformation rules:
$$\left(\frac{d\psi }{d\varphi }\right)^2=\frac{3}{4}\left(\frac{d\mathrm{ln}F(\varphi )}{d\varphi }\right)^2+\frac{1}{2F(\varphi )},$$
(11)
and
$$A(\psi )=F^{1/2}(\varphi ),V(\psi )=V(\varphi )F^2(\varphi ),$$
(12)
$$_m^{}=_mF^2(\varphi ),$$
(13)
one gets the action
$$𝒜_{}=\sqrt{g_{}}\left[R_{}+\frac{1}{2}g_{}^{\mu \nu }\psi _{,\mu }\psi _{,\nu }V(\psi )+_{}^{}{}_{m}{}^{}\right].$$
(14)
The first consequence of such a transformation is that now the non-minimal coupling is transferred on the ordinary matter sector. In fact, the Lagrangian $`_m^{}`$ is dependent not only on the conformally transformed metric $`g_{\mu \nu }^{}`$ and the matter field but it is even characterized by the coupling function $`A(\psi )^2`$. In the same way, the field equations can be recast in the Einstein frame. The energy-momentum tensor is defined as $`T_{\mu \nu }^m=\frac{2}{\sqrt{g^{}}}\frac{\delta _m}{\delta g_{\mu \nu }^{}}`$ and it is related to the Jordan expression as $`T_{\mu \nu }^m=A(\psi )T_{\mu \nu }^m`$. The function:
$$\alpha (\psi )=\frac{d\mathrm{ln}A(\psi )}{d\psi }$$
(15)
establishes a measure of the coupling arising in the Einstein frame between the scalar sector and the matter one as an effect of the conformal transformation (General Relativity is recovered when this quantity vanishes). It is possible even to define a control of the variation of the coupling function through the definition of the parameter $`\beta ={\displaystyle \frac{d\alpha (\psi )}{d\psi }}`$. Regarding the effective gravitational constant, it can be expressed in term of the function $`A(\psi )`$ as $`G_{eff}=\frac{G_N}{F(\varphi )}=G_NA^2(\psi )`$. It has to be remarked that such a quantity is, in reality, well different by the Newton constant measured in the Cavendish-like terrestrial experiments (see Eq.(21) below).
Let us now, concentrate on the scalar-tensor generalization of the local gravitational constraints. Deviations from General Relativity can be characterized through Solar System experiments will and binary pulsar observations which give an experimental estimate of the PPN parameters. These parameters were introduced by Eddington to better determine the deviation from the standard prediction of General Relativity, expanding local metrics as the Schwarzschild one, to higher order terms.
The generalization of this quantities to scalar-tensor theories allows the PPN-parameters to be expressed in term of non-minimal coupling function $`F(\varphi )`$ or, equivalently, in term of the parameter $`\alpha `$ defined in Eq.(15), that is :
$$\gamma ^{PPN}1=\frac{(F^{}(\varphi ))^2}{F(\varphi )+2[F^{}(\varphi )]^2}=2\frac{\alpha ^2}{1+\alpha ^2},$$
(16)
$$\beta ^{PPN}1=\frac{1}{4}\frac{F(\varphi )F^{}(\varphi )}{2F(\varphi )+3[F^{}(\varphi )]^2}\frac{d\gamma ^{PPN}}{d\varphi }=$$
$$=\frac{1}{2}\frac{\alpha ^2}{(1+\alpha ^2)^2}\frac{d\alpha }{d\psi }.$$
(17)
The above definitions imply that the PPN-parameters become dependent on the non-minimal coupling function $`F(\varphi )`$ and its derivatives. They can be directly constrained by the observational data. Actually, Solar System experiments give accurate indications on the ranges of $`\gamma _0^{PPN},\beta _0^{PPN}`$<sup>1</sup><sup>1</sup>1We indicate with the subscript <sub>0</sub> the Solar System measured estimates.. Results are summarized in Tab.1.
The experimental results can be substantially resumed into the two limits schimd05 :
$$|\gamma _0^{PPN}1|2\times 10^3,|\beta _0^{PPN}1|6\times 10^4,$$
(18)
which can be converted into constraints on $`\alpha _0`$ and $`\beta _0`$. In particular, the Cassini spacecraft value induces the bound $`\alpha _0{\displaystyle \frac{F_{0,\varphi }}{F_0}}<\mathrm{\hspace{0.17em}4}\times 10^4`$. At first sight, one can deduce that the first derivative of the coupling function $`A(\psi )`$ has to be very small, which means a very low interaction between matter and the scalar field; conversely the second derivative $`\beta _0`$ can take large values so that the matter sector may be strongly coupled with scalar degrees of freedom esposito-farese .
Together with the Solar System experiments, even binary-pulsar tests can be physically significant to characterize the PPN-parameters. From this analysis damour1 ; damour2 ; esposito-farese descends that the second derivative can be a large number, i.e. $`\beta _0>4.5`$ even for a vanishingly small $`\alpha _0`$.
This constraint allows to achieve a further limit on the two PPN-parameters $`\gamma ^{PPN}`$ and $`\beta ^{PPN}`$, which can be outlined by means of the ratio :
$$\frac{\beta ^{PPN}1}{\gamma ^{PPN}1}<\mathrm{\hspace{0.17em}1.1}.$$
(19)
The singular $`(0/0)`$ nature of this ratio puts in evidence that it was not possible to get such a limit in the case of weak - field experiments (see for details esposito-farese ).
For sake of completeness, we cite here even the shift that the scalar-tensor gravity induces on the theoretical predictions for the local value of the gravitational constant as coming from Cavendish-like experiments. This quantity represents the gravitational coupling measured when the Newton force arises between two masses :
$$G_{Cav}=\frac{Fr^2}{m_1m_2}.$$
(20)
In the case of scalar tensor gravity, the Cavendish coupling reads :
$$G_{Cav}=\frac{G_N}{F(\varphi )}\left[1+\frac{[F^{}(\varphi )]^2}{2F(\varphi )+3[F^{}(\varphi )]^2}\right]=$$
$$=G_NA^2(\psi )(1+\alpha ^2).$$
(21)
From the limit on $`\alpha `$ coming from Cassini spacecraft, the difference between $`G_{Cav}`$ and $`G_{eff}`$ is not more than the $`10^3\%`$.
## IV PPN limit of fourth order gravity inspired by the scalar-tensor analogy
In previous section, we discussed the PPN limit in the case of a scalar-tensor gravity. These results can be extended to the case of fourth order exploiting the analogy with scalar-tensor case developed in Sec. 2.
We have seen that fourth order gravity is equivalent to the introduction of a scalar extra degree of freedom into the dynamics. In particular, from this transformation, it derives a Brans-Dicke type Lagrangian with a vanishing Brans-Dicke parameter $`\omega _{BD}=\mathrm{\hspace{0.17em}0}`$. Performing the change of variables implied by a conformal transformation, the Brans-Dicke Lagrangian can be furtherly transformed into a Lagrangian where the non-minimal coupling is moved onto the matter side as in (14). The net effect is that, as in the case of a “true” scalar-tensor theory, it is possible to develop an Einstein frame formalism which allows a PPN-limit analysis. The basic physical difference between the two descriptions is that the quantities entering the PPN-parameters $`\gamma ^{PPN}`$ and $`\beta ^{PPN}`$, or the derivatives of the non-minimal coupling function $`A(\psi )`$, are now $`f^{}(R)`$ and its derivatives with respect to the Ricci scalar $`R`$ since the non minimal coupling function in the Jordan frame is $`f^{}(R)\varphi `$.
Alternatively, to obtain a more versatile equivalence between the two approaches it is possible to write down fourth order gravity by an analytic function of the Ricci scalar considering the identification induced by the field equations, i.e. $`\phi R`$. In fact, if one takes into account the scalar-tensor Lagrangian :
$$d^4\sqrt{g}\left[F(\phi )+(R\phi )F^{}(\phi )+\kappa _m\right],$$
(22)
the variation with respect $`\phi `$ and the metric provide the above identification and a system of field equations which are completely equivalent to the ordinary ones descending from fourth order gravity. The expression (22) can be recast in the form of the O’Hanlon Lagrangian (10) by means of the substitutions :
$$\varphi F^{}(\phi ),V(\varphi )\phi F^{}(\phi )F(\phi ),$$
(23)
where, in such a case, the prime means the derivative with respect to $`\phi `$. It is evident that the new scalar-tensor description implies a non-minimal coupling function through the term
$$F^{}(\phi )=\frac{df(R)}{dR},$$
(24)
and the identification $`\phi R`$ implies that the higher order derivatives can be straightforwardly generalized.
At this point, it is immediate to extend the results of the PPN-formalism developed for scalar-tensor gravity to the case of a fourth order theory. In fact, it is possible to recast the PPN parameters (16)-(17) in term of the curvature invariants quantities.
This means that the non-minimal coupling function role, in the fourth order scenario, is played by the $`df(R)/dR`$ quantity. As a consequence the PPN-parameters (16) and (17) become :
$$\gamma _R^{PPN}1=\frac{f^{\prime \prime }(R)^2}{f^{}(R)+2f^{\prime \prime }(R)^2},$$
(25)
$$\beta _R^{PPN}1=\frac{1}{4}\frac{f^{}(R)f^{\prime \prime }(R)}{2f^{}(R)+3f^{\prime \prime }(R)^2}\frac{d\gamma _R^{PPN}}{d\varphi }.$$
(26)
These quantities have, now, to fulfill the requirements drawn from the experimental tests resumed in Table 1. The immediate consequence of such definitions is that derivatives of fourth order gravity theories have to satisfy constraints in relation to the actual measured values of the Ricci scalar $`R_0`$. As a matter of fact, one can check these quantities by the Solar System experimental prescriptions and deduce the compatibility between fourth order gravity and General Relativity.
Since the definitions (25) and (26) do not allow to obtain, in general, upper limits on $`f(R)`$ from the constraints of Table 1, one can arbitrarily chose classes of fourth order Lagrangians, in order to check if the approach is working. We shall adopt classes of Lagrangians which are interesting from a cosmological point of view since give viable results to solve the dark energy problem curv-quint ; noi-review ; noi-ijmpd .
In principle, one can try to obtain some hints on the form of $`F(\phi )`$ (or correspondently of the $`f(R)`$) by imposing constraints provided from the Lunar Laser Ranging (LLR) experiments and the Cassini spacecraft measurements which give direct stringent estimates of PPN-parameters. After, one can try to solve these relations and then to verify what is the response to the pulsar upper limit with respect to the ratio $`{\displaystyle \frac{\beta ^{PPN}1}{\gamma ^{PPN}1}}<\mathrm{\hspace{0.17em}1.1}`$. This procedure has shown that generally if the two Solar System relations are verified, the pulsar constraint is well fitted by a modified gravity model. However this result is strictly influenced by the error range of the Cassini and LLR tests.
After this remark, one can consider different fourth order Lagrangians with respect to the two Solar System constraints coming from the perihelion shift of Mercury and the Very Long Baseline Interferometry.
The results are summarized in Table 2. We have listed the fourth order Lagrangians considered in the first column and the limit on the model parameters induced by the Solar System constraints is in the second column.
As it is possible to see, the PPN-limits induced by the Solar System tests can be fulfilled by different kinds of fourth order Lagrangians provided that their parameters remain well defined with respect to the background value of the curvature.
These results corroborate evidences for a defined PPN-limit which does not exclude higher order gravity. They are in contrast with other recent investigations olmo ; ppn-bad , where it has been pointed out that this kind of theories are not excluded by experimental results in the weak field limit and with respect to the PPN prescriptions.
Similar results also hold for Lagrangians as $`f(R)=f_0R^n`$ and $`f(R)=R+\frac{\mu }{R}`$ which have shown interesting properties from a cosmological point of view curv-quint ; noi-review ; noi-ijmpd ; carroll . This fact allows to establish a significant link between gravity at local and cosmological scales.
Finally a remark is in order. It has to be taken into account that the $`f(R)=A\mathrm{ln}[R]`$ does not admit a Minkowski background around which to perform the usual post-Newtonian analysis. Due to this fact this model is essentially different from the others in the weak energy limit.
## V Conclusions
Since the issue of higher order gravity is recently become a very debated matter, we have discussed its low energy limit considering the PPN-formalism in the metric framework. The study is based on the analogy between the scalar-tensor gravity and fourth order gravity. Such an investigation is particularly interesting even in relation to the debate about the real meaning of the curvature fluid which could be a natural explanation for dark energy curv-quint ; noi-review ; noi-ijmpd ; carroll ; odintsov-m ; palatini ; francaviglia ; olmo ; allemandi-ruggiero . The PPN-limit indicates that several fourth order Lagrangians could be viable on the Solar System scales. It has to remarked that the Solar System experiments pose rather tight constraints on the values of coupling constants, e.g. $`f_0`$ (see Table II). Such a result does not agree with the very recent papers olmo which suggest negative conclusions in this sense, based on questionable theoretical assumptions and extrapolations.
It is evident that such discussion does not represent a final answer on this puzzling issue. Nevertheless it is reasonable to affirm that extended gravity theories cannot be ruled out, definitively, by Solar System experiments. Of course, further accurate investigations are needed to achieve some other significant indications in this sense, both from theoretical and experimental points of view. For example the study of higher order gravity PPN-limit directly in the Jordan frame could represent an interesting task for forthcoming investigations.
An important concluding remark is due at this point. A scalar-tensor theory can be recast in the Einstein frame, via a conformal transformation, implying an equivalent framework. Actually, dealing with higher order gravity, there is no more such a conformal transformation able to “equivalently” transform the whole system from the Jordan frame to the Einstein one. Effectively, it is possible to conformally transform a higher order (and, in particular, a fourth order) theory into an Einstein-like with the addiction of some scalar fields as a direct consequence of the equivalence between the higher order framework and the scalar-tensor one at level of the classical field equations. This equivalence addressed, as dynamical equivalence wands:cqg94 , does not holds anymore when one considers configurations which do not follow the classical trajectories, for example in the case of quantum effects. A fundamental result which follows from this considerations is that dealing with the early-time inflationary scenario one can safely perform calculations for the primordial perturbations in the Einstein conformal frame of a scalar-tensor model while it is not possible to develop such calculations in the case of an higher order gravity scenario since the scalar degrees of freedom are no more independent of the gravitational field source. This issue holds, if the effective field is induced from geometrical degrees of freedom. Since the PPN-limit is achieved in the semi-classical limit, when the conformal factor turns out to be well defined, deductions about the PPN-limit for fourth order gravity models, developed exploiting the analogy with the scalar-tensor scheme, are safe from problems. |
warning/0507/astro-ph0507194.html | ar5iv | text | # Morphology of synchrotron emission in young supernova remnants
## 1 Introduction
Shocks in supernova remnants (SNRs) are believed to produce the majority of the Galactic cosmic-rays (CRs) at least up to the “knee” ($`3\times 10^{15}`$ eV). The particle acceleration mechanism most likely responsible for this is known as diffusive shock acceleration (DSA) (e.g., Drury 1983; Blandford & Eichler 1987). This mechanism may transfer a large fraction of the ram kinetic energy (up to $`50\%`$) into relativistic particles and remove it from the thermal plasma (see, for example, Jones & Ellison 1991).
Convincing observational support for the acceleration of particles in shell-type SNRs comes from their nonthermal radio and X-ray emissions due to synchrotron radiation from relativistic GeV and at least TeV electrons, respectively. In radio and X-rays, synchrotron-dominated SNRs display various morphologies: for instance, the synchrotron emission dominates in two bright limbs in SN 1006 (e.g., Rothenflug et al. 2004) whereas it is distorted and complex in RX J1713.7–3946 (e.g., Cassam-Chenaï et al. 2004b). The detection and imaging with the HESS telescopes of TeV $`\gamma `$-rays in RX J1713.7–3946 provides unambiguous evidence for particle acceleration to very high energies. The $`\gamma `$-ray morphology in this remnant is similar to that seen in X-rays (Aharonian et al. 2004).
Recent works based on Chandra (Vink & Laming 2003, for Cas A) and XMM-Newton (Cassam-Chenaï et al. 2004a, for Kepler’s SNR) observations have demonstrated that X-ray synchrotron emission is also present in ejecta-dominated SNRs and largely contributes to the continuum emission at the forward shock. This X-ray emission arises from sharp filaments encircling the SNR’s outer boundary. The observed width of these filaments is a few arcseconds, and has been used to constrain the magnetic field intensity just behind the shock<sup>1</sup><sup>1</sup>1Magnetic field values are found to be at least 30 times higher than the typical Galactic field of $`3\mu `$G and imply that the field has been amplified, perhaps by the particle acceleration process (Bell & Lucek 2001). (Vink & Laming 2003; Berezhko et al. 2003; Berezhko & Völk 2004; Völk et al. 2005; Ballet 2005).
A number of recent hydrodynamical models, including particle acceleration and photon emission, have been presented to explain various features of these observations. Reynolds (1998) has described the morphology and spectrum of the synchrotron X-ray emission from SNRs in the Sedov evolutionary phase. Similar work based on numerical simulations was done by van der Swaluw & Achterberg (2004) who take into account the diffusion of particles. CRs are treated as test-particles in these studies.
Here, we expand on the work of Reynolds (1998) by considering young (ejecta-dominated) SNRs. We investigate the synchrotron emission morphology, both in radio and X-rays, as well as how it can be modified by efficient particle acceleration. Our results show that the radio and X-ray profiles are very different due to the effects of the magnetic field evolution and synchrotron losses in the interaction region between the contact discontinuity and the forward shock. For typical parameters, the radio emission peaks at the contact discontinuity while the X-ray emission forms sheet-like structures at the forward shock.
## 2 Hydrodynamics and particle acceleration
The hydrodynamic evolution of young supernova remnants, including the backreaction from accelerated particles, can be described by self-similar solutions if the initial density profiles in the ejected material (ejecta) and in the ambient medium have power-law distributions (Chevalier 1982, 1983), and if the acceleration efficiency (*i.e.* the fraction of total ram kinetic energy going into suprathermal particles) is independent of time.
Here, we use the self-similar model of Chevalier (1983) which considers a thermal gas ($`\gamma =5/3`$) and the cosmic-ray fluid ($`\gamma =4/3`$), with the boundary conditions calculated from the non-linear diffusive shock acceleration (DSA) model of Berezhko & Ellison (1999) and Ellison et al. (2000) as described in Decourchelle et al. (2000). This acceleration model is an approximate, semi-analytical model that determines the shock modification and particle spectrum from thermal to relativistic energies in the plane-wave, steady state approximation as a function of an arbitrary injection parameter, $`\eta _{\mathrm{inj}}`$ (*i.e.* the fraction of total particles which end up with suprathermal energies). The validity of the self-similar solutions has been discussed by Decourchelle et al. (2000) and direct comparisons between this self-similar model and the more general CR-hydro model of Ellison et al. (2004) showed good correspondence for a range of input conditions.
The hydrodynamic evolution provides the shock characteristics necessary to calculate the particle spectrum at the forward shock<sup>2</sup><sup>2</sup>2We do not consider CR production at the reverse shock since the magnetic field at the reverse shock may be considerably smaller than that at the forward shock due to the dilution by expansion and flux freezing of the progenitor magnetic field (see Ellison et al. 2005)., at any time. Once a particle spectrum has been produced at the shock, it will evolve downstream because of radiative and adiabatic expansion losses. We assume that the accelerated particles remain confined to the fluid element in which they were produced, so adiabatic losses are determined directly from the fluid element expansion. The basic power law spectrum produced by DSA, before losses are taken into account, is modified at the highest energies with a exponential cutoff, $`\mathrm{exp}(p/p_{\mathrm{max}})`$, where $`p_{\mathrm{max}}`$ is determined by matching either the acceleration time to the shock age or the upstream diffusive length to some fraction of the shock radius. In our simulation, the electron-to-proton density ratio at relativistic energies, $`(e/p)_{\mathrm{rel}}`$, is set equal to $`0.01`$ (see Ellison et al. 2000).
Unless explicitly stated, our numerical examples are given for the following supernova parameters: $`M_{\mathrm{ej}}=5\mathrm{M}_{}`$ for the ejected mass, $`E_{51}=1`$ where $`E_{51}`$ is the kinetic energy of the ejecta in units of $`10^{51}`$ ergs and $`n=9`$, where $`n`$ is the index of the initial power-law density profile in the ejecta ($`\rho r^n`$). In our simulations, the SNR age is $`t_f=400`$ years and the shock velocity at the forward shock is $`v_s5\times 10^3\mathrm{km}/\mathrm{s}`$. For the ambient medium parameters, we take a magnetic field $`B_0=10\mu \mathrm{G}`$, a density $`n_0=0.1\mathrm{cm}^3`$, an ambient gas pressure $`p_{\mathrm{g},0}/k=\mathrm{2\hspace{0.25em}300}\mathrm{K}\mathrm{cm}^3`$ and $`s=0`$, where $`s`$ is the index of its initial power-law density profile ($`\rho r^s`$). The case $`s=0`$ corresponds to a uniform interstellar medium ($`s=2`$ describes a stellar wind).
In the next section, we discuss the importance of the magnetic field for the synchrotron emission and particle acceleration. We do not, however, explicitly include the dynamical influence of the magnetic field on the hydrodynamics.
## 3 Results
### 3.1 Magnetic field
To track the synchrotron losses, we are interested in the temporal evolution of the magnetic field behind the shock. We assume the magnetic field to be simply compressed at the shock and passively carried by the flow, frozen in the plasma, so that it evolves conserving flux. In this simple 1-D approach, we do not consider any production of the SNR magnetic field, for instance, by hydrodynamical instabilities which is an additional effect. As for the magnetic field ahead of the forward shock, it is assumed to be isotropic and fully turbulent. Appendix A (see the on-line version) shows how to compute the magnetic field profile for self-similar solutions in both test-particle and nonlinear particle acceleration cases.
#### 3.1.1 Test-Particle limit
We first discuss the behavior of the normal and tangential components of the magnetic field in the test-particle case where the backreaction of the accelerated particles is neglected.
When the SNR evolves in an ambient medium which is uniform in density and magnetic field, the expansion and flux freezing generally cause the tangential component of the magnetic field to increase at the contact discontinuity whereas the normal component falls to zero (Fig. 1). As a result, the magnetic field profile is dominated by the tangential component.
One has often invoked hydrodynamic instabilities to explain the magnetic field increase at the interface between the shocked ejecta and the shocked ambient medium (Jun et al. 1995). The numerical simulations of Jun & Norman (1996) have shown that the magnetic field could be amplified by a factor 60 by Rayleigh-Taylor and Kelvin-Helmholtz instabilities. Here, we note that simple advection of the magnetic field already predicts amplification by a factor 5 (Table 2 top, $`n=9`$).
We note that, if the SNR evolves in a wind with a decreasing initial density profile, advection goes the other way (diluting the magnetic field instead of amplifying it). But when both the ambient density and magnetic field decrease with radius, as would be the case for a pre-supernova stellar wind, the magnetic field is larger close to the contact discontinuity than at the forward shock (by a factor of $`1000`$ in some cases). This is because the dilution of the advected magnetic field is negligible compared to the fact that the ambient magnetic field was much larger at early times.
#### 3.1.2 Nonlinear Particle Acceleration
We now consider the behavior of the normal and tangential components of the magnetic field in the nonlinear case where the backreaction of the accelerated particles on the shock is taken into account.
In the ideal non-linear case, where the acceleration is instantaneous, the magnetic field diverges at the contact discontinuity because of its tangential component, whatever the injection efficiency is, as in the test-particle case. However, the contrast between the magnetic field in a given fluid element and the one just behind the shock, will be always smaller than in the test-particle case (see Table 2). Figure 2 shows the profile of the total downstream magnetic field for different values of the injection efficiency. Table 1 shows the associated compression ratio and immediate post-shock magnetic field.
### 3.2 Synchrotron emission
Once the magnetic field structure and the particle spectrum (attached to a fluid element) modified by the radiative and adiabatic expansion losses as computed in Reynolds (1998) are known, we compute the synchrotron emission (Rybicki & Lightman 1979), averaged over the pitch-angle, in any energy band<sup>3</sup><sup>3</sup>3We did not calculate the synchrotron emission from the precursor..
Figure 3 shows the radial profiles of the synchrotron emission in the radio (top panel) and X-ray (bottom panel) domains for different injection efficiencies, $`\eta _{\mathrm{inj}}`$. An increase in the injection efficiency not only provides a larger number of accelerated electrons, but also a larger compression of the downstream magnetic field (see Table 1) and a narrower interaction region. These effects combine to produce enhanced synchrotron emission as the injection increases.
The radio synchrotron emission is produced by GeV electrons which are not affected by radiative losses. Consequently, the radio synchrotron emission critically depends on the final magnetic field profile (Fig. 2) and, therefore, peaks at the contact discontinuity. In contrast, the X-ray synchrotron emission is produced by the highest momentum electrons ($`10^{35}m_\mathrm{p}c`$) which, depending on the downstream field strength, may suffer radiative losses. The high energy electrons that have been accelerated at the earliest time have suffered strong synchrotron losses as they were advected behind the shock. Because of this, they are not numerous enough at the end to radiate in the X-ray regime despite a strong magnetic field. As a result, the X-ray synchrotron emission rapidly decreases behind the shock. The X-ray profile becomes sharper when the injection efficiency increases because it provides larger compression of the downstream magnetic field and then stronger synchrotron losses.
Figure 4 shows the synchrotron emission after integration along the line-of-sight. The radial profile of the radio emission (top panel) shows a peak at the contact discontinuity. The radial profile of the X-ray projected synchrotron emission (bottom panel) shows bright rims just behind the forward shock whose width decreases as the injection efficiency increases.
## 4 Discussion and Conclusion
We have computed the radio and X-ray synchrotron emission in young ejecta-dominated SNRs. This has been done using a one dimensional, self-similar hydrodynamical calculation coupled with a non-linear diffusive shock acceleration model, and taking into account the adiabatic and radiative losses of the electron spectrum during its advection in the remnant.
We show that the morphology of the synchrotron emission in young ejecta-dominated SNRs is very different in radio and X-ray. This is the result of the increased magnetic field toward the contact discontinuity, to which only low energy electrons that emit radio are sensitive, while the high energy electrons emitting X-rays experience strong radiative losses and are mostly dependent on the post-shock magnetic field.
Briefly, the radio synchrotron emission increases as one moves from the forward shock toward the contact discontinuity due to a compression of the magnetic field (particularly its tangential component), assuming both uniform ambient density and upstream magnetic field. Such a compression naturally results from the dynamical evolution of the SNR. In contrast, because of the radiative losses, the X-ray synchrotron emission decreases behind the forward shock and forms sheet-like structures after line-of-sight projection. Their widths decrease as the acceleration becomes more efficient.
The morphology of the radio synchrotron emission obtained for the young ejecta-dominated stage of SNRs will differ from that of SNRs in the Sedov phase (but not in X-ray). Indeed, Reynolds (1998) has shown that both the normal and tangential components of the magnetic field decrease behind the forward shock in the Sedov phase and, as a result, we expect the radio synchrotron emission to decrease behind the shock (however, less rapidly than the X-ray synchrotron emission since the radio electrons do not experience radiative losses).
Our model qualitatively reproduces the main features of the radio and X-ray observations of emission in young ejecta-dominated SNRs (e.g., Tycho and Kepler), *i.e.* bright radio synchrotron emission at the interface between the shocked ejecta and ambient medium, and a narrow filament of X-ray emission at the forward shock. However, this model is unable to reproduce the thin radio filaments observed at the forward shock in some SNRs (for instance those seen in Tycho’s SNR, Dickel et al. 1991).
We note that extensions of this work to cases with exponential ejecta profiles and/or SNRs evolving in a pre-supernova stellar wind with varying magnetic fields, cannot be done with self-similar solutions. These cases can be calculated in the numerical CR-modified hydrodynamical model described in Ellison et al. (2005) and this work is in progress (Ellison & Cassam-Chenaï 2005).
## Appendix A Magnetic field evolution
The evolution of the normal (subscript $`r`$) and tangential (subscript $`t`$) components of the magnetic field at the downstream position, $`B`$, is given by (Reynolds & Chevalier 1981):
$`B_r(r)`$ $`=`$ $`B_{r,j}\left({\displaystyle \frac{r}{r_j}}\right)^2`$ (1)
$`B_t(r)`$ $`=`$ $`B_{t,j}{\displaystyle \frac{\rho }{\rho _j}}{\displaystyle \frac{r}{r_j}},`$ (2)
and the total magnetic field is simply (Reynolds 1998):
$$B(r)=\left(B_r(r)^2+B_t(r)^2\right)^{1/2}.$$
(3)
In these equations, $`r`$ and $`\rho `$ are, respectively, the radius and density of a fluid element at the current time that was shocked at the previous time $`t_j`$. At time $`t_j`$, the fluid element was just behind the shock at the radius $`r_j`$, with a density $`\rho _j`$ and a magnetic field $`B_j`$.
We assume that the upstream magnetic field at time $`t_j`$, $`B_{0,j}`$ is isotropic and fully turbulent so that the components of the immediate post-shock magnetic field $`B_j`$ in Eqs (1) and (2) are given on average by (Berezhko et al. 2002):
$`B_{r,j}`$ $`=`$ $`1/\sqrt{3}B_{0,j}`$ (4)
$`B_{t,j}`$ $`=`$ $`\sqrt{2/3}r_{\mathrm{tot}}B_{0,j}.`$ (5)
where $`r_{\mathrm{tot}}`$ is the shock compression ratio. In the self-similar approach, $`r_{\mathrm{tot}}`$ is assumed independent of time (see Decourchelle et al. 2000, for details).
We consider that the current magnetic field upstream of the forward shock, $`B_{0,s}`$, can behave like:
$$B_{0,s}=B_{0,j}\left(\frac{r_s}{r_j}\right)^q$$
(6)
where $`r_s`$ is the current shock radius. If the magnetic field is uniform, the index $`q`$ is equal to 0. In a stellar wind ($`s=2`$), the magnetic field profile may be decreasing yielding $`q=1`$ (Lyutikov & Pohl 2004) or $`q=2`$ if we assume that it is frozen in the plasma.
We define the magnetic field contrast factor, $`\sigma _BB/B_s`$, as the ratio between the current magnetic field in a fluid element, $`B`$, and the current one just behind the shock, $`B_s`$. We have:
$$\sigma _B=\left(\frac{\sigma _{B_r}^2+2r_{\mathrm{tot}}^2\sigma _{B_t}^2}{1+2r_{\mathrm{tot}}^2}\right)^{1/2}$$
(7)
where $`\sigma _{B_r}B_r/B_{r,s}`$ and $`\sigma _{B_t}B_t/B_{t,s}`$ are the magnetic field contrast factors of the normal and tangential components of the field, respectively. The components $`B_{r,s}`$ and $`B_{t,s}`$ obey the same relation as in Eqs (4) and (5).
### A.1 Test-Particle limit
Assuming adiabaticity of the thermal gas, the magnetic field contrast factors of the normal and tangential components of the field are given by:
$`\sigma _{B_r}`$ $`=`$ $`\left({\displaystyle \frac{R_s}{R}}\right)^2\left({\displaystyle \frac{v_j}{v_s}}\right)^{\beta _r}`$ (8)
$`\sigma _{B_t}`$ $`=`$ $`\left({\displaystyle \frac{P_{\mathrm{g},s}}{P_\mathrm{g}}}\right)^{3/5}\left({\displaystyle \frac{R_s}{R}}\right)^{(113s)/5}\left({\displaystyle \frac{v_j}{v_s}}\right)^{\beta _t}`$ (9)
where the indexes $`\beta _r`$ and $`\beta _t`$ are given by:
$`\beta _r`$ $`=`$ $`\left(q2\right){\displaystyle \frac{n3}{3s}}`$ (10)
$`\beta _t`$ $`=`$ $`{\displaystyle \frac{5n333s(n5)}{5(3s)}}+q{\displaystyle \frac{n3}{3s}}.`$ (11)
In Eqs (8) and (9), $`R_s/R`$ and $`P_{\mathrm{g},s}/P_\mathrm{g}`$ are the ratio of the self-similar radii and thermal gas pressures, respectively, between the shock (subscript $`s`$) and a fluid element (see Chevalier 1982). They depend on $`n`$ and $`s`$, but also weakly on $`v_j/v_s`$ where $`v_s`$ and $`v_j`$ are the current shock velocity and the shock velocity at the time $`t_j`$, respectively.
In the framework of these self-similar solutions, the forward shock velocity tends to infinity at early times, corresponding to fluid elements close to the contact discontinuity at the current time. To limit the maximum velocity to a realistic value, we look at the value of $`\sigma _B`$ for a shock velocity ratio $`v_j/v_s=10`$. For the typical forward shock velocity $`v_s`$ that we have used for the numerical application, the initial velocity corresponds to $`v_j5\times 10^4\mathrm{km}/\mathrm{s}`$. This shock velocity is the criterion used to define the radial position of the oldest fluid element that is currently located close to the contact discontinuity.
Here, we consider the case of both an uniform ambient medium ($`s=0`$) and upstream magnetic field ($`q=0`$). Under this assumption, $`B_s=B_j`$, since $`r_{\mathrm{tot}}`$ is constant with time. Then, the magnetic field contrast factor, $`\sigma _B`$ is equal to $`B/B_j`$ and can be viewed as a compression or a dilution factor. Table 2 (top) gives the contrast $`\sigma _B`$ for different values of $`n`$.
### A.2 Nonlinear Particle Acceleration
In the ideal non-linear case, where the acceleration is instantaneous and efficient, the thermal gas pressure falls to zero at the contact discontinuity while the relativistic gas pressure goes to infinity. Hence, the contrast factor of the tangential field component, $`\sigma _{B_t}`$, given by Eq. (9), obtained in the test-particle limit, is not defined when $`v_j/v_s`$ tends to infinity.
However, the contrast of the tangential component of the magnetic field can also be found by using the adiabaticity of the relativistic gas:
$`\sigma _{B_t}`$ $`=`$ $`\left({\displaystyle \frac{P_{\mathrm{c},s}}{P_\mathrm{c}}}\right)^{3/4}\left({\displaystyle \frac{R_s}{R}}\right)^{(103s)/4}\left({\displaystyle \frac{v_j}{v_s}}\right)^{\beta _t^{}}`$ (12)
where the index $`\beta _t^{}`$ is given by:
$`\beta _t^{}`$ $`=`$ $`{\displaystyle \frac{4n303s(n5)}{4(3s)}}+q{\displaystyle \frac{n3}{3s}}.`$ (13)
In Eq. (12), $`P_{\mathrm{c},s}/P_\mathrm{c}`$ is the ratio of the self-similar relativistic gas pressures between the shock (subscript $`s`$) and a fluid element. This ratio depends on $`n`$, $`s`$, and $`v_j/v_s`$. The contrast of the normal field component, $`\sigma _{B_r}`$, is still given by Eq. (8). The asymptotic behavior of the contrast factor, $`\sigma _{B_t}`$, can be derived from Eq. (12) because the relativistic gas pressure does not tend to zero at the contact discontinuity.
Because the thermal gas pressure vanishes as we approach the contact discontinuity in the case of ideal particle acceleration, i.e., when the acceleration is instantaneous and efficient, the contrast of the tangential field component, $`\sigma _{B_t}`$, will always be smaller than in the test-particle case where the thermal gas pressure rapidly tends to a constant (see Eq. 9). Table 2 (bottom) gives the lower and upper limits on the magnetic field contrast factor, $`\sigma _B`$, in the case of ideal nonlinear particle acceleration for $`\eta _{\mathrm{inj}}=10^3`$ and for different values of $`n`$ when both the ambient medium and upstream magnetic field are uniform ($`s=0`$ and $`q=0`$). The lower and upper limits on $`\sigma _B`$ are obtained by replacing in Eq. (12) the ratio of the self-similar relativistic gas pressures, $`P_{\mathrm{c},s}/P_\mathrm{c}`$, by the ratio of the self-similar total gas pressures, $`P_s/P(P_{\mathrm{c},s}+P_{\mathrm{g},s})/(P_\mathrm{c}+P_\mathrm{g})`$, and by the ratio between the self-similar relativistic gas pressure at the shock and the self-similar total gas pressure, $`P_{c,s}/P`$, respectively.
However, for an injection efficiency lower than $`5\times 10^4`$, the acceleration is not efficient enough for the shock to be modified at the beginning of the evolution. In that case, the fluid elements that have been shocked at the earliest times are still dominated by the thermal gas so that test-particle solutions could still apply locally. |
warning/0507/cond-mat0507276.html | ar5iv | text | # Photon-assisted electron-hole shot noise in multi-terminal conductors
## I Introduction
Theoretical and experimental investigations of the current and charge noise properties of small conductors are an important frontier of mesoscopic physics. The aim is to analyze noise as an additional source of information on the quantum statistical properties of small conductors. Interest in quantum communication and quantum computation have further drawn attention to this subject. We refer the interested reader to reviews,review1 ; review2 a conference book with extended articles on the subject,nazarov a special issue of a journal,special and to original work on shot noise in mesoscopic conductors. khlus ; lesovik ; buttiker90 ; buttiker91 ; birk ; reznikov ; kumar ; oberholzerprl Investigations have predominantly considered samples subject to stationary applied voltages which drive a dc-current through the sample. The work reported here is motivated by a recent experiment by Reydellet *et al.* \[glattli, \] which applies only an ac-voltage to a contact of a two-terminal sample. There is no dc-current linear in voltage. However, the ac-voltage leads to the generation of electron-hole pairs and the members of this pair are subsequently transmitted through the sample or reflected back into the excited contact.
For several reasons, it is of interest to analyze the noise in terms of electron-hole excitations. First, if an ac-voltage is applied to a conductor that is otherwise at equilibrium, electron-hole pairs are in fact the natural elementary excitations of the system. mosk This is in contrast with most of the literature on shot noise which even in the presence of ac-excitations uses an electron picture only. The understanding and interpretation of the results can differ dramatically depending on whether one relies on an all electron picture or on an electron-hole picture. A second reason to investigate electron-hole pair generation is that recently it was realized that such pairs are a source of entanglement, similar to the optical process in which a pair of photons with entangled polarization state is generated through down conversion.bee1 Indeed an electron-hole pair creation event leads to an electron and hole who’s spins are entangled.bee1 However, it is not only the spin degree which can be useful, but electron-hole sources can be used to generate orbital sam1 quasi-particle entanglement. This leads to simple and controllable geometries sam2 which permit the investigation of entanglement in electrical conductors without the need of superconducting/semiconductor/ferromagnetic hybrid structures. Indeed the dynamic generation of electron-hole pairs through the periodic modulation of potentials in the interior of conductors sam3 ; bee2 or through the application of pulses llb has recently been discussed.
Fig. 1 shows the type of structures which we analyze in this work. Electron-hole pairs are excited in contact $`1`$. The rate with which carriers are excited is much smaller than the inverse time taken by an electron or hole to be transmitted through the conductor. Fig. 1 depicts a particular scattering event in which the pair is split and the electron leaves the conductor through contact $`2`$ and the hole leaves the conductor through contact $`3`$. An essential point of our work is the fact the correlation of currents between contacts $`2`$ and $`3`$ is entirely determined by scattering events in which an electron-hole pair is split. Other scattering processes, for instance one in which only one carrier of the pair leaves the conductor through contact $`2`$ and the other is scattered back into contact $`1`$, obviously give no contribution to the current correlation. Thus in the limit of rare excitations of electron-hole pairs, the current-correlation between contact $`2`$ and $`3`$ is a direct measurement of the coincident creation of an electron and hole.
The term ’photon-assisted transport’ is applied here in the following sense: For weak coupling of the photon density with the electrons in the lead, the photon field appears in the many-body electron Hamiltonian as a weak periodic potential $`V(t)`$. We can neglect the feedback of the weak currents on the photon source. Thus in the following, we consider all photon sources as weak ac-bias potentials applied to the leads of the conductor. tuck
Photon-assisted current in quantum dots have been experimentally investigated in Ref. \[accurrent, \]. But only a few experiments have thus far investigated the noise properties of photon-assisted transport. Following the theoretical work of Lesovik and Levitov,LL Schoelkopf *et al*. schoel illuminated a normal diffusive conductor and found features in the zero-frequency shot noise at the photon energies $`V=nh\nu /e`$. Theoretical work on hybrid-superconducting system by Lesovik *et al*. les99 lead similarly to the experimental observation of features at energies $`V=nh\nu /2e`$ by Kozhevnikov *et al*. kozh In both of these experiments a dc-voltage was applied in addition to the radiation. A similar regime for the systems in the fractional quantum Hall state is discussed in Ref. \[tmart, \]. The theoretical works LL ; les99 consider photon-assisted transport purely as a particle transport problem. However, despite the fact that one measures a dc-current or a zero-frequency noise spectrum, the radiation actually also generates a charge and current response at the excitation frequency. The important role of displacement currents and their rectification effects have been emphasized in the work of Pedersen and Buttiker.pedersen In the work presented here we will also only consider non-interacting particles and derive expressions for electron and hole currents and their correlations. The role of Coulomb interactions, inelastic scattering and dephasing on the photon assisted noise of electron-hole pairs is treated in the Ref. \[poli, \].
In contrast, to the above mentioned experiments, the recent shot noise measurements by Reydellet *et al.* \[glattli, \] report results for the case that only an ac-voltage is applied to the leads. Reydellet *et al.* \[glattli, \] also discuss their results in terms of electron-hole excitations assuming that the electrons and holes generate a partition noise independently from each other and add incoherently. As indicated above our interest is in the correlated nature of the electron-hole processes and the manifestation of this correlation in shot-noise spectra. We first derive general expressions for electron and hole currents and subsequently apply them to two, three and four terminal conductors. Of particular interest is an arrangement which uses two out-of-phase ac-voltages of the same magnitude and frequency. The phase difference between the applied voltages provides an additional degree of control. It turns out that the exchange contribution to the current correlations buttiker90 ; buttiker91 ; VLB ; sam2 (the Hanbury Brown Twiss (HBT) correlations hbt ) are sensitive to the phase difference between the applied voltages. The HBT-correlations reflect indistinguishable two-particle processes in the shot noise. sam2 Sweeping the phase shift allows to maximize (minimize) shot-noise correlations. The positions of these extrema yield information about combinations of phases of the scattering matrix, which we call the HBT-phase. If the phase-difference of the ac-potentials is set to the HBT-phase, the correlation is maximal. This permits to explore the statistical properties of the HBT-phase. We analyze the distribution of the HBT-phase for a chaotic dot coupled to single channel leads.
The additional degree of control generated by two potentials which are out of phase has its analog in quantum pumping where a scatterer is modulated with two out-of-phase parameters. B\_p ; Switkes ; Avron ; VAA ; PVB <sup>,</sup> mosk Adiabatic quantum pumping uses in an essential way this additional degree of control. The difference is that here we deal with potentials applied to contacts and we analyze a situation in which the frequency to voltage ratio is large.
The paper is organized as follows: In Sec. II we define the geometry of the mesoscopic conductors, present the model assumptions and the theoretical approach used. Sec. III describes the electron-hole picture for photon-excited multiprobe conductors and presents sample-specific results for the shot-noise spectra. Sec. IV discusses an arrangement to measure the Hanbury Brown Twiss exchange interference correlations. We conclude in Sec. V. Appendix A gives the quantum state which leads to the HBT-exchange interference correlations, and Appendix B presents a probability theory argument to show the existence of electron-hole correlations.
## II Scattering approach
We consider a multi-terminal mesoscopic conductor connected to metallic contacts. The leads are subject to time-dependent voltages. We briefly recall the main aspects of the scattering approach and introduce the basic notations. We follow Refs. pedersen ; Hartree who distinguish external potentials (applied to leads) from internal potentials (generated in the interior of the conductor). This distinction is particularly useful to a theory formulated in terms of the scattering matrix: external potentials leave the scattering matrix invariant whereas internal potentials lead to emission and absorption of energy inside the mesoscopic conductor. A realistic description of time-dependent transport involves both external and internal oscillating potentials. Here we focus on the effect of external potentials, keeping the internal potential fixed.
Metallic reservoirs are considered as emitters (absorbers) of charge carriers incident on (exiting from) the mesoscopic conductor. Motion of electrons in the leads can be described by their energy, transverse mode number and direction of momenta. Thus we introduce operators of incoming and outgoing states: $`𝐚_\lambda (\epsilon _i)`$ and $`𝐛_\lambda (\epsilon _i)`$ are vectors of operators $`a_{\lambda n}(\epsilon _i)`$ and $`b_{\lambda n}(\epsilon _i)`$. The operator $`a_{\lambda n}(\epsilon _i)`$ annihilates a carrier incident in the $`\lambda `$-th lead with energy $`\epsilon _i`$ in the $`n`$-th transverse channel. The operator $`b_{\lambda n}(\epsilon _i)`$ annihilates an outgoing carrier in the $`\lambda `$-th lead with energy $`\epsilon _i`$ in the $`n`$-th transverse channel. Here and in the following greek indices run over all contacts of the conductor. For a stationary scatterer the connection between incoming and outgoing carriers is governed by scattering matrices which depend on one energy only. We consider the case of spinless electrons. The scattering matrix transforms the operators $`𝐚_\lambda (\epsilon _i)`$ into $`𝐛_\lambda (\epsilon _i)`$ according to $`𝐛_\alpha (\epsilon )=_\beta 𝒮_{\alpha \beta }(\epsilon )𝐚_\beta (\epsilon )`$. In a multichannel conductor, the matrix $`𝒮_{\alpha \beta }(\epsilon )`$ is a sub-block of the scattering matrix $`𝒮`$ for scattering from lead $`\beta `$ ($`N_\beta `$ channels) to lead $`\alpha `$ ($`N_\alpha `$ channels) with energy $`\epsilon `$. It has dimensions $`N_\alpha \times N_\beta `$.
The current operator in contact $`\lambda `$ is buttiker90
$`I_\lambda (t)`$ $`=`$ $`{\displaystyle \frac{e}{2\pi \mathrm{}}}{\displaystyle 𝑑\epsilon _1𝑑\epsilon _2e^{i(\epsilon _1\epsilon _2)t/\mathrm{}}}`$ (1)
$`\times `$ $`{\displaystyle \underset{\alpha \beta }{}}𝐚_\alpha ^{}(\epsilon _1)𝐀_{\alpha \beta }(\lambda ,\epsilon _1,\epsilon _2)𝐚_\beta (\epsilon _2).`$
Using the projector matrix $`𝟙_\lambda `$ which is a unit matrix of size $`N_\lambda \times N_\lambda `$ in the $`\lambda `$-th lead, we introduce the current matrix buttiker90
$$𝐀(\lambda ,\epsilon _1,\epsilon _2)=𝟙_\lambda 𝒮^{}(\epsilon _\mathrm{𝟙})𝟙_\lambda 𝒮(\epsilon _\mathrm{𝟚}).$$
(2)
In the following we use the abbreviated notation $`𝐀(\lambda ,\epsilon )`$ if the energies coincide and $`𝐀(\lambda )`$ if the energy dependence is unimportant. Electrons obey Fermi statistics. In order to transfer charge across the mesoscopic conductor the incoming charge state has to be filled and at the same time the outgoing state has to be open. A dc-voltage applied across the sample opens an energy window where both conditions are fulfilled and transport is possible.
Consider now a time-dependent potential $`eV_\alpha (t)=eV_\alpha \mathrm{cos}(\omega t+\varphi _a)`$ applied to the $`\alpha `$-th lead. This potential can be absorbed in the phase of the wave function. The single particle wave function in the presence of the perturbation is: $`\psi _{\alpha n}(\epsilon ,t)=\varphi _{\alpha n}(\epsilon ,t)\mathrm{exp}\{ieV_\alpha \mathrm{sin}(\omega t+\varphi _a)/\mathrm{}\omega )\}`$. Here $`\varphi _{\alpha n}(\epsilon ,t)`$ is the stationary wave function describing incoming (outgoing) carriers in the $`n`$-th transverse channel with energy $`\epsilon `$. This wave function can be expressed as a series in Bessel functions:
$$\psi _{\alpha n}(\epsilon ,t)=\varphi _{\alpha n}(\epsilon ,t)\underset{l}{}J_l\left(\frac{eV_\alpha }{\mathrm{}\omega }\right)e^{il(\omega t+\varphi _\alpha )}.$$
(3)
We see, that wave function $`\psi _{\alpha n}(\epsilon ,t)`$ in Eq. (3) has the same coordinate dependence as $`\varphi _{\alpha n}(\epsilon ,t)`$ but in energy space it is a superposition of sideband states with amplitudes $`J_l(eV_\alpha /\mathrm{}\omega )\mathrm{exp}(il\varphi _\alpha )`$ and energies $`\epsilon l\mathrm{}\omega `$.
We follow Refs. \[Hartree, ; pedersen, \] and assume that the oscillating potential exists in a region of the lead between the reservoir and the conductor. The potential vanishes as we approach the conductor. Matching wave functions in the regions with and without oscillating potential leads to
$`a_{\alpha n}(\epsilon )={\displaystyle \underset{l}{}}a_{\alpha n}^{}(\epsilon l\mathrm{}\omega )J_l\left({\displaystyle \frac{eV_\alpha }{\mathrm{}\omega }}\right)e^{il\varphi _\alpha }.`$ (4)
Here and in the following $`a_{\alpha n}(\epsilon )`$ are operators corresponding to annihilation of carrier incident on the conductor, $`a_{\alpha n}^{}(\epsilon l\mathrm{}\omega )`$ are operators corresponding to annihilation of carriers in the reservoir. This expression has an accuracy of order $`\mathrm{}\omega /\epsilon _F`$. We neglect corrections Hartree which arise from the difference of momenta of the particles with different sideband energies $`\epsilon `$ and $`\epsilon l\mathrm{}\omega `$. The reservoir is at thermal equilibrium and the statistical average of the operators $`(𝐚_{}^{}{}_{\alpha }{}^{})^{}`$ and $`𝐚_\beta ^{}`$ corresponds to the Fermi distribution of electrons in the absence of the time-dependent voltage: $`(𝐚_{}^{}{}_{\alpha }{}^{})^{}(\epsilon _1)𝐚_\beta ^{}(\epsilon _2)=\delta _{\alpha \beta }\delta (\epsilon _1\epsilon _2)f_\alpha (\epsilon _1)`$.
Let us next discuss the noise. At zero frequency, the noise spectral density is defined as follows:
$`S_{\lambda \mu }`$ $`=`$ $`\underset{T\mathrm{}}{lim}{\displaystyle \frac{1}{T}}{\displaystyle _0^T}𝑑t{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑\tau \mathrm{\Delta }I_\lambda (t+\tau )\mathrm{\Delta }I_\mu (t).`$ (5)
Here angular brackets denote the quantum mechanical and statistical average and $`\mathrm{\Delta }I_\lambda (t)=I_\lambda (t)I_\lambda (t)`$. For time independent voltages the current correlator depends only on the time difference, hence the integral over time $`t`$ in Eq. (5) is trivial and leads to the stationary state expression for shot noise.review1 In general, in the case of a time-dependent perturbation, the current correlator depends on both times $`t`$ and $`\tau `$. One can think about $`\tau `$ as the time difference and $`t`$ as the moment at which the measurement starts. In the steady state situation, a shift of the time $`t`$ by an integer number of periods changes nothing. This is why the averaging over time $`t`$ can be performed only over one period of the perturbation $`T_0=2\pi /\omega `$.
$`S_{\lambda \mu }={\displaystyle \frac{1}{T_0}}{\displaystyle _0^{T_0}}𝑑t{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑\tau \mathrm{\Delta }I_\lambda (t+\tau )\mathrm{\Delta }I_\mu (t).`$ (6)
Below, we will write this current noise in terms of pure electron and pure hole noise contributions and their correlations.
## III Electron-hole picture of noise
In the presence of ac-potentials it is appropriate to consider excitations away from the global equilibrium state and these excitations are electron-hole pairs. Our goal, motivated by the recent experiment of Reydellet *et al*. \[glattli, \], is to develop an electron-hole description of photon-assisted shot noise. The creation of an electron-hole pair is a correlated process. These correlations have been used in proposals for the generation and detection of orbital quasi-particle entanglement. sam1 ; bee1 ; sam2 Therefore the manifestations of electron-hole correlations in the noise properties are of particular interest.
To distinguish between different contributions to the shot noise we express the incident states in terms of electron and hole operators,
$`𝐞_\alpha (\epsilon )=𝐚_\alpha (\epsilon )\theta _e(\epsilon ),𝐡_\alpha (\epsilon )=𝐚_\alpha ^{}(\epsilon )\theta _h(\epsilon ),`$ (7)
where $`\theta _e(\epsilon )=\theta (\epsilon )`$ and $`\theta _h(\epsilon )=\theta (\epsilon )`$ are step functions for electrons and holes. Note that $`𝐡_\alpha (\epsilon )`$ is a transposed vector (row). Equation (6) for the shot noise can be converted to energy space. Using the Fourier transform of the current operator $`I_\lambda (t)=𝑑\epsilon 𝑑\mathrm{\Omega }\mathrm{exp}(i\mathrm{\Omega }t)I_\lambda (\epsilon +\mathrm{\Omega },\epsilon )`$ and recalling the fact that the condition (4) implies $`\epsilon _1\epsilon _2=l\mathrm{}\omega `$ for the current operators we obtain the following result for the zero-frequency shot noise:
$`S_{\lambda \mu }`$ $`=`$ $`2\pi \mathrm{}{\displaystyle 𝑑\epsilon _1𝑑\epsilon _2\mathrm{\Delta }I_\lambda (\epsilon _1)\mathrm{\Delta }I_\mu (\epsilon _2)}.`$ (8)
The current operator $`I_\lambda (\epsilon )I_\lambda (\epsilon ,\epsilon )`$ at equal energies can be represented as a sum of electron and hole currents $`I_\lambda (\epsilon )=I_\lambda ^e(\epsilon )+I_\lambda ^h(\epsilon )`$. Using Eq. (7) and Fourier transform of Eq. (1) we obtain:
$`I_\lambda ^e(\epsilon )`$ $`=`$ $`{\displaystyle \frac{e}{2\pi \mathrm{}}}{\displaystyle \underset{\alpha \beta }{}}𝐞_\alpha ^{}(\epsilon )𝐀_{\alpha \beta }(\lambda ,\epsilon )𝐞_\beta (\epsilon ),`$ (9)
$`I_\lambda ^h(\epsilon )`$ $`=`$ $`{\displaystyle \frac{e}{2\pi \mathrm{}}}{\displaystyle \underset{\alpha \beta }{}}𝐡_\alpha ^{}(\epsilon )𝐀_{\alpha \beta }^T(\lambda ,\epsilon )𝐡_\beta ^T(\epsilon ),`$ (10)
where the electron charge is $`e=|e|`$. It is now interesting to calculate explicitly the correlations between the same or different types of particles:
$`S_{\lambda \mu }`$ $`=`$ $`{\displaystyle \underset{ij=eh}{}}S_{\lambda \mu }^{ij}=2\pi \mathrm{}{\displaystyle \underset{ij=eh}{}}{\displaystyle 𝑑\epsilon _1𝑑\epsilon _2\mathrm{\Delta }I_\lambda ^i(\epsilon _1)\mathrm{\Delta }I_\mu ^j(\epsilon _2)}.`$
A non-zero answer for $`S_{\lambda \mu }^{eh}`$ will show us the presence of intrinsic correlations between electrons and holes. Using the expressions for the currents (9), (10) and Eqs. (4) and (7) we find the correlations of the electron and the hole currents:
$`S_{\lambda \mu }^{ij}`$ $`=`$ $`{\displaystyle \frac{e^2}{2\pi \mathrm{}}}{\displaystyle \underset{klm\alpha \beta }{}}{\displaystyle 𝑑\epsilon _1𝑑\epsilon _2\delta (\epsilon _1\epsilon _2+m\mathrm{}\omega )}`$ (11)
$`\times `$ $`J_k\left({\displaystyle \frac{eV_\alpha }{\mathrm{}\omega }}\right)J_{k+m}\left({\displaystyle \frac{eV_\alpha }{\mathrm{}\omega }}\right)J_{l+m}\left({\displaystyle \frac{eV_\beta }{\mathrm{}\omega }}\right)J_l\left({\displaystyle \frac{eV_\beta }{\mathrm{}\omega }}\right)`$
$`\times `$ $`e^{im(\varphi _\beta \varphi _\alpha )}f_\alpha (\epsilon _2k\mathrm{}\omega )(1f_\beta (\epsilon _2l\mathrm{}\omega ))`$
$`\times `$ $`\text{tr}\left(𝐀_{\alpha \beta }(\lambda ,\epsilon _1)𝐀_{\beta \alpha }(\mu ,\epsilon _2)\right)\theta _i(\epsilon _1)\theta _j(\epsilon _2).`$
Here $`i=e,h`$ stands for electron or hole. For example, in the limit of in-phase applied voltages the sum of Eq. (11) over electron and hole indices gives the result which for a two-probe sample with energy-independent scattering matrix can be found in Ref. \[LL, \] and for a multi-probe conductor is given in Ref. \[pedersen, \].
To probe correlations of electron-hole pairs we impose two conditions. First, the ac-potentials are taken to be weak enough in order to exclude processes of multiple absorbtion or emission of photons which implies $`eV_\alpha \mathrm{}\omega `$. As a consequence, electron-hole pairs are generated one by one. In this case one can prove that the many body wave function of the system will correspond to incoming single electron-hole pairs (see \[sam3, ; sam1, \] and Appendix A). Second, electrons and holes have different energies (the typical energy scale of an electron-hole pair is $`\mathrm{}\omega `$). We neglect the energy dependence of scattering matrices on the scale of $`\mathrm{}\omega `$. The energy dependence of the scattering matrix starts to play a role when the frequency becomes comparable to the inverse of the dwell time $`\tau _\mathrm{d}`$ such that $`\omega \tau _\mathrm{d}1`$. When screening is taken into account the frequency must even be comparable to the inverse charge relaxation time. poli As a consequence of these conditions, the correlations of $`ee`$ and $`hh`$, as well as $`eh`$ and $`he`$ correlations, are identical. Neglecting corrections of the order of $`(eV/\mathrm{}\omega )^4`$, we find the correlations between electron and hole currents:
$`S_{\lambda \mu }^{ee}`$ $`=`$ $`{\displaystyle \frac{e^2\omega }{2\pi }}\text{tr}𝐀(\mu )𝒫𝐀(\lambda ),`$ (12)
$`S_{\lambda \mu }^{eh}`$ $`=`$ $`{\displaystyle \frac{e^2\omega }{2\pi }}\text{tr}𝐀(\mu )\sqrt{𝒫}e^{i\varphi }𝐀(\lambda )\sqrt{𝒫}e^{i\varphi }.`$ (13)
Similarly to the experimental workglattli we introduce the probability to create an electron-hole pair in the $`\alpha `$-th channel, $`𝒫_\alpha =(eV_\alpha )^2/(2\mathrm{}\omega )^2`$ and diagonal matrices $`𝒫=\text{ diag}(𝒫_1,\mathrm{},𝒫_N)`$ and $`\varphi =\text{diag }(\varphi _1,\mathrm{},\varphi _N)`$ are defined by the amplitudes and phases of applied voltages. We next consider a number of different set-ups.
Consider a multi-lead conductor and consider an alternating voltage applied to only one lead, say lead $`1`$. The probability to create an electron-hole pair is $`𝒫_1=(eV_1/2\mathrm{}\omega )^2`$. All other leads are grounded. The auto-correlations and cross-correlations are
$`S_{\lambda \mu }^{ee}`$ $`=`$ $`{\displaystyle \frac{e^2\omega }{2\pi }}𝒫_1\text{tr}𝐀(\mu )𝟙_\mathrm{𝟙}𝐀(\lambda ),`$ (14)
$`S_{\lambda \mu }^{eh}`$ $`=`$ $`{\displaystyle \frac{e^2\omega }{2\pi }}𝒫_1\text{tr}𝐀(\mu )𝟙_\mathrm{𝟙}𝐀(\lambda )𝟙_\mathrm{𝟙}.`$ (15)
Next consider the 2-terminal case. Charge current conservation and particle (electron and hole) current conservation imply that the auto-correlations and the cross-correlations are equal in magnitude and differ only by a sign. Thus it is sufficient to give the auto-correlations,
$`S_{11}^{ee}`$ $`=`$ $`{\displaystyle \frac{e^2\omega }{2\pi }}𝒫_1{\displaystyle \underset{n}{}}𝒯_n,S_{11}^{eh}={\displaystyle \frac{e^2\omega }{2\pi }}𝒫_1{\displaystyle \underset{n}{}}𝒯_n^2.`$ (16)
Here $`𝒯_n`$ is transmission probability of the $`nth`$ eigen-channel, *i.e.* an eigenvalue of $`𝒮^{}𝟙_\mathrm{𝟚}𝒮𝟙_\mathrm{𝟙}`$. Summing up all four terms gives a shot noise of the total measured charge current proportional to $`_n𝒯_n(1𝒯_n)`$. However, unlike the case of a dc-biased conductor, where the $`𝒯_n(1𝒯_n)`$ is proportional to the quantum partition noise of a *fully filled* incident channel,lesovik ; buttiker90 here $`𝒯_n(1𝒯_n)`$ has a completely different origin. The auto-correlations of the electron $`S_{11}^{ee}`$ and hole noise $`S_{11}^{hh}`$ are simply proportional to $`𝒯_n`$ reflecting Poissonian shot noise of a nearly *empty channel* of incident particles. Electron and hole particle currents are correlated and it is this two particle correlation which is proportional to $`𝒯_n^2`$. The existence of electron-hole correlations can also be proved using a simple probability theory argument, which is demonstrated in Appendix B.
In order to make a direct measurement of electron-hole correlations we now analyze cross-correlations in three terminal mesoscopic conductors (see Fig. 1). For a weak ac-perturbation electron-hole pairs are rarely injected into the mesoscopic conductor. If we measure the current-correlation at the grounded leads, then if electron and hole exit through the same lead or if one of the particles of the pair is reflected back then their contribution to the cross-correlation is zero. Thus current cross-correlations are determined only by electron-hole pairs for which one of the particles exits through contact $`2`$ and the other through contact $`3`$ (see Fig. 1). In fact a cross-correlation measurement is a coincidence measurement run over long times. sam1
For a three terminal conductor we find for the particle cross-correlations,
$`S_{23}^{eh}={\displaystyle \frac{e^2\omega }{2\pi }}𝒫_1{\displaystyle \underset{n}{}}{\displaystyle \frac{𝒯_{21n}𝒯_{31n}}{2}},S_{23}^{ee}=0.`$ (17)
Here $`𝒯_{21n}`$ and $`𝒯_{31n}`$ are transmission eigenvalues of $`𝒮^{}𝟙_\mathrm{𝟚}𝒮𝟙_\mathrm{𝟙}`$ and $`𝒮^{}𝟙_\mathrm{𝟛}𝒮𝟙_\mathrm{𝟙}`$. This demonstrates that the current-cross-correlations are a direct measure of electron-hole correlations.
The considerations made above are valid at zero-temperature. It is thus important to consider the effect of thermal noise. After all, thermal noise can be viewed as another mechanism which generates electron-hole pairs. It should not matter how exactly electron-hole pairs are created. From the above consideration we see that electron-hole correlations are of second order in the transmission probability both for auto-correlations and for cross-correlations. We know from FDT that equilibrium noise is proportional to conductance and hence it is proportional to transmission probability. Consequently we do not expect any electron-hole correlations in thermal equilibrium.
Equilibrium noise is given by Eq. (11); only the terms $`k=l=m=0`$ contribute at $`V=0`$. In Eq. (11) the expression $`\theta _e(\epsilon _1)\theta _h(\epsilon _2)\delta (\epsilon _1\epsilon _2)0`$, so the only contribution to the equilibrium noise comes from electron-electron and hole-hole correlations, both are proportional to the dc-conductance matrix $`G_{\lambda \mu }=(e^2/h)(N_\lambda \delta _{\lambda \mu }\text{tr}𝒮^{}𝟙_\lambda 𝒮𝟙_\mu )`$:
$`S_{\lambda \mu }^{ee}`$ $`=`$ $`2k_\mathrm{B}TG_{\lambda \mu },S_{\lambda \mu }^{eh}=0.`$ (18)
Notice that a system at thermal equilibrium does not exhibit electron-hole correlations, as expected, whereas electron-electron and hole-hole correlations do of course exist. Also Eq. (18) shows that the thermal contribution to the noise can be neglected if $`k_\mathrm{B}T(eV)^2/(\mathrm{}\omega )`$.
## IV Hanbury Brown Twiss phase
If currents are injected from two or more contacts there are contributions to current-cross-correlations $`S_{34}`$ which can not be expressed in terms of transmission probabilities. buttiker91 The cross-correlations also depend on terms which contain products of four scattering matrices of the type buttiker90 ; buttiker91 ; VLB $`𝒮_{13}^{}𝒮_{32}𝒮_{24}^{}𝒮_{41}`$. In such a product none of the scattering matrices is the hermitian conjugate of the other. As a consequence the cross-correlation depends on the relative phase of scattering matrix elements. The physical origin of these terms is the quantum mechanical indistinguishability of particles. A particle from source contact $`1`$ can be transmitted to either $`3`$ or $`4`$ and is indistinguishable form a particle injected through contact $`2`$ (see Fig 2). The quantum state is given explicitly in Appendix A.
This exchange interference effect reflects the fact that two particles can be simultaneously in the conductor. It has long been proposed that this can be used to generate a two-particle Aharonov-Bohm effect which exists only in the current correlations while under the same conditions conductance exhibits no Aharonov-Bohm effect at all. buttiker91 Recently a geometry which demonstrates this explicitly has been proposed and analyzed. sam2 The magnitude of the current cross-correlations is shown to depend on the phase
$$\chi =\mathrm{arg}\left(𝒮_{13}^{}𝒮_{32}𝒮_{24}^{}𝒮_{41}\right),$$
(19)
which we call Hanbury Brown Twiss (HBT) phase.
For simplicity, let us consider a four terminal mesoscopic conductor with one mode contacts. AC-voltages of equal magnitude are applied to leads $`1`$ and $`2`$ and generate electron-hole pairs with probability $`𝒫=(eV/2\mathrm{}\omega )^2`$. The phase difference between the oscillating potentials $`V_1(t)`$ and $`V_2(t)`$ is $`\mathrm{\Delta }\varphi `$. Using (12) we find the charge current correlation at contact $`3`$ and $`4`$:
$`S(\mathrm{\Delta }\varphi )={\displaystyle \frac{e^2\omega }{2\pi }}𝒫\left|𝒮_{13}^{}𝒮_{41}e^{i\mathrm{\Delta }\varphi }+𝒮_{23}^{}𝒮_{42}\right|^2.`$ (20)
Once created in the lead $`1`$ or $`2`$ an electron-hole pair contributes to the shot noise only if it is split to the leads $`3`$ and $`4`$. The relative phase of pairs injected from $`1`$ and $`2`$ is $`\mathrm{\Delta }\varphi `$. Indistinguishability of particles emitted from the sources $`1`$ and $`2`$ leads to the phase dependence of the shot noise. From Eq. (20) we calculate the phase shift $`\mathrm{\Delta }\varphi `$ which corresponds to the maximum or minimum in the shot noise, $`\varphi _\pm `$ and the extremal values of the cross-correlations $`S^\pm `$:
$`\mathrm{\Delta }\varphi _+`$ $`=`$ $`\mathrm{arg}\left(𝒮_{13}^{}𝒮_{32}𝒮_{24}^{}𝒮_{41}\right),\mathrm{\Delta }\varphi _{}=\mathrm{\Delta }\varphi _++\pi ,`$ (21)
$`S^\pm `$ $`=`$ $`{\displaystyle \frac{e^2\omega }{2\pi }}𝒫\left(|𝒮_{13}^{}||𝒮_{41}|\pm |𝒮_{23}^{}||𝒮_{42}|\right)^2.`$ (22)
Notice, that when the phase shift becomes equal to the HBT-phase of Eq. (19), $`\mathrm{\Delta }\varphi =\chi `$, the shot noise reaches its maximum value $`S^+`$. For a dc-biased 4 terminal conductor one can extract information about $`Re(𝒮_{13}^{}𝒮_{32}𝒮_{24}^{}𝒮_{41})`$ using results of three different measurements.buttiker91 ; VLB In contrast, the ac-perturbation gives an additional degree of freedom to vary the phase shift between the voltages. Periodic dependence of the noise on the phase shift allows us to reproduce the whole function $`S(\mathrm{\Delta }\varphi )`$ using measurements at three different values of the phase shift $`\mathrm{\Delta }\varphi `$. Maximal $`S^+`$ and minimal $`S^{}`$ values of the shot noise, as well as the corresponding phase shifts $`\mathrm{\Delta }\varphi _\pm `$, yield not only the real part of $`𝒮_{13}^{}𝒮_{32}𝒮_{24}^{}𝒮_{41}`$ but also the full exchange interference correlation, *i.e.* its imaginary part.
We next investigate the statistical properties of the HBT-phase of four-terminal conductors. First, it is useful to mention that especially simple structures like a Mach-Zehnder (MZ) interferometer which was recently realized in a 2DEG mz exhibit only a trivial HBT-phase. A Mach-Zehnder interferometer has the property that there exists only forward scattering, *i.e.* every incoming particle is transmitted in one of two output arms. As a consequence the transmission sub-matrix is a unitary matrix itself. It follows that $`𝒮_{41}𝒮_{13}^{}+𝒮_{42}𝒮_{23}^{}=0`$. Here $`1`$ and $`2`$ are contacts under ac-excitation and $`3`$ and $`4`$ are measurement contacts. As a consequence the HBT-phase is given by $`\chi =\pi `$. In particular there are no correlations if two in-phase voltages are applied to two input contacts of a Mach-Zehder interferometer.
We now investigate the statistical properties of current cross-correlations and HBT-phase in the chaotic quantum dots, connected to four single channel leads. Chaotic scattering inside the dot leads to substantial back-scattering and we expect therefore nontrivial HBT-phase behavior. Scattering properties of an open quantum dot are very sensitive to external conditions such as shape (which can be changed by gate voltages), impurity distribution or applied magnetic fields. This allows one to explore the total ensemble of quantum dots of a proper symmetry (presence or absence of time-reversal symmetry, TRS). Current cross-correlations and HBT-phase are complicated functions of transmission amplitudes and phases of the scattering matrix elements, so the distributions can be obtained only by numerical integration.
Using convenient ’polar decomposition’ of the matrix $`𝒮`$:
$`𝒮=\left(\begin{array}{cc}u^{}& 0\\ 0& v\end{array}\right)\left(\begin{array}{cc}\sqrt{1𝒯}& i\sqrt{𝒯}\\ i\sqrt{𝒯}& \sqrt{1𝒯}\end{array}\right)\left(\begin{array}{cc}u& 0\\ 0& v^{}\end{array}\right),`$ (29)
where $`u,v,u^{},v^{}`$ are unitary $`2\times 2`$ block matrices ($`u^{}=u^\mathrm{T},v^{}=v^\mathrm{T}`$ in time-reversal symmetric case) and $`𝒯=\text{diag}(T_1,T_2)`$ with $`T_i[0,1]`$ distributed according to the relevant symmetry class $`\beta =1(2)`$ with (without of) the TRS,BeeReview we express the noise correlation $`S`$ of Eq. (20) in the form,
$`S(\mathrm{\Delta }\varphi )={\displaystyle \frac{e^2\omega }{2\pi }}𝒫|(v\sqrt{𝒯}u\mathrm{exp}(i\sigma _z{\displaystyle \frac{\mathrm{\Delta }\varphi }{2}})u^{}\sqrt{𝒯}v^{})_{12}|^2.`$ (30)
Numerical integration over random matrix ensembles gives the mesoscopic distribution $`P_\beta (\chi )`$, $`\beta =1,2`$ of the HBT-phase Fig. 3 presents the results for these distributions $`P_\beta (\chi )`$ in the range $`0\chi \pi `$, since they are symmetric with respect to $`\chi \chi `$. The distribution $`P_1(\chi )`$ has peaks at $`\chi =0,\pi `$, and the distribution $`P_2(\chi )`$ is monotonic for $`0\chi \pi `$. Below we present a qualitative explanation of these differences.
The matrices $`u,v`$ are randomly distributed over the unitary group, so the distributions $`P_\beta (\chi )`$ differ only because of different joint distributions of the transmission eigenvalues, $`P_\beta (T_1,T_2)`$. If one analyzes the distribution $`P_\beta (\chi )`$, two simple limits are readily calculated. First, take the limit $`T_1=T_2`$ (unreachable for chaotic quantum dots, since the joint distribution is $`P_\beta (T_1,T_2)|T_1T_2|^\beta (T_1T_2)^{1+\beta /2}`$, but still instructive). It is easy to see that in this case $`\chi =\pi `$. One would expect that for sufficiently close $`T_1,T_2`$ the phase $`\chi `$ does not differ much from $`\chi \pi `$ (for the Mach-Zender interferometer $`T_1=T_2=1`$ and the HBT-phase $`\chi `$ is locked at $`\pi `$). The relative statistical weight of this limit in the mesoscopic distribution $`P_\beta (\chi )`$ might be characterized by the average of $`\sqrt{T_1T_2}/|T_1T_2|`$, this average is 3/4 for $`\beta =1`$ and 4/5 at $`\beta =2`$. This could explain why $`P_1(\pi )`$ is slightly smaller then $`P_2(\pi )`$.
Another example which provides some understanding is the case of $`T_1=0`$ and arbitrary $`T_2`$. At $`\mathrm{\Delta }\varphi =0`$ the noise $`S`$ reaches its maximal value $`|S^+|=e^2\omega 𝒫/\pi `$, so that $`\chi =0`$. The statistical weight of such cases could be characterized by the average of $`\sqrt{T_1/T_2}`$ which is divergent for $`\beta =1`$ and finite for $`\beta =2`$. Thus for $`\beta =1`$ a peak at $`\chi =0`$ could be expected.
The monotonic almost uniform $`P_2(\chi )`$ is hard to explain. We expect that in the multi-channel limit $`N\mathrm{}`$, when the transmission value distribution reaches its symmetry-insensitive shape $`P(T)1/\sqrt{T(1T)}`$, the differences between $`\beta =1,2`$ are washed out and the distributions $`P_{1(2)}(\chi )`$ become uniform. Possibly, the distribution $`P_2(\chi )`$ is much more uniform then $`P_1(\chi )`$ in the four-mode quantum dot because the $`𝒮`$ matrix is characterized by a much larger number of independent variables (16 for $`\beta =2`$ vs. 10 for $`\beta =1`$). The complicated behavior of the distribution of the HBT-phase $`P_\beta (\chi )`$ can be contrasted with that of conductance or concurrence,bk which depend only on transmission eigenvalues of channels.
We next consider the mesoscopic distribution of the maximal (minimal) values of the noise correlations $`S`$, which could be reached in an experiment by tuning the phase shift $`\mathrm{\Delta }\varphi `$ between applied voltages. Using numerical integration, we find a distribution of $`S^+`$ (main figure in Fig. 4) and $`S^{}`$ (shown in the inset), normalized by a factor $`e^2\omega 𝒫/2\pi `$. From the numerical integration we conclude that the distribution of the minimal values $`S^{}`$ diverges at small arguments as $`P(S^{})(S^{})^{1/2}`$ for both $`\beta =1,2`$. The monotonic distributions of $`S^{}`$ displayed in the inset quickly decay, and the averaged values of $`S_\beta ^{}`$ are $`S_1^{}=0.022,S_2^{}=0.027`$. The distributions of $`S^+`$ are broad, as expected from mesoscopic distributions in few-channel systems, and the averaged values are also comparable, $`S_1^+=0.121,S_2^+=0.173`$.
## V Conclusions
We constructed the scattering theory for electron-hole transport in mesoscopic systems photon-excited at contacts. In the limit of a weak ac-excitation we investigate the correlation between electrons and holes. In different geometries signatures of such correlations appear in a different way. In a two terminal mesoscopic conductor subjected to a weak ac-voltage at one of the contacts, the electron-hole correlation effect in the shot noise co-exists with electron-electron correlations. In a three terminal geometry we investigate the correlations between currents at different terminals and find that they are pure electron-hole correlations.
In contrast to dc-biased systems, in the ac-biased systems investigated here we can vary not only the magnitude of the applied voltages but also the phases of the applied voltages. This provides additional controllable parameters. Shot noise measurements in a four terminal mesoscopic conductor provide information about two particle exchange interference in the sample. The shot noise depends on the relative phase between applied voltages, because the particles from different sources are indistinguishable.
We illustrate our theory on the example of a four probe chaotic dot coupled to four single channel leads. At two leads the conductor is subject to ac-voltages. When the phase shift $`\mathrm{\Delta }\varphi `$ of the applied voltages coincides with the HBT-phase $`\chi `$ of the sample, the correlations reach their maximal value $`S^+`$. One might expect the phase to be uniformly distributed. However the quantum dot coupled to single channel leads exhibits a strongly non-uniform mesoscopic distributions $`P(\chi )`$ of the HBT-phase and of $`P(S^\pm )`$ of the extremal correlations values $`S^\pm `$. Recent advances in high-frequency measurement techniquesglattli ; schoel ; kozh will make it possible to measure these distributions. The close link between the two-particle HBT-effect sam2 and quasi-particle entanglement bee1 ; sam2 and recent proposals for dynamic generation of quasi-particle entanglement sam3 ; bee2 ; llb make such experiments highly desirable.
## ACKNOWLEGEMENTS
This work was supported by the Swiss National Science Foundation and by the M. Curie RTN on ”Fundamentals of Nanoelectronics”.
## Appendix A Quantum state and Exchange Interference
In this appendix we construct the quantum state which leads to the Hanbury Brown Twiss exchange interference effect discussed in Sec. IV. We consider the conductor shown in Fig. 2 with four one channel leads. In particular we want to show that only electron hole pairs which are created in either lead $`1`$ or $`2`$ and are subsequently split into lead $`3`$ or $`4`$ generate the HBT-exchange interference term.
Each state incident from a reservoir subject to an oscillating voltage can gain or loose modulation energy quanta. The many particle state incident from the $`\alpha `$-th reservoir, $`_{ϵ<0}a_\alpha ^{}(ϵ)|`$, can be transformed using Eq. (4) into the state $`|in`$ incident on the mesoscopic conductor. $`|`$ is the true vacuum. Taking only the first sidebands into account, we find
$$|in=\underset{E<0}{}(a_\alpha ^{}(ϵ)+V_\alpha ^+a_\alpha ^{}(ϵ^+)+V_\alpha ^{}a_\alpha ^{}(ϵ^{}))|.$$
(31)
Here $`V_\alpha ^\pm =eV_\alpha e^{\pm i\varphi _\alpha }/2\mathrm{}\omega `$ is the probability amplitude for the creation of an electron-hole pair in the $`\alpha `$-th lead and $`ϵ^\pm =ϵ\pm \mathrm{}\omega `$. Using the commutation rules of free fermion operators we find the state incident on the four-terminal mesoscopic conductor in the presence of oscillating potentials at contacts $`1`$ and $`2`$:
$`|in=|0+{\displaystyle _0^\mathrm{}\omega }𝑑ϵa^{}(ϵ)\widehat{V}a(ϵ^{})|0.`$ (32)
Here the vacuum state $`|0=_{(ϵ<0,\alpha =\overline{1,4})}a_\alpha ^{}(ϵ)|`$ is the filled Fermi sea at equilibrium in all four leads. The second term on the r.h.s. of Eq. (32) represents the superposition of the electron-hole pairs coming from leads $`1`$ and $`2`$. The matrix $`\widehat{V}`$ is diagonal $`\widehat{V}=\text{diag}(eV_1e^{i\mathrm{\Delta }\varphi }/2\mathrm{}\omega ,eV_2/2\mathrm{}\omega ,0,0)`$. Using the relation between incoming and outgoing states $`a_\alpha =_\beta 𝒮_{\alpha \beta }^{}b_\beta `$ we find for the outgoing state:
$`|out=|\overline{0}+{\displaystyle _0^\mathrm{}\omega }𝑑ϵb^{}(ϵ)𝒮\widehat{V}𝒮^{}b(ϵ^{})|\overline{0}.`$ (33)
Here the new vacuum $`|\overline{0}=_{(ϵ<0,\alpha =\overline{1,4})}det(𝒮)b_\alpha ^{}(ϵ)|`$ describes the equilibrium state in the basis of out-going states.
The many-particle state (33) contains information about the final states in all the leads. We are specifically interested in the correlation of currents at leads $`3`$ and $`4`$ for a conductor subject to oscillating potentials at contacts $`1`$ and $`2`$. Leads $`3`$ and $`4`$ are grounded and at zero temperature there are no particles injected into the mesoscopic conductor through these leads. Thus the cross-correlations $`S_{34}`$ are determined only by that portion of the state (33) which describes out-going particles in lead $`3`$ and $`4`$,
$`|out_{34}`$ $`=`$ $`{\displaystyle \frac{eV}{2\mathrm{}\omega }}{\displaystyle _0^\mathrm{}\omega }𝑑ϵ\left[\mathrm{}\right]|\overline{0},`$ (34)
$`\left[\mathrm{}\right]`$ $`=`$ $`\left(e^{i\mathrm{\Delta }\varphi }𝒮_{41}𝒮_{13}^{}+𝒮_{42}𝒮_{23}^{}\right)b_4^{}(ϵ)b_3(ϵ^{})`$ (35)
$`+`$ $`\left(e^{i\mathrm{\Delta }\varphi }𝒮_{31}𝒮_{14}^{}+𝒮_{32}𝒮_{24}^{}\right)b_3^{}(ϵ)b_4(ϵ^{}).`$
An excitation with energy above zero corresponds to an electron $`b_\alpha ^{}(ϵ)=e_\alpha ^{}(ϵ)`$ and an excitation with energy below zero to a hole $`b_\alpha (ϵ^{})=h_\alpha ^{}(ϵ^{})`$. Thus the first term in Eq. (34) represents a superposition of amplitudes for an electron-hole pair created in contact $`1`$ or contact $`2`$ with the electron leaving through $`4`$ and the hole through contact $`3`$. It is an orbitally entangled electron-hole pair state and the index of the source contact $`1`$ and $`2`$ is a pseudo-spin index. Similarly the second term represents orbitally entangled electron-hole pairs with the electron leaving through contact $`3`$ and the electron through contact $`4`$. In fact, from a formal point of view the state is identical to the one created by two oscillating potentials acting in spatially separated interior regions of a conductor investigated in Ref. \[sam3, \]. One might think that the oscillating potentials in contacts $`1`$ and $`2`$ serve to mark particles created in these contacts, since they carry the phase factor $`\mathrm{exp}(i\varphi _1)`$ and $`\mathrm{exp}(i\varphi _2)`$ and thus to make them distinguishable. However, also for the applied oscillating voltages it is only the phase difference $`\mathrm{\Delta }\varphi =\varphi _1\varphi _2`$ that counts. As a consequence, the phase of the oscillating voltages only modulates the HBT-interference but does not destroy it.
We now show how the different terms in the quantum state Eq. (34) contribute to the HBT-current-correlation between leads $`3`$ and $`4`$. This correlation can now be derived by using either the full state (33) or the state (34) which contains the out-going particles in leads $`3`$ and $`4`$ only. The results of these two calculations are of course identical. Using Eq. (8) we find:
$`S_{34}`$ $`=`$ $`2\pi \mathrm{}{\displaystyle 𝑑\epsilon 𝑑\epsilon ^{}out|\mathrm{\Delta }I_3(\epsilon )\mathrm{\Delta }I_4(\epsilon ^{})|out}.`$
Since leads $`3`$ and $`4`$ are grounded and there are no particles injected into the mesoscopic conductor through these leads, the current operators in $`3`$ and $`4`$ can be written in terms of operators of the outgoing states only:
$$I_\alpha (\epsilon )=\frac{e}{2\pi \mathrm{}}b_\alpha ^{}(\epsilon )b_\alpha (\epsilon ).$$
For weak ac-potentials, the fact that average currents are equal to zero to linear order in the potentials, permits us to write:
$`S_{34}={\displaystyle \frac{e^2}{h}}{\displaystyle 𝑑\epsilon 𝑑\epsilon ^{}_0^\mathrm{}\omega 𝑑ϵ𝑑ϵ^{}\underset{\begin{array}{c}\alpha \beta \gamma \delta \end{array}}{}\left[𝒮\widehat{V}^{}𝒮^{}\right]_{\alpha \beta }\left[𝒮\widehat{V}𝒮^{}\right]_{\gamma \delta }\overline{0}|b_\alpha ^{}(ϵ^{})b_\beta (ϵ)b_3^{}(\epsilon )b_3(\epsilon )b_4^{}(\epsilon ^{})b_4(\epsilon ^{})b_\gamma ^{}(ϵ^{})b_\delta (ϵ^{})|\overline{0}}.`$ (36)
Since $`\mathrm{}\omega >ϵ>0`$, the operators $`b_\alpha ^{}(ϵ\mathrm{}\omega )`$ and $`b_\beta (ϵ)`$ acting to the right on the vacuum give zero. Applying Wick’s theorem to the quantum mechanical average in Eq. (36) and performing the integration over the energies we find:
$`S_{34}`$ $`=`$ $`{\displaystyle \frac{e^2\omega }{2\pi }}\left|\left(𝒮\widehat{V}𝒮^{}\right)_{34}\right|^2.`$ (37)
This expression coincides with Eq. (20). The diagonal matrix $`\widehat{V}`$ provides additional selection rules for the processes which contribute to the shot noise. Denoting the part of $`\widehat{V}`$ in lead $`\alpha `$ by $`\widehat{V}_\alpha `$ we find from Eq. (37)
$`\left|\left(𝒮\widehat{V}𝒮^{}\right)_{34}\right|^2`$ $`=`$ $`\left(𝒮_{31}\widehat{V}_1^{}𝒮_{14}^{}+𝒮_{32}\widehat{V}_2^{}𝒮_{24}^{}\right)`$ (38)
$`\times \left(𝒮_{41}\widehat{V}_1𝒮_{13}^{}+𝒮_{42}\widehat{V}_2𝒮_{23}^{}\right).`$
Thus we see that of the four terms in the current correlation function, two correspond to electron-hole emission out of the same contact, and two contributions, sensitive to the phase $`\mathrm{\Delta }\varphi `$, correspond to electron-hole emission from both contacts $`1`$ and $`2`$. All contributions arise due to electron-hole pairs which are split into contact $`3`$ and $`4`$.
## Appendix B Current correlations for Poissonian source of $`eh`$ pairs
Here we present a model of a probability game to illustrate correlations between photon-generated $`eh`$ pairs similarly to the probabilistic electron model of Ref. \[binomial, \]. We calculate correlations of electrons and holes for a two-terminal geometry, and show that the total noise can not be interpreted in terms of independent electron and hole noises.
Consider a two-lead geometry with one of the leads exposed to a monochromatic photon beam of frequency $`\omega `$. The photons generate $`eh`$ pairs, and the statistics of these pairs is assumed to be Poissonian. (The details of the photon statistics are quite irrelevant and we take this example as an illustration.) Let $`\lambda `$ be the average number of pairs generated in a period $`2\pi /\omega `$. The number $`N`$ of pairs generated during one period fluctuates according to the Poissonian distribution $`\rho _\lambda (N)=\lambda ^N\mathrm{exp}(\lambda )/N!`$. A barrier of transparency $`𝒯`$ separates the left and right contacts of the conductor. The probability that $`m`$ electrons and $`n`$ holes out of $`N`$ electron-hole pairs are transmitted from the left to the right contact is
$`P(m,n|N)=\left({\displaystyle \genfrac{}{}{0pt}{}{m}{N}}\right)\left({\displaystyle \genfrac{}{}{0pt}{}{n}{N}}\right)𝒯^{m+n}(1𝒯)^{2N(m+n)}.`$ (39)
The measured quantity is the charge current $`j=mn`$ (we omit electron charge $`e`$ for simplicity) and its distribution is symmetric with respect to $`j=0`$, so to be definite we consider $`j>0`$. The distribution $`P(j)`$ is then given by
$`P(j)={\displaystyle \underset{N=1}{\overset{\mathrm{}}{}}}{\displaystyle \underset{m=j}{\overset{N}{}}}P(m,mj|N)\rho _\lambda (N).`$ (40)
The limits of summations result from the fact that to create the current $`j`$ one needs at least $`j`$ electrons out of $`Nm`$ electron-hole pairs. For $`\lambda 1`$ which we assume from now on, the leading contribution to Eq. (40) is
$`P(j)`$ $``$ $`{\displaystyle \frac{(\lambda 𝒯(1𝒯))^j}{j!}},`$ (41)
and thus $`j^{2k+1}=0`$, $`j^{2k}2\lambda 𝒯(1𝒯)`$. We conclude that the noise equals $`2\lambda 𝒯(1𝒯)`$. From comparison with the scattering theory, summing up the Eqs. (16), we find $`j^2=(eV/2\mathrm{}\omega )^2𝒯(1𝒯)`$, which allows us to identify $`\lambda =(eV)^2/8(\mathrm{}\omega )^21`$.
If we now consider the distribution of the electron current $`j=j_e`$ and hole current $`j=j_h`$ separately, we find
$`P(j)`$ $`=`$ $`{\displaystyle \underset{N=j}{\overset{\mathrm{}}{}}}{\displaystyle \frac{N!\rho _\lambda (N)}{j!(Nj)!}}𝒯^j(1𝒯)^{Nj}=\rho _{\lambda 𝒯}(j).`$
The distribution of the current $`j`$ in the r.h.s. is a Poissonian distribution characterized by the parameter $`\lambda 𝒯`$. However, since now the total measured current is the difference between two independent Poissonian processes, it is not a Poissonian process itself (only for two independent Gaussian processes is their difference also a Gaussian process). For $`j=j_ej_h`$ we find
$`P(j)=I_{|j|}(2\lambda 𝒯)\mathrm{exp}(2\lambda 𝒯),`$ (42)
with $`I_{|j|}(x)`$ the modified Bessel function of the $`|j|`$th (integer) order. Thus we find $`j^2=2\lambda 𝒯+𝒪((\lambda 𝒯)^2)`$, unlike the correct result $`2\lambda 𝒯(1𝒯)`$ found from the distribution Eq. (41). |
warning/0507/quant-ph0507024.html | ar5iv | text | # Normal covariant quantization maps
## 1. Introduction
Quantization is a procedure which associates a quantum mechanical observable to a given classical dynamical variable, the latter being represented by a complex valued Borel function on the phase space $`X`$ of a classical system. The phase space $`X`$ can be taken to be $`^{2n}`$ or, more generally, $`G/H`$, where $`G`$ is a locally compact second countable topological group and $`H`$ a closed subgroup. We consider here only the case where $`X=G`$. Quantization can be realized e.g. by integrating the classical variable $`f`$ with respect to a suitable (positive normalized) operator measure $`E:(G)L()`$, where $`(G)`$ is the Borel $`\sigma `$-algebra of subsets of $`G`$ and $`L()`$ the set of bounded operators acting on the Hilbert space $``$ of the quantum system. The resulting operator integral $`L(f,E)=f𝑑E`$ is a (possibly unbounded) linear operator, which is symmetric if $`f`$ is real valued. (See Section 6 for our definition of the domain of the operator integral.) In many cases, the operator $`L(f,E)`$ is essentially selfadjoint, so that it is eligible to represent a quantum observable. The map $`fL(f,E)`$ is linear (in the sense made precise in Section 6). If $`f`$ is bounded, then $`L(f,E)L()`$. The operator integral has a convergence property, which could be called ”quasicontinuity” (see e.g. \[2, p. 22\]): If $`(f_n)`$ is an increasing sequence of positive Borel functions converging pointwise to a Borel function $`f`$, and $`\phi `$ is a vector belonging to the domains of $`L(f,E)`$ and each $`L(f_n,E)`$, then the sequence $`(\psi |L(f_n,E)\phi )`$ of numbers converges for each $`\psi `$ to $`\psi |L(f,E)\phi `$.
As noted above, quantization might be any mapping $`\mathrm{\Gamma }`$ from the set of Borel functions to the set of linear operators on $``$. It is therefore natural to ask which of them can be represented by operator integrals with respect to some positive operator measures. Essential requirements for $`\mathrm{\Gamma }`$ are linearity, positivity, the property that bounded functions are mapped to $`L()`$, and quasicontinuity, as they assure that the association $`B\mathrm{\Gamma }(\chi _B)`$ defines a positive operator measure $`E^\mathrm{\Gamma }`$. Obviously, this does not guarantee that the quantization map $`\mathrm{\Gamma }`$ would coincide with the map given by the operator integral with respect to $`E^\mathrm{\Gamma }`$; in particular, nothing has been said about the domains of the operators $`\mathrm{\Gamma }(f)`$. In the case of a bounded function $`f`$, however, the domain of the operator integral $`L(f,E^\mathrm{\Gamma })`$ is all of $``$, and it follows easily that $`\mathrm{\Gamma }(f)=L(f,E^\mathrm{\Gamma })`$. Thus, if we have a positive and quasicontinuous linear quantization map $`\mathrm{\Gamma }`$, which maps bounded functions to $`L()`$, then (at least) the restriction of $`\mathrm{\Gamma }`$ to the set of bounded functions can be represented as the operator integral $`L(,E^\mathrm{\Gamma })`$.
Since the phase space $`G`$ has a left Haar measure $`\lambda `$, it is convenient to consider the functions in $`L^{\mathrm{}}(G,\lambda )`$ (i.e. $`\lambda `$-equivalence classes of $`\lambda `$-essentially bounded $`\lambda `$-measurable complex functions) instead of bounded Borel functions. Assume that the original quantization map $`\mathrm{\Gamma }`$ (defined on all complex Borel functions) is linear, positive, has the quasicontinuity property, and maps bounded functions to $`L()`$. In addition, we can require that each complex measure $`BE_{\psi ,\phi }^\mathrm{\Gamma }(B)=\psi |E^\mathrm{\Gamma }(B)\phi `$ is $`\lambda `$-continuous. This ensures that $`\mathrm{\Gamma }(f)`$ does not depend on the (Borel) representative of $`fL^{\mathrm{}}(G,\lambda )`$, so we get a well-defined positive linear quantization map $`\stackrel{~}{\mathrm{\Gamma }}:L^{\mathrm{}}(G,\lambda )L()`$ which coincides with the map obtained from $`L(,E^\mathrm{\Gamma })`$ in the similar way.
When we restrict our attention to the positive linear quantization maps $`\mathrm{\Gamma }`$ defined on $`L^{\mathrm{}}(G,\lambda )`$ with values in $`L()`$, the condition of quasicontinuity is not appropriate, as it involves pointwise convergence. Instead, we require the somewhat similar condition of normality, i.e. weak-\* continuity associated with the dualities $`L^1(G,\lambda )^{}=L^{\mathrm{}}(G,\lambda )`$ and $`𝒯()^{}=L()`$, where $`L^1(G,\lambda )`$ is the set of $`\lambda `$-equivalence classes of $`\lambda `$-integrable complex functions and $`𝒯()`$ is the set of all trace class operators on $``$. Thus, if the $`\lambda `$-continuity of the complex measures $`E_{\psi ,\phi }^\mathrm{\Gamma }`$ is assumed, we have $`\mathrm{\Gamma }=L(,E^\mathrm{\Gamma })`$. Conversely, if a positive operator measure $`E`$ is given, for which each $`E_{\psi ,\phi }`$ is $`\lambda `$-continuous, then the map $`L^{\mathrm{}}(G,\lambda )fL(f,E)L()`$ is linear, positive, and normal (see Section 5).
An important property of a quantization map $`\mathrm{\Gamma }:L^{\mathrm{}}(G,\lambda )L()`$ (or the corresponding operator measure) is covariance (see Section 2), which connects it to the structure of the phase space. Covariance also conveniently implies the $`\lambda `$-continuity of the complex measures $`E_{\psi ,\phi }^\mathrm{\Gamma }`$ (see section 5). Covariant positive phase space operator measures have proved highly useful also in various other applications of quantum mechanics, including for instance the fundamental questions on joint measurements of position and momentum observables and the problem of quantum state estimation (quantum tomography). Consequently, the structure of such operator measures has been studied extensively: the canonical examples of the covariant phase space observables are constructed e.g. in , whereas a complete group theoretical characterization is given in .
The characterization of is based on a generalization of Mackey’s imprimitivity theorem . However, in the concrete case where the phase space is $`^{2n}`$, there is another, more direct (and completely different) approach, outlined by Holevo , and further elaborated by Werner . In fact, in , Werner characterizes all the positive normal phase space translation covariant maps $`\mathrm{\Gamma }:L^{\mathrm{}}(^{2n})L()`$. The essential part of both Holevo’s and Werner’s proofs relies on the fact that the Banach space of trace class operators on a separable Hilbert space has the Radon-Nikodým property.
In this paper we generalize Werner’s approach to the case where the phase space is a locally compact unimodular topological group, paying due attention to the details arising in this context. In addition, we consider briefly the question of quantization of unbounded functions.
## 2. Preliminaries
If $``$ is a Hilbert space, we let $`L()`$ and $`𝒯()`$ denote the sets of bounded operators and trace class operators on $``$, respectively.
Let $`\mu _L`$ denote the Lebesgue measure of $`^{2n}`$. Denote the Weyl operators on the Hilbert space $`L^2(^n)`$ by $`W(x)`$, $`x=(q,p)^{2n}`$, so that $`W(q,p)`$ acts according to
$$(W(q,p)\psi )(t)=e^{i\frac{1}{2}qp}e^{ipt}\psi (t+q).$$
They satisfy the relation
(1)
$$W(x)W(y)=e^{i\frac{1}{2}\{x,y\}}W(x+y),$$
where $`\{(q,p),(q^{},p^{})\}=qp^{}pq^{}`$ for all $`(q,p),(q^{},p^{})^{2n}`$.
For each $`x^{2n}`$, define $`\gamma (x):𝒯(L^2(^n))𝒯(L^2(^n))`$ by $`\gamma (x)(T)=W(x)TW(x)`$. Then the map $`x\gamma (x)`$ has the following well-known properties. The proof is included for the reader’s convenience.
###### Lemma 1.
* $`\gamma (x+y)=\gamma (x)\gamma (y)`$ for all $`x,y^{2n}`$.
* $`\gamma (x)^{}(A)=W(x)AW(x)`$ for all $`AL(L^2(^n))`$ and $`x^{2n}`$.
* $`\gamma (x)`$ is a positive trace-norm isometry for all $`x^{2n}`$.
* For each $`AL(L^2(^n))`$ and $`S𝒯(L^2(^n))`$, the function $`x\mathrm{Tr}[A\gamma (x)(S)]`$ is continuous.
* $`\mathrm{Tr}[P_1\gamma (x)(P_2)]𝑑\mu _L(x)=(2\pi )^n`$ for all one-dimensional projections $`P_1`$ and $`P_2`$ on $`L^2(^n)`$.
Proof. (a) is a direct consequence of the relation (1), and (b) follows from a basic property of the trace. If $`U`$ is a unitary operator, $`|USU^{}|=U|S|U^{}`$ for each $`SL(L^2(^n))`$. Therefore, since $`W(x)`$ is unitary and $`W(x)^{}=W(x)`$, $`\gamma (x)(S)_{\mathrm{tr}}=\mathrm{Tr}[|W(x)SW(x)|]=S_{\mathrm{tr}}`$ for each $`S𝒯(L^2(^n))`$. This proves (c), as it is clear that $`\gamma (x)`$ is positive. To prove (d), take $`AL(L^2(^n))`$ and $`S𝒯(L^2(^n))`$. Let $`x^{2n}`$, and $`(x_n)`$ be a sequence converging to $`x`$. Since $`xW(x)`$ is strongly continuous, $`\gamma (x_n)^{}(A)=W(x_n)AW(x_n)W(x)AW(x)=\gamma (x)^{}(A)`$ weakly. Since all $`W(x)`$ are unitary, the sequence $`(\gamma (x_n)^{}(A))`$ is norm bounded, from which it follows that it converges to $`\gamma (x)^{}(A)`$ also ultraweakly. Thus we get
$$\mathrm{Tr}[A\gamma (x_n)(S)]=\mathrm{Tr}[\gamma (x_n)^{}(A)S]\mathrm{Tr}[\gamma (x)^{}(A)S]=\mathrm{Tr}[A\gamma (x)(S)],$$
which proves (d). The proof of (e) goes as follows. Assume that $`P_1=|\psi \psi |`$ and $`P_2=|\phi \phi |`$, where $`\psi ,\phi `$ are unit vectors. Define the function $`\varphi _q`$ for each $`q^n`$ by $`\varphi _q(t)=\psi (t)\phi (t+q)`$. Then
$$1=\psi ^2\phi ^2=\left(|\psi (t)|^2|\phi (q)|^2𝑑q\right)𝑑t=\left(|\psi (t)|^2|\phi (t+q)|^2𝑑t\right)𝑑q$$
by the Fubini-Tonelli theorem, so that $`\varphi _q`$ for almost all $`q`$. By the unitarity of the inverse Fourier-Plancherel operator $`F`$, we have now
$$1=|(F\varphi _q)(p)|^2𝑑p𝑑q.$$
But since $`\psi `$ and $`\phi (+q)`$ are in $`L^2(^n)`$, $`\varphi _q`$ is also integrable, so
$$(F\varphi _q)(p)=\frac{1}{\sqrt{(2\pi )^n}}e^{ipt}\varphi _q(t)𝑑t=\frac{1}{\sqrt{(2\pi )^n}}e^{i\frac{1}{2}pq}\psi |W(q,p)\phi ,$$
from which it follows that
$$(2\pi )^n=|\psi |W(x)\phi |^2𝑑\mu _L(x)=\mathrm{Tr}[P_1\gamma (x)(P_2)]𝑑\mu _L(x).$$
$`\mathrm{}`$
Now we proceed to a more abstract case.
If $`(\mathrm{\Omega },𝒜,\nu )`$ is a $`\sigma `$-finite (positive) measure space, we let $`L^1(\mathrm{\Omega },\nu )`$ denote the Banach space of (equivalence classes of) complex valued, $`\nu `$-integrable functions, and $`L^{\mathrm{}}(\mathrm{\Omega },\nu )`$ the Banach space of (equivalence classes of) complex valued, $`\nu `$-measurable, $`\nu `$-essentially bounded functions.
A function $`g`$ defined on $`\mathrm{\Omega }`$ and having values in some Banach space is said to be $`\nu `$-measurable, if for each $`B𝒜`$ of finite measure there is a sequence of $`\nu `$-simple functions converging to $`\chi _Bg`$ in $`\nu `$-measure (or, equivalently, there is a sequence of $`\nu `$-simple functions which converges $`\nu `$-almost everywhere to $`\chi _Bg`$) \[8, pp. 106, 150\]. In the case where the value space of $`g`$ is separable (in particular, if $`g`$ is scalar-valued), $`\nu `$-measurability is equivalent to the measurability with respect to the Lebesgue extension of the $`\sigma `$-algebra $`𝒜`$ with respect to $`\nu `$ \[8, p. 148\]. If $`X`$ is a Banach space, $`\mathrm{Iso}(X)`$ denotes the group of linear homeomorphisms from $`X`$ onto $`X`$.
Let $``$ be a separable Hilbert space. Let $`\mathrm{Aut}(𝒯())`$ denote the subgroup of $`\mathrm{Iso}(𝒯())`$ consisting of the positive maps which preserve the trace norm. The set $`\mathrm{Aut}(𝒯())`$ is equipped with the weak topology given by the set of functionals $`u\mathrm{Tr}[Au(T)]`$, where $`AL()`$, $`T𝒯()`$. For $`u\mathrm{Aut}(𝒯())`$, the adjoint map $`u^{}:L()L()`$ restricted to $`𝒯()`$ is equal to $`u^1`$. It follows from the Wigner theorem that for each $`u\mathrm{Aut}(𝒯())`$ there is an either unitary or antiunitary operator $`U`$, such that $`u(T)=UTU^{}`$ for all $`T𝒯()`$.
Let $`G`$ be a locally compact unimodular second countable (Hausdorff) topological group, with Haar measure $`\lambda `$, such that there is a continuous group homomorphism $`\beta :G\mathrm{Aut}(𝒯())`$ and a constant $`d>0`$, satisfying
(2)
$$\mathrm{Tr}[P_1\beta (g)(P_2)]𝑑\lambda (g)=d$$
for all one-dimensional projections $`P_1`$ and $`P_2`$ on $``$. The system $`(G,\beta ,d)`$ will remain fixed throughout the paper.
Remark.
* It follows from Lemma 1 that the additive group $`^{2n}`$, with the homomorphism $`\gamma `$ and the constant $`(2\pi )^n`$ constitute an example of the abstract system $`(G,\beta ,d)`$.
* The fact that each $`\beta (g)`$ has the form $`\beta (g)(T)=U(g)TU(g)^{}`$ for some unitary or antiunitary operator $`U(g)`$ implies that, in the case where $`G`$ is connected, the map $`gU(g)`$ is a projective unitary representation of $`G`$ which satisfies the square integrability condition
$$|\psi |U(g)\phi |^2𝑑\lambda (g)=d$$
for all unit vectors $`\psi ,\phi `$. The theory of such representations and the associated covariant operator measures is well developed, see e.g. . It can be noted that in the case of a nonunimodular group, the square integrability condition is no longer of the above form for some fixed $`d`$ .
In this paper, however, we do not need the explicit structure of the map $`\beta `$ given by the projective representation $`gU(g)`$. Thus we will use only the abstract definition, with the map $`\gamma `$ associated with the group $`^{2n}`$ as a concrete example.
If $`S`$ is a positive trace class operator and $`A`$ a bounded positive operator, the function $`Gg\mathrm{Tr}[A\beta (g)(S)]`$ is positive. Concerning the integrability of such a function, the following lemma holds (with the understanding that $`\mathrm{}0=0`$):
###### Lemma 2.
Let $`S𝒯()`$ and $`AL()`$ be positive operators. Then
$$d^1\mathrm{Tr}[A\beta (g)(S)]𝑑\lambda (g)=\mathrm{Tr}[A]\mathrm{Tr}[S].$$
In particular, if $`SO`$, the function $`g\mathrm{Tr}[A\beta (g)(S)]`$ is integrable if and only if $`A𝒯()`$.
Proof. The proof consists of several stages.
* Assume that $`A`$ and $`S`$ are one-dimensional projections. Since now $`\mathrm{Tr}[A]\mathrm{Tr}[S]=1`$, it follows directly from the relation (2) that $`d^1\mathrm{Tr}[A\beta (g)(S)]𝑑\lambda (g)=\mathrm{Tr}[A]\mathrm{Tr}[S]`$.
* Assume that $`S`$ is a positive nonzero trace class operator and $`A`$ a one-dimensional projection. Then $`S=_{i=1}^{\mathrm{}}w_i|\phi _i\phi _i|`$, where $`(\phi _i)`$ is an orthonormal sequence in $``$, the series converging in the trace norm, and $`w_i0`$, $`_iw_i=\mathrm{Tr}[S]`$. Since the map $`T\mathrm{Tr}[A\beta (g)(T)]`$ is linear and trace-norm continuous, we have $`\mathrm{Tr}[A\beta (g)(S)]=_iw_i\mathrm{Tr}[A\beta (g)(|\phi _i\phi _i|)]`$ for all $`g`$. Now the result 1) and the monotone convergence theorem give
$`\mathrm{Tr}[A]\mathrm{Tr}[S]`$ $`=`$ $`{\displaystyle \underset{i}{}}w_i\mathrm{Tr}[A]\mathrm{Tr}[|\phi _i\phi _i|]={\displaystyle \underset{i}{}}w_id^1{\displaystyle \mathrm{Tr}[A\beta (g)(|\phi _i\phi _i|)]𝑑\lambda (g)}`$
$`=`$ $`d^1{\displaystyle \mathrm{Tr}[A\beta (g)(S)]𝑑\lambda (g)}.`$
* Let $`S`$ be as before, and $`A`$ a bounded positive operator such that the set $`\sigma _p(A)`$ of eigenvalues of $`A`$ equals either the spectrum $`\sigma (A)`$ or the set $`\sigma (A)\{0\}`$. (In particular, all the positive compact operators are like this.) Now $`E^A(\sigma _p(A))=I`$, where $`E^A`$ is the spectral measure of $`A`$. It follows that the eigenvectors of $`A`$ constitute an orthonormal basis of $``$. Since $``$ is separable, the set $`\sigma _p(A)`$ is at most countable. Let $`\sigma _p(A)=\{a_1,a_2,\mathrm{}\}`$, and $`(\phi _{nk})`$ be an orthonormal basis of $``$, such that for each $`n`$, the vectors $`\phi _{nk}`$ span the eigenspace of $`A`$ associated with the eigenvalue $`a_n`$. Now $`\mathrm{Tr}[A]=_{nk}\phi _{nk}|A\phi _{nk}=_na_nd_n`$, where $`d_n\mathrm{}`$ is the degree of the eigenvalue $`a_n`$. Moreover,
$$\mathrm{Tr}[A\beta (g)(S)]=\underset{nk}{}a_n\phi _{nk}|\beta (g)(S)\phi _{nk}=\underset{nk}{}a_n\mathrm{Tr}[|\phi _{nk}\phi _{nk}|\beta (g)(S)].$$
It now follows from 2) and the monotone convergence theorem that
$$d^1\mathrm{Tr}[A\beta (g)(S)]𝑑\lambda (g)=\underset{n}{}a_nd_n\mathrm{Tr}[S]=\mathrm{Tr}[A]\mathrm{Tr}[S].$$
In particular, if $`A`$ has an eigenspace of infinite dimension corresponding to a nonzero eigenvalue, then $`\mathrm{Tr}[A\beta (g)(S)]𝑑\lambda (g)=\mathrm{}`$.
* Let again $`S`$ be a positive nonzero trace class operator. Assume that $`A`$ is a positive bounded operator, such that the set of eigenvalues of $`A`$ equals neither the whole spectrum $`\sigma (A)`$ nor the set $`\sigma (A)\{0\}`$. Then $`\sigma (A)`$ contains a point $`a_0>0`$, which is not an eigenvalue of $`A`$. Now $`E^A(\{a_0\})=O`$. It follows that $`E^A(I_ϵ)`$, where $`I_ϵ=(a_0ϵ,a_0+ϵ)`$, is infinite-dimensional for all $`ϵ>0`$. Let $`t=\frac{a_0}{2}`$. Then $`t\chi _{I_t}(x)x`$ for all $`x0`$, so that $`tE^A(I_t)=t\chi _{I_t}(x)𝑑E^A(x)x𝑑E^A(x)=A`$. Since $`E^A(I_t)`$ is an infinite dimensional projection, $`\mathrm{}=t\mathrm{Tr}[E^A(I_t)]\mathrm{Tr}[A]`$, and hence also $`\mathrm{Tr}[A]=\mathrm{}`$. In addition, since $`\beta (g)(S)`$ is positive, $`t\mathrm{Tr}[E^A(I_t)\beta (g)(S)]\mathrm{Tr}[A\beta (g)(S)]`$. Since the projection $`E^A(I_t)`$ is infinite-dimensional, 3) implies that the function $`gt\mathrm{Tr}[E^A(I_t)\beta (g)(S)]`$ is not integrable. Thus $`d^1\mathrm{Tr}[A\beta (g)(S)]𝑑\lambda (g)=\mathrm{}=\mathrm{Tr}[A]\mathrm{Tr}[S]`$.
The lemma is proved. $`\mathrm{}`$
In the following definition, note that the class of the function $`f(g)L^{\mathrm{}}(G,\lambda )`$ is independent of the representative of $`f`$.
Definition. A linear map $`\mathrm{\Gamma }:L^{\mathrm{}}(G,\lambda )L()`$ is said to be *$`\beta `$-covariant*, if $`\beta (g)^{}(\mathrm{\Gamma }(f))=\mathrm{\Gamma }(f(g))`$ for all $`fL^{\mathrm{}}(G,\lambda )`$, $`gG`$.
The main result in this paper, Theorem 2, has the following rather straightforward and, at least in special cases, well-known converse.
###### Theorem 1.
Let $`T`$ be a positive operator of trace one. Then for each $`fL^{\mathrm{}}(G,\lambda )`$, the integral
(3)
$$d^1f(g)\beta (g)(T)𝑑\lambda (g)$$
exists as an operator $`\mathrm{\Gamma }(f)L()`$ in the ultraweak sense. In addition, $`\mathrm{\Gamma }(g1)=I`$, and the map $`f\mathrm{\Gamma }(f)`$ is linear, positive, normal, and $`\beta `$-covariant.
Proof. By Lemma 2, the function $`g\mathrm{Tr}[S\beta (g)(T)]`$ is in $`L^1(G,\lambda )`$ for each trace class operator $`S`$ (the operator $`S`$ can be written as a linear combination of four positive trace-class operators). Thus for each $`fL^{\mathrm{}}(G,\lambda )`$ we can define the (clearly linear) functional $`\mathrm{\Phi }_f:𝒯()`$ by
$$\mathrm{\Phi }_f(S)=d^1f(g)\mathrm{Tr}[S\beta (g)(T)]𝑑\lambda (g).$$
Let now $`fL^{\mathrm{}}(G,\lambda )`$ be real valued. If $`S𝒯()`$ is positive, we have by Lemma 2
$$|\mathrm{\Phi }_f(S)|d^1M_f\mathrm{Tr}[S\beta (g)(T)]𝑑\lambda (g)=M_fS_{\mathrm{tr}},$$
where $`M_f<\mathrm{}`$ is such that $`f(g)M_f`$ for almost all $`g`$. If $`S𝒯()`$ is selfadjoint, it can be written in the form $`S=S^+S^{}`$, where $`S^\pm 𝒯()`$ are positive and $`|S|=S^++S^{}`$. Now
$$|\mathrm{\Phi }_f(S)||\mathrm{\Phi }_f(S^+)|+|\mathrm{\Phi }_f(S^{})|M_f(S^+_{\mathrm{tr}}+S^{}_{\mathrm{tr}})=M_fS_{\mathrm{tr}},$$
so that for real valued $`f`$, the map $`\mathrm{\Phi }_f`$ restricted to the set of selfadjoint trace class operators is a real valued trace-norm continuous linear functional. Hence, there is a selfadjoint operator $`\mathrm{\Gamma }(f)L()`$, such that $`\mathrm{\Phi }_f(S)=\mathrm{Tr}[S\mathrm{\Gamma }(f)]`$ for all selfadjoint $`S𝒯()`$. For an arbitrary $`S𝒯()`$, we have $`S=S_1+iS_2`$, where each $`S_i𝒯()`$ are selfadjoint, and so
$$\mathrm{\Phi }_f(S)=\mathrm{\Phi }_f(S_1)+i\mathrm{\Phi }_f(S_2)=\mathrm{Tr}[S_1\mathrm{\Gamma }(f)]+i\mathrm{Tr}[S_2\mathrm{\Gamma }(f)]=\mathrm{Tr}[S\mathrm{\Gamma }(f)].$$
Let now $`fL^{\mathrm{}}(G,\lambda )`$ be complex valued: $`f=f_1+if_2`$, where $`f_1`$ and $`f_2`$ are real. Then clearly $`\mathrm{\Phi }_f(S)=\mathrm{\Phi }_{f_1}(S)+i\mathrm{\Phi }_{f_2}(S)=\mathrm{Tr}[S(\mathrm{\Gamma }(f_1)+i\mathrm{\Gamma }(f_2))]`$ for all $`S𝒯()`$. Define $`\mathrm{\Gamma }(f):=\mathrm{\Gamma }(f_1)+i\mathrm{\Gamma }(f_2)L()`$. Now
$$\mathrm{\Phi }_f(S)=\mathrm{Tr}[S\mathrm{\Gamma }(f)]$$
for all $`S𝒯()`$ and $`fL^{\mathrm{}}(G,\lambda )`$, implying the existence of the integral (3) in the ultraweak sense as the operator $`\mathrm{\Gamma }(f)L()`$.
The statement $`\mathrm{\Gamma }(g1)=I`$ follows from Lemma 2.
Clearly $`\mathrm{\Gamma }:L^{\mathrm{}}(G,\lambda )L()`$ is linear. If $`f0`$ and $`\phi `$,
$$\phi |\mathrm{\Gamma }(f)\phi =\mathrm{Tr}[|\phi \phi |\mathrm{\Gamma }(f)]=\mathrm{\Phi }_f(|\phi \phi |)=d^1f(g)\phi |\beta (g)(T)\phi 𝑑\lambda (g)0,$$
which proves the positivity of $`\mathrm{\Gamma }`$. Since $`\mathrm{\Phi }_f(S)=\mathrm{Tr}[S\mathrm{\Gamma }(f)]`$ for all $`S𝒯()`$ and $`fL^{\mathrm{}}(G,\lambda )`$, $`\mathrm{\Gamma }`$ is the dual of the map $`𝒯()Sd^1\mathrm{Tr}[S\beta ()(T)]L^1(G,\lambda )`$, and hence normal.
Covariance is seen from the calculation
$`\mathrm{Tr}[S\mathrm{\Gamma }(f(g))]`$ $`=`$ $`d^1{\displaystyle f(gg^{})\mathrm{Tr}[S\beta (g^{})(T)]𝑑\lambda (g^{})}=d^1{\displaystyle f(g^{})\mathrm{Tr}[S\beta (g^1g^{})(T)]𝑑\lambda (g^{})}`$
$`=`$ $`d^1{\displaystyle f(g^{})\mathrm{Tr}[\beta (g)(S)\beta (g^{})(T)]𝑑\lambda (g^{})}=\mathrm{Tr}[\beta (g)(S)\mathrm{\Gamma }(f)]=\mathrm{Tr}[S\beta (g)^{}(\mathrm{\Gamma }(f))],`$
where $`gG`$ and $`fL^{\mathrm{}}(G,\lambda )`$ are arbitrary, and the left invariance of $`\lambda `$, along with the properties of the map $`\beta `$, are used. $`\mathrm{}`$
## 3. General covariant maps
In this section we formulate the essential part of the characterization in yet a slightly more general context. Let $`X`$ a Banach space having the Radon-Nikodým property, i.e., if $`(\mathrm{\Omega },𝒜,\nu )`$ is a finite (positive) measure space and $`\mu :𝒜X`$ a $`\nu `$-continuous vector measure with bounded variation, there is a $`\nu `$-(Bochner-)integrable function $`g_\mu :\mathrm{\Omega }X`$, such that $`\mu (B)=_Bg_\mu 𝑑\nu `$ for all $`B𝒜`$ (see \[6, p. 61\]). The function $`g_\mu `$ is $`\nu `$-essentially unique \[6, p. 47, Corollary 5\].
The statement of the following Lemma is called the Riesz Representation Theorem. It is proved in \[6, pp. 62-63\], in the case where $`\nu `$ is a finite measure. The Lemma here is an obvious generalization of that result to the $`\sigma `$-finite case. As it constitutes the very starting point of the proof of the main result of the paper, we give its proof here.
###### Lemma 3.
Let $`(\mathrm{\Omega },𝒜,\nu )`$ be a $`\sigma `$-finite measure space, $`X`$ a Banach space having the Radon-Nikodým property, and $`\mathrm{\Gamma }:L^1(\mathrm{\Omega },\nu )X`$ a continuous linear map. Then there is a $`\nu `$-essentially unique $`\nu `$-measurable function $`v:\mathrm{\Omega }X`$, such that $`sup_{x\mathrm{\Omega }}v(x)=\mathrm{\Gamma }`$, and
$$\mathrm{\Gamma }(f)=fv𝑑\nu $$
for all $`fL^1(\mathrm{\Omega },\nu )`$.
Proof. Choose a disjoint sequence $`(K_n)`$ of sets in $`𝒜`$, such that $`\mathrm{\Omega }=_nK_n`$, and $`\nu (K_n)<\mathrm{}`$. Denote by $`\nu _n`$ the restriction of $`\nu `$ to the $`\sigma `$-algebra $`𝒜(K_n)=\{BK_n|B𝒜\}`$. Define the set function $`\mu _n:𝒜(K_n)X`$ by $`\mu _n(B)=\mathrm{\Gamma }(\chi _B)`$. Now $`\mu _n(B)\mathrm{\Gamma }\chi _B_1=\mathrm{\Gamma }\nu _n(B)`$ for all $`B𝒜(K_n)`$. It follows that $`\mu _n`$ is a $`\nu _n`$-continuous vector measure of bounded variation, with the variation satisfying $`|\mu _n|(B)\mathrm{\Gamma }\nu (B)`$ for all $`B𝒜(K_n)`$. Since $`X`$ has the Radon-Nikodým property and $`\nu _n`$ is a finite measure, there is a $`\nu _n`$-integrable function $`v_n:K_nX`$, such that $`\mu _n(B)=_Bv_n𝑑\nu _n`$ for all $`B𝒜(K_n)`$. For each $`fL^1(K_n,\nu _n)`$, let $`\stackrel{~}{f}`$ be the function $`\mathrm{\Omega }`$ which coincides with $`f`$ in $`K_n`$ and is zero elsewhere. Since the map $`L^1(K_n,\nu _n)f\mathrm{\Gamma }(\stackrel{~}{f})X`$ is linear and continuous, it follows from \[6, Lemma 4, p. 62\] that $`v_n(x)\mathrm{\Gamma }`$ for $`\nu _n`$-almost all $`xK_n`$, and $`\mathrm{\Gamma }(\stackrel{~}{f})=fv_n𝑑\nu _n`$ for each $`fL^1(K_n,\nu _n)`$. By \[6, Corollary 5, p. 47\], $`v_n`$ is $`\nu _n`$-essentially unique, and $`v_n`$ can be redefined to be zero in the null set in which originally $`v_n(x)>\mathrm{\Gamma }`$. Now we have $`sup_{xK_n}v_n(x)\mathrm{\Gamma }`$.
We denote by $`v_n`$ also the function $`\mathrm{\Omega }X`$ which coincides with $`v_n`$ in $`K_n`$ and is zero elsewhere. Since the sets $`K_n`$ are disjoint, we can define $`v=_nv_n`$, where the sum converges pointwise. Since $`v`$ is a pointwise limit of $`\nu `$-measurable functions, it is itself $`\nu `$-measurable. Denote $`M=sup_{x\mathrm{\Omega }}v(x)\mathrm{\Gamma }`$.
Let $`fL^1(\mathrm{\Omega },\nu )`$. Now the sequence $`(f_k)`$, where $`f_k=\chi _{_{n=1}^kK_n}f`$ converges pointwise, and hence (by the dominated convergence theorem) in $`L^1(\mathrm{\Omega },\nu )`$ to $`f`$. By continuity, $`(\mathrm{\Gamma }(f_k))`$ converges to $`\mathrm{\Gamma }(f)`$ in $`X`$. On the other hand, since $`f_k(x)v(x)|f(x)|\mathrm{\Gamma }`$ for all $`x\mathrm{\Omega }`$, the dominated convergence theorem gives
$$\mathrm{\Gamma }(f_k)=\underset{n=1}{\overset{k}{}}\mathrm{\Gamma }(\chi _{K_n}f)=\underset{n=1}{\overset{k}{}}(f|K_n)v_n𝑑\nu _n=f_kv𝑑\nu fv𝑑\nu .$$
Thus,
$$\mathrm{\Gamma }(f)=fv𝑑\nu .$$
Since $`\mathrm{\Gamma }(f)|f(x)|v(x)𝑑\nu (x)f_1M`$ for all $`fL^1(\mathrm{\Omega },\nu )`$, we get $`\mathrm{\Gamma }M`$, so $`M=\mathrm{\Gamma }`$. Since $`\nu `$ is $`\sigma `$-additive, $`v`$ is $`\nu `$-essentially unique by \[6, Corollary 5, p. 47\]. The Lemma is proved. $`\mathrm{}`$
The next Proposition allows us to specify the nature of the function $`v`$ obtained in the previous Lemma, in the case where $`\mathrm{\Omega }`$ is a locally compact topological group possessing certain additional properties. The next Lemma is essential to its proof.
###### Lemma 4.
Let $`\mathrm{\Omega }`$ be a locally compact second countable topological group with a left Haar measure $`\nu `$.
* Let $`h:\mathrm{\Omega }`$ be a $`\nu `$-measurable $`\nu `$-essentially bounded function such that for each $`y\mathrm{\Omega }`$, the function $`h(y)`$ coincides with $`h`$ $`\nu `$-almost everywhere. Then there is a constant $`c`$, such that $`h(x)=c`$ for $`\nu `$-almost all $`x\mathrm{\Omega }`$.
* Let $`X`$ be a Banach space, and $`h:\mathrm{\Omega }X`$ a $`\nu `$-measurable $`\nu `$-essentially bounded function such that for each $`y\mathrm{\Omega }`$, the function $`h(y)`$ coincides with $`h`$ $`\nu `$-almost everywhere. Then there is an $`sX`$, such that $`h(x)=s`$ for $`\nu `$-almost all $`x\mathrm{\Omega }`$.
Proof. (a) Clearly the positive functions $`h_i^\pm =\frac{1}{2}(|h_i|\pm h_i)`$, $`i=1,2`$, for which $`h=(h_1^+h_1^{})+i(h_2^+h_2^{})`$, share the property assumed to hold for $`h`$. Therefore, it suffices to prove the result in the case where $`h`$ is positive. Since $`h`$ is $`\nu `$-essentially bounded and $`\nu `$-measurable, the $`\nu `$-measurable function $`fh`$ is $`\nu `$-integrable for all $`fL^1(\mathrm{\Omega },\nu )`$. Let $`𝒞_c(\mathrm{\Omega })`$ denote the space of compactly supported continuous complex functions on $`\mathrm{\Omega }`$. We notice that the positive functional $`I_h:𝒞_c(\mathrm{\Omega })`$, defined by $`I_h(f)=fh𝑑\nu `$, satisfies the relation
$$I_h(f)=f(x)h(x)d\nu (x)=f(yx)h(yx)d\nu (x)=f(yx)h(x)d\nu (x)=I_h(f(y))$$
for each $`y\mathrm{\Omega }`$, and hence is a left Haar integral in the group $`\mathrm{\Omega }`$. By the uniqueness theorem of Haar integrals, there is a $`c>0`$, such that $`I_h(f)=cf𝑑\nu `$ for all $`f𝒞_c(\mathrm{\Omega })`$. Since $`𝒞_c(\mathrm{\Omega })`$ is dense in $`L^1(\mathrm{\Omega },\nu )`$, it follows that $`h(x)=c`$ for almost all $`x\mathrm{\Omega }`$.
(b) Fix some $`A(\mathrm{\Omega })`$, such that $`0<\nu (A)<\mathrm{}`$, and denote $`s=\nu (A)^1_Ah𝑑\nu X`$. Let $`w^{}X^{}`$. Since $`h`$ is $`\nu `$-measurable, so is the complex valued function $`xw^{},h(x)`$, which thus coincides almost everywhere with some Borel function $`h_w^{}`$. Since $`(x,y)xy`$ is continuous, the function $`(x,y)h_w^{}(xy)`$ is $`\nu \times \nu `$-measurable. By assumption, $`h_w^{}`$ satisfies the conditions of (a), so there is a constant $`c`$, and a $`\nu `$-null set $`N`$, such that $`h_w^{}(y)=c`$ for all $`y\mathrm{\Omega }N`$. Let $`x\mathrm{\Omega }`$. Since the left and right Haar measures have the same null sets, also $`Nx^1x^1N`$ is a $`\nu `$-null set. Thus, for each $`x\mathrm{\Omega }`$, we have $`h_w^{}(yx)=c=h_w^{}(xy)`$ for almost all $`y\mathrm{\Omega }`$. Using this fact, the assumption and the Fubini-Tonelli theorem, we get for each $`fL^1(\mathrm{\Omega },\nu )`$,
$`w^{},\nu (A)s{\displaystyle f𝑑\nu }`$ $`=`$ $`{\displaystyle _A}h_w^{}(x)𝑑\nu (x){\displaystyle f(y)𝑑\nu (y)}`$
$`=`$ $`{\displaystyle \left(\chi _A(x)h_w^{}(x)f(y)𝑑\nu (x)\right)𝑑\nu (y)}`$
$`=`$ $`{\displaystyle \left(\chi _A(x)h_w^{}(yx)f(y)𝑑\nu (x)\right)𝑑\nu (y)}`$
$`=`$ $`{\displaystyle \left(\chi _A(x)h_w^{}(yx)f(y)𝑑\nu (y)\right)𝑑\nu (x)}`$
$`=`$ $`{\displaystyle \left(\chi _A(x)h_w^{}(xy)f(y)𝑑\nu (y)\right)𝑑\nu (x)}`$
$`=`$ $`{\displaystyle \left(\chi _A(x)h_w^{}(y)f(y)𝑑\nu (y)\right)𝑑\nu (x)}=w^{},\nu (A){\displaystyle f(y)h(y)𝑑\nu (y)}.`$
The use of the Fubini-Tonelli theorem is justified because $`\nu `$ is $`\sigma `$-finite, $`(x,y)\chi _A(x)h_w^{}(yx)f(y)`$ is $`\nu \times \nu `$-measurable, and
$$\left(\chi _A(x)h_w^{}(yx)f(y)𝑑\nu (x)\right)𝑑\nu (y)\nu (A)w^{}Mf_1<\mathrm{},$$
where $`M>0`$ is such that $`h(x)M`$ for almost all $`x\mathrm{\Omega }`$. Since $`w^{}X^{}`$ was arbitrary, we get $`_Bh(y)𝑑\nu (y)=_Bs𝑑\nu (y)`$ for each $`B(\mathrm{\Omega })`$ of finite measure. Thus, from \[6, Corollary 5, p. 47\] and the $`\sigma `$-finiteness of $`\nu `$ it follows that $`h(x)=s`$ for almost all $`x\mathrm{\Omega }`$. $`\mathrm{}`$
###### Proposition 1.
Assume that $`\mathrm{\Omega }`$ is a locally compact second countable topological group with a left Haar measure $`\nu `$, and $`X`$ a Banach space having the Radon-Nikodým property. In addition, assume that there is a homomorphism $`\alpha :\mathrm{\Omega }\mathrm{Iso}(X)`$, such that
* $`sup_{x\mathrm{\Omega }}\alpha (x)<\mathrm{}`$;
* for all $`wX`$, the map $`x\alpha (x^1)(w)`$ is $`\nu `$-measurable.
If $`\mathrm{\Gamma }:L^1(\mathrm{\Omega },\nu )X`$ is a continuous linear map satisfying $`\alpha (x)(\mathrm{\Gamma }(f))=\mathrm{\Gamma }(f(x^1))`$ for all $`fL^1(\mathrm{\Omega },\nu )`$ and $`x\mathrm{\Omega }`$, then there is a unique vector $`sX`$, such that
$$\mathrm{\Gamma }(f)=f(x)\alpha (x)(s)𝑑\nu (x)$$
for all $`fL^1(\mathrm{\Omega },\nu )`$. If each $`\alpha (x)`$ is an isometry, then $`s=\mathrm{\Gamma }`$.
Proof. Let $`v:\mathrm{\Omega }X`$ be the function obtained in Lemma 3. We have to prove that for some unique $`sX`$, it satisfies $`v(x)=\alpha (x)(s)`$ for almost all $`x\mathrm{\Omega }`$. To that end, let $`B(\mathrm{\Omega })`$ be such that $`\nu (B)<\mathrm{}`$, and $`y\mathrm{\Omega }`$. Then, by the continuity of the linear map $`\alpha (y)`$ we get, by using the left invariance of $`\nu `$,
$`{\displaystyle _B}\alpha (y)(v(x))𝑑\nu (x)`$ $`=`$ $`\alpha (y)(\mathrm{\Gamma }(\chi _B))=\mathrm{\Gamma }(\chi _B(y^1))={\displaystyle }\chi _B(y^1x)v(x)d\nu (x)`$
$`=`$ $`{\displaystyle _B}v(yx)𝑑\nu (x).`$
Since the measure $`\nu `$ is $`\sigma `$-finite, it follows from \[6, Corollary 5, p. 47\] that for each $`y\mathrm{\Omega }`$,
(4)
$$\alpha (y)(v(x))=v(yx)\text{ for almost all }x\mathrm{\Omega }.$$
Now define the map $`v_0:\mathrm{\Omega }X`$ by $`v_0(x)=\alpha (x^1)(v(x))`$. Then $`v_0`$ is $`\nu `$-measurable. Indeed, let $`B(\mathrm{\Omega })`$ be such that $`\nu (B)<\mathrm{}`$. Since $`v`$ is $`\nu `$-measurable, there is a sequence $`(v_n)`$ of $`\nu `$-simple functions vanishing outside $`B`$ and converging $`\nu `$-a.e. to $`\chi _Bv`$. For each $`wX`$, the map $`x\alpha (x^1)(w)`$ is $`\nu `$-measurable by assumption (ii), so that also the functions $`x\alpha (x^1)(v_n(x))`$, are $`\nu `$-measurable. Now $`\alpha (x^1)(v_n(x))\chi _B(x)\alpha (x^1)(v(x))=\chi _Bv_0(x)`$ for $`\nu `$-almost all $`x`$, because $`\alpha (x^1)`$ is continuous, so the limit $`\chi _Bv_0`$ is $`\nu `$-measurable \[8, p. 150\]. Thus $`v_0`$ is $`\nu `$-measurable.
Let $`fL^1(\mathrm{\Omega },\nu )`$. Since $`v_0`$ is $`\nu `$-measurable, so is $`fv_0`$ \[8, p. 106\]. In addition, since $`sup_{x\mathrm{\Omega }}v(x)=\mathrm{\Gamma }`$, we get $`f(x)v_0(x)M|f(x)|v(x)|f(x)|M\mathrm{\Gamma }`$ for all $`x`$, where $`M=sup_{x\mathrm{\Omega }}\alpha (x)<\mathrm{}`$, so $`fv_0`$ is $`\nu `$-integrable. In particular, $`v_0`$ is integrable over any set $`B(\mathrm{\Omega })`$ of finite measure. Also, $`v_0(x)M\mathrm{\Gamma }`$ for all $`x`$, so $`v_0`$ is $`\nu `$-essentially bounded. Since $`\alpha `$ is a homomorphism, $`v(x)=\alpha (x)(v_0(x))`$ for all $`x`$. Let $`y\mathrm{\Omega }`$. The result (4) gives $`\alpha (y)(\alpha (x)(v_0(x)))=\alpha (yx)(v_0(yx))`$ for almost all $`x`$, so that
(5)
$$\text{for each }y\mathrm{\Omega },v_0(x)=v_0(yx)\text{ for almost all }x\mathrm{\Omega }.$$
By Lemma 4 there is an $`sX`$, such that $`v_0(x)=s`$ for $`\nu `$-almost all $`x`$. Thus
$$\mathrm{\Gamma }(f)=f(x)v(x)𝑑\nu (x)=f(x)\alpha (x)(v_0(x))𝑑\nu (x)=f(x)\alpha (x)(s)𝑑\nu (x)$$
for all $`fL^1(\mathrm{\Omega },\nu )`$. The vector $`s`$ in the above representation is uniquely determined, because if $`s^{}X`$ had the same properties, then by the uniqueness of the map $`xv(x)`$ in the representation of Lemma 3, $`\alpha (x)(s)=v(x)=\alpha (x)(s^{})`$ for almost all $`x\mathrm{\Omega }`$, so that $`s=s^{}`$.
If $`\alpha (x)`$ is an isometry for each $`x\mathrm{\Omega }`$, we have in addition,
$$\mathrm{\Gamma }=\underset{x\mathrm{\Omega }}{sup}v(x)=\underset{x\mathrm{\Omega }}{sup}\alpha (x)(s)=s$$
The proof is complete. $`\mathrm{}`$
## 4. Positive normal covariant maps
Now we return to the concept of $`(G,\beta ,d)`$ introduced earlier. Theorem 2 below characterizes all positive normal $`\beta `$-covariant maps $`\mathrm{\Gamma }:L^{\mathrm{}}(G,\lambda )L()`$. The proof is based on the fact that $`𝒯()`$, being a separable dual space, has the Radon-Nikodým property by \[6, p. 79\]. Therefore, the following Lemma is needed. We give the proof for completeness. (The result is given without proof e.g. in \[14, Exercise 5.7, p. 131\].)
###### Lemma 5.
The space $`𝒯()`$ is separable (with respect to the trace norm).
Proof. If $`\phi ,\psi `$ are such that $`\phi =1`$, and $`\phi \psi 1`$, then
(6)
$$|\psi \psi ||\phi \phi |_{\mathrm{tr}}3\psi \phi .$$
Indeed, since the map $`𝒯()T\mathrm{Tr}[T]𝒞()^{}`$, where $`𝒞()`$ denotes the set of compact operators, is an isometry, we have $`T_{\mathrm{tr}}=sup\{|\mathrm{Tr}[TA]|A𝒞(),A1\}`$ for each $`T𝒯()`$. Let $`\phi ,\psi `$ be such that $`\phi =1`$, and $`\phi \psi 1`$. If $`A𝒞()`$, $`A1`$, we have
$`|\mathrm{Tr}[(|\psi \psi ||\phi \phi |)A]|`$ $`=`$ $`|\psi |A\psi \phi |A\phi ||\psi |A\psi \psi |A\phi |+|\psi |A\phi \phi |A\phi |`$
$``$ $`\psi \psi \phi +\psi \phi \phi `$
$``$ $`(\psi \phi +\phi )\psi \phi +\psi \phi \phi 3\psi \phi .`$
Thus (6) holds.
Let $`M`$ be a countable dense set in the separable space $``$. Define $``$ to be the set of operators of the form $`_{\psi F}\lambda _\psi |\psi \psi |`$, where $`F`$ is a finite subset of $`M`$ and each $`\lambda _\psi `$ is a positive rational number (the vectors $`\psi `$ need not be of unit length). Since $`M`$ and $``$ are countable sets, $``$ is countable. Clearly $``$ is a subset of the set $`𝒯()^+`$ of positive trace-class operators. We proceed to show that $``$ is $`_{\mathrm{tr}}`$-dense in $`𝒯()^+`$.
Let $`S𝒯()^+`$ and $`ϵ>0`$. Using the decomposition $`S=_nt_n|\phi _n\phi _n|`$, which converges in the trace norm, with $`t_n0`$ and the $`\phi _n`$ orthonormal unit vectors, we find that there is a $`k`$, such that
(7)
$$S\underset{n=1}{\overset{k}{}}t_n|\phi _n\phi _n|_{\mathrm{tr}}<\frac{ϵ}{3}.$$
Now we choose positive rational numbers $`\lambda _n`$, $`n=1,\mathrm{},k`$, such that $`|t_n\lambda _n|<\frac{ϵ}{3k}`$ for all $`n=1,\mathrm{},k`$. Then
(8)
$$\underset{n=1}{\overset{k}{}}t_n|\phi _n\phi _n|\underset{n=1}{\overset{k}{}}\lambda _n|\phi _n\phi _n|_{\mathrm{tr}}<\frac{ϵ}{3}.$$
Since $`M`$ is dense, we can pick vectors $`\psi _nM`$, $`n=1,\mathrm{},k`$, such that $`\psi _n\phi _n<\frac{ϵ}{9_{n=1}^k\lambda _n}`$ for all $`n=1,\mathrm{},k`$. It can be assumed that $`ϵ<1`$, so that we can use the result (6) to get
(9)
$$\underset{n=1}{\overset{k}{}}\lambda _n|\phi _n\phi _n|\stackrel{~}{S}_{\mathrm{tr}}<\frac{ϵ}{3},$$
where $`\stackrel{~}{S}=_{n=1}^k\lambda _n|\psi _n\psi _n|`$. The inequalities (7)-(9) now imply $`S\stackrel{~}{S}<ϵ`$. Thus $``$ is $`_{\mathrm{tr}}`$-dense in $`𝒯()^+`$.
Since $`𝒯()=𝒯()^+𝒯()^++i(𝒯()^+𝒯()^{})`$, the set $`+i()`$ is a countable dense subset of $`𝒯()`$. $`\mathrm{}`$
###### Theorem 2.
Let $`\mathrm{\Gamma }:L^{\mathrm{}}(G,\lambda )L()`$ be a normal positive $`\beta `$-covariant linear map satisfying $`\mathrm{\Gamma }(g1)=I`$. Then $`\mathrm{\Gamma }`$ is of the form of Theorem 1 for a unique positive operator $`T𝒯()`$ of trace one.
Proof. Since $`\mathrm{\Gamma }:L^1(G,\lambda )^{}𝒯()^{}`$ is a weak-\* continuous linear map, there is a linear map $`\mathrm{\Gamma }_{}:𝒯()L^1(G,\lambda )`$, such that $`(\mathrm{\Gamma }_{})^{}=\mathrm{\Gamma }`$. The map $`\mathrm{\Gamma }_{}`$ is also positive, since $`(\mathrm{\Gamma }_{}(S))(g)f(g)𝑑\lambda (g)=\mathrm{Tr}[\mathrm{\Gamma }(f)S]0`$ for all positive $`S𝒯()`$ and $`fL^{\mathrm{}}(G,\lambda )`$, $`f0`$. Let $`S𝒯()`$ be positive and $`fL^{\mathrm{}}(G,\lambda )L^1(G,\lambda )`$ a positive function. Then $`\mathrm{\Gamma }(f)`$ is a positive operator and $`\mathrm{\Gamma }_{}(S)`$ a positive function. By covariance, we have
$$\mathrm{Tr}[\mathrm{\Gamma }(f)\beta (g)(S)]=\mathrm{Tr}[\beta (g)^{}(\mathrm{\Gamma }(f))S]=\mathrm{Tr}[S\mathrm{\Gamma }(f(g))]=(\mathrm{\Gamma }_{}(S))(g^{})f(gg^{})d\lambda (g^{}),$$
from which it follows by the Fubini-Tonelli theorem and the right invariance of $`\lambda `$ that
$$\mathrm{Tr}[\mathrm{\Gamma }(f)\beta (g)(S)]𝑑\lambda (g)=(\mathrm{\Gamma }_{}(S))(g^{})\left(f(gg^{})𝑑\lambda (g)\right)𝑑\lambda (g^{})=\mathrm{\Gamma }_{}(S)_1f_1<\mathrm{}.$$
Lemma 2 now implies that $`\mathrm{\Gamma }(f)𝒯()`$, and since $`\mathrm{\Gamma }_{}(S)_1=\mathrm{\Gamma }_{}(S)(g)𝑑\lambda (g)=\mathrm{Tr}[S\mathrm{\Gamma }(g1)]=\mathrm{Tr}[S]`$, we find (by using Lemma 2 again) that for positive $`fL^{\mathrm{}}(G,\lambda )L^1(G,\lambda )`$, $`\mathrm{\Gamma }(f)_{\mathrm{tr}}=d^1f_1`$. If $`fL^1(G,\lambda )L^{\mathrm{}}(G,\lambda )`$ is arbitrary, we can write $`f=(f_1^+f_1^{})+i(f_2^+f_2^{})`$, where the $`f_i^\pm `$ are positive, and $`f_1^++f_1^{}+f_2^++f_2^{}=|f_1|+|f_2|2|f|`$. It then follows by the linearity of $`\mathrm{\Gamma }`$ that $`\mathrm{\Gamma }(f)_{\mathrm{tr}}2d^1f_1`$, implying that the restriction $`\mathrm{\Gamma }|L^1(G,\lambda )L^{\mathrm{}}(G,\lambda ):L^1(G,\lambda )L^{\mathrm{}}(G,\lambda )𝒯()`$ is continuous with respect to the norms $`_1`$ and $`_{\mathrm{tr}}`$. Since the set $`L^1(G,\lambda )L^{\mathrm{}}(G,\lambda )`$ contains all integrable simple functions, it is dense in $`L^1(G,\lambda )`$. Therefore (since $`𝒯()`$ is complete), $`\mathrm{\Gamma }|L^1(G,\lambda )L^{\mathrm{}}(G,\lambda )`$ can be extended to a continuous linear map $`\stackrel{~}{\mathrm{\Gamma }}:L^1(G,\lambda )𝒯()`$.
The map $`\stackrel{~}{\mathrm{\Gamma }}`$ is positive. In fact, if $`fL^1(G,\lambda )`$ is positive, there is an increasing sequence $`(f_n)`$ of integrable positive simple functions converging pointwise to $`f`$. By the monotone convergence theorem, $`f_nf`$ in the $`_1`$-norm, so that the trace-class operator $`\stackrel{~}{\mathrm{\Gamma }}(f)`$, being the trace-norm (and hence weak) limit of the sequence $`\mathrm{\Gamma }(f_n)`$ of positive trace-class operators, must be positive.
Now we show that the conditions of Proposition 1 are satisfied by the measure space $`(G,(G),\lambda )`$, the Banach space $`𝒯()`$, the homomorphism $`\beta `$, and the linear map $`\stackrel{~}{\mathrm{\Gamma }}`$.
Since $`𝒯()𝒞()^{}`$ is separable by Lemma 5, it has the Radon-Nikodým property \[6, p. 79\]. Since each $`\beta (g)`$ is an isometry, the condition (i) is holds. Let $`S𝒯()`$ and $`AL()`$. Since $`\lambda `$ is a Borel measure, the map $`g\mathrm{Tr}[A\beta (g^1)(S)]`$, being continuous, is also $`\lambda `$-measurable. Thus $`Ggw^{}(\beta (g^1)(S))`$ is $`\lambda `$-measurable for each $`w^{}𝒯()^{}L()`$. Since $`𝒯()`$ is separable, this implies by \[8, p. 149\] that the map $`g\beta (g^1)(S)`$ is measurable, so that the condition (ii) of Proposition 1 is satisfied. To verify condition (iii), let $`fL^1(G,\lambda )`$, $`gG`$. Choose a sequence $`(f_n)`$ of integrable simple functions converging to $`f`$ in the $`_1`$-norm. Thus, by the continuity of the mappings involved, the covariance of $`\mathrm{\Gamma }`$, and the fact that the map $`\beta (g^1)^{}=(\beta (g)^1)^{}`$ coincides with $`\beta (g)`$ on $`𝒯()`$, we get
$$\beta (g)(\stackrel{~}{\mathrm{\Gamma }}(f))=\underset{n}{lim}\beta (g^1)^{}(\mathrm{\Gamma }(f_n))=\underset{n}{lim}\mathrm{\Gamma }(f_n(g^1))=\stackrel{~}{\mathrm{\Gamma }}(f(g^1)),$$
where the limits are in the trace norm and the $`_1`$-norm. This proves that (iii) holds.
Thus, we can apply Proposition 1 to the map $`\stackrel{~}{\mathrm{\Gamma }}`$. There is a unique $`T^{}𝒯()`$, such that
$$\stackrel{~}{\mathrm{\Gamma }}(f)=f(g)\beta (g)(T^{})𝑑\lambda (g)$$
for all $`fL^1(G,\lambda )`$. Since $`L^{\mathrm{}}(G,\lambda )L^1(G,\lambda )`$ is weak-\* dense in $`L^{\mathrm{}}(G,\lambda )`$ and $`\mathrm{\Gamma }`$ is normal, we also have
$$\mathrm{\Gamma }(f)=f(g)\beta (g)(T^{})𝑑\lambda (g)$$
in the ultraweak sense for all $`fL^{\mathrm{}}(G,\lambda )`$.
It remains to prove that $`T^{}`$ is positive and of trace $`d^1`$.
Let $`S𝒯()`$ be positive. Since $`\mathrm{\Gamma }(\chi _B)`$ is a positive operator, we have
$$0\mathrm{Tr}[S\mathrm{\Gamma }(\chi _B)]=_B\mathrm{Tr}[S\beta (g)(T^{})]𝑑\lambda (g)$$
for all $`B(G)`$, from which it follows by the continuity of $`g\mathrm{Tr}[S\beta (g)(T^{})]`$ that $`\mathrm{Tr}[S\beta (g)(T^{})]0`$ for all $`gG`$. Thus $`T^{}`$ must be positive.
In addition, by the condition $`\mathrm{\Gamma }(g1)=I`$, and Lemma 2,
$$\mathrm{Tr}[S]d^1=d^1\mathrm{Tr}[S\mathrm{\Gamma }(\chi _G)]=d^1\mathrm{Tr}[S\beta (g)(T^{})]𝑑\lambda (g)=\mathrm{Tr}[S]\mathrm{Tr}[T^{}]$$
for any positive $`S𝒯()`$. Thus $`\mathrm{Tr}[T^{}]=d^1`$, so that by defining $`T=T^{}d`$, we get the required form for $`\mathrm{\Gamma }`$. $`\mathrm{}`$
## 5. Covariant observables
An observable (i.e. a positive normalized operator measure) $`E:(G)L()`$ is said to be *$`\beta `$-covariant* if $`\beta (g)^{}(E(B))=E(g^1B)`$ for all $`gG`$ and $`B(G)`$. The following Lemma shows that Theorem 2 can be used to characterize the covariant observables. The result (b) of the Lemma is obtained in for the more general case where the group need not be unimodular, and the condition (2) is not assumed. In the context of this paper, the proof following is more simple, as it can be formulated so that it uses Lemma 2. The proof is therefore given here.
###### Lemma 6.
Let $`E:(G)L()`$ be an observable.
* Assume that for each trace class operator $`S`$, the measure $`B\mathrm{Tr}[SE(B)]`$ is continuous with respect to the measure $`\lambda `$. Then for each $`fL^{\mathrm{}}(G,\lambda )`$, the operator integral $`f𝑑E`$ exists in $`L()`$ in the ultraweak sense, and the linear map $`ff𝑑E`$ is normal, positive, and satisfies $`1𝑑E(g)=I`$. If $`E`$ is $`\beta `$-covariant, so is the map $`ff𝑑E`$.
* If $`E`$ is $`\beta `$-covariant, the measure $`B\mathrm{Tr}[SE(B)]`$ is continuous with respect to the measure $`\lambda `$ for each trace class operator $`S`$.
Proof. (a) Let $`S𝒯()`$. Then $`S=_nt_n|\psi _n\phi _n|`$, where $`(\phi _n)`$ and $`(\psi _n)`$ are orthonormal sequences, $`t_n0`$, and $`t_n<\mathrm{}`$. The series converges in the trace norm. The map $`\mu `$, defined by $`B\mu (B)=\mathrm{Tr}[SE(B)]`$ is a complex valued finite measure, and (by the $`_{\mathrm{tr}}`$-continuity of the trace functional) it is a pointwise limit of the measures $`_{n=1}^k\mu _n`$, where $`\mu _n(B)=t_n\mathrm{Tr}[|\psi _n\phi _n|E(B)]`$ for each $`B(G)`$. Since the total variation norm of $`\mu _n`$ satisfies $`\mu _n4sup_{B(G)}|\mu _n(B)|4t_n`$, the series $`\mu =_n\mu _n`$ converges absolutely in the total variation norm.
Let $`fL^{\mathrm{}}(G,\lambda )`$. Since $`\mu `$ and each $`\mu _n`$ are $`\lambda `$-continuous, $`|f(g)|f_{\mathrm{}}`$ also $`\mu _n`$-, and $`\mu `$-almost everywhere. Thus, $`|f|d|\mu _n|f_{\mathrm{}}\mu _n4f_{\mathrm{}}t_n`$ so that $`_n|f|d|\mu _n|4f_{\mathrm{}}_nt_n=4f_{\mathrm{}}S_{\mathrm{tr}}<\mathrm{}`$. It now follows e.g. from \[11, Lemma 1\] that $`f`$ is $`\mu `$-integrable, and
$$fd(\mathrm{Tr}[SE()])=f𝑑\mu =\underset{n}{}f𝑑\mu _n=\underset{n}{}t_nfd(\mathrm{Tr}[|\psi _n\phi _n|E()]).$$
Since $`\mu `$ is $`\lambda `$-continuous, the integral does not depend on the representative of $`fL^{\mathrm{}}(G,\lambda )`$. In addition,
(10)
$$\left|fd(\mathrm{Tr}[SE()])\right|\underset{n}{}|f|d|\mu _n|=4f_{\mathrm{}}S_{\mathrm{tr}},$$
so that the functional $`Sfd(\mathrm{Tr}[SE()])`$ is $`_{\mathrm{tr}}`$-continuous. Thus the integral $`f𝑑E`$ exists in the ultraweak sense as an operator in $`L()`$, i.e., for each $`S𝒯()`$,
(11)
$$\mathrm{Tr}[S(f𝑑E)]=fd(\mathrm{Tr}[SE()]).$$
Since $`B\mathrm{Tr}[SE(B)]`$ is $`\lambda `$-continuous, it has a density $`g_SL^1(G,\lambda )`$. Since $`L^{\mathrm{}}(G,\lambda )ff𝑑EL()`$ is the dual map of $`𝒯()Sg_SL^1(G,\lambda )`$, it is normal.
Let $`fL^{\mathrm{}}(G,\lambda )`$ be positive and $`S𝒯()`$ a positive operator. Since the measure $`\mathrm{Tr}[SE()]`$ is positive, so is $`\mathrm{Tr}[S(f𝑑E)]=fd(\mathrm{Tr}[SE()])`$. It follows that $`f𝑑E`$ is positive. Thus the map $`ff𝑑E`$ is positive. Since $`E`$ is normalized, $`1𝑑E(g)=E(G)=I`$.
Assume now that $`E`$ is $`\beta `$-covariant. Let $`gG`$, $`B(G)`$, and $`S𝒯()`$. Since the measure $`\mathrm{Tr}[SE()]`$ has the density $`g_SL^1(G,\lambda )`$, the measure $`\mathrm{Tr}[SE(g^1)]`$ has the density $`g_S(g^1)`$. Using the left invariance of $`\lambda `$ and the covariance of $`E`$, we get
$`\mathrm{Tr}[S\beta (g)^{}({\displaystyle f𝑑E})]`$ $`=`$ $`\mathrm{Tr}[\beta (g)(S)({\displaystyle }fdE)]={\displaystyle }fd(\mathrm{Tr}[\beta (g)(S)E()])={\displaystyle }fd(\mathrm{Tr}[SE(g^1)])`$
$`=`$ $`{\displaystyle }f(g^{})g_S(g^1g^{})d\lambda (g^{})={\displaystyle }f(gg^{})g_S(g^{})d\lambda (g^{})={\displaystyle }f(g)d(\mathrm{Tr}[SE()])`$
$`=`$ $`\mathrm{Tr}[S({\displaystyle }f(g)dE)],`$
which proves that the map $`ff𝑑E`$ is $`\beta `$-covariant.
(b) Let $`S𝒯()`$ be positive and of trace one, and $`\mu `$ the probability measure $`B\mathrm{Tr}[SE(B)]`$. Now for each $`B(G)`$, covariance implies
$$\mathrm{Tr}[\beta (g)^{}(E(B))S]=\mathrm{Tr}[SE(g^1B)]=\chi _{g^1B}𝑑\mu =\chi _B(gg^{})𝑑\mu (g^{}).$$
Thus, by Lemma 2, the Fubini-Tonelli theorem, and the right invariance of $`\lambda `$, we get
$`\mathrm{Tr}[E(B)]`$ $`=`$ $`d^1{\displaystyle \mathrm{Tr}[E(B)\beta (g)(S)]𝑑\lambda (g)}=d^1{\displaystyle \left(\chi _B(gg^{})𝑑\lambda (g)\right)𝑑\mu (g^{})}`$
$`=`$ $`d^1\lambda (B){\displaystyle 𝑑\mu }=d^1\lambda (B).`$
Now let $`S𝒯()`$ be arbitrary. Then, if $`B(G)`$ is such that $`\lambda (B)<\mathrm{}`$, we have $`|\mathrm{Tr}[SE(B)]|SE(B)_{\mathrm{tr}}=d^1S\lambda (B)`$. This implies that the measure $`B\mathrm{Tr}[SE(B)]`$ is $`\lambda `$-continuous. $`\mathrm{}`$
###### Theorem 3.
Let $`E:(G)L()`$ be a positive normalized $`\beta `$-covariant operator measure. Then
$$E(B)=d^1_B\beta (g)(T)𝑑\lambda (g)$$
in the ultraweak sense, for some unique positive operator $`T𝒯()`$ of trace one.
Proof. By the previous Lemma, the linear map $`L^{\mathrm{}}(G,\lambda )ff𝑑EL()`$ satisfies the conditions of Theorem 2 and hence is of the form
$$f𝑑E=d^1f(g)\beta (g)(T)𝑑\lambda (g)$$
for some unique positive operator $`T`$ of trace one. In particular,
(12)
$$E(B)=\chi _B𝑑E=d^1_B\beta (g)(T)𝑑\lambda (g)$$
for each $`B(G)`$. The operator $`T`$ in the representation (12) of $`E`$ is also uniquely determined. In fact, if $`S𝒯()`$ is such that $`E(B)=d^1_B\beta (g)(S)𝑑\lambda (g)`$ for each $`B(G)`$, then by the uniqueness of $`T`$ in the representation of the linear map $`ff𝑑E`$, we get $`\chi _B(g)\beta (g)(S)𝑑\lambda (g)=\chi _B(g)\beta (g)(T)𝑑\lambda (g)`$ for all $`B(G)`$, so $`\beta (g)(S)=\beta (g)(T)`$ for almost all $`g`$, showing that $`S=T`$. $`\mathrm{}`$
Remark. Consider the concrete case $`(^{2n},\gamma ,(2\pi )^n)`$. For a linear map $`\mathrm{\Gamma }:L^{\mathrm{}}(^{2n},\mu _L)L(L^2(^n))`$, covariance means that $`\gamma (x)(\mathrm{\Gamma }(f))=f(x)`$ for all $`x^{2n}`$ and $`fL^{\mathrm{}}(^{2n},\mu _L)`$, whereas a covariant observable $`E:(^{2n})L(L^2(^n))`$ is such that $`\gamma (x)(E(B))=E(x+B)`$ for each $`x^{2n}`$ and $`B(^{2n})`$. Thus Theorem 2 gives, in particular, a characterization of positive covariant linear maps $`\mathrm{\Gamma }:L^{\mathrm{}}(^{2n},\mu _L)L(L^2(^n))`$, and Theorem 3 a characterization of the covariant phase space observables.
## 6. A note on quantization maps on the set of unbounded functions
Since many of the important dynamical variables in classical mechanics are unbounded functions, it is rather restrictive to consider only the quantization maps $`\mathrm{\Gamma }:L^{\mathrm{}}(G,\lambda )L()`$.
Let $`(G)`$ denote the set of all complex Borel functions on $`G`$, and $`𝒪()`$ the set of all (not necessarily bounded) linear operators in $``$. We call a map $`\mathrm{\Gamma }:(G)𝒪()`$ linear if $`\alpha \mathrm{\Gamma }(f)+\beta \mathrm{\Gamma }(h)\mathrm{\Gamma }(\alpha f+\beta h)`$ for all $`\alpha ,\beta `$ and $`f,h(G)`$. For each $`f(G)`$, we let $`D(\mathrm{\Gamma }(f))`$ denote the domain of $`\mathrm{\Gamma }(f)`$.
Let $`E:(G)L()`$ be a positive operator measure. For $`f(G)`$ let $`D(f,E)`$ be the set of those vectors $`\phi `$ for which $`f`$ is $`E_{\psi ,\phi }`$-integrable for all $`\psi `$. The operator integral $`L(f,E)=f𝑑E`$ is defined to be the unique (possibly unbounded) linear operator on the domain $`D(f,E)`$, for which $`\psi |L(f,E)\phi =f𝑑E_{\psi ,\phi }`$ for all $`\phi D(f,E)`$ and $`\psi `$ (cf. ). If $`f`$ is real valued, then $`L(f,E)`$ is a symmetric operator.
Consider the map $`\mathrm{\Gamma }_E:(G)𝒪()`$, defined by $`\mathrm{\Gamma }_E(f)=L(f,E)`$. If $`f,h(G)`$, $`\alpha ,\beta `$, then (since $`|f+h||f|+|h|`$) $`\alpha \mathrm{\Gamma }(f)+\beta \mathrm{\Gamma }(h)\mathrm{\Gamma }(\alpha f+\beta h)`$, so $`\mathrm{\Gamma }_E`$ is linear. It follows from the dominated convergence theorem that it is quasicontinuous in the sense of the following definition (already given in the Introduction).
Definition. A linear map $`\mathrm{\Gamma }:(G)𝒪()`$ is *quasicontinuous*, if for each increasing sequence $`(f_n)`$ of positive Borel functions converging pointwise to an $`f(G)`$ the numerical sequence $`(\psi |\mathrm{\Gamma }(f_n)\phi )`$ converges to $`\psi |\mathrm{\Gamma }(f)\phi `$ for all $`\psi `$ and $`\phi D(\mathrm{\Gamma }(f))_{nN}D(\mathrm{\Gamma }(f_n))`$.
In the Introduction we mentioned that in order to be represented as an operator integral, a quantization map $`\mathrm{\Gamma }`$ must be at least positive, linear and quasicontinuous, and map bounded functions to $`L()`$, for then the map $`E^\mathrm{\Gamma }:(G)L()`$, given by $`B\mathrm{\Gamma }(\chi _B)`$ is a positive operator measure, and $`\mathrm{\Gamma }(f)=L(f,E^\mathrm{\Gamma })`$ for each bounded function $`f(G)`$. In order to claim that $`\mathrm{\Gamma }=L(,E^\mathrm{\Gamma })`$, something must be assumed on the domains of the operators $`\mathrm{\Gamma }(f)`$. The following simple result follows readily from the definition of the operator integral:
###### Proposition 2.
Let $`\mathrm{\Gamma }:(G)𝒪()`$ be a linear map satisfying the following conditions:
* $`\mathrm{\Gamma }`$ is positive and quasicontinuous;
* $`\mathrm{\Gamma }`$ maps bounded functions to $`L()`$;
* for $`f(G)`$, the domain of $`\mathrm{\Gamma }(f)`$ consists of those vectors $`\phi `$ for which $`f`$ is $`E_{\psi ,\phi }^\mathrm{\Gamma }`$-integrable for all $`\psi `$.
Then $`\mathrm{\Gamma }(f)=L(f,E^\mathrm{\Gamma })`$ for all $`f(G)`$.
Proof. As (iii) asserts that the domains of the operators $`\mathrm{\Gamma }(f)`$ and $`L(f,E^\mathrm{\Gamma })`$ are the same, we are left to show that $`\mathrm{\Gamma }(f)\phi =L(f,E^\mathrm{\Gamma })\phi `$ for all $`\phi `$ in the common domain $`𝒟`$. Let $`f(G)`$, $`\phi 𝒟`$ and $`\psi `$. Assume first that $`f`$ is positive. Pick an increasing sequence $`(f_n)`$ of $`(G)`$-simple functions converging pointwise to $`f`$. By (ii), $`D(\mathrm{\Gamma }(f_n))=`$, so quasicontinuity implies that the sequence $`(z_n^\psi )`$, where $`z_n^\psi =\psi |\mathrm{\Gamma }(f_n)\phi `$, converges to $`\psi |\mathrm{\Gamma }(f)\phi `$ for all $`\psi `$. Since each $`f_n`$ is bounded, $`\mathrm{\Gamma }(f_n)=L(f_n,E^\mathrm{\Gamma })`$ for all $`n`$, so $`z_n^\psi =f_n𝑑E_{\psi ,\phi }^\mathrm{\Gamma }`$. But now (iii) and the dominated convergence theorem imply that $`z_n^\psi `$ converges to $`f𝑑E_{\psi ,\phi }^\mathrm{\Gamma }=\psi |L(f,E^\mathrm{\Gamma })\phi `$, so $`\psi |\mathrm{\Gamma }(f)\phi =\psi |L(f,E^\mathrm{\Gamma })\phi `$. Since $`\psi `$ was arbitrary, this gives $`\mathrm{\Gamma }(f)\phi =L(f,E^\mathrm{\Gamma })\phi `$. For a general $`f(G)`$, we write $`f=f_1^+f_1^{}+i(f_2^+f_2^{})`$, where $`f_j^\pm `$ are the positive and negative parts of $`f_j`$. Let $`\phi 𝒟`$ and $`\psi `$. Since $`0f_j^\pm |f|`$, we have that also $`f_j^\pm `$ is $`E_{\psi ,\phi }^\mathrm{\Gamma }`$-integrable for all $`\psi `$, i.e. $`\phi D(f_j^\pm ,E^\mathrm{\Gamma })=D(\mathrm{\Gamma }(f_j^\pm ))`$. Thus, $`\mathrm{\Gamma }(f_j^\pm )\phi =L(f_j^\pm ,E^\mathrm{\Gamma })\phi `$. By linearity, we get $`\mathrm{\Gamma }(f)\phi =L(f,E^\mathrm{\Gamma })\phi `$, completing the proof. $`\mathrm{}`$ |
warning/0507/math0507613.html | ar5iv | text | # Values of decomposable forms at 𝑆-integer points and tori orbits on homogeneous spaces
## 1. Introduction
Let $`𝐆`$ be a reductive algebraic group defined over a number field $`K`$ and let $`𝒮`$ be a finite set of (normalized) valuations of $`K`$ containing all archimedean ones. If $`v𝒮`$ we set $`G_v=𝐆(K_v)`$, where $`K_v`$ is the completion of $`K`$ with respect to $`v`$. Every $`G_v`$ is a locally compact group with a topology induced by the topology of $`K_v`$. Let $`G=_{v𝒮}G_v`$. The group of $`K`$-rational points $`𝐆(K)`$ is identified with its diagonal imbedding in $`G`$. We denote by $`\mathrm{\Gamma }`$ an $`𝒮`$-arithmetic subgroup of $`G`$, that is, $`\mathrm{\Gamma }`$ is a subgroup of $`G`$ such that $`\mathrm{\Gamma }𝐆(𝒪)`$ has finite index in both $`\mathrm{\Gamma }`$ and $`𝐆(𝒪)`$, where $`𝒪`$ is the ring of $`𝒮`$-integers of $`K`$. We fix a maximal $`K`$-split torus $`𝐃`$ of $`𝐆`$ and, for every $`v𝒮`$, we fix a $`K_v`$-torus $`𝐓_v`$ of $`𝐆`$ such that $`𝐓_v`$ contains both $`𝐃`$ and a maximal $`K_v`$-split torus of $`𝐆`$. Let $``$ be a non-empty subset of $`𝒮`$. Recall that the $``$-rank of $`𝐆`$ (or $`G`$) is $`\mathrm{rank}_{}𝐆\stackrel{def}{=}_v\mathrm{rank}_{K_v}𝐆`$. (If $`F`$ is a field containing $`K`$ then $`\mathrm{rank}_F𝐆`$ is by definition the dimension of any maximal $`F`$-split torus of $`𝐆`$.) We set $`T_{}=_vT_v`$ and $`D_{}=_vD_v`$, where $`T_v=𝐓_v(K_v)`$ and $`D_v=𝐃(K_v)`$. Then $`T_{}`$ is a torus of maximal $``$-rank and it acts on $`G/\mathrm{\Gamma }`$ by left translations
$$t\pi (g)=\pi (tg),$$
where $`\pi :GG/\mathrm{\Gamma }`$ is the quotient map. An orbit $`T_{}\pi (g)`$ is called divergent if the orbit map $`tt\pi (g)`$ is proper, i.e. if $`\{t_i\pi (g)\}`$ leaves compacts of $`G/\mathrm{\Gamma }`$ whenever $`\{t_i\}`$ leaves compacts of $`T_{}`$. In particular, the divergent orbits are closed.
We prove the following:
###### Theorem 1.1.
Let $`\mathrm{rank}_{}𝐆>0`$ and $`gG`$.
1. The orbit $`T_{}\pi (g)`$ is closed if and only if $``$ is a singleton or $`=𝒮`$, and there exists a $`K`$-torus $`𝐋`$ of $`𝐆`$ such that
$$g^1T_{}g=CL_{},$$
where $`C`$ is a compact group and $`L_{}=_v𝐋(K_v)`$;
2. The orbit $`T_{}\pi (g)`$ is divergent if and only if the following conditions are satisfied: $``$ is a singleton equal to $`v`$, $`\mathrm{rank}_{K_v}𝐆=\mathrm{rank}_K𝐆`$ and
$$g𝒵_G(D_v)𝐆(K),$$
where $`D_v`$ is identified with its natural projection in $`G`$ and $`𝒵_G(D_v)`$ is the centralizer of $`D_v`$ in $`G`$.
Theorem 1.1 generalizes the following result by B.Weiss and the author, the second part of which has been earlier proved (though unpublished) by G.Margulis for $`𝐆=\mathrm{𝐒𝐋}_𝐧`$ endowed with the standard $``$-structure (cf. \[To-We, Appendix\]).
###### Theorem 1.2.
(\[To-We, Theorem 1.1\]) Let $`𝐆`$ be a reductive $``$-algebraic group, $`𝐓`$ an $``$-torus containing a maximal $``$-split torus, $`T=𝐓()`$ and let $`xG`$. Then:
* $`T\pi (x)`$ is a closed orbit if and only if $`x^1𝐓x`$ is a product of a $``$-subtorus and an $``$-anisotropic $``$-subtorus;
* $`T\pi (x)`$ is a divergent orbit if and only if the maximal $``$-split subtorus of $`x^1𝐓x`$ is defined over $``$ and $``$-split.
When $`\mathrm{\#}>1`$, Theorem 1.1 implies a specific phenomenon:
###### Corollary 1.3.
If $`\mathrm{\#}𝒮>1`$ and $`T_{}\pi (g)`$ is a closed orbit then either $`=𝒮`$ and $`T_{}\pi (g)`$ is never divergent, or $``$ is a singleton and $`T_{}\pi (g)`$ is always divergent.
An orbit $`T_{}\pi (g)`$ is called locally divergent if $`T_v\pi (g)`$ is divergent for every $`v`$. Theorem 1.1 will be deduced from the next theorem about the locally divergent orbits.
###### Theorem 1.4.
Let $`\mathrm{rank}_{}(𝐆)>0`$. Then the orbit $`T_{}\pi (g)`$ is closed and locally divergent if and only if the following conditions are fulfilled:
1. $`=𝒮`$ or $``$ is a singleton;
2. $`\mathrm{rank}_{}(𝐆)=\mathrm{\#}\mathrm{rank}_K(𝐆)`$;
3. $`g𝒩_G(D_{})𝐆(K)`$, where $`𝒩_G(D_{})`$ is the normalizer of $`D_{}`$ in $`G`$.
When $`\mathrm{\#}=1`$ we can replace the normalizer $`𝒩_G(D_{})`$ in the formulation of Theorem 1.4 (iii) by the centralizer $`𝒵_G(D_{})`$. This is not possible when $`=𝒮`$ (see 6.2 (b)).
As a consequence of Theorem 1.4, one can easily see that the locally divergent $`T_{}`$-orbits are also all ”standard”:
###### Corollary 1.5.
Let $`gG`$. The orbit $`T_{}\pi (g)`$ is locally divergent if and only if
$$\mathrm{rank}_{}(𝐆)=\mathrm{\#}\mathrm{rank}_K(𝐆)$$
and
$$g\underset{v}{}𝒵_G(D_v)𝐆(K).$$
We also get the following result:
###### Corollary 1.6.
1. If $`\mathrm{rank}_{}(𝐆)>\mathrm{\#}\mathrm{rank}_K(𝐆)`$ then there are no locally divergent orbits for $`T_{}`$;
2. Let $`𝐆`$ be semisimple, $`\mathrm{\#}>1`$ and $`\mathrm{rank}_{}(𝐆)=\mathrm{\#}\mathrm{rank}_K(𝐆)>0`$. Then there exist locally divergent but non-closed orbits for $`T_{}`$.
We apply Theorem 1.1 to obtain a characterization of the rational decomposable homogeneous forms in terms of their values at the integer points. Such forms appear in a very natural way in both the algebraic number theory and the Diophantine approximation of numbers in connection with the notable Littlewood conjecture. (See, \[Bor-Sh, ch.2\] and \[Ma, §2\], respectively.)
We will first formulate our result in technically simpler particular cases. Given a commutative ring $`R`$, we denote by $`R[\stackrel{}{x}]`$ the ring of polynomials with coefficients from $`R`$ in $`n`$ variables $`\stackrel{}{x}=(x_1,\mathrm{},x_n)`$.
###### Theorem 1.7.
Let $`f(\stackrel{}{x})=l_1(\stackrel{}{x})\mathrm{}l_m(\stackrel{}{x})`$, where $`l_1(\stackrel{}{x}),\mathrm{},l_m(\stackrel{}{x})[\stackrel{}{x}]`$ are real linear forms. Suppose that $`l_1(\stackrel{}{x}),\mathrm{},l_m(\stackrel{}{x})`$ are linearly independent over $``$ and that the set $`f(^n)`$ is discrete in $``$. Then $`f(\stackrel{}{x})=\alpha g(\stackrel{}{x})`$, where $`g(\stackrel{}{x})[\stackrel{}{x}]`$ and $`\alpha ^{}`$.
The hypotheses that the form $`f(\stackrel{}{x})`$ is decomposable and $`l_1(\stackrel{}{x}),\mathrm{},`$ $`l_m(\stackrel{}{x})`$ are linearly independent over $``$ are essential. (See §7 for simple examples.) It is easy to prove (see \[Bor-Sh, ch.2, Theorem 2\]) that the form $`g(\stackrel{}{x})`$ in the formulation of the theorem is a constant multiple of a product of forms of the type $`\mathrm{N}_{K/}(x_1+x_2\mu _2+\mathrm{}+x_n\mu _n)`$, where $`\mu _2,\mathrm{},\mu _n`$ are algebraic numbers linearly generating a totally real number field $`K`$ of degree $`n`$ and $`\mathrm{N}_{K/}`$ is the algebraic norm of $`K`$.
If $`f`$ is a decomposable homogeneous form with complex coefficients and we are considering the values of $`f`$ at the Gaussian integer vectors, we get:
###### Theorem 1.8.
Let $`f(\stackrel{}{x})=l_1(\stackrel{}{x})\mathrm{}l_m(\stackrel{}{x})`$, where $`l_1(\stackrel{}{x}),\mathrm{},l_m(\stackrel{}{x})[\stackrel{}{x}]`$ are complex linear forms. Suppose that $`l_1(\stackrel{}{x}),\mathrm{},l_m(\stackrel{}{x})`$ are linearly independent over $``$ and that the set $`f([i]^n)`$ is discrete in $``$. Then $`f(\stackrel{}{x})=\alpha g(\stackrel{}{x})`$, where $`g(\stackrel{}{x})[i][\stackrel{}{x}]`$ and $`\alpha ^{}`$.
Let $`K`$, $`𝒮`$ and $`𝒪`$ be as in the formulation of Theorem 1.1. For every $`v𝒮`$, let $`f_v=l_1^{(v)}\mathrm{}l_m^{(v)}K_v[\stackrel{}{x}]`$, where $`l_1^{(v)},\mathrm{},l_m^{(v)}`$ are linearly independent over $`K_v`$ linear forms in $`K_v[\stackrel{}{x}]`$. Denote by $`K_𝒮`$ the direct product of the topological fields $`K_v`$, $`v𝒮`$. Both Theorems 1.7 and 1.8 are particular cases for $`K=`$ and $`K=(i)`$, respectively, of the next general theorem:
###### Theorem 1.9.
With the above notation, assume that $`\{(f_v(\stackrel{}{z}))_{v𝒮}K_𝒮|\stackrel{}{z}𝒪^n\}`$ is a discrete subset of $`K_𝒮`$. Then there exist an homogeneous form $`g`$ with coefficients from $`𝒪`$ and an element $`(\alpha _v)_{v𝒮}K_𝒮^{}`$ such that $`f_v=\alpha _vg`$ for all $`v𝒮`$.
In connection with Theorem 1.9 it seems natural to formulate the following conjecture which generalizes a well known conjecture for the real forms $`f`$:
$`\mathrm{𝐂𝐨𝐧𝐣𝐞𝐜𝐭𝐮𝐫𝐞}.`$ Let $`f_v,v𝒮`$, be as in the formulation of Theorem 1.9 with $`n=m`$ and $`\mathrm{\#}𝒮.n>2`$. Additionally, assume that there exists a neighborhood $`W`$ of $`0`$ in $`K_𝒮`$ such that $`(f_v(\stackrel{}{z}))_{v𝒮}W`$ for every $`\stackrel{}{z}𝒪^n,\stackrel{}{z}0`$. Then there exist an homogeneous form $`g`$ with coefficients from $`𝒪`$ and an element $`(\alpha _v)_{v𝒮}K_𝒮^{}`$ such that $`f_v=\alpha _vg`$ for all $`v𝒮`$.
Using the $`𝒮`$-adic version of Malher’s criterion (see Theorem 3.1 below), it is easy to see that the above conjecture can be reformulated in terms of Theorem 1.1 as follows: If $`𝐆=\mathrm{𝐒𝐋}_n`$ and $`\mathrm{rank}_𝒮𝐆>1`$ then $`T_𝒮\pi (g)`$ is compact whenever $`T_𝒮\pi (g)`$ is relatively compact. In the case $`n=3`$ and $`K=`$ the conjecture implies (cf.\[Ma, §2\]) the Littlewood conjecture which states that
$$\underset{n\mathrm{}}{lim\; inf}nn\alpha n\beta =0$$
for all $`\alpha ,\beta `$, where $`x`$ denotes the distance from $`x`$ to $``$. In \[Ei-Ka-Li\], using the dynamical approach, M.Einsiedler, A.Katok and E.Lindenstrauss proved that the Littlewood conjecture fails at most on a set of Hausdorff dimension zero. Similar results in the $`p`$-adic setting have recently appeared in the M.Einsiedler and D.Kleinbock paper \[Ei-Kl\].
The paper is organized as follows. The notation and the terminology are introduced in §2. Our starting point is the paper \[To-We\]. In §3, using \[To-We\], we prove an $`𝒮`$-adic compactness criterium in terms of intersections of so-called quasiballs with horospherical subsets. In §4 we prove Proposition 4.3 which plays a crucial role in revealing the dichotomy in Corollary 1.3. In §5 we describe the locally divergent orbits in terms of minimal parabolic $`K`$-algebras. In order to do this, we have to apply more intrinsic arguments than in \[To-We, §5\] for the proof of a similar result. For instance, the Galois type arguments are replaced by Proposition 5.4 from the algebraic group theory. Theorems 1.1, 1.4 and their corollaries are proved in §6. The proof of Theorem 1.9 is given in §7.
The author is grateful to Manfred Einsiedler, Dima Kleinbock, Gregory Margulis and Barak Weiss for the useful discussions and to the Max Planck Institut für Mathematik, where the main part of this work was accomplished, for its hospitality.
## 2. Preliminaries: notation and basic concepts
### 2.1. Numbers
As usual $``$, $``$, $``$ and $``$ denote the complex, real, rational and integer numbers, respectively.
In this paper $`K`$ denotes a number field, that is, a finite extension of $``$. All valuations of $`K`$ which we consider are supposed to be normalized (see \[Ca-F, ch.2, §7\]) and, therefore, pairwise non-equivalent. If $`v`$ is a valuation of $`K`$ then $`K_v`$ is the completion of $`K`$ with respect to $`v`$ and $`|.|_v`$ is the corresponding norm on $`K_v`$. If $`v`$ is non-archimedean then $`𝒪_v=\{xK_v:|x|_v1\}`$ is the ring of integers of $`K_v`$.
We fix a finite set $`𝒮`$ of valuations of $`K`$ containing all archimedean valuations of $`K`$. The latter set is denoted by $`𝒮_{\mathrm{}}`$ or, simply, $`\mathrm{}`$, if this does not lead to confusion. We also put $`𝒮_f=𝒮𝒮_{\mathrm{}}`$.
We denote by $`𝒪`$ the ring of $`𝒮`$-integers of $`K`$, i.e., $`𝒪=K(_{v𝒮}𝒪_v)`$.
For any non-empty subset $``$ of $`𝒮`$, $`K_{}\stackrel{def}{=}_vK_v`$ is a direct product of locally compact fields. Note that $`K_{}`$ is a topological ring and that the diagonal imbedding of $`K`$ in $`K_{}`$ is dense. As usual, we denote by $`K_{}^{}`$ the multiplicative group of all invertible elements in the ring $`K_{}`$.
### 2.2. Norms
Let $`𝐕`$ be a finite dimensional vector space defined over $`K`$. For every $`𝒮`$ (respectively $`v𝒮`$) we write $`V_{}`$ for $`𝐕(K_{})`$ (respectively, $`V_v`$ for $`𝐕(K_v)`$). Fixing a basis of $`K`$-rational vectors $`e_1,\mathrm{},e_n`$, for every $`K`$-algebra $`A`$, we identify $`𝐕(A)`$ with $`A^n`$. For every $`v𝒮`$ we define a normalized norm $`_v`$ on $`V_v`$ as follows. If $`v`$ is real (respectively, complex) then $`_v`$ is the standard norm on $`^n`$ (respectively, the square of the standard norm on $`^n`$). If $`v`$ is non-archimedean, then $`_v`$ is defined by $`𝐱_v=\mathrm{max}_i|x_i|_v`$, where $`(x_1,\mathrm{},x_n)`$ are the coordinates of the vector $`𝐱V_v`$ with respect to the bases $`e_1,\mathrm{},e_n`$.
For $`\text{x}=(\text{x}^{(v)})_{v𝒮}`$ in $`V_{}`$ we define the norm of $`𝐱`$ as
$$𝐱_{}=\underset{v}{\mathrm{max}}𝐱^{(v)}_v.$$
Also, if $`=𝒮`$ we define the content of x as
$$𝐜_𝒮(\text{x})=\underset{v𝒮}{}𝐱^{(v)}_v.$$
Since all our norms are normalized and $`_{v𝒮}|\xi |_v=1`$ for every $`\xi 𝒪^{}`$ \[Ca-F, ch.2, Theorem 12.1\], we have that
(1)
$$𝐜_𝒮(\text{x})=𝐜_𝒮(\xi \text{x}),\xi 𝒪^{}.$$
By a pseudoball in $`V_𝒮`$ of radius $`r>0`$ centered at $`0`$ we mean the set $`_𝒮(r)=\{𝐱V_𝒮|𝐜_𝒮(\text{x})<r\}`$. We preserve the notation $`B_𝒮(r)`$ to denote the usual ball in $`V_𝒮`$ of radius $`r`$ centered at $`0`$ with respect to the norm $`._𝒮`$.
### 2.3. $`K`$-algebraic groups and their Lie algebras
We use boldface upper case letters to denote the algebraic groups and boldface lower case Gothic letters to denote their Lie algebras.
In this paper $`𝐆`$ is a reductive algebraic group defined over $`K`$. Recall that the Lie algebra $`𝖌`$ of $`𝐆`$ is equipped with a $`K`$-structure compatible with the $`K`$-structure of $`𝐆`$ \[Bo1, Theorem 3.4\]. An algebraic subgroup of $`𝐆`$ defined over $`K`$ is called shortly $`K`$-subgroup.
Given $`𝒮`$ and a $`K`$-subgroup $`𝐇`$ of $`𝐆`$, we usually denote $`H_{}\stackrel{def}{=}𝐇(K_{})`$ and $`𝔥_{}\stackrel{def}{=}𝖍(K_{})`$. The group $`H_{}`$ (respectively, its Lie algebra $`𝔥_{}`$) is identified with the direct product $`_vH_v`$ (respectively, $`_v𝔥_v`$), where $`H_v\stackrel{def}{=}𝐇(K_v)`$ (respectively, $`𝔥_v\stackrel{def}{=}𝖍(K_v)`$). But if $`=𝒮`$ and this does not lead to confusion we prefer the simpler notation $`H`$ (respectively, $`𝔥`$) for $`H_𝒮`$ (respectively, $`𝔥_𝒮`$).
We will use the notation $`\mathrm{pr}_{}`$ to denote both the natural projections $`GG_{}`$ and $`𝔤𝔤_{}`$. (The exact use of $`\mathrm{pr}_{\mathrm{}}`$ will follow from the context.)
On every $`G_v`$ we have a Zariski topology induced by the Zariski topology on $`𝐆`$ and a Hausdorff topology induced by the locally compact topology on $`K_v`$. The formal product of the Zariski (respectively, Hausdorff) topologies on $`G_v`$, $`v`$, is the Zariski (respectively, Hausdorff) topology on $`G_{}`$. In order to distinguish the two topologies, all topological notions connected with the first one will be used with the prefix ”Zariski”.
An element $`g=(g_v)_vG_{}`$ is called unipotent (respectively, semisimple) if each $`v`$-component $`g_v`$ of $`g`$ is unipotent (respectively, semisimple). A subgroup $`U`$ of $`G_{}`$ is called unipotent if it consists of unipotent elements. A subalgebra $`𝔲`$ of $`𝔤_{}`$ is unipotent if it corresponds to a Zariski closed unipotent subgroup $`U`$ of $`G_{}`$, i.e. if there exists a subgroup $`UG_{}`$ such that $`U=_vU_v`$, each $`U_v`$ is Zariski closed in $`G_v`$, and $`𝔲=_v𝔲_v`$ where $`𝔲_v`$ is the Lie algebra of $`U_v`$.
If $`𝐏`$ is a parabolic $`K`$-subgroup of $`𝐆`$ then $`R_u(𝐏)`$ denotes the unipotent radical of $`𝐏`$. The unipotent radical of the Lie algebra of $`𝐏`$ is by definition the Lie algebra of $`R_u(𝐏)`$.
If $`H`$ is a subgroup of $`G`$ then $`𝒩_G(H)`$ (respectively, $`𝒵_G(H)`$) denotes the normalizer (respectively, the centralizer) of $`H`$ in $`G`$.
For any non-empty $`𝒮`$ the adjoint representation $`\mathrm{Ad}_{}:G_{}\mathrm{GL}(𝔤_{})`$, where $`\mathrm{GL}(𝔤_{})=_v\mathrm{GL}(𝔤_v)`$, is the direct product of the adjoint representations $`\mathrm{Ad}_v:G_v\mathrm{GL}(𝔤_v)`$, $`v`$. We will use the notation $`\mathrm{Ad}`$ (respectively, $`\mathrm{Ad}_{\mathrm{}}`$) when $`=𝒮`$ (respectively, $`=𝒮_{\mathrm{}}`$).
### 2.4. $`𝒮`$-arithmetic subgroups
Recall that $`\mathrm{\Gamma }`$ is an $`𝒮`$-arithmetic subgroup of $`G`$, i.e., $`\mathrm{\Gamma }𝐆(𝒪)`$ has finite index in both $`\mathrm{\Gamma }`$ and $`𝐆(𝒪)`$. We assume that $`𝐆`$ is imbedded in $`\mathrm{𝐒𝐋}_n`$ in such a way that $`𝐆(𝒪)=\mathrm{𝐒𝐋}_n(𝒪)𝐆`$ and $`𝖌(𝒪)=`$ $`𝖘𝖑`$$`{}_{n}{}^{}(𝒪)𝖌`$. In particular, $`𝖌(𝒪)`$ is invariant under the adjoint action of $`𝐆(𝒪)`$. Let $`\mathrm{\Gamma }^{}`$ be a subgroup of finite index in $`\mathrm{\Gamma }`$ and let $`\varphi :G/\mathrm{\Gamma }^{}G/\mathrm{\Gamma }`$ be the natural map. Since $`\varphi `$ is a proper map it is easy to see that Theorems 1.1, 1.4 and their corollaries are valid for $`\mathrm{\Gamma }`$ if and only if they are valid for $`\mathrm{\Gamma }^{}`$. Therefore, we may suppose without loss of generality that $`\mathrm{\Gamma }=𝐆(𝒪)`$.
Let $`\pi :GG/\mathrm{\Gamma }`$ be the natural projection. For every $`xG/\mathrm{\Gamma }`$ we introduce the following notation. If $`x=\pi (g)`$, $`gG`$, we denote
$$𝔤_x=\mathrm{Ad}(g)𝖌(𝒪).$$
Since $`𝖌(𝒪)`$ is $`\mathrm{Ad}(\mathrm{\Gamma })`$-invariant, $`𝔤_x`$ does not depend on the choice of the element $`g`$.
## 3. Compactness criteria in $`𝒮`$-adic setting
### 3.1. $`𝒮`$-adic Mahler’s criterion
Let $`G=\mathrm{SL}_n(K_𝒮)`$, $`\mathrm{\Gamma }=\mathrm{SL}_n(𝒪)`$ and $`\pi :GG/\mathrm{\Gamma }`$ be the natural projection. The group $`G`$ is acting naturally on $`K_𝒮^n`$ and $`\mathrm{\Gamma }`$ is the stabilizer of $`𝒪^n`$ in $`G`$. If $`r>0`$ then $`B_𝒮(r)`$ (resp., $`_𝒮(r)`$) is the ball (resp. pseudoball) in $`K_𝒮^n`$ centered in $`0`$ and with radius $`r`$ (see §2.3).
We have
###### Theorem 3.1.
$`(\mathrm{Mahler}^{}\mathrm{s}\mathrm{criterion})`$ With the above notation, given a subset $`MG`$ the following conditions are equivalent:
1. $`\pi (M)`$ is relatively compact in $`G/\mathrm{\Gamma }`$;
2. There exists $`r>0`$ such that $`g𝒪^n_𝒮(r)=\{0\}`$ for all $`gM`$;
3. There exists $`r>0`$ such that $`g𝒪^nB_𝒮(r)=\{0\}`$ for all $`gM`$ .
The equivalence between (i) and (iii) is proved in \[Kl-To, Theorem 5.12\] and it is obvious that (ii) implies (iii). In order to prove that (iii) implies (ii) note that, in view of the formula (1), every $`_𝒮(r)`$ is invariant under the multiplication by elements from $`𝒪^{}`$. Now the implication easily follows from the following lemma:
###### Lemma 3.2.
There exists a constant $`\kappa >1`$ with the following property. Let $`\text{x}=(\text{x}^{(v)})_{v𝒮}K_𝒮^n`$ be such that $`\text{x}^{(v)}0`$ for all $`v𝒮`$. For each $`v𝒮`$ we choose a positive real number $`a_v`$ in such a way that $`𝐜_𝒮(\text{x})=_{v𝒮}a_v`$. Then there exists $`\xi 𝒪^{}`$ such that
(2)
$$\frac{a_v}{\kappa }\xi \text{x}^{(v)}_v\kappa a_v$$
for all $`v𝒮`$. In particular, for every x as above there exists $`\xi 𝒪^{}`$ such that
(3)
$$\frac{𝐜_𝒮(\text{x})^{1/m}}{\kappa }\xi \text{x}_𝒮\kappa 𝐜_𝒮(\text{x})^{1/m},$$
where $`m=\mathrm{\#}𝒮`$.
Proof. Let $`K_𝒮^1=\{y=(y^{(v)})K_𝒮^{}|_{v𝒮}|y^{(v)}|_v=1\}`$. Then $`𝒪^{}K_𝒮^1`$ and $`K_𝒮^1/𝒪^{}`$ is compact \[Ca-F, ch.2, Theorem 16.1\]. Therefore there exists a constant $`\kappa _0>1`$ such that for every $`y=(y^{(v)})K_𝒮^1`$ there exists $`\xi 𝒪^{}`$ such that
(4)
$$\frac{1}{\kappa _0}|\xi y^{(v)}|_v\kappa _0,v𝒮.$$
Let x and $`a_v,v𝒮`$, be as in the formulation of the proposition. There exists a constant $`c>1`$, depending only on $`𝒮`$, such that for every $`v𝒮`$ there exists $`\alpha ^{(v)}K_v^{}`$ with
(5)
$$\frac{c}{|\alpha ^{(v)}|_v}a_vc|\alpha ^{(v)}|_v$$
and $`_{v𝒮}|\alpha ^{(v)}|_v=_{v𝒮}a_v`$. So, $`𝐜_𝒮(\alpha ^1𝐱)=1`$ where $`\alpha =(\alpha ^{(v)})_{v𝒮}K_𝒮^{}`$. Put $`\kappa =\kappa _0c`$. In view of (4) and (5) there exists $`\xi 𝒪^{}`$ such that
$$\frac{|\alpha ^{(v)}|_v}{\kappa }|\xi 𝐱^{(v)}|_v\kappa |\alpha ^{(v)}|_v,v𝒮,$$
which proves (2).
In order to prove (3) it is enough to apply (2) with $`a_v=𝐜_𝒮(\text{x})^{1/n}`$. ∎
### 3.2. Horospherical subsets
We need to prove a compactness criterion which reflects the group structure of $`𝐆`$.
We generalize the notion of horospherical subset from \[To-We, Definition 3.4\].
###### Definition 3.3.
Let $`𝒮`$. A finite subset $``$ of $`𝔤_{}`$ is called $``$-horospherical (or, simply, horospherical when $``$ is implicit) if $`=\mathrm{pr}_{}(\mathrm{Ad}(g)(_0))`$, where $`gG`$ and $`_0`$ is a subset of $`𝖌(𝒪)`$ which spans linearly the unipotent radical of a maximal parabolic $`K`$-subalgebra of $`𝖌`$.
The next proposition provides a compactness criterion in terms of the intersection of pseudo-balls (and balls) in $`𝔤`$ with $`𝔤_x`$, $`xG/\mathrm{\Gamma }`$ (see 2.1 for the notation). It generalizes \[To-We, Propositions 3.3 and 3.5\].
###### Proposition 3.4.
Assume that $`𝐆`$ is a semisimple algebraic group. Then the following assertions hold:
1. There exists $`r>0`$ $`(`$respectively, $`t>0`$$`)`$ such that for any $`x=\pi (g)`$ the subalgebra of $`𝔤`$ spanned by $`_𝒮(r)𝔤_x`$ $`(`$respectively, $`B_𝒮(t)𝔤_x`$$`)`$ is unipotent;
2. $`(\mathrm{𝐂𝐨𝐦𝐩𝐚𝐜𝐭𝐧𝐞𝐬𝐬}\mathrm{𝐂𝐫𝐢𝐭𝐞𝐫𝐢𝐨𝐧})`$ A subset $`M`$ of $`G/\mathrm{\Gamma }`$ is relatively compact if and only if there exists $`r>0`$ $`(`$respectively, $`t>0`$$`)`$ such that $`_𝒮(r)𝔤_x`$ $`(`$respectively, $`B_𝒮(t)𝔤_x`$$`)`$ does not contain a horospherical subset for any $`xM`$.
### 3.3. Proof of Proposition 3.4
For every $`t>0`$ we let $`r=\left(\frac{t}{\kappa }\right)^m`$, where $`\kappa `$ and $`m`$ are as in the formulation of Lemma 3.2. It follows from Lemma 3.2 that
$$B_𝒮(t/\kappa )_𝒮(r)𝒪^{}B_𝒮(t).$$
Now the validity of the proposition for the balls $`B_𝒮(t)`$ implies easily its validity for the pseudoballs $`_𝒮(r)`$.
Further on, the proof of the proposition breaks in two cases. (In view of 2.4, we will assume that $`\mathrm{\Gamma }=𝐆(𝒪)`$.)
#### 3.3.1. The case $`𝒮=𝒮_{\mathrm{}}`$
Let $`\mathrm{R}_{\mathrm{K}/}`$ be the Weil restriction of scalars functor. Then $`𝐇=\mathrm{R}_{\mathrm{K}/}(𝐆)`$ is a semisimple $``$-algebraic group and $`𝖍=\mathrm{R}_{\mathrm{K}/}(𝖌)`$ is its $``$-Lie algebra. Denote $`\mathrm{\Delta }=𝐇()`$, $`H=𝐇()`$ and $`𝔥=𝖍()`$. The following properties of the functor $`\mathrm{R}_{\mathrm{K}/}`$ are well known and easily follow from its definition (see, for example, \[Pl-R, ch.2, §2.1.1\]). There exist continuous isomorphisms $`\mu :GH`$ and $`\nu :𝔤𝔥`$ such that $`\mu (\mathrm{\Gamma })=\mathrm{\Delta }`$, $`\nu (𝖌(𝒪))=𝖍()`$ and
$$\nu (\mathrm{Ad}_G(g)x)=\mathrm{Ad}_H(\mu (g))\nu (x)$$
for all $`gG`$ and $`x𝔤`$. Moreover, $`\nu `$ maps bijectively the family of the horospherical subsets of $`𝔤`$ to the family of the horospherical subsets of $`𝔥`$ and $`\mu `$ induces an homeomorphism $`G/\mathrm{\Gamma }H/\mathrm{\Delta }`$. Hence, when $`𝒮=𝒮_{\mathrm{}}`$ the proposition follows from the case $`K=`$ considered in \[To-We, Propositons 3.3 and 3.5\].
#### 3.3.2. The case $`𝒮𝒮_{\mathrm{}}`$
We introduce the topological rings $`𝒪_f\stackrel{def}{=}_{v𝒮_f}𝒪_v`$ and $`K_f\stackrel{def}{=}K_{\mathrm{}}\times 𝒪_f`$ (see 2.1). So, $`𝒪_{\mathrm{}}=𝒪(K_{\mathrm{}}\times 𝒪_f)`$ is the ring of integers of $`K`$.
If $`\stackrel{~}{𝐆}`$ is the simply connected covering of the algebraic group $`𝐆`$ then $`\stackrel{~}{G}/\stackrel{~}{\mathrm{\Gamma }}`$ is naturally homeomorphic to $`G/\mathrm{\Gamma }`$, where $`\stackrel{~}{G}=\stackrel{~}{𝐆}(K_𝒮)`$ and $`\stackrel{~}{\mathrm{\Gamma }}=\stackrel{~}{𝐆}(𝒪)`$. In view of this and of Theorem 4.1 below, we may (and will) assume without loss of generality that $`𝐆`$ is simply connected and without $`K`$-anisotropic factors. Then the diagonal imbedding of $`\mathrm{\Gamma }`$ into $`_{v𝒮_f}𝐆(K_v)`$ is dense. (This fact follows immediately from the strong approximation theorem \[Pl-R, Theorem 7.12\].) Therefore
$$G=𝐆(K_f)\mathrm{\Gamma }.$$
Every $`gG`$ can be writhen in the following way
(6)
$$g=g_{\mathrm{}}g_f\gamma ,$$
where $`g_{\mathrm{}}G_{\mathrm{}},g_f𝐆(𝒪_f)`$ and $`\gamma \mathrm{\Gamma }`$. Let $`\mathrm{\Gamma }_{\mathrm{}}=𝐆(K_f)\mathrm{\Gamma }`$. Then $`G/\mathrm{\Gamma }`$ is homeomorphic to $`𝐆(K_f)/\mathrm{\Gamma }_{\mathrm{}}`$ and the projection of $`𝐆(K_f)`$ on $`G_{\mathrm{}}`$ yields the following map
$$\phi :G/\mathrm{\Gamma }G_{\mathrm{}}/\mathrm{\Gamma }_{\mathrm{}},\phi (\pi (g))\stackrel{def}{=}\pi _{\mathrm{}}(g_{\mathrm{}}),gG,$$
where $`\pi _{\mathrm{}}:G_{\mathrm{}}G_{\mathrm{}}/\mathrm{\Gamma }`$ is the natural map. In view of the compactness of $`𝐆(𝒪_f)`$, $`\phi `$ is a proper continuous map.
Let $`A`$ be a subset of $`𝔤_{\mathrm{}}`$ and $`x=\pi (g)`$ for some $`gG`$. Set $`A_f=A\times 𝖌(𝒪_f)`$. Using (6) and the fact that $`𝖌(𝒪)`$ is invariant under the adjoint action of $`\mathrm{\Gamma }`$, we obtain
(7)
$$\begin{array}{cc}\hfill \mathrm{pr}_{\mathrm{}}(𝔤_xA_f)=\mathrm{pr}_{\mathrm{}}(\mathrm{Ad}_𝒮& (g)\left(𝖌(𝒪)\right)A_f)=\hfill \\ \hfill \mathrm{pr}_{\mathrm{}}(\mathrm{Ad}_𝒮(g_{\mathrm{}}g_f)(𝖌(𝒪)& A_f))=𝔤_{\mathrm{},y}A,\hfill \end{array}$$
where $`y=\phi (x)`$ and $`𝔤_{\mathrm{},y}=\mathrm{Ad}_{\mathrm{}}(g_{\mathrm{}})𝖌(𝒪_{\mathrm{}})`$. (Recall that $`\mathrm{pr}_{\mathrm{}}`$ denotes the natural projection $`𝔤𝔤_{\mathrm{}}`$.)
Let $`\stackrel{~}{B}(t)=B_{\mathrm{}}(t)\times 𝖌(𝒪_f)`$. Applying (7) with $`A=B_{\mathrm{}}(t)`$, we get
$$\mathrm{pr}_{\mathrm{}}\left(𝔤_x\stackrel{~}{B}(t)\right)=𝔤_{\mathrm{},y}B_{\mathrm{}}(t).$$
Since the restriction of $`\mathrm{pr}_{\mathrm{}}`$ to $`𝔤_x`$ is injective, we obtain that the subalgebra spanned by $`𝔤_x\stackrel{~}{B}(t)`$ is unipotent if and only if the subalgebra spanned by $`𝔤_{\mathrm{},y}B_{\mathrm{}}(t)`$ is unipotent. This, in view of 3.3.1, proves (a).
Let us prove (b). If $`M`$ is compact, it follows from the continuity of the adjoint action that if $`t>0`$ is sufficiently small then $`B_𝒮(t)𝔤_x`$ does not contain horospherical subsets for all $`xG/\mathrm{\Gamma }`$. In order to prove the inverse implication, let $`MG/\mathrm{\Gamma }`$ and $`t>0`$ be such that $`B_𝒮(t)𝔤_x`$ does not contain horospherical subsets for any $`xM`$. Assume the contrary, that is, that there exists a divergent sequence $`\{x_i\}`$ of elements in $`M`$. Then the sequence $`\{y_i=\phi (x_i)\}`$ is divergent in $`G_{\mathrm{}}/\mathrm{\Gamma }_{\mathrm{}}`$ (because $`\phi `$ is proper). Since the proposition is true for $`G_{\mathrm{}}/\mathrm{\Gamma }_{\mathrm{}}`$, for every $`\epsilon >0`$ there exists $`i0`$ such that $`B_{\mathrm{}}(\epsilon )𝔤_{\mathrm{},y_i}`$ contains a horospherical subset. Set $`\stackrel{~}{B}(\epsilon )=B_{\mathrm{}}(\epsilon )\times 𝖌(𝒪_f)`$. By (7) (applied with $`A=B_{\mathrm{}}(\epsilon )`$) and the injectivity of the restriction of $`\mathrm{pr}_{\mathrm{}}`$ to $`𝔤_x`$, we obtain that $`\stackrel{~}{B}(\epsilon )𝔤_{x_i}`$ contains a horospherical subset. Now, using Lemma 3.2, we conclude that $`B_𝒮(t)𝔤_{x_i}`$ contains horospherical subsets for all sufficiently large $`i`$. Contradiction. ∎
### 3.4. Expanding transformations
For every $`v𝒮`$, we fix a maximal $`K_v`$-split torus $`𝐓_v`$ of $`𝐆`$. We denote $`T_v=𝐓_v(K_v)`$ and $`T_{}=_vT_v`$ where $``$ is a non empty subset of $`𝒮`$.
###### Proposition 3.5.
With the above notation, for every real $`\tau >1`$ there exists a finite set $`t_1,\mathrm{},t_s`$ of elements in $`T_{}`$ such that if $`𝔲`$ is a unipotent subalgebra of $`𝔤_{}`$ then there exists an element $`t_i`$ such that
(8)
$$\mathrm{Ad}(t_i)(𝐱)_{}\tau 𝐱_{}$$
for all $`𝐱𝔲`$.
Proof. It is easy to see that it is enough to prove the proposition when $``$ is a singleton. Let $`=\{v\}`$. If $`v`$ is real then the proposition is proved in \[To-We, Proposition 4.1\]. Here we present a shorter proof for an arbitrary $`v`$.
Let $`𝔲_v^+`$ and $`𝔲_v^{}`$ be invariant under the adjoint action of $`T_v`$ maximal unipotent subalgebras of $`𝔤_v`$ which are opposite to each other. Then
(9)
$$C_v=\{dT_v|\underset{n+\mathrm{}}{lim}\mathrm{Ad}(d^n)x=\mathrm{},x𝔲_v^+\}$$
is the interior of the Weil chamber corresponding to $`𝔲_v^+`$ (see \[Bo1\]). Denote by $`U_v^+`$ and $`U_v^{}`$ the unipotent subgroups of $`G_v`$ with Lie algebras $`𝔲_v^+`$ and $`𝔲_v^{}`$, respectively.
Now let $`𝔲_v`$ be any maximal unipotent subalgebra of $`𝔤_v`$. There exists $`gG_v`$ such that $`\mathrm{Ad}(g)𝔲_v^+=𝔲_v`$. By Bruhat decomposition $`g=a\omega b`$, where $`\omega 𝒩_{G_v}(T_v)`$, $`a`$ and $`bU_v^+`$ and $`\omega ^1a\omega U_v^{}`$. We can write $`𝔲_v=\mathrm{Ad}(\omega a^{})𝔲_v^+`$, where $`a^{}=\omega ^1a\omega `$. Let $`x𝔲_v^+`$ and $`f_vC_v`$. We put $`y=\mathrm{Ad}(\omega a^{})x`$ and $`d_v=\omega f_v\omega ^1`$. Using (9) and the fact that $`lim_{n+\mathrm{}}f_v^na^{}f_v^n=0`$, we get
$`\underset{n+\mathrm{}}{lim}\mathrm{Ad}(d_v^n)y=\underset{n+\mathrm{}}{lim}\mathrm{Ad}(\omega (f_v^na^{}f_v^n))\mathrm{Ad}(f_v^n)(x)=\mathrm{}.`$
Therefore, taking $`t=d^n`$ with $`n`$ sufficiently large, we obtain that
(10)
$$\mathrm{Ad}(t)z_v>\tau z_v$$
for all non-zero $`z𝔲_v`$.
Since the stabilizer of every maximal unipotent subalgebra is a minimal parabolic subgroup and all minimal parabolic subgroups are conjugated, the set of all maximal unipotent subalgebras can be identified with the compact homogeneous space $`G_v/P_v^+`$, where $`P_v^+`$ is the parabolic subgroup of $`G_v`$ with Lie algebra $`𝔲_v^+`$. It is easy to see that (10) is true for all subalgebras in a neighborhood of $`𝔲_v`$. Now the existence of the elements $`t_1,\mathrm{},t_s`$ as in the formulation of the theorem follows from the compactness of $`G_v/P_v^+`$ by a standard argument. ∎
## 4. Closed orbits of reductive $`K`$-groups
### 4.1. Reductive groups
Recall the $`𝒮`$-adic version of a well-known theorem of Borel and Harish-Chandra. (As usual, $`G=𝐆(K_𝒮)`$ and $`\mathrm{\Gamma }=𝐆(𝒪)`$.)
###### Theorem 4.1.
(cf.\[Pl-R, Theorem 5.7 \]) Let $`𝐆`$ be a reductive $`K`$-group and let $`\mathrm{X}_K(𝐆)`$ be the group of $`K`$-rational characters of $`𝐆`$. Then
1. $`G/\mathrm{\Gamma }`$ has a finite invariant volume if and only if $`\mathrm{X}_K(𝐆)=\{1\}`$;
2. $`G/\mathrm{\Gamma }`$ is compact if and only if $`𝐆`$ is anisotropic over $`K`$.
Because of the lack of appropriate reference we will prove the following known proposition.
###### Proposition 4.2.
With the above notation, let $`𝐇`$ be a reductive subgroup of $`𝐆`$ defined over $`K`$ and $`H=𝐇(K_𝒮)`$. Then $`H\pi (e)`$ is closed in $`G/\mathrm{\Gamma }`$.
Proof. Using the Weil restriction of scalars, one can reduce the proof to the case when $`K=`$. In view of \[Bo2, Proposition 7.7\] there exists a $``$-rational action of $`𝐆`$ on an affine $``$-variety $`𝐕`$ admitting an element $`a𝐕()`$ such that $`𝐇=\{g𝐆|ga=a\}`$. Since the map $`𝐆𝐕`$, $`gga`$, is polynomial with rational coefficients, there exists a non-zero integer $`n`$ such that $`\gamma na𝐕(𝒪)`$ for all $`\gamma \mathrm{\Gamma }`$. Therefore $`\mathrm{\Gamma }H`$ is closed in $`G`$, equivalently, $`H\pi (e)`$ is closed. ∎
### 4.2. Algebraic tori
We will need the following
###### Proposition 4.3.
Let $`𝐓`$ be a $`K`$-torus in $`𝐆`$ and let $``$ be a non-empty subset of $`𝒮`$. Suppose that $`T_{}`$ is not compact. Then the orbit $`T_{}\pi (e)`$ is divergent if and only if the following conditions are fulfilled:
1. $`=\{v_{}\}`$ is a singleton, and
2. $`\mathrm{rank}_K𝐓=\mathrm{rank}_{K_v_{}}𝐓>0`$.
Proof. In view of Proposition 4.2 the orbit $`𝐓(K_𝒮)\pi (e)`$ is closed and, therefore, homeomorphic to $`𝐓(K_𝒮)/(𝐓(K_𝒮)\mathrm{\Gamma })`$. So, we may suppose, with no loss or generality, that $`𝐓=𝐆`$.
Assume that the orbit $`T_{}\pi (e)`$ is divergent. Let $`𝐓_a`$ (respectively, $`𝐓_d`$) be the largest $`K`$-anisotropic (respectively, split over $`K`$) subtorus of $`𝐓`$. It is well known that $`𝐓`$ is an almost direct product of $`𝐓_a`$ and $`𝐓_d`$. This implies that if there exists $`v`$ such that $`\mathrm{rank}_{K_v}𝐓>\mathrm{rank}_K𝐓`$ then $`𝐓_a(K_{})`$ is not compact. But $`𝐓_a(K_𝒮)\pi (e)`$ is compact (Theorem 4.1). Therefore, $`T_{}\pi (e)`$ can not be divergent, a contradiction. So, $`\mathrm{rank}_{K_v}𝐓=\mathrm{rank}_K𝐓`$ for all $`v`$. In this case $`𝐓_a(K_{})`$ is compact and, since $`T_{}`$ is not compact, $`𝐓_d`$ is not trivial. Note that $`T_{}\pi (e)`$ is divergent if and only if $`𝐓_d(K_{})\pi (e)`$ is divergent.
In order to prove (i) consider the character group $`\mathrm{X}_K(𝐓)`$ of $`𝐓`$. It is well known that $`\mathrm{X}_K(𝐓)`$ is a free $``$-module of rank equal to $`dim𝐓_d`$ (cf. \[Bo1, 8.15\]). Let $`\chi _1,\mathrm{},\chi _r`$ be a basis of $`\mathrm{X}_K(𝐓)`$. Define a homomorphism of $`K`$-algebraic groups $`\chi =(\chi _1,\mathrm{},\chi _r):𝐓𝐆_m^r`$, where $`𝐆_m`$ denotes the one-dimensional $`K`$-split torus. Let $`T=𝐓(K_𝒮)`$ and $`T_{}=\{(t_v)_{v𝒮}T|_{v𝒮}|\chi _i(t_v)|_v=1\mathrm{for}\mathrm{all}i\}`$. It follows from \[Ca-F, ch.2, Theorem 16.1\] that $`\mathrm{\Gamma }`$ is a co-compact lattice in $`T_{}`$. Set $`\phi :T^r`$, $`\phi ((t_v)_{v𝒮})=(\mathrm{log}(_{v𝒮}|\chi _1(t_v)|_v),\mathrm{},`$ $`\mathrm{log}(_{v𝒮}|\chi _r(t_v)|_v))`$. It is clear that $`\phi `$ is a continuous surjective homomorphism of locally compact topological groups with $`\mathrm{ker}(\phi )=T_{}`$. Since $`T_{}/\mathrm{\Gamma }`$ is compact, $`\phi `$ induces a proper homomorphism $`\psi :T/\mathrm{\Gamma }T/T_{}`$. Now let $``$ contain two different valuations $`v_1`$ and $`v_2`$. It is easy to find sequences $`\{a_i\}`$ in $`K_{v_1}^{}`$ and $`\{b_i\}`$ in $`K_{v_2}^{}`$ such that $`\mathrm{log}|a_i|_{v_1}+\mathrm{}`$, $`\mathrm{log}|b_i|_{v_2}\mathrm{}`$ and the sequence $`\{\mathrm{log}|a_i|_{v_1}+\mathrm{log}|b_i|_{v_2}\}`$ is bounded. We define a sequence $`\{s_i=(s_i^{(v)})_v\}`$ in $`T_{}`$ as follows:
$$s_i^{(v)}=\{\begin{array}{cc}1,\text{if}v\{v_1,v_2\};\hfill & \\ \chi _1(s_i^{(v_1)})=a_i\text{and}\chi _j(s_i^{(v_1)})=1\text{for all}j>1;\hfill & \\ \chi _1(s_i^{(v_2)})=b_i\text{and}\chi _j(s_i^{(v_2)})=1\text{for all}j>1.\hfill & \end{array}$$
We have that $`\{s_i\}`$ is unbounded and that $`\{\phi (s_i)\}`$ is bounded. (Recall that $`T_{}`$ is considered as a subgroup of $`T`$, so that the notation $`\phi (s_i)`$ makes sense.) Since $`\psi `$ is proper, $`s_i\pi (e)`$ is bounded. Therefore the orbit $`T_{}\pi (e)`$ is not divergent. This contradiction completes the proof of (i).
Assume that $``$ contains only one valuation $`v_{}`$ and that $`\mathrm{rank}_K𝐓=\mathrm{rank}_{K_v_{}}𝐓>0`$. It follows from the above definition of $`\phi `$ and the fact that $`\chi `$ is an homomorphism with compact kernel, that if a sequence $`\{t_i\}`$ in $`T_{}`$ diverges then $`\{\phi (t_i)\}`$ does too. Therefore $`T_{}\pi (e)`$ is a divergent orbit. ∎
Proposition 4.3 implies:
###### Proposition 4.4.
Let $`𝐓`$ be a $`K`$-torus and let $``$ be a non-empty subset of $`𝒮`$. Then the orbit $`T_{}\pi (e)`$ is closed if and only if one of the following conditions holds:
1. $`=𝒮`$;
2. $`\mathrm{rank}_{K_v}𝐓=0`$ for all $`v`$, equivalently, $`T_{}`$ is compact;
3. $`=\{v_{}\}`$ and $`\mathrm{rank}_K𝐓=\mathrm{rank}_{K_v_{}}𝐓`$.
Proof. Note that if $`𝒮`$ and $`T_{}`$ is not compact then $`T_{}\pi (e)`$ is closed if and only if it is divergent. Now the proposition follows easily from Proposition 4.3. ∎
## 5. Parabolic subgroups and divergent orbits
### 5.1. Main proposition
Recall that, given a subset $`𝒮`$, we use the notation $`\mathrm{pr}_{}`$ to denote depending on the context the projection $`GG_{}`$ or the projection $`𝔤𝔤_{}`$.
The goal of this section is to prove the following
###### Proposition 5.1.
Let $`𝐆`$ be a reductive $`K`$-algebraic group, $``$ be a non-empty subset of $`𝒮`$, $`g=(g_v)_{v𝒮}G`$ and $`x=\pi (g)`$. Assume that $`\mathrm{rank}_K𝐆>0`$ and that for every minimal parabolic $`K`$-subalgebra $`𝖇`$ of $`𝖌`$ containing the Lie algebra of $`𝐃`$ there exists a horospherical subset $`_𝔟`$ of $`𝔤_{}`$ such that $`_𝔟\mathrm{pr}_{}(𝔤_x)𝔟_{}`$. Then the following assertions hold:
1. For every $`v`$ the orbit $`D_v\pi (g)`$ is divergent;
2. If $`g_{}=\mathrm{pr}_{}(g)`$ then
(11)
$$g_{}𝒵_G_{}(D_{})\mathrm{pr}_{}(𝐆(K));$$
3. There exists a maximal $`K`$-split torus $`𝐒`$ of $`𝐆`$ such that
(12)
$$S_v=g_{v}^{}{}_{}{}^{1}D_vg_v$$
for all $`v`$, where $`S_v=𝐒(K_v)`$.
In order to prove Proposition 5.1 we will need some facts from algebraic group theory.
### 5.2. Intersections of parabolic subgroups
The next three propositions remain valid for any field $`K`$.
###### Proposition 5.2.
\[Bo1, Propositions 14.22 and 21.13\] Let $`𝐏`$ and $`𝐐`$ be parabolic $`K`$-subgroups of $`𝐆`$.
1. $`(𝐏𝐐)R_u(𝐏)`$ is a parabolic $`K`$-subgroup;
2. If $`𝐐`$ is conjugate to $`𝐏`$ and contains $`R_u(𝐏)`$ then $`𝐐=𝐏`$.
We also have
###### Proposition 5.3.
\[To-We, Proposition 5.2\] For every minimal parabolic $`K`$-subgroup $`𝐁`$ containing $`𝐃`$ we let $`𝐏_𝐁`$ be a proper parabolic $`K`$-subgroup containing $`𝐁`$. Then
(13)
$$\underset{𝐁}{}𝐏_𝐁=𝒵_𝐆(𝐃).$$
Keeping the notation and assumptions of Proposition 5.3, we prove:
###### Proposition 5.4.
Let $`n𝒩_𝐆(𝒵_𝐆(𝐃))`$. Assume that for every $`𝐁`$ the group $`n𝐏_𝐁n^1`$ is defined over $`K`$. Then $`n𝒩_𝐆(𝐃)`$. The projection of $`n`$ into the Weyl group $`\mathrm{W}_K=𝒩_𝐆(𝐃)/𝒵_𝐆(𝐃)`$ is uniquely defined by the map $`𝐁n𝐏_𝐁n^1`$.
Proof. The uniqueness of the projection of $`n`$ into $`\mathrm{W}_K`$ follows immediately from Proposition 5.3 and the fact that every parabolic subgroup coincides with its normalizer.
We will assume that for every $`𝐁`$ the group $`𝐏_𝐁`$ is minimal among the parabolic $`K`$-subgroups $`𝐏`$ containing $`𝐁`$ and such that $`n𝐏n^1`$ is defined over $`K`$.
Assume that there exists $`𝐁`$ such that $`𝐁=𝐏_𝐁`$. Let $`𝐁^{}=n𝐁n^1`$. Since all minimal parabolic $`K`$-subgroups are conjugated under the action of $`\mathrm{W}_K`$ and $`𝒩_𝐆(𝐃)=𝒩_𝐆(𝐃)(K)𝒵_𝐆(𝐃)`$ \[Bo1, Theorem 21.2\], there exists $`n_{}𝒩_𝐆(𝐃)(K)`$ such that $`𝐁=n_{}𝐁^{}n_{}^1`$. Therefore, $`𝐁=n_{}n𝐁(n_{}n)^1`$ which implies that $`n_{}n𝐁`$. Since $`𝒩_𝐆(𝐃)𝒩_𝐆(𝒵_𝐆(𝐃))`$, we get $`n_{}n𝒩_𝐁(𝒵_𝐆(𝐃))`$. Now, the proposition follows from the fact that $`𝒵_𝐆(𝐃)=𝒩_𝐁(𝒵_𝐆(𝐃))`$ \[Bo1, Corollary 14.19\].
Assume that $`𝐏_𝐁𝐁`$ for all $`𝐁`$. Choose a $`𝐏_𝐁`$ with the minimal dimension and set $`𝐏=𝐏_𝐁`$. Let $`\mathrm{\Phi }(𝐃,𝐆)`$ be the relative root system of $`𝐆`$ with respect to $`𝐃`$. (See \[Bo1, 21.1 and 8.17\] for the standard definition of a system of $`K`$-roots.) Since $`𝐏𝐁`$, there exists a long root $`\alpha \mathrm{\Phi }(𝐃,𝐆)`$ such that $`\pm \alpha `$ are roots of the group $`𝐏`$ with respect to $`𝐃`$. Recall that all roots of the same length in $`\mathrm{\Phi }(𝐃,𝐆)`$ are conjugated under the action of $`\mathrm{W}_K`$ \[Hu, 10.4, Lemma C and 10.3, Theorem\]. Therefore there exists a minimal parabolic $`K`$-subgroup $`𝐁^+`$ containing $`𝐃`$ such that $`\alpha `$ is a maximal long root of $`𝐁^+`$ relative to $`𝐃`$. Let $`\mathrm{\Delta }^+`$ be the set of simple roots corresponding to $`𝐁^+`$. Then in the expression of $`\alpha `$ as a linear combination of the roots in $`\mathrm{\Delta }^+`$ all coefficients are strictly positive \[Hu, 10.4, Lemma A\]. It follows from the explicit description of the standard parabolic $`K`$-subgroups (see \[Bo1, 21.11\]), that $`\alpha `$ is not a root of any parabolic $`K`$-subgroup containing $`𝐁^+`$. Similarly, $`\alpha `$ is not a root of any parabolic $`K`$-subgroup containing $`𝐁^{}`$, where $`𝐁^{}`$ is the minimal parabolic $`K`$-subgroup opposite to $`𝐁^+`$. As a consequence, one of the $`K`$-subgroups $`(𝐏_{𝐁^+}𝐏)R_u(𝐏)`$ or $`(𝐏_𝐁^{}𝐏)R_u(𝐏)`$ is strictly smaller than $`𝐏`$. Let $`𝐏(𝐏_{𝐁^+}𝐏)R_u(𝐏)`$. Since $`(𝐏_{𝐁^+}𝐏)R_u(𝐏)`$ is a parabolic $`K`$-subgroup (Proposition 5.2(i)) and $`n(𝐏_{𝐁^+}𝐏)R_u(𝐏)n^1`$ is defined over $`K`$. The latter contradicts the choice of $`𝐏`$, which completes our proof. ∎
###### Remark 5.5.
In connection with the above proposition, let us note that in certain cases $`𝒩_𝐆(𝐃)𝒩_𝐆(𝒵_𝐆(𝐃))`$. As a simple example one can consider the special unitary group $`\mathrm{𝐒𝐔}_3(h)`$, where $`h`$ is an hermitian form with coefficients from $`K`$ of indice 1. This is a quasisplit group of type $`A_2`$. Therefore $`𝒩_𝐆(𝒵_𝐆(𝐃))/𝒵_𝐆(𝐃)`$ is isomorphic to the symmetric group $`S_3`$ and $`𝒩_𝐆(𝐃)/𝒵_𝐆(𝐃)`$ is a group of order two.
### 5.3. Proof of Proposition 5.1
We start the proof with a general remark. We keep the notation from the formulation of the proposition. For every $`𝖇`$ there exists a finite subset $`_𝔟^{}`$ of $`𝖌(𝒪)`$ which spans linearly the unipotent radical of a maximal parabolic $`K`$-subgroup $`𝐏_𝔟^{}`$ of $`𝐆`$ and such that $`_𝔟=\mathrm{pr}_{}(\mathrm{Ad}(g)(_𝔟^{})`$. So, if $`v`$, we have
$$g_vR_u(𝐏_𝔟^{})(K_v)g_v^1𝐁(K_v),$$
where $`𝐁`$ is the $`K`$-algebraic subgroup of $`𝐆`$ the Lie algebra of which is $`𝖇`$. It follows from Proposition 5.2(ii) that there exists a parabolic $`K`$-subgroup $`𝐏_𝔟`$ containing $`𝐁`$ such that
(14)
$$𝐏_𝔟=g_v𝐏_𝔟^{}g_v^1$$
for all $`v`$.
Let us prove (a). (Remark that (a) follows a posteriori from (b) and Proposition 4.3.) Fix $`v`$. We want to prove that the orbit $`D_v\pi (g)`$ diverges. Let $`\{d_i\}`$ be a divergent sequence in $`D_v`$. Put $`s_i=g_v^1d_ig_v`$. It is enough to prove that the sequence $`\{s_i\pi (e)\}`$ is divergent. Passing to a subsequence we may assume that $`\{d_i^1\}`$ belongs to the Weyl chamber corresponding to some minimal parabolic $`K`$-subgroup $`𝐁`$. Let $`𝖚`$ be the Lie algebra of $`R_u(𝐏_𝔟^{})`$. Let $`m`$ be the dimension of $`𝖚`$ and let $`^m\mathrm{Ad}`$ be the adjoint representation of $`𝐆`$ on the $`m`$-th exterior power $`^m𝖌`$. Since $`𝖚`$ is defined over $`K`$, there exists a non-zero $`K`$-rational vector $`z^m𝖌`$ corresponding to $`𝖚`$. It is known (see the proof of Proposition 5.4) that if $`\alpha `$ is a maximal root of $`𝐁`$ with respect to $`𝐃`$ then $`\alpha `$ is a root of every standard parabolic subgroup containing $`𝐁`$ and, given the choice of $`\{d_i\}`$, $`lim_i\mathrm{}\alpha (d_i)=0`$. Since $`𝐏_𝔟=g_v𝐏_𝔟^{}g_v^1`$ and $`𝐏_𝔟`$ is a parabolic containing $`𝐁`$, we obtain that
$$\underset{i\mathrm{}}{lim}\stackrel{m}{}\mathrm{Ad}(d_i)g_vz_v=0.$$
This implies
$$\underset{i\mathrm{}}{lim}𝐜_𝒮(\stackrel{m}{}\mathrm{Ad}(s_i)z)=0.$$
It follows from Theorem 3.1 (ii) that $`\{s_i\pi (e)\}`$ diverges. This completes the proof of (a).
Note that (c) follows immediately from (b). So, it remains to prove (b). Let $`𝐏_𝔟^{}`$ be as above. Set $`𝐇=_𝔟𝐏_𝔟^{}`$. Since $`𝐏_𝔟`$ is a $`K`$-parabolic subgroup of $`𝐆`$ containing $`𝐁`$, in view of Proposition 5.3, we get that
(15)
$$𝐇=\underset{𝔟}{}g_v^1𝐏_𝔟g_v=g_v^1\left(\underset{𝔟}{}𝐏_𝔟\right)g_v=g_v^1𝒵_𝐆(𝐃)g_v$$
for all $`v`$.
Note that the groups $`𝒵_𝐆(𝐃)`$ and $`𝐇`$ are reductive and defined over $`K`$. Let $`𝐙`$ (respectively, $`𝐙^{}`$) be the Zariski connected component of the center of $`𝒵_𝐆(𝐃)`$ (respectively, $`𝐇`$). It follows from (15) that
(16)
$$𝐙^{}=g_v^1𝐙g_v$$
for all $`v`$. Since $`𝐃`$ is a maximal $`K`$-split torus of $`𝐆`$, we have that $`𝐃=𝐙_d`$, where $`𝐙_d`$ is the largest $`K`$-split subtorus of $`𝐙`$.
Denote by $`𝐙_d^{}`$ the largest $`K`$-split subtorus of $`𝐙^{}`$ and assume that $`𝐙_d^{}`$ is not maximal in $`𝐆`$. Let $`𝐙_a^{}`$ be the largest $`K`$-anisotropic subtorus of $`𝐙^{}`$. Fix $`v`$. Since every $`K`$-torus is an almost direct product over $`K`$ of its largest $`K`$-split and its largest $`K`$-anisotropic subtori \[Bo1, Proposition 8.15\], it follows from (16) that there exists an element $`t𝐙_a^{}(K_v)g{}_{v}{}^{}{}_{}{}^{1}𝐃(K_v)g_v`$ such that $`\{t^n|n\}`$ is a divergent sequence. In view of (a), $`\{g_vt^ng{}_{v}{}^{}{}_{}{}^{1}\pi (g)\}`$, and therefore $`\{t^n\pi (e)\}`$, are also divergent sequences. The latter contradicts the fact that the orbit $`𝐙_a^{}(K_{})\pi (e)`$ is compact (see Theorem 4.1). Therefore $`𝐙_d^{}`$ is a maximal $`K`$-split torus of $`𝐆`$.
Since the maximal $`K`$-split tori are conjugated under $`𝐆(K)`$ \[Bo1, Theorem 20.9\], there exists $`q𝐆(K)`$ such that $`𝐙_d^{}=q^1𝐃q`$. Also, $`𝒵_𝐆(𝐙_d^{})=q^1𝒵_𝐆(𝐃)q`$, $`𝒵_𝐆(𝐙_d^{})𝐇`$ and $`dim𝐇=dim𝒵_𝐆(𝐃)`$. Therefore,
$$𝐇=q^1𝒵_𝐆(𝐃)q.$$
In view of (15), we have
$$g_vq^1𝒩_𝐆(𝒵_𝐆(𝐃)),v.$$
Given $`v`$, the group
$$qg_v^1𝐏_𝔟(qg_v^1)^1=q𝐏_𝔟^{}q^1$$
is defined over $`K`$ for every $`𝔟`$. It follows from Proposition 5.4 that there exists $`n𝒩_𝐆(𝐃)(K)`$ such that
(17)
$$nqg_v^1𝒵_𝐆(𝐃),v.$$
Since $`n`$ is the same for all $`v`$, (17) implies (11), which completes the proof. ∎
## 6. Proofs of Theorem 1.4 and of its corollaries
### 6.1. Proof of Theorem 1.4.
Let the conditions (i)-(iii) in the formulation of the theorem hold. Since $`\mathrm{rank}_{K_v}𝐆\mathrm{rank}_K𝐆`$, it follows from (ii) that $`\mathrm{rank}_{K_v}𝐆=\mathrm{rank}_K𝐆`$ for all $`v`$. Therefore, $`T_{}/D_{}`$ is compact. So, $`T_{}\pi (g)`$ is closed and locally divergent if and and only if $`D_{}\pi (g)`$ has this property. In view of (iii), $`g^1D_{}g=\stackrel{~}{D}_{}`$, where $`\stackrel{~}{D}_{}=\stackrel{~}{𝐃}(K_{})`$ and $`\stackrel{~}{𝐃}`$ is a $`K`$-split torus. Using (i) and Proposition 4.3, it is easy to see that $`\stackrel{~}{D}_{}\pi (g)`$, and therefore $`D_{}\pi (g)`$, are closed locally divergent orbits.
Let the orbit $`T_{}\pi (e)`$ be closed and locally divergent. In view of Theorem 4.1(b), $`\mathrm{rank}_K𝐆>0`$. Moreover, since every $`𝐓_v`$ is a product of a maximal $`K_v`$-split torus and a compact, we can suppose without loss of generality that $`𝐓_v`$ is a maximal $`K_v`$-split torus.
Denote by $`𝐒`$ the connected component of the Zariski closure of $`g^1T_{}g\mathrm{\Gamma }`$ in $`𝐆`$. Suppose that $`𝐒`$ is not trivial. Then $`=𝒮`$. Set $`S=𝐒(K_𝒮)`$. Since $`S`$ is not compact, $`S\pi (e)`$ is locally divergent and $`𝐒`$ is $`K_v`$-split, $`v𝒮`$, it follows from Proposition 4.3 that $`𝐒`$ is $`K`$-split. Set $`𝐇=𝒵_𝐆(𝐒)`$, $`H=𝐇(K_𝒮)`$ and $`\mathrm{\Delta }=H\mathrm{\Gamma }`$. Let $`\pi _H:HH/\mathrm{\Delta }`$ be the natural projection. Remark that $`𝐇`$ is a reductive group \[Bo1, 13.17, Corollary 2\]. Choose a maximal $`K`$-split torus $`\stackrel{~}{𝐒}`$ of $`𝐇`$. Then $`\stackrel{~}{𝐒}𝐒`$ and there exists $`q𝐆(K)`$ such that
(18)
$$\stackrel{~}{𝐒}=q^1𝐃q.$$
Denote $`\stackrel{~}{S}_v=\stackrel{~}{𝐒}(K_v)`$, $`v𝒮`$, and $`\stackrel{~}{S}=\stackrel{~}{𝐒}(K_𝒮)`$. There exists $`h=(h_v)_{v𝒮}H`$ such that $`h_v^1\stackrel{~}{S}_vh_vg_v^1T_vg_v`$ for every $`v𝒮`$. Denote $`\stackrel{~}{T}_v=h_vg_v^1T_vg_vh_v^1`$ and $`\stackrel{~}{T}=_{v𝒮}\stackrel{~}{T}_v`$. Then $`\stackrel{~}{S}\stackrel{~}{T}H`$ and $`\stackrel{~}{T}\pi _H(h)`$ is a closed locally divergent orbit. Suppose for a moment that the theorem is valid for $`𝐇`$. Then the conditions (i) and (ii) in the formulation of the theorem are automatically fulfilled because $`\mathrm{rank}_K𝐆=\mathrm{rank}_K𝐇`$ and $`\mathrm{rank}_{K_v}𝐆=\mathrm{rank}_{K_v}𝐇`$, $`v𝒮`$. Since $`h=zd`$, where $`z𝒩_H(\stackrel{~}{S})`$ and $`d𝐇(K)`$, using (18), we obtain
$`D=gh^1\stackrel{~}{S}hg^1=`$ $`gd\stackrel{~}{S}d^1g^1=`$
$`=gdq^1Dqd^1`$ $`g^1.`$
Therefore, $`g𝒩_G(D)𝐆(K)`$, which proves (iii). The above discussion reduces the proof to the case when $`𝐒`$ is a central $`K`$-split torus in $`𝐆`$. In this case $`𝐆`$ is an almost direct product over $`K`$ of $`𝐒`$ and a reductive $`K`$-group. Factorizing by $`𝐒`$, we can further reduce the proof to the case when $`𝐒`$ is trivial.
So, in order to complete the proof of the theorem, it is enough to consider the case when $`T_{}\pi (g)`$ is a divergent orbit. The rest of the proof breaks in two cases according to whether or not the assumptions in the formulation of Proposition 5.1 are satisfied.
Assume that for every $`K`$-subalgebra $`𝖇`$ of $`𝖌`$ containing $`\mathrm{Lie}(𝐃)`$ the intersection $`\mathrm{pr}_{}(𝔤_x)𝔟_{}`$, where $`x=\pi (g)`$, contains a horospherical subset. Then (iii) follows from Proposition 5.1(b), and (ii) from Proposition 5.1(c) and Theorem 4.1(b). The condition (i) follows easily from (ii), (iii) and Proposition 4.3.
Now assume the contrary, that is, that there exists a minimal parabolic $`K`$-subalgebra $`𝖇`$ of $`𝖌`$ containing $`\mathrm{Lie}(𝐃)`$ and such that $`\mathrm{pr}_{}(𝔤_x)𝔟_{}`$ does not contain a horospherical subset. We will prove that this assumption leads to contradiction. (As in \[To-We\], our argument is inspired by Margulis’ one, cf.\[To-We, Appendix\].) Let $`𝖚^{\mathbf{}}`$ be the unipotent radical of the minimal parabolic $`K`$-subalgebra opposite to $`𝖇`$. For every positive integer $`n`$ we let $`B_n`$ be a ball of radius $`n`$ in $`𝔤`$. Since $`𝔤_x`$ is discrete in $`𝔤`$, the family of the horospherical subsets in $`\mathrm{pr}_{}(𝔤_x)𝔟_{}`$ is finite. In view of this and the assumption that $`\mathrm{pr}_{}(𝔤_x)𝔟_{}`$ does not contain horospherical subsets, for every $`n`$ there exists an element $`s_nD_{}`$ such that $`\mathrm{Ad}(s_n)`$ acts as an expansion on $`𝔲_{}^{}`$ and
(19)
$$\mathrm{Ad}(s_n)B_n$$
for every horospherical subset $`𝔤_xB_n`$.
Using Proposition 3.4(a), we fix a compact neighborhood $`W_0`$ of 0 in $`𝔤`$ such that $`W_0B_n`$ and for every $`xG/\mathrm{\Gamma }`$ the subalgebra of $`𝔤`$ spanned by $`𝔤_xW_0`$ is unipotent.
Proposition 3.5 and the choice of $`W_0`$ imply that there exist a constant $`\tau >1`$ and a finite set $`t_1,\mathrm{},t_l`$ in $`D_{}`$ such that for every $`yG/\mathrm{\Gamma }`$ there exists $`t\{t_1,\mathrm{},t_l\}`$ satisfying
(20)
$$\mathrm{Ad}(t)a_{}\tau a_{},a𝔤_yW_0.$$
We put
$$W=W_0\left(\underset{i=1}{\overset{l}{}}\mathrm{Ad}(t_i)W_0\right).$$
Given a positive $`n`$, we define inductively a finite sequence $`p_0,p_1,\mathrm{},p_{r_n}`$ as follows. We put $`p_0=s_n`$. Assume that $`p_0,p_1,\mathrm{},p_i`$ are already defined. If $`\mathrm{Ad}(p_i\mathrm{}p_0)(𝔤_x)W`$ does not contain a horospherical subset then $`p_0,p_1,\mathrm{},p_i`$ is the required sequence. If not, we put $`p_{i+1}=t`$, where $`t`$ satisfies (20) with $`y=p_i\mathrm{}p_0x`$. With the same $`y`$ and $`p_{i+1}`$, remark that if $`b𝔤_y`$ and $`bW_0`$ then $`\mathrm{Ad}(p_{i+1})bW`$. This and (20) imply the following
$`\mathrm{𝐂𝐥𝐚𝐢𝐦}`$: If $`p_0,p_1,\mathrm{},p_r`$ are already defined, $`0i<r`$, $`y=p_i\mathrm{}p_0x`$, $`b𝔤_y`$ and $`bW_0`$ then $`\mathrm{Ad}(p_j\mathrm{}p_{i+1})bW`$ for every $`j`$ such that $`ijr`$.
The claim implies that the cardinality of $`\mathrm{Ad}(p_i\mathrm{}p_0)(𝔤_x)W`$ does not increase with $`i`$ and, moreover, the sequence $`\{p_i\}`$ is finite. Put $`g_n=p_{r_n}\mathrm{}p_1p_0`$. It follows from Proposition 3.4(b) that the sequence $`\{g_nx\}`$ is bounded in $`G/\mathrm{\Gamma }`$. Since the orbit $`T_{}x`$ is divergent, the sequence $`\{g_n\}`$ is bounded in $`T_{}`$. Also note that, given the above definition of $`s_n`$, the sequence $`\{s_n\}`$ is unbounded. Again by Proposition 3.4(b), passing to a subsequence, we may assume that $`r_n>0`$ for all $`n`$.
Let $`h_n=p_{r_n}^1g_n`$ and $`_n`$ be a horospherical subset of $`\mathrm{Ad}(h_n)(𝔤_x)W`$. Assume that $`\mathrm{Ad}(h_n^1)(_n)B_n`$. Then it follows from (19) that $`\mathrm{Ad}(p_0h_n^1)(_n)B_n`$. The Claim implies that $`_nW`$, which contradicts the choice of $`_n`$. Therefore,
$$\mathrm{Ad}(h_n^1)(_n)B_n.$$
Since $`_nW`$ and $`W`$ is compact, the sequence $`\{h_n^1\}`$ is not bounded. Therefore, $`\{g_n\}`$ is not either. Contradiction. ∎
### 6.2. Remarks
(a) It follows from the proof of Theorem 1.4 that if $`\mathrm{\#}𝒮>1`$ and the orbit $`Tx`$ $`(`$where $`T=T_𝒮`$$`)`$ is closed and locally divergent then the Zariski closure of $`g^1Tg\mathrm{\Gamma }`$ in $`𝐆`$ contains a maximal $`K`$-split torus.
(b) Since $`𝒩_𝐆(𝐃)(K)`$ meets every coset of the quotient $`𝒩_𝐆(𝐃)/𝒵_𝐆(𝐃)`$, we have that $`𝒵_G(D_v)𝐆(K)=𝒩_G(D_v)𝐆(K)`$ for every $`v`$. On the other hand, it is easy to see that $`𝒵_G(D_{})𝐆(K)𝒩_G(D_{})𝐆(K)`$ whenever $`\mathrm{\#}>1`$ and $`𝐆`$ is a semisimple $`K`$-isotropic group.
### 6.3. Proof of Theorem 1.1
Let us first prove (b). Since the divergent orbits are locally divergent and closed we can apply Theorem 1.4. If $``$ is not a singleton it follows from Theorem 1.4 (i) that $`=𝒮`$. Also it follows from Theorem 1.4 (ii) an (iii) that $`T_𝒮`$ is a compact extension of $`D_𝒮`$. So, $`D_𝒮\pi (g)`$ diverges. This contradicts Proposition 4.3. Therefore $`=\{v\}`$. Again by Theorem 1.4, $`\mathrm{rank}_{K_v}𝐆=\mathrm{rank}_K𝐆`$ and $`g𝒩_G(D_v)𝐆(K_v)`$. Now in order to complete the proof of (b) it remains to apply the remark 6.2 (b).
Let us prove (a). The implication $``$ follows trivially from Propositions 4.2 and 4.4. Suppose that $`T_{}\pi (g)`$ is closed. If $`T_{}\pi (g)`$ is divergent it follows from (b) that $``$ is a singleton. Let $`=\{v\}`$. Since $`\mathrm{rank}_{K_v}𝐆=\mathrm{rank}_K𝐆`$, $`T_v`$ is a compact extension of $`D_v`$. But $`g=zq`$, where $`z𝒵_G(D_v)`$ and $`q𝐆(K)`$. Therefore $`g^1T_vg`$ is a compact extension of $`𝐋(K_v)`$, where $`𝐋=q^1𝐃q`$, which proves (a) when $`T_{}\pi (g)`$ is divergent. Let $`T_{}\pi (g)`$ be not divergent. Then $`g^1Tg\mathrm{\Gamma }`$ is not finite, in particular, $`=𝒮`$. Let $`𝐋`$ be the connected component of the Zariski closure of $`g^1Tg\mathrm{\Gamma }`$ in $`𝐆`$. Set $`𝐇=𝒵_𝐆(𝐋)`$. Since $`𝐇`$ is an almost direct product over $`K`$ of $`𝐋`$ and of a reductive $`K`$-group, factorizing by $`𝐋`$, we can reduced the proof to the case when $`𝐋`$ is trivial. In the latter case either $`T_𝒮`$ is compact and there is nothing to prove or $`T_𝒮\pi (g)`$ is divergent. This complets the proof of (a). ∎
### 6.4. Proof of Corollaries 1.3, 1.5 and 1.6
Corollary 1.3 follows from Theorem 1.1 (a) and Remark 6.2 (a), and Corollary 1.5 follows from Theorem 1.4 and Remark 6.2 (b).
Let us prove Corollary 1.6. The part (a) is immediate from Theorem 1.4. In order to prove (b), remark that $`\left(𝒩_𝐆(𝐃)\times 𝒩_𝐆(𝐃)\right)\mathrm{diag}(𝐆)𝐆\times 𝐆`$, where $`\mathrm{diag}(𝐆)`$ is the diagonal imbedding of $`𝐆`$ into $`𝐆\times 𝐆`$. Therefore, there exists $`(g_1,g_2)(𝐆\times 𝐆)(K)`$ such that $`(g_1,g_2)\left(𝒩_𝐆(𝐃)\times 𝒩_𝐆(𝐃)\right)\mathrm{diag}(𝐆)`$. Let $`v_1`$ and $`v_2`$ be two different valuations in $`𝒮`$ and let $`g=(g_v)_{v𝒮}G`$ be such that $`g_{v_1}=g_1`$, $`g_{v_2}=g_2`$ and $`g_v=1`$ for all $`v𝒮\{v_1,v_2\}`$. It follows from Theorem 1.4 (iii) and Proposition 4.3 that the orbit $`T_{}\pi (g)`$ is locally divergent but not closed. ∎
### 6.5. Remark
In connection with Corollary 1.6 (a), note that if $`G`$ is a real $``$-algebraic group and $`D_{\mathrm{}}`$ is an $``$-split algebraic torus of $`G`$ with $`dimD_{\mathrm{}}>\mathrm{rank}_{}G`$, it was proved by B.Weiss \[We\] that there are no divergent orbits for the action of $`D_{\mathrm{}}`$ on $`G/\mathrm{\Gamma }`$. The following generalization of this result is proved \[To2\]: Let $`G`$ and $`\mathrm{\Gamma }`$ be as in the formulation of Theorem 1.1, $`v𝒮`$ and $`𝐃_v`$ be a $`K_v`$-split torus of $`𝐆`$. Assume that $`dim𝐃_v>\mathrm{rank}_K𝐆`$. Then $`G/\mathrm{\Gamma }`$ does not admit divergent orbits for the action of $`D_v=𝐃_v(K_v)`$.
## 7. Number theoretical application
Let $`K_𝒮[\stackrel{}{x}]`$ be the ring of polynomials in $`n`$ variables $`\stackrel{}{x}=`$ $`(x_1,\mathrm{},x_n)`$ with coefficients from the topological ring $`K_𝒮`$. Let $`f(\stackrel{}{x})=l_1(\stackrel{}{x})\mathrm{}l_m(\stackrel{}{x})`$ $`K_𝒮[\stackrel{}{x}]`$, where $`l_1(\stackrel{}{x}),\mathrm{},`$ $`l_m(\stackrel{}{x})`$ are linearly independent over $`K_𝒮`$ linear forms.
The following is a reformulation of Theorem 1.9 from the Introduction:
###### Theorem 7.1.
With the above notation and assumptions, suppose that $`f(𝒪^n)`$ is a discrete subset of $`K_𝒮`$. Then $`f(\stackrel{}{x})=\alpha g(\stackrel{}{x})`$ for some $`\alpha K_𝒮^{}`$ and some $`g(\stackrel{}{x})𝒪[\stackrel{}{x}]`$ .
The following examples show that the hypotheses in the formulations of Theorem 7.1 are essential and can not be omitted.
$`\mathrm{𝐄𝐱𝐚𝐦𝐩𝐥𝐞𝐬}.`$ Let $`\alpha `$ be a badly approximable number, i.e. there exists a $`c=c(\alpha )>0`$ such that
$$\left|\alpha \frac{p}{q}\right|\frac{c}{q^2}$$
for all $`p/q`$. (Recall that the quadratic irrationals, such as $`\sqrt{2}`$, and the golden ratio $`(\sqrt{5}+1)/2`$ are badly approximable.) Consider the form $`f(x,y)=x^2(\alpha xy)`$. Then the set of values of $`f`$ at the integer points is discrete but $`f`$ is not a multiple of a form with rational coefficients. The reason is that $`f`$ is a product of linearly dependent linear forms.
The hypothesis that $`f`$ is decomposable is also essential. In order to see this it is enough to consider a form $`f(x,y)=x^2+\beta y^2`$ where $`\beta `$ is a positive irrational real number. It is obvious that $`f(^2)`$ is discrete in $``$.
We put $`𝐆=\mathrm{𝐒𝐋}_n`$. So, $`G=\mathrm{𝐒𝐋}_n(K_𝒮)`$ and $`\mathrm{\Gamma }=\mathrm{𝐒𝐋}_n(𝒪)`$.) The group $`G`$ is acting on $`K_𝒮[\stackrel{}{x}]`$ according to the law $`(\sigma f)(\stackrel{}{x})=f(\sigma ^1\stackrel{}{x})`$, where $`\sigma G`$ and $`fK_𝒮[\stackrel{}{x}]`$. We denote $`f_0(\stackrel{}{x})=x_1x_2\mathrm{}x_m`$. It is clear that if $`fK_𝒮[\stackrel{}{x}]`$ is as in the formulation of Theorem 7.1 then $`f(\stackrel{}{x})=\alpha (\sigma f_0)(\stackrel{}{x})`$ for some $`\sigma G`$ and $`\alpha K_𝒮^{}`$. We will denote by $`H_f`$ the stabilizer of $`f`$ in $`G`$.
We precede the proof of Theorem 7.1 by the following general proposition.
###### Proposition 7.2.
Let $`f(\stackrel{}{x})=(\sigma f_0)(\stackrel{}{x})`$ for some $`\sigma G`$ . Assume that $`f(𝒪^n)`$ is a discrete subset of $`K_𝒮`$. Then $`H_f\pi (e)`$ is closed in $`G/\mathrm{\Gamma }`$.
Proof. Let $`\pi (a)`$, $`aG`$, belong to the closure of $`H_f\pi (e)`$. Fix a sequence $`h_iH_f`$ such that $`lim_i\mathrm{}h_i\pi (e)=\pi (a)`$. There exist $`\gamma _i\mathrm{\Gamma }`$ and $`b_iG`$ such that $`lim_i\mathrm{}b_i=e`$ and $`h_i\gamma _i=b_ia`$. Since $`f(𝒪^n)`$ is discrete, for every $`\stackrel{}{z}𝒪^n`$ there exists a real number $`c(\stackrel{}{z})>0`$ such that
(21)
$$f(\gamma _i\stackrel{}{z})=f(h_i\gamma _i\stackrel{}{z})=f(b_ia\stackrel{}{z})=f(a\stackrel{}{z})f(a𝒪^n)f(𝒪^n)$$
for all $`i>c(\stackrel{}{z})`$.
Let $`\chi _1,\chi _2,\mathrm{},\chi _lK[\stackrel{}{x}]`$ be the set of all monomials of degree $`m`$. We consider $`\chi _1,\chi _2,\mathrm{},\chi _l`$ as homomorphisms of multiplicative groups $`K_{}^{}{}_{}{}^{n}K^{}`$. Since $`\chi _1,\chi _2,\mathrm{},\chi _l`$ are linearly independent over $`K`$, i.e. whenever we have a relation
$$\alpha _1\chi _1+\alpha _2\chi _2+\mathrm{}+\alpha _l\chi _l=0,$$
with $`\alpha _iK`$ then all $`\alpha _i=0`$, there exist $`\stackrel{}{z}_1,\stackrel{}{z}_2,\mathrm{},\stackrel{}{z}_l𝒪^n`$ such that $`det(\chi _k(\stackrel{}{z}_s))0`$. In view of (21), there exists $`c>0`$ such that
(22)
$$f(b_ia\stackrel{}{z}_s)=f(a\stackrel{}{z}_s)$$
for all $`s`$ and $`i>c`$.
The form $`f`$ can be regarded as a collection of forms $`f_vK_v[\stackrel{}{x}],v𝒮`$. Since $`det(\chi _k(\stackrel{}{z}_s))0`$, using (22), we get that
$$f_v(b_{iv}a_v\stackrel{}{x})=f_v(a_v\stackrel{}{x})$$
for all $`v𝒮`$ and $`i>c`$, where $`b_{iv}`$ is the $`v`$-component of $`b_i`$ and $`a_v`$ is the $`v`$-component of $`a`$. Hence $`b_iH_f`$ for all $`i>c`$. So, we obtain that
$$\pi (a)=b_i^1h_i\pi (e)H_f\pi (e),$$
which proves that $`H_f\pi (e)`$ is closed. ∎
Given a subgroup $`L`$ of $`G`$, we will write $`L_u`$ for the subgroup generated by the Zariski closed in $`G`$ unipotent subgroups of $`L`$.
The following is a particular case of Theorem 3 from \[To1\].
###### Proposition 7.3.
Let $`L`$ be a closed (for the Euclidean topology) subgroup of $`G`$. Assume that $`L\pi (e)`$ is closed and $`L_u\pi (e)`$ is dense in $`L\pi (e)`$. Let $`𝐏`$ be the connected component of the Zariski closure of $`L\mathrm{\Gamma }`$ in $`𝐆`$ and let $`P=𝐏(K_𝒮)`$. Then
1. $`PL_u`$ and there exists a subgroup of finite index $`P^{}`$ in $`P`$ such that $`L\pi (e)=P^{}\pi (e)`$;
2. If $`𝐐`$ is a proper normal $`K`$-subgroup of $`𝐏`$, there exists $`v𝒮`$ such that $`(𝐏/𝐐)(K_v)`$ contains a unipotent element different from the identity.
Proof of Theorem 7.1. Let $`H_0`$ be the Zariski connected component of $`H_{f_0}`$. It is easy to see that
(23)
$$H_0=\left\{\left(\begin{array}{cc}d& a\\ 0& s\end{array}\right)\right|dD_m,a\mathrm{M}_{m\times (nm)}(K_𝒮)\text{ and }s\mathrm{SL}_{nm}(K_𝒮)\},$$
where $`D_m`$ is the group of all diagonal matrices in $`\mathrm{SL}_m(K_𝒮)`$. Since $`f=\sigma f_0`$, we have that
$$H=\sigma H_0\sigma ^1$$
is the Zariski connected component of $`H_f`$.
Let $`_m`$ be the $`K_𝒮`$-module of all homogeneous polynomials of degree $`m`$ in $`K_𝒮[\stackrel{}{x}]`$. A simple calculation shows that $`K_𝒮f_0`$ is the submodule of all $`H_0`$-invariant elements in $`_m`$. Therefore,
(24)
$$K_𝒮f=\{h_m|\sigma h=h,\sigma H\}.$$
It follows from \[Ra, Theorem 2\] that there exists a closed subgroup $`L`$ of $`G`$ such that $`L\pi (e)=\overline{H_u\pi (e)}`$. Let $`𝐏`$ be the connected component of the Zariski closure of $`L\mathrm{\Gamma }`$ in $`𝐆`$ and let $`P=𝐏(K_𝒮)`$. By Proposition 7.3, $`L\pi (e)=P^{}\pi (e)`$ where $`P^{}`$ is a subgroup of finite index in $`P`$. On the other hand, since $`H_f\pi (e)`$ is closed (Proposition 7.2) and $`H`$ has finite index in $`H_f`$, $`H\pi (e)`$ is also closed. Therefore, $`P^{}H`$. Since $`H_uP^{}`$, it follows from Proposition 7.3 (ii) and from the description (23) of $`H_0`$ that $`H_u=P`$ and $`L\pi (e)=P\pi (e)`$.
Let $`𝐐`$ be the commutator subgroup of $`𝒩_𝐆(𝐏)`$. It follows from (23) that $`𝐐`$ is a semidirect product over $`K`$ of $`𝐏`$ and of an algebraic group $`𝐑`$ defined over $`K`$ which is isomorphic over $`K_v`$ to $`\mathrm{𝐒𝐋}_m`$ for all $`v𝒮`$. (Note that $`𝐑`$ is isomorphic to $`\mathrm{𝐒𝐋}_m`$ over a finite extension of $`K`$ but, in general, $`𝐑`$ is not isomorphic to $`\mathrm{𝐒𝐋}_m`$ over $`K`$ itself.) Let $`R=_{v𝒮}𝐑_v(K_v)`$ and $`T=RH`$. Then $`T=_{v𝒮}𝐓_v(K_v)`$, where $`𝐓_v`$ is a maximal $`K_v`$-split torus in $`𝐑`$, and $`H=TP`$. Since the projection of $`H`$ into $`Q/(Q\mathrm{\Gamma })`$, where $`Q=𝐐(K_𝒮)`$, is closed, the projection of $`T`$ into $`R/(R\mathrm{\Gamma })`$ is closed too. Applying Theorem 1.1, we get a torus $`𝐓`$ in $`𝐑`$ defined over $`K`$ such that $`T=𝐓(K_𝒮)`$. Therefore, $`H=𝐇(K_𝒮)`$, where $`𝐇=\mathrm{𝐓𝐏}`$ is an algebraic group defined over $`K`$.
It follows from the above that $`𝐇(K)`$ is Zariski dense in $`H`$. Note that given $`\sigma 𝐇(K)`$ the coefficients of all $`h_m`$ such that
$$\sigma h=h$$
can be regarded as the space of solutions of a system of linear equations with coefficients from $`K`$. Therefore, in view of (24), there exist $`g(\stackrel{}{x})𝒪[\stackrel{}{x}]`$ and $`\alpha K_𝒮^{}`$ such that $`f(\stackrel{}{x})=\alpha g(\stackrel{}{x})`$. ∎ |
warning/0507/cs0507020.html | ar5iv | text | # First-order queries on structures of bounded degree are computable with constant delay
## Introduction
Evaluating the expressive power of logical formalisms is an important task in theoretical computer science. It has many applications in numerous fields such as complexity theory, verification or databases. In this latter case, it often amounts to determine how difficult it is to compute a query written in a given language. In this vein, determining which fragments of first-order logic defines tractable query languages has deserved much attention.
It is well known, that over an arbitrary signature, computing a first-order query can be done in time polynomial in the size of the structure (and even in logarithmic space and $`AC^0`$). However the exponent of this polynomial depends heavily on the formula size (more precisely, on the number of variables). Nevertheless, for particular kinds of structures or formulas the complexity bound can be substantially improved. In \[See96\], it is proved that checking if a given first-order sentence $`\phi `$ is true (i.e., the Boolean query or model-checking problem) in a structure $`𝒮`$ all of whose relations are of bounded degree can be done in linear time in the size of $`𝒮`$. The method used to prove this result relies on old model-theoretic technics (see \[Han65\]). It is perfectly constructive but hardly implementable. Later, still using such kind of methods, several other tractability results have been shown for the complexity of the model-checking of first-order formulas over structures or formulas that admit nice (tree) decomposition properties (see \[FFG02\]).
In this paper, a bounded degree structure is either a relational structure all of whose relations are of bounded degree or a functional structure involving bijective functions only.
The main goal of this paper is to revisit the complexity of the evaluation problem of not necessarily Boolean first-order queries over structures of bounded degree. We regard query evaluation as a dynamical process. Instead of considering the cost of the evaluation globally, we measure the delay between consecutive tuples, i.e., query problems are viewed as enumeration problems. This latter kind of problems appears widely in many areas of computer science (see for example \[EG95, EGM03, BGKM00, KSS00, Gol94\] or \[JYP88\] for basic complexity notions on enumeration). However, to our knowledge, relation to query evaluation has not been investigated so far.
We prove that any query on bounded degree structures is Constant-Delay<sub>lin</sub>, i.e., can be computed by an algorithm that has two separate parts: it has a precomputation step whose time complexity is linear in the size of the structure and then, outputs all the solution tuples one by one with a constant (i.e., depending on the size of the formula only) delay between two successive tuples. Seen as a global process, this implies that queries on bounded structures can be evaluated in total time $`O(f(|\phi |).(|𝒮|+|\phi (𝒮)|))`$ and space $`O(f(|\phi |).|𝒮|)`$ where $`|𝒮|`$ is the size of the structure $`𝒮`$, $`|\phi |`$ is that of the formula $`\phi `$, $`|\phi (𝒮)|`$ is the size of the result $`\phi (𝒮)`$ of the query and $`f`$ is some function. As a corollary, it implies that the time complexity of the model-checking problem is $`O(f(|\phi |).|𝒮|)`$ thus providing an alternative proof of the result of \[See96\].
A particularity of the main method used in this paper is that it does not rely on model-theoretic technic as previous results of the same kind (see, for example, \[See96\] or \[Lin04\] for a generalization to least-fixed point formulas). Instead, we develop a quantifier elimination method suitable for bijective unary functions and apply it to obtain our complexity bound. An advantage of this method is that it is effective and easily implementable. Another advantage is that our paper is completely self-contained.
Besides, the Constant-Delay<sub>lin</sub> class is an interesting notion by itself and is, to our knowledge, a new complexity class for enumeration problems: as proved for linear time complexity (the class DLIN studied in \[GS02\]) it can be shown that Constant-Delay<sub>lin</sub> is a robust class and is in some sense the minimal robust complexity class of enumeration problems.
The paper is organized as follows. First, basic definitions are given in Section 1. In particular, in Subsection 1.3, we recall definitions about enumeration problems and introduce the notion of constant delay computation and prove some basic properties about it. In Section 2, the quantifier elimination method is introduced and is applied to the evaluation problem of first-order formulas over functional structures all of whose functions are bijective. In Section 3, using classical logical interpretation technics, this later problem is reduced in linear time to the first-order query problem over structures of bounded degree thus providing the same bound for it. Finally, in Subsection 3.3, consequences about the complexity of the subgraph (resp. induced subgraph) isomorphism problem are given.
## 1 Definitions
### 1.1 Logical definitions and query problems
We suppose the reader to be familiar with basic notions of first-order logic. A signature $`\sigma `$ is a finite set of relational and functional symbols of given arities ($`0`$-ary function symbols are constants symbols). The arity of $`\sigma `$ is the maximal arity of its symbols. The set $`\sigma `$ is called unary functional if all its symbols are of arity bounded by one.
A (finite) $`\sigma \text{-structure}`$ consists of a domain $`D`$ together with an interpretation of each symbol of $`\sigma `$ over $`D`$ (the same notation is used here for each signature symbol and its interpretation).
In this paper, we will distinguish between two kinds of signatures on which semantical restrictions on their possible interpretation are imposed:
* Either $`\sigma `$ is made of constant and monadic (i.e., unary) relation symbols and unary function symbols whose interpretation is taken among bijective functions (i.e., permutations) only,
* Or $`\sigma `$ contains relation symbols only whose degrees are bounded by some given constant (detailed definitions about bounded degree relations are delayed till section 3).
Structures defined by either of semantical restrictions will be called bounded degree structures.
In what follows we make precise notions and problems about first-order logic over bijective structures.
###### Definition 1
Let $`\sigma =\{\overline{c},\overline{U},f_1,\mathrm{},f_k\}`$ be a signature consisting of constant symbols $`c_i\overline{c}`$, of monadic predicates $`U_i\overline{U}`$ and of unary function symbols $`f_i`$, $`i=1,\mathrm{},k`$. A bijective $`\sigma \text{-structure}`$ is a $`\sigma \text{-structure}`$ $`𝒮`$ of the form $`𝒮=D;\overline{c},\overline{U},f_1,\mathrm{},f_k`$ where each $`f_i`$ is a permutation on domain $`D`$.
One of the main results of this paper provides a quantifier elimination method over bijective structures. As it is usual for such kind of result, the elimination will be done in a richer language. The following definition is required.
###### Definition 2
A bijective term $`\tau (x)`$ is of the form $`f_1^{ϵ_1}\mathrm{}f_l^{ϵ_l}(x)`$ where $`l0`$, $`x`$ is a variable and where each $`f_i^{ϵ_i}`$ is either the function symbol $`f_i`$ or its reciprocal $`f_i^1`$. The term $`\tau ^1(x)`$ denotes the reciprocal of the term $`\tau (x)`$.
A bijective atomic formula is of one of the following four forms where $`\tau (x)`$ and $`\tau _1(x)`$ are bijective terms:
* either a bijective equality $`\tau (x)=\tau _1(y)`$,
* or $`\tau (x)=c`$ where $`c`$ is a constant symbol,
* or $`U(\tau (x))`$ where $`U`$ is a monadic predicate,
* or a cardinality statement $`_x^k\mathrm{\Psi }(x)`$ where the quantifier $`_x^k`$ is interpreted as ”there exist at least $`k`$ values of $`x`$ such that” and $`\mathrm{\Psi }`$ is a Boolean combination of bijective atoms $`\alpha (x)`$ over variable $`x`$ only.
As the reciprocal of each function symbol can be used, each bijective equality $`\tau (x)=\tau _1(y)`$ can be rephrased as $`\tau _2(x)=y`$ where $`\tau _2(x)=\tau _1^1\tau (x)`$. A bijective literal is a bijective atomic formula or its negation.
###### Definition 3
The set $`\mathrm{𝐅𝐎}_{\mathrm{𝐁𝐢𝐣}}`$ of bijective first-order formulas is the set of first-order formulas built over bijective atomic formulas of some unary signature $`\sigma `$.
Let $`\overline{t}=(t_1,\mathrm{},t_k)`$ be a $`k`$-tuple of variables and $`\phi (\overline{t})`$ and $`\phi ^{}(\overline{t})`$ be two $`\sigma `$-formulas with free variables $`\overline{t}`$. Formulas $`\phi (\overline{t})`$ and $`\phi ^{}(\overline{t})`$ are equivalent if for all $`\sigma \text{-structures}`$ $`𝒮`$ and all tuples $`\overline{a}`$ of element of the domain with $`|\overline{a}|=|\overline{t}|`$ it holds that:
$$(𝒮,\overline{a})\phi (\overline{t})\text{ iff }(𝒮,\overline{a})\phi ^{}(\overline{t}).$$
In this paper query problems are considered for specific classes of first-order formulas (and structures). One of the specific problems under consideration here is the following.
$`\text{Query}(\mathrm{𝐅𝐎}_{\mathrm{𝐁𝐢𝐣}})`$
Input: a unary functional signature $`\sigma `$, a bijective $`\sigma `$-structure $`𝒮`$ and a first-order bijective $`\sigma `$-formula $`\phi (\overline{x})`$ with $`k`$ free variables $`\overline{x}=(x_1,\mathrm{},x_k)`$
Parameter: $`\phi `$
Output: $`\phi (𝒮)=\{\overline{a}D^k:(𝒮,\overline{a})\phi (\overline{x})\}`$.
The Boolean query problem (the subproblem where $`k=0`$) is often called a model-checking problem. It will be denoted by $`\text{MC}(\mathrm{𝐅𝐎}_{\mathrm{𝐁𝐢𝐣}})`$ here. As suggested by the formulation of the query problem, we are interested in its parameterized complexity and the complexity results given here consider the size of the query formula $`\phi `$ as the parameter (see \[DF99\]).
### 1.2 Model of computation and measure of time
The model of computation used in this paper is the Random Access Machine (RAM) with uniform cost measure (see \[AHU74, GS02, GO04, FFG02\]). As query problems are the main subject of this paper, instances of problems always consist of two kinds of objects: first-order structures and first-order formulas.
The size $`|I|`$ of an object $`I`$ is the number of registers used to store $`I`$ in the RAM. If $`E`$ is the set $`[n]`$, $`|E|=card(E)=n`$. If $`RD^k`$ is a $`k`$-ary relation over domain $`D`$, with $`|D|=card(D)`$, then $`|R|=k.card(R)`$: all the tuples $`(x_1,\mathrm{},x_k)`$ for which $`R(x_1,\mathrm{},x_k)`$ holds must be stored, each in a separate $`k`$-tuple of registers. Similarly, if $`f`$ is a unary function from $`D`$ to $`D`$, all values $`f(x)`$ must be stored and $`|f|=|D|`$.
If $`\phi `$ is a first-order formula, $`|\phi |`$ is the number of occurrences of variables, relation or function symbols and syntactic symbols: $`,,,,\neg ,=,\mathrm{"}(\mathrm{"},\mathrm{"})\mathrm{"},\mathrm{"},\mathrm{"}`$. For example, if $`\phi xyR(x,y)\neg (x=y)`$ then $`|\phi |=17`$.
All the problems we consider in this paper are parameterized problems: they take as input a list of objects made of a $`\sigma `$-structure $`𝒮`$ and a formula $`\phi `$ and as output the result of the query size $`\phi (𝒮)`$. Due to the much larger size, in practice, of the structure $`𝒮`$ than the size of formula $`\phi `$, $`|𝒮|>>|\phi |`$, this latter one, $`|\phi |`$ , in considered here as the parameter.
A problem P is said to be computable in time $`f(|\phi |).T(|𝒮|,|\phi (𝒮)|)`$ for some function $`f:NR^+`$ if there exists a RAM that computes P in time (i.e., the number of instructions performed) bounded by $`f(|\phi |).T(|𝒮|,|\phi (𝒮)|)`$ using space, i.e., addresses and register contents also bounded by $`f(|\phi |).T(|𝒮|,|\phi (𝒮)|)`$. The notation $`O_\phi (T(|𝒮|,|\phi (𝒮)|))`$ is used when one does not want to make precise the value of function $`f`$. It is also assumed that the function $`T`$ is at least linear and at most polynomial, i.e., $`T(n,p)=\mathrm{\Omega }(n+p)`$ and $`T(n,p)=(n+p)^{O(1)}`$. To give an example and to relate our complexity measure to the logarithmic cost measure, in case $`T`$ is linear, i.e., $`T(n,p)=n+p`$, the number of bits manipulated by the RAM is well linear in the number of bits needed to encode the input and the output.
### 1.3 Enumeration algorithms and constant delay computation
In this section, $`A`$ is a binary predicate. Enumeration problems will be defined by reference to such a predicate.
###### Definition 4
Given a binary relation $`A`$, the enumeration function $`\text{Enum}A`$ associated to $`A`$ is defined as follows. For each input $`x`$:
$$\text{Enum}A(x)=\{y:A(x,y)\text{ holds }\}$$
###### Remark 1
Query problems may evidently be seen as enumeration problems. The input $`x`$ is made of the structure $`𝒮`$ and the formula $`\phi (\overline{x})`$, a witness $`y`$ is a tuple $`\overline{a}`$ and evaluating predicate $`A`$ amounts to check whether $`(𝒮,\overline{a})\phi (\overline{x})`$.
One may consider the delay between two consecutive solutions as an important point in the complexity of enumeration problems. In \[JYP88\] several complexity measures for enumeration have been defined. One of the most interesting is that of polynomial delay algorithm. An algorithm $`𝒜`$ is said to run within a polynomial delay if there is no more than a (fixed) polynomial delay between two consecutive solutions it outputs (and no more than a polynomial delay to output the first solution and between the last solution and the end of the algorithm). Polynomial delay is often considered as the right notion of feasability for enumeration problems.
In this paper, we introduce a much stronger complexity measure that forces constant delay between outputs.
###### Definition 5
An enumeration problem $`\text{Enum}A`$ is constant delay with linear precomputation, which is written $`\text{Enum}A\text{Constant-Delay}\text{lin}`$, if there exists a RAM algorithm $`𝒜`$ which, for any input $`x`$, enumerates all the elements of the set $`\text{Enum}A(x)`$ with a constant delay, i.e., that satisfies the following properties.
1. $`𝒜`$ uses linear input space, i.e., space $`O(|x|)`$
2. $`𝒜`$ can be decomposed into the two following successive steps
1. $`\text{precomp}(𝒜)`$ which runs some precomputations in time $`O(|x|)`$, and
2. $`\text{enum}(𝒜)`$ which outputs all solutions within a delay bounded by some constant $`\text{delay}(𝒜)`$. This delay applies between two consecutive solutions and after the last one.
Allowing polynomial time precomputations (and polynomial space) instead of linear time, one may define a larger class called Constant-Delay<sub>poly</sub>.
###### Remark 2
As proved for the linear time class DLIN (see \[GS02\]), it can be shown that the complexity enumeration class Constant-Delay<sub>lin</sub> is robust, i.e., is not modified if the set of allowed operations and statements of the RAMs is changed in many ways. This is because linear time (and linear space) precomputations give the ability to precompute the tables of new allowed operations.
The following result is immediate, it evaluates the total time cost of any constant delay algorithm.
###### Lemma 1
Let $`\text{Enum}A`$ be an enumeration problem belonging to Constant-Delay<sub>lin</sub> then, for any input $`x`$, the set $`\text{Enum}A(x)`$ can be computed in $`O(|x|+|\text{Enum}A(x)|)`$ total time, i.e., in time linear in the size of $`|Input|+|Output|`$, and linear input space $`O(|x|)`$.
###### Remark 3
In the query problem we consider, the size of $`\phi `$ is considered as a parameter. Then, $`|x|=|𝒮|`$ and the constant delay depends on $`|\phi |`$ only.
The two lemmas below give basic properties of constant delay computations.
###### Lemma 2
An enumeration problem $`\text{Enum}A`$ computable in linear time $`O(|x|)`$ for any input $`x`$ belongs to Constant-Delay<sub>lin</sub>.
Proof. For any input $`x`$, one only has to compute the set $`\text{Enum}A(x)`$, to sort it and to eliminate the possible multiple occurrences of solutions. These steps can be viewed as the precomputation part of the algorithm running in time $`O(|x|)`$. Then, one has to enumerate one by one the solutions of the sorted list. This is obviously a constant delay process. $`\mathrm{}`$
###### Lemma 3
Let $`\text{Enum}A`$ and $`\text{Enum}B`$ be two disjoint enumeration problems, i.e., such that, for any input $`x`$, $`\text{Enum}A(x)\text{Enum}B(x)=\mathrm{}`$. Let $`\text{Enum}(AB)`$ be the union of this two enumeration problems defined by, for any $`x`$:
$$\text{Enum}(AB)(x)=\{y:A(x,y)\text{ or }B(x,y)\text{ holds }\}.$$
If $`\text{Enum}A`$ and $`\text{Enum}B`$ belong to Constant-Delay<sub>lin</sub> then, problem $`\text{Enum}AB`$ also belongs to Constant-Delay<sub>lin</sub>.
Proof. Due to the disjointness of the two solutions sets for any input, the proof is evident. Given $`𝒜`$ and $``$ the algorithms for problems $`\text{Enum}A`$ and $`\text{Enum}B`$, the following algorithm correctly computes for the problem $`\text{Enum}AB`$.
Obviously, the delay is bounded by the maximum of $`\text{delay}(𝒜)`$ and $`\text{delay}()`$. $`\mathrm{}`$
###### Remark 4
Note that the disjointness condition in the Lemma above is not always necessary. In case there exist a total ordering $``$ and constant delay enumeration algorithms for $`\text{Enum}A`$ and $`\text{Enum}B`$ that enumerate solutions with respect to this unique ordering $``$ then, it is easily seen that $`\text{Enum}AB`$ belongs also to Constant-Delay<sub>lin</sub> even if the problems are not disjoints.
## 2 First-order queries on bijective structures
### 2.1 Quantifier elimination on bijective structures
The key result of this paper consists of a quantifier elimination method for $`\mathrm{𝐅𝐎}_{\mathrm{𝐁𝐢𝐣}}`$ formulas.
###### Theorem 4 (quantifier elimination for $`\mathrm{𝐅𝐎}_{\mathrm{𝐁𝐢𝐣}}`$)
Each bijective first-order formula is equivalent to a Boolean combination of bijective atomic formulas. More precisely, let $`\phi (\overline{t})\mathrm{𝐅𝐎}_{\mathrm{𝐁𝐢𝐣}}`$ with free variables $`\overline{t}`$ then, there exists a Boolean combination of bijective atomic formulas $`\phi ^{}(\overline{t})`$ over the same free variables $`\overline{t}`$ equivalent to $`\phi (\overline{t})`$.
In the special case where $`\phi `$ is closed (i.e., without free variable) then, $`\phi `$ is equivalent to a Boolean combination of cardinality statements.
Proof. As $`x\phi \neg (x\neg \phi )`$, we only have to consider elimination of existentially quantified variables. W.l.o.g., we consider formulas in disjunctive normal form and, as existential quantifier commutes with disjunction we may consider the case of the elimination of a single existentially quantified variable $`y`$ in a formula of the form:
$$\phi (\overline{x})y(\alpha _1\mathrm{}\alpha _r)$$
(1)
where each $`\alpha _i`$ is a bijective literal among variables $`\overline{x}`$ and $`y`$. Literals depending on $`\overline{x}`$ only and cardinality statements need not be considered since they do not involve $`y`$, so $`\phi (\overline{x})`$ may be supposed of the following form:
$$\phi (\overline{x})y[\psi (y)y=_{ϵ_1}\tau _1(x_{i_1})\mathrm{}y=_{ϵ_k}\tau _k(x_{i_k})]$$
(2)
where each $`y=_{ϵ_j}\tau _j(x_{i_j})`$ with $`ϵ_j=\pm 1`$ is $`y=\tau _j(x_{i_j})`$ if $`ϵ_j=1`$ or $`y\tau _j(x_{i_j})`$ if $`ϵ_j=1`$. To eliminate quantified variable $`y`$ two cases may happen.
Suppose first there is at least one index $`j`$ such that $`ϵ_j=1`$. In this case, the equality $`y=\tau _j(x_{i_j})`$ is used to replace each occurrence of $`y`$ in the formula by the term $`\tau _j(x_{i_j})`$. The process results in a new formula $`\phi ^{}(\overline{x})`$ without variable $`y`$.
The second possibility leads to a more complicated replacement scheme. Suppose that for every $`j`$, $`ϵ_j=1`$. Then,
$$\phi (\overline{x})y[\psi (y)\underset{jk}{}y\tau _j(x_j)]$$
(3)
(For simplicity of notations but w.l.o.g. we have supposed that $`i_j=j`$ for $`j=1,\mathrm{},k`$). The basic idea is now the following : suppose $`hk`$ is the number of distinct values among the $`k`$ terms $`\tau _j(x_j)`$ such that $`\psi (\tau _j(x_j))`$ is true; then, formula $`\phi (\overline{x})`$ is true if and only if the number of $`y`$ such that $`\psi (y)`$ holds is strictly greater than $`h`$ (i.e., $`_y^{h+1}\psi (y)`$ is true). Introducing (new) cardinality statements in the formula, $`\phi (\overline{x})`$ can be equivalently rephrased as the following Boolean combination of bijective atomic formulas:
$$\phi (\overline{x})\begin{array}{c}\underset{h=0}{\overset{k}{}}\underset{P[k],QP,|Q|=h}{}\hfill \\ \left[\underset{jQ}{}\psi (\tau _j(x_j))\underset{iP}{}\underset{jQ}{}\tau _i(x_i)=\tau _j(x_j)\underset{j[k]P}{}\neg \psi (\tau _j(x_j))_y^{h+1}\psi (y)\right]\hfill \end{array}$$
(4)
where $`[k]=\{1,\mathrm{},k\}`$.
More generally, starting from a prenex bijective first-order formula $`\phi (\overline{t})`$ with free variables $`\overline{t}`$, one eliminates all quantified variables from the innermost to the outermost one. This will result in an equivalent Boolean combination of bijective atomic formulas over $`\overline{t}`$. In the case where $`\phi `$ is without free variable (i.e., $`\overline{t}`$ is empty), it is easily seen that the elimination process results in a Boolean combination of cardinality statements (note that, of course, $`x\phi (x)_x^1\phi (x)`$). $`\mathrm{}`$
One interesting consequence of Theorem 4 is the following result.
###### Corollary 5 (Seese \[See96\])
The problem $`\text{MC}(\mathrm{𝐅𝐎}_{\mathrm{𝐁𝐢𝐣}})`$ is decidable in time $`O_\phi (|𝒮|)`$.
Proof. From Theorem 4, we know that there exists a Boolean combination of cardinality statements over the same signature $`\sigma `$ equivalent to $`\mathrm{\Phi }`$. Given a formula $`_x^k\mathrm{\Psi }(x)`$ one can test whether a given $`\sigma \text{-structure}`$ $`𝒮`$ satisfies $`𝒮_x^k\mathrm{\Psi }(x)`$ in time $`O_\mathrm{\Psi }(|𝒮|)`$: it suffices to enumerate all the elements $`a`$ of the domain, test whether $`(𝒮,a)\mathrm{\Psi }(x)`$ in constant time and count those for which the answer is positive. If this number is greater than or equal to $`k`$ then $`_x^k\mathrm{\Psi }(x)`$ is true in $`𝒮`$. The final answer for $`\mathrm{\Phi }`$ is given by the boolean combination of the answers for each cardinality statement. $`\mathrm{}`$
#### 2.1.1 Considerations on an efficient implementation of the algorithm
Compared to the method of \[See96\], the proofs given in this paper are constructive and easily implementable. But, due to the case of Formula 3 in Theorem 4 which leads to the equivalent Formula 4 the whole process is in $`O_\phi (|𝒮|)=O(f(|\phi |).|𝒮|)`$ for some function $`f`$ that may be a tower of exponentials. It can be shown that it heavily depends on the number of variables and of quantifier alternations of the formula. However, the size of the function $`f`$ can be substantially reduced in case there are few quantifier alternations.
In what follows, we revisit the method of the proof of Theorem 4 to prove a slightly different result in a specific case. We focus on formulas with existentially quantified variables only and show that the model-checking problem for such formulas can be efficiently evaluated. A $`\mathrm{𝐅𝐎}_{\mathrm{𝐁𝐢𝐣}}`$ formula is in $`\mathrm{\Sigma }_1\mathrm{𝐅𝐎}_{\mathrm{𝐁𝐢𝐣}}`$ if it is of the form:
$$\overline{y}\phi $$
where $`\phi `$ is quantifier-free and in disjunctive normal form (DNF).
###### Corollary 6
The model-checking problem for $`\mathrm{\Sigma }_1\mathrm{𝐅𝐎}_{\mathrm{𝐁𝐢𝐣}}`$ formulas can be evaluated in time $`O(|\phi |^d.|𝒮|)`$ where $`d`$ is the number of distinct variables of $`\phi `$.
Proof. The result obviously holds for $`d=1`$. So, assume $`d>1`$. For the same reason as in Theorem 4, we may consider any formula of the form:
$$\phi (\overline{x})y(\alpha _1\mathrm{}\alpha _r)$$
(5)
where each $`\alpha _i`$ is a bijective literal <sup>1</sup><sup>1</sup>1In this proof, bijective literals do not involve cardinality statements with variables among $`\overline{x}`$ and $`y`$. For sake of completeness here, we consider also terms not containing $`y`$. Then, $`\phi (\overline{x})`$ is of the form:
$$\phi (\overline{x})y[\psi (y)y=_{ϵ_1}\tau _1(x_{i_1})\mathrm{}y=_{ϵ_k}\tau _k(x_{i_k})\gamma (\overline{x})]$$
(6)
with the same notation $`ϵ_j`$ as in the proof of Theorem 4 and $`\gamma (\overline{x})`$ involves variables of $`\overline{x}`$ only. Again, if $`ϵ_j=1`$, for some $`j`$, then all the occurences of $`y`$ are replaced by $`\tau _j(x_{i_j})`$ and $`\phi (\overline{x})`$ is equivalent to a conjunction of literals without variable $`y`$.
Suppose now that $`ϵ_j=1`$ for all $`jk`$. Let $`A=\{aD:(𝒮,a)\psi (y)\}`$. Since $`\psi (y)`$ is quantifier-free, $`A`$ can be computed in time $`O(|\psi |.|𝒮|)`$. Two cases need to be considered now. If $`|A|>k`$, since there are at most $`k`$ different values $`\tau _j(x_j)`$ for $`j=1,\mathrm{},k`$, then the conjunction $`y[\psi (y)y\tau _1(x_{i_1})\mathrm{}y\tau _k(x_{i_k})]`$ is always true and $`\phi (\overline{x})`$ is simply equivalent to $`\gamma (\overline{x})`$. If $`|A|k`$ let $`A=\{a_1,\mathrm{},a_h\}`$, with $`hk`$. Formula $`\phi (\overline{x})`$ is replaced by the equivalent formula below over the richer signature $`\sigma \{a_1,\mathrm{},a_h\}`$:
$$\underset{ih}{}(\underset{jk}{}a_i\tau _j(x_{i_j})\gamma (\overline{x}))$$
In all cases, the formula obtained is also in DNF. Time $`O(|\phi |.|𝒮|)`$ is needed to eliminate variable $`y`$ and the new formula is of size bounded by $`O(k.|\phi |)`$, i.e., less than $`O(|\phi |^2)`$. Elimination of all the $`d`$ existentially quantified variables except the last one can be pursued from this new formula (without need for a normalisation). In the worst case (where all literals are of the form $`x_i\tau _1(x_j)`$), the process will result in a disjunction of less than $`|\phi |^{d1}`$ conjunctions of at most $`|\phi |`$ literals. $`\mathrm{}`$
### 2.2 Constant delay algorithm for first-order queries on bijective structures
We are now ready to state the main result of this section.
###### Theorem 7
The problem $`\text{Query}(\mathrm{𝐅𝐎}_{\mathrm{𝐁𝐢𝐣}})\text{Constant-Delay}\text{lin}`$. In particular, from Lemma 1, it can be computed in time $`O_\phi (|𝒮|+|\phi (𝒮)|)`$ and space $`O_\phi (|𝒮|)`$.
###### Definition 6
A bijective literal is a bijective atomic formula or its negation.
Before proving Theorem 7, we establish the following lemma.
###### Lemma 8
Let $`S`$ be a bijective structure and $`\mathrm{\Psi }`$ be a conjunction of bijective literals. Computing query $`𝒮\mathrm{\Psi }(𝒮)`$ can be done in Constant-Delay<sub>lin</sub>.
Proof. The result is proved by induction on $`k`$ the number of free variables of $`\mathrm{\Psi }(\overline{x})`$ where $`\overline{x}=(x_1,\mathrm{},x_k)`$. We even assume that $`\mathrm{\Psi }`$ makes use of explicit constants from domain $`D`$ of $`𝒮`$.
For the case $`k=1`$, it is evident that the one variable query $`Q=\{aD:(𝒮,a)\mathrm{\Psi }(x)\}`$ can be evaluated in time $`O_\mathrm{\Psi }(|D|)=O_\mathrm{\Psi }(|𝒮|)`$ and hence, by Lemma 2, is in Constant-Delay<sub>lin</sub>.
The result is supposed to be true for $`k`$ ($`k1`$) and proved now for $`k+1`$. Let’s consider the query:
$$Q=\{(\overline{a},b)D^{k+1}:𝒮\mathrm{\Psi }(\overline{x},y)\}$$
where the conjunction of bijective literals $`\mathrm{\Psi }`$ is over variables $`\overline{x}=(x_1,\mathrm{},x_k)`$ and $`y`$. As for Theorem 4, two cases need to be distinguished.
1. $`\mathrm{\Psi }`$ contains at least one literal of the form $`\tau _1(y)=\tau _2(x_{i_0})`$, $`1i_0k`$, that can also be rephrased as $`y=\tau (x_{i_0})`$,
2. $`\mathrm{\Psi }`$ does not contain such a literal.
In the first case, $`\mathrm{\Psi }`$ can rewritten as:
$$\mathrm{\Psi }(\overline{x},y)=\mathrm{\Psi }_0(\overline{x},y)y=\tau (x_{i_0}).$$
Query $`Q`$ is then equivalent to:
$$Q=\{(\overline{a},\tau (a_{i_0}))D^{k+1}:(𝒮,\overline{a})\mathrm{\Psi }_0(\overline{x},\tau (x_{i_0}))\},$$
which is essentially the following $`k`$ variable query $`Q^{}`$:
$$Q^{}=\{\overline{a}D^k:(𝒮,\overline{a})\mathrm{\Psi }_0(\overline{x},\tau (x_{i_0})\}.$$
To be precise, $`Q=\{(\overline{a},\tau (a_{i_0})):\overline{a}Q^{}\}`$. By the induction hypothesis, query $`Q^{}`$ can be computed by some algorithm $`𝒜^{}`$ in constant delay. This provides the following constant delay procedure for query $`Q`$.
Case 2 is a little more complicated. Formula $`\mathrm{\Psi }`$ can be put under the following form:
$$\mathrm{\Psi }\mathrm{\Psi }_1(\overline{x})\mathrm{\Psi }_2(y)\underset{1ir}{}y\tau _i(x_{j_i})$$
with $`1j_ik`$ for $`1ir`$. By induction hypothesis, the $`k`$ variable query:
$$Q_1=\{\overline{a}D^k:(𝒮,\overline{a})\mathrm{\Psi }_1(\overline{x})\}$$
can be computed by an algorithm $`𝒜_1`$ on input $`𝒮`$ with constant delay. For similar reason, the $`k`$ variable query $`Q_b`$ over structure $`(𝒮,b)`$ defined by:
$$Q_b=\{\overline{a}D^k:(𝒮,\overline{a},b)\mathrm{\Psi }(\overline{x},y)\}\}$$
can be enumerated by an algorithm using constant delay. Let now $`Q_2`$ be:
$$Q_2=\{bD:(𝒮,b)\mathrm{\Psi }_2(y)\}.$$
If $`|Q_2|r`$ then, by Lemma 3, there exists an algorithm $`𝒜_0`$ which enumerates the disjoint union $`_{bQ_2}Q_b\times \{b\}`$ with constant delay. Note that $`_{bQ_2}Q_b\times \{b\}=Q`$. From what has been said Algorithm 3 below correctly computes query $`Q`$.
Up to step 10 of the algorithm, all can be done in linear time.
It remains to show that, in the case where $`|Q_2|r+1`$, the delay between two successive solutions is bounded by some constant. Since $`|Q_2|r+1`$ and the number of $`bQ_2`$ that verify $`(𝒮,\overline{a},b)\vDash ̸_{1ir}y=\tau _i(x_{j_i})`$ is bounded by $`r`$, the algorithm outputs at least one $`(\overline{a},b)`$ for each $`\overline{a}Q_1`$. More precisely, it outputs $`|Q_2|r`$ such tuples. For the same reasons, the maximal delay between two successive outputs is then bounded by $`2r`$. The same arguments apply for the delay between the last solution and the end of the algorithm. Then, computing $`Q`$ can be done in constant delay. $`\mathrm{}`$
Proof of Theorem 7. Let $`𝒮`$ and $`\phi (\overline{x})`$ be instances of the $`\text{Query}(\mathrm{𝐅𝐎}_{\mathrm{𝐁𝐢𝐣}})`$ problem. From Theorem 4, one can transform $`\phi (\overline{x})`$ into the following equivalent formula in disjunctive normal form:
$$\phi (\overline{x})\mathrm{\Psi }_1(\overline{x})\mathrm{}\mathrm{\Psi }_q(\overline{x})$$
where each $`\mathrm{\Psi }_i`$ is a conjunction of bijective literals and for all $`i,j`$, $`1i<jq`$ and all bijective structures $`𝒮`$, $`\mathrm{\Psi }_i(𝒮)\mathrm{\Psi }_j(𝒮)=\mathrm{}`$. The Theorem immediately follows from Lemma 3 since the enumeration problem of each query $`𝒮\mathrm{\Psi }_i(𝒮)`$, $`1iq`$, belongs to Constant-Delay<sub>lin</sub> by Lemma 8. $`\mathrm{}`$
## 3 Relational structures of bounded degree
### 3.1 Two equivalent definitions
Let $`\rho =\{R_1,\mathrm{},R_q\}`$ be a relational signature, i.e., a signature made of relational symbols $`R_i`$ each of arity $`a_i`$. Recall that $`arity(\rho )=max_{1iq}(a_i)=m`$.
Let $`𝒮=D;R_1,\mathrm{},R_q`$ be a $`\rho `$-structure. For each $`iq`$, $`R_iD^{a_i}`$. The degree of an element $`x`$ in $`𝒮`$ is defined as follows:
$$degree_𝒮(x)=\underset{1iq}{}\underset{1ja_i}{}\mathrm{}\{(y_1,\mathrm{},y_{a_i})D^{a_i}:ja_i\text{ s.t. }x=y_j\text{ and }𝒮R_i(y_1,\mathrm{},y_{a_i})\}.$$
Intuitively, $`degree_𝒮(x)`$ is the total number of tuples of relations $`R_i`$ to which $`x`$ belongs to. One defines the degree of a structure as $`degree(𝒮)=max_{xD}(degree_𝒮(x))`$.
###### Remark 5
In \[See96\] a different definition of the degree of a structure is given. It counts, for each $`x`$, the number of distinct elements $`yx`$ adjacent to $`x`$, i.e., that appear in some tuple with $`x`$. More precisely,
$$degree_𝒮^1(x)=\mathrm{}\{y:yx\text{ and }iq,\overline{t}D^{a_i},\text{ s.t. }𝒮R_i(\overline{t})\text{ and }x,y\overline{t}\},$$
and $`degree^1(𝒮)=max_{xD}(degree_𝒮^1(x))`$.
Since each tuple containing $`x`$ contains at most $`m1`$ elements different from $`x`$, it is easily seen that:
$$degree^1(𝒮)(m1).degree(𝒮)\text{ where }m=arity(\rho ).$$
.
Conversely, for each $`x`$, if there exist at most $`d`$ elements $`yD`$ adjacent to $`x`$ then, the number of distinct tuples involving $`x`$ and $`y`$ is bounded by $`q.m.d^{m1}`$. Hence,
$$degree(𝒮)q.m.(degree^1(𝒮))^{m1}.$$
So, the two measures yield the same notion of bounded degree structure.
We are interested in the complexity of the following query problem for bounded degree structures (which is clearly independent of either measure of degree we choose).
$`\text{Query}(\mathrm{𝐅𝐎}_{\mathrm{𝐃𝐞𝐠}})`$
Input: an integer $`d`$, a relational signature $`\rho `$, a $`\rho `$-structure $`𝒮`$ with $`degree(𝒮)d`$ and a first-order $`\rho `$-formula $`\phi (\overline{x})`$ with $`k`$ free variables $`\overline{x}=(x_1,\mathrm{},x_k)`$
Parameter: $`d,\phi `$
Output: $`\phi (𝒮)=\{\overline{a}D^k:(𝒮,\overline{a})\phi (\overline{x})\}`$.
### 3.2 Interpreting a structure of bounded degree into a bijective structure
In this section, we present a natural reduction from $`\text{Query}(\mathrm{𝐅𝐎}_{\mathrm{𝐃𝐞𝐠}})`$ to $`\text{Query}(\mathrm{𝐅𝐎}_{\mathrm{𝐁𝐢𝐣}})`$ which is obtained by interpreting any structure of bounded degree into a bijective one.
Let $`𝒮=D;R_1,\mathrm{},R_q`$ be a $`\rho `$-structure of domain $`D`$, of arity $`m=max_{1iq}arity(R_i)`$ and of degree bounded by some constant $`d`$. One associates to $`𝒮`$ a bijective $`\sigma `$-structure $`𝒮^{}=D^{};D,T_1,\mathrm{},T_q,g,f_1,\mathrm{},f_m`$ of domain $`D^{}`$ where $`D,T_1,\mathrm{},T_q`$ are pairwise disjoints unary relations (i.e. subsets of $`D^{}`$) and $`g,f_1,\mathrm{},f_m`$ are permutations of $`D^{}`$. Structure $`𝒮^{}`$ is precisely defined as follows:
* $`D`$ corresponds to the domain of $`𝒮`$.
* $`T_i`$ ($`1iq`$) is a set of elements each representing a tuple of $`R_i`$ (hence, $`card(T_i)=card(R_i)`$).
The new domain $`D^{}`$ is the disjoint union: $`D(D\times \{1,\mathrm{},d\})T_1\mathrm{}T_q`$. Let us use the following convenient abbreviations: $`U=D(D\times \{1,\mathrm{},d\})`$ and $`T=_{1iq}T_i`$.
* $`g`$ creates a cycle that relates $`d`$ copies of each element $`x`$ of the domain. More precisely, for each $`xD`$, it holds $`g(x)=(x,1)`$, $`g((x,i))=(x,i+1)`$ for $`1i<d`$, and $`g((x,d))=x`$. We also set $`g(x)=x`$ for all other $`x`$ ($`xT`$).
* Each $`f_i`$ is an involutive permutation and essentially represents a projection of $`T`$ into $`D`$ as follows. Let $`R_i(x_1,\mathrm{},x_k)`$ be true in $`𝒮`$ for some relation $`R_i`$ of arity $`km`$ and some $`k`$-tuple $`(x_1,\mathrm{},x_k)D^k`$. Suppose $`R_i(x_1,\mathrm{},x_k)`$ is represented by element $`tT_i`$, then, for each $`jk`$, set $`f_j(t)=(x_j,h)`$ and set the reciprocal $`f((x_j,h))=t`$ if $`R(x_1,\mathrm{},x_k)`$ is the $`h^{th}`$ tuple in which $`x_j`$ appears (with $`hd`$). The construction is completed by loops $`f_j(x)=x`$ for all other $`xD^{}`$.
Figure 1 details the reduction on an example.
It is clear that, by construction, $`𝒮^{}`$ is a bijective structure and that we have the following interpretation Lemma.
###### Lemma 9
Let $`\theta _i`$ be the $`\sigma `$-formula below associated to any symbol $`R_i\rho `$ of arity $`k`$:
$$\theta _i(x_1,\mathrm{},x_k)t(T_i(t)\underset{1jk}{}\underset{1hd}{}f_j(t)=g^h(x_j)).$$
Then, for all $`(a_1\mathrm{},a_k)D^k`$:
$$(𝒮,a_1,\mathrm{},a_k)R_i(x_1,\mathrm{},x_k)(𝒮^{},a_1,\mathrm{},a_k)\theta _i(x_1,\mathrm{},x_k).$$
To each first-order $`\rho `$-formula $`\phi (x_1,\mathrm{},x_p)`$, one associates the $`\sigma `$-formula $`\phi ^{\prime \prime }(x_1,\mathrm{},x_p)`$ obtained by replacing each quantification $`v`$ (resp. $`v`$) by the relativized quantification $`(vD(v))`$ (resp. $`(vD(v))`$) (that can be written respectively as $`v(D(v)\mathrm{})`$ and $`v(D(v)\mathrm{})`$) and by replacing each subformula $`R_i(x_1,\mathrm{},x_k)`$ by $`\theta _i(x_1,\mathrm{},x_k)`$.
The following proposition and lemma express that our reduction is correct and linear in $`|𝒮|`$. Because of Lemma 9, Proposition 10 can be easily proved by induction on formula $`\phi `$.
###### Proposition 10 (interpretation of $`𝒮`$ into $`𝒮^{}`$)
For all $`(x_1\mathrm{},x_p)D^p`$:
$$(𝒮,a_1,\mathrm{},a_p)\phi (x_1,\mathrm{},x_p)(𝒮^{},a_1,\mathrm{},a_p)\phi ^{\prime \prime }(x_1,\mathrm{},x_p).$$
In other words: $`\phi (𝒮)=\phi ^{\prime \prime }(𝒮^{})D^p`$. Then, setting $`\phi ^{}(x_1,\mathrm{},x_p)\phi ^{\prime \prime }(x_1,\mathrm{},x_p)_{ip}D(x_i)`$, it holds: $`\phi (𝒮)=\phi ^{}(𝒮^{})`$
###### Lemma 11
Computing $`𝒮^{}`$ from $`𝒮`$ can be done in linear time $`O_{\rho ,d}(|𝒮|)`$.
Proof. As computing $`𝒮^{}`$ from $`𝒮`$ is easy, one has only to compare the size of the two structures. The size of $`𝒮`$ is:
$$|𝒮|=\mathrm{\Theta }(|D|+\underset{i=1}{\overset{q}{}}card(R_i).arity(R_i))=\mathrm{\Theta }_\rho (|D|+\underset{i=1}{\overset{q}{}}card(R_i)).$$
For $`𝒮^{}`$, by construction, it holds that:
$$|D^{}|=(d+1).|D|+\underset{i=1}{\overset{q}{}}card(R_i)=\mathrm{\Theta }_{d,\rho }(|𝒮|).$$
Hence, $`|𝒮^{}|=\mathrm{\Theta }(m|D^{}|)=\mathrm{\Theta }_{d,\rho }(|𝒮|)`$. $`\mathrm{}`$
We are now ready to state and prove the main result of this section.
###### Theorem 12
$`\text{Query}(\mathrm{𝐅𝐎}_{\mathrm{𝐃𝐞𝐠}})`$ belongs to Constant-Delay<sub>lin</sub>.
Proof. Let $`𝒜`$ be a constant delay algorithm that computes queries of $`\text{Query}(\mathrm{𝐅𝐎}_{\mathrm{𝐁𝐢𝐣}})`$. By using Proposition 10, the algorithm below correctly evaluates queries in $`\text{Query}(\mathrm{𝐅𝐎}_{\mathrm{𝐃𝐞𝐠}})`$.
The cost of instruction 4 is $`O_{\phi ,d}(1)`$, that of instruction 6 is $`O_{\phi ,d}(|𝒮|)`$ (by Lemma 11) and the precomputation part of algorithm $`𝒜`$ (included in instruction 8) is $`O_\phi ^{}(|𝒮^{}|)`$ (hence $`O_{\phi ,d}(|𝒮|)`$) by Theorem 7. These steps form a precomputation phase of time complexity $`O_{\phi ,d}(|𝒮|)`$. Finally, the effective enumeration of $`\phi (𝒮)=\phi ^{}(𝒮^{})`$ is handled on $`𝒮^{},\phi ^{}`$ by $`𝒜`$ and is performed with constant delay. $`\mathrm{}`$
### 3.3 Complexity of subgraphs problems
In this part, we present a simple application of our result to a well-known graph problem. Given two graphs $`G=V;E`$ and $`H=V_H;E_H`$, $`H`$ is said to be a subgraph (resp. induced subgraph) of $`G`$ if there is a one-to-one function $`g`$ from $`V_H`$ to $`V`$ such that, for all $`u,vV_H`$, $`E(g(u),g(v))`$ holds if (resp. if and only if) $`E_H(u,v)`$ holds.
generate subgraph (resp. generate induced subgraph)
Input: any graph $`H`$ and a graph $`G`$ of degree bounded by $`d`$ Parameter: $`|H|,d`$. Output: All the subgraphs (resp. induced subgraphs) of $`G`$ isomorphic to $`H`$.
The treewidth of a graph $`G`$ is the maximal size of a node in a tree decomposition of $`G`$ (see, for example, \[DF99\]). In \[PV90\] it is proved that for graphs $`H`$ of treewidth at most $`w`$, testing if a given graph $`H`$ is an induced subgraph of a graph $`G`$ of degree at most $`d`$ can be done in time $`f(|H|,d).|G|^{w+1}`$. In what follows, we show that there is no reason to focus on graphs of bounded treewidth and that a better bound can be obtained for any graph $`H`$ (provided $`G`$ is of bounded degree). In the result below, we prove that not only the complexity of this decision problem is $`f(|H|,d).|G|`$ but that generating all the (induced) subgraphs isomorphic to $`H`$ can be done with constant delay.
###### Corollary 13
The problem generate subgraph (resp. generate induced subgraph) belongs to Constant-Delay<sub>lin</sub>
Proof. The proof is given for the erate geinduced subgraph problem. Let $`G=V;E`$ and $`H=V_H=\{h_\mathit{1},\mathrm{},h_k\};E_H`$ ($`|V_H|=k`$) be the two inputs of the problem. Since $`G`$ is of maximum degree $`d`$, we can partition its vertex set $`V`$ into $`d`$ sets $`V^0,\mathrm{},V^d`$ where each $`V^\alpha `$ is the set of vertices of degree $`\alpha `$. This can be done in linear time $`O(|G|)`$. We proceed the same for graph $`H`$ and obtain the sets $`V_H^0,\mathrm{},V_H^d`$. In case there exists a vertex in $`H`$ of degree greater than $`d`$, it can be concluded immediately that the problem has no solution. Now, let $`Q`$ be the following formula:
$$Q(x_1,\mathrm{},x_k)\underset{i<jk}{}x_ix_j\underset{V_H^\alpha (h_i)}{}V_G^\alpha (x_i)\underset{E_H(h_i,h_j)}{}E(x_i,x_j).$$
Formula $`Q`$ simply checks that $`H`$ is a subgraph of $`G`$ and that each distinguished vertex $`x_i`$ of $`G`$ has the same degree as its associated vertex $`h_i`$ in $`H`$. Note that formula $`Q`$ only depends on $`H`$ and $`d`$. The result follows now from Theorem 12. $`\mathrm{}`$
## 4 Conclusion
In this paper, we study the complexity of evaluating first-order queries on bounded degree structures and consider this evaluation as a dynamical process, i.e., as an enumeration problem. Our main contributions are two-fold. First, we define a simple quantifier elimination method suitable for first-order formulas which have to be evaluated against a bijective structure. Second, we define a new complexity class, called Constant-Delay<sub>lin</sub>, for enumeration problem which can be seen as the minimal robust complexity class for this kind of problems and we prove that our query problem on bounded degree structures belong to this class.
There are several interesting directions for further researches. Among them, the two following series of questions seem worth to be studied:
* Which ”natural” query problems belong to Constant-Delay<sub>lin</sub> ? More generally, which kind of combinatorial or algorithmic enumeration problems admit constant delay procedures ?
The same questions can be asked for the larger class Constant-Delay<sub>poly</sub> of constant delay enumeration problems for which polynomial time (instead of linear time) precomputations are allowed.
* What are the structural properties of the class Constant-Delay<sub>lin</sub> or of the larger Constant-Delay<sub>poly</sub> ? Do they have complete problems ? Under which kind of reductions ? Could they be proved to be different from the classes of enumeration problems solvable with linear or polynomial delay ?
Acknowledgment. We thank Ron Fagin for a very fruitful email exchange that lead us to define complexity notions about constant delay computation. |
warning/0507/math0507308.html | ar5iv | text | # Cohomological structure of the mapping class group and beyond
## 1. Introduction
We begin by fixing our notations for various groups appearing in this paper. Let $`\mathrm{\Sigma }_g`$ denote a closed oriented surface of genus $`g`$ which will be assumed to be greater than or equal to $`2`$ unless otherwise specified. We denote by $`\mathrm{Diff}_+\mathrm{\Sigma }_g`$ the group of orientation preserving diffeomorphisms of $`\mathrm{\Sigma }_g`$ equipped with the $`C^{\mathrm{}}`$ topology. The same group equipped with the discrete topology is denoted by $`\mathrm{Diff}_+^\delta \mathrm{\Sigma }_g`$. The mapping class group $`_g`$ is the group of path components of $`\mathrm{Diff}_+\mathrm{\Sigma }_g`$. The Torelli group, denoted by $`_g`$, is the subgroup of $`_g`$ consisting of mapping classes which act on the homology group $`H_1(\mathrm{\Sigma }_g;)`$ trivially. Thus we have an extension
(1)
$$1_g_g\mathrm{Sp}(2g,)1$$
where $`\mathrm{Sp}(2g,)`$ denotes the Siegel modular group. Choose an embedded disk $`D\mathrm{\Sigma }_g`$ and a base point $`D\mathrm{\Sigma }_g`$. We denote by $`_{g,1}`$ and $`_{g,1}`$ (resp. $`_{g,}`$ and $`_{g,}`$) the mapping class group and the Torelli group relative to $`D`$ (resp. the base point $``$).
Next let $`F_n`$ denote a free group of rank $`n2`$. Let $`\mathrm{Aut}F_n`$ (resp. $`\mathrm{Out}F_n`$) denote the automorphism group (resp. outer automorphism group) of $`F_n`$. Let $`\mathrm{IAut}_n`$ (resp. $`\mathrm{IOut}_n`$) denote the subgroup of $`\mathrm{Aut}F_n`$ (resp. $`\mathrm{Out}F_n`$) consisting of those elements which act on the abelianization $`H_1(F_n;)`$ of $`F_n`$ trivially. Thus we have an extension
(2)
$$1\mathrm{IOut}_n\mathrm{Out}F_n\mathrm{GL}(n,)1.$$
The fundamental group $`\pi _1(\mathrm{\Sigma }_g\mathrm{Int}D)`$ is a free group of rank $`2g`$. Fix an isomorphism $`\pi _1(\mathrm{\Sigma }_g\mathrm{Int}D)F_{2g}`$. By a classical result of Dehn and Nielsen, we can write
$$_{g,1}=\{\phi \mathrm{Aut}F_{2g},\phi (\gamma )=\gamma \}$$
where the element $`\gamma `$ is defined by
$$\gamma =[\alpha _1,\beta _1]\mathrm{}[\alpha _g,\beta _g]$$
in terms of appropriate free generators $`\alpha _1,\beta _1,\mathrm{},\alpha _g,\beta _g`$ of $`F_{2g}`$. Then we have the following commutative diagram
(3)
$$\begin{array}{ccccccccc}1& & _{g,1}& & _{g,1}& & \mathrm{Sp}(2g,)& & 1\\ & & & & & & & & & & \\ 1& & \mathrm{IAut}_{2g}& & \mathrm{Aut}F_{2g}& & \mathrm{GL}(2g,)& & 1.\end{array}$$
Similarly, for the case of the mapping class group with respect to a base point, we have
(4)
$$\begin{array}{ccccccccc}1& & _{g,}& & _{g,}& & \mathrm{Sp}(2g,)& & 1\\ & & & & & & & & & & \\ 1& & \mathrm{IOut}_{2g}& & \mathrm{Out}F_{2g}& & \mathrm{GL}(2g,)& & 1.\end{array}$$
Next we fix an area form (or equivalently a symplectic form) $`\omega `$ on $`\mathrm{\Sigma }_g`$ and we denote by $`\mathrm{Symp}\mathrm{\Sigma }_g`$ the subgroup of $`\mathrm{Diff}_+\mathrm{\Sigma }_g`$ consisting of those elements which preserve the form $`\omega `$. Also let $`\mathrm{Symp}_0\mathrm{\Sigma }_g`$ be the identity component of $`\mathrm{Symp}\mathrm{\Sigma }_g`$. Moser’s theorem implies that the quotient group $`\mathrm{Symp}\mathrm{\Sigma }_g/\mathrm{Symp}_0\mathrm{\Sigma }_g`$ can be naturally identified with the mapping class group $`_g`$ and we have the following commutatvie diagram
(5)
$$\begin{array}{ccccccccc}1& & \mathrm{Symp}_0\mathrm{\Sigma }_g& & \mathrm{Symp}\mathrm{\Sigma }_g& & _g& & 1\\ & & & & & & & & & & \\ 1& & \mathrm{Diff}_0\mathrm{\Sigma }_g& & \mathrm{Diff}_+\mathrm{\Sigma }_g& & _g& & 1\end{array}$$
where $`\mathrm{Diff}_0\mathrm{\Sigma }_g`$ is the identity component of $`\mathrm{Diff}_+\mathrm{\Sigma }_g`$.
In this paper, we also consider two other groups. Namely the arithmetic mapping class group and the group of homology cobordism classes of homology cylinders. They will be mentioned in $`\mathrm{\S }8`$ and $`\mathrm{\S }11`$ respectively.
## 2. tautological algebra of the mapping class group
Let $`𝐌_g`$ be the moduli space of smooth projective curves of genus $`g`$ and let $`^{}(𝐌_g)`$ be its tautological algebra. Namely it is the subalgebra of the Chow algebra $`𝒜^{}(𝐌_g)`$ generated by the tautological classes $`\kappa _i𝒜^i(𝐌_g)(i=1,2,\mathrm{})`$ introduced by Mumford . Faber made a beautiful conjecture about the structure of $`^{}(𝐌_g)`$. There have been done many works related to and inspired by Faber’s conjecture (we refer to survey papers for some of the recent results including enhancements of Faber’s original conjecture). However the most difficult part of Faber’s conjecture, which claims that $`^{}(𝐌_g)`$ should be a Poincaré duality algebra of dimension $`2g4`$, remains unsettled.
Here we would like to describe a topological approach to Faber’s conjecture, in particular this most difficult part. For this, we denote by
$$e_iH^{2i}(_g;)(i=1,2,\mathrm{})$$
the $`i`$-th Mumford-Morita-Miller tautological class which was defined in as follows. For any oriented $`\mathrm{\Sigma }_g`$-bundle $`\pi :EX`$, the tangent bundle along the fiber of $`\pi `$, denoted by $`\xi `$, is an oriented plane bundle over the total space $`E`$. Hence we have its Euler class $`e=\chi (\xi )H^2(E;)`$. If we apply the Gysin homomorphism (or the integration along the fibers) $`\pi _{}:H^{}(E;)H^2(X;)`$ to the power $`e^{i+1}`$, we obtain a cohomology class
$$e_i(\pi )=\pi _{}(e^{i+1})H^{2i}(X;)$$
of the base space $`X`$. By the obvious naturality of this construction, we obtain certain cohomology classes
$$eH^2(\mathrm{EDiff}_+\mathrm{\Sigma }_g;),e_iH^{2i}(\mathrm{BDiff}_+\mathrm{\Sigma }_g;)$$
where $`\mathrm{EDiff}_+\mathrm{\Sigma }_g\mathrm{BDiff}_+\mathrm{\Sigma }_g`$ denotes the universal oriented $`\mathrm{\Sigma }_g`$-bundle. In the cases where $`g2`$, a theorem of Earle and Eells implies that the two spaces $`\mathrm{EDiff}_+\mathrm{\Sigma }_g`$ and $`\mathrm{BDiff}_+\mathrm{\Sigma }_g`$ are Eilenberg-MacLane spaces $`K(_{g,},1)`$ and $`K(_g,1)`$ respectively. Hence we obtain the universal Euler class $`eH^2(_{g,};)`$ and the Mumford-Morita-Miller classes $`e_iH^{2i}(_g;)`$ as group cohomology classes of the mapping class groups. It follows from the definition that, over the rationals, the class $`e_i`$ is the image of $`(1)^{i+1}\kappa _i`$ under the natural projection $`𝒜^{}(𝐌_g)H^{}(_g;)`$.
Now we define $`^{}(_g)`$ (resp. $`^{}(_{g,})`$) to be the subalgebra of $`H^{}(_g;)`$ (resp. $`H^{}(_{g,};)`$) generated by the classes $`e_1,e_2,\mathrm{}`$ (resp. $`e,e_1,e_2,\mathrm{}`$) and call them the tautological algebra of the mapping class group $`_g`$ (resp. $`_{g,}`$). There is a canonical projection $`^{}(𝐌_g)^{}(_g)`$.
Let us denote simply by $`H`$ (resp. $`H_{}`$) the homology group $`H_1(\mathrm{\Sigma }_g;)`$ (resp. $`H_1(\mathrm{\Sigma }_g;)`$). Also we set
$$U=\mathrm{\Lambda }^3H/\omega _0H,U_{}=U$$
where $`\omega _0\mathrm{\Lambda }^2H`$ denotes the symplectic class. $`U_{}`$ is an irreducible representation of the algebraic group $`\mathrm{Sp}(2g,)`$ corresponding to the Young diagram $`[1^3]`$ consisting of $`3`$ boxes in a single column. Recall here that, associated to any Young diagram whose number of rows is less than or equal to $`g`$, there corresponds an irreducible representation of $`\mathrm{Sp}(2g,)`$ (cf. ). In our papers , we constructed a morphism
(6)
$$\begin{array}{ccc}_{g,}& \stackrel{\rho _2}{}& \left(\left([1^2][2^2]\right)\stackrel{~}{\times }_{\text{torelli}}\mathrm{\Lambda }^3H_{}\right)\mathrm{Sp}(2g,)\\ & & & & \\ _g& \underset{\rho _2}{}& \left([2^2]\stackrel{~}{\times }U_{}\right)\mathrm{Sp}(2g,)\end{array}$$
where $`[2^2]\stackrel{~}{\times }U_{}`$ denotes a central extension of $`U_{}`$ by $`[2^2]`$ corresponding to the unique copy $`[2^2]H^2(U_{})`$ and $`(([1^2][2^2])\stackrel{~}{\times }_{\text{torelli}}\mathrm{\Lambda }^3H_{}`$ is defined similarly (see for details). The diagram (6) induces the following commutative diagram.
(7)
$$\begin{array}{ccc}\left(\mathrm{\Lambda }^{}\mathrm{\Lambda }^3H_{}^{}/\left([1^2]^{\text{torelli}}[2^2]\right)\right)^{Sp}& \stackrel{\rho _2^{}}{}& H^{}(_{g,};)\\ & & & & \\ \left(\mathrm{\Lambda }^{}U_{}^{}/([2^2])\right)^{Sp}& \underset{\rho _2^{}}{}& H^{}(_g;).\end{array}$$
On the other hand, we proved in that the images of the above homomorphisms $`\rho _2^{}`$ are precisely the tautological algebras. Here the concept of the generalized Morita-Mumford classes defined by Kawazumi played an important role. Then in , the effect of unstable degenerations of $`Sp`$-modules appearing in (7) was analized and in particular a part of Faber’s conjecture claiming that $`^{}(_g)`$ is already generated by the classes $`e_1,e_2,\mathrm{},e_{[g/3]}`$ was proved (later Ionel proved this fact at the level of $`^{}(𝐌_g)`$). Although the way of degenerations of $`Sp`$-modules is by no means easy to be studied, it seems natural to expect the following.
###### Conjecture 1.
The natural homomorphisms
$$\left(\mathrm{\Lambda }^{}\mathrm{\Lambda }^3H_{}\right)^{Sp}H^{}(_{g,};),(\mathrm{\Lambda }^{}U_{})^{Sp}H^{}(_g;)$$
induce isomorphisms
$`\left(\mathrm{\Lambda }^{}\mathrm{\Lambda }^3H_{}^{}/\left([1^2]^{\text{torelli}}[2^2]\right)\right)^{Sp}`$ $`^{}(_{g,})`$
$`\left(\mathrm{\Lambda }^{}U_{}^{}/([2^2])\right)^{Sp}`$ $`^{}(_g).`$
Furthermore, the algebras on the left hand sides are Poincaré duality algebras of dimensions $`2g2`$ and $`2g4`$ respectively.
Here we mention that for a single Riemann surface $`X`$, the cohomology $`H^{}(\mathrm{Jac}(X);)`$ is a Poincaré duality algebra of dimension $`2g`$ while it can be shown that there exists a canonical isomorphism
$$H^{}(\mathrm{Jac}(X);)/([1^2])H^{}(X;)$$
which is a Poincaré duality algebra of dimension $`2`$. Here
$$[1^2]H^2(\mathrm{Jac}(X);)$$
denotes the kernel $`\mathrm{Ker}(\mathrm{\Lambda }^2H_{}^{})`$ of the intersection pairing and $`([1^2])`$ denotes the ideal generated by it. Observe that we can write $`\mathrm{\Lambda }^{}U_{}^{}=H^{}(PH^3(\mathrm{Jac}))`$ which is a Poincaré duality algebra of dimension $`\left(\genfrac{}{}{0pt}{}{2g}{3}\right)2g`$, where $`PH^3(\mathrm{Jac})`$ denotes the primitive part of the third cohomology of the Jacobian variety. Hence the above conjecture can be rewritten as
$$\left(H^{}(PH^3(\mathrm{Jac}))/([2^2])\right)^{Sp}^{}(_g)$$
so that it could be phrased as the family version of the above simple fact for a single Riemann surface.
## 3. Higher geometry of the mapping class group
Madsen and Weiss recently proved a remarkable result about the homotopy type of the classifying space of the stable mapping class group. As a corollary, they showed that the stable rational cohomology of the mapping class group is isomorphic to the polynomial algebra generated by the Mumford-Morita-Miller classes
$$\underset{g\mathrm{}}{lim}H^{}(_g;)[e_1,e_2,\mathrm{}].$$
We also would like to mention fundamental results of Tillmann and Madsen and Tillmann .
As was explained in , the classes $`e_i`$ serve as the (orbifold) Chern classes of the tangent bundle of the moduli space $`𝐌_g`$ and it may appear that, stably and quantitatively, the moduli space $`𝐌_g`$ is similar to the classifying space of the unitary group, namely the complex Grassmannian. However, qualitatively the situation is completely different and the moduli space has much deeper structure than the Grassmannian. Here we would like to present a few problems concerning “higher geometry” of the mapping class group where we understand the Mumford-Morita-Miller classes as the primary characteristic classes.
First we recall the following problem, because of its importance, which was already mentioned in (Conjecture 3.4).
###### Problem 2.
Prove (or disprove) that the even Mumford-Morita-Miller classes $`e_{2i}H^{4i}(_g;)`$ are non-trivial, in a suitable stable range, as cohomology classes of the Torelli group.
The difficulty of the above problem comes from the now classical fact, proved by Johnson , that the abelianization of the Torelli group is very big, namely $`H_1(_g;)U_{}(g3)`$. Observe that if $`_g`$ were perfect, then the above problem would have been easily solved by simply applying the Quillen plus construction to each group of the group extension (1) and then looking at the homotopy exact sequence of the resulting fibration. The work of Igusa (in particular Corollary 8.5.17) shows a close connection between the above problem with another very important problem (see Problem 11 in $`\mathrm{\S }`$ 4) of non-triviality of Igusa’s higher Franz-Reidemeister torsion classes in $`H^{4i}(\mathrm{IOut}_n;)`$ (Igusa uses the notation $`\mathrm{Out}^hF_n`$ for the group $`\mathrm{IOut}_n`$). We also refer to a recent work of Sakasai which is related to the above problem.
Next we recall the following two well-known problems about the structure of the Torelli group which are related to a foundational work of Hain .
###### Problem 3.
Determine whether the Torelli group $`_g(g3)`$ is finitely presentable or not (note that $`_g(g3)`$ is known to be finitely generated by Johnson ).
###### Problem 4.
Let $`𝔲_g`$ denote the graded Lie algebra associated to the prounipotent radical of the relative Malcev completion of $`_g`$ defined by Hain and let $`𝔲_g𝔥_g^{}`$ be the natural homomorphism (here $`𝔥_g^{}`$ denotes the graded Lie algebra consisting of symplectic derivations, with positive degrees, of the Malcev Lie algebra of $`\pi _1\mathrm{\Sigma }_g`$). Determine whether this homomorphism is injective or not.
In , we defined a series of secondary characteristic classes for the mapping class group. However there was ambiguity coming from possible odd dimensional stable cohomology classes of the mapping class group. Because of the result of Madsen-Weiss cited above, we can now eliminate the ambiguity and give a precise definition as follows. For each $`i`$, we constructed in explicit group cocycles $`z_iZ^{2i}(_g;)`$ which represent the $`i`$-th Mumford-Morita-Miller class $`e_i`$ by making use of the homomorphism $`_gU\mathrm{Sp}(2g,)`$ constructed in which extends the (first) Johnson homomorphism $`_gU`$. These cocycles are $`_g`$-invariant by the definition. Furthermore we proved that such cocycles are unique up to coboundaries. On the other hand, as is well known, any odd class $`e_{2i1}`$ comes from the Siegel modular group $`\mathrm{Sp}(2g,)`$ so that there is a cocycle $`z_{2i1}^{}Z^{4i2}(_g;)`$ which comes from $`\mathrm{Sp}(2g,)`$. This cocycle is uniquely defined up to coboundaries and $`_g`$-invariant. Now consider the difference $`z_{2i1}z_{2i1}^{}`$. It is a coboundary so that there exists a cochain $`y_iC^{4i3}(_g;)`$ such that $`\delta y_i=z_{2i1}z_{2i1}^{}`$. Since $`H^{4i3}(_g;)=0`$ by (in a suitable stable range), the cochain $`y_i`$ is well-defined up to coboundaries.
Now let $`𝒦_g`$ be the kernel of the Johnson homomorphism so that we have an extension
(8)
$$1𝒦_g_gU1.$$
Recall that Johnson proved that $`𝒦_g`$ is the subgroup of $`_g`$ generated by Dehn twists along separating simple closed curves on $`\mathrm{\Sigma }_g`$. The cocycle $`z_{2i1}^{}`$ is trivial on the Torelli group $`_g`$ while the cocycle $`z_{2i1}`$ (in fact any $`z_i`$) vanishes on $`𝒦_g`$. It follows that the restriction of the cochain $`y_i`$ to $`𝒦_g`$ is a cocycle. Hence we obtain a cohomology class
$$d_i=[y_i|_{𝒦_g}]H^{4i3}(𝒦_g;).$$
This cohomology class is $`_g`$-invariant where $`_g`$ acts on $`H^{}(𝒦_g;)`$ via outer conjugations. This can be shown as follows. For any element $`\phi _g`$, let $`\phi _{}(y_i)`$ be the cochain obtained by applying the conjugation by $`\phi `$ on $`y_i`$. Since both cocycles $`z_{2i1},z_{2i1}^{}`$ are $`_g`$-invariant, we have $`\delta \phi _{}(y_i)=\delta y_i`$. Hence $`\phi _{}(y_i)y_i`$ is a cocycle of $`_g`$. By the result of again, we see that $`\phi _{}(y_i)y_i`$ is a coboundary. Hence the restrictions of $`\phi _{}(y_i)`$ and $`y_i`$ to $`𝒦_g`$ give the same cohomology class.
###### Definition 5.
We call the cohomology classes $`d_iH^{4i3}(𝒦_g;)^_g`$ $`(i=1,2\mathrm{})`$ obtained above the secondary characteristic classes of the mapping class group.
The secondary classes $`d_i`$ are stable in the following sense. Namely the pull back of them in $`H^{4i3}(𝒦_{g,1};)`$ are independent of $`g`$ under natural homomorphisms induced by the inclusions $`𝒦_{g,1}𝒦_{g+1,1}`$ where $`𝒦_{g,1}`$ denotes the subgroup of $`_{g,1}`$ generated by Dehn twists along separating simple closed curves on $`\mathrm{\Sigma }_gD`$. This is because the cocycles $`z_{2i1},z_{2i1}^{}`$ are stable with respect to $`g`$. It follows that the secondary classes $`d_i`$ are defined for all $`g`$ as elements of $`H^{4i3}(𝒦_{g,1};)^{_{g,1}}`$ although we have used the result of , which is valid only in a stable range. However as elements of $`H^{4i3}(𝒦_g;)^_g`$ the class $`d_i`$ is defined only for $`g12i9`$ at present, although it is highly likely that it is defined for all $`g`$. It was proved in that $`d_1`$ is the generator of $`H^1(𝒦_g;)^_g`$ for all $`g2`$. See for another approach to the secondary classes.
###### Problem 6.
Prove that all the secondary classes $`d_2,d_3,\mathrm{}`$ are non-trivial.
Here is a problem concerning the first class $`d_1`$. Let $`C`$ be a separating simple closed curve on $`\mathrm{\Sigma }_g`$ which divides $`\mathrm{\Sigma }_g`$ into two compact surfaces of genera $`h`$ and $`gh`$ and let $`\tau _C𝒦_g`$ be the Dehn twist along $`C`$. Then we know that the value of $`d_1`$ on $`\tau _C𝒦_g`$ is $`h(gh)`$ (up to a constant depending on $`g`$). This is a very simple formula. However at present there is no known algorithm to calculate the value $`d_1(\phi )`$ for a given element $`\phi 𝒦_g`$, say by analyzing the action of $`\phi `$ on $`\pi _1\mathrm{\Sigma }_g`$.
###### Problem 7.
Find explicit way of calculating $`d_1(\phi )`$ for any given element $`\phi 𝒦_g`$. In particular, determine whether the Magnus representation $`_{g,1}\mathrm{GL}(2g;[H])`$ of the Torelli group detects $`d_1`$ or not.
Suzuki proved that the Magnus representation of the Torelli group mentioned above is not faithful so that it may happen that the intersection of the kernel of the Magnus representation with $`𝒦_g`$ is not contained in the kernel of $`d_1`$. We may also ask whether the representation of the hyperelliptic mapping class group given by Jones , restricted to the intersection of this group with $`𝒦_g`$, detects $`d_1`$ or not (cf. Kasahara for a related work for the case $`g=2`$). There are also various interesting works related to the class $`d_1`$ such as Endo and Morifuji treating the hyperelliptic mapping class group, Kitano as well as Hain and Reed .
Recently Biss and Farb proved that the group $`𝒦_g`$ is not finitely generated for all $`g3`$ ($`𝒦_2`$ is known to be an infinitely generated free group by Mess ). However it is still not yet known whether the abelianization $`H_1(𝒦_g)`$ is finitely generated or not (cf. Problem 2.2 of ).
Finally we would like to mention that Kawazumi is developing a theory of harmonic Magnus expansions which gives in particular a system of differential forms representing the Mumford-Morita-Miller classes on the universal family of curves over the moduli space $`𝐌_g`$.
## 4. Outer automorphism group of free groups
As already mentioned in $`\mathrm{\S }`$ 1, let $`F_n`$ denote a free group of rank $`n2`$ and let $`\mathrm{Out}F_n=\mathrm{Aut}F_n/\mathrm{Inn}F_n`$ denote the outer automorphism group of $`F_n`$. In 1986, Culler and Vogtmann defined a space $`X_n`$, called the Outer Space, which plays the role of the Teichmüller space where the mapping class group is replaced by $`\mathrm{Out}F_n`$. In particular, they proved that $`X_n`$ is contractible and $`\mathrm{Out}F_n`$ acts on it properly discontinuously. The quotient space
$$𝐆_n=X_n/\mathrm{Out}F_n$$
is called the moduli space of graphs which is the space of all the isomorphism classes of metric graphs with fundamental group $`F_n`$. Recently many works have been done on the structure of $`\mathrm{Out}F_n`$ as well as $`𝐆_n`$, notably by Vogtmann (see her survey paper ), Bestvina (see ) and many others.
It is an interesting and important problem to compare similarity as well as difference between the mapping class group and $`\mathrm{Out}F_n`$ which will be discussed at several places in this book. Here we would like to concentrate on the cohomological side of this problem.
Hatcher and Vogtmann (see also Hatcher ) proved that the homology of $`\mathrm{Out}F_n`$ stabilizes in a certain stable range. This is an analogue of Harer’s stability theorem for the mapping class group. More precisely, they proved that the natural homomorphisms
$$\mathrm{Aut}F_n\mathrm{Aut}F_{n+1},\mathrm{Aut}F_n\mathrm{Out}F_n$$
induce isomorphisms on the $`i`$-dimensional homology group for $`n2i+2`$ and $`n2i+4`$, respectively. Thus we can speak of the stable cohomology group
$$\underset{n\mathrm{}}{lim}\stackrel{~}{H}^{}(\mathrm{Out}F_n)$$
of $`\mathrm{Out}F_n`$.
In the case of the mapping class group, it was proved in that the natural homomorphism $`_g\mathrm{Sp}(2g,)`$ induces an injection
$$\underset{g\mathrm{}}{lim}H^{}(\mathrm{Sp}(2g,);)[c_1,c_3,\mathrm{}]\underset{g\mathrm{}}{lim}H^{}(_g;)$$
on the stable rational cohomology group where the stable rational cohomology of $`\mathrm{Sp}(2g,)`$ was determined by Borel . In the case of $`\mathrm{Out}F_n`$, Igusa proved the following remarkable result which shows a sharp contrast with the case of the mapping class group (see Theorem 8.5.3 and Remark 8.5.4 of ).
###### Theorem 8 (Igusa).
The homomorphism
$$\stackrel{~}{H}^k(\mathrm{GL}(n,);)\stackrel{~}{H}^k(\mathrm{Out}F_n;)$$
induced by the natural homomorphism $`\mathrm{Out}F_n\mathrm{GL}(n,)`$ is the $`0`$-map in the stable range $`n2k+1`$.
Recall here that the stable cohomology of $`\mathrm{GL}(n,)`$ is given by
$$\underset{n\mathrm{}}{lim}H^{}(\mathrm{GL}(n,);)\mathrm{\Lambda }_{}(\beta _5,\beta _9,\beta _{13},\mathrm{})$$
due to Borel in the above cited papers.
On the other hand, the first non-trivial rational cohomology of the group $`\mathrm{Aut}F_n`$ was given by Hatcher and Vogtmann . They showed that, up to cohomology degree $`6`$, the only non-trivial rational cohomology is
$$H^4(\mathrm{Aut}F_4;).$$
Around the same time, by making use of a remarkable theorem of Kontsevich given in , the author constructed many homology classes in $`H_{}(\mathrm{Out}F_n;)`$ (see and $`\mathrm{\S }10`$ below). The simplest one in this construction gave a series of elements
$$\mu _iH_{4i}(\mathrm{Out}F_{2i+2};)(i=1,2,\mathrm{})$$
and the first one $`\mu _1`$ was shown to be non-trivial by a computer calculation. Responding to an inquiry of the author, Vogtmann communicated us that she modified the argument in to obtain an isomorphism $`H^4(\mathrm{Out}F_4;)`$. Thus we could conclude that $`\mu _1`$ is the generator of this group (see ). Recently Conant and Vogtmann proved that the second class $`\mu _2H_8(\mathrm{Out}F_6;)`$ is also non-trivial in their paper where they call $`\mu _i`$ the Morita classes. Furthermore they constructed many cycles of the moduli space $`𝐆_n`$ of graphs by explicit constructions in the Outer Space $`X_n`$.
More recently, Ohashi determined the rational cohomology group of $`\mathrm{Out}F_n`$ for all $`n6`$ and in particular he showed
$$H_8(\mathrm{Out}F_6;).$$
It follows that $`\mu _2`$ is the generator of this group. At present, the above two groups (and one more group, $`H_7(\mathrm{Aut}F_5;)`$ proved by Gerlits ) are the only known non-trivial rational homology groups of $`\mathrm{Out}F_n`$ (and $`\mathrm{Aut}F_n`$). Now we would like to present the following conjecture based on our expectation that the classes $`\mu _i`$ should concern not only the cohomology of $`\mathrm{Out}F_n`$ but also the structure of the arithmetic mapping class group (see $`\mathrm{\S }8`$) as well as homology cobordism invariants of homology $`3`$-spheres as will be explained in $`\mathrm{\S }11`$ below and .
###### Conjecture 9.
The classes $`\mu _i`$ are non-trivial for all $`i=1,2,\mathrm{}`$.
More generally we have the following.
###### Problem 10.
Produce non-trivial rational (co)homology classes of $`\mathrm{Out}F_n`$.
Next we consider the group $`\mathrm{IOut}_n`$. In Igusa defined higher Franz-Reidemeister torsion classes
$$\tau _{2i}H^{4i}(\mathrm{IOut}_n;)$$
as a special case of his general theory. These classes reflect Igusa’s result mentioned above (Theorem 8) that the pull back of the Borel classes $`\beta _{4i+1}H^{4i+1}(\mathrm{GL}(n,);)`$ in $`H^{4i+1}(\mathrm{Out}F_n;)`$ vanish. However it seems to be unknown whether his classes are non-trivial or not.
###### Problem 11 (Igusa).
Prove that the higher Franz-Reidemeister torsion classes $`\tau _{2i}H^{4i}(\mathrm{IOut}_n;)`$ are non-trivial in a suitable stable range.
In the unstable range, where the Borel classes vanish in $`H^{}(\mathrm{GL}(n,);)`$, there seem to be certain relations between the classes $`\tau _{2i}`$, (dual of) $`\mu _i`$ and unstable cohomology classes in $`H^{}(\mathrm{GL}(n,);)`$. As the first such example, we would like to ask the following specific problem.
###### Problem 12.
Prove (or disprove) that the natural homomorphism
$$H^4(\mathrm{Out}F_4;)H^4(\mathrm{IOut}_4;)^{GL}$$
is an isomorphism where the right hand side is generated by (certain non-zero multiple of) $`\tau _2`$.
Here is another very specific problem. We know the following groups explicitly by various authors:
$`H^8(_{3,};)`$ $`^2(\text{Looijenga }\text{[61]})`$
$`H^8(\mathrm{GL}(6,);)`$ $`(\text{Elbaz-Vincent, Gangl, Soulé }\text{[13]})`$
$`H^8(\mathrm{Out}F_6;)`$ $`(\text{Ohashi }\text{[91]})`$
On the other hand, we have the following natural injection $`i`$ as well as projection $`p`$
(9)
$$_{3,}\stackrel{𝑖}{}\mathrm{Out}F_6\stackrel{𝑝}{}\mathrm{GL}(6,).$$
###### Problem 13.
Determine the homomorphisms
(10)
$$H^8(_{3,};)\stackrel{i^{}}{}H^8(\mathrm{Out}F_6;)\stackrel{p^{}}{}H^8(\mathrm{GL}(6,);)$$
induced by the above homomorphisms in (9).
###### Remark 14.
It seems to be natural to conjecture that the right map in (10) is an isomorphism while the left map is trivial. The former part is based on a consideration of possible geometric meaning of the classes $`\mu _iH_{4i}(\mathrm{Out}F_{2i+2};)`$. For the particular case $`i=2`$ here, it was proved in that $`H^9(\mathrm{GL}(6,);)=0`$. It follows that the Borel class in $`H^9(\mathrm{GL}(6,);)`$ vanishes. Because of this, it is likely that the Igusa class $`\tau _4H^8(\mathrm{IOut}_6;)`$ would vanish as well and the class $`\mu _2`$ would survive in $`H_8(\mathrm{GL}(6,);)`$. For the latter part, see Remark 19 below.
###### Problem 15.
Define unstable (co)homology classes of $`\mathrm{GL}(n,)`$. In particular, what can be said about the image of $`\mu _iH_{4i}(\mathrm{Out}F_{2i+2};)`$ in $`H_{4i}(\mathrm{GL}(2i+2,);)`$ under the projection $`\mathrm{Out}F_{2k+2}\mathrm{GL}(2k+2,)`$?
The above known results as well as explicit computation made so far seem to support the following conjecture (which might be something like a folklore).
###### Conjecture 16.
The stable rational cohomology of $`\mathrm{Out}F_n`$ is trivial. Namely
$$\underset{n\mathrm{}}{lim}\stackrel{~}{H}^{}(\mathrm{Out}F_n;)=0.$$
We can aslo ask how the cohomology of $`\mathrm{Out}F_n`$ with twisted coefficients look like.
###### Problem 17.
Compute the cohomology of $`\mathrm{Aut}F_n`$ and $`\mathrm{Out}F_n`$ with coefficients in various $`\mathrm{GL}(n,)`$-modules.
For example, we could ask how Looijenga’s result for the case of the mapping class group can be generalized in these contexts. We refer to the work of Kawazumi and also Satoh for recent results concerning the above problem.
Finally we recall the following well known problem.
###### Problem 18.
Determine whether the natural homomorphisms
$`\stackrel{~}{H}^{}(\mathrm{Aut}F_{2g};)\stackrel{~}{H}^{}(_{g,1};)`$
$`\stackrel{~}{H}^{}(\mathrm{Out}F_{2g};)\stackrel{~}{H}^{}(_{g,};)`$
induced by the inclusions $`_{g,1}\mathrm{Aut}F_{2g},_{g,}\mathrm{Out}F_{2g}`$ are trivial or not.
We refer to a result of Wahl for a homotopy theoretical property of the homomorphism $`_{g,1}\mathrm{Aut}F_{2g}`$ where $`g`$ tends to $`\mathrm{}`$.
###### Remark 19.
The known results as well as explicit computations made so far seem to suggest that the above maps are trivial. According to a theorem of Kontsevich (see $`\mathrm{\S }9`$ below), the triviality of the second map above is equivalent to the statement that the natural inclusion
$$𝔩_{\mathrm{}}^+𝔞_{\mathrm{}}^+$$
between two infinite dimentional Lie algebras (see $`\mathrm{\S }9`$ for the definition) induces the trivial map
$$H^{}(𝔞_{\mathrm{}}^+)^{Sp}H^{}(𝔩_{\mathrm{}}^+)^{Sp}$$
in the $`Sp`$-invariant cohomology groups. Here the trivial map means that it is the $`0`$-map except for the bigraded parts which correspond to the $`0`$-dimensional homology groups of $`_{g,}`$ and $`\mathrm{Out}F_{2g}`$.
###### Remark 20.
In this paper, we are mainly concerned with the rational cohomology group of the mapping class group, $`\mathrm{Out}F_n`$ and other groups. As for cohomology group with finite coefficients or torsion classes, here we only mention the work of Galatius which determines the mod $`p`$ stable cohomology of the mapping class group and also Hatcher’s result that the stable homology of $`\mathrm{Out}F_n`$ contains the homology of $`\mathrm{\Omega }^{\mathrm{}}S^{\mathrm{}}`$ as a direct summand.
## 5. The derivation algebra of free Lie algebras and the traces
As in $`\mathrm{\S }1`$, let $`F_n`$ be a free group of rank $`n2`$ and let us denote the abelianization $`H_1(F_n)`$ of $`F_n`$ simply by $`H_n`$. Also let $`H_n^{}=H_n`$. Sometimes we omit $`n`$ and we simply write $`H`$ and $`H_{}`$ instead of $`H_n`$ and $`H_n^{}`$. Let
$$_n=_{k=1}^{\mathrm{}}_n(k)$$
be the free graded Lie algebra generated by $`H_n`$. Also let $`_n^{}=_n`$. We set
$$\mathrm{Der}^+(_n)=\{\text{derivation}D\text{of}_n\text{with positive degree}\}$$
which has a natural structure of a graded Lie algebra over $``$. The degree $`k`$ part of this graded Lie algebra can be expressed as
$$\mathrm{Der}^+(_n)(k)=\mathrm{Hom}(H_n,_n(k+1))$$
and we have
$$\mathrm{Der}^+(_n)=\underset{k=1}{\overset{\mathrm{}}{}}\mathrm{Der}^+(_n)(k).$$
Similarly we consider $`\mathrm{Der}^+(_n^{})`$ which is a graded Lie algebra over $``$.
In the case where we are given an identification $`\pi _1(\mathrm{\Sigma }_g\mathrm{Int}D)F_{2g}`$, we have the symplectic class $`\omega _0_{2g}(2)=\mathrm{\Lambda }^2H_{2g}`$ and we can consider the following graded Lie subalgebra
$`𝔥_{g,1}`$ $`=\{D\mathrm{Der}^+(_{2g});D(\omega _0)=0\}`$
$`={\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}𝔥_{g,1}(k).`$
Similarly we have $`𝔥_{g,1}^{}=𝔥_{g,1}`$.
In our paper , for each $`k`$, we defined a certain homomorphism
$$\mathrm{trace}(k):\mathrm{Der}^+(_n)(k)S^kH_n$$
where $`S^kH_n`$ denotes the $`k`$-th symmetric power of $`H_n`$. We call this “trace” because it is defined as the usual trace of the abelianized non-commutative Jacobian matrix of each homogeneous derivation. Here we recall the definition briefly from the above cited paper. Choose a basis $`x_1,\mathrm{},x_n`$ of $`H_n=H_1(F_n;)`$. We can consider $`_n(k+1)`$ as a natural submodule of $`H_n^{(k+1)}`$ consisting of all the Lie polynomials of degree $`k+1`$. For example $`[x_1,x_2]_n(2)`$ corresponds to the element $`x_1x_2x_2x_1H^2`$. By using the concept of the Fox free differential, we can also embed $`_n(k+1)`$ into the set $`(H_n^k)^n`$ of all the $`n`$-dimensional column vectors with entries in $`H_n^k`$ by the following correspondence
$$_n(k+1)\eta \left(\frac{\eta }{x_i}\right)(H_n^k)^n.$$
Here for each monomial $`\eta _n(k+1)H_n^{(k+1)}`$ which is uniquely expressed as
$$\eta =\eta _1x_1+\mathrm{}+\eta _nx_n(\eta _iH_n^k),$$
we have
$$\frac{\eta }{x_i}=\eta _i.$$
###### Definition 21.
In the above terminologies, the $`k`$-th trace $`\mathrm{trace}(k):\mathrm{Der}^+(_n)(k)S^kH_n`$ is defined by
$$\mathrm{trace}(k)(f)=\left(\underset{i=1}{\overset{n}{}}\frac{f(x_i)}{x_i}\right)^{\mathrm{ab}}$$
where $`f\mathrm{Der}^+(_n)(k)=\mathrm{Hom}(H_n,_n(k+1))`$ and the superscript $`\mathrm{ab}`$ denotes the natural projection $`H_n^kS^kH_n`$.
###### Remark 22.
If we identify the target $`\mathrm{Hom}(H_n,_n(k+1))`$ of $`\mathrm{trace}(k)`$ with
$$H_n^{}_n(k+1)H_n^{}H_n^{(k+1)}$$
where $`H_n^{}=\mathrm{Hom}(H_n,)`$ denotes the dual space of $`H_n`$, then it follows immediately from the definition that $`\mathrm{trace}(k)`$ is equal to the restriction of the contraction
$$C_{k+1}:H_n^{}H_n^{(k+1)}H_n^k$$
followed by the abelianization $`H_n^kS^kH_n`$. Here
$$C_{k+1}(fu_1\mathrm{}u_{k+1})=f(u_{k+1})u_1\mathrm{}u_k$$
for $`fH_n^{},u_iH_n`$. Also it is easy to see that, if we replace $`C_{k+1}`$ with $`C_1`$ defined by
$$C_1(fu_1\mathrm{}u_{k+1})=f(u_1)u_2\mathrm{}u_{k+1}$$
in the above discussion, then we obtain $`(1)^k\mathrm{trace}(k)`$.
###### Example 23.
Let $`\mathrm{ad}_{x_2}(x_1)^k\mathrm{Der}^+(_n)(k)`$ be the element defined by
$`\mathrm{ad}_{x_2}(x_1)^k(x_2)`$ $`=[x_1,[x_1,[\mathrm{},[x_1,x_2]\mathrm{}](k\text{-times }x_1)`$
$`\mathrm{ad}_{x_2}(x_1)^k(x_i)`$ $`=0(i2).`$
Then a direct computation shows that
$$\mathrm{trace}(k)(\mathrm{ad}_{x_2}(x_1)^k)=x_1^k.$$
As was mentioned in , the traces are $`GL(H_n)`$-equivariant in an obvious sense. Since $`x_1^k`$ generates $`S^kH_n`$ as a $`GL(H_n)`$-module, the above example implies that the mapping $`\mathrm{trace}(k):\mathrm{Der}^+(_n)(k)S^kH_n`$ is surjective for any $`k`$. Another very important property of the traces proved in the above cited paper is that they vanish identically on the commutator ideal $`[\mathrm{Der}^+(_n),\mathrm{Der}^+(_n)]`$. Hence we have the following surjective homomorphism of graded Lie algebras
$$(\tau _1,_k\mathrm{trace}(k)):\mathrm{Der}^+(_n)\mathrm{Hom}(H_n,\mathrm{\Lambda }^2H_n)\underset{k=2}{\overset{\mathrm{}}{}}S^kH_n$$
where the target is understood to be an abelian Lie algebra. We have also proved that, for any $`k`$, $`\mathrm{trace}(2k)`$ vanishes identically on $`𝔥_{g,1}`$ and that $`\mathrm{trace}(2k+1):𝔥_{g,1}^{}(2k+1)S^{2k+1}H_{2g}^{}`$ is surjective. Thus we have a surjective homomorphism
(11)
$$(\tau _1,_k\mathrm{trace}(2k+1)):𝔥_{g,1}^{}\mathrm{\Lambda }^3H_{2g}^{}\underset{k=1}{\overset{\mathrm{}}{}}S^{2k+1}H_{2g}^{}$$
of graded Lie algebras which we conjectured to give the abelianization of the Lie algebra $`𝔥_{g,1}^{}`$ (see Conjecture 6.10 of ).
Recently Kassabov (Theorem 1.4.11) proved the following remarkable result. Let $`x_1,\mathrm{},x_n`$ be a basis of $`H_n`$ as before.
###### Theorem 24 (Kassabov).
Up to degree $`n(n1)`$, the graded Lie algebra $`\mathrm{Der}^+(_n^{})`$ is generated as a Lie algebra and $`𝔰𝔩(n,)`$-module by the elements $`\mathrm{ad}(x_1)^k(k=1,2,\mathrm{})`$ and the element $`D`$ which sends $`x_1`$ to $`[x_2,x_3]`$ and $`x_i(i1)`$ to $`0`$.
If we combine this theorem with the concept of the traces, we obtain the following.
###### Theorem 25.
The surjective Lie algebra homomorphism
$$(\tau _1,_k\mathrm{trace}(k)):\mathrm{Der}^+(_n^{})\mathrm{Hom}(H_n^{},\mathrm{\Lambda }^2H_n^{})\underset{k=2}{\overset{\mathrm{}}{}}S^kH_n^{},$$
induced by the degree $`1`$ part and the traces, gives the abelianization of the graded Lie algebra $`\mathrm{Der}^+(_n^{})`$ up to degree $`n(n1)`$ so that any element of degree $`2dn(n1)`$ with vanishing trace belongs to the commutator ideal $`[\mathrm{Der}^+(_n^{}),\mathrm{Der}^+(_n^{})]`$. Furthermore, any $`𝔰𝔩(n,)`$-equivariant splitting to this abelianization generates $`\mathrm{Der}^+(_n^{})`$ in this range. Hence stably there exists an isomorphism
$$H_1\left(\mathrm{Der}^+(_{\mathrm{}}^{})\right)\mathrm{Hom}(H_{\mathrm{}}^{},\mathrm{\Lambda }^2H_{\mathrm{}}^{})\underset{k=2}{\overset{\mathrm{}}{}}S^kH_{\mathrm{}}^{}$$
and the degree $`1`$ part and (any $`𝔰𝔩(n,)`$-equivariant splittings of) the traces generate $`\mathrm{Der}^+(_{\mathrm{}}^{})`$.
Although the structure of $`𝔥_{g,1}^{}`$ is much more complicated than that of $`\mathrm{Der}^+(_n^{})`$, Kassabov’s argument adapted to this case together with some additional idea will produce enough information about the generation as well as the abelianization of $`𝔥_{g,1}^{}`$ in a certain stable range. Details will be given in our forthcoming paper . It follows that any element in the Lie algebra $`𝔥_{\mathrm{},1}^{}`$ can be expressed in terms of the degree $`1`$ part and the traces.
## 6. The second cohomology of $`𝔥_{g,1}^{}`$
In this section, we define a series of elements in $`H^2(𝔥_{g,1}^{})^{Sp}`$ which denotes the $`Sp`$-invariant part of the second cohomology of the graded Lie algebra $`𝔥_{g,1}^{}`$.
As is well known, $`U_{}=\mathrm{\Lambda }^3H_{}/\omega _0H_{}`$ and $`S^{2k+1}H_{}(k=1,2,\mathrm{})`$ are all irreducible representations of $`\mathrm{Sp}(2g,)`$. It is well known in the representation theory that, for any irreducible representation $`V`$ of the algebraic group $`\mathrm{Sp}(2g,)`$, the tensor product $`VV`$ contains a unique trivial summand $`VV`$ (cf. for generalities of the representations of the algebraic group $`\mathrm{Sp}(2g,)`$). In our case where $`V`$ is any of the above irreducible representations, it is easy to see that the trivial summand appears in the second exterior power part $`\mathrm{\Lambda }^2VVV`$. It follows that each of
$$\mathrm{\Lambda }^2U_{},\mathrm{\Lambda }^2S^3H_{},\mathrm{\Lambda }^2S^5H_{},\mathrm{}$$
contains a unique trivial summand $``$. Let
$$\iota _1:\mathrm{\Lambda }^2U_{},\iota _{2k+1}:\mathrm{\Lambda }^2S^{2k+1}H_{}(k=1,2,\mathrm{})$$
be the unique (up to scalars) $`Sp`$-equivariant homomorphism. We would like to call them higher intersection pairing on surfaces which generalize the usual one $`\mathrm{\Lambda }^2H_{}`$. We can write
$$\iota _1H^2(U_{})^{Sp},\iota _{2k+1}H^2(S^{2k+1}H_{})^{Sp}.$$
###### Definition 26.
We define the cohomology classes
$$e_1,t_3,t_5,\mathrm{}H^2(𝔥_{g,1}^{})^{Sp}$$
by setting
$$e_1=\overline{\tau }_1^{}(\iota _1),t_{2k+1}=\mathrm{trace}(2k+1)^{}(\iota _{2k+1})$$
where $`\overline{\tau }_1`$ denotes the composition $`𝔥_{g,1}^{}\mathrm{\Lambda }^3H_{}U_{}`$.
###### Conjecture 27.
The classes $`e_1,t_3,t_5,\mathrm{}`$ are all non-trivial. Furthermore they are linearly independent and form a basis of $`H^2(𝔥_{g,1}^{})^{Sp}`$.
###### Remark 28.
The element $`e_1`$ is the Lie algebra version of the first Mumford-Morita-Miller class (we use the same notation).
###### Remark 29.
The Lie algebra $`𝔥_{g,1}^{}`$ is graded so that the cohomology group $`H^2(𝔥_{g,1}^{})^{Sp}`$ is bigraded. Let $`H^2(𝔥_{g,1}^{})_n^{Sp}`$ denote the weight $`n`$ part of $`H^2(𝔥_{g,1}^{})^{Sp}`$ (see $`\mathrm{\S }9`$ for more details). Then by definition we have
$$e_1H^2(𝔥_{g,1}^{})_2^{Sp},t_{2k+1}H^2(𝔥_{g,1}^{})_{4k+2}^{Sp}.$$
Hence if the elements $`e_1,t_3,t_5,\mathrm{}`$ are non-trivial, then they are automatically linearly independent. Thus the above conjecture can be rewritten as
$$H^2(𝔥_{g,1}^{})_n^{Sp}\{\begin{array}{cc}\hfill & (n=2,6,10,14,\mathrm{})\hfill \\ 0\hfill & (\text{otherwise})\hfill \end{array}$$
where the summands $``$ in degrees $`2,6,10,\mathrm{}`$ are generated by the above classes.
As for the non-triviality, all we know at present is the non-triviality of $`e_1,t_3,t_5`$. The non-triviality of the class $`t_{2k+1}`$ is the same as that of the class $`\mu _k`$ because of the theorem of Kontsevich described in $`\mathrm{\S }9`$. This will be explained in that section.
## 7. Constructing cohomology classes of $`𝔥_{g,1}^{}`$
In this section, we describe a general method of constructing $`Sp`$-invariant cohomology classes of the Lie algebra $`𝔥_{g,1}^{}`$ which generalize the procedure given in the previous section. As was already mentioned in our paper , the homomorphism (11) induces the following homomorphim in the $`Sp`$-invariant part of the cohomology
(12)
$$H^{}\left(\mathrm{\Lambda }^3H_{}\underset{k=1}{\overset{\mathrm{}}{}}S^{2k+1}H_{2g}^{}\right)^{Sp}H^{}(𝔥_{g,1}^{})^{Sp}.$$
By the same way as in , the left hand side can be computed by certain polynomial algebra
$$[\mathrm{\Gamma };\mathrm{\Gamma }𝒢^{odd}]$$
generated by graphs belonging to $`𝒢^{odd}`$ which denotes the set of all isomorphism classes of connected graphs with valencies in the set
$$3,3,5,7,\mathrm{}$$
of odd integers. Here we write two copies of $`3`$ because of different roles: the first one is related to the target $`\mathrm{\Lambda }^3H_{}`$ of $`\tau _1`$ (alternating) while the second one is related to the target $`S^3H_{}`$ of $`\mathrm{trace}(3)`$ (symmetric). The other $`2k+1(k=2,3,\mathrm{})`$ are related to the target $`S^{2k+1}H_{}`$ of $`\text{trace}(2k+1)`$. Thus we obtain a homomorphism
(13)
$$\mathrm{\Phi }:[\mathrm{\Gamma };\mathrm{\Gamma }𝒢^{odd}]H^{}(𝔥_{g,1}^{})^{Sp}.$$
The elements $`e_1,t_3,t_5,\mathrm{}`$ defined in Definition 26 arise as the images, under $`\mathrm{\Phi }`$, of those graphs with exactly two vertices which are connected by $`3,3,5,7,\mathrm{}`$ edges.
###### Remark 30.
As was mentioned already in the previous section $`\mathrm{\S }6`$, the cohomology of $`𝔥_{g,1}^{}`$ is bigraded. Let $`\mathrm{\Gamma }𝒢^{odd}`$ be a connected graph whose valencies are $`v_3^a`$ times $`3`$ (alternating), $`v_3^s`$ times $`3`$ (symmetric) and $`v_{2k+1}`$ times $`2k+1(2k+1>3)`$. Then
(14)
$$\mathrm{\Phi }(\mathrm{\Gamma })H^d(𝔥_{g,1}^{})_n^{Sp}$$
where
$`d`$ $`=v_3^a+v_3^s+v_5+v_7+\mathrm{}`$
$`n`$ $`=v_3^a+3v_3^s+5v_5+7v_7+\mathrm{}.`$
Observe that $`n+2v_3^a`$ is equal to twice of the number of edges of $`\mathrm{\Gamma }`$. It follows that $`n`$ (and hence $`d`$) is always an even integer.
###### Problem 31.
Find explicit graphs $`\mathrm{\Gamma }𝒢^{odd}`$ such that the corresponding homology classes $`\mathrm{\Phi }(\mathrm{\Gamma })`$ are non-trivial.
## 8. Three groups beyond the mapping class group
In view of the definition of the Lie algebra $`𝔥_{g,1}`$ (see $`\mathrm{\S }5`$), we may say that it is the “Lie algebra version” of the mapping class group $`_{g,1}`$. However the result of the author showed that it is too big to be considered so and the following question arose: what is the algebraic and/or geometric meaning of the complement of the image of $`_{g,1}`$ in $`𝔥_{g,1}`$ ? Two groups came into play in this framework in the 1990’s. One is the arithmetic mapping class group through the works of number theorists, notably Oda, Nakamura and Matsumoto, and the other is $`\mathrm{Out}F_n`$ via the theorem of Kontsevich described in the next section $`\mathrm{\S }9`$. In this section, we would like to consider the former group briefly from a very limited point of view (see and references therein for details). The latter group was already introduced in $`\mathrm{\S }4`$ and will be further discussed in $`\mathrm{\S }10`$.
More recently, it turned out that we have to treat one more group in the above setting and that is the group of homology cobordism classes of homology cylinders. This will be discussed in $`\mathrm{\S }11`$ below. We strongly expect that the traces will give rise to meaningful invariants in each of these three groups beyond the mapping classs group.
Now we consider the first group above. The action of $`_{g,1}`$ on the lower central series of $`\mathrm{\Gamma }=\pi _1(\mathrm{\Sigma }_g\mathrm{Int}D)`$ induces a filtration $`\{_{g,1}(k)\}_k`$ on $`_{g,1}`$ as follows. Let $`\mathrm{\Gamma }_1=[\mathrm{\Gamma },\mathrm{\Gamma }]`$ be the commutator subgroup of $`\mathrm{\Gamma }`$ and inductively define $`\mathrm{\Gamma }_{k+1}=[\mathrm{\Gamma },\mathrm{\Gamma }_k](k=1,2,\mathrm{})`$. The quotient group $`N_k=\mathrm{\Gamma }/\mathrm{\Gamma }_k`$ is called the $`k`$-th nilpotent quotient of $`\mathrm{\Gamma }`$. Note that $`N_1`$ is canonically isomorphic to $`H_{2g}=H_1(\mathrm{\Sigma }_g;)`$. Now we set
$$_{g,1}(k)=\{\phi _{g,1};\phi \text{ acts on }N_k\text{ trivially}\}.$$
Thus the first group $`_{g,1}(1)`$ in this filtration is nothing but the Torelli group $`_{g,1}`$. As is well known, the quotient group $`\mathrm{\Gamma }_k/\mathrm{\Gamma }_{k+1}`$ can be identified with the $`(k+1)`$-st term $`_{2g}(k+1)`$ of the free graded Lie algebra generated by $`H_{2g}`$ (see $`\mathrm{\S }`$ 5). It can be checked that the correspondence
$`_{g,1}(k)`$ $`\phi `$
$`\mathrm{\Gamma }\alpha \phi _{}(\alpha )\alpha ^1\mathrm{\Gamma }_k/\mathrm{\Gamma }_{k+1}_{2g}(k+1)`$
descends to a homomorphism
$$\tau _k:_{g,1}(k)\mathrm{Hom}(H_{2g},_{2g}(k+1))$$
which is now called the $`k`$-th Johnson homomorphism because it was introduced by Johnson (see ). Furthermore it turns out that the totality $`\{\tau _k\}_k`$ of these homomorphims induces an injective homomorphism of graded Lie algebras
(15)
$$\mathrm{Gr}^+(_{g,1})=\underset{k=1}{\overset{\mathrm{}}{}}_{g,1}(k)/_{g,1}(k+1)𝔥_{g,1}$$
(see for details). Although there have been obtained many important results concerning the image of the above homomorphism, the following is still open.
###### Problem 32.
Determine the image as well as the cokernel of the homomorphism (15) explicitly.
Note that Hain proved that the image of (15), after tensored with $``$, is precisely the Lie subalgebra generated by the degree $`1`$ part. However it is unclear which part of $`𝔥_{g,1}^{}`$ belongs to this Lie subalgebra.
In relation to this problem, Oda predicted, in the late 1980’s, that there should arise “arithmetic obstructions” to the surjectivity of Johnson homomorphism. More precisely, based on the theory of Ihara in number theory which treated mainly the case $`g=0,n=3`$, he expected that the absolute Galois group $`\mathrm{Gal}(\overline{}/)`$ should “appear” in $`𝔥_{g,1}_{\mathrm{}}`$ outside of the geometric part and which should be $`Sp`$-invariant for any genus $`g`$ and for any prime $`\mathrm{}`$. In 1994, Nakamura proved, among other results, that this is in fact the case (see also Matsumoto ). This was the second obstruction to the surjectivity of Johnson homomorphism, the first one being the traces in . This raised the following problem.
###### Problem 33.
Describe the Galois images in $`𝔥_{g,1}_{\mathrm{}}`$.
The above result was proved by analyzing the number theoretical enhancement of the Johnson homomorphism where the geometric mapping class group is replaced by the arithmetic mapping class group which is expressed as an extension
$$1\widehat{}_g^n\pi _1^{alg}𝐌_g^n/\mathrm{Gal}(\overline{}/)1$$
and studied by Grothendieck, Deligne, Ihara, Drinfel’d and many number theorists. Nakamura continued to study the structure of the mapping class group from the point of view of number theory extensively (see e.g. ). On the other hand, Hain and Matsumoto recently proved remarkable results concerning this subject (see ). In view of deep theories in number theory, as well as the above explicit results, it seems to be conjectured that there should exist an embedding
$$\text{FreeLie}_{}(\sigma _3,\sigma _5,\mathrm{})𝔥_{g,1}$$
of certain free graded Lie algebra over $``$ generated by certain elements $`\sigma _3,\sigma _5,\mathrm{}`$, corresponding to the Soulé elements, into $`𝔥_{g,1}`$ such that the tensor product of it with $`_{\mathrm{}}`$ coincides with the image of $`\mathrm{Gal}(\overline{}/)`$ for any prime $`\mathrm{}`$.
We expect that the above conjectured free graded Lie algebra (the motivic Lie algebra) can be realized inside $`𝔥_{g,1}`$ (in fact inside the commutator ideal $`[𝔥_{g,1},𝔥_{g,1}]`$) explicitly in terms of the traces. In some sense, the elements $`\sigma _{2k+1}`$ should be decomposable in higher genera. Here we omit the precise form of the expected formula which will be given in a forthcoming paper.
Finally we mention the analogue of the Johnson homomorphisms for the group $`\mathrm{Aut}F_n`$ very briefly. Prior to the work of Johnson, Andreadakis introduced and studied the filtration on $`\{\mathrm{Aut}F_n(k)\}_k`$ which is induced from the action of $`\mathrm{Aut}F_n`$ on the lower central series of $`F_n`$. The first group $`\mathrm{Aut}F_n(1)`$ in this filtration is nothing but the group $`\mathrm{IAut}_n`$. It can be checked that an analogous procedure as in the case of the mapping class group gives rise to certain homomorphisms
$$\tau _k:\mathrm{Aut}F_n(k)\mathrm{Hom}(H_n,_n(k+1))$$
and the totality $`\{\tau _k\}_k`$ of these homomorphims induces an injective homomorphism of graded Lie algebras
(16)
$$\mathrm{Gr}^+(\mathrm{Aut}F_n)=\underset{k=1}{\overset{\mathrm{}}{}}\mathrm{Aut}F_n(k)/\mathrm{Aut}F_n(k+1)\mathrm{Der}^+(_n).$$
We refer to for some of the recent works related to the above homomorphism.
## 9. A theorem of Kontsevich
In this section, we recall a theorem of Kontsevich described in which is the key result for the argument given in the next section. See also the paper by Conant and Vogtmann for a detailed proof as well as discussion of this theorem in the context of cyclic operads. In the above cited papers, Kontsevich considered three kinds of infinite dimensional Lie algebras denoted by $`𝔠_g,𝔞_g,𝔩_g`$ (commutative, associative, and lie version, respectively). The latter two Lie algebras are defined by
$`𝔞_g`$ $`=\{\text{derivation }D\text{ of the tensor algebra }T^{}(H_{})`$
$`\text{such that }D(\omega _0)=0\}`$
$`𝔩_g`$ $`=\{\text{derivation }D\text{ of the free Lie algebra }_{2g}^{}T^{}(H_{})`$
$`\text{such that }D(\omega _0)=0\}.`$
There is a natural injective Lie algebra homomorphism $`𝔩_g𝔞_g`$. The degree $`0`$ part of both of $`𝔞_g,𝔩_g`$ is the Lie algebra $`𝔰𝔭(2g,)`$ of $`\mathrm{Sp}(2g,)`$. Let $`𝔞_g^+`$ (resp. $`𝔩_g^+`$) denote the Lie subalgebra of $`𝔞_g`$ (resp. $`𝔩_g`$) consisting of derivations with positive degrees. Then the latter one $`𝔩_g^+`$ is nothing other than the Lie algebra $`𝔥_{g,1}^{}`$ considered in $`\mathrm{\S }5`$. Now Kontsevich described the stable homology groups of the above three Lie algebras (where $`g`$ tends to $`\mathrm{}`$) in terms of cohomology groups of graph complexes, moduli spaces $`𝐌_g^m`$ of Riemann surfaces and the outer automorphism groups $`\mathrm{Out}F_n`$ of free groups (or the moduli space of graphs), respectively. Here is the statement for the cases of $`𝔞_{\mathrm{}},𝔩_{\mathrm{}}`$.
###### Theorem 34 (Kontsevich).
There are isomorphisms
$`PH_{}(𝔞_{\mathrm{}})`$ $`PH_{}(𝔰𝔭(\mathrm{},)){\displaystyle \underset{g0,m1,2g2+m>0}{}}H^{}(𝐌_g^m;)^{𝔖_m},`$
$`PH_{}(𝔩_{\mathrm{}})`$ $`PH_{}(𝔰𝔭(\mathrm{},)){\displaystyle \underset{n2}{}}H^{}(\mathrm{Out}F_n;).`$
Here $`P`$ denotes the primitive parts of $`H_{}(𝔞_{\mathrm{}}),H_{}(𝔩_{\mathrm{}})`$ which have natural structures of Hopf algebras and $`𝐌_g^m`$ denotes the moduli space of genus $`g`$ smooth curves with $`m`$ punctures.
Here is a very short outline of the proof of the above theorem. Using natural cell structure of the Riemann moduli space $`𝐌_g^m(m1)`$ and the moduli space $`𝐆_n`$ of graphs, which serves as the (rational) classifying space of $`\mathrm{Out}F_n`$ by , Kontsevich introduced a natural filtration on the cellular cochain complex of these moduli spaces. Then he proved that the associated spectral sequence degenerates at the $`E_2`$-term and only the diagonal terms remain to be non-trivial. On the other hand, by making use of classical representation theory for the group $`\mathrm{Sp}(2g,)`$, he constructed a quasi isomorphism between the $`E_1`$-term and the chain complexes of the relevant Lie algebras ($`𝔞_{\mathrm{}}`$ or $`𝔩_{\mathrm{}}`$). For details, see the original papers cited above as well as .
There is also the dual version of the above theorem which connects the primitive cohomology of the relevant Lie algebras with the homology groups of the moduli space or $`\mathrm{Out}F_n`$. We would like to describe it in a detailed form because this version is most suitable for our purpose. The Lie algebras $`𝔩_{\mathrm{}}^+,𝔞_{\mathrm{}}^+`$ are graded. Hence their $`Sp`$-invariant cohomology groups are bigraded. Let $`H^k(𝔩_{\mathrm{}}^+)_n^{Sp}`$ and $`H^k(𝔞_{\mathrm{}}^+)_n^{Sp}`$ denote the weight $`n`$ part of $`H^k(𝔩_{\mathrm{}}^+)^{Sp}`$ and $`H^k(𝔞_{\mathrm{}}^+)^{Sp}`$ respectively. Then we have the following isomorphisms.
(17) $`PH^k(𝔩_{\mathrm{}}^+)_{2n}^{Sp}`$ $`H_{2nk}(\mathrm{Out}F_{n+1};)`$
(18) $`PH^k(𝔞_{\mathrm{}}^+)_{2n}^{Sp}`$ $`{\displaystyle \underset{2g2+m=n}{}}H_{2nk}(𝐌_g^m;)_{𝔖_m}`$
## 10. Constructing homology classes of $`\mathrm{Out}F_n`$
In this section, we combine our construction of many cohomology classes in $`H^{}(𝔥_{g,1}^{})^{Sp}`$ given in $`\mathrm{\S }6,\mathrm{\S }7`$ with Kontsevich’s theorem given in $`\mathrm{\S }9`$ to produce homology classes of the group $`\mathrm{Out}F_n`$.
First we see that the homomorphism given in (13) is stable with respect to $`g`$. More precisely, the following diagram is commutative
(19)
$$\begin{array}{ccc}[\mathrm{\Gamma };\mathrm{\Gamma }𝒢^{odd}]& \stackrel{\mathrm{\Phi }}{}& H^{}(𝔥_{g+1,1}^{})^{Sp}\\ & & & & \\ [\mathrm{\Gamma };\mathrm{\Gamma }𝒢^{odd}]& \underset{\mathrm{\Phi }}{}& H^{}(𝔥_{g,1}^{})^{Sp},\end{array}$$
where the right vertical map is induced by the inclusion $`𝔥_{g,1}𝔥_{g+1,1}`$. This follows from the fact that the traces $`\mathrm{trace}(2k+1)`$ as well as $`\tau _1`$ are all stable with respect to $`g`$ in an obvious way.
Next, we see that the cohomology class $`\mathrm{\Phi }(\mathrm{\Gamma })H^{}(𝔥_{\mathrm{},1}^{})^{Sp}`$ obtained in this way is primitive if and only if $`\mathrm{\Gamma }`$ is connected. Keeping in mind the fact $`𝔥_{\mathrm{},1}^{}=𝔩_{\mathrm{}}^+`$, the property (14), as well as the version of Kontsevich’s theorem given in (17) we now obtain the following theorem.
###### Theorem 35.
Associated to each connected graph $`\mathrm{\Gamma }𝒢^{odd}`$ whose valencies are $`v_3^a`$ times $`3`$ (alternating), $`v_3^s`$ times $`3`$ (symmetric) and $`v_{2k+1}`$ times $`2k+1(2k+1>3)`$, we have a homology class
$$\mathrm{\Phi }(\mathrm{\Gamma })H_{2nd}(\mathrm{Out}F_{n+1};)$$
where
$$d=v_3^a+v_3^s+v_5+v_7+\mathrm{},2n=v_3^a+3v_3^s+5v_5+7v_7+\mathrm{}.$$
###### Remark 36.
Let $`\mathrm{\Gamma }_{2k+1}`$ be the connected graph with two vertices both of which have valency $`2k+1`$. Then $`d=2,2n=4k+2`$ and $`\mathrm{\Phi }(\mathrm{\Gamma }_{2k+1})H_{4k}(\mathrm{Out}F_{2k+2};)`$ is the class $`\mu _k`$ already mentioned in $`\mathrm{\S }4`$.
The following problem is an enhancement of Problem 31 in the context of the homology of the moduli space of graphs rather than the cohomology of the Lie algebra $`𝔥_{g,1}^{}`$.
###### Problem 37.
Give examples of odd valent graphs $`\mathrm{\Gamma }`$ whose associated homology classes $`\mathrm{\Phi }(\mathrm{\Gamma })H_{}(\mathrm{Out}F_n;)`$ are non-trivial as many as possible. Also compare these classes with the homology classes constructed by Conant and Vogtmann as explicit cycles in the moduli space of graphs. Furthermore investigate whether these classes survive in $`H_{}(\mathrm{GL}(n,);)`$, or else come from $`H_{}(\mathrm{IOut}_n;)`$, or not.
###### Remark 38.
It seems that the geometric meaning of Kontsevich’s theorem is not very well understood yet. In particular, there is almost no known relations between the classes $`\mathrm{\Phi }(\mathrm{\Gamma })H_{}(\mathrm{Out}F_n;)`$ where the rank $`n`$ varies. However, there should be some unknown structures here. For example, there are graphs $`\mathrm{\Gamma }`$ whose associated classes $`\mathrm{\Phi }(\mathrm{\Gamma })`$ lie in $`H_8(\mathrm{Out}F_n;)`$ for $`n=7,8`$ which might be closely related to the class $`\mu _2H_8(\mathrm{Out}F_6;)`$.
## 11. Group of homology cobordism classes of homology cylinders
In his theory developed in , Habiro introduced the concept of homology cobordism of surfaces and proposed interesting problems concerning it. Goussarov also studied the same thing in his theory. It played an important role in the classification theory of $`3`$-manifolds now called the Goussarov-Habiro theory. Later Garoufalidis and Levine and Levine used this concept to define a group $`_{g,1}`$ which consists of homology cobordism classes of homology cylinders on $`\mathrm{\Sigma }_g\mathrm{Int}D`$. We refer to the above papers for the definition (we use the terminology homology cylinder following them) as well as many interesting questions concerning the structure of $`_{g,1}`$. It seems that the importance of this group is growing recently.
Here we summarize some of the results of which will be necessary for our purpose here. Consider $`\mathrm{\Gamma }=\pi _1(\mathrm{\Sigma }_g\mathrm{Int}D)`$, which is isomorphic to $`F_{2g}`$, and let $`\{\mathrm{\Gamma }_k\}_k`$ be its lower central series as before. Note that $`\mathrm{\Gamma }`$ contains a particular element $`\gamma \mathrm{\Gamma }`$ which corresponds to the unique relation in $`\pi _1\mathrm{\Sigma }_g`$. They define
$`\mathrm{Aut}`$ $`{}_{0}{}^{}(\mathrm{\Gamma }/\mathrm{\Gamma }_k)=\{f\mathrm{Aut}(\mathrm{\Gamma }/\mathrm{\Gamma }_k);`$
$`f\text{lifts to an endomorphism of}\mathrm{\Gamma }\text{which fixes}\gamma \mathrm{mod}\mathrm{\Gamma }_{k+1}\}.`$
By making a crucial use of a theorem of Stallings , for each $`k`$ they obtain a homomorphism
$$\sigma _k:_{g,1}\mathrm{Aut}_0(\mathrm{\Gamma }/\mathrm{\Gamma }_k).$$
The following theorem given in is a basic result for the study of the structure of the group $`_{g,1}`$.
###### Theorem 39 (Garoufalidis-Levine).
The above homomorphism $`\sigma _k`$ is surjective for any $`k`$.
They use the homomorphisms $`\{\sigma _k\}_k`$ to define a certain filtration $`\{_{g,1}(k)\}_k`$ of $`_{g,1}`$ and show that the Johnson homomorphisms are defined also on this group. Furthermore they are surjective so that there is an isomorphism
$$\mathrm{Gr}^+(_{g,1})=\underset{k=1}{\overset{\mathrm{}}{}}_{g,1}(k)/_{g,1}(k+1)𝔥_{g,1}.$$
They concluded from this fact that $`_{g,1}`$ contains $`_{g,1}`$ as a subgroup. The homomorphisms $`\sigma _k`$ fit together to define a homomorphism
$$\sigma :_{g,1}\underset{}{\mathrm{lim}}\mathrm{Aut}_0(\mathrm{\Gamma }/\mathrm{\Gamma }_k).$$
They show that this homomorphism is not surjective by using the argument of Levine . We mention a recent paper of Sakasai for a related work. Also they point out that, although the restriction of $`\sigma `$ to $`_{g,1}`$ is injective, $`\sigma `$ has a rather big kernel because $`\mathrm{Ker}\sigma `$ at least contains the group
$$\mathrm{\Theta }_{}^3=\{\text{oriented homology }3\text{-sphere}\}/\text{homology cobordism}.$$
It is easy to see that $`\mathrm{\Theta }_{}^3`$ is contained in the center of $`_{g,1}`$ so that we have a central extension
(20)
$$0\mathrm{\Theta }_{}^3_{g,1}\overline{}_{g,1}1$$
where $`\overline{}_{g,1}=_{g,1}/\mathrm{\Theta }_{}^3`$.
The group $`\mathrm{\Theta }_{}^3`$ is a very important abelian group in low dimensional topology. In , Furuta first proved that this group is an infinitely generated group by making use of gauge theory. See also the paper by Fintushel and Stern for another proof. However, only a few additive invariants are known on this group at present besides the classical surjective homomorphism
(21)
$$\mu :\mathrm{\Theta }_{}^3/2$$
defined by the Rokhlin invariant. One is a non-trivial homomorphism $`\mathrm{\Theta }_{}^3`$ constructed by Fr$`\varphi `$yshov and the other is given by Ozsváth and Szabó as an application of their Heegaard Floer homology theory. Neumann and Siebenmann defined an invariant $`\overline{\mu }`$ for plumbed homology $`3`$-spheres and Saveliev introduced his $`\nu `$-invariant for any homology sphere by making use of the Floer homology. On the other hand, Fukumoto and Furuta (see ) defined certain invariants for plumbed homology $`3`$-spheres using gauge theory. According to a recent result of Saveliev , these invariants fit together to give a candidate of another homomorphism on $`\mathrm{\Theta }_{}^3`$.
The situation being like this, $`\mathrm{\Theta }_{}^3`$ remains to be a rather mysterious group. Thus we have the following important problem.
###### Problem 40.
Study the central extension (20) from the point of view of group cohomology as well as geometric topology. In particular determine the Euler class of this central extension which is an element of the group
$$H^2(\overline{}_{g,1};\mathrm{\Theta }_{}^3)\mathrm{Hom}(H_2(\overline{}_{g,1}),\mathrm{\Theta }_{}^3)\mathrm{Ext}_{}(H_1(\overline{}_{g,1}),\mathrm{\Theta }_{}^3).$$
This should be an extremely difficult problem. Here we would like to indicate a possible method of attacking it, and in particular a possible way of obtaining additive invariants for the group $`\mathrm{\Theta }_{}^3`$, very briefly. Details will be given in a forthcoming paper.
Using the traces, we can define a series of certain cohomology classes
$$\stackrel{~}{t}_{2k+1}H^2(\underset{}{\mathrm{lim}}\mathrm{Aut}_0(\mathrm{\Gamma }/\mathrm{\Gamma }_m))(k=1,2,\mathrm{}).$$
These are the “group version” of the elements $`t_{2k+1}H^2(𝔥_{g,1}^{})^{Sp}`$ defined in $`\mathrm{\S }6`$ (Definition 26). The homomorphism $`\sigma `$ is trivial on $`\mathrm{\Theta }_{}^3`$ so that we have the induced homomorphism $`\overline{\sigma }:\overline{}_{g,1}\underset{}{\mathrm{lim}}\mathrm{Aut}_0(\mathrm{\Gamma }/\mathrm{\Gamma }_m)`$.
###### Conjecture 41.
1. $`\overline{\sigma }^{}(\stackrel{~}{t}_{2k+1})`$ is non-trivial in $`H^2(\overline{}_{g,1})`$ for any $`k`$
$`\text{ 2. }\sigma ^{}(\stackrel{~}{t}_{2k+1})\text{ is trivial in }H^2(_{g,1})\text{ for any }k.`$
The first part of the above conjecture is the “group version” of Conjecture 27 and it should be even more difficult to prove. On the other hand, if the classes $`\sigma ^{}(\stackrel{~}{t}_{2k+1})`$ were non-trivial, then they would serve as invariants for certain $`4`$-manifolds ($`2`$-dimensional family of homology cylinders). This seems unlikely to be the case. Thus the second part is related to the following problem.
###### Problem 42.
Determine the abelianization of the group $`_{g,1}`$. Is it trivial? Also determine the second homology group $`H_2(_{g,1};)`$. Is the rank of it equal to $`1`$ given by the signature?
If everything will be as expected, we would obtain non-trivial homomorphisms
$$\widehat{t}_{2k+1}:\mathrm{\Theta }_{}^3$$
as secondary invariants associated to the cohomology classes $`\stackrel{~}{t}_{2k+1}`$. There should be both similarity and difference between these cases and the situation where we interpreted the Casson invariant as the secondary invariant associated to the first Mumford-Morita-Miller class $`e_1`$ (see ). More precisely, they are similar because they are all related to some cohomology classes in $`H^2(𝔥_{g,1}^{})`$. They are different because $`e_1`$ is non-trivial in $`H^2(_{g,1})`$ whereas we expect that the other classes $`\sigma ^{}(\stackrel{~}{t}_{2k+1})`$ would be all trivial in the same group.
Also recall here that Matumoto and Galewsky and Stern proved that every topological manifold (of dimension $`n7`$) is simplicially triangulable if and only if the homomorphism (21) splits. In view of this result, it should be important to investigate the mod $`2`$ structure of the extension (20) keeping in mind the works of Birman and Craggs as well as Johnson for the case of the mapping class group.
However we have come too far and surely many things have to be clarified before we would understand the structure of the group $`_{g,1}`$.
Finally we would like to propose a problem.
###### Problem 43.
Generalize the infinitesimal presentation of the Torelli Lie algebra given by Hain to the case of the group of homology cobordism classes of homology cylinders.
## 12. Diffeomorphism groups of surfaces
Let us recall the important problem of constructing (and then detecting) characteristic classes of smooth fiber bundles as well as those of foliated fiber bundles whose fibers are diffeomorphic to a general closed $`C^{\mathrm{}}`$ manifold $`M`$. This is equivalent to the problem of computing the cohomology groups $`H^{}(\mathrm{BDiff}M)`$ (resp. $`H^{}(\mathrm{BDiff}^\delta M)`$) of the classifying space of the diffeomorphim group $`\mathrm{Diff}M`$ (resp. the same group $`\mathrm{Diff}^\delta M`$ equipped with the discrete topology) of $`M`$. Although the theory of higher torsion invariants of fiber bundles developed by Igusa on the one hand and by Bismut, Lott and Goette on the other and also the Gel’fand-Fuks cohomology $`H_{GF}^{}(M)`$ of $`M`$ (see e.g. ) produce characteristic classes for the above two types of $`M`$-bundles, there seem to be only a few known results concerning explicit computations for specific manifolds. The same problems are also important for various subgroups of $`\mathrm{Diff}M`$, in particular the symplectomorphism group $`\mathrm{Symp}(M,\omega )`$ (in the case where there is given a symplectic form $`\omega `$ on $`M`$) as well as the volume preserving diffeomorphism group.
Here we would like to propose two problems for the special, but at the same time very important case of surfaces. Note that we have two characteristic classes
(22)
$$_{fiber}u_1c_1^2,_{fiber}u_1c_2H^3(\mathrm{BDiff}_+^\delta \mathrm{\Sigma }_g;)$$
which are defined to be the fiber integral of the characteristic classes $`u_1c_1^2,u_1c_2`$ of codimention two foliations. In the case of $`g=0`$ (namely $`S^2`$), Thurston and then Rasmussen (see also Boullay ) proved that these classes are linearly independent and vary continuously. Although it is highly likely that their results can be extended to the cases of surfaces of higher genera, explicit construction seems to be open.
###### Problem 44.
Prove that the above characteristic classes induce surjective homomorphism
$$H_3(\mathrm{BDiff}_+^\delta \mathrm{\Sigma }_g;)^2$$
for any $`g`$.
The cohomology classes in (22) are stable with respect to $`g`$. On the other hand, in we found an interesting interaction between the twisted cohomology group of the mapping class group and some well known concepts in symplectic topology such as the flux homomorphism as well as the Calabi homomorphism (see for generalities of the symplectic topology). By making use of this, we defined certain cohomology classes of $`\mathrm{BSymp}^\delta \mathrm{\Sigma }_g`$ and proved non-triviality of them.
In view of the fact that all the known cohomology classes of $`\mathrm{BDiff}_+^\delta \mathrm{\Sigma }_g`$ as well as $`\mathrm{BSymp}^\delta \mathrm{\Sigma }_g`$ are stable with respect to the genus, we would like to ask the following problem.
###### Problem 45.
Study whether the homology groups of $`\mathrm{BDiff}_+^\delta \mathrm{\Sigma }_g`$ stabilize with respect to $`g`$ or not. The same problem for the group $`\mathrm{Symp}^\delta \mathrm{\Sigma }_g`$.
Acknowledgments The author would like to express his hearty thanks to R. Hain, N. Kawazumi, D. Kotschick, M. Matsumoto, H. Nakamura, T. Sakasai for enlightening discussions as well as useful informations concerning the problems treated in this paper. |
warning/0507/physics0507027.html | ar5iv | text | # REMARKS ON PHOTONS AND THE AETHER
## 1. INTRODUCTION
In the book we discussed many aspects of e.g. electromagnetism (EM), the aether, and the Schrödinger equation (SE) partly in connection with our study of the quantum potential (QP). We now want to examine further the nature of photons and radiation in connection with a putative aether. It is essential that we review some of the background from in order to motivate the aether treatment in Section 6.
## 2. PHOTONS
We begin with (which is also sketched in in a somewhat different manner) and in Section 5 give a related description following . We will also examine further various points of view concerning the massless Klein-Gordon (KG) equation, the SE, the Maxwell equations (ME), and the quantum vacuum. For background we mention here . One takes massless photons as objects with energy E, momentum P, and internal angular momentum (or spin) S with $`E=c|𝐏|`$ and $`𝐒\times 𝐏=0`$. It is presumed to have velocity c in the direction k and to spin in a plane perpendicular to k, which is spanned by two vectors e and b where
(2.1)
$$𝐤𝐞=𝐤𝐛=0;𝐤\times 𝐞=𝐛;𝐤\times 𝐛=𝐞;𝐞|=|𝐛|;𝐞𝐛=0$$
One sets $`\omega =e=b`$ (frequency) and $`E=\mathrm{}\omega `$ historically (with $`|𝐒|=\pm \mathrm{}`$) while $`\lambda =2\pi c/\omega `$ (which will eventually be identified with a wave length). The photon is considered as following a right of left handed helix generated by the tip of e where the plane of $`𝐞,𝐛`$ moves along the direction k with velocity c. These objects are exhibited via a photon tensor
(2.2)
$$f^{\mu \nu }=\left(\begin{array}{cccc}0& e_1& e_2& e_3\\ e_1& 0& b_3& b_2\\ e_2& b_3& 0& b_1\\ e_3& b_2& b_1& 0\end{array}\right)$$
which is $`\underset{¯}{not}`$ a field like the EM tensor $`F^{\mu \nu }`$ (see Section 5 for more comments on the tensor nature of $`f^{\mu \nu }`$). The dual tensor is
(2.3)
$$f^{\mu \nu }=\frac{1}{2}ϵ^{\mu \nu \sigma \rho }f_{\sigma \rho }=\left(\begin{array}{cccc}0& b_1& b_2& b_3\\ b_1& 0& e_3& e_2\\ b_2& e_3& 0& e_1\\ b_3& e_2& e_1& 0\end{array}\right)$$
with $`f^{\mu \nu }f_{\mu \nu }=2(e^2b^2)=0`$ and $`f^{\mu \nu }f_{\mu \nu }^{}=4𝐞𝐛=0`$. One works here in a Hilbert space $`H=H^SH^K`$ with $`𝐒𝐒1,𝐏1𝐏`$, and $`𝐑1𝐑`$. Now spin is colinear with momentum (recall $`𝐒\times 𝐏=0`$) and the spin eigenstates $`\chi _\pm `$ correspond to helicities $`\pm 1`$ satisfying
(2.4)
$$(𝐤𝐒)\chi _\pm =\pm \mathrm{}\chi _\pm $$
where k is a unit vector in the direction of P. The spin operators will be expressed via
(2.5)
$$S_x=\mathrm{}\left(\begin{array}{ccc}0& 0& 0\\ 0& 0& i\\ 0& i& 0\end{array}\right);S_y=\mathrm{}\left(\begin{array}{ccc}0& 0& i\\ 0& 0& 0\\ i& 0& 0\end{array}\right);S_z=\mathrm{}\left(\begin{array}{ccc}0& i& 0\\ i& 0& 0\\ 0& 0& 0\end{array}\right)$$
with $`(S_j)_k\mathrm{}=i\mathrm{}ϵ_{jk\mathrm{}}`$. One must distinguish here $`𝐤H^K`$ and $`𝐒H^S`$; the 2-dimensional spin space is orthogonal to k with $`𝐤𝐒\pm \mathrm{}`$ as indicated in (2.4). Now write $`\psi _jH^SH^K`$ with $`j=1,2,3`$ denoting components in $`H^S`$ and set $`𝐤=𝐩/|𝐩|`$. An operator leaving invariant a photon state $`\chi _\pm ^k\varphi _p`$ is $`𝐒𝐏=𝐤𝐒|𝐏|`$ where $`|𝐏|\varphi _p=|𝐩|\varphi _p`$ (with $`E=c|𝐩|=c|p|`$) and one has
(2.6)
$$𝐒𝐏\chi _\pm ^k\varphi _p=\pm \frac{\mathrm{}E}{c}\chi _\pm ^k\varphi _p$$
(the Hamiltonian is $`H=(c/\mathrm{})𝐒𝐏`$ and a minus sign should be interpreted as positive energy but negative helicity). Then the time evolution of a general photon state is
(2.7)
$$i\mathrm{}_t\psi _j=(H)_{jk}\psi _k=\frac{c}{\mathrm{}}(𝐒𝐏)_{jk}\psi _k$$
(H is a $`3\times 3`$ matrix in $`H^S`$ whose components are operators in $`H^K`$). Putting $`𝐏=i\mathrm{}`$ one has a SE for the photon, namely
(2.8)
$$\frac{i}{c}_t\psi _j(t,𝐫)=ϵ_{jk\mathrm{}}_{\mathrm{}}\psi _k(t,𝐫)$$
(since $`(c/\mathrm{})(i\mathrm{}ϵ_{jk\mathrm{}})(i\mathrm{}_{\mathrm{}})=c\mathrm{}ϵ_{jk\mathrm{}}_{\mathrm{}}`$). Note $`\mathrm{}`$ has disappeared and although this is a QM equation it does not have a classical limit.
REMARK 2.1. It is pointed out in that there are conceptual errors in writing $`\psi _j=E_j+iB_j`$ and deriving the Maxwell equations via $`(i/c)_t(E_j+iB_j)=ϵ_{jk\mathrm{}}_{\mathrm{}}(E_k+iB_k)`$ in the form
(2.9)
$$\frac{1}{c}_tE_j=ϵ_{jk\mathrm{}}_{\mathrm{}}B_k;\frac{1}{c}_tB_j=ϵ_{jk\mathrm{}}_{\mathrm{}}E_k$$
(e.g. $`(1/c)_tE_1=ϵ_{123}_3B_2ϵ_{132}_2B_3=_2B_3_3B_2`$, etc. - note $`ϵ_{jk\mathrm{}}=ϵ_{j\mathrm{}k}`$ in ). The equations are correct but the derivation is faulty since it identifies the 3-D space of states with the 3-D physical space!$`\mathrm{}`$
REMARK 2.2. Using the momentum representation one can write as in
(2.10)
$$\frac{\mathrm{}}{c}_t\psi _j(t,𝐩)=ϵ_{jk\mathrm{}}p_{\mathrm{}}\psi _k(t,𝐩)$$
but this is not $`\stackrel{}{\psi }\times \stackrel{}{p}`$ because the two vectors belong to different spaces. One can look also at stationary state solutions $`\psi _j=exp[(i/\mathrm{})Et)\mathrm{\Phi }_{j,E}`$ where $`(𝐤=𝐩/|𝐩|`$)
(2.11)
$$(𝐒𝐏)_{jk}\mathrm{\Phi }_{k,E}=\frac{E\mathrm{}}{c}\mathrm{\Phi }_{j,E};\mathrm{\Phi }_{j,E}(𝐩)=\chi _\pm ^k\delta \left(|𝐩|\frac{E}{c}\right)$$
For the corresponding position representation one would use $`\chi _\pm ^{k_0}\varphi _{p_0}`$ where
(2.12)
$$\varphi _{p_0}(𝐫)=\frac{1}{\sqrt{2\pi \mathrm{}}^3}exp\left(\frac{i}{\mathrm{}}𝐩_0𝐫\right)$$
(where $`𝐤_0=𝐩_0/|𝐩_0|`$.$`\mathrm{}`$
## 3. THE EM FIELDS
In the last paper of it is shown how to construct the EM fields from knowledge of photons. First one defines
(3.1)
$$ϵ_+=\frac{1}{\sqrt{2}}(\widehat{𝐞}+i\widehat{𝐛});ϵ_{}=\frac{1}{\sqrt{2}}(i\widehat{𝐞}+\widehat{𝐛})$$
where $`𝐞=\omega \widehat{𝐞}`$ and $`𝐛=\omega \widehat{𝐛}`$. Then write
(3.2)
$$𝐞_+(t)=(\frac{\omega }{\sqrt{2}}ϵ_+e^{i\omega t}+c.c.);𝐞_{}(t)=(\frac{\omega }{\sqrt{2}}ϵ_{}e^{i\omega t}+c.c.)$$
One writes $`𝐞_s(t)=[(\omega /\sqrt{2})ϵ_sexp(i\omega t)+c.c.)`$ and $`𝐛_s(t)=𝐤\times 𝐞_s(t)`$, uses momentum eigenfunctions as in Remark 2.2, and writes $`\varphi _{s,p}=\chi _s^k\varphi _p`$. For a state with n photons having helicity $`s_j`$ and momentum $`p_j`$ annihilation and creation operators are defined via
(3.3)
$$a_s^{}(p)\varphi _{s_1p_1,\mathrm{},s_np_n}=\sqrt{n+1}\varphi _{sp,s_1p_1,\mathrm{},s_np_n};$$
$$a_s(p)\varphi _{s_1p_1,\mathrm{},s_np_n}=\frac{1}{\sqrt{n}}\underset{1}{\overset{n}{}}\delta _{s,s_i}\delta (𝐩𝐩_i)\varphi _{s_1p_1,\mathrm{},\widehat{s_ip_i},\mathrm{},s_np_n}$$
A vacuum state $`\varphi _0`$ with zero photons is defined via $`a_s(p)\varphi _0=0`$ and n-photon states are built up via
(3.4)
$$\varphi _{s_1p_1,\mathrm{},s_np_n}=\frac{1}{\sqrt{n!}}a_{s_1}^{}(p_1)\mathrm{}a_{s_n}^{}(p_n)\varphi _0$$
A number operator is defined via $`N_s(p)=a_s^{}(p)a_s(p)`$ and $`N=_sd^3pN_s(p)`$. The total energy, momentum, and spin of a system of photons (each with energy $`E=c|p|=\mathrm{}\omega `$ and spin $`\pm \mathrm{}`$) is then
(3.5)
$$H=\underset{s}{}d^3p\mathrm{}\omega N_s(p);𝐏=\underset{s}{}d^3p𝐩N_s(p);$$
$$𝐒=d^3p\mathrm{}𝐤(N_+(p)N_{}(p))$$
One defines then Hermitian operators
(3.6)
$$𝐄(𝐫,t)=\frac{1}{2\pi \mathrm{}}\underset{s}{}d^3p\sqrt{\omega }(ia_s(p)ϵ_se^{(i/\mathrm{})(𝐩𝐫Et)}+h.c.);$$
$$𝐁(𝐫,t)=\frac{1}{2\pi \mathrm{}}\underset{s}{}d^3p\sqrt{\omega }(ia_s(p)(𝐤\times ϵ_s)e^{(i/\mathrm{})(𝐩𝐫Et)}+h.c.);$$
$$𝐀(𝐫,t)=\frac{c}{2\pi \mathrm{}}\underset{s}{}d^3p\frac{1}{\sqrt{\omega }}(a_s(p)ϵ_se^{(i/\mathrm{})(𝐩𝐫Et)}+h.c.)$$
Then
(3.7)
$$𝐄=\frac{1}{c}_t𝐀;𝐁=\times 𝐀;H=\frac{1}{8\pi }d^3r(𝐄^2+𝐁^2);$$
$$𝐏=\frac{1}{8\pi c}d^3r(𝐄\times 𝐁𝐁\times 𝐄;𝐒=\frac{1}{8\pi c}d^3r(𝐄\times 𝐀𝐀\times 𝐄)$$
and one checks the Maxwell equations
(3.8)
$$\times 𝐄=\frac{1}{c}_t𝐁;\times 𝐁=\frac{1}{c}_t𝐄;𝐄=𝐁=0$$
Thus photons are posited as the fundamental objects and they generate EM fields as a collective manifestation.
Next one defines the “singular” function (cf. for details)
(3.9)
$$D(\stackrel{}{\rho },\tau )=\frac{1}{(2\pi \mathrm{})^3}d^3pe^{(i/\mathrm{})𝐩\stackrel{}{\rho }}\frac{Sin(\omega \tau )}{\omega }=$$
$$=\frac{1}{8\pi ^2c\rho }[\delta (\rho c\tau )\delta (\rho +c\tau )]$$
Here $`\rho =|\stackrel{}{\rho }|`$ where $`\stackrel{}{\rho }𝐫_1𝐫_2`$ and one can say that $`D(\stackrel{}{\rho },\tau )`$ has support on the light cone (cf. also ). This leads to
(3.10)
$$[E_i(𝐫_1,t_1),E_j(𝐫_2,t_2)]=4\pi i\mathrm{}c^2\left(\frac{\delta _{ij}}{c^2}_{t_1}_{t_2}+_{r_1,i}_{r_2,j}\right)D(𝐫_1𝐫_2,t_1t_2)$$
$$[B_i(𝐫_1,t_1),B_j(|bfr_2,t_2)]=4\pi i\mathrm{}c^2(\frac{\delta _{ij}}{c^2}_{t_1}_{t_2}+_{r_1,i}_{r_2,j})D(𝐫_1𝐫_2,t_1t_2)$$
$$[E_i(𝐫_1,t_1),B_j(𝐫_2,t_2)]=4\pi i\mathrm{}cϵ_{ijk}_{t_1}_{r_1,k}D(𝐫_1𝐫_2,t_2t_1)$$
Note that the singular nature of D is really unacceptable in QM (e.g. because of the uncertainty principle) and one could conclude that the field strengths are not measurable quantities (cf. the first paper in ). On the other hand field averages can be accepted in QM. This is one feature leading to the approch in based on the photon as fundamental. The EM fields are considered essentially as a classical macroscopic ideas and are not “basic”. Such an argument might be extendable quite generally to cast suspicion on many results involving singular behavior or generalized solutions of partial differential equations (distributions). The “classical” theory might require e.g. averaging of dependent variables or some new physics (not necessarily QM) in order to retain any meaning.
One looks next at the expectation values of fields in the quantum state describing a system of photons. For the vacuum described via $`\varphi _0`$ one has
(3.11)
$$<\varphi _0,E(𝐫,t)\varphi _0>=<\varphi _0,𝐁(𝐫,t)\varphi _0>=0$$
as expected. However one can show that e.g.
(3.12)
$$<\varphi _0,𝐄^2(𝐫,t)\varphi _0>=\frac{2}{(2\pi \mathrm{})^2}d^3p\omega $$
indicating that there are fluctuations of the electric field in vacuum. For a quantum state of n photons in the same state with fixed helicity and momentum one has (cf. )
(3.13)
$$\varphi =\varphi _{n(s_1p_1)}=\frac{1}{\sqrt{n!}}(a_{s_1}^{}(𝐩_1))^n\varphi _0;<\varphi ,𝐄(𝐫,t)\varphi >=<\varphi ,𝐁(𝐫,t)\varphi >=0$$
which is somewhat strange. However for an indefinite number of photons in a superposition of states $`\psi =_nC_n\varphi _{n(s_1p_1)}`$ one has
(3.14)
$$<\psi ,𝐄(𝐫,t)\psi >=\frac{\sqrt{\omega _1}}{2\pi \mathrm{}}(i\underset{n}{}C_n^{}C_{n+1}ϵ_{s_1}e^{(i/\mathrm{})(𝐩_1𝐫E_1t)}+c.c.)$$
One concludes here that the EM field of an indefinite number of photons all with the same helicity and momentum is a plane wave with circular polarization. The quantum state where all photons are in the same one photon state of fixed helicity and momentum is a Bose-Einstein condensate (?).
REMARK 2.1. We extract here from for a few philosophical observations. The photon, as an elementary “particle” is unique; it is the only elementary particle of energy. A relativistic energy equation should be $`E^2=p^2c^2+m_0c^2=p^2c^2`$ since the rest mass $`m_0=0`$. In the frame of the moving photon the photon’s energy is stored as rotational (spin) energy where $`E=\mathrm{}\nu =\mathrm{}c/\lambda `$ with $`\nu `$ the frequency and $`\lambda `$ the wave length. Hence the greater the energy the smaller the wave length and one expects to find a lower bound for the wavelength. For a “particle” the angular momentum is $`L=mrw`$ limited by $`L=mrc`$ and replacing L by the spin S one has $`(\mathrm{})\mathrm{}=mrc`$ where m is a putative mass presumably “generated” by the spin (see here also for toy models with extended energy distributions). Assume the concept of Schwartzschild radius R is valid for the photon where for a black hole $`R=2Gm/c^2`$ or $`(\mathrm{})(R/m)=(2G/c^2)`$. The right side of $`(\mathrm{})`$ is a constant but for the photon the radius decreases as the “mass” increases; hence there is a unique value of radius and mass for which a photon can behave as a black hole. Combining $`(\mathrm{})`$ with $`(\mathrm{})`$ one finds $`(\mathrm{})m=\sqrt{\mathrm{}c/2G}`$ for the Planck mass, which here is the maximum “pseudomass” permitted for the photon. This corresponds to a maximum energy of $`mc^2=(\sqrt{\mathrm{}c/2G})c^2=8.61\times 10^{22}`$ MeV and the highest energy so far observed for a photon is apparently less than this. It is suggested that pair production or photon “splitting” will ensue at the energy limit.$`\mathrm{}`$
## 4. THE ZERO POINT FIELD - ZPF
This is a murky subject and essentially involves understanding the quantum vacuum, which of course still retains some mysteries. We gave some hesitant and heuristic comments on ZPF in , based on , which upon hindsight seem woefully inadequate. Some of this is also summarized and enhanced in a recent paper (first paper). However we go here to the lovely collection of papers by J. Field (see e.g. ) for an aperçu of basic physical connections between QM, thermodynamics, and special relativity. This will serve as a complement to Sections 2-3. We begin with (first paper) which in a sense follows the spirit of Feynman’s QED where the fundamental concepts of QM are explained in terms of the interactions of photons and electrons. One recalls first the energy momentum vector $`P=m(dX/d\tau )((E/c),p_x,p_y,p_z)`$ with $`X=(ct,x,y,z)`$ and $`\tau `$ the proper time (time observed in the rest frame). If the inertial frame S’ is moving with uniform velocity $`\beta c`$ relative to the frame S along the common $`x,x^{}`$ axis with $`0y`$ parallel to $`0y^{}`$ then the 4-vectors as observed in S, S’ are related by Lorentz transform (LT) equations
(4.1)
$$p_x^{}=\gamma (p_x\beta p_t);p_y^{}=p_y;p_z^{}=p_z;p_t^{}=\gamma (p_t\beta p_x);$$
$$\gamma =\frac{1}{\sqrt{1\beta ^2}};p_t=\frac{E}{c}$$
As $`m0`$ P is still well defined so one has an energy momentum vector say $`P_\gamma =[(E_\gamma /c),(E_\gamma /c)Cos(\varphi ),(E_\gamma /c)Sin(\varphi ),0]`$ for a photon of energy $`E_\gamma `$ moving in the $`(x,y)`$ plane in a direction making an angle $`\varphi `$ with the $`x`$ axis. A plane EM wave will be associated with a large number of photons in general and for such a collection, all with the same 4-vector $`P_\gamma `$, one finds from the LT equations an EM wave with
(4.2)
$$\nu ^{}=\nu \gamma (1\beta Cos(\varphi ));E_T^{}=E_T\gamma (1\beta Cos(\varphi ))$$
(here $`\nu `$ frequency and $`E_T`$ total energy). Using (4.1) the energies of the photons in the EM wave transform via $`E_\gamma ^{}=E_\gamma \gamma (1\beta Cos(\varphi ))`$. If $`n_\gamma `$ is the total number of photons then $`E_T=n_\gamma E_\gamma `$ which yields (4.2). Further one sees immediately that $`E_\gamma /\nu =E_\gamma ^{}/\nu ^{}=constant`$ and calling this constant $`\mathrm{}`$ one finds that $`E_\gamma =\mathrm{}\nu `$ which identifies $`\mathrm{}`$ with Planck’s constant. Consequently Planck’s constant arises from consistency between the relativistic kinematics of photons, considered to be massless particles, and the relativistic Doppler effect for classical EM waves. Note also that using $`\lambda =c/\nu `$ and $`E_\gamma =\mathrm{}\nu =p_\gamma c`$ one arrives at the deBroglie relation $`p_\gamma =\mathrm{}/\lambda =\mathrm{}\nu /c`$.
Now from the energy density of a plane EM wave, namely
(4.3)
$$\rho _W=\frac{𝐄^2+𝐁^2}{8\pi }$$
the photon interpretation gives immediately Poynting’s formula for the energy flow F per unit area per unit time, namely $`F=c\rho _W`$ as well as the formula for the radiation pressure $`P_{rad}`$ of a plane wave at normal incidence on a perfect reflector, namely $`P_{rad}=2\rho _W`$. Note the number of photons incident is $`F/E_\gamma `$ per unit area per unit time so the total momentum transferred is then $`p_\gamma (F/E_\gamma )=F/c=\rho _W`$; but a perfect reflector will not absorb energy so an equal number of photons are re-emitted, yielding the factor of 2. Now consider a plane EM wave of wavelength $`\lambda `$ moving in free space parallel to the positive $`x`$ direction in the frame S, written as
(4.4)
$$E_y=E_0e^\mathrm{\Phi };H_z=H_0e^\mathrm{\Phi };\mathrm{\Phi }=2\pi i\frac{(xct)}{\lambda };E_0=H_0=A$$
The time averaged energy density per unit volume $`\overline{\rho }_W`$ is $`\overline{\rho }_W=(E_0^2/8\pi )=(H_0^2/8\pi )=(A^2/8\pi )`$. Assuming that the wave consists of a beam of photons of energy $`\mathrm{}\nu `$ the average number density of photons $`\overline{\rho }_\gamma `$ in the wave is $`\overline{\rho }_W/\mathrm{}\nu `$ so one gets $`\overline{\rho }_\gamma =A^2/8\pi \mathrm{}\nu `$. This is the point where one now leaps across the chasm separating the classical and quantum worlds. First the use of a complex exponential to represent a classical EM wave is convenient but it is really the real or imaginary parts that come into play (Cosines and Sines); for QM the complex exponential is mandatory. Second one uses the definition of wavelength together with $`E_\gamma =\mathrm{}\nu `$ and $`p_\gamma =\mathrm{}/\lambda `$ to replace in the complex exponential the wave parameter $`\lambda `$ by the particle parameters $`E_\gamma `$ and $`p_\gamma `$. The parameter c is part of both descriptions (photons and EM waves) and this leads to the complex exponential describing photons in the form
(4.5)
$$u_p=u_0exp\left[\frac{2\pi i}{\mathrm{}}(p_\gamma xE_\gamma t)\right]=exp\left[\frac{2\pi i}{\mathrm{}}(PX)\right];u_0=\frac{A}{\sqrt{8\pi E_\gamma }}$$
In this situation $`\overline{\rho }_\gamma =|u_p|^2=u_0^2`$ and for the case of very weak EM fields such that $`\overline{\rho }_\gamma <<1`$ it follows that $`|u_p|^2dV`$ can be thought of as the probability that a photon is in the volume $`dV`$; for large numbers of photons or strong EM fields this probabilistic interpretation is not appropriate. Note also that one can write
(4.6)
$$𝒫_x=i\frac{\mathrm{}}{2\pi }\frac{}{x};=i\frac{\mathrm{}}{2\pi }\frac{}{t};𝒫_xu_p=p_\gamma u_p;u_p=E_\gamma u_p$$
Note also for $`f`$ an arbitrary function on space-time
(4.7)
$$𝒫_x(xf)=\frac{i\mathrm{}}{2\pi }\frac{(xf)}{x}=\frac{i\mathrm{}}{2\pi }f+x𝒫_xf$$
Repeated use of (4.6), (4.7), etc. and the relation $`E_\gamma =p_\gamma c`$ gives
(4.8)
$$(c^2𝒫_x^2^2)u_p=\left(\frac{\mathrm{}c}{2\pi }\right)\mathrm{}u_p=0;\mathrm{}=\frac{1}{c^2}\frac{^2}{t^2}\frac{^2}{x^2}$$
so $`u_p`$ will satisfy the Maxwell-Lorentz equation $`\mathrm{}u_P=0`$. If one uses instead of $`E_\gamma =p_\gamma c`$ the general energy momentum relation for massive particles $`E^2=p^2c^2+m^2c^4`$ one can arrive at the Klein-Gordon (KG) equation and expanding $`Emc^2+(p^2/2m)+\mathrm{}`$ the Schrödinger equation (SE) will result.
Consider next ($`p_\gamma \mathrm{}/\lambda `$ and $`E_\gamma \mathrm{}c/\lambda `$)
(4.9)
$$\chi =\frac{1}{2}(u_p+u_p^{})=\mathrm{}(u_p)=u_0Cos\left[\frac{2\pi }{\mathrm{}}(p_\gamma xE_\gamma t)\right]u_0Cos\left(\frac{2\pi (xct)}{\lambda }\right)$$
This equation is then a bridge back across the chasm from QM to the classical world (cf. (4.5)). Just as the quantum wave function is only meaningful in the limit of very low photon density so the function $`\chi `$ is meaningful only in the limit of high photon density. $`\chi `$ is not an eigenfunction of either $`E_\gamma `$ or $`p_\gamma `$ and is a real function. The time average of $`\chi ^2`$ is $`1/2`$ the mean photon density $`\overline{\rho }_\gamma `$ and $`\overline{\rho }_W=\mathrm{}\nu \overline{\rho }_\gamma `$. In a typical situation $`\overline{\rho }_\gamma \mathrm{\Delta }V`$ is much larger than 1 and no probabilistic meaning can be attached to it.
We show next following how to derive the Maxwell equations using only Coulomb’s inverse square law, special relativity, and Hamilton’s principle. Thus take two objects $`O_i`$ of masses $`m_i`$ and electric charges $`q_i`$ with no external forces. The spatial distance separating them in the common center of mass frame is $`𝐱_{12}=𝐱_1𝐱_2`$. One constructs a most general Lorentz invariant Lagrangian in a nonrelativistic reference frame via ($`x_i𝐱_i`$)
(4.10)
$$L(x_1,u_1,x_2,u_2)=\frac{m_1u_1^2}{2}\frac{m_2u_2^2}{2}\frac{j_1j_2}{c^2\sqrt{(x_1x_2)^2}}$$
where the $`j_i=q_1u_i`$ are current 4-vectors; this is then put into the machinery of Hamilton’s principle so that
(4.11)
$$\frac{d}{d\tau }\left(\frac{L}{u_i^\mu }\right)\frac{L}{x_i^\mu }=0;(i=1,2;\mu =1,2,3,4)$$
Since the Lagrangian is a Lorentz scalar this provides a description of the motion of the $`O_i`$ in any inertial reference frame. Note that if one introduces a 4-vector potential $`𝐀_2=𝐣_2/cr_{12},r_{12}=|𝐫_{12}|`$ the standard Lorentz invariant Lagrangian, describing the motion of $`O_1`$ in the EM field created by $`O_2`$, namely $`L(x_1,u_1)=(m_1u_1^2/2)(1/c)q_1𝐮_1𝐀_2`$, is recovered (and similarly for motion of $`O_2`$ in the field of $`O_1`$). Now write $`_i=^1(/x^i)_i`$ and set $`𝐩=m𝐮`$ along with
(4.12)
$$E^i=^iA^0\frac{1}{c}\frac{A^i}{t}=^iA^0^0A^i;B^k=ϵ_{ijk}(^iA^j^jA^i)=(\times 𝐀)^k$$
Some calculation (cf. ) gives then the 3-D Lorentz force equation and a relativistic Biot-Savart Law in the form
(4.13)
$$\frac{d𝐩}{dt}=q\left[𝐄+\frac{𝐯}{c}\times 𝐁\right];𝐁=\frac{q_2\gamma _2(𝐯_2\times 𝐫)}{cr^3}=\frac{𝐣\times 𝐫}{cr^3};$$
$$𝐄=\frac{j_2^0𝐫}{cr^3}\frac{1}{c^2r}\frac{d𝐣_2}{dt}\frac{𝐣_2}{c^2}\frac{(𝐫𝐯_2)}{r^3}$$
where $`𝐫=𝐫_{12}`$. The Maxwell equations can be derived immediately from (4.12) along with the Faraday-Lenz law, Ampere’s law, etc. (cf. for details).
Now concerning the ZPF we collect some background information as follows.
1. It seems well established that there is a unique Lorentz invariant spectral energy density in the EM vacuum of the form $`\rho (\omega )=\rho _0(\omega )=\mathrm{}\omega ^2/2\pi ^2c^3`$ (cf. ). An observer moving with constant velocity in the EM vacuum perceives no force.
2. Following an object undergoing uniform constant acceleration $`a`$ in the vacuum perceives himself to be immersed in a thermal bath at temperature $`T=\mathrm{}a/2\pi kc`$ ($`k`$ Boltzman constant).
3. One recalls also that there is a zero point energy $`(1/2)\mathrm{}\omega `$ attached to a quantum harmonic oscillator. Also since there are $`(\omega ^2/2\pi ^2c^3)d\omega `$ field nodes per unit volume in the frequency interval $`[\omega ,\omega +d\omega ]`$ one obtains the spectral density $`\rho _0(\omega )=\mathrm{}\omega ^4/2\pi ^2c^3`$ of Item 1 (cf. ).
4. In one derives the Planck radiation law for the blackbody spectrum without the formalism of quantum theory. It is assumed only that (i) There is classical, homogeneous, and fluctuating EM EM radiation at absolute zero with Lorentz invariant spectrum. (ii) Classical EM theory holds for a dipole oscillator. (iii) A free particle in equilibrium with blackbody radiation has classical kinetic energy $`(1/2)kT`$ per degree of freedom. This leads then to the zero point energy density shown above and to Planck’s formula
(4.14)
$$\rho (\omega ,T)=\frac{\omega ^2}{\pi ^2c^3}\left[\frac{\mathrm{}\omega }{exp[(\mathrm{}\omega /kT)1]}+\frac{1}{2}\mathrm{}\omega \right]$$
If the zero point energy is ignored one obtains the Rayleigh-Jeans formula
(4.15)
$$\rho (\omega ,T)=\left(\frac{\omega ^2}{\pi ^2c^3}\right)kT$$
Here (the quantum number) $`\mathrm{}`$ arises in (4.14) as a linear factor in calculating the Lorentz invariant spectral density and can later be identified with Planck’s constant (so the derivation is classical).
5. Going again to one finds a lovely discussion involving entropy and energy fluctuations following and modifying Einstein’s arguments. Thus one considers a cavity containing thermal radiation separated into large and small volumes V and $`𝒱`$. The energy $`𝒰`$ of EM radiantion in $`𝒱`$ between frequencies $`\omega `$ and $`\omega +d\omega `$ undergoes spontaneous fluctuations creating a change in the corresponding entropy. Let $`\mathrm{\Sigma }`$ (resp. $`𝔖`$) be the entropy contributed between $`\omega `$ and $`\omega +d\omega `$ for V (resp. ($`𝒱`$). Then for $`ϵ`$ the entropy fluctuation in $`𝒱`$
(4.16)
$$S(ϵ)=\mathrm{\Sigma }+𝔖=\mathrm{\Sigma }_0+𝔖_0+(_ϵ\mathrm{\Sigma }+_ϵ𝔖)ϵ+\frac{1}{2}\left(\frac{^2\mathrm{\Sigma }}{ϵ^2}+\frac{^2𝔖}{ϵ^2}\right)ϵ^2+\mathrm{}$$
where $`\mathrm{\Sigma }_0,𝔖_0`$ signify equilibrium entropies where the fluctuation is zero. The first derivatives vanish at $`ϵ=0`$ and if $`V>>𝒱`$ one finds $`S(ϵ)\mathrm{\Sigma }_0+𝔖_0+(1/2)(^2𝔖/𝒰^2)ϵ^2`$. Now there is probabilistic entropy $`(\mathrm{})S_{prob}=(S_{prob})_0+klog(W)`$ (or $`W=cexp(S_{prob}/k)`$) where W is the number of microstates giving the same macrostate. There is also caloric entropy $`S_{cal}`$ where $`dS_{cal}=dQ/T`$ for reversible processes. Then write
(4.17)
$$dW=cexp\left[\frac{S_{prob}}{k}\right]dϵ=\widehat{c}exp\left[\frac{1}{2k}\frac{^2S_{prob}}{𝒰^2}ϵ^2\right]dϵ$$
Some classical argument (cf. , paper 2) involving $`<ϵ^2>=ϵ^2𝑑W(\pi ^2c^3/\omega ^2)\rho ^2d\omega ,𝒰=\rho d\omega ,`$ and $`^2𝔖_{prob}/𝒰^2=k/<ϵ^2>`$ leads then to $`(\mathrm{})^2S_{prob}/E^2=(k/E^2)`$ for average oscillator energy E. Note in fact directly from the definition $`(\mathrm{})`$ one has $`S_{prob}/E=k/E`$ leading to $`(\mathrm{})`$. Now Einstein assumed that $`S_{prob}=S_{cal}`$ in $`(\mathrm{})`$ and produced $`E=kT`$ along with the Planck formula (cf. (4.14)) $`E=\mathrm{}\omega /[exp(\mathrm{}\omega /kT)1]`$ (with the zero point term missing). Note here (using (4.14)) that the average energy of an oscillator is
(4.18)
$$<ϵ>=\frac{\pi ^2c^3}{\omega ^2}\rho (\omega ,T)=\frac{\mathrm{}\omega }{exp(\mathrm{}\omega /kT)1}+\frac{1}{2}\mathrm{}\omega =E$$
Now Boyer modifies Einstein’s argument in a way which recovers the zero point term (and (4.18)). Indeed he writes $`<ϵ^2>=<ϵ^2>_{ZPF}+<ϵ^2>_{cal}`$ and finds that
(4.19)
$$\frac{^2S_{cal}}{E^2}=\frac{k}{E^2(\mathrm{}\omega /2)^2}$$
leading to (4.18).
## 5. MORE ON PHOTONS
We go here to for some interesting developments concerning the localization of photons and their structure. One knows of course that the methods of quantum field theory (QFT) work for a description of photon activity but we want to examine more direct connections to EM fields, Maxwell’s equations, and wave-particle duality. First from one argues that a photon wave function can be introduced if one is willing to redefine in a physically meaningful manner what one wishes to mean by such a wave function. First one introduces a naive single photon wave function. Then one produces a second quantized many photon theory approached via many particle physics (which will correspond to the quantization of the free radiation field) and then recovers the naive single photon wave function by looking at the manifold of one photon states. There are apparently connections to the work of to which we don’t have access at the moment.
Now photons can be of positive or negative helicity and being massless one has $`E=cp`$ where $`p=|𝐩|`$. If one introduces probability amplitudes for photons of momentum p and helicity $`\pm `$, namely $`\gamma _\pm (𝐩,t)`$ which would be expected to satisfy a Schrödinger type equation $`(\mathrm{})i\mathrm{}_t\gamma _\pm (𝐩,t)=cp\gamma _\pm (𝐩,t)`$. Next for each p introduce two unit vectors $`\widehat{𝐞}_i(\widehat{𝐩})`$ where $`\widehat{𝐩}=𝐩/|𝐩|`$ such that $`\widehat{𝐞}_1,\widehat{𝐩}_2,\widehat{𝐩}`$ form a right handed triad (cf. here Section 1). Then define helicity vectors $`𝐞_\pm (\widehat{𝐩})=(1/\sqrt{2})[\widehat{𝐞}_1\pm i\widehat{𝐞}_2]`$ and write $`\stackrel{}{\gamma }_\pm =𝐞_\pm \gamma _\pm `$ with
(5.1)
$$\stackrel{}{\gamma }_+^{}\stackrel{}{\gamma }_+d𝐩=\gamma _+^{}\gamma _+d𝐩$$
for the probability of detecting a photon of positive helicity and momentum p between p and $`𝐩+d𝐩`$ (similarly for negative helicity). Note also that $`\stackrel{}{\gamma }_+^{}\stackrel{}{\gamma }_{}=0`$. Then define Fourier transforms
(5.2)
$$\mathrm{\Phi }_\pm (𝐫,t)=\frac{d𝐩}{(2\pi \mathrm{})^{3/2}}\gamma _\pm (𝐩,t)e^{i𝐩𝐫/\mathrm{}}$$
One then checks that $`(\mathrm{})`$ is satisfied if
(5.3)
$$i\mathrm{}_t\mathrm{\Phi }_\pm (𝐫,t)=\pm c\mathrm{}\times \mathrm{\Phi }_\pm (𝐫,t)$$
From the assumption that one is dealing with a single photon there results
(5.4)
$$[\stackrel{}{\gamma }_+^{}\stackrel{}{\gamma }_++\stackrel{}{\gamma }_{}^{}\stackrel{}{\gamma }_{}]𝑑𝐩=1[\mathrm{\Phi }_+^{}\mathrm{\Phi }_++\mathrm{\Phi }_{}^{}\mathrm{\Phi }_{}]𝑑𝐫=1$$
The dynamical equations $`(\mathrm{})`$ or (5.3) guarantee that if these equations (5.4) are true at one time then they are satisfied at all later times. In fact there results
(5.5)
$$\mathrm{\Phi }_\pm (𝐫,t)=\frac{d𝐩}{(2\pi \mathrm{})^{3/2}}\stackrel{}{\gamma }_\pm (𝐩,0)e^{icpt/\mathrm{}}e^{i𝐩𝐫/\mathrm{}}$$
There is then a temptation to try and identify the $`\mathrm{\Phi }_\pm `$ as position representation probability amplitudes for photons of positive or negative helicity or perhaps their sum $`\mathrm{\Phi }_++\mathrm{\Phi }_{}`$ as a position representation of a photon. However photons are not localizable so this doesn’t work. One way around (cf. ) is to show that an operator representating the number of photons in an arbitrary volume V can be defined but not as the integral over V of a photon density operator. Another approach (cf. ) is to determine an operator representing the number of photons in a volume V as the integral over V of a so called detection operator which (when the linear dimensions of V are large compared to the photon wavelengths) leads to a simple formula for the probability that n photons are present in V (cf. also and Section 5.1 for coarse grained photon density and current density operators). Here one proceeds differently following and looks for a probability amplitude for the photon energy to be detected about $`d𝐫`$ of $`𝐫`$ in the form $`\mathrm{\Psi }^{}\mathrm{\Psi }d𝐫`$ with normalizations in the sense that
(5.6)
$$\mathrm{\Psi }^{}\mathrm{\Psi }𝑑𝐫=cp[\stackrel{}{\gamma }_+^{}\stackrel{}{\gamma }_++\stackrel{}{\gamma }_{}^{}\stackrel{}{\gamma }_{}]𝑑𝐩$$
To do this one sets $`\mathrm{\Psi }=\mathrm{\Psi }_++\mathrm{\Psi }_{}`$ with
(5.7)
$$\mathrm{\Psi }_\pm (𝐫,t)=\frac{\sqrt{cp}d𝐩}{(2\pi \mathrm{})^{3/2}}\stackrel{}{\gamma }_\pm (𝐩,t)e^{i𝐩𝐫/\mathrm{}}$$
One notes then that $`i\mathrm{}_t\mathrm{\Psi }_\pm =\pm c\mathrm{}\times \mathrm{\Psi }_\pm `$ and one must satisfy (cf. (5.5)) an initial condition given by the $`t=0`$ case of
(5.8)
$$\mathrm{\Psi }_\pm (𝐫,t)=\frac{\sqrt{cp}d𝐩}{(2\pi \mathrm{})^{3/2}}\stackrel{}{\gamma }_\pm (𝐩,0)e^{icpt/\mathrm{}}e^{i𝐩𝐫/\mathrm{}}$$
Note here that p (the usual photon momentum) and r, the position associated with the photon energy, are not conjugate variables. One then builds up a QFT of of the free radiation field via many particle physics (not from a canonical formulation of the EM fields) and this is equivalent to standard canonical quantization. Moreover upon specializing to one photon the energy functions $`\mathrm{\Psi }_\pm `$ above are recovered. Thus it is reasonable to describe the single photon energy distribution in a region $`d𝐫`$ about r via $`\mathrm{\Psi }^{}(𝐫,t)\mathrm{\Psi }(𝐫,t)d𝐫`$. Further it is shown that in a spontaneous emission process the wave function $`\mathrm{\Psi }(𝐫,t)`$ generated is a causal field, propagating out from the emitting atom at the speed of light.
Now one goes to where the Bohr photon having a specific size and shape is discussed. This involves a circularly polarized photon being a monochromatic EM traveling wave confined within a circular ellipsoid of length equal to the wavelength ($`\lambda `$) and diameter $`\lambda /\pi `$ propagating along the long axis of the ellipsoid. In this model the quantization of the photon’s angular momentum (corresponding to spin $`\mathrm{}`$) arises from an appropriately chosen of Maxwell’s equations and the energy is quantized to be $`\mathrm{}\nu `$. In a sense, not entirely clear (cf. ), one can think here of an ellipsoidal soliton arising from the imposition of causality upon the solution of the linear Maxwell equations where EM energy $`𝐄^2+𝐇^2`$ integrated over the volume of the ellipsoid equals $`\mathrm{}\nu `$ leading to an average intensity within the photon-soliton of $`I_p=4\pi \mathrm{}c^2/\lambda ^4`$. The word wavicle is also used in . For a wave traveling with the speed of light parallel to the z-axis the solution of Maxwell’s equations can be any function of $`zct`$ and if monochromatic one has a term $`S(zct)=exp[2\pi i(zct)/\lambda ]`$. Setting $`x=rCos(\varphi )`$ and $`y=rSin(\varphi )`$ in the already separated d’Alembert equation then leads to
(5.9)
$$\frac{1}{\mathrm{\Phi }(\varphi )}\frac{d^2\mathrm{\Phi }}{d\varphi ^2}=m^2=\frac{1}{R(r)}\left[\frac{d^2R}{dr^2}+\frac{1}{r}\frac{dR}{dr}\right]$$
where $`m^2`$ is the real separation constant. The simple plane wave solutions with $`m^2=0`$ are rejected here since light is observed to travel along very narrow beams and for $`m^2=1`$ one has factors of $`r`$ or $`1/r`$ with angular factors $`exp(\pm i\varphi )`$. This corresponds to angular momentum $`L_z=(\mathrm{}/i)_\varphi `$ leading to solutions
(5.10)
$$\psi (r,\varphi ,zct)=(\alpha r+\beta /r)(Ae^{i\varphi }+Be^{i\varphi })e^{2\pi i(zct)/\lambda }$$
This yields then
(5.11)
$$E_z=H_z=0;E_x=(\alpha r+\beta /r)\left[Ae^{i\varphi }+Be^{i\varphi }\right]e^{2\pi i(zct)/\lambda }=\mu _0cH_y;$$
$$E_y=i(\alpha r\beta /r)\left[Ae^{i\varphi }Be^{i\varphi }\right]e^{2\pi i(zct)/\lambda }=\mu _0cH_x$$
Imposing causality leads to the result that if A or B is zero then the field must be contained within a circular ellipsoid of length $`\lambda `$ and cross sectional diameter $`\lambda /\pi `$ (cf. ). The amplitude is determined by integration of the energy $`𝐄^2+𝐇^2`$ and the $`1/r`$ term is then discarded to preserve the ellipsoidal shape; there results $`A^2+B^2=1`$ and $`\alpha ^2=120h\mathrm{}c\pi ^44/ϵ_0\lambda ^6`$ (in suitable units). In addition one expects an evanescent wave decaying like $`1/r`$ (with $`\alpha =0`$) described via
(5.12)
$$E_r=\frac{\beta }{r}[A+B]=\mu _0cH_\varphi ;E_\varphi =i\frac{\beta }{r}[AB]=\mu _0cH_r$$
where $`\alpha r=\beta /r`$ for $`r=\lambda /2\pi `$ and $`\beta ^2=(\lambda /2\pi )^4\times 120n\mathrm{}c\pi ^4/(ϵ_0\lambda ^6)`$. The evanescent wave is believed to be responsible for diffraction and interference and some experimental material is sketched.
### 5.1. PHOTON DYNAMICS
We sketch here from where it is shown that one can define the notions of photon density and photon current density with certain limits. As a trivial example think of geometrical optics where light is treated as an ensemble of point photons moving along definite trajectories with speed c. In the geometric limit the photon number density and current density are perfectly well defined as are the density and current density for any collection of point particles. As a second example one refers to where Mandel defines an operator $`n_V`$ representing the number of photons in V as the integral over V of the photon density $`D_M(𝐱)=𝐀^{}(𝐱)𝐀(𝐱)`$ where
(5.13)
$$𝐀(𝐱)=L^{3/2}\underset{𝐤,\lambda }{}\stackrel{}{ϵ}_{𝐤,\lambda }a_{𝐤,\lambda }e^{i(𝐤𝐱\omega t)}$$
is the so-called detection operator (here $`\stackrel{}{ϵ},a,`$ and $`\omega =c𝐤`$ are respectively the polarization unit vector, the annihilation operator and the frequency of a transverse photon of wave vector k and polarization $`\lambda (=1,2)`$ with $`L^3`$ the quantization volume). It is shown that when the linear dimensions of V are large compared to the photon wavelengths this definitiion of $`n_V`$ yields a simple for the probability $`p_V(n)`$ that n photons are present in V. It was later shown by Amrein that $`n_V`$ agrees with that derived from the theory of and this all has motivated the study in that a coarse grained photon density operator can exist even though a fine grained or microscopic one may not.
Thus one derives a photon density $`D(𝐱)`$ and a photon current density $`𝐂(𝐱)`$ to satisfy $`()_tD+𝐂=0`$ (conservation of photons ignoring absorption and emission). These will be defined in terms of vector field operators $`\stackrel{}{\psi }(𝐱)`$ and $`\stackrel{}{\varphi }(𝐱)`$ which will be referred to as the photon field (cf. here also Section 2 again). For a volume V large compared to the photon wavelengths $`D(𝐱)`$ will correctly predict the number statistics of photons in that volume while for a time interval $`[t,t+T]`$ long compared to $`\lambda /c`$ $`𝐂(𝐱)`$ correctly predicts the statistics of the number of photons that cross the surface S in time T. This was worked out in the first paper of for a discrete situation and is redone in the second paper in a continuum context; we sketch this here for the free field case and, following , show that photon dynamics is a relativistically covariant theory. Thus write
(5.14)
$$\stackrel{}{\psi }(𝐱,t)=\frac{1}{\sqrt{2(2\pi )^3}}\underset{\lambda =1}{\overset{2}{}}d^3k\stackrel{}{ϵ}_\lambda (𝐤)a_\lambda (𝐤)e^{i(𝐤𝐱\omega t)};$$
$$\stackrel{}{\varphi }(𝐱,t)=\frac{1}{\sqrt{2(2\pi )^3}}\underset{\lambda }{}d^3k\left(\frac{𝐤}{k}\times \stackrel{}{ϵ}_\lambda (𝐤)\right)a_\lambda (𝐤)e^{i(𝐤𝐱\omega t)}$$
where $`a_\lambda (𝐤)`$ is the annihilation operator and $`\stackrel{}{ϵ}_\lambda (𝐤)`$ the polarization vector of a transverse photon of wave vector k and polarization $`\lambda (=1,2)`$. Evidently we have the free field equations
(5.15)
$$\stackrel{}{\psi }=0;\stackrel{}{\varphi }=0;\times \stackrel{}{\psi }+\frac{1}{c}_t\stackrel{}{\varphi }=0;\times \stackrel{}{\varphi }\frac{1}{c}_t\stackrel{}{\psi }=0$$
Then one defines
(5.16)
$$D=\stackrel{}{\psi }^{}\stackrel{}{\psi }+\stackrel{}{\varphi }^{}\stackrel{}{\varphi };𝐂=c(\stackrel{}{\psi }^{}\times \stackrel{}{\varphi }\stackrel{}{\varphi }^{}\times \stackrel{}{\psi })$$
Evidently $`()_tD+𝐂=0`$ as required. Note that D is a positive definite operator whose integral over all space is the usual photon number operator
(5.17)
$$d^3xD(𝐱=\underset{\lambda }{}d^3ka_\lambda ^{}(𝐤)a_\lambda (𝐤)$$
The interpretation of C as the photon current density is justified by $`()`$ and by a calculation showing that the integral of the inward normal component of C over the surface of an ideal photon detector equals the counting rate of the detector (cf. first paper). The operators for the number of photons in V and the number of photons crossing a given surface in the time interval $`[t,t+T]`$ are
(5.18)
$$n_V=_Vd^3xD(𝐱);n_T=_t^{t+T}𝑑t^{}_S𝑑a𝐧𝐂(𝐱,t)$$
(n is the unit normal to S in the direction of interest). For a volume V large as described the probability that V contains m photons is $`()p_V(m)=Tr[\rho :n_V^mexp(n_V):]/m!`$ where $`\rho `$ is the density operator of the radiation field and $`::`$ means normal ordering. Similarly for sufficently large T the probability that m photons cross the surface S in time T is $`(\mathrm{}\mathrm{})p_T(m)=Tr[\rho :n_T^mexp(n_T):]/m!`$ where S is the sensitive surface of an ideal photon detector (with one unit quantum efficiency) and $`p_T(m)`$ is the photon count distribution measured by the detector. For these calculations see the first paper of and ($`n_V`$ and $`n_T`$ are treated as number operators).
Now the transverse EM field operators $`𝐄=𝐄^++𝐄^{}`$ and $`𝐁=𝐁^++𝐁^{}`$ can be expressed via
(5.19)
$$𝐄=\frac{i}{2\pi }\underset{\lambda }{}d^3k(\mathrm{}\omega )^{1/2}\stackrel{}{ϵ}(𝐤)a_\lambda (𝐤)e^{i(𝐤𝐱\omega t)};$$
$$𝐁=\frac{i}{2\pi }\underset{\lambda }{}d^3k(\mathrm{}\omega )^{1/2}\left(\frac{𝐤}{k}\times \stackrel{}{ϵ}(𝐤)\right)a_\lambda (𝐤)e^{i(𝐤𝐱\omega t)}$$
and $`𝐄^{}=(𝐄^+)^{}`$ with $`𝐁^{}=(𝐁^+)^{}`$. One sees that the photon field vectors in (5.14) are obtained from $`𝐄^+`$ and $`𝐁^+`$ by multiplying the momentum components of $`𝐄^+`$ and $`𝐁^+`$ by $`i[4\pi \mathrm{}\omega (𝐤)]^{1/2}`$ which corresponds to a convolution in position space
(5.20)
$$\stackrel{}{\psi }(𝐱,t)=d^3yg(𝐱𝐲)𝐄^+(𝐲,t);\stackrel{}{\varphi }(𝐱,t)=d^3yg(𝐱𝐲)𝐁^+(𝐲,t)$$
where
(5.21)
$$g(𝐱)=\frac{i}{(2\pi )^3}d^3k(4\pi \mathrm{}\omega )^{1/2}e^{i𝐤𝐱};d^3yg^1(𝐱𝐲)g(𝐲𝐳=\delta ^3(𝐱𝐳);$$
$$g^1(𝐱)=\frac{i}{(2\pi )^3}d^3k(4\pi \mathrm{}\omega )^{1/2}e^{i𝐤}𝐱$$
One has also
(5.22)
$$𝐄^+=d^3yg^1(𝐱𝐲)\stackrel{}{\psi }(𝐲,t);𝐁^+=d^3yg^1(𝐱𝐲)\stackrel{}{\varphi }(𝐲,t)$$
It is convenient now to express the photon field as a matrix
(5.23)
$$\psi _{\mu \nu }=\left(\begin{array}{cccc}0& \psi _1& \psi _2& \psi _3\\ \psi _1& 0& \varphi _3& \varphi _2\\ \psi _2& \varphi _3& 0& \varphi _1\\ \psi _3& \varphi _2& \varphi _1& 0\end{array}\right)$$
Note here the similarity to (2.2) in Section 2 (apparently A. de la Torre was unaware of Cook’s work). An immediate relation to the EM field strength tensor is exhibited via
(5.24)
$$F_{\mu \nu }^+=\left(\begin{array}{cccc}0& E_1^+& E_2^+& E_3^+\\ E_1^+& 0& B_3^+& B_2^+\\ E_2^+& B_3^+& 0& B_1^+\\ E_3^+& B_2^+& B_1^+& 0\end{array}\right)$$
One is using coordinates $`x^\mu =(ct,x,y,z)`$ and metric $`g^{00}=1,g^{ii}=1`$ and will raise and lower indices with $`g^{\alpha \beta }`$or $`g_{\alpha \beta }`$ as in $`\psi _{\mu \nu }`$ were a tensor (it turns out to transform as a tensor under displacements and spatial rotations but not for boosts - cf. ). Now define
(5.25)
$$G(x)=\frac{i}{(2\pi )^4}d^4k[4\pi \mathrm{}\omega (k)]^{1/2}e^{ikx}$$
where $`xx^\mu ,kk^\mu =(k^0,𝐤),kxk^\mu x_\mu ,`$ and $`\omega (k)=c|𝐤|`$. Clearly G has an inverse with $`d^4yG^1(xy)G(yz)=\delta ^4(xz)`$ etc. Then one can write
(5.26)
$$\psi _{\mu \nu }(x)=d^4yG(xy)F_{\mu \nu }^+(y);F_{\mu \nu }^+(x)=d^4yG^1(xy)\psi _{\mu \nu }(y)$$
To see that these are equivalent to (5.20) and (5.22) note that since $`\omega (k)=c|𝐤|`$ does not depend on $`k^0`$ the $`k^0`$ integrals can be evaluated immediately to give $`G(x)=\delta (x^0)g(𝐱)`$ etc. Although (5.20) and (5.22) were derived from (5.14) and (5.19) for transverse photon and EM fields one assumes that they and hence (5.26) remain valid when the EM fields have a longitudinal component (for the free field this is of no concern). Now one shows that the photon field equations (5.22) are a direct consequence of the free field Maxwell equations
(5.27)
$$^\nu F_{\mu \nu }^+=0;_\alpha F_{\beta \gamma }^++_\beta F_{\gamma \alpha }^++_\gamma F_{\alpha \beta }^+=0$$
The second of these equations is a general operator relation following from $`F_{\mu \nu }^+=_\nu A_\mu ^+_\mu A_\nu ^+`$ while the first equation is valid in the sense that $`^\nu F_{\mu \nu }^+|>=0`$ for all physically admissable states $`|>`$ (which are defined as those satisfying the Gupta-Bleuler condition $`(\mathrm{})^\mu A_\mu ^+|>=0`$ \- note the free field vector potential satisfies the wave equation $`^\nu _\nu A_\mu ^+=0`$). Now consider
(5.28)
$$\frac{\psi _{\mu \nu }(x)}{x^\alpha }=d^4y\frac{G(xy)}{x^\alpha }F_{\mu \nu }^+(y)=$$
$$=d^4y\frac{G(xy)}{y^\alpha }F_{\mu \nu }^+(y)=d^4yG(xy)\frac{F_{\mu \nu }^+(y)}{y^\alpha }$$
Neglect of the integrated part is justified via $`G(x)0`$ for $`x\mathrm{}`$ and $`F_{\mu \nu }^+(x)0`$ at spatial infinity. From (5.27) one has then
(5.29)
$$^\nu \psi _{\mu \nu }=0;_\alpha \psi _{\beta \gamma }+_\beta \psi _{\gamma \alpha }+_\gamma \psi _{\alpha \beta }=0$$
and these are equivalent to the original photon field equations(5.15); again one restricts to the subspace of physical states. Although these have the appearance of tensor equations they are not manifestly covariant since $`\psi _{\mu \nu }`$ is not a tensor (cf. Section 2 for comments in this direction). Nevertheless the equations are shown to be invariant under Lorentz transformations because the photon field $`\psi _{\mu \nu }`$ is defined in terms of the tensor $`F_{\mu \nu }^+`$ in the same way in each Lorentz frame (see here for details).
Finally one considers the matrix of Hermitian operators
(5.30)
$$N^{\alpha \beta }=\psi _\lambda ^\alpha \psi ^{\lambda \beta }+\psi ^{\lambda \beta }\psi _\lambda ^\alpha +\frac{1}{2}g^{\alpha \beta }\psi ^{\mu \nu }\psi _{\mu \nu }$$
with is analogous to the EM energy momentum tensor. One checks easily that $`(\mathrm{}\mathrm{})_\beta N^{\alpha \beta }=0M^\alpha =d^3xN^{\alpha 0}`$ is conserved as the photon field develops in time. In fact one can write
(5.31)
$$N^{\alpha \beta }=\left(\begin{array}{cccc}D& C_1/c& C_2/c& C_3/c\\ C_1/c& S_{11}/c^2& S_{12}/c^2& C_{13}/c^2\\ C_2/c& S_{21}/c^2& S_{22}/c^2& S_{23}/c^2\\ C_3/c& S_{31}/c^2& S_{32}/c^2& S_{33}/c^2\end{array}\right)$$
Here $`(\mathrm{}\mathrm{})S_{ij}=c^2[D\delta _{ij}(\psi _i^{}\psi _j+\psi _j^{}\psi _i)(\varphi _i^{}\varphi _j+\varphi _j^{}\varphi _i)]`$ is a $`3\times 3`$ matrix analogous to the Maxwell stress tensor and $`(\mathrm{}\mathrm{})`$ now takes the form
(5.32)
$$_tD+𝐂=0;_tC_i+\frac{S_{ij}}{x^j}=0;$$
$$M^0=d^3xD(𝐱)=const.;cM^i=d^3xC_i(𝐱=const.$$
These equations express the local and global conservation of photons. One shows that $`N^{\alpha \beta }`$ does not transform as a tensor (except for coordinate displacements and spatial rotations) and in fact there is no general transformation law relating the components of $`N^{\alpha \beta }`$ in different Lorentz frames. Nevertheless (5.32) are covariant since the photon field equations are covariant and $`N^{\alpha \beta }`$ is constructed from the photon field in the same way in each frame. In particular the number of photons $`M^0`$ is independent of time in each Lorentz frame and considerable calculation also shows that $`M^0`$ is a scalar (using the condition $`(\mathrm{})`$).
## 6. SOME SPECULATIONS ON THE AETHER
The aether has been reviewed in to a certain extent and in some speculations were advanced concerning a possible geometry for the aether. These were based on work of and we sketch here some variations and embellishments. First we note from (5.15) that the components $`\psi _i`$ and $`\varphi _i`$ satisfy the massless KG equation so for analysis of photons one needs 6 components $`(\psi _i,\varphi _i)`$ each satisfying a massless KG equation. However the equations (5.15) are exactly the same as the Maxwell equations (3.7) so one could also imagine introducing a vector $`\mathrm{\Psi }=(𝐀,\varphi )`$ with $`A_{\mu \nu }=\mathrm{\Psi }_{\mu ,\nu }\mathrm{\Psi }_{\nu ,\mu }`$ to generate the photon equations for a free field with $`\mathrm{}\mathrm{\Psi }=0`$ (see e.g. ). In this spirit then one would have a 4-vector $`\mathrm{\Psi }`$ satisfying the massless KG equation to serve as a generator of photon activity. In any event we will think of fields labeled $`\psi _i`$ for $`i=0,1,2,3`$ as characterizing photon dynamics with each component satisfying the massless KG equation. Then we will apply the machinery of $`(x,\psi )`$ duality of Faraggi-Matone and Vancea (see especially ) to express the coordinates $`x^\mu `$ in terms of the fields $`\psi _i`$ arising from $`\mathrm{\Psi }`$ (which will be called aether fields); they are seen to be “potential” fields for the photon fields $`\psi _i,\varphi _i`$ of Section 5.1.
As background here we refer to a lovely paper of P. Isaev where he makes conjectures, with supporting arguments, which arrive at a definition of the aether as a Bose-Einstein condensate of neutrino-antineutrino pairs of Cooper type (Bose-Einstein condensates of various types have been considered by others in this context - cf. ). The equation for the $`\psi `$-aether is then a solution of the massless Klein-Gordon (KG) equation (photon equation) $`(\mathrm{}^2\mathrm{\Delta }(\mathrm{}^2/c^2)_t^2)\psi =0`$. This $`\psi `$ field heuristically acts as a carrier of waves (playground for waves) and one might say that special relativity (SR) is a way of including the influence of the aether on physical processes and consequently SR does not see the aether (cf. here also the idea of a Dirac aether in and Einstein-aether theories as in \- this is discussed further in ). In the electromagnetic (EM) theory in one looks at $`\stackrel{}{\psi }=(\varphi ,\stackrel{}{A})`$ with $`\mathrm{}\psi _i=0`$ as the defining equation for a real $`\psi `$-aether, in terms of the potentials $`\varphi `$ and $`\stackrel{}{A}`$ which therefore define the $`\psi `$-aether. EM waves are then considered as oscillations of the $`\psi `$ aether and wave processes in the aether accompanying a moving particle determine wave properties of the particle. Interesting examples involving standing EM waves in a spherical resonator are attributed to oscillations of the $`\psi `$ aether and references to superconductivity à la Volovik are indicated.
In Faraggi and Matone develop a theory of $`x\psi `$ duality, related to Seiberg-Witten theory in the string arena, which was expanded in various ways in . Here one works from a stationary SE $`[(\mathrm{}^2/2m)\mathrm{\Delta }+V(x)]\psi =E\psi `$, and, assuming for convenience one space dimension, the space variable $`x`$ is determined by the wave function $`\psi `$ from a prepotential $`𝔉`$ via Legendre transformations. The theory suggests that $`x`$ plays the role of a macroscopic variable for a statistical system with a scaling term involving $`\mathrm{}`$. Thus define a prepotential $`𝔉_E(\psi )=𝔉(\psi )`$ such that the dual variable $`\psi ^D=𝔉/\psi `$ is a (linearly independent) solution of the same SE. Take V and E real so that $`\overline{\psi }=\psi ^D`$ qualifies and write $`_x𝔉=\psi ^D_x\psi =(1/2)[_x(\psi \psi ^D)+W)]`$ where W is the Wronskian. This leads to ($`\psi ^D=\overline{\psi }`$) the relation $`𝔉=(1/2)\psi \overline{\psi }+(W/2)x`$ (setting the integration constant to zero). Consequently, scaling W to $`2i\sqrt{2m}/\mathrm{}`$ one obtains
(6.1)
$$\frac{i\sqrt{2m}}{\mathrm{}}x=\frac{1}{2}\psi \frac{𝔉}{\psi }𝔉\frac{i\sqrt{2m}}{\mathrm{}}x=\psi ^2\frac{𝔉}{\psi ^2}𝔉$$
which exhibits $`x`$ as a Legendre transform of $`𝔉`$ with respect to $`\psi ^2`$. Duality of the Legendre transform then gives also
(6.2)
$$𝔉=\varphi _\varphi \left(\frac{i\sqrt{2m}x}{\mathrm{}}\right)\left(\frac{i\sqrt{2m}x}{\mathrm{}}\right);\varphi =_{\psi ^2}𝔉=\frac{\overline{\psi }}{2\psi }$$
so that $`𝔉`$ and $`(i\sqrt{2m}x/\mathrm{})`$ form a Legendre pair. In particular one has $`\rho =|\psi |^2=\frac{2i\sqrt{2m}}{\mathrm{}}x+2𝔉`$ which also relates $`𝔉`$ and the probability density. In any event one sees that the wave function $`\psi `$ specifically determines the location of the “particle” whose quantum evolution is described by $`\psi `$. We mention here also that the (stationary) SE can be replaced by a third order equation
(6.3)
$$4𝔉^{\prime \prime \prime }+(V(x)E)(𝔉^{}\psi 𝔉^{\prime \prime })^3=0;𝔉^{}\frac{𝔉}{\psi }$$
and a dual stationary SE has the form
(6.4)
$$\frac{\mathrm{}^2}{2m}\frac{^2x}{\psi ^2}=\psi [EV]\left(\frac{x}{\psi }\right)^3$$
A noncommutative version of this is developed in the second paper of .
We mention for some material on the aether and the vacuum and refer to the bibliography for other references. We sketch first some material from which extends the SE theory to the Klein-Gordon (KG) equation. Following take a spacetime manifold M with a metric field $`g`$ and a scalar field $`\psi `$ satisfying the KG equation. Locally one has cartesian coordinates $`x^\alpha (\alpha =0,1.\mathrm{},n1)`$ in which the metric is diagonal with $`g_{\alpha \beta }(x)=\eta _{\alpha \beta }(x)`$ and the KG equation has the form $`(\mathrm{}_x+m^2)\psi (x)=0`$ ($`\mathrm{}_x(\mathrm{}^2/c^2)[(_t^2/c^2)^2]`$). Defining prepotentials such that $`\stackrel{~}{\psi }^{(\alpha )}=𝔉^{(\alpha )}[\psi ^{(\alpha )}]/\psi ^{(\alpha )}`$ where $`\psi ^{(\alpha )}`$ and $`\stackrel{~}{\psi }^{(\alpha )}`$ are two linearly independent solutions of the KG equation depending on parameters $`x^\alpha `$ one has as above (with a different scaling factor)
(6.5)
$$\frac{\sqrt{2m}}{\mathrm{}}x^\alpha =\frac{1}{2}\psi ^{(\alpha )}\frac{𝔉^{(\alpha )}[\psi ^{\alpha )}]}{\psi ^{(\alpha )}}𝔉^{(\alpha )};[^\alpha _\alpha V^\alpha ]\psi ^\alpha =0$$
This is suggested in and used in ; the factor $`\sqrt{2m}/\mathrm{}`$ is simply a scaling factor (possibly too stringent here) and it would be more productive to scale $`x^0ct`$ differently or in fact to scale all variables as indicated in (cf. below for a general scaling). Locally $`𝔉^{(\alpha )}`$ satisfies the third order equation
(6.6)
$$4𝔉^{(\alpha )^{^{\prime \prime \prime }}}+[V^{(\alpha )}(x^\alpha )+m^2](\psi ^{(\alpha )}𝔉^{(\alpha )^{^{\prime \prime }}}𝔉^{(\alpha )^{})})^3=0$$
where $`{}_{}{}^{}/\varphi ^{(\alpha )}`$ and a (quantum) potential $`V^\alpha `$ has the form
(6.7)
$$V^{(\alpha )}(x^\alpha )=\left[\frac{1}{\psi (x)}\underset{\beta =0,\beta \alpha }{\overset{n1}{}}^\beta _\beta \psi (x)\right]|_{x^{\beta \alpha }fixed}$$
We go back to now and derive equations for the KG equation with $`m=0`$ from the beginning (rather than rescaling and then taking $`m0`$). Further we proceed with more detail and show how a general scaling will involve insertion of some variable factors (cf. also for various scaling factors). Thus consider $`(1/c^2)\psi _{tt}\mathrm{\Delta }\psi =0`$ with $`x^0=ct`$ and write out explicitly ($`i=1,2,3`$)
(6.8)
$$\frac{1}{c^2}_t^2\psi ^0V^0\psi ^0=0;V^0=\frac{\mathrm{\Delta }\psi }{\psi }=\frac{(1/c^2)\psi _{tt}}{\psi };$$
$$_i^2\psi ^iV^i\psi ^i=0;V^i=\frac{\left(\frac{1}{c^2}_t^2\psi _{ji}_j^2\psi \right)}{\psi }$$
Here $`V^i`$ is thought of as $`V^i(x^i)`$ (where in fact $`V^i=V^i(x^i,x^j,x^0)`$ with $`ji`$ and $`x^0,x^j`$ are considered as parameters). Similarly $`V^0=V^0(x^0)`$ ($`V^0(x^0,x^i)`$). Now e.g. for $`\psi ^0`$ and $`\stackrel{~}{\psi }^0`$ linearly independent solutions of the first equation in (6.8) one has $`\psi _{tt}^0\stackrel{~}{\psi }^0=\psi ^0\stackrel{~}{\psi }_{tt}^0`$ which implies
(6.9)
$$W^0(t)=(\psi ^0\stackrel{~}{\psi }_t^0(t)\stackrel{~}{\psi }^0\psi _t^0)(t)=2c\gamma (x^i)$$
Here, as specified above, $`\stackrel{~}{\psi }^0=𝔉^0/\psi ^0`$, and
(6.10)
$$_t𝔉^0=𝔉_\psi ^0\psi _t=\stackrel{~}{\psi }^0\psi _t$$
$$\frac{1}{2}_t(\psi ^0\stackrel{~}{\psi }^0)\stackrel{~}{\psi }^0\psi _t^0=\frac{1}{2}(\psi ^0\stackrel{~}{\psi }_t^0\psi _t^0\stackrel{~}{\psi }^0)=\frac{1}{2}W^0=c\gamma (x^i)$$
and consequently one can write
(6.11)
$$c\gamma (x^i)t=\frac{1}{2}\psi ^0\frac{𝔉^0}{\psi ^0}𝔉^0=𝔈^0$$
This leads to (for $`\psi ^0\varphi `$)
(6.12)
$$c\gamma (x^i)=\frac{𝔈^0}{\varphi }\frac{d\varphi }{dt}=\left[\frac{1}{2}\left(𝔉_\varphi ^0+\varphi \frac{^2𝔉^0}{\varphi ^2}\right)𝔉_\varphi ^0\right]\frac{d\varphi }{dt}=$$
$$=\frac{1}{2}\left(\varphi \frac{^2𝔉^0}{\varphi ^2}𝔉_\varphi ^0\right)\frac{d\varphi }{dt}=\frac{1}{2}E^0\frac{d\varphi }{dt}$$
Similarly we write, using (6.8),
(6.13)
$$\stackrel{~}{\psi }^i=\frac{𝔉^i}{\psi ^i};W^i=\psi ^i_t\stackrel{~}{\psi }^i\stackrel{~}{\psi }^i_t\psi ^i;\beta ^i(x^0,x^j)x^i=\frac{1}{2}\psi ^i\frac{𝔉^i}{\psi ^i}𝔉^i=𝔈^i$$
Consequently ($`\psi _i\psi ^i`$)
(6.14)
$$\gamma dx^0=c\gamma dt=\frac{1}{2}E^0d\psi ^0;\beta ^idx^i=\frac{1}{2}E^id\psi ^i=\frac{1}{2}\left(\psi ^i\frac{^2𝔉^i}{\psi _i^2}\frac{𝔉^i}{\psi ^i}\right)d\psi ^i$$
Since $`_t=_\varphi (d\varphi /dt)`$, etc. one can write then
(6.15)
$$_t=\left(\frac{2c\gamma }{E^0}\right)_\varphi ;_i=\left(\frac{2\beta }{E^i}\right)\frac{}{\psi ^i}$$
The extraneous variables are considered as parameters when concentrating on one $`x^i`$ or $`x^0`$ and we note from (6.11) or (6.13) that $`x^0`$ or $`x^i`$ can be considered as a function of $`\varphi =\psi ^0`$ or $`\psi ^i`$ and $`𝔉^i`$ is a function of $`\psi ^i`$ (satisfying ordinary differential equations as in (6.6) - with $`m=0`$). Here (6.14)-(6.15) represents an induced parametrization on the spaces $`T_P(U)`$ and $`T_P^{}(U)`$ ($`PU`$ \- local tangent and cotangent spaces). Now using the linearity of the metric tensor field one sees that the components of the metric in the $`\{(\psi ^\alpha ,𝔉^\alpha )\}`$ parametrization are ($`\beta ^0=c\gamma `$)
(6.16)
$$G_{\alpha \sigma }(\psi )=\frac{E^\alpha E^\sigma }{4\beta ^\alpha \beta ^\sigma }\eta _{\alpha \sigma }(x)$$
(cf. ). Now following let $`z^\mu (\mu =0,1,\mathrm{},n1)`$ be a general coordinate system in U and write the coordinate transformation matrices via
(6.17)
$$A_\mu ^\alpha =\frac{x^\alpha }{z^\mu };(A^1)_\alpha ^\mu =\frac{z^\mu }{x^\alpha }$$
The metric then takes the form
(6.18)
$$g_{\mu \nu }(z)=\frac{4\beta ^\alpha \beta ^\sigma }{E^\alpha E^\sigma }A_\mu ^\alpha A_\nu ^\sigma G_{\alpha \sigma }(\psi )$$
The components of the metric connection can be computed via
(6.19)
$$\mathrm{\Gamma }_{\mu \nu }^\rho =\frac{1}{2}g^{\rho \sigma }(z)\underset{𝒫}{}ϵ_𝒫𝒫\left[\frac{g_{\sigma \nu }(z)}{z^\mu }\right]$$
where $`𝒫`$ is a cyclic permutation of the ordered set of indices $`\{\sigma \nu \mu \}`$ and $`ϵ_𝒫`$ is the signature of $`𝒫`$. Via the coordinate transformation (6.17) the function $`\psi ^\alpha `$ depends on all the $`z^\mu `$. The metric connection (6.19) can be expressed in the $`\{\psi ^\alpha ,𝔉^\alpha \}`$ parametrization and in one computes also the components of the curvature tensor, the Ricci tensor, and the scalar curvature and gives an expression for the Einstein equations (we omit the details here). The same procedure apply to our formulas above which leads us to state heuristically
###### THEOREM 6.1.
The formulas (6.14), (6.15), (6.16), (6.17), (6.18), and (6.19), and their continuations determine a geometry for a putative aether, expressed in terms of our so-called aether fields $`\psi _i`$.
REMARK 1.1. These matters are taken up again in for a general curved spacetime and some sufficient constraints are isolated which make the theory work. Also in both papers a quantized version of the KG equation is also treated and the relevant $`x\psi `$ duality is spelled out in operator form. We omit this also in remarking that the main feature here for our purposes is the fact that one can describe spacetime geometry (at least locally) in terms of (field) solutions of a KG equation and prepotentials (which are themselves functions of the fields). In other words the coordinates are programmed by fields and if the motion of some particle of mass m is involved then its coordinates are choreographed by the fields with a quantum potential eventually entering the picture via (6.7). In a similar duality is worked out for the Dirac field and cartesian coordinates and to connect this with the aether idea one should examine such formulas for $`m0`$.$`\mathrm{}`$
###### EXAMPLE 6.1.
One knows that general solutions of the massless KG equation will have the form $`\psi =\psi (𝐚𝐱ct)`$ with $`|𝐚|=1`$. For example take $`\psi =exp(a_ix_ict)`$ with $`(1/c^2)\psi _{tt}=\psi `$ and $`\psi _{ii}=a_i^2\psi `$. This leads to
(6.20)
$$V^0=1;V^i=1\underset{ji}{}a_j^2$$
Hence
(6.21)
$$\frac{1}{c^2}_t^2\psi ^0\psi ^0=0;_i^2\psi ^i(1\underset{ji}{}a_j^2)\psi ^i=0$$
On the other hand if $`\psi =f(𝐚𝐱ct)`$ one gets
(6.22)
$$V^0=\left(\frac{f^{\prime \prime }}{f}\right)(𝐚𝐱ct);V^i=\left(1\underset{ji}{}a_j^2\right)\left(\frac{f^{\prime \prime }}{f}\right)(𝐚𝐱ct)$$
Setting $`f^{\prime \prime }/f=g(x^i,x^0)`$ one has
(6.23)
$$_0^2\psi ^0g(x^i,x^0)\psi ^0=0;_i^2\psi ^i\left(1\underset{ji}{}a_j^2\right)g(x^i,x^j,x^0)\psi ^i$$
Here the $`x^i`$ or $`(x^j,x^0)`$ are considered as parameters.$`\mathrm{}`$
###### EXAMPLE 6.2.
Consider a simple situation with two $`x^i`$ variables and $`x^0=ct`$ and take $`a_1=a_2=1/\sqrt{2}`$. Then $`V^0=1`$ and $`V^i=1(1/2)=1/2`$ with
(6.24)
$$\frac{1}{c^2}_t^2\psi ^0=\psi ^0;\frac{^2\psi ^i}{(x^i)^2}\psi ^i=\frac{1}{2}\psi ^i$$
Hence we can take
(6.25)
$$\psi ^0=A_0e^{ct};\psi ^i=A_ie^{(1/\sqrt{2})x^i};\stackrel{~}{\psi }^i=\stackrel{~}{A}_ie^{(1/\sqrt{2})x^i};\stackrel{~}{\psi }^0=\stackrel{~}{A}_0e^{ct}$$
Now $`\psi ^i\stackrel{~}{\psi }^i=\kappa _i`$ for $`i=0,1,2`$ so (recall $`\beta ^0=\gamma `$ and $`x^0=ct`$)
(6.26)
$$\beta ^ix^i=\frac{1}{2}\kappa _i𝔉^i=𝔈^i(i=0,1,2)$$
and
(6.27)
$$\frac{1}{2}E^i=\frac{}{\psi ^i}\left(\frac{1}{2}\kappa _i𝔉^i\right)=\frac{𝔉^i}{\psi ^i}=\stackrel{~}{\psi }^i(i=0,1,2)$$
(the $`\beta ^i`$ here need not depend on other variables). Consequently one has
(6.28)
$$G_{\alpha \sigma }(\psi )=\frac{E^\alpha E^\sigma }{4\beta ^\alpha \beta ^\sigma }\eta _{\alpha \sigma }(x)=\frac{\stackrel{~}{\psi }^\alpha \stackrel{~}{\psi }^\sigma }{\beta ^\alpha \beta ^\sigma }\eta _{\alpha \sigma }(x)$$
and this exhibits in a simple example the manner in which the metric can depend on the fields.$`\mathrm{}`$
###### EXAMPLE 6.3.
We look now at the more complicated situation for $`\psi =f(𝐚𝐱ct)`$ as in (6.22)-(6.23). Here $`f^{\prime \prime }/f=g`$ could be a fairly general function with argument $`𝐚𝐱ct`$ and in the equations $`_i^2\psi ^i=\alpha _ig\psi ^i`$ the function $`g_i`$ is considered as a function of $`x^i`$ with the other $`x^j`$ as parameters. Let $`\psi ^i`$ and $`\stackrel{~}{\psi }^i`$ be two solutions ($`i=0,1,2,3`$ say) and look at ($`\psi _i\psi ^i`$)
(6.29)
$$𝔈^i=\beta ^ix^i=\frac{1}{2}\psi ^i\frac{𝔉}{\psi ^i}𝔉^i;E^i=\psi ^i\frac{^2𝔉}{\psi _i^2}\frac{𝔉}{\psi ^i}$$
Recall $`\stackrel{~}{\psi }^i=𝔉/\psi ^i`$ and we can write, from Item 3 in Section 2, $`\varphi ^i=𝔉/(\psi ^i)^2=\stackrel{~}{\psi }^i/2\psi ^i`$ (although this will not be used here). In terms of the two fields $`\psi `$ and $`\stackrel{~}{\psi }^i`$ one has
(6.30)
$$𝔈^i=\frac{1}{2}\psi ^i\stackrel{~}{\psi }^i𝔉^i=\beta ^ix^i;𝔉^i=𝔉^i(\psi ^i,\stackrel{~}{\psi }^i,x^i,\beta ^i);$$
$$E^i=\psi ^i\frac{\stackrel{~}{\psi }^i}{\psi ^i}\stackrel{~}{\psi }^i;\beta ^idx^i=\frac{1}{2}E^id\psi ^i$$
In particular $`E^i`$ is expresed directly in terms of the fields $`\psi ^i`$ and $`\stackrel{~}{\psi }^i`$; no extraneous variables are explicit. Now $`\psi ^i`$ and $`\stackrel{~}{\psi }^i`$ are linearly independent solutions of $`_i^2\psi ^i=\alpha _ig\psi `$ but they are linked by a Wronskian $`W_i=(_x\psi ^i)\stackrel{~}{\psi }^i\psi ^i(_x\stackrel{~}{\psi }^i)=2\beta ^i`$ where $`\beta ^i`$ does not depend on $`x^i`$ (only perhaps on the other $`x^j`$). One can write now
(6.31)
$$_x\left(\frac{\stackrel{~}{\psi ^i}}{\psi ^i}\right)=\frac{W_i}{\psi _i^2}\stackrel{~}{\psi }^i=\psi ^i^x\frac{W_idx}{\psi _i^2}+c\psi ^i$$
Formally this suggests
(6.32)
$$\frac{\stackrel{~}{\psi }^i}{\psi ^i}=2\beta ^i^x\frac{dx}{\psi _i^2}+4\beta ^i\psi ^i^x\frac{dx}{\psi _i^3}+c$$
from which follows
(6.33)
$$E^i=\psi ^i\left[2\beta ^i^x\frac{dx^i}{\psi _i^2}+c+4\beta ^i\psi ^i^x\frac{dx^i}{\psi _i^3}\frac{2\beta ^i}{\psi ^i}\frac{dx^i}{d\psi ^i}\right]+$$
$$+2\beta ^i\psi ^i^x\frac{dx^i}{d\psi ^i}c\psi ^i=4\beta ^i\psi _i^2^x\frac{dx^i}{\psi _i^3}2\beta ^i\frac{E^i}{2\beta ^i}E^i=2\beta ^i\psi _i^2^x\frac{dx^i}{\psi _i^3}$$
Thus $`E^i`$ can be expressed entirely in terms of the field $`\psi ^i`$. $`\mathrm{}`$
One notes here that these arguments and results hold for any $`\psi ^\alpha ,V^\alpha `$ as in (6.5)-(6.7) so we state heuristically
###### THEOREM 6.2.
The objects $`E^\alpha `$ used in constructing the geometry can be expressed in terms of fields $`\psi ^\alpha `$ as in (6.34). |
warning/0507/gr-qc0507117.html | ar5iv | text | # 1 Introduction
## 1 Introduction
The story of searching for black holes is not completed yet. Although it is worthwhile emphasizing that about 200 massive and highly compact objects have been discovered so far which possess properties very similar to those of black holes. A black hole (BH) is an object which has an escape velocity equal to the speed of light in vacuum, $`c=300,000`$ km/s. Conception of black hole arose after discovery of Newton’s law of universal gravitation in 1687. In 1783 John Michell expressed an idea of dark stars having gravitational field so strong that even the light cannot escape outwards. The same idea was put forward in 1798 by Pierre-Simon Laplace. The existence of black holes is predicted in Einstein’s General Theory of Relativity (GTR). A characteristic size of a black hole is defined by the Schwarzschild radius (gravitational radius), $`r_g=2GM/c^2`$, where M is a mass of the object, c is the speed of light, $`G=6.6710^8\text{dyn}\text{cm}^2\text{g}^2`$ is the gravitational constant.
The gravitational radius is equal to
$$r_g=\{\begin{array}{ccccc}9\hfill & \text{mm}\hfill & \text{for}\hfill & \text{the Earth}\hfill & (M=610^{27}\text{g})\hfill \\ 30\hfill & \text{km}\hfill & \text{for}\hfill & 10M_{}\hfill & (M=210^{34}\text{g})\hfill \\ 40\hfill & \text{AU}\hfill & \text{for}\hfill & M=210^9M_{}\hfill & (M=410^{42}\text{g}),\hfill \end{array}$$
where $``$ is a symbol for the Sun, 1 Astronomical Unit (AU) is the Earth’s average distance from the Sun, $`1.510^{13}`$ cm.
The boundary of a black hole is identified with the event horizon where time is stopped for a distant observer. Therefore, all the events occuring under the event horizon are beyond the reach of the observer. The radius of the event horizon is equal to the gravitational radius for a non-rotating (Schwarzschild) black hole and is less than the gravitational radius for a rotating black hole. In this case the event horizon is located inside the black hole’s ergosphere where the vortical gravitational field is created. It is John Wheeler who called these objects "black holes"(BH) in 1968.
Properties of BH are described, for example, in as well as in a recent review by the same authors .It is worth mentioning that the event horizon is not a solid observable surface. It can be removed with the use of a proper reference frame. For example, there is no event horizon at all for an observer freely falling onto a BH. Such an observer could get into the BH and see the central singularity which contains all the compressed matter but he could transmit no information outward. Because of extraordinary properties of BH their existence in the Universe is doubted and has been debated for some decades. Some versions of the gravitational theory deny BH (see, e.g., ), the problem of searching for them getting more intriguing and interesting. Moreover, because of relativistic effect of time retardation near the event horizon, "contemporary"BH do not seem to have the event horizons completely formed. Astronomers consider these objects to be "virtual"BH with "virtual"event horizons.
In 1964 a pioneering work by Ya.B.Zel’dovich and E.E.Salpeter appeared where the possibility to observe black holes was predicted due to the powerful energy release in the process of non-spherical accretion of matter to the BH. In a theory of disk accretion to neutron stars (NS) and BH was developed which was very helpful in understanding the nature of compact X-ray sources discovered aboard UHURU satellite and identifying them as accreting NS and BH in stellar binary systems. Optical investigations of X-ray binary systems stimulated development of reliable methods for determining NS and BH masses . 3D gasdynamic models of gas flow in binary systems made mechanisms of accretion disk formation more clear . Models of advection-dominated disks around BH proposed in account for anomalously low luminosity of accreting BH in nuclei of normal galaxies and low-mass X-ray binary systems. Searching for stellar-mass BH being rather succesful, a certain progress is likely to have appeared recently in researching super-mass BH in galactic nuclei. The most convincing evidence for super-mass compact objects has been obtained recently while investigating "quiescent"galactic nuclei (see, for example, the recent Symposium results and the review ), although quasars and nuclei of active galaxies were considered to be the first super-mass BH candidates .
## 2 Methods for Searching Black Holes
Three types of BH are suggested to exist:
1. Stellar mass BH, $`M=(350)M_{}`$, which are formed at advanced stages of massive star evolution. If an evolved stellar core has got a mass $`M_c(1.21.4)M_{}`$ it yields a white dwarf, if its final mass is $`M_c<3M_{}`$ it produces a neutron star, and if the mass is $`M_c3M_{}`$ a BH is likely to be the final object.
2. Super-mass BH in galactic nuclei, $`M=10^610^9M_{}`$.
3. Primary BH formed at early stages of the evolution of the Universe.
As to intermediate-mass BH with masses $`M=(10^210^4)M_{}`$, their existence is still debated. Whether or not they can appear is not clear yet. Actually a number of stellar mass BH has been found to be nearly 20, and a number of super-mass BH in galactic nuclei around 200.
Two main problems are to be solved in the process of searching for BH:
1. Looking for massive, $`M>3M_{}`$, compact objects as possible BH candidates.
2. Looking for sufficient observational criteria for them being real BH.
Astronomical observations of BH are possible owing to X-ray halos around them occuring in non-spherical accretion of matter . But terrestrial atmosphere is opaque for X-rays. It is in 1962 that the first X-ray source beyond the solar system, Sco X-1, was discovered aboard an American Aerobee rocket in an experiment leaded by Riccardo Giacconi who was awarded the Nobel Prize in 2002.
Astronomical observations of BH involve three steps.
1. Analysing motions of "trial bodies"(stars, gas clouds, gas disks) in the BH gravitational fields provides estimates of BH masses. Characteristic distances being much greater than gravitational radii, Newton law may be used to make such estimates.
2. BH radii are measured by indirect approaches: through analyses of fast X-ray variability, X-ray line profiles, etc.
3. The most challenging problem is to look for observational evidence of the event horizon.
A final proof whether or not an object is a BH can be obtained if a reliable event horizon or an ergosphere is revealed (for a rotating BH). Special groundbased and space projects are planned to solve this significant problem. The main observational criteria of an accreting stellar mass BH are great mass, powerful X-ray radiation, and absence of X-ray pulses or 1st type X-ray bursts. Pulses or bursts are peculiar to accreting NS which have got observable surface and fast rotation. In a strong magnetic field ($`10^{12}`$ Gs) matter from the inner parts of the NS accretion disk is tunnelled via magnetic lines of force to the NS magnetic poles and, after having been collided with the NS surface, is heated up to temperatures over 10 million degrees, up to some hundred million degrees. Hot spots are formed in the areas of collision. Since the rotational axis is not coincident with the magnetic dipole axis, these hot X-ray spots on the NS surface are alternately visible to an Earth observer or shielded by the NS body (‘‘lighthouse effect’’). Thus a phenomenon of X-ray pulsar appears when a strictly periodic X-ray radiation comes out, periods lying between less than a second and some minutes (Fig. 1).
If the NS magnetic field is rather small (less than $`10^8`$ Gs) matter from the inner parts of the accretion disk is spreaded about and accumulated on the NS surface resulting in nuclear explosion of accumulated matter. Thus a phenomenon of the 1st type X-ray burster appears, i.e. short (for some seconds) and powerful X-ray bursts. So, X-ray pulses and the 1st type X-ray bursts are considered to justify that an accreting relativistic object has got the observable surface. Another evidence of such observable surface is when a relativistic object displays pulses in radio occuring due to ejection of charged relativistic particles in a strong magnetic field of a fast rotating NS. In this case, short (between milliseconds and seconds) and strictly periodic pulses of radio emission come out from the NS.
X-ray pulses, radio pulses or the 1st type X-ray bursts are not likely to occur on an accreting BH. What may be expected, however, is just X-ray irregular variability on time scales $`r_g/c10^310^4`$ Fig. 2.
Sometimes NS may be observed which display neither X-ray pulses, radio pulses, nor the 1st type X-ray bursts. Hence, their absence is a necessary but insufficient condition to identify a compact object with a real BH.
There are no so far sufficient observational criteria to select a BH. It is worthwhile mentioning, however, that all the necessary criteria taking into account GR-effects are fullfilled for all known BH candidates ( 200). So modern astronomers, with a kind of forced argumentation, call the BH candidates "black holes".
## 3 Stellar mass Black Holes
There are two types of X-ray binary systems which can contain BH (see the Catalog ):
1. Quasistationary X-ray binaries with massive hot companion stars.
2. Transient (flashing) X-ray binaries, i.e. X-ray novae, with low-mass cool companion stars.
About a thousand X-ray binary systems in the Milky Way and near galaxies have been discovered so far owing to specialized X-ray observatories launched into orbits around the Earth. Noticeable contribution into discovery and researches of X-ray binary systems was made by the Soviet and Russian Mir Kvant and Granat X-ray observatories under R.A.Sunyaev (see, for example, \[22, SChG91\]). Successful observations of BH candidates in a hard X-ray spectrum that is the most appropriate range to search for BHs have been carried out by the international X-ray and gamma observatory INTEGRAl launched by a Russian Proton carrier rocket in October 2002 . The Russian scientific co-leader of the project is R.A.Sunyaev.
A 3D gas-dynamic model of an X-ray binary system elaborated by A.A.Boyarchuk’s team is displayed in Fig. 2. Powerful X-ray radiation is produced in the inner parts of the accretion disk close to a BH.
Spectrum of optical radiation of X-ray Nova Oph1997 in a quiescent state is shown in Fig. 3.
Measuring the doppler shifts of numerous metallic absorption lines that correspond to a K-type star yields the radial velocity curve representing projected orbital velocity of the star on the line of sight. The radial velocity semi-amplitude of the optical star is about 400 km/s. Since radial velocities are measured within 1-3 km/s error, the radial velocity curve is highly reliable. While interpreting it, however, one should take into account that the optical star is not a material point, it has got considerable size and is pear-shaped, temperature distribution over its surface being rather complicated. A mathematical technique has been elaborated so far which allows for these effects to be taken into consideration .
Optical investigations of X-ray binary systems and measurements of BH masses were carried out by American, Canadian and British scientists (Charles, Cowley, Murdin, Hutchings, McClintock, Remillard, Hynes, Martin, Casares, Orosz, Bailyn, Filippenko, Shahbaz, Greiner, et al.) as well as Soviet and Russian researches (Pavlenko, Lyuty, Sokolov, Fabrika, Goransky, Kurochkin, Shugarov, et al.). The optical light curve of the X-ray binary system Cyg X-1, the most plausible candidate for BH, first obtained by Lyuty, Sunyaev and Cherepashchuk is shown in Fig. 1. The amplitude of optical variability due to optical star ellipticity effect gave inclination angle, i, between the orbital plane and the picture plane and allowed one of the earliest estimates of the BH mass to be obtained in Cyg X-1 system. A BH mass is calculated using the formula $`m_{BH}=f_{opt}(m)(1+q^1)^2\mathrm{sin}^3i`$ where $`q=m_{BH}/m_{opt}`$ is a ratio of the BH and optical star masses, $`f_{opt}(m)`$ is the mass function of the optical star determined over its radial velocity curve ($`m_{BH}>f_{opt}(m)`$). Parameters q and i are found from additional information: optical light curve analyses, rotational Doppler broadening of lines in the spectrum of the optical star, distance to the binary system, duration of an X-ray eclipse. The technique available is used to determine reliable BH masses having spectral and photometric observations of X-ray binary systems. Parameters for 18 X-ray binary systems with measured BH masses are given in Table 1. Fig. 4 displays masses of relativistic objects (NS and BH) versus companion masses in the binaries.
Companions of X-ray pulsars, 1st type X-ray bursters, and BH in binary systems are O-M type optical stars. Companions of radiopulsars are non-active NS, white dwarfs, and massive type B stars (we do not consider here radiopulsars with planets as companions). Radiopulsar masses are determined at a high level of accuracy taking into account relativistic effects in their orbital motions. Fig. 4 indicates that masses of relativistic objects do not depend on masses of their companions: both NS and BH can belong to binary systems with massive and low-mass companions. Orbital periods of X-ray binaries with BH are shown to lie in wide intervals between 0.17 days and 33.5 days. About a half of the systems have the mass function of the optical star more than $`3M_{}`$, i.e. the absolute upper limit for a BH mass predicted by General Relativity. In these cases relativistic objects may be considered to be BH candidates, their masses being over $`3M_{}`$. The mathematical technique available allows one to find BH and NS masses, together with their errors, from the mass function of the optical star (see Table 1). Masses of 19 NS are in the interval $`M_{NS}=(12)M_{}`$, a mean value of their mass being $`(1.35\pm 0.15)M_{}`$. Masses of 18 BH have been measured to lie in the interval $`M_{BH}=(416)M_{}`$, a mean value of their mass being $`(810)M_{}`$.
The more numerous a number of relativistic objects with measured masses becomes (19 NS and 18 BH in the Table 1), the stronger is the conviction that there is a systematic difference both in NS and BH masses and in their observational manifestations according to Einstein’s GR theory. For all objects with clearly observed surface (radiopulsars, X-ray pulsars, or 1st type X-ray bursters) their masses (of NS) do not exceed $`3M_{}`$ what is in full accordance with GR. At the same time, among 18 massive ($`M>3M_{}`$) binary X-ray sources studied (BH candidates) there are neither radiopulsars nor X-ray pulsars, nor 1st type X-ray bursters. Hence, in full accordance with GR predictions, massive ($`M>3M_{}`$) X-ray sources, BH candidates, do not reveal any observable surface. A great number of relativistic objects with measured masses (37) makes the conclusion rather reliable. It is an argument, though not a final proof, that 18 BH candidates with measured masses are real BH in the sense of GR. Thus, presense or absence of pulses or bursts is an observational manifestation crucial while defining whether or not an accreting object is a NS or a BH. Moreover, there are finer spectral distinctions (in the range 1-10 KeV) which indicate that NS have got surfaces observed whereas BH have not got .
Systems GRS1915+105, SAX J1819.3-2525, GRO J1655-40, and 1E1740.7-2942 called microquasars display relativistic collimated jets in X-ray bursts which have velocities $`v0.92`$ of the speed of light and plasma cloud motions which have apparent velocities in excess of the speed of light (apparent superluminal motions in the plane of the sky are due to the Special Theory of Relativity effects).
Recently interesting results have been obtained concerning rotation of stellar-mass BH. If an accretion disk around the BH rotates in the same direction as the BH itself, such a rotating disk penetrates much closer to the BH than it would in the case of a non-rotating BH. This is due to the fact that the radius of the final stable orbit for a rotating BH is less than for a non-rotating BH, $`3r_g`$. Hence, luminosity and temperature of the thermal component of X-rays emitted by rotating accreting BH are enhanced because of more powerful release of energy in the process of accretion. As a matter of fact, two transient X-ray binary systems with BH, microquasars GRS1915+105 and GRO J1655-40, display such enhanced characteristics and are likely to contain fast rotating BH.
Radii of stellar-mass BH may be restricted while analyzing observational results on fast variability of X-ray radiation. For example, system Cyg X-1 displays fast irregular X-ray intensity variability on a typical time scale $`\mathrm{\Delta }t`$ up to $`10^3`$ s. A typical size of the region near a BH emitting X-rays is, therefore, no more than $`r=c\mathrm{\Delta }t300`$ km/s $`10r_g`$. Observations of binary systems with BH have exhibited wide-ranging quasiperiodic oscillations (QPOs). Detailed findings on QPOs in X-ray binary systems with BH are given in the recent review . If high-frequency QPOs (typical frequencies between 41 Hz and 450 Hz) are associated with orbital motions of plasma condensations near a BH, the corresponding distancies are not more than some gravitational radii. High-frequency QPOs may be also connected with seismic oscillations of inner parts of the accretion disk as the GR theory predicts or they may be caused by relativistic dragging of the inertial reference frame near a fast rotating BH.
Fig. 5 displays NS and BH masses versus mass distributions, $`M_{CO}^f`$, of CO cores at the end of massive star evolution (Wolf-Rayet stars).
Different models of stellar wind flowing from WR are designated by parameters $`\alpha =1`$ and $`\alpha =2`$. Distribution of $`M_{CO}^f`$ in the interval $`(112)M_{}`$ is seen to be continuous while mass distribution of NS and BH resulting from the collapse of massive stellar CO-cores is bimodal with two maxima and a dip in the interval $`24M_{}`$. It seems like there is a deep reason which would prevent the formation of massive NS with $`M>2M_{}`$ and low-mass BH with $`M<4M_{}`$ in binary systems. Such an "avoidance zone"for relativistic objects may be shown to be due to other reasons than observational selection. If the conclusion is confirmed with more observational data, a more serious interpretation will be needed.
## 4 Supermassive BH in Galactic Nuclei
Most galaxies have got compact condensations of stars and gas in their centers which are called "nuclei". Usually the cores are well visible in spiral galaxies and hardly discernible in irregular ones. Among all galaxies a relatively small group may be singled out ( 1 per cent) which involves galaxies with active nuclei: Seyfert galaxies, radio galaxies, BL Lac galaxies, and quasars. The quasars are the most powerful sources of stationary radiation in the Universe. Their total luminosity reaches $`10^{47}`$ erg/s, which is three orders of magnitude more than that of a host galaxy. Intense non-stationary processes occuring in active nuclei result in variability of optical radiation on time scales of days to many years. Spectra of these active nuclei exhibit strong and often wide emission lines of hydrogen, helium and other elements. Many nuclei of active galaxies are observed to have strongly collimated jets of matter moving with relativistic velocities. A galactic nucleus is currently considered to be a supermassive BH with accretion of stellar matter and gas .
To determine BH masses in galactic nuclei a hypothesis is used according to which the gravitational field of a central object controls the motion of gas and stars near the nucleus . As was mentioned above, Newton’s law can be used since $`r>r_g`$. In this case the velocity $`v`$ of a star or a gas cloud depends on the distance $`r`$ to the center of a galaxy as $`v^2r^1`$. Hence the BH mass in the nucleus can be estimated as $`M_{BH}=\eta v^2`$ r/G, where $`\eta =13`$ depending on a kinematic model of body motion around the galactic center (for circular motion, $`\eta =1`$).
It is possible in many cases to see the moving gas directly, for our Galaxy even individual stars, near the galactic center owing to modern observational facilities (the Hubble Space Telescope, very large groundbased telescopes provided with techniques for compensation of atmospheric distortions, intercontinental interferometers, etc.). Therefore the BH masses are determined unequivocally using the formula given.
If a disk of gas and dust surrounding the galactic center cannot be seen and its rotation cannot be investigated another method is applied based on statistical researches of stellar kinematics in the central parts of the galaxy which is defined by the BH gravitational influence.
BH masses in active galactic nuclei with observed strong and wide emission lines can be determined using the formula given. Velocities, $`v`$, of gas clouds near the nucleus which are responsible for a wide component of emission line profiles can be estimated with the help of the Doppler semi-amplitude of this wide line component. The distance, $`r`$, of gaseous clouds to the nucleus can be estimated by two ways: by means of a photoionization model of the nucleus or by time delay, $`\mathrm{\Delta }t`$, which reveals in the fast variability of a wide emission line component with regard to the variability of the continuous spectrum, $`rc\mathrm{\Delta }t`$ (so called reverberation mapping). The time delay effect was discovered by Cherepashchuk and Lyutyi in 1973 . It was mentioned there that the time delay of the line variability with regard to the continuous spectrum is equal to the time needed for ionizing radiation to cover the distance from the galactic center to the gaseous clouds emitting in the line. Thus an independent estimate of the distance can be obtained and the mass of an active galaxy nucleus can be reliably determined.
The first method to estimate BH masses in active galactic nuclei suggested that the bolometric luminosity of the nucleus should be close to the Eddington limit . Such estimates give the following values for quasar nuclei masses: $`M>10^8M_{}`$.
So far there are some dozens of BH masses in active galactic nuclei estimated by means of the time delay effect according to which emission line variability lags behind continuum .
‘‘Normal galaxies’’ have got nuclei which are characterized by rather faint optical activity as compared to their stellar constituent. In such galaxies stars and gas moving near the nucleus can be observed directly what permits the most exact and model-independent mass estimates of supermassive BH to be obtained. Recently disks of gas and dust spanning some dozens or hundreds of parsecs around the nuclei of many galaxies and rotating according to Newton’s law were discovered aboard the Hubble space telescope with high angular resolution (see and references therein). To check the validity of the Keplerian rotation law for a disk ($`vr^{1/2}`$) and to obtain the inclination angle $`i`$ between the disk axis and the line of sight Doppler shifts are investigated in emission lines using the projection of the near-nucleus region of the disk onto the picture plane. Then the mass in the volume with radius $`r`$ is estimated unequivocally. Since the near-nucleus region in the galaxy may be observed directly the mass-to-luminosity ratio, $`M(r)/L(r)`$, can be estimated and compared with the corresponding value for external parts of the galaxy, $`M/L1÷10`$, where $`M`$ and $`L`$ are the solar mass and luminosity respectively. The first galaxy to have been used for determining the mass of the central BH using the near-nucleus disk of gas and dust with a luminous and stretched jet was M87 . The mass of the central BH is $`(3.2\pm 0.9)10^9M_{}`$, the mass-to-luminosity ratio is $`M/L>110`$. If the central mass was due to a dense cluster of ordinary stars rather than a BH, the nucleus of M87 would be dozens of times brighter than what is actually observed. An average density of dark matter in the nucleus of M87 is estimated to be $`10^7M_{}/\text{pc}^3`$ whereas star density in external parts of the galaxy is $`0.5M_{}/\text{pc}^3`$ and in the most dense stellar clusters $`10^5M_{}/\text{pc}^3`$. All these findings allow one to believe with a good reason that there is a supermassive BH in the nucleus of M87 (three billion solar masses) which undergoes accretion of matter causing many aspects of M87 activity including formation of a relativistic jet. A number of mass estimates obtained so far for supermassive BH by means of researching gas and star kinematics near galactic nuclei reaches many dozens (see, for example, review ). Table 2 gives some results on determining BH masses in galactic nuclei.
Outstanding results for estimating BH masses in galactic nuclei have been obtained recently while researching compact maser sources in near-nucleus molecular disks by means of intercontinental radioastronomy (see review by Moran et al. and references therein). Observations of NGC4258 galactic nucleus revealed 17 compact maser sources emitting very narrow $`H_2O`$ lines and being located in a disk-like envelope with a radius of $`10^{17}`$ cm which is seen nearly edge-on. Velocities of maser sources are distributed according to the Keplerian law. The mass of a central BH is $`3.910^7M_{}`$. This method has been used so far to measure masses of about a ten BHs in nuclei of galaxies (see reviews and references therein).
## 5 A Supermassive Black Hole in the Nucleus of Our Galaxy
The most convincing evidence for a supermassive BH have been obtained recently while researching motions of individual stars in the closest surroundings of SgrA\* source, the center of our Galaxy. Beginning from the 90s of the past century the motions of individual stars have been investigated in the picture plane near the center of our Galaxy . Observations are carried out in the IR range using special techniques for compensation of atmospheric distortions of the image (the Galactic Center is hidden from optical view by thick layers of gas and dust). The stars near the Galactic Center are found to shift considerably, their velocities being the higher the closer to the Center. Recently R.Schoedel et al. constructed an orbit of one of the closest stars to the Galactic Center (SO2) (see Fig. 6).
The star SO2 has 15.2-year orbit with an eccentricity of 0.87 and a semi-major axis of $`4.6210^3`$ pc ($`20000r_g`$). Kepler’s third law yields, for a BH mass, $`(3.7\pm 1)10^6M_{}`$. The dark gravitating matter density in the field measured reaches $`10^{17}M_{}/\text{pc}^3`$ while a typical dynamical break-up time of a supposed cluster of individual dark bodies in the galactic nucleus (due to collective collisions) is estimated to be $`10^5`$ years, the age of the Galaxy being $`10^{10}`$ years. This argues strongly that the massive compact object in the center of the Galaxy forms a whole dark body rather than a cluster of individual low-mass objects. Moreover, orbits of eight individual stars near the galactic center were measured lately: SO-16, SO-19, SO-20, SO-1, SO-2, SO-3, SO-4, SO-5. The BH mass in the Galactic nucleus is estimated to be $`(4\pm 0.3)10^6M_{}`$, it is situated at the dynamic center of the Galaxy within $`\pm 10^3`$ arcsec. The BH proper motion is $`(0.8\pm 0.7)10^3\text{s}\text{year}^3`$ what is actually zero within the limits of error. These findings argue strongly in favour of Gurevich’s idea that supermassive BH form in galactic nuclei due to accretion of baryonic matter which falls into potential wells in the center of galactic halos of dark matter. The SO-16 star approaches the BH as close as 90 AU ($`1700r_g`$) while moving in the orbit around.
## 6 Observational Restrictions upon Radii of Black Holes in Galactic Nuclei
According to X-ray image data with resolution $`0.^{\prime \prime }5`$ obtained by the Chandra observatory the center of the Galaxy is shown to emit variable X-rays. On the time scale of a year X-ray luminosity changes between $`210^{33}`$ and $`10^{35}`$ erg/s, the galactic nucleus displaying rapid variability (as high as 5 times on the scales $`t_{min}10`$ min) . Hence the size of the region emitting in X-rays, $`rct_{min}`$, is $`20r_g`$. The center of the Galaxy shows large variations in IR flux density, a factor of 2 over 40 min, as was discovered with the W. M. Keck II 10-meter telescope . This variability implies that the size of the IR emitting field does not exceed $`80r_g`$. IR luminosity of the galactic center is $`10^{34}`$ erg/s at 3.8 $`\mu `$m. The variable IR source is coincident with the center of the Galaxy to within $`610^3`$ arcsec and does not move, its velocity being at least $`v<300`$ km/s, whereas the stars near the galactic center have got constant luminosity and move around the center with velocities of thousands of km/s.
Therefore, observations indicate that there is a massive compact object with a mass of $`410^6M_{}`$ and a radius less than 20 gravitational radii in the center of our Galaxy, the parameters arguing in favour of the object being a supermassive BH.
Direct measuring a supermassive BH radius in the center of the Galaxy (as well as in the centers of nearby galaxies) will be possible after launching space interferometers: an X-ray interferometer with $`10^7`$ arcsec resolution and the RadioAstron interferometer with $`10^6`$ arcsec resolution in the radio range. Angular sizes of the supermassive BH in the centers of our Galaxy and the Andromeda Galaxy are $`710^6`$ and $`310^6`$ arcsec, respectively. Launching these interferometers will allow not only supermassive BH radii to be measured but also physical phenomena to be observed connected with plasma moving near the event horizon. Such experiments are likely to provide sufficient criteria to select BH and to prove, once and for all, their existence in the Universe. Using contemporary intercontinental interferometry techniques made it possible to research, in the millimeter range, the formation of jet in the inner parts of the M87 galaxy as well as to confine directly a value of the supermassive BH radius within $`r<30100`$ gravitational radii .
In addition, iron $`K_\alpha `$ emission line profile in X-rays at 6.4 keV in spectra of active galactic nuclei observed aboard X-ray observatories ASCA, CHANDRA, and XMM with high spectral resolution also restricts strongly the values of BH radii. Relativistic effects near the event horizon of the central BH result in the redshift of a spectral line, its specific asymmetric profile and a huge width (up to 100,000 km/s) thus providing bounds on the supermassive BH radius in the center of a galaxy. For instance, in case of MCG-6-30-15 galactic nucleus, analysis of the wide spectral component of Fe XXV X-ray line profile indicates that the inner edge of the accretion disk is located less than $`3r_g`$ from the central supermassive BH which seems to rotate (see Fig. 7).
## 7 Demography of Supermassive Black Holes
A number of supermassive BH with their masses measured approaches now 200. The radii estimated are available for many of them: $`r<(10100)r_g`$. Hence a new field in astrophysics is intensively developing now, the demography of BH. The basic results achieved in this field are briefly stated below.
1. Supermassive BH in galactic nuclei have masses correlated with those of galactic bulges, spherical condensations of old low-mass stars near the nucleus with the large dispersion of velocities :
$`M_{BH}=0.0012M_{bulge}^{0.95\pm 0.05}`$.
2. There is the correlation between supermassive BH masses and velocity dispersion of stars, $`\sigma `$, belonging to the bulge : $`M_{BH}\sigma _{bulge}^4`$.
3. Masses of supermassive BH are correlated with linear rotation velocities of galaxies in the range of a costant value of the velocity. Since a linear rotation velocity of a galaxy far away from its center is mainly due to the gravitational pull of the galactic halo consisting of dark matter, the mass of the central supermassive BH should be correlated with the mass of the galactic halo : $`M_{BH}M_{halo}^{1.27}`$. The result obtained is an important evidence in favour of Gurevich’s model .
## 8 Conclusion
The actual state of the problem connected with searching for both stellar-mass BH and supermassive BH in the galactic nuclei has been described. We did not touch upon the problem concerning searches for primary BH because of the lack of observational data and ambiguity revealed in their interpretation. Primary BH may as well exist among isolated stellar-mass BH. It is worthwhile mentioning that, using gravitational microlensing effect , the masses of two isolated BH have been measured so far: $`m_{BH}6M_{}`$ (the brightness of a distant background star being strenghened because of the influence of a foreground object playing a part of a "gravitational lense the duration of its strengthened brightness is proportional to a square root of the gravitational lense mass). The problem for searching BH of intermediate mass ($`m_{BH}=10^2÷10^4M_{}`$) was not considered either because there were not convincing findings in this field. The review on intermediate mass BH located both in galaxies and in stellar clusters is available in .
It is important that the problem with searching for BH is actually supported with a firm observational basis and a number of BH being discovered is gradually increasing (around 200 at the present moment). It should be especially emphasized that all necessary constraints imposed upon BH manifestations by Einstein’s General Relativity are fulfilled as it follows from observational data. This strengthens considerably our confidence in actual existence of BH in the Universe.
The main task to be solved in the decade to come is to find sufficient criteria for BH candidates being real BH. The following experiments may be expected to help in solving this problem:
1. Using space interferometers with $`10^610^7`$ arcsec angular resolution and direct observations of matter moving close to event horizons of supermassive BH in nuclei of our and nearest galaxies.
2. Searches and investigations of gravitational wave bursts from BH coalescence in binary systems with the use of laser gravitational wave interferometry (LIGO, VIRGO, LISA, etc.).
3. Detection and researches of radiopulsar motion in binary systems with BH (among 1000 pulsars one pulsar is anticipated to be paired with a BH, about 1500 pulsars being known so far).
4. Detailed investigations of spectra, intensity, polarization and variability of X-ray and gamma-radiation from accreting BH with the use of new generation orbital observatories.
5. Observations and interpretations of gravitational microlensing effects in galactic nuclei caused by closer galaxies (gravitational lenses).
6. Routine storage of evidence concerning masses of black holes and neutron stars (NS) as well as statistical comparison between observational manifestations of BH and NS.
The work is supported by the Russian Fund for Fundamental Research (grant 02-02-17524).
TABLE 1. Parameters of Binary Systems with Black Holes
| System | Opt. Star | $`P_{orb}`$ | $`f_{opt}\left(M\right)`$ | $`M_{BH}`$ | $`M_{opt}`$ | $`V_{pec}`$ | Note |
| --- | --- | --- | --- | --- | --- | --- | --- |
| | Spectrum | (days) | $`\left(M_{}\right)`$ | $`\left(M_{}\right)`$ | $`\left(M_{}\right)`$ | (km/s) | |
| Cyg X-1 | O 9.7 Iab | 5.6 | 0.24 | 16 | 33 | 49 | stat. |
| V 1357 Cyg | | | $`\pm 0.01`$ | $`\pm 5`$ | $`\pm 9`$ | $`\pm 14`$ | |
| LMC X-3 | B3 Ve | 1.7 | 2.3 | 9 | 6 | - | stat. |
| | | | $`\pm 0.3`$ | $`\pm 2`$ | $`\pm 2`$ | | |
| LMC X-1 | O(7-9) III | 4.2 | 0.14 | 7 | 22 | - | stat. |
| | | | $`\pm 0.05`$ | $`\pm 3`$ | $`\pm 4`$ | | |
| SS 433 | $``$ A7 Ib | 13.1 | $`1.3`$ | 11 | 19 | - | stat. |
| | | | | $`\pm 5`$ | $`\pm 7`$ | | |
| A0 620-00 | K5 V | 0.3 | 2.91 | 10 | 0.6 | -15 | trans. |
| (V 616 Mon) | | | $`\pm 0.08`$ | $`\pm 5`$ | $`\pm 0.1`$ | $`\pm 5`$ | |
| GS 2023+338 | K0 IV | 6.5 | 6.08 | 12 | 0.7 | 8.5 | trans. |
| (V 444 Cyg) | | | $`\pm 0.06`$ | $`\pm 2`$ | $`\pm 0.1`$ | $`\pm 2.2`$ | |
| GRS 1124-68 | K2 V | 0.4 | 3.01 | 6 | 0.8 | 26 | trans. |
| (GU Mus) | | | $`\pm 0.15`$ | (+5,-2) | $`\pm 0.1`$ | $`\pm 5`$ | |
| GS 2000+25 | K5 V | 0.3 | 4.97 | 10 | 0.5 | - | trans. |
| (QZ Vul) | | | $`\pm 0.10`$ | $`\pm 4`$ | $`\pm 0.1`$ | | |
| GRO J0422+32 | M2 V | 0.2 | 1.13 | 10 | 0.4 | - | trans. |
| (V 518 Per) | | | $`\pm 0.09`$ | $`\pm 5`$ | $`\pm 0.1`$ | | |
| GRO J1655-40 | F5 IV | 2.6 | 2.73 | 6.3 | 2.4 | -114 | trans. |
| (XN Sco 1994) | | | $`\pm 0.09`$ | $`\pm 0.5`$ | $`\pm 0.4`$ | $`\pm 19`$ | |
| H 1705-250 | K5 V | 0.5 | 4.86 | $`6\pm 1`$ | 0.4 | 38 | trans. |
| (V 2107 Oph) | | | $`\pm 0.13`$ | | $`\pm 0.1`$ | $`\pm 20`$ | |
| 4U 1543-47 | A2 V | 1.1 | 0.22 | 4.0- | $`2.5`$ | - | trans. |
| (HL Lup) | | | $`\pm 0.02`$ | 6.7 | | | |
| GRS 1009-45 | (K6-M0) V | 0.3 | 3.17 | 3.6- | 0.5- | - | trans. |
| (MM Vel) | | | $`\pm 0.12`$ | 4.7 | 0.7 | | |
| SAX J1819.3-2525 | B9 III | 2.8 | 2.74 | 9.61 | 6.53 | - | trans. |
| (V 4641 Sgr) | | | $`\pm 0.12`$ | (+2.08- | (+1.6- | | |
| | | | | 0.88) | 1.03) | | |
| XTE 1118+480 | (K7-M0)V | 0.17 | 6.0 | | 0.09- | 126 | trans. |
| | | | -7.7 | | 0.5 | | |
| GRS 1915+105 | (K-M)III | 33.5 | 9.5 | $`14\pm 4`$ | 1.2 | - | trans. |
| | | | $`\pm 3.0`$ | | $`\pm 0.2`$ | | |
| XTE J1550-564 | $`K3`$ | 1.54 | 6.86 | 8.36- | $`1`$ | - | trans. |
| | | | $`\pm 0.71`$ | 10.76 | | | |
| XTE J1859+226 | $`K7`$ | 0.38 | 7.4 | 7.6- | $`0.7`$ | - | trans. |
| | | | $`\pm 1.1`$ | 12.0 | | | |
Note: $`P_{orb}`$ stands here for an orbital period, $`f_{opt}(M)=\frac{M_{BH}^3\mathrm{sin}^3i}{(M_{BH}+M_{opt})^2}`$ is a mass function of an optical star, $`M_{BH}`$, $`M_{opt}`$ are masses of black holes and optical companions, respectively, $`V_{pec}`$ is a peculiar velocity of the center of gravity of a binary system. |
warning/0507/astro-ph0507482.html | ar5iv | text | # Perturbations of the Quintom Models of Dark Energy and the Effects on Observations
## I Introduction
In 1998 two groups Riess98 ; Perl99 have independently discovered the accelerating expansion of our current universe based on the analysis of Type Ia Supernovae (SN) observations of the redshift-distance relations. In the framework of Friedmann-Robertson-Walker (FRW) cosmology, the acceleration has been attributed to the mysterious source dubbed dark energy. The simplest candidate for dark energy is a small positive cosmological constant, but it suffers from the difficulties associated with the fine tuning and the coincidence problem SW89 ; ZWS99 . The most popular alternative to the cosmological constant is the model of rolling scalar field–quintessence pquint ; quint . In most cases the quintessence equation of state (EOS) $`w`$ changes slowly with time and can be well approximated with a constant $`w`$ with $`w1`$ HWDCS99 ; WCOS2000 . In the early probes of new physics, cosmologists have assumed a cosmological constant as the new component CPT92 ; KT95 ; OS95 ; Riess98 ; Perl99 and later fitted directly to the dynamical quintessence models CDF96 ; ZWS99 , or used a constant $`w`$ CDS98 ; GSST98 ; HWDCS99 ; WCOS2000 where $`w`$ was restricted in the region of $`w1`$. In Ref.phant99 the author firstly extended the fitting of dark energy to include $`w<1`$ and found some mild preferences. The author constructed a toy model of rolling scalar field with a negative kinetic term and called it phantom phant99 . The model of phantom has some theoretical problems Phtproblms and there have been many attempts towards resolving them SPhtproblms .
The accumulation of the observational data Bennett03 ; Tonry03 ; Knop03 ; Tegmark03 ; Riess04 ; Allen04 has opened a robust window for probing the more detailed behaviors of dark energy. There have been many studies in reconstructing the evolution of its energy density wangy04 or equation of state sahni00 ; Huterer ; dd03 as a function of the redshift. Various parametrizations of $`w`$ as well as dark energy models have also been considered to fit directly to the observational data (e.g. DES ; DES1 ; DES2 ; sahni ; cooray ; ex ; ex1 ; quintom ; seljak04 ; tao05 ). Based on the fact that current observations cannot exclude dark energy models with the equation of state getting across $`1`$ during evolution with the redshift, we proposed a model dubbed quintomquintom . The model of quintom is a new scenario in the sense that the conventional quintessence or phantom models cannot realize the crossing of the cosmological boundary. Along this line the author in Ref.Vikman04 has demonstrated that in the framework of general relativity the model of k-essence kessence , where the scalar field of dark energy has non-canonical kinetic terms, cannot realize such a crossing behavior. A toy model with two rolling scalar fields which have opposite kinetic energy terms can easily realize the transition and it can be regarded as the simplest quintom model quintom ; quintom1 . Recently a single field quintom model was proposed in MFZ05 by adding higher derivative operators in the Lagrangian. In the simplest case such a model is equivalent to the two-field case as proposed in quintom . In addition, the quintom model of dark energy are different from the quintessence or the phantom in the determination of the evolution and the fate of the universe. Due to its distinctive properties, the quintom model with oscillating equation of state across $`1`$ can lead to the oscillations of the Hubble constant and a new scenairo of recurrent universe FLPZ , which to some extent unifies the early inflation Pinflation and the current acceleration of the universe. Recently there have been a lot of interests in the phenomenological studies relevant to quintom models in the literature wei ; zhang ; hu ; xia ; michael ; li ; xfzhang ; relevnt ; relevnt1 ; NOT05 ; 0504518 .
Current supernovae data alone, which make the only direct detection of dark energy, seem to favor a quintom-like model at around 2-$`\sigma `$ level cooray ; sahni ; quintom . The quintom model is also mildly favored in the combined analysis with the cosmic microwave background (CMB), large scale structure (LSS) and supernovae data ex ; ex1 . However, when some other observational data sets (such as the new observational data based on Chandra measurements of the X-ray gas mass fraction in 26 X-ray luminous galaxy clusters Allen04 or the recent new constraints from the bias and Ly$`\alpha `$ forest of the Sloan Digital Sky Survey (SDSS)) have been taken into account the situation changes and the preference for quintom-like dark energy models becomes weak quintom ; seljak04 . However the previous fittings in the literature on quintom-like dark energy models have either fully or partially neglected the perturbations, which in some sense do not describe the realistic models with EOS across $`1`$ and will lead to some bias in the fittings. The aim of this paper is to develop a self-consistent way to include the perturbations of quintom in light of the observations. We will present a simple new method to show that conventional single perfect fluid and k-essence dark energy models cannot act as quintom, which is due to the singularities and classical instabilities of perturbations. Based on the realistic quintom models in this paper we will provide one way to include the perturbations for dark energy models with parametrized equation of state across $`1`$. Compared with those assuming no dark energy perturbations, we find that when including the perturbations the parameter space which allows the equation of state to get across $`1`$ will be enlarged in general.
This paper is organized as follows: in section II we discuss the difficulty of quintom model buildings and provide a new proof regarding the impossibility of single perfect fluid and k-essence model as quintom, then present some viable quintom models; in section III we study in detail the perturbations of the quintom models; in section IV we investigate the possible signatures of quintom models of scalar fields and the effects of quintom perturbations on the observations; in section V we provide one way to include the perturbations for models of dark energy with a parametrized equation of state across $`1`$; we conclude in section VI.
## II Quintom Model Building
### II.1 Difficulties of Quintom Model Buildings
We start with a brief overview on the arguments against the possibility of realizing the quintom scenario with a single fluid or a single scalar field in the conventional framework.
Consider firstly a single perfect fluid, the energy-momentum tensor has the conventional form,
$$T_{\mu \nu }=Pg_{\mu \nu }+(\rho +P)u_\mu u_\nu ,$$
(1)
where $`\rho `$ and $`P`$ are proper energy density and pressure, $`u_\mu `$ is 4-velocity with $`u^\mu u_\mu =1`$. The energy density and the pressure of the fluid can be parametrized as jackiw
$`\rho `$ $`=`$ $`f(n),`$
$`P`$ $`=`$ $`nf^{}(n)f(n),`$ (2)
where $`f(n)`$ is a positive function of $`n`$. The introduced variable $`n`$ can be identified with the number density and the prime represents derivative with $`n`$. The equations of motion are just the covariant conservation equations of the momentum tensor, $`_\mu T^{\mu \nu }=0`$. In spatially flat FRW spacetime
$$ds^2=a^2(\tau )(d\tau ^2dx^idx_i),$$
(3)
there is only one equation,
$$\dot{\rho }+3(\rho +P)=0,$$
(4)
where the dot is the derivative with the conformal time $`\tau `$ and $`=\dot{a}/a`$. Combining Eq.(4) with Eq. (II.1), we get
$$f^{}(n)(\dot{n}+3n)=0.$$
(5)
Since $`f^{}(n)`$ does not vanish everywhere (otherwise it corresponds to the cosmological constant), one has the conservation equation of the number density $`\dot{n}+3n=0`$. In the expanding universe, $`n`$ will decrease monotonically with time.
In the following we will demonstrate that the system suffers from the problem of singularity and classical instability when the equation of state of the perfect fluid crosses the boundary of $`1`$. Let us assume that the system crosses $`1`$ at the point of $`n=n_00`$. At this point $`\rho (n_0)+P(n_0)=0`$, $`f^{}(n_0)=0`$ and $`f^{}(n)`$ will change the sign after the crossing. So, in the neighborhood of $`n_0`$, we can expand $`f^{}(n)`$ in terms of $`(nn_0)`$. The adiabatic sound speed square in this neighborhood is
$$c_s^2\frac{dP}{d\rho }=\frac{nf^{\prime \prime }(n)}{f^{}(n)}\frac{n}{nn_0}.$$
(6)
We can see that $`c_s^2`$ is singular at the crossing point. Moreover, $`c_s^2`$ is negative in the region of $`n<n_0`$. And when $`n`$ approaches $`n_0`$ from this side, it will approach $`\mathrm{}`$. A negative $`c_s^2`$ will induce a serious classical instability to the system, the perturbations on small scales will increase quickly with time and the late time history of the structure formations will get significantly modified. This will inevitably lead to the fact that such models can never be compatible with the observations relevant to structure formations, such as CMB and LSS.
As shown above it is impossible to realize the quintom scenario with a single perfect fluid, now we turn to consider the model of scalar field. The general model of dark energy with a single scalar field and an arbitrary function of its first derivative in the Lagrangian was proposed in Ref. kessence and named as k-essence. Its Lagrangian usually has a non-canonical form
$$=P(\varphi ,X),$$
(7)
with
$$X\frac{1}{2}_\mu \varphi ^\mu \varphi .$$
(8)
The energy-momentum tensor of this system has the same form as that of the single perfect fluid Eq. (1), where
$`\rho =2XP,_XP,`$
$`u_\mu ={\displaystyle \frac{_\mu \varphi }{\sqrt{2X}}}.`$ (9)
Let’s see under what conditions the system will be able to cross the barrier of $`w=1`$. In order to do that, one requires $`\rho +P`$ to vanish at a point of $`(\varphi _0,X_0)`$ and change the sign after the crossing. This can only be achieved by requiring $`P,_X(\varphi _0,X_0)=0`$ and $`P,_X`$ has different signs before and after the crossing since $`X`$ cannot be negative. The covariant conservation law of the energy-momentum tensor gives the equation of motion,
$$(P,_Xg^{\mu \nu }+P,_{XX}^\mu \varphi ^\nu \varphi )_\mu _\nu \varphi +\rho ,_\varphi =0.$$
(10)
From this equation, we obtain the equation for the background field
$$\rho ,_X[\ddot{\varphi }+(3c_{sk}^21)\dot{\varphi }]+a^2\rho ,_\varphi =0,$$
(11)
and the perturbation to the first order (neglecting the metric perturbations for the time being):
$$\ddot{u}+[c_{sk}^2^2\ddot{z}/z3c_{sk}^2(\dot{}^2)]u=0,$$
(12)
where
$$uaz\frac{\delta \varphi }{\dot{\varphi }},z\sqrt{\dot{\varphi }^2|\rho ,_X|},$$
(13)
and the effective sound speed is given by
$$c_{sk}^2\frac{P,_X}{\rho ,_X}.$$
(14)
This sound speed $`c_{sk}^2`$ is often used in describing the perturbations of the scalar fields instead of the isentropic sound speed, which behave differently due to the intrinsic properties of the scalar fields KS84 . For a conventional quintessence or phantom field, $`c_{sk}^21`$. The dispersion relation from Eq. (12) is
$$\omega ^2=c_{sk}^2k^2\ddot{z}/z3c_{sk}^2(\dot{}^2).$$
(15)
One of the conditions for the stability of k-essence perturbations is that $`c_{sk}^2`$ must be positive kessence . This requires that $`\rho ,_X`$ has the same behavior as that of $`P,_X`$, i.e., it must vanish at the crossing point and change the sign after the crossing.
Similar to the analysis in the case of single fluid, we can see that $`\ddot{z}/z`$ diverges at the point $`(\varphi _0,X_0)`$. This singularity is unavoidable in the perturbation equation and the physical quantities describing the fluctuations are not well defined. Generally, $`\ddot{z}`$ does not vanish at the crossing point, hence there exists a region in which $`\omega ^2<0`$ and the perturbation is unstable. Furthermore, the canonical momentum defined by the Lagrangian (7) is
$$\mathrm{\Pi }=\frac{P}{\dot{\varphi }}=P,_X\frac{\dot{\varphi }}{a^2}.$$
(16)
Its derivative with respect to $`\dot{\varphi }`$,
$$\frac{\mathrm{\Pi }}{\dot{\varphi }}=\frac{\rho ,_X}{a^2},$$
(17)
vanishes at the point of crossing. This shows that $`\dot{\varphi }`$ is not a single valued function of the momentum $`\mathrm{\Pi }`$ and we cannot get a canonical Hamiltonian transformed from the Lagrangian unambiguously susskind . The theory cannot be quantized in a canonical way. Hence we have shown that the conventional k-essence model cannot give rise to $`w`$ across $`1`$. A different proof is given in Ref. Vikman04 .
We should stress again that in realistic quintom model buildings one must consider the aspects of perturbations, where there are often dangerous instabilities in the conventional case. The concordance cosmology is based on the precise observations where many of them are tightly connected to the growth of perturbations and we must ensure the stability of perturbations. If we start with parametrizations of the scale factorbarrow or EOS to construct quintom models, it can be realized arbitrarily if we do not consider the stability of perturbations. On the other hand when we start from scalar fields and use some phenomenological parametrizations it is in some sense very easy to resemble fluid behavior in the background evolutions. However the stability of perturbations must be considered.
### II.2 Some Viable Quintom Models
As we demonstrated above in the conventional cases with a single fluid or a k-essence one cannot realize a viable model of quintom, we need to introduce extra degrees of freedom to realize the transition of $`w`$ across $`1`$. One of the possibilities is a system including two fluids with one being $`w>1`$ and another $`w<1`$. Specifically, consider a model which consists of two Chaplygin gases CGAS with $`P_1=\lambda _1/\rho _1`$ and $`P_2=\lambda _2/\rho _2`$, in which $`\lambda _1`$ and $`\lambda _2`$ are positive constants. If $`\rho _1^2>\lambda _1`$ and $`\rho _2^2<\lambda _2`$, one has $`0>w_1>1`$ and $`w_2<1`$. This system will cross the boundary of $`1`$ at some time because $`\rho _2`$ is always increasing and $`\rho _1`$ decreasing. The sound speed squares are $`c_{si}^2=w_i>0`$ with $`i=1,2`$, hence the system will be free of the difficulties associated with the singularity and the classical instability which exist in the model of a single fluid. Furthermore, the final state of this system will be characterized by $`w=1`$, the universe will approach the de Sitter space in the far future. In such a scenario there will be no big rip. In the framework of the field theory, the simple way to introduce the extra degree of freedom for the quintom model is the double scalar fields model with one being quintessence-like and one phantom-like. We should point out that when adding extra degrees of freedom in the above way, this does not help solve the cosmological constant problem and nor can it help solve the coincidence problem, since for the component where $`w<1`$ it cannot have the property of tracking behavior and has to be fine tuned <sup>1</sup><sup>1</sup>1For the k-essence field where $`w<1`$, it also needs to be fine tuned.. The above way of introducing more components provides the simplest possibility of quintom model building.
There is another possibility of introducing the extra degrees of freedom for the realization of the transition from the quintessence phase to the phantom phase. This is the model proposed in Ref. MFZ05 by introducing higher derivative operators to the Lagrangian. Specifically in MFZ05 we considered a model with the Lagrangian
$$=\frac{1}{2}_\mu \varphi ^\mu \varphi +\frac{c}{2M^2}\mathrm{}\varphi \mathrm{}\varphi V(\varphi ),$$
(18)
where $`\mathrm{}_\mu ^\mu `$ is the d’Alembertian operator. The term related to the d’Alembertian operator is absent in the quintessence, phantom and the k-essence model, which is the key to make the model possible for $`w`$ to cross over $`1`$. We have proven in MFZ05 this Lagrangian is equivalent to an effective two-field model
$$=\frac{1}{2}_\mu \psi ^\mu \psi +\frac{1}{2}_\mu \chi ^\mu \chi V(\psi \chi )\frac{M^2}{2c}\chi ^2,$$
(19)
with the following definition
$`\chi `$ $`=`$ $`{\displaystyle \frac{c}{M^2}}\mathrm{}\varphi ,`$ (20)
$`\psi `$ $`=`$ $`\varphi +\chi .`$ (21)
Note that the redefined fields $`\psi `$ and $`\chi `$ have opposite signs in their kinetic terms. One might be able to derive the higher derivative terms in the effective Lagrangian (18) from fundamental theories. In fact it has been shown in the literature that this type of operators does appear as some quantum corrections or due to the non-local physics in the string theory simon ; woodard ; gross . With the higher derivative terms to the Einstein gravity, the theory is shown to become renormalizable stelle which has attracted many attentions. In fact the canonical form for the higher derivative theory has been put forward by Ostrogradski about one and a half century ago ostrogradski . In short, it is interesting and worthwhile to study further the implications of models with higher derivatives in cosmology (for a recent study see e.g. relevnt1 ).
## III Perturbations of the quintom model
The quintom scenario as we have argued above needs extra degrees of freedom to the conventional models of a single scalar field, such as quintessence, phantom and k-essence and the simple realization of the quintom is a model with two scalar fields or “equivalently” two scalar fields for the case with the higher derivative operators. In the discussions below on the perturbations we will restrict ourselves to the two-field model of quintom with the following lagrangian:
$$=_Q+_P$$
(22)
where
$$_Q=\frac{1}{2}_\mu \varphi _1^\mu \varphi _1V_1(\varphi _1)$$
(23)
describes the quintessence component, and
$$_P=\frac{1}{2}_\mu \varphi _2^\mu \varphi _2V_2(\varphi _2)$$
(24)
for the phantom component. The background equations of motion for the two scalar fields $`\varphi _i(i=1,2)`$ are
$$\ddot{\varphi _i}+2\dot{\varphi _i}\pm a^2\frac{V_i}{\varphi _i}=0,$$
(25)
where the positive sign is for the quintessence and the minus sign for the phantom. In general there will be couplings between the two scalar fields, here for simplicity we neglect them.
For a complete study on the perturbations, the fluctuations of the fields as well as those of the metric need to be considered. In the conformal Newtonian gauge the perturbed metric is given by
$$ds^2=a^2(\tau )[(1+2\mathrm{\Psi })d\tau ^2(12\mathrm{\Phi })dx^idx_i],$$
(26)
Using the notations of ma , the perturbation equations satisfied by each of the components of the quintom model (22) are:
$`\dot{\delta }_i`$ $`=`$ $`(1+w_i)(\theta _i3\dot{\mathrm{\Phi }})3({\displaystyle \frac{\delta P_i}{\delta \rho _i}}w_i)\delta _i,`$ (27)
$`\dot{\theta }_i`$ $`=`$ $`(13w_i)\theta _i{\displaystyle \frac{\dot{w_i}}{1+w_i}}\theta _i+k^2({\displaystyle \frac{\delta P_i/\delta \rho _i}{1+w_i}}\delta _i\sigma _i+\mathrm{\Psi }),`$ (28)
where
$$\theta _i=(k^2/\dot{\varphi _i})\delta \varphi _i,\sigma _i=0,$$
(29)
$$w_i=\frac{P_i}{\rho _i},$$
(30)
and
$$\delta P_i=\delta \rho _i2V_i^{}\delta \varphi _i=\delta \rho _i+\frac{\rho _i\theta _i}{k^2}[3(1w_i^2)+\dot{w_i}].$$
(31)
Combining Eqs. (27), (28) and (31), we have
$`\dot{\delta }_i`$ $`=`$ $`(1+w_i)(\theta _i3\dot{\mathrm{\Phi }})3(1w_i)\delta _i3{\displaystyle \frac{\dot{w}_i+3(1w_i^2)}{k^2}}\theta _i,`$ (32)
$`\dot{\theta }_i`$ $`=`$ $`2\theta _i+{\displaystyle \frac{k^2}{1+w_i}}\delta _i+k^2\mathrm{\Psi }.`$ (33)
Since the quintom model in (22) is essentially a combination of a quintessence and a phantom field, one obtains the perturbation equations of quintom by combining the equations above together. The corresponding variables for the quintom system are
$$w_{quintom}=\frac{_iP_i}{_i\rho _i},$$
(34)
$$\delta _{quintom}=\frac{_i\rho _i\delta _i}{_i\rho _i},$$
(35)
and
$$\theta _{quintom}=\frac{_i(\rho _i+p_i)\theta _i}{_i(\rho _i+P_i)}.$$
(36)
Note that for the quintessence component, $`1w_11`$, while for the phantom component, $`w_21`$.
With the two fields the quintom model in (22) will be characterized by the potential $`V_i`$ in (23) and (24). In this paper we take $`V_i(\varphi _i)=\frac{1}{2}m_i^2\varphi _i^2`$. In general the perturbations of $`\varphi _i`$ today stem from two origins, the adiabatic and the isocurvature modes. If we use the gauge invariant variable $`\zeta _i=\mathrm{\Phi }\frac{\delta \rho _i}{\dot{\rho _i}}`$ instead of $`\delta _i`$, and the relation $`\mathrm{\Phi }=\mathrm{\Psi }`$ in the universe without anisotropic stress, the equations (32) and (33) can be rewritten as,
$`\dot{\zeta }_i`$ $`=`$ $`{\displaystyle \frac{\theta _i}{3}}C_i(\zeta _i+\mathrm{\Phi }+{\displaystyle \frac{}{k^2}}\theta _i),`$ (37)
$`\dot{\theta }_i`$ $`=`$ $`2\theta _i+k^2(3\zeta _i+4\mathrm{\Phi }),`$ (38)
where
$$C_i=\frac{\dot{w}_i}{1+w_i}+3(1w_i)=_0[\mathrm{ln}(a^6|\rho _i+p_i|)].$$
(39)
$`\zeta _\alpha `$ is the curvature perturbation on the uniform-density hypersurfaces for the $`\alpha `$-component in the universe wands . Usually, the isocurvature perturbations of $`\varphi _i`$ are characterized by the differences between the curvature perturbation of the uniform-$`\varphi _i`$-density hypersurfaces and that of the uniform-radiation-density hypersurfaces,
$$S_{ir}3(\zeta _i\zeta _r),$$
(40)
where the subscript $`r`$ represents radiation. In this paper, we assume there is no matter isocurvature perturbations, so $`\zeta _m=\zeta _r`$. Eliminating $`\zeta _i`$ in equations (37) and (38), we obtain a second order equation for $`\theta _i`$,
$$\ddot{\theta }_i+(C_i2)\dot{\theta }_i+(C_i2\dot{}+k^2)\theta _i=k^2(4\dot{\mathrm{\Phi }}+C_i\mathrm{\Phi }).$$
(41)
This is an inhomogeneous differential equation, the general solution to it is the sum of a general solution to its homogeneous part and a special integration. In the following, we will show that the special integration corresponds to the adiabatic perturbation. Before the era of dark energy domination, the universe is dominated by some background fluids, for instance, the radiation or the matter. The perturbation equations of the background fluid are,
$`\dot{\zeta }_f=\theta _f/3,`$
$`\dot{\theta }_f=(13w_f)\theta _f+k^2[3w_f\zeta _f+(1+3w_f)\mathrm{\Phi }].`$ (42)
From the Poisson equation
$$\frac{k^2}{^2}\mathrm{\Phi }=\frac{9}{2}\underset{\alpha }{}\mathrm{\Omega }_\alpha (1+w_\alpha )(\zeta _\alpha +\mathrm{\Phi }+\frac{}{k^2}\theta _\alpha )\frac{9}{2}(1+w_f)(\zeta _f+\mathrm{\Phi }+\frac{}{k^2}\theta _f),$$
(43)
we have approximately on large scales,
$$\mathrm{\Phi }\zeta _f\frac{}{k^2}\theta _f.$$
(44)
Combining these equations above with $`=2/[(1+3w_f)\tau ]`$, we get (note numerically $`\theta _f𝒪(k^2)\zeta _f`$)
$`\zeta _f={\displaystyle \frac{5+3w_f}{3(1+w_f)}}\mathrm{\Phi }=\mathrm{Const}.,`$
$`\theta _f={\displaystyle \frac{k^2(1+3w_f)}{3(1+w_f)}}\mathrm{\Phi }\tau .`$ (45)
So, we can see from Eq. (41) that there is a special solution to it which is given approximately on large scales by ,
$$\theta _i^{ad}=\theta _f,$$
(46)
and from Eq. (38) we have,
$$\zeta _i^{ad}=\zeta _f.$$
(47)
This shows that the special integration to Eq. (41) has the meaning that it corresponds to the adiabatic perturbation. Hence, for the sake of isocurvature perturbations of $`\varphi _i`$, we can only consider the solution to the homogeneous part of Eq. (41),
$$\ddot{\theta }_i+(C_i2)\dot{\theta }_i+(C_i2\dot{}+k^2)\theta _i=0.$$
(48)
These solutions are represented by $`\theta _i^{iso}`$ and $`\zeta _i^{iso}`$. The relation between them is
$$\zeta _i^{iso}=\frac{\dot{\theta }_i^{iso}2\theta _i^{iso}}{3k^2}.$$
(49)
Since the general solution of $`\zeta _i`$ is
$$\zeta _i=\zeta _i^{ad}+\zeta _i^{iso}=\zeta _r+\zeta _i^{iso},$$
(50)
the isocurvature perturbations are simply $`S_{ir}=3\zeta _i^{iso}`$.
In order to solve Eq. (48), we need to know the forms of $`C_i`$ and $``$ as functions of time $`\tau `$. For this purpose, we solve the background equations (25). In radiation dominated period, $`a=A\tau ,=1/\tau `$ and we have
$$\varphi _1=\tau ^{1/2}[A_1J_{1/4}(\frac{A}{2}m_1\tau ^2)+A_2J_{1/4}(\frac{A}{2}m_1\tau ^2)],$$
(51)
and
$$\varphi _2=\tau ^{1/2}[\stackrel{~}{A}_1I_{1/4}(\frac{A}{2}m_2\tau ^2)+\stackrel{~}{A}_2I_{1/4}(\frac{A}{2}m_2\tau ^2)],$$
(52)
respectively, where $`A`$, $`A_i`$ and $`\stackrel{~}{A}_i`$ are constants, $`J_\nu (x)`$ is the $`\nu `$th order of Bessel function and $`I_\nu (x)`$ is the $`\nu `$th order of modified Bessel function. Usually the masses are small in comparison with the expansion rate in the early universe $`m_i/a`$, we can approximate the (modified) Bessel functions as $`J_\nu (x)x^\nu (c_1+c_2x^2)`$ and $`I_\nu (x)x^\nu (\stackrel{~}{c}_1+\stackrel{~}{c}_2x^2)`$. We note that $`J_{1/4}`$ and $`I_{1/4}`$ are divergent when $`x0`$. Given these arguments one can see that this requires large initial values of $`\varphi _i`$ and $`\dot{\varphi }_i`$ if $`A_2`$ and $`\stackrel{~}{A}_2`$ are not vanished. If we choose small initial values, which is the natural choice if the dark energy fields are assumed to survive after inflation, only $`A_1`$ and $`\stackrel{~}{A}_1`$ modes exist, so $`\dot{\varphi }_i`$ will be proportional to $`\tau ^3`$ in the leading order. Thus, the parameters $`C_i`$ in equation (39) will be $`C_i=10/\tau `$ (we have used $`|\rho _i+p_i|=\dot{\varphi }_i^2/a^2`$). So, we get the solution to Eq. (48),
$$\theta _i^{iso}=\tau ^4[D_{i1}\mathrm{cos}(k\tau )+D_{i2}\mathrm{sin}(k\tau )].$$
(53)
$`\theta _i^{iso}`$ oscillates with an amplitude damping with the expansion of the universe. The isocurvature perturbations $`\zeta _i^{iso}`$ decrease rapidly. If we choose large initial values for $`\varphi _i`$ and $`\dot{\varphi }_i`$, $`A_2`$ and $`\stackrel{~}{A}_2`$ modes are present, $`\dot{\varphi }_i`$ will be proportional to $`\tau ^2`$ in the leading order and $`C_i=0`$. Now the solution to Eq. (48) is
$$\theta _i^{iso}=\tau [D_{i1}\mathrm{cos}(k\tau )+D_{i2}\mathrm{sin}(k\tau )].$$
(54)
$`\theta _i^{iso}`$ will oscillate with a increasing amplitude, so $`\zeta _i^{iso}`$ remains constant on large scales.
Similarly, in matter dominated era, $`a=B\tau ^2,=2/\tau `$, the solutions for the fields $`\varphi _i`$ are
$$\varphi _1=\tau ^3[B_1\mathrm{sin}(\frac{B}{3}m_1\tau ^3)+B_2\mathrm{cos}(\frac{B}{3}m_1\tau ^3)],$$
(55)
and
$$\varphi _2=\tau ^3[\stackrel{~}{B}_1\mathrm{sinh}(\frac{B}{3}m_2\tau ^3)+\stackrel{~}{B}_2\mathrm{cosh}(\frac{B}{3}m_2\tau ^3)],$$
(56)
respectively. We get the same conclusions as those reached by the above analysis for the radiation dominated era. If we choose small initial values at the beginning of the matter domination, we will get the isocurvature perturbations in $`\varphi _i`$ decrease with time. On the contrary for large initial values, the isocurvature perturbations remain constant on large scales. This conclusion is expectable. In the case of large initial velocity, the energy density in the scalar field is dominated by the kinetic term and it behaves like the fluid with $`w=1`$. The isocurvature perturbation in such a fluid remains constant on large scales. In the opposite case, however, the energy density in the scalar field will be dominated by the potential energy due to the slow rolling. It behaves like a cosmological constant, and there is only tiny isocurvature perturbation in it.
We have seen that the isocurvature perturbations in quintessence-like or phantom-like field with quadratical potential decrease or remain constant on large scales depending on the initial velocities. In this sense the isocurvature perturbations are stable on large scales. The amplitude of these perturbations will be proportional to the value of Hubble rate evaluated during the period of inflation $`H_{inf}`$ if their quantum origins are from inflation. For a large $`H_{inf}`$ isocurvature dark energy perturbations may be non-negligible and will contribute to the observed CMB anisotropyKMT01 ; MT04 . In the cases discussed here, however, these isocurvature perturbations are negligible. Firstly, large initial velocities are not possible if these fields survive after inflation as mentioned above. Secondly, even though the initial velocities are large at the beginning of the radiation domination, these velocities will be reduced to a small value due to the small masses and the damping effect of Hubble expansion. In general the contributions of dark energy isocurvature perturbations are not very largeGW05 and here for simplicity we assume $`H_{inf}`$ is small enough that the isocurvature contributions are negligible<sup>2</sup><sup>2</sup>2We assume in the next section when the mass of quintessence is larger by an order and oscillates during late time evolutions, the adiabatic condition still satisfies well.. Thus we will concentrate on in next sections the effects of the adiabatic perturbations of the quintom model with two scalars considered in this paper.
## IV Signatures of quintom and the effects of perturbations on observations
Based on the perturbation equations(35) and (36), we modify the code of CAMB camb and will study preliminarily in this chapter the observational signatures of quintom. Throughout this paper we assume a flat universe. In showing the illustrative effects for quintom we have assumed the fiducial background parameters to be $`\mathrm{\Omega }_b=0.042,\mathrm{\Omega }_c=0.231,\mathrm{\Omega }_{DE}=0.727`$, where $`DE`$ denotes dark energy and today’s Hubble constant is fixed at $`H_0=69.255`$ km/s Mpc<sup>-2</sup>. We will calculate the effects of perturbed quintom on CMB and LSS.
In the quintom model we focus on there are two parameters: one is the quintessence mass and the other being the phantom mass. When the mass of quintessence is heavier than Hubble parameter the field will start to oscillate and consequently one will get an oscillating quintom. In the numerical discussions we will fix the mass of the phantom field to be $`m_P2.0\times 10^{60}m_{pl}`$. We vary the quintessence mass with the typical values being $`m_Q=10^{60}m_{pl}`$ and $`4\times 10^{60}m_{pl}`$ respectively. We plot in Fig. 1 the equations of state as function of the scale factor for the above two sets of the parameters and their corresponding effects on the observations. One can see the obvious oscillating feature of quintom as the mass of quintessence component goes heavier. After touching the $`w=1`$ pivot for several times, $`w`$ crosses $`1`$ consequently where the phantom part dominates dark energy. The quintom field modifies the metric perturbations: $`\delta g_{00}=2a^2\mathrm{\Psi },\delta g_{ii}=2a^2\mathrm{\Phi }\delta _{ij}`$ and consequently contribute to the late time Integrated Sachs-Wolfe (ISW) effect. The ISW effect is an integrant of $`\dot{\mathrm{\Phi }}+\dot{\mathrm{\Psi }}`$ over conformal time and wavenumber k. The above two quintom models yield quite different evolving $`\mathrm{\Phi }+\mathrm{\Psi }`$ as shown in the right panel of Fig.1, where the scale is $`k10^3`$ Mpc<sup>-1</sup>. We can see the late time ISW effects differ significantly when dark energy perturbations are taken into account(solid lines) or not(dashed lines).
ISW effects take an important part on large angular scales of CMB and on the matter power spectrum of LSS. For a constant EOS of phantom Ref.WL03 has shown that the low multipoles of CMB will get significantly enhanced when dark energy perturbations are neglected. On the other hand for a matter like scalar field where the equation of state is around zero, perturbations will also play an important role on the large scales of CMB, as shown in ref.CDS98 . Our results on CMB and LSS reflect the two combined effects of phantom and oscillating quintessence. Note that in the early studies of quintessence effects on CMB, one usually considers a constant $`w_{eff}`$ instead:
$$w_{eff}\frac{𝑑a\mathrm{\Omega }(a)w(a)}{𝑑a\mathrm{\Omega }(a)},$$
(57)
however this is not enough for the study of effects on SN, nor for CMB when the EOS of dark energy has a very large variation with redshift, such as the model of oscillating quintom considered in this paper.
To face the oscillating model of quintom with the current observations, we make a preliminary fitting to the first year Wilkinson Microwave Anisotropy Probe(WMAP) TT and the TE temperature–polarization cross-power spectrum as well as the recently released 157 “Gold” SN dataRiess04 . Following Refs.smalll ; Fengl1 in all the fittings below we will fix $`\tau =0.17`$, $`\mathrm{\Omega }_mh^2=0.135`$ and $`\mathrm{\Omega }_bh^2=0.022`$, we set the spectral index as $`n_S=0.95`$ and the amplitude of the primordial spectrum will be used as a continuous parameter. In the fittings of oscillating quintom we’ve fixed the mass of phantom to be $`m_P6.2\times 10^{61}m_{pl}`$. Fig.2 delineates 3$`\sigma `$ WMAP and SN constraints on the two-field quintom model, it also shows the corresponding best fit values. In the labels $`m_Q`$ and $`m_P`$ stand for the mass of quintessence and phantom respectively. The left panel of Fig.2 shows the separate WMAP and SN constraints. The green(shaded) area is WMAP constraints on models where dark energy perturbations have been included and the blue area(contour with solid lines) is without dark energy perturbations. The perturbations of dark energy have no effects on the geometric constraint of SN. The right panel shows the combined WMAP and SN constraints on the two-field quintom model with perturbations (green/shaded region) and without perturbations (red region/contour with solid lines). We find the confidence regions do show a large difference when the perturbations of dark energy have been taken into account or not.
So far we have investigated the imprints of oscillating quintom on CMB and LSS. Now we consider another example where $`w`$ crosses $`1`$ smoothly without oscillation. It is interesting to study the effects of this type of quintom model with its effective equation of state defined in (57) exactly equal to $`1`$ on CMB and matter power spectrum. This study will help to distinguish the quintom model from the cosmological constant. We have realized such a model of quintom in the lower right panel of Fig.3, which can be easily given in the two-field model with lighter quintessence mass. In this example we have set $`m_Q2.6\times 10^{61}m_{pl},m_P6.2\times 10^{61}m_{pl}`$. We assume there are no initial kinetic energy. The initial values of the quintessence component is set as $`\varphi _{1i}=0.226m_{pl}`$ and the phantom part: $`\varphi _{2i}=6.64\times 10^3m_{pl}`$. We find the EOS of quintom crosses $`1`$ at $`z0.15`$, which is consistent with the latest SN results.
The model of quintom, which is mainly favored by current SN only, needs to be confronted with other observations in the framework of concordance cosmology. SN making the only direct detection of dark energy, this model is most promising to be distinguished from the cosmological constant and other dynamical dark energy models which do not get across $`1`$ by future SN projects on the low redshift(for illustrations see e.g. cooray ). This is also the case for the model of quintom in the full parameter space: it can be most directly tested in low redshift Type Ia supernova surveys. In the upper left panel of Fig.3 we delineate the different ISW effects among the cosmological constant (red/light solid), the quintom model which gives $`w_{eff}=1`$ with (blue/dark solid) and without(blue dashed) perturbations. Similar to the previous oscillating case, the difference is very large when switching off quintom perturbations and much smaller when including the perturbations. In the upper right panel we find the quintom model cannot be distinguished from a cosmological constant in light of WMAP. The two models almost give exactly the same results in CMB TT and TE power spectra when including the perturbations. We find the difference in CMB is hardly distinguishable even by cosmic variance.
Given the fact aobve that from CMB observations quintom with $`w_{eff}=1`$ makes no distinctive signatures, now we discuss briefly the signatures in some other observations. To do that we need to consider the physical observables which can be affected by the evolving $`w`$ sensitively. In comparison with the cosmological constant such a quintom model gives a different evolution of expansion history of universe, such as altering the epoch of matter-radiation equality. The Hubble expansion rate $`H`$ is :
$$H\frac{\dot{a}}{a^2}=H_0[\mathrm{\Omega }_ma^3+\mathrm{\Omega }_ra^4+X]^{1/2}$$
(58)
where X, the energy density ratio of dark energy between the early epochs and today, is quite different for the $`quintom`$-CDM and $`\mathrm{\Lambda }`$CDM. In the $`\mathrm{\Lambda }`$CDM scenario, X is simply a constant while in general for dark energy models with varying energy density or EOS,
$$X=\mathrm{\Omega }_{DE}a^3e^{3{\scriptscriptstyle w(a)d\mathrm{ln}a}}.$$
(59)
The two models will give different Hubble expansion rates. This is also the case between the quintom model with $`w_{eff}=1`$ in the left panel of Fig.3 and a cosmological constant. Different $`H`$ leads directly to different behaviors of the growth factor. In the linear perturbation theory all Fourier modes of the matter density perturbations grow at the same rate. The matter density perturbations are independent of $`k`$:
$$\ddot{\delta }_k+\dot{\delta }_k4\pi Ga^2\rho _\mathrm{M}\delta _k=0.$$
(60)
The growth factor $`D_1(a)`$ characterizes the growth of the matter density perturbations: $`D_1(a)=\delta _k(a)/\delta _k(a=1)`$ and is normalized to unity today. In the matter-dominated epoch we have $`D_1(a)=a`$. Analytically $`D_1(a)`$ is often approximated by the Meszaros equation Dodsbk :
$$D_1(a)=\frac{5\mathrm{\Omega }_mH(a)}{2H_0}_0^a\frac{da^{}}{(a^{}H(a^{})/H_0)^3},$$
(61)
where we can easily see the difference between the model of quintom and a cosmological constant due to the different Hubble expansion rates. More strictly one needs to solve Eq.(60) numerically. In the lower left panel of Fig.3 we show the difference of $`D_1(a)`$ between the quintom with $`w_{eff}=1`$ and the cosmological constant. The difference in the linear growth function is considerably large in the late time evolution and possibly distinguishable in future LSS surveys and in weak gravitational lensing (WGL) observations. WGL has emerged with a direct mapping of cosmic structures and it has been recently shown that the method of cosmic magnification tomography can be extremely efficientlensmagn , which leaves a promising future for breaking the degeneracy between quintom and a cosmological constant.
## V Perturbations of parametrized quintom and the effects on the observations
There have been many studies in the literature in the fittings of the dark energy with parametrized EOS, such as the linear parametrization $`w=w_0+w_1z`$ linearP to SN and other observations such as CMB and LSS. For the latter observations the perturbations of dark energy need to be considered. However, at the point of $`w=1`$, as pointed out in Section II one would be encountered with the singularity of the isentropic sound speed. Moreover in the perturbation equation (28) one will get infinite $`\dot{\theta }`$. For the physical quantity $`(\rho +P)\theta `$ in the model of the single field of quintessence, it is not divergent at $`w=1`$, i.e. $`\theta \mathrm{}`$ but $`(\rho +P)\theta =k^2\dot{\varphi }\delta \varphi =0`$, however for the model with parametrized EOS one will generically have an unphysical divergence when $`\dot{w}0`$ at the cosmological constant boundary. The detailed explanation is given as follows: firstly from Eq.(28) one will get infinite $`\dot{\theta }`$ and the physical continuity implies that one will also get $`\theta \mathrm{}`$ at $`w=1`$. Introducing the new physical quantity which is relevant to the CMB observations:
$$𝒱(1+w)\theta ,$$
(62)
Eqs.(32, 33) can be rewritten now as:
$`\dot{\delta }`$ $`=`$ $`𝒱+(1+w)3\dot{\mathrm{\Phi }}3(1w)\delta 3{\displaystyle \frac{\dot{w}/(1+w)+3(1w)}{k^2}}𝒱,`$ (63)
$`\dot{𝒱}`$ $`=`$ $`2𝒱+k^2\delta +{\displaystyle \frac{\dot{w}}{1+w}}𝒱+k^2(1+w)\mathrm{\Psi }.`$ (64)
We can easily see that $`\dot{𝒱}\mathrm{}`$ when $`\dot{w}0`$ at $`w=1`$.
We should point out that both the scalar fields and fluids obey the same form of equations on the evolution of perturbations: Eqs.(27,28), and the only difference comes from the term of $`\delta P_i/\delta \rho _i`$. If one starts from Eqs.(32, 33) and study the effects of dark energy by parametrizing the EOS, this is equivalent to the description of the effects of the scalar field and is identical to work starting with dark energy potentials. If in models with the parametrized EOS we have $`w`$ always in the range $`[1,1]`$, or $`w1`$ for $`0<a<\mathrm{}`$, there will be no unphysical divergence and this equivalently describes the single field of quintessence or phantomguo05 . For example in model with $`w=w_0+w_1\mathrm{sin}(\mathrm{ln}a)`$, if we restrict $`w_0=0`$ and $`|w_1|1`$ then Eqs.(63, 64) will always be continuous. However when the parameter space is enlarged to include $`\dot{w}0`$ at $`w=1`$ Eqs.(63, 64) will be unphysical.
We emphasize that the above discussions are valid only for models with a single field. For models with multi fields we have shown explicitly in the previous sections the perturbation equations Eqs.(32, 33) are continuous during the crossing of the cosmological constant boundary. It is similar for models with two fluids or models with two components of parametrized EOS: $`w=\mathrm{\Sigma }\mathrm{\Omega }_iw_i`$ where each component $`w_i`$ does not evolve across $`1`$. This implies, however, in the fitting of the models to the observational data the parameters introduced for the EOS should be doubled if allowing the EOS $`w`$ to vary and get across $`1`$. Certainly this is not practically applicable. It would be nice to develop a technique to include the perturbations which approximates well to the quintom, meanwhile not introducing the extra degrees of freedom to the models considered widely in the literature with parametrized EOS. We will make a proposal for it below.
First of all we consider a system of quintom with two fields as above, $`\varphi _1`$ being quintessence-like and $`\varphi _2`$ being phantom-like, but restrict the EOS of the system not to cross over $`1`$. In this case we will show the background of this system is equivalent to a model with an effective single scalar field denoted by $`\chi `$. By definition the pressure P and energy density $`\rho `$ of the $`\chi `$ field should be equal to the two-field case. When the kinetic term of $`\varphi _1`$ is larger than that of the phantom part $`\varphi _2`$, the whole system of dark energy gives rise to an EOS larger than $`1`$ and the effective $`\chi `$ behaves like a quintessence. On the contrary when $`\dot{\varphi _1}^2\dot{\varphi _2}^20`$, $`\chi `$ is a phantom field. Hence the kinetic and potential terms of $`\chi `$, in terms of $`\varphi _1`$ and $`\varphi _2`$, can be expressed as
$$\pm \dot{\chi }^2=\dot{\varphi _1}^2\dot{\varphi _2}^2$$
(65)
and
$$V(\chi )=V(\varphi _1)+V(\varphi _2),$$
(66)
where the ”+” sign in Eq. (65) is for the case where the total EOS of dark energy is quintessence-like and the ”-” sign for phantom-like evolutions. We can directly reconstruct the potential and time evolutions of $`\chi `$. For example if we set the potentials of the two fields to be both linear:
$$V_i(\varphi _i)=V_{0i}+\lambda _i\varphi _i,$$
(67)
in the early epochs of radiation and matter domination dark energy fields are slow rolling and
$$\varphi _{1}^{}{}_{}{}^{}\lambda _1/3H,\varphi _{2}^{}{}_{}{}^{}\lambda _2/3H,$$
(68)
where prime denotes derivative respects to the physical time. For the whole system in the quintessence phase $`\varphi _1`$ will have a larger kinetic energy, and in the radiation dominant epoch
$$H=1/2t,\chi ^{}=\pm \frac{2}{3}t\sqrt{\lambda _1^2\lambda _2^2}.$$
(69)
On the other hand from Eq.(66) we have
$$V_{,\chi }(\chi )\chi ^{}=V_{,\varphi _1}(\varphi _1)\varphi _1^{}+V_{,\varphi _2}(\varphi _2)\varphi _2^{},$$
(70)
combining Eqs.(68,69,70) we can easily get
$$V_{,\chi }(\chi )=\sqrt{\lambda _1^2\lambda _2^2},$$
(71)
consequently the effective potential of $`\chi `$ analytically is
$$V(\chi )=\pm \sqrt{\lambda _1^2\lambda _2^2}(\chi \chi _0),$$
(72)
where $`\chi _0`$ can be easily set by the initial conditions of $`\varphi _i`$ and the sign of “$`+`$” or “$``$” is somewhat optional. The arguments above applies for the case when the total EOS of the system is restricted to be no larger than $`1`$, the effective scalar will behave like phantom.
On the evolution of perturbations we can see from Eqs.(27,28) that the phantom and the quintessence fields obey the same equations. As shown in Section III although generically the two field model would have non-vanishing isocurvature perturbations we can choose suitable initial conditions so that the isocurvature contributions can be safely negligible. In this sense when the total EOS does not evolve across minus unity, the whole system can be equally described by an adiabatic field: both the background evolution and the adiabatic perturbations, as shown similarly in Refs.Gordon01 ; Fengl1 ; Fengl2 in the inflationary universe.
We have demonstrated in the previous paragraphs the equivalence between the two-field quintom model and the single scalar field model when the EOS of the system does not cross over $`1`$. However if the total equation of state for the double fields does cross over $`1`$, this system will not be able to be described effectively by a single scalar field. To study the perturbations of the dark energy models with EOS across $`1`$, we introduce a small positive constant $`c`$ to divide the whole region of the allowed value of the EOS $`w`$ into three parts: 1) $`w>1+c`$; 2) $`1+c>w>1c`$; and 3) $`w<1c`$. For 1) the EOS is always larger than $`1`$ and for 3) $`w`$ is always less than $`1`$. For both cases the system with two fields as shown above can be described effectively by a single scalar field with a potential satisfying DCS02
$$\pm a^2\frac{d^2V}{d\chi ^2}=\frac{3}{2}(1w)\left[\frac{\ddot{a}}{a}^2\left(\frac{7}{2}+\frac{3}{2}w\right)\right]+\frac{1}{1+w}\left[\frac{\dot{w}^2}{4(1+w)}\frac{\ddot{w}}{2}+\dot{w}(3w+2)\right],$$
where “$`+`$” is for the case 1) and “$``$” for the case 3). One can see that $`\frac{d^2V}{d\chi ^2}`$ is divergent and there would be a discontinuity in the derivative $`V^{}`$ at the turning point of $`w=1`$, which corresponds to $`c0`$. As an example in Fig. 4 we give the reconstructed potential of the effective $`\chi `$ field for an oscillating quintom. One can see $`\chi `$ behaves like quintessence when $`w>1`$ and like phantom when $`w<1`$. The reconstructed potential is well defined except in region 2) when the EOS gets across $`1`$, where there is a sharp discontinuity on $`V^{}(\chi )`$.
For the case 2), different from those in 1) and 3), the perturbations cannot be fully described by a single adiabatic field. However as we learn from the above, for the realistic quintom models the perturbations in the region 2) will be continuous and not divergent, i.e. $`\delta `$ and $`\theta `$ are continuous, and the derivatives of $`\delta `$ and $`\theta `$ are finite. A good approximation to the perturbation in region 2) is requiring it to match to the regions 1) and 3) at the boundary. Practically we take $`\delta `$ and $`\theta `$ to be constant matching to regions 1) and 3) at the boundary and set
$$\dot{\delta }=0,\dot{\theta }=0.$$
(73)
In the numerical calculation the constant $`c`$ is a very small number, the approximation above lies in a very close neighborhood of $`w=1`$. In practice in our numerical calculations we’ve limited the range to be $`|\mathrm{\Delta }w=c|<10^5`$. Since the region 2) is extremely limited, neglecting the evolutions of perturbations as shown in (74) is quite safe and well approximated. Thus we can use Eqs.(32, 33) to study the effects of perturbations in models with parametrized EOS. We have also numerically checked the validity of Eq.(74) and found their contributions to the observed CMB and LSS power spectra are very small. The procedure of our checking is listed as follows:
1. Start with the two-field model of quintom and record $`w(a)`$, compute CMB and LSS spectra with perturbations.
2. Build a code in CAMBcamb to include dark energy perturbations with parametrized EOS. Include perturbations by setting Eq.(73) and treating $`\delta ,\theta `$ as continuous.
3. Interpolate w(a) in the code with parametrized EOS, compute CMB and LSS spectra and make comparisons with the results from step 1.
With this procedure we have considered a model of oscillating quintom and found the difference is no more than $`10^4`$, which is safely negligible.
As examples now we study the effects of perturbations for several models with parametrized EOS in light of WMAP and SN data. The first example is given by Ref.FLPZ , where $`w`$ is parametrized by
$$w(\mathrm{ln}a)=w_a+w_0\mathrm{cos}[w_1\mathrm{ln}(a/a_c)]$$
(74)
with $`a`$ being the scale factor. This model has a nice feature of unifying the early inflation and the current accelerated expansions. In Ref.FLPZ the period of oscillation has been set as long as $`200`$ e-folds. It is interesting to study the consequences with a shorter period. Here for illustrations we fix $`w_a=1`$, $`w_1=20`$ and $`a_c=1`$. In the upper panels of Fig.5 we show the illustrative fittings when with and without the perturbations. We can see the parameter space has been enlarged a lot when including the contributions of the perturbations. For a second example we parametrize $`w`$ as
$$w(\mathrm{ln}a)=w_a+w_0a\mathrm{cos}[w_1a+a_c].$$
(75)
In the numerical calculation we’ve fixed $`w_a=1`$, $`w_1=50`$ and $`a_c=0`$. In the lower panels of Fig.5 we can see the effects are still very prominent both in the separate and combined constraints, although not as strong as the example in (75). For the third example we take $`w`$ to be non-oscillatory:
$$w=w_0/(1\mathrm{ln}a),$$
(76)
where the original form was firstly proposed in Ref.CW04 . We find in our case SN constraints are very weak due to the fixed background parameters, the 1$`\sigma `$ regions have not been affected much by the perturbations, but the 2, 3$`\sigma `$ regions have been enlarged significantly when the perturbations are taken into account.
## VI Conclusions
In this paper we have studied the perturbations of the dynamical quintom model of dark energy in a self-consistent way. It is physically significant for the inclusion of quintom perturbations, both on the theoretical grounds of model buildings and on the fittings to the observations. Due to the singularities and instabilities of perturbations at the cosmological constant boundary, we have shown a new method regarding the impossibility of k-essence as a viable quintom model. In the realistic quintom model buildings one must include the perturbations. In general one needs to add extra degrees of freedom to realize the model of quintom. In the two-field model and the model with a d’Alembertian operator the isocurvature contributions may be safely negligible in the simplest case. We have considered the implications of quintom perturbations on the observations of CMB and LSS. We have shown that the parameter space is different when one includes the perturbations of dark energy or not. In trying to constrain dark energy in a model independent way we have also proposed a method to include the perturbations for models of dark energy with parametrized EOS across $`1`$. With some specific examples of the parametrized EOS, we show that the parameter space which characterizes the properties of the model will get enlarged in general when including the perturbations. This will lead to important consequences in the phenomenological studies on the cosmological imprints of dynamical dark energy, including the model of quintom. A thorough investigation of current constraints on the quintom model of dark energy where dark perturbations are taken into account is beyond the scope of current paper and will be presented elsewhere toappear ; zhang05 .
Overall, a dynamical quintom model is favored by current SN data and not ruled out by the combined observational constraints. There are still some inconsistencies today among different observations in the precision cosmology and the concordance $`\mathrm{\Lambda }`$CDM model has not yet fitted well to the observations in a high enough confidence level, in this sense we might not be adopting the Ockham’s razor with a cosmological constant. When we start from a $`\mathrm{\Lambda }`$CDM model in the probe of our universe we cannot achieve more subtle physics beyond that. This is necessary to bear in mind for us to understand the nature of dark energy with the accumulation of the observational data.
Acknowledgements: We thank the anonymous referee for helpful suggestions. We thank Robert Brandenberger, Xue-Lei Chen, Zuhui Fan, Pei-Hong Gu, Hong Li, Hiranya Peiris, Yunsong Piao, Yong-Zhong Xu and Peng-Jie Zhang for helpful discussions. We acknowledge the using of CMBfastcmbfast ; IEcmbfast in our early studies and CAMBcamb ; IEcamb for all the numerical calculations. In the fitting to WMAP we’ve used the code developed in Ref. Verde03 . We thank Antony Lewis for early miscellaneous help and discussions on the cosmocoffeecoffee . This work is supported in part by National Natural Science Foundation of China under Grant Nos. 90303004 and 19925523 and by Ministry of Science and Technology of China under Grant No. NKBRSF G19990754. B. F. would like to thank the hospitality of the National Astronomical Observatories, Chinese Academy of Sciences where part of this work was finished and M. L. is supported by Alexander von Humboldt Foundation. |
warning/0507/q-bio0507035.html | ar5iv | text | # A stochastic model for wound healing
## 1 Introduction
The biology of wound healing is fairly well understood . A simplified version of the process may be given as follows: a layer of undamaged cells is usually quiescent, so that the birth rate of cells matches the death rate, and both are quite small. When a wound is suffered, there is a rapid signal the wakes the cells up – perhaps a pulse of ATP or a calcium wave. Cells at the edge of the wound become more mobile, and also enhance their proliferation rate. (Otherwise the healed layer would not have the right density.) A typical experiment to study this process consists in plating suitable cells (e.g. epithelial cells) on a substrate so that they form a confluent monolayer. Then a scratch is made in the layer, and the process of filling in the scratch is studied. For example, the speed of advance of the invading cells, $`v`$, is easily measured.
There have been many modeling studies of wound healing . In many cases ( is an exception) the process is studied using some variant of the Fisher-Kolmogorov (FK) equation . This is an obvious model to use. It builds in diffusion with diffusion constant $`D`$ and proliferation with growth rate $`k`$ (related to inverse doubling time). It also shuts off growth for the confluent layer at density $`c_o`$.
$$c/t=D^2c+kc(1c/c_o)$$
(1)
The justification for using a continuum equation for a cellular process relies on the common experience that coarse-graining is reasonable for dynamic processes involving a large number of agents. In this particular case, we expect that the FK equation should be useful if the characteristic length of the pattern predicted by Eq. (1) is much larger that the size of a cell.
However, it is well known that coarse-graining the FK equation has many pitfalls even in this limit, and that the transition to the continuum limit is often very slow. This motivates the present investigation: we present a discrete stochastic model for wound healing, and study it in various limiting regimes. It is quite similar to a model previously introduced and studied for flame-front propagation . Thus, our results and methods should be of interest beyond the explicit biological context. We will give new numerical and analytical results, and show how, in one and two dimensions, our model aproaches the FK limit. We will show that in the biologically relevant regime there are corrections to FK due to discreteness.
## 2 Formulation of the model and known properties
Consider a set of sites that form a linear or square lattice, corresponding to one or two dimensional ‘tissues’. We allow each site to be occupied by zero or one cells. Our initial configuration is an occupied half space: if $`i`$ labels the $`x`$ coordinates of the sites, then we have all sites with $`i0`$ occupied. The dynamical rules are as follows: we choose a parameter $`p`$ which specifies the proliferation rate of the cells. Then at any time step we choose a cell at random, and an adjacent site at random as a target for diffusion or proliferation. E.g., in 1d if we choose a cell at site $`i`$, we also pick site $`i+1`$ or $`i1`$ as a target. If the target site is empty, with probability $`p`$ we put a new cell at the target, and with probability $`q=1p`$ we move the chosen cell to the target. If the target site is filled, we do nothing.
These are examples of the elementary processes allowed:
* (…1111000…) $``$ (…1111100…); probability $`p`$
* (…1111000…) $``$ (…1110100..); probability $`q`$
As time advances cells appear for $`i>0`$. These form a front or chemical wave. We will examine the speed, $`v(p)`$, and front width, $`w(p)`$ for the invading cells. Precise definitions for these quantities will be given below.
An essentially identical model was devised by Kerstein to describe flame-front propagation. He studied it numerically in 1d, and Bramson *et al.* found some analytic results, also in 1d. In their formulation there is a parameter $`\gamma `$ which may be identified as $`(1p)/p`$ in our notation. Also, in their model the time unit is different from ours by a factor $`1+\gamma `$. If $`V(\gamma )`$ denotes the front speed in the Kerstein model, we have:
$$v(p)=V(\gamma )/(1+\gamma ).$$
(2)
In there are two exact results. In our notation these are:
* $`v(p)1/2+𝒪(q^2)`$ as $`p1`$,
* $`v(p)\sqrt{2p}`$ as $`p0`$.
The first of these two is obvious. In the limit $`p=1`$ there is no diffusion, only proliferation, and the half space advances with no vacancies. The only process allowed is to choose the leading cell at site $`i`$ and proliferate at site $`i+1`$. Since half the moves are wasted by choosing as a target the filled site at $`i1`$ the front speed is 1/2. The lack of a term linear in $`q`$ will be derived below.
The second result may be understood by comparison with Eq. (1). Consider the coarse-grained limit of our discrete model using the lattice constant as the unit of space, and a computer time step as the time unit. It is elementary to see that the diffusion coefficient, $`D`$, is 1/2. Now consider a collection of cells distant from one another with concentration $`c`$. In unit time the number will increase to $`(1+p)c`$. By integrating Eq. (1) over space in the low density limit, we see that we must identify $`k=p`$. The front velocity given by Eq. (1) is well known for bounded initial conditions :
$$v=2\sqrt{Dk}=\sqrt{2p}.$$
(3)
Kerstein verified both limits numerically. We will extend these one dimensional results below both numerically and analytically, and also investigate the two-dimensional case.
We note for future reference that the solutions of Eq. (1) generate an interface with an intrinsic width (see Figure (1)) given by:
$$w=\sqrt{D/k}1/\sqrt{p}.$$
(4)
Previous authors have not discussed the front width, but, as we will see, it is relevant to a biological interpretation of the results.
## 3 Numerical results
### 3.1 Defining the front
The solution to Eq. (1) is a traveling front of the general form shown in the sketch in Figure 1. Our data for the discrete model is the form of occupancies of sites as a function of time. We present here a useful way to analyze such data that allows easy comparison to continuum theories.
We start by defining the occupancy of a given column of our numerical data, $`P(i)`$. In 1d this is simply 1 or 0, depending on whether site $`i`$ is occupied. In 2d it is the average occupancy of column $`i`$, that is, the number of occupied sites with first coordinate $`i`$ divided by the width of the system (the total number of such sites) which we denote by $`L`$. We also define the negative of the discrete derivative of $`P`$; it is localized near the interface:
$$\mathrm{\Delta }(i)=P(i)P(i+1).$$
(5)
Note that at long times we certainly have $`P(0)=1`$, and for large enough $`i,P(i)=0`$. Thus:
$$\underset{i=0}{\overset{\mathrm{}}{}}\mathrm{\Delta }(i)=1.$$
(6)
That is, we can use $`\mathrm{\Delta }`$ as a weight function to define averages. We put:
$`i`$ $`=`$ $`{\displaystyle \underset{i=0}{\overset{\mathrm{}}{}}}i\mathrm{\Delta }(i)={\displaystyle \underset{i=1}{\overset{\mathrm{}}{}}}P(i)=n_p`$
$`i^2`$ $`=`$ $`{\displaystyle \underset{i=0}{\overset{\mathrm{}}{}}}i^2\mathrm{\Delta }(i)={\displaystyle \underset{i=1}{\overset{\mathrm{}}{}}}(2i1)P(i),\mathrm{}`$ (7)
Here, $`n_p`$ is the number of particles for $`i1`$ in 1d, or that number divided by $`L`$ in 2d. That is, we get the position of the front by the total mass of created particles.
The front speed is defined as
$$v=\underset{t\mathrm{}}{lim}i/t.$$
The front width is given by
$$w=\sqrt{i^2i^2}.$$
Other moments of the distribution can be defined similarly.
### 3.2 One dimension
The results of our simulations are shown in Figures (2) and (3). The front speeds were found by fitting $`i`$ to $`vt`$. It is remarkable that the front speed is quite well defined even for very small $`p`$. Of course, for $`p=0`$ the speed is not defined at all.
The convergence to the continuum predictions is quite evident in the figures. Note that the prediction for $`v(p)`$ does not contain an adjustable constant, so the agreement is quite remarkable. However, following the work of we would expect the corrections to the continuum prediction, Eq. (3), to follow:
$$v(p)=\sqrt{2p}A/\mathrm{ln}^2(p),$$
(8)
where $`A`$ is a factor of order unity. In fact, this expression does not fit our results. Rather, the correction to the continuum formula is more like a power law with a power near 2/3.
### 3.3 Two dimensions
Our numerical results for $`v(p)`$ in 2d are given in Figure 4. Note that as $`p1,v(p)>0.5`$. Analysis of the processes the contribute to front motion in 2d is more complex than in 1d where $`v(1)=0.5`$ is an exact result. In particular the front will always be rough (see below) so that particles behind the leading particle will not be blocked from advancing.
For $`p1`$ we expect that we should converge to the result of the FK equation, namely that $`v(p)\sqrt{p}`$. However, for 2d we have no convincing *a priori* estimate of the prefactor. Following the treatment above, we might proceed by noting that in 2d $`D=1/4`$ for the discrete model. We then have:
$$v(p)\sqrt{p}.$$
(9)
As we can see from the figure, this is a reasonable estimate for small $`p`$.
In two dimensions the width of the interface is a more complicated object than in one dimension . The reason for this is that in the presence of fluctuations the front can do two different things: it can spread so that it has an intrinsic width (as in 1d) by having a reduced density in the interface region, but also it can *wander*. Indeed, for $`p=1`$ wandering is the only effect possible. (Recall that in 1d $`w(p=1)=0`$.) In fact, in the large $`p`$ regime this model is identical to the Eden model where perimeter sites all grow with equal probability.
The phenomenology of the Eden model is well understood . The wandering of the interface is time-dependent and obeys (in our system of units):
$`w`$ $``$ $`t^{1/3}tL^{3/2}`$ (10)
$``$ $`L^{1/2}tL^{3/2}.`$
This is indeed the case here: see Figure 5. The agreement with the Eq. (10) is reasonable. We have verified that the scaling behavior given in Eq. (10) persists down to $`p=0.5`$.
However, as $`p`$ decreases the intrinsic width grows rapidly. As soon as the intrinsic width exceeds the saturated width from wandering (the second line of Eq. (10)) we will loose the power-law time dependence of $`w`$. In Figure 6 we show the saturated width for a range of $`p`$.
## 4 Results for $`p1`$ in one dimension
For $`p1`$ the dominant process is proliferation. For $`p=1`$ this gives rise to a simple configuration as we have mentioned above: all sites behind the front are occupied, and the front advances because the leading cell proliferates. For $`q1`$ there is a small probability $`q/2`$ of creating a configuration with a ‘hole’. Because the model is very simple we can use this observation to work out the power series expansion of $`v(q)`$.
### 4.1 Exact solution of model for states with one hole
Suppose we consider only states with zero holes or one hole at any position. We expect these to be the dominant configurations small $`q`$. Define the states:
* $`|0=(\mathrm{}11111000\mathrm{})`$
* $`|1=(\mathrm{}11101000\mathrm{})`$
* $`|2=(\mathrm{}11011000\mathrm{})`$
* $`|3=(\mathrm{}10111000\mathrm{})`$, etc.
We allow transitions only between these states. The transitions and their associated probabilities $`W_{ij}W(|i|j)`$ are:
$`W_{00}=p/2W_{01}=q/2`$
$`W_{10}=(1+p)/2W_{12}=1/2`$
$`W_{n0}=pW_{n,n1}=q/2W_{n,n+1}=1/2(n>1)`$ (11)
Note that in many of these transitions the actual location of the rightmost 1 changes. We always define states in a frame moving with the front.
The equations for the probabilities are:
$`P_0W_{01}`$ $`=`$ $`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}P_nW_{n0}`$
$`P_1(W_{10}+W_{12})`$ $`=`$ $`P_0W_{01}+P_2W_{21}`$
$`P_n(W_{n0}+W_{n,n1}+W_{n,n+1})`$ $`=`$ $`P_{n1}W_{n1,n}+P_{n+1}W_{n+1,n}(n>1)`$
Using Eq. (11) we have:
$`\left({\displaystyle \frac{q}{2}}\right)P_0`$ $`=`$ $`\left({\displaystyle \frac{1+p}{2}}\right)P_1+p{\displaystyle \underset{n=2}{\overset{\mathrm{}}{}}}P_n`$
$`\left({\displaystyle \frac{3q}{2}}\right)P_1`$ $`=`$ $`\left({\displaystyle \frac{q}{2}}\right)P_0+\left({\displaystyle \frac{q}{2}}\right)P_2`$
$`\left({\displaystyle \frac{3q}{2}}\right)P_n`$ $`=`$ $`\left({\displaystyle \frac{1}{2}}\right)P_{n1}+\left({\displaystyle \frac{q}{2}}\right)P_{n+1}(n>1)`$ (13)
For $`n>1`$, we make the ansatz $`P_n=a^{n1}P_1`$, and inserting this in the last equation above, we find:
$`\left({\displaystyle \frac{3q}{2}}\right)a`$ $`=`$ $`{\displaystyle \frac{1}{2}}+\left({\displaystyle \frac{q}{2}}\right)a^2`$
$`a`$ $`=`$ $`{\displaystyle \frac{3q\sqrt{910q+q^2}}{2q}}={\displaystyle \frac{1}{3}}+{\displaystyle \frac{4q}{27}}+{\displaystyle \frac{2q^2}{243}}\mathrm{}.`$ (14)
Next, we substitute $`P_2=aP_1`$ into the second line of Eq. (13) and use $`_{n=2}^{\mathrm{}}P_n=1P_0P_1`$, in the first line of Eq. (13). Solving these two equations we find:
$`P_0`$ $`=`$ $`{\displaystyle \frac{2(1q)(3(1+a)q)}{6(5+2a)q+aq^2}}`$ (15)
$`=`$ $`1{\displaystyle \frac{q}{2}}{\displaystyle \frac{q^2}{12}}{\displaystyle \frac{11q^3}{216}}{\displaystyle \frac{137q^3}{3888}}\mathrm{}`$
Thus
$`P_1`$ $`=`$ $`{\displaystyle \frac{q}{3}}{\displaystyle \frac{q^2}{54}}{\displaystyle \frac{19q^3}{972}}+\mathrm{}`$
$`P_2`$ $`=`$ $`aP_1={\displaystyle \frac{q}{9}}+{\displaystyle \frac{7q^2}{162}}+{\displaystyle \frac{53q^3}{2916}}+\mathrm{}`$
$`P_3`$ $`=`$ $`aP_2={\displaystyle \frac{q}{27}}+{\displaystyle \frac{5q^2}{162}}+{\displaystyle \frac{7q^3}{324}}+\mathrm{}`$
The velocity can be found from
$$v=\frac{1}{2}\frac{qP(\times 01)}{2},$$
(16)
where $`\times `$ is any string of 0’s and 1’s, and we omit the zeros to the right. In this case, $`P(\times 01)=P_1`$, because $`|1`$ is the only one-hole state that ends with $`(01)`$. Thus, we have
$$v=\frac{1}{2}\frac{q^2}{6}+\frac{q^3}{108}+\frac{19q^4}{1944}\mathrm{}$$
(17)
We have precise numerical data for $`v(q)`$ for small $`q`$; see Figure 7. We find that Eq. (17) is correct only up to quadratic order, as we might expect in the one-hole approximation. For example, for $`q=0.1`$, we find numerically that $`v=0.498292`$, while $`1/2q^2/6=0.49833`$, so that the coefficient of $`q^3`$ should be negative, not positive. Note $`q^3/1080.00001`$.
### 4.2 Reduced distribution functions.
In order to go beyond quadratic terms in $`q`$, we introduce *reduced distribution functions*. This method would, in principle, allow the power-series expansion to be carried to arbitrary order.
A reduced distribution function is probability to have a given pattern near the front for any pattern to the left. For example, as in Eq. (16):
$$P(\times 01)=\text{prob}(\mathrm{}xxx0100\mathrm{})$$
where the sites marked as $`x`$ are any string. Likewise, we define $`P(\times 11)`$, $`P(\times 001),P(\times 101)`$, and so forth. Note that, for example:
$`P(\times 001)+P(\times 101)`$ $`=`$ $`P(\times 01)`$
$`P(\times 011)+P(\times 111)`$ $`=`$ $`P(\times 11).`$
We can derive a hierarchy of equations based on events that change the last $`n`$ sites. For $`n=2`$, consider all events that change the probability that the last two sites are (11). We have:
$`({\displaystyle \frac{p}{2}}+{\displaystyle \frac{p}{2}})P(\times 001)`$ $`+`$ $`({\displaystyle \frac{1}{2}}+{\displaystyle \frac{1}{2}}+{\displaystyle \frac{p}{2}})P(\times 101)`$
$`({\displaystyle \frac{q}{2}}+{\displaystyle \frac{q}{2}})P(\times 011)`$ $``$ $`\left({\displaystyle \frac{q}{2}}\right)P(\times 111)=0.`$ (18)
The positive terms represent events that increase the population of states ending with (11), and the negative terms represent events that decrease that population. Next we write all states in terms of $`P(\times 01),P(\times 001),P(\times 011)`$, i.e., states that have a leading zero on the left. For the other states, we use:
$`P(\times 101)`$ $`=`$ $`P(\times 01)P(\times 001)`$ (19)
$`P(\times 111)`$ $`=`$ $`P(\times 11)P(\times 011)=1P(\times 01)P(\times 011).`$ (20)
Then Eq. (18) becomes:
$$\frac{q}{2}\left(\frac{3}{2}\right)P(\times 01)+\left(\frac{1+q}{2}\right)P(\times 001)+\left(\frac{q}{2}\right)P(\times 011)=0.$$
(21)
Now, we expect that $`P(\times 01)=𝒪(q),P(\times 011)=𝒪(q),`$ and $`P(\times 001)=𝒪(q^2)`$, because a diffusive move (weight $`q/2`$) is required to produce each empty site starting from state $`|0`$. To order $`q`$ we find from Eq. (21)
$$P(\times 01)=\frac{q}{3}+𝒪(q^2)$$
(22)
which agrees with the leading behavior found in the one-hole approximation. This implies that the velocity is given by:
$$v=\frac{1}{2}\frac{qP(\times 01)}{2}=\frac{1}{2}\frac{q^2}{6}+𝒪(q^3)$$
(23)
Note that there is no linear term, as mentioned above.
For the next-order behavior, we use:
$`P(\times 011)`$ $`=`$ $`{\displaystyle \frac{q}{9}}+𝒪(q^2)`$ (24)
$`P(\times 001)`$ $`=`$ $`{\displaystyle \frac{q^2}{9}}+𝒪(q^3)`$ (25)
where the leading behavior in Eq. (24) is from the one-hole approximation. The second line follows from a simple argument: the leading behavior of $`P(\times 001)`$ is determined by $`P(\mathrm{}1001)`$, and its leading behavior is determined by the equation:
$$P(\mathrm{}1001)(3/2)=P(\mathrm{}101)(q/2)+P(\mathrm{}10001)(p/2+q/2)+\mathrm{}$$
Again $`\mathrm{}`$ represents a string of all 1’s to the left. The second and higher-order terms on the right-hand-side are of order $`q^3`$, so to leading order we find $`P(\mathrm{}1001)=(q/3)P(\mathrm{}101)=q^2/9`$, which proves Eq. (25).
Using these results, Eq. (21) implies
$$P(\times 01)=\frac{q}{3}+\frac{2q^2}{27}+𝒪(q^3)$$
which yields
$$v=\frac{1}{2}\frac{q^2}{6}\frac{q^3}{27}+𝒪(q^4).$$
For the case $`q=0.1`$, these three terms give $`v=0.498296`$, in close agreement with the numerical simulations, which give 0.498292.
We have carried this procedure to the next order, $`n=3`$, by straightforward extensions of what we have given above. The result for the velocity is:
$$v=\frac{1}{2}\frac{q^2}{6}\frac{q^3}{27}\frac{49q^4}{1215}+𝒪(q^5)$$
(26)
For $`q=0.1`$, the predicted velocity is now 0.4982923, in complete agreement with the numerical result 0.498292. Even at $`q=0.5`$, the prediction of Eq. (26), $`v=0.4511831`$, is within 0.2% of the measured value, 0.45014.
## 5 Application to biology
Our emphasis in this paper has been an analysis of the model introduced in the introduction. It is interesting, nevertheless, to make some comments on the relationship of this model to real biological systems. Needless to say, our view of wound repair is very much oversimplified. In a real tissue there are various types of cells such as stem cells which have different behavior with respect to proliferation than others. Further, the proliferation cycle is complex, and involves time delays that we have not considered except in a rough way. Also, the initiation of wound healing is probably mediated by chemical signals rather than cell proximity as we have assumed .
However, if we are interested in macroscopic features such as the velocity and shape of the moving front, we are entitled to hope that many of these details will be unimportant. We can then ask how to translate the parameters of our model to a real system. We will take as an example the experiment of Sheardown and Cheng on the wounding of rabbit corneas.
In the emphasis was on modeling with the FK equation. To this end the authors measured $`D`$ in Eq. (1) by looking at the initial stage of invasion of cultured cells, and found $`D=1.6110^6\mathrm{mm}^2/\mathrm{s}`$. The parameter $`k`$ in Eq. (1) is related to the mitotic rate of cells which was measured by labeling with a dye: $`k^1=4.3\mathrm{days}`$. Using these parameters the authors found reasonable agreement for the velocity of the front. Further, for these parameters, the shape of the front is quite ‘fuzzy’, that is, the width, $`w`$, is many cells across so that the wound fills in gradually, as observed. We should note that this is in sharp contrast to other observations where the advancing front is quite sharp. We will return to this point below.
We now attempt to translate these observations into the parameters of our discrete model. We need to define units of length and time. For length it is natural to take a typical cell size, $`d=10\mu `$ as the lattice unit. It is clear that we can define the *hopping time*, $`\tau _{hop}`$ by $`D=d^2/2\tau _{hop}`$. This turns out to be about 108 seconds for the rabbit cornea. However, there is another characteristic time, the cell *cycling time*, $`\tau _{cyc}=1/k`$. This is $`3.710^5`$ seconds for the same experiment. In our model, in $`N`$ computer cycles there are $`Np`$ cell cycles and $`Nq`$ hops. Thus our time unit should be $`q\tau _{hop}+p\tau _{cyc}`$.
To determine the biological $`p`$ note that $`p/q=\tau _{cyc}/\tau _{hop}`$. For the rabbit experiment we get $`p=310^4`$. This is the regime of very diffuse, fuzzy interfaces, as observed. In this regime FK modeling should be reasonable, though, as Figure 2 shows, there are still differences between FK and the discrete model in this regime.
If we apply the same set of considerations to the systems studied in we find a contradiction. The width of the interface should be quite substantial for the small $`p`$’s relevant to biological experiments, cf. Figure 3. In fact, FK modeling shows the same thing. However, direct observation in these cases shows that the front is quite sharp.
A possible solution to this quandary is given by where it is pointed out that cell-cell adhesion can have an effect on wound healing in some systems. In fact, for the cells that they study they can regulate the adhesion, and hence the front width, by controlling the supply of Ca<sup>++</sup> ion in the solution bathing the cells. A detailed discrete model in their paper shows these effects too. This is an interesting avenue for future work.
## 6 Summary and conclusions
In this paper we have extended the work of on a discrete model. We have shown that the model can be interpreted as a representation of the important biological process of wound healing. We have given numerical results in one and two dimensions, and a power-series expansion of the velocity around $`q=0`$ in one dimension. We have shown that the biologically interesting regime is that of $`p1`$.
There are a number of further extensions of this work that could be pursued. Our method of reduced distribution functions should be applicable to models with more complex rules as long as a sensible expansion parameter, analogous to $`q`$, is present. We do not understand why the convergence to the FK limit is different in our case than in the generic cases discussed in . A more extensive numerical study may be called for.
We think that the most interesting extension of the model would be to include cell-cell adhesion, in the spirit of . This work is in progress.
#### Acknowledgements
LMS would like to thank P. Maini and J. Devita for helpful discussions. Supported in part by NSF grant DMS-0244419. EK is supported by NIH grant CA085139-01A2. |
warning/0507/hep-ph0507183.html | ar5iv | text | # Dual parameterization of generalized parton distributions and description of DVCS data
## I Introduction
Generalized parton distributions (GPDs) have become a standard QCD tool for analyzing and parameterizing the non-perturbative parton structure of hadronic targets, for reviews see Ji:1998pc ; Radyushkin:2000uy ; Goeke:2001tz ; Belitsky:2001ns ; Diehl:2003ny ; Belitsky:2005qn . In general, GPDs are more general and complex objects than structure functions and form factors. In addition, experimentally measured observables do not access the GPDs themselves but only their convolution with the hard scattering coefficients. Therefore, the experimental determination of the GPDs is an extremely difficult task. Hence, when dealing with the GPDs, one invariably uses models, the known limiting behavior and general properties of the GPDs and the physical intuition.
The GPDs have been modeled using virtually all known models of the nucleon structure: bag models Ji:1997gm , the chiral quark-soliton model Petrov:1998kf , light-front models Tiburzi:2001ta ; Tiburzi:2002tq , constituent quark models Scopetta:2003et , Bethe-Salpeter approach Tiburzi:2004mh , a NJL model Mineo:2005qr . In addition, a double distribution model of the GPDs Radyushkin:1998es ; Musatov:1999xp and modeling by perturbative diagrams Pobylitsa:2002vw were suggested.
The factorization theorem for deeply virtual Compton scattering (DVCS) Collins:1998be gives a practical possibility to measure the GPDs by studying various processes involving the GPDs: DVCS, exclusive electroproduction of vector mesons, wide angle Compton scattering Radyushkin:1998rt ; Diehl:1998kh , exclusive $`p\overline{p}\gamma \gamma `$ annihilation Freund:2002cq , the $`p\overline{p}\gamma \pi ^0`$ process Kroll:2005ni , $`\gamma ^{}\gamma \pi \pi `$ near threshold Diehl:1998dk . However, in order to accommodate such a potentially large number of data, parameterizations of the GPDs should be sufficiently flexible and versatile. In particular, they should allow for the connection of the DVCS with the $`p\overline{p}\gamma \gamma `$ process.
The commonly used double distribution parameterization of the GPDs Radyushkin:1998es ; Musatov:1999xp is one example of the model of the GPDs which could be used to connect different physical channels Teryaev:2001qm . However, the parameterization of the GPDs based on the double distribution has a number of problems from the phenomenological point of view. First, in order to have the full form of polynomiality, the so-called $`D`$-term Polyakov:1999gs has to added by hand. Second, the model dramatically overestimates the low-$`x`$ HERA data on the total DVCS cross section because it involves proton parton distributions at extremely small and unmeasured values of Bjorken $`x`$ Freund:2001hd . Third, the model does not allow for an intuitive physical motivation and interpretation, see Freund:2002ff for a discussion of the physics of GPDs.
In this paper, we offer a new model for the GPDs, which was in a general form introduced in Polyakov:2002wz . Unlike the models of the GPDs mentioned above, the present model has a simple physical interpretation and direct correspondence to the mechanical properties of the target Polyakov:2002yz . The suggested parameterization of the GPDs can be analytically continued in $`t`$ to the physical region of the $`p\overline{p}\gamma \gamma `$ reaction and also allows for flexible modeling of the $`t`$-dependence of the GPDs.
The considered parameterization of the GPDs is called dual because the GPDs are presented as an infinite series of $`t`$-channel exchanges, which reminds of the ideas of duality in hadron-hadron scattering.
In this work, we formulate the minimal version of the dual parameterization and determine the free parameters of the model. Using the resulting dual parameterization of the GPDs, we successfully describe the HERA data on the DVCS cross section Adloff:2001cn ; Chekanov:2003ya ; Aktas:2005ty . We explain that our parameterization suits the low-$`x_{Bj}`$ kinematics especially well because the quark singlet parton distributions are never probed at the unmeasurably low values of Bjorken $`x`$ and because the final expression for the DVCS amplitude is numerically stable. Thus, the dual parameterization of the GPDs gives an opportunity to have a physically intuitive model of the GPDs, which agrees with the DVCS experiments and which can serve as an alternative to the popular double distribution model.
## II The dual parameterization of GPDs
The dual representation of the GPDs was first introduced for the pion GPDs in Polyakov:1998ze . The essence of that derivation is presented below. The starting point was the decomposition of the two-pion distribution amplitude ($`2\pi `$DA) in terms of the eigenfunctions of the ERBL evolution equation (Gegenbauer polynomials $`C_n^{3/2}`$), the partial waves of produced pions (Legendge polynomials $`P_l`$) and generalized form factors $`B_{nl}`$. The moments of the $`2\pi `$DA, being the matrix elements of certain local operators, could be related by crossing to the moments of the pion GPDs. Then, the pion GPDs could be formally reconstructed using the explicit form of their moments.
Based on the result of Polyakov:1998ze , the dual representation for the proton GPDs was suggested in Polyakov:2002wz . In this paper, we will consider only the singlet ($`C`$-even) combination of the GPDs $`H`$, which give the dominant contribution to the total DVCS cross section at high energies and small $`t`$. We will work in the leading order approximation and, hence, we will consider only quark GPDs.
The dual representation of the singlet GPD $`H^i`$ of the quark flavor $`i`$ is Polyakov:2002wz
$$H^i(x,\xi ,t,\mu ^2)=\underset{n=1,\mathrm{odd}}{\overset{\mathrm{}}{}}\underset{l=0,\mathrm{even}}{\overset{n+1}{}}B_{nl}^i(t,\mu ^2)\theta \left(1\frac{x^2}{\xi ^2}\right)\left(1\frac{x^2}{\xi ^2}\right)C_n^{3/2}\left(\frac{x}{\xi }\right)P_l\left(\frac{1}{\xi }\right),$$
(1)
where $`x`$, $`\xi `$ and $`t`$ are the usual GPD variables. The series (1) is divergent at fixed $`x`$ and $`\xi `$, and, hence, it should be understood as a formal series. In particular, it is incorrect to evaluate the series term by term. As a result, the GPD $`H^i`$ of Eq. (1) has a support over the entire $`1x1`$ region, regardless that each term of the series is non-vanishing only for $`\xi x\xi `$. The formal representation (1) can be equivalently rewritten as a converging series using the technique developed in Polyakov:2002wz .
The derivation of Eq. (1) used the idea of duality of hadronic physics, when the scattering amplitude in the $`s`$-channel is represented as an infinite series of the $`t`$-channel exchanges. This explains the name “the dual representation” for the representation of Eq. (1).
As a double series, Eq. (1) is inconvenient for phenomenological applications. For the evaluation of the LO DVCS amplitude, it is useful to introduce the functions $`Q_k(x,t)`$ whose Mellin moments generate the $`B_{nl}^i`$ form factors Polyakov:2002wz
$$B_{nn+1k}^i(t,\mu ^2)=_0^1𝑑xx^nQ_k^i(x,t,\mu ^2),$$
(2)
where $`k`$ is even. A remarkable property of the dual representation is that the $`\mu ^2`$-evolution of the functions $`Q_k^i`$ is given by the usual leading order (LO) DGLAP evolution.
Since the $`B_{nl}^i`$ form factors are related to the moments of $`H^i`$, the $`Q_k`$ functions are also constrained by these moments. In particular, from
$$_1^1𝑑xxH^i(x,\xi ,t)=M_2^i(t)+\frac{4}{5}\xi ^2d^i(t)=\frac{6}{5}\left[B_{12}(t)\frac{1}{3}\left(B_{12}(t)2B_{10}(t)\right)\xi ^2\right],$$
(3)
it follows that
$`{\displaystyle _0^1}𝑑xxQ_0^i(x,t,\mu ^2)={\displaystyle \frac{5}{6}}M_2^i(t,\mu ^2),`$
$`{\displaystyle _0^1}𝑑xxQ_2^i(x,t,\mu ^2)={\displaystyle \frac{5}{12}}M_2^i(t,\mu ^2)+d^i(t,\mu ^2),`$ (4)
where $`M_2^i`$ at $`t=0`$ is the proton light-cone momentum fraction carried by the quarks; $`d^i(t)`$ is the first moment of the quark $`D`$-term.
In addition, the $`B_{nn+1}`$ form factors at the zero momentum transfer are fixed by the Mellin moments of the quark singlet parton distribution functions (PDFs). In particular,
$$\frac{3}{4}_0^1𝑑x\left(q^i(x,\mu ^2)+\overline{q}^i(x,\mu ^2)\right)=B_{10}(0)=_0^1𝑑xQ_0^i(x,0,\mu ^2).$$
(5)
The $`Q_0^i`$ functions at $`t=0`$ are completely fixed in terms of the forward proton PDFs Polyakov:2002wz
$$Q_0^i(x,0,\mu ^2)=q^i(x,\mu ^2)+\overline{q}^i(x,\mu ^2)\frac{x}{2}_x^1\frac{dz}{z^2}\left(q^i(z,\mu ^2)+\overline{q}^i(z,\mu ^2)\right).$$
(6)
As suggested in Polyakov:2002wz , keeping only the functions $`Q_0^i`$ and $`Q_2^i`$ constitutes the minimal version of the dual parameterization of the GPDs. The functions $`Q_0^i`$ and $`Q_2^i`$ are defined by Eqs. (6) and (4), where $`M_2^i(t)`$ and $`d^i(t)`$ have a clear physical interpretation since they are the form factors of the energy-momentum tensor evaluated between the states representing the given target. At $`t=0`$, $`M_2^i(0)`$ is the fraction of the plus-momentum of the nucleon carried by the quarks of flavor $`i`$; $`d^i(0)`$ characterizes the shear forces experienced by the quarks in the target.
Next we discuss the minimal version of the dual representation in detail. While $`Q_0^i`$ at $`t=0`$ is defined by Eq. (6), only the first $`x`$-moment of $`Q_2^i`$ is constrained. We simply assume that $`Q_2^iQ_0^i`$ and take
$$Q_2^i(x,0,\mu ^2)=\beta ^iQ_0^i(x,0,\mu ^2),$$
(7)
where $`\beta ^i`$ are constants. From Eq. (4), we obtain
$$\beta ^i=\frac{6}{5}\frac{d^i(0)}{M_2^i(0)}+\frac{1}{2},$$
(8)
which gives
$$\beta ^u=4.4,\beta ^d=8.9,\beta ^s=0.5.$$
(9)
In this numerical estimate, we assume that $`d^u=d^d2`$ and $`d^s0`$, which agrees with the SU(3)-symmetric chiral quark soliton model calculation of the nucleon $`D`$-term Kivel:2000fg : $`_id^i(0)4`$, and takes into account the SU(3) symmetry breaking in the nucleon PDFs (the suppression of the strange quark PDF at the low resolution scale). The momentum fractions $`M_2^i`$ were evaluated at $`\mu _0=0.6`$ GeV using the LO GRV parton PDFs Gluck:1998xa .
In general, $`\beta ^i`$ also depend on $`\mu ^2`$. However, as will be seen from the general expression for the DVCS amplitude, at small values of $`\xi `$ typical for the HERA data on the total DVCS cross section, the contribution of the $`Q_2^i`$ function is kinematically suppressed. Therefore, the goodness of the description of the data is not affected by the exact values of $`\beta ^i`$, and we simply used Eq. (9) at all $`\mu ^2`$.
Until recently, the $`t`$-dependence of the DVCS cross section was not measured. One would simply assume that the DVCS cross section exponentially depends on $`t`$,
$$\frac{d\sigma _{\mathrm{DVCS}}(x_{Bj},Q^2,t)}{dt}=\mathrm{exp}\left(B|t|\right)\left(\frac{d\sigma _{\mathrm{DVCS}}(x_{Bj},Q^2,t)}{dt}\right)_{t=0},$$
(10)
such that the total DVCS cross section is
$$\sigma _{\mathrm{DVCS}}(x_{Bj},Q^2)=\frac{1}{B}\left(\frac{d\sigma _{\mathrm{DVCS}}(x_{Bj},Q^2,t)}{dt}\right)_{t=0}.$$
(11)
The value of the slope parameter $`B`$ was rather uncertain, $`5B9`$ GeV<sup>-2</sup>. The range of the values covers the experimentally measured range of the $`t`$-slope of electroproduction of light vector mesons at HERA. However, very recently the $`t`$-dependence of the total DVCS cross section for $`0.1|t|0.8`$ GeV<sup>2</sup> and at $`Q^2=8`$ GeV<sup>2</sup> was measured by the H1 collaboration at HERA and was fitted by the exponential form of Eq. (10) with the result $`B=6.02\pm 0.35\pm 0.39`$ GeV<sup>-2</sup> Aktas:2005ty .
In our numerical estimates of the DVCS cross section, we calculate the DVCS amplitude at $`t=0`$ and then use Eq. (11) in order to find the $`t`$-integrated DVCS cross section. In general, the slope $`B`$ should decrease with increasing $`Q^2`$. A particular model for the $`Q^2`$-dependent slope was suggested in Freund:2002qf : $`B(Q^2)=8(10.15\mathrm{ln}(Q^2/2))`$ GeV<sup>-2</sup>. In our analysis, we use the same $`Q^2`$-dependence,
$$B(Q^2)=7.6\left(10.15\mathrm{ln}(Q^2/2)\right)\mathrm{GeV}^2,$$
(12)
but with a slightly smaller constant $`7.6`$ GeV<sup>-2</sup>, which is chosen such that Eq. (12) reproduces the H1 value of the slope at $`Q^2=8`$ GeV<sup>2</sup>.
In summary, our parameterization of the GPDs $`H^i`$ is defined by Eqs. (4), (6) and (7). The $`t`$-dependence of the DVCS cross section is given by Eqs. (10) and (12). This is the minimal version of the dual representation of the GPDs, which can be readily extended by considering more $`Q_k^i`$ functions, a more elaborate $`t`$-dependence and by taking into account the other GPDs of the proton. The main practical advantage of our representation is that the $`\mu ^2`$-evolution of $`Q_{0,2}^i`$ is given by the usual DGLAP evolution of the singlet PDFs, see Eq. (6).
## III Description of low-$`x`$ HERA data on DVCS cross section
In this section, we evaluate the total DVCS cross section using the minimal model of the dual representation of the GPDs and compare the results to the HERA data Chekanov:2003ya ; Aktas:2005ty .
The total unpolarized DVCS cross section on the photon level reads, see e.g. Freund:2001hm ,
$$\sigma _{\mathrm{DVCS}}(x_{Bj},Q^2)=\frac{\alpha _{e.m.}^2x_{Bj}^2\pi (1\xi ^2)}{Q^4\sqrt{1+4x_{Bj}^2m_N^2/Q^2}}_{t_{min}}^{t_{max}}𝑑t|\overline{𝒜}_{\mathrm{DVCS}}(\xi ,t,Q^2)|^2,$$
(13)
where $`\alpha _{e.m.}`$ the fine-structure constant; $`\xi =1/2x_{Bj}/(1x_{Bj}/2)`$ in the Bjorken limit; $`t_{max}0`$ and $`t_{min}1`$ GeV<sup>-2</sup>; $`|\overline{𝒜}_{\mathrm{DVCS}}|^2`$ is the squared and spin-averaged DVCS amplitude.
To the leading order in $`\alpha _s`$, the DVCS amplitude is expressed in terms of the singlet combination of the GPDs $`H^i`$,
$$𝒜_{\mathrm{DVCS}}(\xi ,t,Q^2)=\underset{i}{}e_i^2_0^1𝑑xH^i(x,\xi ,t,Q^2)\left(\frac{1}{x\xi +i0}+\frac{1}{x+\xi i0}\right).$$
(14)
Using our model for the GPDs and the results of Polyakov:2002wz , the DVCS amplitude can be presented in a compact form in terms of the $`Q_0^i`$ and $`Q_2^i`$ functions
$$𝒜_{\mathrm{DVCS}}(\xi ,t,Q^2)=\underset{i}{}e_i^2_0^1\frac{dx}{x}\underset{k=0}{\overset{2}{}}x^kQ_k^i(x,t,Q^2)\left(\frac{1}{\sqrt{1\frac{2x}{\xi }+x^2}}+\frac{1}{\sqrt{1+\frac{2x}{\xi }+x^2}}2\delta _{k0}\right).$$
(15)
Using the exponential ansatz for the $`t`$-dependence of the DVCS cross section, the total DVCS cross section is expressed in terms of the DVCS amplitude at $`t=0`$ \[see Eq. (11)\]
$$\sigma _{\mathrm{DVCS}}(x_{Bj},Q^2)=\frac{\alpha _{e.m.}^2x_{Bj}^2\pi (1\xi ^2)}{Q^4\sqrt{1+4x_{Bj}^2m_N^2/Q^2}}\frac{1}{B(Q^2)}|\overline{𝒜}_{\mathrm{DVCS}}(\xi ,t=0,Q^2)|^2,$$
(16)
where $`𝒜_{\mathrm{DVCS}}(\xi ,t=0,Q^2)`$ is given by Eq. (15) evaluated with $`Q_{0,2}^i(x,0,Q^2)`$.
Our predictions for the $`Q^2`$ and $`W`$-dependence of the total DVCS cross section are presented in Figs. 1 and 2, respectively. For comparison, we also present the H1 Aktas:2005ty and ZEUS Chekanov:2003ya data.
Note that the ZEUS data points, which were taken at $`W=89`$ GeV and at $`Q^2=9.6`$ GeV<sup>2</sup>, have been rescaled to the H1 values of $`W=82`$ GeV and $`Q^2=8`$ GeV<sup>2</sup> using the fitted $`W`$ and $`Q^2`$-dependence of the DVCS cross section: $`\sigma _{\mathrm{DVCS}}W^{0.75}`$ and $`\sigma _{\mathrm{DVCS}}1/(Q^2)^{1.54}`$ Chekanov:2003ya .
For the proton forward PDFs, which are required to evaluate $`Q_0^i`$, we used the LO CTEQ5L parameterization Lai:1999wy .
One can see from Fig. 1 that the absolute value and the $`Q^2`$-dependence of the total DVCS cross section is described very well. The agreement with the data at the highest values of $`Q^2`$ would have been worse, if we had used the $`Q^2`$-independent slope $`B`$.
From Fig. 2 one can see that the absolute value and the $`W`$-dependence of the DVCS cross section is also reproduced rather well. However, one should note a slight discrepancy between the ZEUS and H1 data points at lower values of $`W`$ and large experimental errors at the high end of $`W`$.
It is important to emphasize that our predictions for the total DVCS cross section were made using the parameterization of the GPD, which contains no free parameters (the role of $`Q_2^i`$ and $`\beta `$, see Eq. (7), is negligible in the H1 and ZEUS kinematics). It is very remarkable that the agreement with the data is so good.
In order to understand, at least partially, the success of the dual parameterization of GPDs in the description of the low-$`x`$ HERA data, it is instructive to analyze the DVCS amplitude $`𝒜_{\mathrm{DVCS}}`$ of Eq. (15) in some detail. Evaluating the imaginary and real parts of Eq. (15), one obtains Polyakov:2002wz
$`\mathrm{Im}𝒜(\xi ,Q^2)={\displaystyle \underset{i}{}}e_i^2{\displaystyle _a^1}{\displaystyle \frac{dx}{x}}{\displaystyle \frac{1}{\sqrt{2x/\xi x^21}}}{\displaystyle \underset{k=0}{\overset{2}{}}}x^kQ_k(x,0,Q^2),`$
$`\mathrm{Re}𝒜(\xi ,t)={\displaystyle \underset{i}{}}e_i^2{\displaystyle _a^1}{\displaystyle \frac{dx}{x}}{\displaystyle \underset{k=0}{\overset{2}{}}}x^kQ_k(x,0,Q^2)\left({\displaystyle \frac{1}{\sqrt{1+2x/\xi +x^2}}}2\delta _{k0}\right)`$
$`{\displaystyle \underset{i}{}}e_i^2{\displaystyle _0^a}{\displaystyle \frac{dx}{x}}{\displaystyle \underset{k=0}{\overset{2}{}}}x^kQ_k(x,0,Q^2)\left({\displaystyle \frac{1}{\sqrt{12x/\xi +x^2}}}+{\displaystyle \frac{1}{\sqrt{1+2x/\xi +x^2}}}2\delta _{k0}\right),`$ (17)
where $`a=(1\sqrt{1\xi ^2})/\xi `$.
Al low $`x_{Bj}`$, $`\xi x_{Bj}/2`$ and the integration limit is $`a\xi /2=x_{Bj}/4`$. Thus, the functions $`Q_0^i`$ and $`Q_2^i`$ are never sampled at $`x<x_{Bj}/4`$. This is clearly an advantage over the double distribution parameterization of GPDs, where the forward parton distributions are sampled all the way down to $`x=0`$, which results in the acute sensitivity to the unmeasured, very low-$`x`$ behavior of the proton PDFs and leads to a gross overestimate of the data Freund:2001hd .
In addition, Eqs. (17) are convenient for the numerical implementation since the integrands do not contain large end-point contributions, as can be explicitly seen by changing the integration variables.
## IV Discussion and Conclusions
We presented and discussed a new leading order parameterization of GPDs introduced in Polyakov:2002wz . In its minimal form, the parameterization is defined by the forward singlet quark PDFs and the form factors of the energy-momentum tensor, see Eqs. (4) and (6). The $`t`$-dependence of the DVCS cross section was assumed in a simple factorized form with the $`Q^2`$-dependent slope, see Eqs. (10) and (12).
We showed that our parameterization of the GPDs describes very well the absolute value, the $`Q^2`$-dependence and $`W`$-dependence of the HERA data on the total DVCS cross section. Moreover, since the data is at low $`x_{Bj}`$, our parameterization can be simplified by omitting the contribution of the $`Q_2^i`$ function. This means that we achieved a remarkably good description of a large set of the data on DVCS using a parameterization of the GPDs which contains no free parameters!
We discuss that our parameterization suits the low-$`x_{Bj}`$ kinematics especially well because the quark singlet PDFs are never probed at the unmeasurably low values of Bjorken $`x`$, as is the case for the popular double distribution model Freund:2001hd , and because the expression for the DVCS amplitude is numerically stable. This allows us to advertise our model as a better alternative to the popular double distribution parameterization of the GPDs, at least in the low-$`\xi `$ region.
The parameterization presented in this work can be readily generalized by including more $`Q_k^i`$ functions, considering the GPDs $`E`$, $`\stackrel{~}{H}`$ and $`\stackrel{~}{E}`$ and by using more elaborate models of the $`t`$-dependence. This was not necessary in the H1 Aktas:2005ty and ZEUS Chekanov:2003ya kinematics, but might be required for the HERMES and CLAS kinematics.
Also, the role of next-to-leading order (NLO) corrections and higher twist effects should be investigated. In particular, it is important to compare the size of the NLO corrections using the dual parameterization with the results of the analysis using the double distribution parameterization, where the NLO corrections were found to be large Freund:2001hd .
## Acknowledgements
We would like to thank M. Strikman and P. Pobylitsa for valuable discussions and comments. It is a pleasure to thank A. Freund for carefully reading the initial draft of the manuscript and making valuable suggestions. This work is supported by the Sofia Kovalevskaya Program of the Alexander von Humboldt Foundation. |
warning/0507/hep-ex0507047.html | ar5iv | text | # 1 Introduction
## 1 Introduction
In the Standard Model of elementary particle interactions, the strong interaction is described by the theory of Quantum Chromodynamics (QCD), and depends on just one fundamental parameter, the strong coupling $`\alpha _\mathrm{s}`$. The value of $`\alpha _\mathrm{s}`$ is expected to depend on the energy scale of the interaction. It is therefore an important test of the theory to determine the value of $`\alpha _\mathrm{s}`$ experimentally at as many different energies as possible. It is also important to use as many different techniques as possible, as different measurements of $`\alpha _\mathrm{s}`$ are sensitive to different theoretical and hadronization variations.
Indeed, many methods have already been employed to evaluate $`\alpha _\mathrm{s}`$ . At very low energies the value of $`\alpha _\mathrm{s}`$ can be measured using the hadronic decays of the $`\tau `$ lepton and heavy quarkonia. Low energy determinations are also available using scaling violations and sum rules from deep inelastic scattering experiments. Higher energy determinations of $`\alpha _\mathrm{s}`$ come from collider experiments ($`\mathrm{e}^+\mathrm{e}^{}`$ , pp, p$`\overline{\mathrm{p}}`$ or ep) using properties of the created hadron system which are explicitly dependent on the value of $`\alpha _\mathrm{s}`$($`Q`$), where $`Q`$ corresponds to the energy scale at which the interaction takes place<sup>1</sup><sup>1</sup>1For $`\mathrm{e}^+\mathrm{e}^{}`$ collisions $`Q=`$$`\sqrt{s}`$..
During the LEP1.5 ($`\sqrt{s}`$ $``$133 GeV) and LEP2 (above $`\mathrm{W}^+\mathrm{W}^{}`$ threshold) operational phases of the Large Electron-Positron collider at CERN, events were recorded with centre-of-mass collision energies ranging from 91 GeV to 209 GeV. Events of the form $`\mathrm{e}^+\mathrm{e}^{}`$ $``$hadrons can be used to determine distributions based on the ensemble of final state hadrons (event shapes) or on the ensemble of jets (jet rates). Previous results by OPAL for an $`\alpha _\mathrm{s}`$ determination based on event shapes and jet rates using the $`\mathrm{Z}`$ dataset collected during the LEP1 phase can be found in . Determinations of $`\alpha _\mathrm{s}`$ from LEP1.5 and LEP2 datasets up to 189 GeV have already been reported by OPAL based on event shape distributions and on jet rates . Another OPAL paper uses the same data that have been presented here to measure event shapes.
For the analysis presented in this paper we used data collected during the LEP1.5 and LEP2 phases to construct jet rate distributions using several jet clustering algorithms. The differential two-jet rate, $`D_2`$, and the average jet rate, $`N`$, were used to determine values of $`\alpha _\mathrm{s}`$($`\sqrt{s}`$) at the four combined centre-of-mass energies composed of data within the LEP1.5 and LEP2 datasets. Theoretical predictions were fitted to these distributions to extract the value of $`\alpha _\mathrm{s}`$($`\sqrt{s}`$).
The paper is organized as follows. Section 2 contains a brief description of the OPAL detector. A summary of the data and the Monte Carlo samples used in the analysis is given in Section 3. In Section 4, we define the jet rate distributions. The methods used to select signal events and reject backgrounds are presented in Section 5. The variations used for systematic studies are detailed in Section 6. Finally, the results of this analysis are given in Section 7, followed by a conclusion and summary in Section 8.
## 2 The OPAL Experiment
A full description of the OPAL detector can be found in . The critical components of the detector in the identification of jets were the central tracking chambers, which were used to reconstruct charged particles, and the electromagnetic calorimeters, which measured the total energy deposited by electrons and photons.
The tracking chambers were located inside a solenoidal magnet which provided a 0.435 T axial magnetic field along the beam axis. The main component of the tracking system was a large-volume jet chamber, which was approximately 4.0 m long with an outer radius of 1.85 m. The jet chamber was separated into 24 sectors, each with a radial plane of 159 sense wires separated by 1 cm. The momenta of tracks in the $`xy`$ plane<sup>2</sup><sup>2</sup>2The right-handed OPAL coordinate system is defined so that $`z`$ is the coordinate parallel to the e<sup>-</sup> beam direction and the $`x`$ axis points to the centre of the LEP ring, $`r`$ is the distance normal to the $`z`$ axis, $`\theta `$ is the polar angle with respect to the $`z`$ axis and $`\varphi `$ is the azimuthal angle with respect to the $`x`$ axis. were measured with a precision parametrized by $`\sigma _p/p=\sqrt{0.02^2+(0.0015p[\mathrm{GeV}/c])^2}`$.
The calorimetry systems were outside the solenoidal magnet. The electromagnetic calorimeter was composed of 11704 lead glass blocks in the barrel and endcap regions, representing about 25 radiation lengths in the barrel and more than 22 in the endcap. The iron sampling hadron calorimeter was located just outside the electromagnetic calorimeter, and provided the stopping power to contain most hadronic showers. Luminosity was determined using small-angle Bhabha events detected in the forward detectors and silicon-tungsten calorimeter .
After an event was triggered , data were collected from the subdetectorsand processed by the OPAL data acquisition system . The raw event data were transferred to a farm of computer processors where the events were fully reconstructed and written to tape for offline analysis.
## 3 Data and Monte Carlo Samples
The data used in this analysis were collected by OPAL between 1995 and 2000 and correspond to integrated luminosities of 14.7 pb<sup>-1</sup> of data taken with centre-of-mass energy 91 GeV, 11.3 pb<sup>-1</sup> of LEP1.5 data with centre-of-mass energies between 130 GeV and 136 GeV and 707.4 pb<sup>-1</sup> of LEP2 data with centre-of-mass energies ranging from 161 to 209 GeV. The 91 GeV data, known as $`\mathrm{Z}`$-calibration data, were primarily collected for calibrating parameters used in the OPAL reconstruction algorithms. This $`M_\mathrm{Z}`$ sample had the same detector configuration as the other centre-of-mass energy points. The exact breakdown of the centre-of-mass energies together with the respective luminosities and numbers of selected events are given in Table 1. The thirteen points in Table 1 represent the main samples of the spread of energies in the LEP1.5 and LEP2 data.
The data were combined into four datasets. The LEP1.5 data provided a single energy point at an event-weighted centre-of-mass energy of 133 GeV, while the LEP2 data were split into two energy points, one with an event-weighted centre-of-mass energy of 177 GeV using data in the range 161–185 GeV (with a total integrated luminosity of 78.1 pb<sup>-1</sup>) and another at 197 GeV (with a total integrated luminosity of 628.3 pb<sup>-1</sup>) using data in the range 188–209 GeV. Together with the $`\mathrm{Z}`$-calibration data this provided for a determination of $`\alpha _\mathrm{s}`$ at four centre-of-mass energies.
A number of Monte Carlo samples were created to correct for detector acceptance and resolution effects, to correct for hadronization effects and to estimate the contribution of background processes. These Monte Carlo samples were produced using a full simulation of the detector , followed by the same reconstruction and selection algorithms applied to the real data, and are referred to as “detector-level” samples. Other samples without the full detector simulation are discussed in Section 5.2.
PYTHIA 6.150 was used to provide the default Monte Carlo samples (for the process $`\mathrm{e}^+\mathrm{e}^{}`$ $`\mathrm{Z}/\gamma ^{}`$$`\mathrm{q}\overline{\mathrm{q}}`$ $``$ hadrons) which were used to correct the high energy datasets. The $`\mathrm{Z}`$-calibration dataset was corrected using JETSET 7.408 . The use of JETSET for the lower energy data is a matter of convenience only, and not due to any inconsistencies in PYTHIA at this energy. Any differences between the two generators is expected to be negligible. Hadronization corrections were evaluated by comparing results with an alternative Monte Carlo sample, HERWIG 6.2 which uses the cluster model of hadronization. This was compared with the string model of hadronization in PYTHIA. The parameters which were involved in the Monte Carlo simulation, both for JETSET/PYTHIA and HERWIG, were tuned to OPAL data collected at the $`\mathrm{Z}`$ peak, including global event shapes, particle multiplicities and fragmentation functions . The generation of the initial quark-antiquark pair for each Monte Carlo sample was implemented at LEP2 using the $`𝒦𝒦`$2f 4.13 event generator , which has an improved description of photon production in the initial and final states with respect to the one currently implemented in the PYTHIA generator. The available detector-level Monte Carlo samples are listed in Table 1.
Above the $`\mathrm{W}^+\mathrm{W}^{}`$ production threshold (161 GeV), the main background was expected to come from four-fermion events ($`\mathrm{e}^+\mathrm{e}^{}`$ $``$$`\mathrm{W}^+\mathrm{W}^{}`$$``$4f), in particular those events in which two or all four of the fermions were quarks. The contribution of these backgrounds in data was estimated using Monte Carlo samples generated using KORALW 1.42 (for q$`\overline{\mathrm{q}}`$q$`{}_{}{}^{}\overline{\mathrm{q}}_{}^{}`$ and q$`\overline{\mathrm{q}}\mathrm{}\overline{\mathrm{}}^{()}`$ where $`\mathrm{}=\mathrm{e},\mu ,\tau ,\nu `$ but $`\mathrm{}\overline{\mathrm{}}\mathrm{e}^+\mathrm{e}^{}`$) and grc4f 2.1 (for eeq$`\overline{\mathrm{q}}`$). Grc4f 2.1 was used to generate all the expected four-fermion background samples for the 161 and 172 GeV data. The background distributions were normalized to the luminosity of the dataset and subsequently subtracted from the measured distributions. The LEP1.5 energies were well below the $`\mathrm{W}^+\mathrm{W}^{}`$ and $`\mathrm{ZZ}`$ production thresholds and were therefore expected to have no significant four-fermion backgrounds. The total expected background contribution from “four-fermion” $`\mathrm{e}^+\mathrm{e}^{}\mathrm{q}\overline{\mathrm{q}}\mathrm{q}\overline{\mathrm{q}}`$ events is 1.2% of the combined LEP1.5 data sample and it was neglected in the analysis.
## 4 Jet Rate Distributions
Jets were formed from the final state objects by applying jet clustering algorithms. These algorithms use the kinematic and spatial (geometric) properties of the individual objects in order to classify them as belonging to a specific jet. We used here the Durham , Cambridge , JADE and the $`R`$ and $`\epsilon `$ variants of the Cone jet clustering algorithms.
The Durham and Cambridge algorithms construct a test variable built from the energy and angular separation between two particles,
$`y_{ij}={\displaystyle \frac{2\mathrm{min}\{E_i^2,E_j^2\}(1\mathrm{cos}\theta _{ij})}{E_{\mathrm{vis}}^2}}`$
where $`E_i`$ is the energy of particle $`i`$, $`\theta _{ij}`$ the angle between the particle $`i`$ and $`j`$ and $`E_{\mathrm{vis}}`$ is the total visible energy in the event. The pair that produces the smallest value of $`y_{ij}`$ is chosen first. The value of this test variable is compared to a predefined parameter, $`y_{\mathrm{cut}}`$, called the jet resolution parameter. If the test variable is smaller than $`y_{\mathrm{cut}}`$ particles $`i`$ and $`j`$ are merged into a pseudo-particle. Merging means that the momenta of particles $`i`$ and $`j`$ are removed from the set of momenta and the the sum of their four-momenta is added to the set of momenta. After the merging, the clustering starts again using the momentum set and it continues until all test variables become larger than $`y_{\mathrm{cut}}`$. After the clustering stops, all remaining (pseudo-) particles are classified as jets.
The Cambridge algorithm differs slightly from Durham in its implementation. In the Cambridge algorithm particles are first paired together by minimizing the variable $`v_{ij}=2(1\mathrm{cos}\theta _{ij})`$. The standard test variable is then constructed and compared to the jet resolution parameter, $`y_{\mathrm{cut}}`$. The procedure followed is then identical to that of the Durham algorithm, except that Cambridge freezes out soft jets by accepting only the lowest energy (pseudo-)particle as the jet when $`y_{ij}>y_{\mathrm{cut}}`$. The number of jets reconstructed in the event, using Durham or Cambridge, is therefore a function of the jet resolution parameter. The JADE algorithm follows the same procedure as the Durham algorithm; however, it uses the scaled invariant mass, $`y_{ij}=2E_iE_j(1\mathrm{cos}\theta _{ij})/E_{\mathrm{vis}}^2`$, of particles $`i`$ and $`j`$ as the test variable.
In the Cone jet finding algorithm, a jet is defined as a set of particles whose three-momentum vectors lie inside a cone of half angle $`R`$, where the direction of the sum of their three-momentum vectors defines the cone axis. In addition, the total energy of the particles assigned to a jet is required to exceed some minimum value $`\epsilon `$. Typical values are $`R=0.7`$ rad and $`\epsilon =7`$ GeV for jets in $`\mathrm{e}^+\mathrm{e}^{}`$ annihilation at LEP 1 energies. When analysing events at the detector level, we replaced $`\epsilon `$ by $`\epsilon ^{}=\epsilon E_{\mathrm{vis}}/`$$`\sqrt{s}`$ to compensate for the incomplete detection of the energy of the event. In our studies, the jet rate was computed at fixed $`\epsilon =7`$ GeV as $`R`$ was varied, and at fixed $`R=0.7`$ as $`\epsilon `$ was varied. The former is sensitive to the angular structure of jets, and the latter to their energy distribution.
The fraction of multihadronic events in a given sample that are classified as containing $`n`$ jets for a given value of the jet resolution parameter ($`y_{\mathrm{cut}}`$, $`R`$ or $`\epsilon `$) is referred to as the $`n`$-jet rate. This $`n`$-jet rate is explicitly defined as
$$R_n(y_{\mathrm{cut}})=\frac{\sigma _n(y_{\mathrm{cut}})}{\sigma _{\mathrm{tot}}}\frac{N_n(y_{\mathrm{cut}})}{N_{\mathrm{tot}}},$$
(1)
where $`\sigma _n`$ is the cross-section for the production of a hadronic event with $`n`$ jets at fixed $`y_{\mathrm{cut}}`$, $`\sigma _{tot}`$ is the total hadronic cross-section, $`N_n(y_{\mathrm{cut}})`$ is the number of events in a sample with $`n`$ jets for a given value of $`y_{\mathrm{cut}}`$ and $`N_{tot}`$ is the total number of events in that sample.
The differential $`n`$-jet rate was also determined. It is the derivative of the $`n`$-jet rate with respect to $`y_{\mathrm{cut}}`$,
$`D_n(y_{\mathrm{cut}})`$ $`=`$ $`{\displaystyle \frac{\mathrm{d}R_n(y_{\mathrm{cut}})}{\mathrm{d}y_{\mathrm{cut}}}}.`$ (2)
For the case when $`n=2`$, the differential 2-jet rate reduces to $`D_2=y_{23}`$, where $`y_{23}`$ is the value of the jet resolution parameter where the event flips from a 2- to a 3-jet event. When the Durham algorithm is used to define jets the value of $`D_2`$ (denoted $`y_{23}^D`$) is also an event shape variable.
The average number of jets per event in a given sample, as a function of the jet resolution parameter, is defined to be
$`N(y_{\mathrm{cut}})`$ $`=`$ $`{\displaystyle \frac{1}{\sigma _{\mathrm{tot}}}}{\displaystyle \underset{n}{}}n\sigma _n(y_{\mathrm{cut}})`$ (3)
$`=`$ $`{\displaystyle \frac{1}{N_{\mathrm{tot}}}}{\displaystyle \underset{n}{}}nN_n(y_{\mathrm{cut}}).`$
A QCD prediction which matches an $`𝒪(\alpha _\mathrm{s}^2)`$ (next-leading order) prediction, based on the QCD matrix elements , with a resummed, next-leading logarithmic approximation (NLLA) prediction, such that terms that appear in both predictions are not double counted, was fitted to data. In this analysis we used the $`\mathrm{ln}R`$ matched $`D_2`$ and $`N`$ predictions to fit to the distributions of the observables. The differential and average jet rates were determined using the Durham and Cambridge clustering algorithms, since resummed predictions only exist for these algorithms. This provided four separate observables ($`D_2^D`$, $`D_2^C`$, $`N^D`$ and $`N^C`$) which were used to determine a value of $`\alpha _\mathrm{s}`$ at the four different centre-of-mass energy values.
## 5 Analysis Procedure
### 5.1 Selection Method
#### 5.1.1 Preselection
All events within a dataset were required to contain information from both the central jet chamber and the electromagnetic calorimeter, meaning both these subdetectors must have been flagged as being on and in good operational condition. In addition events were required to be tagged as multihadronic in order to be preselected for analysis. Multihadronic events were identified using the criteria described in for events with $`\sqrt{s}`$ $`>`$$`M_\mathrm{Z}`$ and in for $`\mathrm{Z}`$-calibration events. To pass the preselection, an event was required to contain at least seven good tracks to reduce potential backgrounds arising from the production of $`\tau `$ leptons ($`\mathrm{e}^+\mathrm{e}^{}`$ $`\tau ^+\tau ^{}`$) decaying into hadrons and from two-photon interactions producing quarks. Good tracks were defined as those which had
* at least 40 hits in the jet chamber
* at least 150 MeV/$`c`$ transverse momentum relative to the beam axis.
* the distance of closest approach to the interaction point in the $`r\varphi `$ plane satisfying $`d_02`$ cm
* the point of closest approach $`25`$ cm from the interaction point in the $`z`$-direction
Clusters of energy in the calorimeters were also used in the analysis; good clusters were defined as those which produced a signal in at least one block in the barrel electromagnetic calorimeter corresponding to an uncorrected energy of at least 100 MeV or of 2 blocks in the endcap electromagnetic calorimeter corresponding to an uncorrected energy of 250 MeV. The hadron calorimeter was not used in this analysis.
All of the good quality tracks and clusters in the event were used to define “objects” representing particles using an algorithm (MT) to correct for double counting of energy. This MT algorithm produced a uniquely defined array of track and cluster objects. The trajectories of the tracks measured in the central tracking chambers were extrapolated to the clusters in the electromagnetic calorimeters. If the energy of the cluster was less than expected from the track, then the cluster was omitted to avoid double counting of energy, since the momentum resolution for tracks was typically better than the calorimeter energy resolution. If the energy of the cluster was larger than expected the energy of the cluster was reduced by the expected amount with the remaining energy interpreted as due to photons or neutral hadrons. These remaining clusters and those which were not matched defined the four-vectors of “neutral” particles. In all cases tracks were treated as charged pions and neutral particles were treated as being massless.
#### 5.1.2 Containment
We ensured that most particles in the event were well contained in the detector and not lost down the beam line by imposing a cut on the direction of the thrust axis ,
* $`|\mathrm{cos}\theta _\mathrm{T}|<0.9`$,
where $`\theta _\mathrm{T}`$ is the angle between the beam axis and the direction of the thrust axis. The thrust axis direction was determined from all tracks and clusters in the event, without correcting for double counting with the MT algorithm.
#### 5.1.3 Initial State Radiation (ISR) Cuts
The events of interest for this analysis were $`\mathrm{e}^+\mathrm{e}^{}\mathrm{q}\overline{\mathrm{q}}`$ events where the final-state $`\mathrm{q}\overline{\mathrm{q}}`$ pair had the full centre-of-mass energy. The effective centre-of-mass energy of the $`\mathrm{e}^+\mathrm{e}^{}`$ collision can be reduced by the emission of one or more ISR photons. At LEP2, approximately three quarters of the multihadronic events were such “radiative return events”, where the invariant mass of the $`\mathrm{q}\overline{\mathrm{q}}`$ pair was close to the $`\mathrm{Z}`$ mass. The effective centre-of-mass energy of the collision after ISR, $`\sqrt{s^{}}`$ , was evaluated, and the requirement
* $`\sqrt{s}\sqrt{s^{}}<`$10 GeV
was imposed to select full-energy events.
To calculate $`\sqrt{s^{}}`$, all isolated photon candidates with energies greater than 10 GeV were identified. The Durham jet reconstruction algorithm was then used to group the remaining tracks and clusters into jets. ISR photons are often emitted close to the beam direction. Three kinematic fits were performed, under the assumptions that
* there were two undetected photons (in opposite directions along the beam pipe)
* there was one undetected photon
* all photons were observed in the detector,
respectively. The fit with the most acceptable $`\chi ^2`$ was selected, and $`\sqrt{s^{}}`$ was calculated from the invariant mass of the jets, excluding any photons.
The power of this cut can be seen in Figure 1. The efficiency for selecting non-radiative $`\mathrm{q}\overline{\mathrm{q}}`$ events is given in Table 2. The purity of non-radiative events was found to be approximately 73% in all of the LEP1.5 and LEP2 data samples. Non-radiative $`\mathrm{q}\overline{\mathrm{q}}`$ events are defined as those in which $`\sqrt{s}\sqrt{s_{\mathrm{true}}^{}}<1`$ GeV, where $`s_{\mathrm{true}}^{}`$ was determined from generator-level information in the PYTHIA samples. This ISR cut was applied to all analyzed datasets with the exception of the $`\mathrm{Z}`$ calibration data.
#### 5.1.4 Final Cuts
The dominant background to the process $`\mathrm{e}^+\mathrm{e}^{}`$ $`\mathrm{Z}/\gamma ^{}`$ $`\mathrm{q}\overline{\mathrm{q}}`$ $``$ hadrons at LEP2 came from the four-fermion process $`\mathrm{e}^+\mathrm{e}^{}`$ $``$$`\mathrm{W}^+\mathrm{W}^{}`$ in which one or both of the bosons decayed hadronically, producing two or four quarks in the final state. This background was expected to make up approximately 30% of all observed events which pass the first stage of cuts in each of the LEP2 datasets. These backgrounds were addressed by placing a cut on two likelihood values which indicate how likely an event is to be a non-QCD four-quark or a semi-leptonic event:
* $``$$`_{\mathrm{q}\overline{\mathrm{q}}\mathrm{q}\overline{\mathrm{q}}}`$ $`<0.25`$
* $``$$`_{\mathrm{q}\overline{\mathrm{q}}\mathrm{}\nu }`$ $`<0.50`$
The effect of these cuts in each of the LEP2 datasets and the expected backgrounds can be seen on Figure 2.
The four-quark likelihood value , $``$$`_{\mathrm{q}\overline{\mathrm{q}}\mathrm{q}\overline{\mathrm{q}}}`$, was estimated from four kinematic variables describing characteristics of hadronic $`\mathrm{W}^+\mathrm{W}^{}`$ decays like their four-jet nature and angular structure. These variables were used to construct event probabilities based on two hypotheses: first, that the event was due to a hadronically decaying $`\mathrm{W}^+\mathrm{W}^{}`$ pair ($`\mathrm{W}^+\mathrm{W}^{}`$$``$$`\mathrm{q}\overline{\mathrm{q}}`$ $`\mathrm{q}\overline{\mathrm{q}}`$) and, second, that the event was due to a hadronically decaying $`\mathrm{Z}/\gamma ^{}`$ ($`\mathrm{e}^+\mathrm{e}^{}`$ $`\mathrm{Z}/\gamma ^{}`$$`\mathrm{q}\overline{\mathrm{q}}`$). The probabilities were combined to produce the discriminating likelihood, $``$$`_{\mathrm{q}\overline{\mathrm{q}}\mathrm{q}\overline{\mathrm{q}}}`$. This cut reduced the expected background by approximately 80% so that it constituted only 9% of the observed number of events.
The semi-leptonic likelihood , $``$$`_{\mathrm{q}\overline{\mathrm{q}}\mathrm{}\nu }`$, was based upon three separate likelihoods, one for each lepton species ($`\mathrm{}=`$e, $`\mu `$, $`\tau `$). Each of these likelihoods was based on ten variables describing the properties of the lepton, the jets produced by the $`\mathrm{q}\overline{\mathrm{q}}`$ pair and the missing energy carried away by the neutrino. This cut in conjunction with the cut on the four-quark likelihood removed almost 90% of the background expected in the observed LEP2 dataset. The effect of these cuts can be seen in Figure 2.
These likelihood cuts also reduced the backgrounds arising from $`\mathrm{e}^+\mathrm{e}^{}`$ $``$$`\mathrm{ZZ}`$ in which one or both of the $`\mathrm{Z}`$ bosons decay hadronically. $`\mathrm{ZZ}`$ production contributed only a small fraction of the background due to its lower cross-section compared to $`\mathrm{W}^+\mathrm{W}^{}`$ production in the energy ranges used in this experiment. The likelihood cuts were applied only to those datasets with $`\sqrt{s}`$ $``$161 GeV.
The expected size of the total background contribution to each dataset was determined by Monte Carlo predictions after scaling to the luminosity of the dataset. The effect of the final cuts and the expected four-fermion backgrounds for each centre-of-mass energy dataset can be seen in Table 2. As seen in the table the likelihood cuts greatly increased the purity of selected non-radiative $`\mathrm{q}\overline{\mathrm{q}}`$ events. The LEP2 datasets data with $`\sqrt{s}`$ $``$ 183 GeV were typically $``$70% pure following the ISR cuts; however, after the final cuts this increased to a purity of 94–95%.
### 5.2 Monte Carlo Corrections
The values of the variables $`R_n`$, $`D_n`$ and $`N`$ were determined for each accepted event using the MT-corrected tracks and clusters. These values were then compiled into histograms as a function of $`y_{\mathrm{cut}}`$ with bins of varying size. The background rejection cut did not completely remove all of the expected background events from $`\mathrm{W}^+\mathrm{W}^{}`$ and $`\mathrm{ZZ}`$ production (referred to as four-fermion background in this paper). The remaining backgrounds, taken from the Monte Carlo, were subtracted from the corresponding measured distributions on a bin-by-bin basis. Systematic uncertainties in this procedure will be discussed in Section 6.
Corrections to the distributions were also made for effects arising from finite detector resolution and a limited detector acceptance (recall that the cut on $`|\mathrm{cos}\theta _\mathrm{T}|`$ reduced the fiducial volume) and for residual ISR events which were not removed by the $`\sqrt{s^{}}`$ cut. These corrections were done separately for each variable and were accomplished by comparing distributions from two separate Monte Carlo samples, one of which had gone through a full detector simulation, including effects of detector resolution and acceptance and initial state radiation, called the “detector” level. The other sample used only the generator-level hadrons and had not gone through the detailed detector simulation. In this sample all short-lived particles ($`\tau 3\times 10^{10}`$s) had decayed and a requirement that $`\sqrt{s}\sqrt{s_{\mathrm{true}}^{}}<1`$ GeV was imposed. This “hadron level” sample was thus expected to produce distributions arising solely from the properties of the underlying hadrons, free of any detector biases determined over the full acceptance without any limitations arising from limited resolution.
Correction factors for each bin of the distributions were determined from the ratio of the two Monte Carlo distributions. Thus, any bin, $`i`$, of the measured distributions was corrected via
$$_i^{\mathrm{data}}=\left(\frac{_i^{\mathrm{MC}}}{𝒟_i^{\mathrm{MC}}}\right)(𝒟_i^{\mathrm{data}}𝒟_i^{\mathrm{bkgd}})$$
(4)
where $``$ and $`𝒟`$ represent distributions at the hadron level and the detector level respectively, and $`𝒟_i^{\mathrm{bkgd}}`$ corresponds to the expected size of the total background in bin $`i`$. The hadron level was used in this analysis when determining jet rate distributions; theoretical predictions which were fitted to these distributions were obtained from computations valid at the “parton level”. The parton level corresponds to distributions that would be produced if only the partons created immediately following the $`\mathrm{e}^+\mathrm{e}^{}`$ annihilation and before the hadronization phase were used in the analysis. The parton level Monte Carlo sample was built from quarks and gluons that were produced during the parton shower simulated by the generator before the hadronization phase began. As in the case of the hadron level, the parton level sample gave rise to distributions that were free of initial state radiation without any detector simulation applied. The correction factor determined from the ratio of the hadron level to the parton level was applied to the theoretical predictions before the fitting procedure. This factor corrected the prediction to the hadron level so that it could be compared to the corrected hadron level distribution determined from the data,
$$_i^{\mathrm{pred}}=\left(\frac{_i^{\mathrm{MC}}}{𝒫_i^{\mathrm{MC}}}\right)𝒫_i^{\mathrm{pred}}$$
(5)
where $``$ and $`𝒫`$ represent distributions at the hadron level and the parton level respectively. Statistical errors on the data distributions included the effects of Monte Carlo statistics.
## 6 Systematic Uncertainties
### 6.1 Experimental Systematic Variations
Several selection algorithms and selection cuts were varied to determine their impact on the results of the analysis. In all cases the result from the variation was compared to the result from the standard selection, and the difference was taken as a contribution to the total systematic error. The systematic variations that were used (given in descending order of their contributions to the overall experimental uncertainty) are:
Systematic errors relevant to the definition of objects defined within the tracking chambers and calorimeters, and hence used for jet clustering, were estimated by comparing the results using the MT package to a method using all selected tracks and clusters without taking into account the possibility of double counting.
The uncertainty in modelling the detector was investigated by using HERWIG Monte Carlo datasets in place of PYTHIA to correct from the detector level to the hadron level. The hadronization correction was still performed using the PYTHIA Monte Carlo samples.
The constraint on the direction of the thrust axis was tightened to $`|\mathrm{cos}\theta _T|<0.7`$, restricting events to the barrel region.
A possible systematic effect introduced through the selection of events with little or no initial state radiation was estimated by repeating the selection using a second method of ISR determination . This alternative method assumed while performing the kinematic fit that there was always a single photon which either escaped undetected down the beam line or was detected in the electromagnetic calorimeter.
To account for the systematic uncertainty that arises from the value of the cut placed on the $`\mathrm{W}^+\mathrm{W}^{}`$ hadronic and semi-leptonic likelihoods, the values of the likelihood cuts were changed to 0.1 and 0.4 for $``$$`_{\mathrm{q}\overline{\mathrm{q}}\mathrm{q}\overline{\mathrm{q}}}`$ and to 0.25 and 0.75 for $``$$`_{\mathrm{q}\overline{\mathrm{q}}\mathrm{}\nu }`$. In each case the largest deviation was taken as the systematic uncertainty.
To account for uncertainties introduced during background subtraction, arising from imprecise knowledge of the four-fermion cross-sections, these cross-sections were conservatively varied by $`\pm `$5% and the largest deviation from the standard value was used to determine an overall systematic error.
It should be noted that there was no single dominant contribution to the overall experimental systematic error. In general, the largest contributions occurred for the track-cluster matching and detector correction variations, while the least significant contributions came from the variations of the background cross-section and the semi-leptonic likelihood.
### 6.2 Hadronization Systematic Variations
Systematic uncertainties arise from the modelling of the hadronization. These were estimated by using HERWIG Monte Carlo samples, which employ a different hadronization model, in the determination of the theoretical prediction correction factor (Eqn 5). The PYTHIA-HERWIG differences were taken to be the systematic uncertainties from hadronization. The statistical component due to limited Monte Carlo statistics was included in the total determination of the hadronization uncertianty.
### 6.3 Theoretical Systematic Variations
Three further systematic variations were considered when fitting the predictions to the data to determine $`\alpha _\mathrm{s}`$.
We investigated the choice of the range of bins used in performing the fit of the theoretical prediction to the data. The fit range was first increased by two bins, by adding one bin to each endpoint of the fit range. In the case where one of the endpoints was already at the maximum allowed bin value, only the other point was extended by one bin. A second variation decreased the fit range by two bins, by removing a bin from each endpoint of the fit range. The largest deviation from the standard fit value was taken as the contribution to the systematic error.
The second fit-related systematic variation accounted for the uncertainty due to the dependence of the fixed order and resummed predictions on the renormalization scale, $`x_\mu `$, where $`x_\mu =\mu /\sqrt{s}`$. The value of $`x_\mu `$, which was set to 1.0 in the standard fits, was varied to 0.5 and 2.0 respectively. The largest deviation from the standard fit value was taken as the contribution to the systematic error representing renormalization scale uncertainty.
In the resummation process for event shape variables, like $`D_2`$, there is an arbitrariness due to the definition of the logarithms which were resummed. In this analysis we used $`\alpha _\mathrm{s}\mathrm{ln}(1/y_{\mathrm{cut}})`$; however, this could be generalized to powers of $`\alpha _\mathrm{s}\mathrm{ln}[1/(x_Ly_{\mathrm{cut}})]`$ . The standard value for this rescaling ($`x_L`$=1) was varied to $`\frac{4}{9}`$ and $`\frac{9}{4}`$, to investigate the systematic effect of this arbitrariness. No analoguous rescaling prescription has yet been developed for the case of the average jet rates, so the $`x_L`$ variations are only shown for the $`D_2`$ distributions for comparison to the $`x_\mu `$ variation. Hence this variation was not used in the determination of the total theoretical systematic error and was included only as a cross-check and for comparison with other analyses. A comprehensive study of the combination of various theoretical variations to produce a global theoretical uncertainty for event shape distributions is given in .
The differences between the standard result and those due to the above variations were separated into three categories: experimental, hadronization and theory (see Table 3). The variations in each of these categories were added in quadrature and the result taken as the systematic uncertainty for that category. In the case of asymmetric errors, the error was symmetrized by taking the largest systematic variation and applying it as the full systematic contribution. A summary of the systematic variations is provided in Table 3.
## 7 Results
Data from thirteen datasets were used in this analysis: one $`\mathrm{Z}`$-calibration dataset, two LEP1.5 datasets and ten datasets from LEP2. These thirteen datasets were combined to produce four higher statistics datasets ($`\sqrt{s}`$ =$`M_\mathrm{Z}`$, 133, 177 and 197 GeV) which were analysed separately. The raw distributions ($`n`$-jet fractions, $`D_2`$ and $`N`$) for each of the datasets were determined as functions of the jet resolution parameters defined using four different jet clustering algorithms. The distributions underwent a bin-by-bin correction which included subtraction of expected backgrounds and correction for detector and residual ISR effects. Systematic effects were examined by varying the parameters used in selecting events (see Table 3). The difference between the corrected distributions using these variations and those from the standard selection then determined the size of the systematic error on each bin of the distribution.
Matched predictions of NLO and resummed calculations were fitted to the corrected Durham and Cambridge $`D_2`$ and $`N`$ distributions over a predetermined fit range (see Section 7.2.1), taking into account bin-to-bin correlations in $`N`$ (see Section 7.2.2). These fits provided four values of $`\alpha _\mathrm{s}`$ with statistical and systematic errors at each of the four centre-of-mass energies. Taking into account the statistical and systematic correlations between the four measurements, they were combined into a single $`\alpha _\mathrm{s}`$ result at each energy.
### 7.1 $`𝒏`$-Jet Fractions
The $`n`$-jet fractions for the $`\mathrm{Z}`$-calibration sample and those for the LEP1.5 and the two LEP2 samples are shown in Figures 3 to 7. Each plot shows the fraction of events in a given sample that were determined to be $`n`$-jets for a given value of the jet resolution parameter at the hadron level<sup>3</sup><sup>3</sup>3Further details of the data will be made available in the HEPDATA database, http://durpdg.dur.ac.uk/HEPDATA.. The jet fractions were calculated using four different clustering algorithms. For the Cone algorithm, results from using both the $`R`$ and $`\epsilon `$ resolution parameters are plotted, showing the individual $`n`$-jet fractions for $`n2`$, $`n=3`$ and $`n4`$ in Figures 3 to 4. Similarly the $`n`$-jet fractions for the JADE algorithm are shown in Figure 5 for $`n=2,3,4,5`$. Figures 6 to 7 show the individual $`n`$-jet fractions for $`n=2,3,4,5`$ for the Durham and Cambridge algorithms respectively. The error bars on the points represent the total statistical and systematic errors added in quadrature. The Monte Carlo expectations corresponding to PYTHIA and HERWIG are also displayed on each plot, for each algorithm and energy. The Monte Carlo expectations match the measured $`n`$-jet fractions reasonably well.
The differential two-, three- and four-jet rates, $`D_n`$, are plotted as a function of $`y_{\mathrm{cut}}`$ for the Durham and Cambridge algorithms for the $`\mathrm{Z}`$-calibration, the LEP1.5 and the two LEP2 data samples in Figures 8 and 9 respectively. Similarly, the average jet rates are plotted as a function of $`y_{\mathrm{cut}}`$ for the Durham and Cambridge algorithms for the $`\mathrm{Z}`$-calibration, the LEP1.5 and the two LEP2 data samples in Figures 10 and 11 respectively. The curves on all the plots represent the expected Monte Carlo distributions. There is good agreement between the data and the expectations from both PYTHIA and HERWIG.
### 7.2 Fits to Determine the Value of $`𝜶_𝐬`$
#### 7.2.1 Differential Two-jet Rates
The range over which the $`D_2`$ and $`N`$ distributions were fitted was determined by splitting the 91 GeV (and 189 GeV) PYTHIA Monte Carlo samples into 100 statistically independent subsamples. The QCD predictions for each distribution were fitted to each of these subsamples, with $`\alpha _\mathrm{s}`$ as a free parameter, for a number of possible end-points of the fit range (requiring that the fit range be at least six bins). A $`\chi ^2`$ per degree of freedom was determined for each fit range. The $`\chi ^2`$ values were averaged over the 100 subsamples. The fit range that produced the smallest average $`\chi ^2`$ per degree of freedom was chosen to be the default fit range for the $`\mathrm{Z}`$-calibration (and high energy) datasets. Where no clear $`\chi ^2`$ minimum was found, the largest reasonable range was chosen. The size of the fit range was then adjusted to ensure the range did not extend into a region where the hadronization corrections exceeded 10%, in particular at smaller $`y_{\mathrm{cut}}`$ values. A further adjustment was made to exclude fit ranges where one of the endpoints produced a contribution of more than 30% to the overall $`\chi ^2`$ value. Potential correlations introduced in the correction of the distributions were accounted by including a covariance matrix in the $`\chi ^2`$ fit. The covariance matrices were determined in the manner detailed in Section 7.2.2.
The fits of the $`\mathrm{ln}R`$ matching prediction to the $`D_2`$ distribution are shown in Figure 12 for the Cambridge algorithm and in Figure 13 for the Durham algorithm. The values of $`\alpha _\mathrm{s}`$ determined from the $`\mathrm{ln}R`$ fits for the four datasets, together with a complete breakdown of statistical and systematic uncertainties, are given in Tables 5 and 5 for the Cambridge and Durham<sup>4</sup><sup>4</sup>4Note that also determines $`\alpha _\mathrm{s}`$ using $`D_2^D`$, denoted there as $`y_{23}^D`$. The small differences between the results have been investigated in detail, and are not significant. They may be attributed to differences in fit regions, the use of statistically different Monte Carlo samples, and the adoption of slightly different strategies for the assessment of theoretical errors. algorithms respectively.
#### 7.2.2 Average Jet Rates
For the average jet rates, $`N`$, the statistical errors were strongly correlated between points, since the same events were used to determine $`N`$ at each value of $`y_{\mathrm{cut}}`$. The fit was performed using a correlated $`\chi ^2`$ fit in which the covariance matrix was determined from the PYTHIA Monte Carlo sample divided into many detector level subsamples. Each subsample was then corrected to the hadron level using a second, statistically independent, PYTHIA sample. There were 1500 subsamples created for the $`\mathrm{Z}`$-calibration dataset and 1000 subsamples created for the high energy datasets. The number of subsamples was chosen so that the elements of the covariance matrix would be stable and would have negligible fluctuations. These subsamples were used to build a standard covariance matrix, which was then converted into a correlation matrix. The statistical errors on the bins of the data distributions were then applied to this correlation matrix to produce the covariance matrix used in the fits.
An example of the fit of the $`\mathrm{ln}R`$ matching prediction to the $`N`$ distribution is seen in Figure 14 for the Cambridge algorithm and in Figure 15 for the Durham algorithm. The values of $`\alpha _\mathrm{s}`$ determined from the $`\mathrm{ln}R`$ fits for the four datasets, together with a complete breakdown of statistical and systematic uncertainties, are given in Tables 7 and 7 for the Cambridge and Durham algorithms respectively.
#### 7.2.3 Running of $`𝜶_𝐬`$
The four values of $`\alpha _\mathrm{s}`$ determined from the $`D_2`$ and $`N`$ distributions were combined into a single value at each centre-of-mass energy. The large statistical correlations between the four values were handled in a manner similar to that for the bin-to-bin correlations of the average jet rates. For each of the four distributions, 1000 Monte Carlo samples were used to determine the statistical correlations between the $`\alpha _\mathrm{s}`$ values. The correlation matrices determined for the four separate energy points are given in Table 8. Using the statistical error for the $`\alpha _\mathrm{s}`$ value from each observable, the statistical covariance matrix was then determined. The full covariance matrix also included contributions from experimental, hadronization and theoretical uncertainties
$$V=V_{\mathrm{stat}}+V_{\mathrm{expt}}+V_{\mathrm{hadr}}+V_{\mathrm{theo}}.$$
(6)
A weight was determined for each of the $`\alpha _\mathrm{s}`$ values from the inverse of the covariance matrix, $`w_i=_j(V^1)_{ij}/_{ij}(V^1)_{ij}`$. The combined $`\alpha _\mathrm{s}`$ value was then determined from the weighted sum of the $`\alpha _\mathrm{s}`$ values. Only statistical and experimental systematic uncertainties were allowed to contribute to the off diagonal elements of the covariance matrix $`V`$, to ensure undesirable features such as negative weights were avoided. Hadronization and theoretical systematics were added only to the diagonal elements of the covariance matrix.
The statistical and experimental uncertainties on the combined $`\alpha _\mathrm{s}`$ value were determined from the product of the weights with the individual covariance matrices ,
$$\sigma _{\mathrm{err}}^2=w^TV_{\mathrm{err}}w\mathrm{where}\mathrm{err}=\mathrm{stat},\mathrm{expt}.$$
(7)
The hadronization and theoretical systematic uncertainties were determined by repeating the combination for each systematic variation separately using the same weights. The difference between these systematic combinations and the central value is taken as the systematic contribution to the error on the central value. For the experimental systematic covariance matrix, off diagonal elements were determined using a “minimum overlap” method,
$$(V_{\mathrm{expt}})_{ij}=\mathrm{max}[(V_{\mathrm{expt}})_{ii},(V_{\mathrm{expt}})_{jj}].$$
(8)
The values of $`\alpha _\mathrm{s}`$ determined for each centre-of-mass energy are given in Table 9 along with the breakdown of the uncertainties, both statistical and systematic. A comparison of the combined $`\alpha _\mathrm{s}`$ values in Table 9 with those determined from the individual $`D_2`$ and $`N`$ distributions is seen in Figure 16. Taking the combined $`\alpha _\mathrm{s}`$ values at each centre-of-mass energy as an input, the value of $`\alpha _\mathrm{s}`$ can be run back to the $`\mathrm{Z}`$ pole using an $`𝒪(\alpha _\mathrm{s}^3)`$ prediction. These $`\alpha _\mathrm{s}`$($`M_\mathrm{Z}`$) values are also shown in Table 9 and plotted against the world average value of $`\alpha _\mathrm{s}`$($`M_\mathrm{Z}`$)=0.1187$`\pm `$0.0020 in Fig 17. Using these values and taking into account proper statistical and systematic correlations, a weighted mean of $`\alpha _\mathrm{s}`$($`M_\mathrm{Z}`$)=$`0.1177\pm 0.0006`$(stat.)$`\pm 0.0012`$(expt.)$`\pm 0.0010`$(had.)$`\pm 0.0032`$(theo.) is determined. The four combined $`\alpha _\mathrm{s}`$ values are plotted in Figure 18 as a function of the centre-of-mass energy, compared to the $`𝒪(\alpha _\mathrm{s}^2)`$ energy evolution of $`\alpha _\mathrm{s}`$ based on the determined value of $`\alpha _\mathrm{s}`$($`M_\mathrm{Z}`$).
## 8 Summary and Conclusion
Data from twelve LEP 1.5 and LEP 2 datasets, with centre-of-mass energies ranging from 130 GeV to 209 GeV, were combined into three higher statistics datasets. These datasets and one combining several $`\mathrm{Z}`$-calibration runs at 91 GeV were used to determine the $`n`$-jet fractions, the differential $`n`$-jet rates and the average jet rates for each of the energies. The different jet multiplicity distributions were compared to both PYTHIA and HERWIG Monte Carlo expectations.
Hadron-level $`n`$-jet fractions were determined using four jet-clustering algorithms, Cone, JADE, Durham and Cambridge. For the Cone algorithm, measurements of the fraction of events with $`n2`$, 3, $``$4 jets were presented as functions of $`R`$ and $`\epsilon `$. In the case of JADE, Durham and Cambridge, measurements of the 2-, 3-, 4-, and 5-jet fractions were presented as functions of the jet resolution parameter, $`y_{\mathrm{cut}}`$. In all cases there was generally good agreement between the measured jet fractions and the Monte Carlo expectations.
Hadron-level determinations of the differential $`n`$-jet and average jet rates were performed for the Durham and Cambridge algorithms. The differential two-jet rate, $`D_2`$, and the average jet rate, $`N`$ were used to determine the value of $`\alpha _\mathrm{s}`$($`\sqrt{s}`$) for each of the four combined energy points. The determinations were carried out by fitting the ln$`R`$ matching predictions, appropriately corrected to the hadron level, to the hadron level data distributions over an appropriate range of $`y_{\mathrm{cut}}`$, with $`\alpha _\mathrm{s}`$($`\sqrt{s}`$) as the variable parameter. The running of $`\alpha _\mathrm{s}`$($`\sqrt{s}`$) was demonstrated by comparing the four values of $`\alpha _\mathrm{s}`$ as determined from the combined datasets as a function of their centre-of-mass energies.
Using the measured values of $`\alpha _\mathrm{s}`$($`\sqrt{s}`$) a value of $`\alpha _\mathrm{s}`$($`M_\mathrm{Z}`$) was determined for each of the four datasets. A weighted mean taking account of correlations determined a final value of the strong coupling at $`\sqrt{s}`$ =$`M_\mathrm{Z}`$ of
$$\alpha _\mathrm{s}(M_\mathrm{Z})=0.1177\pm 0.0006(\mathrm{stat}.)\pm 0.0012(\mathrm{expt}.)\pm 0.0010(\mathrm{had}.)\pm 0.0032(\mathrm{theo}.)$$
where the error contains contributions from statistical, experimental, hadronization and theoretical uncertainties. The error on the determined value is slightly larger than that for the previous OPAL publication which also used resummed predictions for $`D_2`$ and average jet rate distributions, but explored slightly different energy ranges, including 35 and 45 GeV, and all LEP energies up to only 189 GeV. There is good agreement between these four values of $`\alpha _\mathrm{s}`$($`M_\mathrm{Z}`$) measured in this analysis and previous determinations of $`\alpha _\mathrm{s}`$ summarized in and with the world average value of 0.1187$`\pm `$0.0020 .
## Acknowledgements
We particularly wish to thank the SL Division for the efficient operation of the LEP accelerator at all energies and for their close cooperation with our experimental group. In addition to the support staff at our own institutions we are pleased to acknowledge the
Department of Energy, USA,
National Science Foundation, USA,
Particle Physics and Astronomy Research Council, UK,
Natural Sciences and Engineering Research Council, Canada,
Israel Science Foundation, administered by the Israel Academy of Science and Humanities,
Benoziyo Center for High Energy Physics,
Japanese Ministry of Education, Culture, Sports, Science and Technology (MEXT) and a grant under the MEXT International Science Research Program,
Japanese Society for the Promotion of Science (JSPS),
German Israeli Bi-national Science Foundation (GIF),
Bundesministerium für Bildung und Forschung, Germany,
National Research Council of Canada,
Hungarian Foundation for Scientific Research, OTKA T-038240, and T-042864,
The NWO/NATO Fund for Scientific Research, the Netherlands. |
warning/0507/gr-qc0507054.html | ar5iv | text | # Accelerated Detector - Quantum Field Correlations: From Vacuum Fluctuations to Radiation Flux
## I Introduction
Inasmuch as studies of the interaction between a particle and a quantum field are basic to particle physics and field theory, understanding the interaction between an atom and a quantum field is essential to current atomic and optical physics research CTDG ; CTmovatom ; Milonni ; CPP ; scully . The interaction of an accelerated charge or detector (an object with some internal degrees of freedom such as an atom or harmonic oscillator) in a quantum field is a simple yet fundamental problem with many implications in quantum field theory BD ; DeW75 , thermodynamics BH ; Haw75 and applications in radiation theory and atomic-optical physics.
It is common knowledge that accelerating charges give rise to radiation Jackson . But it is not entirely straightforward to derive the radiation formula from quantum field theory. How are vacuum fluctuations related to the emitted radiation? When an atom or detector moves at constant acceleration, according to Unruh Unr76 , it would experience a thermal bath at temperature $`T_U=\mathrm{}a/(2\pi ck_B)`$, where $`a`$ is the proper acceleration. Is there emitted radiation with an energy flux in the Unruh effect?
Unruh effect, and the related effect for moving mirrors studied by Davies and Fulling DavFul , were intended originally to mimic Hawking radiation from black holes. Because of this connection, for some time now there has been a speculation that there is real radiation emitted from a uniformly accelerated detector (UAD) under steady state conditions (i.e., for atoms which have been uniformly accelerated for a time sufficiently long that transient effects have died out), not unlike that of an accelerating charge Jackson ; Boulware . In light of pending experiments both for electrons in accelerators Chen ; ChenTaj ; BelLen and for accelerated atoms in optical cavities Scullyetal this speculation has acquired some realistic significance. There is a need for more detailed analysis for both the uniform acceleration of charges or detectors and for transient motions because the latter can produce radiation and as explained below, sudden changes in the dynamics can also produce emitted radiation with thermal characteristics.
After Unruh and Wald’s UnrWal earlier explication of what a Minkowski observer sees, Grove Grove questioned whether an accelerated atom actually emits radiated energy. Raine, Sciama and Grove RSG (RSG) analyzed what an inertial observer placed in the forward light cone of the accelerating detector would measure and concluded that the oscillator does not radiate. Unruh Unr92 , in an independent calculation, basically concurred with the findings of RSG but he also showed the existence of extra terms in the two-point function of the field which would contribute to the excitation of a detector placed in the forward light cone. Massar, Parantani and Brout MPB93 (MPB) pointed out that the missing terms in RSG contribute to a “polarization cloud” around the accelerating detector. For a review of earlier work on accelerated detectors, see e.g., AccDetRev . For work after that, see, e.g., Hinterleitner Hin , Audretsch, Müller and Holzmann AMH , Massar and Parantani MP . Our present work follows the vein of Raval, Hu, Anglin (RHA) and Koks RavalPhD ; RHA ; RHK ; CapHR on the minimal coupling model and uses some results of Lin lin03b on the Unruh-DeWitt model Unr76 ; DeW79 .
With regard to the question “Is there a radiation flux emitted from an Unruh detector?” the findings of RSG, Unruh, MPB, RHA and others show that, at least in (1+1) dimension model calculations, there is no emitted radiation from a linear uniformly accelerated oscillator under equilibrium conditions, even though, as found before, that there exists a polarization cloud around it. Hu and Johnson CapHJ emphasized the difference between an equilibrium condition (steady state uniform acceleration) where there is no emitted radiation, and nonequilibrium conditions where there could be radiation emitted. Nonequilibrium conditions arise for non-uniformly accelerated atoms (for an example of finite time acceleration, see Raval, Hu and Koks (RHK) RHK ), or during the initial transient time for an atom approaching uniform acceleration, when its internal states have not yet reached equilibrium through interaction with the field. Hu and Raval (HR) CapHR ; RavalPhD presented a more complete analysis of the two-point function, calculated for two points lying in arbitrary regions of Minkowski space. This generalizes the results of MPB in that there is no restriction for the two points to lie to the left of the accelerated oscillator trajectory. They show where the extra terms in the two-point function are which were ignored in the RSG analysis. More important to answering the theme question, they show that at least in (1+1) dimension the stress-energy tensor vanishes everywhere except on the horizon. This means that there is no net flux of radiation emitted from the uniformly accelerated oscillator in steady state in (1+1)D case.
Most prior theoretical work on this topic was done in (1+1) dimensional spacetimes. However since most experimental proposals on the detection of Unruh effect are designed for the physical four dimensional spacetime, it is necessary to do a detailed analysis for (3+1) dimensions. Although tempting, one cannot assume that all (3+1) results are equivalent to those from (1+1) calculations. First, there are new divergences in the (3+1) case to deal with. Second, the structure of the retarded field in (3+1) dimensional spacetime is much richer: it consists of a bound field (generalized Coulomb field) and a radiation field with a variety of multipole structure, while the (1+1) case has only the radiation field in a different form. Third, an earlier work of one of us lin03b showed that there is some constant negative monopole radiation emitted from a detector initially in the ground state and uniformly accelerated in (3+1)D Minkowski space, and claimed that this signal could be an evidence of the Unruh effect. This contradicts the results reported by HR CapHR and others from the (1+1)D calculations. We need to clarify this discrepancy and determine the cause of it, by studying the complete process from transient to steady state. In particular, since radiation only exists under nonequilibrium conditions in the (1+1) case, it is crucial to understand the transient effects in the (3+1) case to gauge our expectation of what could be, against what would be, observed in laboratories.
In conceptual terms, one is tempted to invoke stationarity and thermality conditions for the description of an UAD. This is indeed a simple and powerful way to understand its physics if the detector undergoes uniform acceleration and interacts with the field all throughout (e.g., because of the stationarity of the problem in the Rindler proper time it is guaranteed that the total boost energy-operator is conserved). However, this argument is inapplicable for transient epochs during which the physics is quite different (see, e.g., the inertial to uniform acceleration motion treated in RHK ). Likewise, one can invoke the thermality condition (i.e., the thermal radiance experienced by a UAD is equivalent to that of an inertial detector in a thermal bath) to obtain results based on simple reasonings. But then we note that the thermality condition does not uniquely arise from uniform acceleration conditions. For example if the motion is rapidly altered HRthermal the radiation produced can be approximately thermal. This thermality in emitted radiation (e.g., from sudden injection of atoms into a cavity) is similar to those encountered in cosmological particle creation Parker76 , but has a different physical origin from Unruh effect which is similar to particle creation from black holes (Hawking effect) Haw75 (see, e.g., Scullyetal ; HuRouraPRL ).
In terms of methodology, instead of using the more sophisticated influence functional method as in the earlier series of papers on accelerated detectors RHA ; RHK and moving charges JH1 ; IARD ; GH1 , our work here follows more closely the work of HR who used the Heisenberg operator method to calculate the two-point function and the stress-energy tensor of a massless quantum scalar field. In our analysis based on the (3+1)D Unruh-DeWitt detector theory we found the full and exact dynamics of the detector and the field in terms of their Heisenberg operator evolution, thus making available the complete quantum and statistical information for this detector-field system, enabling us to address the interplay of thermal radiance in the detector, vacuum polarization cloud around the detector, quantum fluctuations and radiation, and emitted flux of classical radiation.
The paper is organized as follows. In Sec.II we introduce the Unruh-DeWitt detector theory. Then in Sec.III we describe the quantum dynamics of the detector-field system in the Heisenberg picture, yielding the expectation values of the detector two-point function with respect to the Minkowski vacuum and a detector coherent state in Sec.IV. With these results we derive the two-point function of the quantum field and describe what constitutes the “vacuum polarization” around the detector in Sec.V. Then in Sec.VI we calculate the quantum expectation values of the stress-energy tensor induced by the uniformly accelerated detector. This allows us to explore the conservation law and derive the quantum radiation formula. A comparison with the results in Ref.lin03b follows in Sec.VII. Finally, we summarize our findings in Sec.VIII.
## II Unruh-DeWitt Detector Theory
The total action of the detector-field system is given by
$$S=S_Q+S_\mathrm{\Phi }+S_I,$$
(1)
where $`Q`$ is the internal degree of freedom of the detector, assumed to be a harmonic oscillator with mass $`m_0`$ and a (bare) natural frequency $`\mathrm{\Omega }_0`$:
$$S_Q=𝑑\tau \frac{m_0}{2}\left[\left(_\tau Q\right)^2\mathrm{\Omega }_0^2Q^2\right].$$
(2)
Here $`\tau `$ is the detector’s proper time. Henceforth we will use an overdot on $`Q`$ to denote $`dQ(\tau )/d\tau `$. The scalar field $`\mathrm{\Phi }`$ is assumed to be massless,
$$S_\mathrm{\Phi }=d^4x\frac{1}{2}_\mu \mathrm{\Phi }^\mu \mathrm{\Phi }.$$
(3)
The interaction action $`S_I`$ for the Unruh-DeWitt (UD) detector theory has the form Unr76 ; DeW79 ,
$$S_I=\lambda _0𝑑\tau d^4xQ(\tau )\mathrm{\Phi }(x)\delta ^4\left(x^\mu z^\mu (\tau )\right).$$
(4)
where $`\lambda _0`$ is the coupling constant. This can be regarded as a simplified version of an atom.
Below we will consider the UD detector moving in a prescribed trajectory $`z^\mu (\tau )`$ in a four-dimensional Minkowski spacetime with metric $`\eta _{\mu \nu }=`$ diag$`(1,1,1,1)`$ and line element $`ds^2=\eta _{\mu \nu }dx^\mu dx^\nu `$. By “prescribed” we mean the trajectory of the detector is not considered as a dynamical variable, thus we ignore the backreaction effect of the field on the trajectory. (See Ref.JH1 for an example where the trajectory and the field are determined self-consistently by each other.) The detector is made (by the act of an external agent) to go along the worldline
$$z^\mu (\tau )=(a^1\mathrm{sinh}a\tau ,a^1\mathrm{cosh}a\tau ,0,0),$$
(5)
parametrized by its proper time $`\tau `$. This is the trajectory of a uniformly accelerated detector situated in Rindler wedge R (the portion $`tx^1<0`$ and $`t+x^1>0`$ of Minkowski space; see Chapter 4 of Ref.BD ).
## III Quantum Theory in Heisenberg Picture
The conjugate momenta ($`P(\tau ),\mathrm{\Pi }(x)`$) of dynamical variables ($`Q(\tau ),\mathrm{\Phi }(x)`$) are defined by
$`P(\tau )`$ $`=`$ $`{\displaystyle \frac{\delta S}{\delta \dot{Q}(\tau )}}=m_0\dot{Q}(\tau ),`$ (6)
$`\mathrm{\Pi }(x)`$ $`=`$ $`{\displaystyle \frac{\delta S}{\delta _t\mathrm{\Phi }(x)}}=_t\mathrm{\Phi }(x).`$ (7)
By treating the above dynamical variables as operators and introducing the equal time commutation relations,
$`[\widehat{Q}(\tau ),\widehat{P}(\tau )]`$ $`=`$ $`i\mathrm{},`$ (8)
$`[\widehat{\mathrm{\Phi }}(t,𝐱),\widehat{\mathrm{\Pi }}(t,𝐱^{})]`$ $`=`$ $`i\mathrm{}\delta ^3(𝐱𝐱^{}),`$ (9)
one can write down the Heisenberg equations of motion for the operators and obtains
$`_\tau ^2\widehat{Q}(\tau )+\mathrm{\Omega }_0^2\widehat{Q}(\tau )`$ $`=`$ $`{\displaystyle \frac{\lambda _0}{m_0}}\widehat{\mathrm{\Phi }}(\tau ,𝐳(\tau )),`$ (10)
$`\left(_t^2^2\right)\widehat{\mathrm{\Phi }}(x)`$ $`=`$ $`\lambda _0{\displaystyle _0^{\mathrm{}}}𝑑\tau \widehat{Q}(\tau )\delta ^4(xz(\tau )),`$ (11)
which have the same form as the classical Euler-Lagrange equations.
Suppose the system is prepared before $`\tau =0`$, and the coupling $`S_I`$ is turned on precisely at the moment $`\tau =0`$ when we allow all the dynamical variables to begin to interact and evolve under the influence of each other. (The consequences of this sudden switch-on and the assumption of a factorizable initial state for the combined system a quantum Brownian oscillator plus oscillator bath is described in some details in HPZ ). By virtue of the linear coupling $`(\text{4})`$, the time evolution of $`\widehat{\mathrm{\Phi }}(𝐱)`$ is simply a linear transformation in the phase space spanned by the orthonormal basis $`(\widehat{\mathrm{\Phi }}(𝐱),\widehat{\mathrm{\Pi }}(𝐱),\widehat{Q},\widehat{P})`$, that is, $`\widehat{\mathrm{\Phi }}(x)`$ can be expressed in the form
$$\widehat{\mathrm{\Phi }}(t,𝐱)=d^3x^{}\left[f^\mathrm{\Phi }(t,𝐱,𝐱^{})\widehat{\mathrm{\Phi }}(0,𝐱^{})+f^\mathrm{\Pi }(t,𝐱,𝐱^{})\widehat{\mathrm{\Pi }}(0,𝐱^{})\right]+f^Q(x)\widehat{Q}(0)+f^P(x)\widehat{P}(0).$$
(12)
Here $`f^\mathrm{\Phi }(x,𝐱^{}),f^\mathrm{\Pi }(x,𝐱^{}),f^Q(x)`$ and $`f^P(x)`$ are c-number functions of spacetime. Similarly, the operator $`\widehat{Q}(\tau )`$ can be written as
$$\widehat{Q}(\tau )=d^3x^{}\left[q^\mathrm{\Phi }(\tau ,𝐱^{})\widehat{\mathrm{\Phi }}(0,𝐱^{})+q^\mathrm{\Pi }(\tau ,𝐱^{})\widehat{\mathrm{\Pi }}(0,𝐱^{})\right]+q^Q(\tau )\widehat{Q}(0)+q^P(\tau )\widehat{P}(0),$$
(13)
with c-number functions $`q^Q(\tau ),q^P(\tau ),q^\mathrm{\Phi }(\tau ,𝐱^{})`$ and $`q^\mathrm{\Pi }(\tau ,𝐱^{})`$.
For the case with initial operaters being the free field operators, namely, $`\widehat{\mathrm{\Phi }}(0,𝐱)=\widehat{\mathrm{\Phi }}_0(𝐱)`$, $`\widehat{\mathrm{\Phi }}(0,𝐱)=\widehat{\mathrm{\Pi }}_0(𝐱)`$, $`\widehat{Q}(0)=\widehat{Q}_0`$ and $`\widehat{P}(0)=\widehat{P}_0`$, one can go further by introducing the complex operators $`\widehat{b}_𝐤`$ and $`\widehat{a}`$:
$`\widehat{\mathrm{\Phi }}_0(𝐱)`$ $`=`$ $`{\displaystyle \frac{d^3k}{(2\pi )^3}\sqrt{\frac{\mathrm{}}{2\omega }}\left[e^{i𝐤𝐱}\widehat{b}_𝐤+e^{i𝐤𝐱}\widehat{b}_\mathrm{k}^{}\right]},`$ (14)
$`\widehat{\mathrm{\Pi }}_0(𝐱)`$ $`=`$ $`{\displaystyle \frac{d^3k}{(2\pi )^3}\sqrt{\frac{\mathrm{}}{2\omega }}(i\omega )\left[e^{i𝐤𝐱}\widehat{b}_𝐤e^{i𝐤𝐱}\widehat{b}_𝐤^{}\right]}`$ (15)
with $`\omega |𝐤|`$, and
$$\widehat{Q}_0=\sqrt{\frac{\mathrm{}}{2\mathrm{\Omega }_rm_0}}(\widehat{a}+\widehat{a}^{}),\widehat{P}_0=i\sqrt{\frac{\mathrm{}\mathrm{\Omega }_rm_0}{2}}(\widehat{a}\widehat{a}^{}).$$
(16)
Note that, instead of $`\mathrm{\Omega }_0`$, we use the renormalizd natural frequency $`\mathrm{\Omega }_r`$ (to be defined in $`(\text{40})`$) in the definition of $`\widehat{a}`$. Then the commutation relations $`(\text{8})`$ and $`(\text{9})`$ give
$$[\widehat{a},\widehat{a}^{}]=1,[\widehat{b}_𝐤,\widehat{b}_𝐤^{}^{}]=(2\pi )^3\delta ^3(𝐤𝐤^{}),$$
(17)
and the expressions $`(\text{12})`$ and $`(\text{13})`$ can be re-written as
$`\widehat{\mathrm{\Phi }}(t,𝐱)`$ $`=`$ $`\widehat{\mathrm{\Phi }}_b(x)+\widehat{\mathrm{\Phi }}_a(x),`$ (18)
$`\widehat{Q}(\tau )`$ $`=`$ $`\widehat{Q}_b(\tau )+\widehat{Q}_a(\tau )`$ (19)
where
$`\widehat{\mathrm{\Phi }}_b(x)`$ $`=`$ $`{\displaystyle \frac{d^3k}{(2\pi )^3}\sqrt{\frac{\mathrm{}}{2\omega }}\left[f^{(+)}(t,𝐱;𝐤)\widehat{b}_𝐤+f^{()}(t,𝐱;𝐤)\widehat{b}_𝐤^{}\right]},`$ (20)
$`\widehat{\mathrm{\Phi }}_a(x)`$ $`=`$ $`\sqrt{{\displaystyle \frac{\mathrm{}}{2\mathrm{\Omega }_rm_0}}}\left[f^a(t,𝐱)\widehat{a}+f^a(t,𝐱)\widehat{a}^{}\right],`$ (21)
$`\widehat{Q}_b(\tau )`$ $`=`$ $`{\displaystyle \frac{d^3k}{(2\pi )^3}\sqrt{\frac{\mathrm{}}{2\omega }}\left[q^{(+)}(\tau ,𝐤)\widehat{b}_𝐤+q^{()}(\tau ,𝐤)\widehat{b}_𝐤^{}\right]},`$ (22)
$`\widehat{Q}_a(\tau )`$ $`=`$ $`\sqrt{{\displaystyle \frac{\mathrm{}}{2\mathrm{\Omega }_rm_0}}}\left[q^a(\tau )\widehat{a}+q^a(\tau )\widehat{a}^{}\right].`$ (23)
The whole problem therefore can be transformed to solving c-number functions $`f^s(x)`$ and $`q^s(\tau )`$ from $`(\text{10})`$ and $`(\text{11})`$ with suitable initial conditions. Since $`\widehat{Q}`$ and $`\widehat{\mathrm{\Phi }}`$ are hermitian, one has $`f^{()}=(f^{(+)})^{}`$ and $`q^{()}=(q^{(+)})^{}`$. Hence it is sufficient to solve the c-number functions $`f^{(+)}(t,𝐱;𝐤)`$, $`q^{(+)}(\tau ,𝐤)`$, $`f^a(t,𝐱)`$ and $`q^a(\tau )`$. To place this in a more general setting, let us perform a Lorentz transformation shifting $`\tau =0`$ to $`\tau =\tau _0`$, and define
$$\eta \tau \tau _0.$$
(24)
This does not add any complication to our calculation. Now the coupling between the detector and the field would be turned on at $`\tau =\tau _0`$. We are looking for solutions with the initial conditions
$`f^{(+)}(t(\tau _0),𝐱;𝐤)=e^{i𝐤𝐱},_tf^{(+)}(t(\tau _0),𝐱;𝐤)=i\omega e^{i𝐤𝐱},q^{(+)}(\tau _0;𝐤)=\dot{q}^{(+)}(\tau _0;𝐤)=0,`$ (25)
$`f^a(t(\tau _0),𝐱)=_tf^a(t(\tau _0),𝐱)=0,q^a(\tau _0)=1,\dot{q}^a(\tau _0)=i\mathrm{\Omega }_r.`$ (26)
### III.1 Solving for $`f^{(+)}`$ and $`q^{(+)}`$
The method to solve for $`f`$ and $`q`$ are analogous to what we did in classical field theory. First, we find an expression relating the harmonic oscillator and the field amplitude right at the detector. Then substituting this relation into the equation of motion for the oscillator, we obtain a complete equation of motion for $`q`$ with full information of the field. Last, we solve this complete equation of motion for $`q`$, and from its solution determine the field $`f`$ consistently.
Eq.$`(\text{11})`$ implies that
$$(_t^2^2)f^{(+)}(x;𝐤)=\lambda _0_{\tau _0}^{\mathrm{}}𝑑\tau \delta ^4(xz(\tau ))q^{(+)}(\tau ;𝐤).$$
(27)
The general solution for $`f^{(+)}`$ reads
$$f^{(+)}(x;𝐤)=f_0^{(+)}(x;𝐤)+f_1^{(+)}(x;𝐤),$$
(28)
where
$$f_0^{(+)}(x;𝐤)e^{i\omega t+i𝐤𝐱}$$
(29)
is the free field solution, and
$$f_1^{(+)}(x;𝐤)\lambda _0_{\tau _0}^{\mathrm{}}𝑑\tau G_{\mathrm{ret}}(x;z(\tau ))q^{(+)}(\tau ;𝐤)$$
(30)
is the retarded solution, which looks like the retarded field in classical field theory. Here $`\omega =|𝐤|`$ and the retarded Green’s function $`G_{\mathrm{ret}}`$ in Minkowski space is given by
$$G_{\mathrm{ret}}(x,x^{})=\frac{1}{4\pi }\delta (\sigma )\theta (tt^{})$$
(31)
with $`\sigma (x_\mu x_\mu ^{})(x^\mu x^\mu )/2`$. Applying the explicit form of the retarded Green’s function, one can go further to write
$$f_1^{(+)}(x;𝐤)=\frac{\lambda _0\theta (\eta _{})}{2\pi aX}q^{(+)}(\tau _{};𝐤),$$
(32)
where
$`X`$ $``$ $`\sqrt{(UV+\rho ^2+a^2)^2+4a^2UV},`$ (33)
$`\tau _{}`$ $``$ $`{\displaystyle \frac{1}{a}}\mathrm{ln}{\displaystyle \frac{a}{2|V|}}\left(XUV+\rho ^2+a^2\right),`$ (34)
$`\eta _{}`$ $``$ $`\tau _{}\tau _0,`$ (35)
with $`\rho \sqrt{x_2{}_{}{}^{2}+x_3^2}`$, $`Utx^1`$ and $`Vt+x^1`$.
The formal retarded solution $`(\text{32})`$ is singular on the trajectory of the detector. To deal with the singularity, note that the UD detector here is a quantum mechanical object, and the detector number would always be one. This means that at the energy threshold of detector creations, there is a natural cutoff on frequency, which sets an upper bound on the resolution to be explored in our theory. Thus it is justified to assume here that the detector has a finite extent $`O(\mathrm{\Lambda }^1)`$, which will introduce the back reaction on the detector.
Let us regularize the retarded Green’s function by invoking the essence of effective field theory:
$$G_{\mathrm{ret}}^\mathrm{\Lambda }(x,x^{})=\frac{1}{4\pi }\sqrt{\frac{8}{\pi }}\mathrm{\Lambda }^2e^{2\mathrm{\Lambda }^4\sigma ^2}\theta (tt^{}).$$
(36)
(For more details on this regularization scheme, see Refs.JH1 ; GH1 .) With this, right on the trajectory, the retarded solution for large $`\mathrm{\Lambda }`$ is
$$f_1^{(+)}(z(\tau );𝐤)=\frac{\lambda _0}{4\pi }\left[\mathrm{\Lambda }\zeta q^{(+)}(\tau ;𝐤)_\tau q^{(+)}(\tau ;𝐤)+O(\mathrm{\Lambda }^1)\right],$$
(37)
where $`\zeta =2^{7/4}\mathrm{\Gamma }(5/4)/\sqrt{\pi }`$. Substituting the above expansion into $`(\text{10})`$ and neglecting $`O(\mathrm{\Lambda }^1)`$ terms, one obtains the equation of motion for $`q^{(+)}`$ with back reaction,
$$(_\tau ^2+2\gamma _\tau +\mathrm{\Omega }_r^2)q^{(+)}(\tau ;𝐤)=\frac{\lambda _0}{m_0}f_0^{(+)}(z(\tau );𝐤).$$
(38)
Fortunately, there is no higher derivatives of $`q`$ present in the above equation of motion. Now $`q^{(+)}`$ behaves like a damped harmonic oscillator driven by the vacuum fluctuations of the scalar field, with the damping constant
$$\gamma \frac{\lambda _0^2}{8\pi m_0},$$
(39)
and the renormalized natural frequency
$$\mathrm{\Omega }_r^2\mathrm{\Omega }_0^2\frac{\lambda _0^2\mathrm{\Lambda }\zeta }{4\pi m_0}.$$
(40)
In $`(\text{38})`$, the solution for $`q^{(+)}`$ compatible with the initial conditions $`q^{(+)}(\tau _0;𝐤)=\dot{q}^{(+)}(\tau _0;𝐤)=0`$ is
$$q^{(+)}(\tau ;𝐤)=\frac{\lambda _0}{m_0}\underset{j=+,}{}_{\tau _0}^\tau 𝑑\tau ^{}c_je^{w_j(\tau \tau ^{})}f_0^{(+)}(z(\tau ^{});𝐤),$$
(41)
where $`f_0^{(+)}`$ has been given in $`(\text{29})`$, $`c_\pm `$ and $`w_\pm `$ are defined as
$$c_\pm =\pm \frac{1}{2i\mathrm{\Omega }},w_\pm =\gamma \pm i\mathrm{\Omega },$$
(42)
with
$$\mathrm{\Omega }\sqrt{\mathrm{\Omega }_r^2\gamma ^2}.$$
(43)
Throughout this paper we consider only the under-damped case with $`\gamma ^2<\mathrm{\Omega }_r^2`$, so $`\mathrm{\Omega }`$ is always real.
### III.2 Solving for $`f^a`$ and $`q^a`$
Similarly, from $`(\text{10})`$, $`(\text{11})`$, $`(\text{18})`$ and $`(\text{19})`$, the equations of motion for $`f^a`$ and $`q^a`$ read
$`\left(_t^2^2\right)f^a(x)=\lambda _0{\displaystyle 𝑑\tau \delta ^4(xz(\tau ))q^a(\tau )},`$ (44)
$`(_\tau ^2+\mathrm{\Omega }_0^2)q^a(\tau )={\displaystyle \frac{\lambda _0}{m_0}}f^a(z(\tau )).`$ (45)
The general solutions for $`f^a`$, similar to $`(\text{28})`$, is
$$f^a(x)=f_0^a(x)+\lambda _0_{\tau _0}^{\mathrm{}}𝑑\tau G_{\mathrm{ret}}(x;z(\tau ))q^a(\tau _{})$$
(46)
However, according to the initial condition $`(\text{26})`$, one has $`f_0^a=0`$, hence
$$f^a(x)=\frac{\lambda _0\theta (\eta _{})}{2\pi aX}q^a(\tau _{}).$$
(47)
Again, the value of $`f^a`$ is singular right at the position of the detector. Performing the same regularization as those for $`q^{(+)}`$, $`(\text{45})`$ becomes (cf. $`(\text{38})`$)
$$\left(_\tau ^2+2\gamma _\tau +\mathrm{\Omega }_r^2\right)q^a(\tau )=0,$$
(48)
which describes a damped harmonic oscillator free of driving force. The solution consistent with the initial condition $`q^a(\tau _0)=1`$ and $`\dot{q}^b(\tau _0)=i\mathrm{\Omega }_r`$ reads
$$q^a(\tau )=\frac{1}{2}\theta (\eta )e^{\gamma \eta }\left[\left(1\frac{\mathrm{\Omega }_r+i\gamma }{\mathrm{\Omega }}\right)e^{i\mathrm{\Omega }\eta }+\left(1+\frac{\mathrm{\Omega }_r+i\gamma }{\mathrm{\Omega }}\right)e^{i\mathrm{\Omega }\eta }\right].$$
(49)
## IV Two-Point Functions of the Detector
As shown in the previous section, as $`\widehat{Q}`$ evolves, some non-zero terms proportional to $`\widehat{\mathrm{\Phi }}`$ and $`\widehat{\mathrm{\Pi }}`$ will be generated. Suppose the detector is initially prepared in a state that can be factorized into the quantum state $`|q`$ for $`Q`$ and the Minkowski vacuum $`|0_M`$ for the scalar field $`\mathrm{\Phi }`$, that is,
$$|\tau _0=|q|0_M$$
(50)
then the two-point function of $`Q`$ will split into two parts,
$`Q(\tau )Q(\tau ^{})`$ $`=`$ $`0_M|q\left|\left[\widehat{Q}_b(\tau )+\widehat{Q}_a(\tau )\right]\left[\widehat{Q}_b(\tau )+\widehat{Q}_a(\tau )\right]\right|q|0_M`$ (51)
$`=`$ $`q|qQ(\tau )Q(\tau ^{})_\mathrm{v}+Q(\tau )Q(\tau ^{})_\mathrm{a}0_M|0_M.`$
where, from $`(\text{19})`$,
$`Q(\tau )Q(\tau ^{})_\mathrm{v}`$ $`=`$ $`0_M|\widehat{Q}_b(\tau )\widehat{Q}_b(\tau ^{})|0_M,`$ (52)
$`Q(\tau )Q(\tau ^{})_\mathrm{a}`$ $`=`$ $`q\left|\widehat{Q}_a(\tau )\widehat{Q}_a(\tau )\right|q.`$ (53)
Similar splitting happens for every two-point function of $`\widehat{\mathrm{\Phi }}(x)`$ as well as for the stress-energy tensor.
Observe that $`Q(\tau )Q(\tau ^{})_\mathrm{v}`$ depends on the initial state of the field, or the Minkowski vacuum, while $`Q(\tau )Q(\tau ^{})_\mathrm{a}`$ depends on the initial state of the detector only. One can thus interpret $`Q(\tau )Q(\tau ^{})_\mathrm{v}`$ as accounting for the response to the vacuum fluctuations, while $`Q(\tau )Q(\tau ^{})_\mathrm{a}`$ corresponds to the intrinsic quantum fluctuations in the detector.
In the following, we will demonstrate the explicit forms of some two-point functions we have obtained and analyze their behavior. To distinguish the quantum or classical natures of these quantities, the initial quantum state $`|q`$ will be taken to be the coherent state scully ,
$$|q=e^{\alpha ^2/2}\underset{n=0}{\overset{\mathrm{}}{}}\frac{\alpha ^n}{\sqrt{n!}}|n,$$
(54)
where $`|n`$ is the $`n`$-th excited state for the free detector, and $`\alpha =q_0\sqrt{\mathrm{\Omega }_r/2\mathrm{}}`$ with a constant $`q_0`$. The representation of $`|q`$ in $`Q`$-space reads
$$\psi (Q,\tau _0)=\left(\frac{\mathrm{\Omega }_r}{\pi \mathrm{}}\right)^{1/4}e^{\mathrm{\Omega }_r(Qq_0)^2/2\mathrm{}},$$
(55)
which is a wave-packet centered at $`q_0`$ with the spread identical to the one for the ground state.
### IV.1 Expectation value of the detector two-point function with respect to the Minkowski vacuum
Along the trajectory $`z^\mu (\tau )`$ in $`(\text{5})`$, performing a Fourier transformation with respect to $`\tau `$ on $`(\text{29})`$, one has
$$f_0^{(+)}(z(\tau );𝐤)𝑑\kappa e^{i\kappa \tau }\phi (\kappa ,𝐤),$$
(56)
where the frequency spectrum of the Minkowski mode from the viewpoint of the UAD,
$$\phi (\kappa ,𝐤)=\frac{e^{\pi \kappa /2a}}{\pi a}\left(\frac{\omega k_1}{\omega +k_1}\right)^{\frac{i\kappa }{2a}}K_{\frac{i\kappa }{a}}\left(\sqrt{k_2{}_{}{}^{2}+k_3^2}/a\right),$$
(57)
is not trivial any more. Given the result of the integration,
$`{\displaystyle \frac{\mathrm{}d^3k}{(2\pi )^32\omega }\phi (\kappa ,𝐤)\phi ^{}(\kappa ^{},𝐤)}`$ $`=`$ $`{\displaystyle \frac{\mathrm{}}{(2\pi )^2}}{\displaystyle \frac{\kappa }{1e^{2\pi \kappa /a}}}\delta (\kappa \kappa ^{}),`$ (58)
a Planck factor with the Unruh temperature $`a/2\pi `$ emerges. Then from $`(\text{52})`$, $`(\text{41})`$ and $`(\text{56})`$, one has
$`Q(\eta )Q(\eta ^{})_\mathrm{v}`$ $`=`$ $`\mathrm{}{\displaystyle \frac{d^3k}{(2\pi )^32\omega }q^{(+)}(\tau ;𝐤)q^{()}(\tau ;𝐤)}`$ (60)
$`=`$ $`{\displaystyle \frac{\lambda _0^2\mathrm{}}{(2\pi )^2m_0^2}}{\displaystyle \underset{j,j^{}}{}}{\displaystyle }{\displaystyle \frac{\kappa d\kappa }{1e^{2\pi \kappa /a}}}{\displaystyle \frac{c_jc_j^{}^{}e^{i\kappa (\tau _0\tau _0^{})}}{(w_j+i\kappa )(w_j^{}^{}i\kappa )}}\times `$
$`\left[e^{w_j(\tau \tau _0)}e^{i\kappa (\tau \tau _0)}\right]\left[e^{w_j^{}^{}(\tau ^{}\tau _0^{})}e^{i\kappa (\tau ^{}\tau _0^{})}\right],`$
where the integrand has poles at $`\kappa =\pm \mathrm{\Omega }i\gamma `$ and $`\kappa =\pm ina`$, $`nN`$. Let $`\tau _0^{}<\tau _0<\tau <\tau ^{}`$ and taking the coincidence limit, one obtains
$`Q(\eta )^2_\mathrm{v}`$ $`=`$ $`\underset{\eta ^{}\eta }{lim}{\displaystyle \frac{1}{2}}\{Q(\eta ),Q(\eta ^{})\}_\mathrm{v}`$ (61)
$`=`$ $`{\displaystyle \frac{\mathrm{}\lambda _0^2}{(2\pi m_0\mathrm{\Omega })^2}}\theta (\eta )\mathrm{Re}\{(\mathrm{\Lambda }_0\mathrm{ln}a)e^{2\gamma \eta }\mathrm{sin}^2\mathrm{\Omega }\eta `$
$`+`$ $`{\displaystyle \frac{a}{2}}e^{(\gamma +a)\eta }\left[{\displaystyle \frac{F_{\gamma +i\mathrm{\Omega }}(e^{a\eta })}{\gamma +i\mathrm{\Omega }+a}}\left({\displaystyle \frac{i\mathrm{\Omega }}{\gamma }}\right)e^{i\mathrm{\Omega }\eta }+{\displaystyle \frac{F_{\gamma i\mathrm{\Omega }}(e^{a\eta })}{\gamma +i\mathrm{\Omega }a}}\left(\left(1+{\displaystyle \frac{i\mathrm{\Omega }}{\gamma }}\right)e^{i\mathrm{\Omega }\eta }e^{i\mathrm{\Omega }\eta }\right)\right]`$
$``$ $`{\displaystyle \frac{1}{4}}[({\displaystyle \frac{i\mathrm{\Omega }}{\gamma }}+e^{2\gamma \eta }({\displaystyle \frac{i\mathrm{\Omega }}{\gamma }}+1e^{2i\mathrm{\Omega }\eta }))(\psi _{\gamma +i\mathrm{\Omega }}+\psi _{\gamma i\mathrm{\Omega }})`$
$`({\displaystyle \frac{i\mathrm{\Omega }}{\gamma }}+e^{2\gamma \eta }({\displaystyle \frac{i\mathrm{\Omega }}{\gamma }}+1e^{2i\mathrm{\Omega }\eta }))i\pi \mathrm{coth}{\displaystyle \frac{\pi }{a}}(\mathrm{\Omega }i\gamma )]\}.`$
Here $`F_s(t)`$ is defined by the hyper-geometric function as
$$F_s(t){}_{2}{}^{}F_{1}^{}(1+\frac{s}{a},1,2+\frac{s}{a};t),$$
(62)
and
$$\psi _s\psi \left(1+\frac{s}{a}\right)$$
(63)
is the poly-gamma function. The divergent $`\mathrm{\Lambda }_0`$-term is produced by the coincidence limit: as $`\eta ^{}\eta `$, $`\mathrm{\Lambda }_0\gamma _e\mathrm{ln}|\tau _0^{}\tau _0|`$ with the Euler’s constant $`\gamma _e`$. Since $`|\tau _0^{}\tau _0|`$ characterizes the time scale that the interaction is turned on, $`\mathrm{\Lambda }_0`$ could be finite in real processes. In any case, for every finite value of $`\mathrm{\Lambda }_0`$, the first line of the result in $`(\text{61})`$ vanishes as $`\eta \mathrm{}`$.
In FIG. 1, we show the $`Q(\eta )^2_\mathrm{v}`$ without $`\mathrm{\Lambda }_0`$-term in dotted line. Roughly speaking the curve saturates exponentially in the detector’s proper time. As $`\eta \mathrm{}`$, $`Q(\eta )^2_\mathrm{v}`$ saturates to the value
$$\underset{\eta \mathrm{}}{lim}Q(\eta )^2_\mathrm{v}=\frac{\mathrm{}}{2\pi m_0\mathrm{\Omega }}\mathrm{Re}\left[\frac{ia}{\gamma +i\mathrm{\Omega }}2i\psi _{\gamma +i\mathrm{\Omega }}\right].$$
(64)
For $`\gamma <a`$, the time scale of the rise is about $`1/2\gamma `$, which can be read off from the $`e^{2\gamma \eta }`$ in $`(\text{61})`$. From there one can also see that the small oscillation around the rising curve has a frequency of $`O(\mathrm{\Omega })`$.
For $`Q(\eta )\dot{Q}(\eta )_\mathrm{v}`$, it will be clear that what is interesting for the calculation of the flux is the combined quantities like $`Q(\eta )\dot{Q}(\eta )_\mathrm{v}+\dot{Q}(\eta )Q(\eta )_\mathrm{v}`$. Notice that
$$Q(\eta )\dot{Q}(\eta )_\mathrm{v}+\dot{Q}(\eta )Q(\eta )_\mathrm{v}=_\tau Q(\eta )^2_\mathrm{v}.$$
(65)
With the result of $`Q(\eta )^2_\mathrm{v}`$, this calculation is straightforward. Let us turn to the two-point functions of $`\dot{Q}`$. Similar to $`Q(\eta )^2_\mathrm{v}`$, one has
$`\dot{Q}(\eta )\dot{Q}(\eta ^{})_\mathrm{v}`$ $`=`$ $`{\displaystyle \frac{\mathrm{}d^3k}{(2\pi )^32\omega }\dot{q}^{(+)}(\tau ;𝐤)\dot{q}^{()}(\tau ;𝐤)}`$ (67)
$`=`$ $`{\displaystyle \frac{\lambda _0^2\mathrm{}}{(2\pi )^2m_0^2}}{\displaystyle \underset{j,j^{}}{}}{\displaystyle }{\displaystyle \frac{\kappa d\kappa }{1e^{2\pi \kappa /a}}}{\displaystyle \frac{c_jc_j^{}^{}e^{i\kappa (\tau _0\tau _0^{})}}{(w_j+i\kappa )(w_j^{}^{}i\kappa )}}\times `$
$`\left[w_je^{w_j(\tau \tau _0)}+i\kappa e^{i\kappa (\tau \tau _0)}\right]\left[w_j^{}^{}e^{w_j^{}^{}(\tau ^{}\tau _0^{})}i\kappa e^{i\kappa (\tau ^{}\tau _0^{})}\right]`$
from $`(\text{41})`$, $`(\text{56})`$ and $`(\text{58})`$. The coincidence limit of the above two-point function reads
$`\dot{Q}(\eta )^2_\mathrm{v}`$ $`=`$ $`{\displaystyle \frac{\mathrm{}\lambda _0^2}{(2\pi m_0\mathrm{\Omega })^2}}\theta (\eta )\mathrm{Re}\{(\mathrm{\Lambda }_1\mathrm{ln}a)\mathrm{\Omega }^2+(\mathrm{\Lambda }_0\mathrm{ln}a)e^{2\gamma \eta }(\mathrm{\Omega }\mathrm{cos}\mathrm{\Omega }\eta \gamma \mathrm{sin}\mathrm{\Omega }\eta )^2`$ (68)
$`+`$ $`{\displaystyle \frac{a}{2}}(\gamma +i\mathrm{\Omega })^2e^{(\gamma +a)\eta }\left[{\displaystyle \frac{F_{\gamma +i\mathrm{\Omega }}(e^{a\eta })}{\gamma +i\mathrm{\Omega }+a}}\left({\displaystyle \frac{i\mathrm{\Omega }}{\gamma }}\right)e^{i\mathrm{\Omega }\eta }+{\displaystyle \frac{F_{\gamma i\mathrm{\Omega }}(e^{a\eta })}{\gamma +i\mathrm{\Omega }a}}\left(\left(1{\displaystyle \frac{i\mathrm{\Omega }}{\gamma }}\right)e^{i\mathrm{\Omega }\eta }e^{i\mathrm{\Omega }\eta }\right)\right]`$
$`+`$ $`{\displaystyle \frac{1}{4}}(\gamma +i\mathrm{\Omega })^2[({\displaystyle \frac{i\mathrm{\Omega }}{\gamma }}+e^{2\gamma \eta }({\displaystyle \frac{i\mathrm{\Omega }}{\gamma }}1+e^{2i\mathrm{\Omega }\eta }))(\psi _{\gamma +i\mathrm{\Omega }}+\psi _{\gamma i\mathrm{\Omega }})`$
$`({\displaystyle \frac{i\mathrm{\Omega }}{\gamma }}+e^{2\gamma \eta }({\displaystyle \frac{i\mathrm{\Omega }}{\gamma }}1+e^{2i\mathrm{\Omega }\eta }))i\pi \mathrm{coth}{\displaystyle \frac{\pi }{a}}(\mathrm{\Omega }i\gamma )]\},`$
where $`\mathrm{\Lambda }_1=\gamma _elim_{\tau ^{}\tau }\mathrm{ln}|\tau \tau ^{}|`$ can be subtracted safely. This will be justified later.
The subtracted $`\dot{Q}(\eta )^2_\mathrm{v}`$ is illustrated in FIG. 2 (in which $`\mathrm{\Lambda }_0`$-term has also been excluded). One can immediately recognize that $`\dot{Q}(\eta )^2_\mathrm{v}\mathrm{ln}\eta `$ when $`\eta `$ approaches zero; a new divergence occurs at $`\eta =0`$. Mathematically, this logarithmic divergence comes about because the divergences in the hyper-geometric functions in $`(\text{68})`$ do not cancel each other, unlike in $`Q(\eta )^2_\mathrm{v}`$. Physically, this divergence at the initial time $`\tau _0`$ could be another consequence of the sudden switch-on at $`\tau =\tau _0`$ or $`\eta =0`$. We expect that these ill-behaviors at the start could be tamed if we turn on the coupling adiabatically. (See HPZ for a discussion on this issue.)
For large $`\eta `$, the behavior of the dotted curve in FIG. 2 is quite similar to the one in FIG. 1 for $`Q(\tau )^2_\mathrm{v}`$. It saturates to
$$\underset{\eta \mathrm{}}{lim}\dot{Q}(\eta )^2_\mathrm{v}=\frac{\mathrm{}}{2\pi m_0\mathrm{\Omega }}\mathrm{Re}\left\{(\mathrm{\Omega }i\gamma )^2\left[\frac{ia}{\gamma +i\mathrm{\Omega }}2i\psi _{\gamma +i\mathrm{\Omega }}\right]4\gamma \mathrm{\Omega }\mathrm{ln}a\right\}.$$
(69)
Comparing $`(\text{68})`$ and $`(\text{61})`$, their time scales of saturation ($`1/2\gamma `$ for $`\gamma <a`$) and the frequency of the small ripples on the rising curve ($`O(\mathrm{\Omega })`$) are also the same. Note that when $`\gamma 1`$ and $`a`$ is finite, $`(\text{69})`$ and $`(\text{64})`$ implies
$$\dot{Q}(\mathrm{})^2_\mathrm{v}\mathrm{\Omega }^2Q(\mathrm{})^2_\mathrm{v},$$
(70)
which justifies the subtraction of $`\mathrm{\Lambda }_1`$-term in $`(\text{68})`$.
### IV.2 Expectation values of the detector two-point functions with respect to a coherent state
We now derive the expectation values of the detector two-point functions with respect to the coherent state $`(\text{54})`$. Subsituting $`(\text{23})`$ into $`(\text{53})`$ and using $`(\text{49})`$ and $`(\text{54})`$, one finds that
$$Q(\tau )Q(\tau ^{})_\mathrm{a}=Q(\tau )Q(\tau ^{})_\mathrm{a}^{\mathrm{qm}}+Q(\tau )Q(\tau ^{})_\mathrm{a}^{\mathrm{cl}},$$
(71)
where
$`Q(\tau )Q(\tau ^{})_\mathrm{a}^{\mathrm{qm}}`$ $``$ $`{\displaystyle \frac{\mathrm{}}{2\mathrm{\Omega }_rm_0}}q^a(\tau )q^a(\tau ^{}),`$ (72)
$`Q(\tau )Q(\tau ^{})_\mathrm{a}^{\mathrm{cl}}`$ $``$ $`{\displaystyle \frac{q_0^2}{m_0}}\mathrm{Re}\left[q^a(\tau )\right]\mathrm{Re}\left[q^a(\tau ^{})\right]=\overline{Q}(\tau )\overline{Q}(\tau ^{}),`$ (73)
with the mean value
$$\overline{Q}(\tau )Q(\tau )=\frac{q_0}{\sqrt{m_0}}\theta (\eta )e^{\gamma \eta }\left(\mathrm{cos}\mathrm{\Omega }\eta +\frac{\gamma }{\mathrm{\Omega }}\mathrm{sin}\mathrm{\Omega }\eta \right).$$
(74)
While the “qm” term is of purely quantum nature, the “cl” term is of classical nature: $`\overline{Q}`$ is real and $`Q(\tau )Q(\tau ^{})_\mathrm{a}^{\mathrm{cl}}`$ does not involve $`\mathrm{}`$. Thus the “cl” term is identified as the semiclassical part of the two-point functions. The coincidence limits of the above two-point functions are
$$Q(\eta )^2_\mathrm{a}^{\mathrm{qm}}=\frac{\mathrm{}\theta (\eta )}{2\mathrm{\Omega }^2\mathrm{\Omega }_rm_0}e^{2\gamma \eta }\left[\mathrm{\Omega }_r^2\gamma ^2\mathrm{cos}2\mathrm{\Omega }\eta +\gamma \mathrm{\Omega }\mathrm{sin}2\mathrm{\Omega }\eta \right],$$
(75)
and $`Q(\eta )^2_\mathrm{a}^{\mathrm{cl}}=\overline{Q}(\eta )^2`$.
Similarly, it is easy to find $`\dot{Q}(\tau )\dot{Q}(\tau ^{})_\mathrm{a}=\dot{Q}(\tau )\dot{Q}(\tau ^{})_\mathrm{a}^{\mathrm{qm}}+\dot{Q}(\tau )\dot{Q}(\tau ^{})_\mathrm{a}^{\mathrm{cl}}`$:
$`\dot{Q}(\tau )\dot{Q}(\tau ^{})_\mathrm{a}^{\mathrm{qm}}`$ $`=`$ $`{\displaystyle \frac{\mathrm{}}{2\mathrm{\Omega }_rm_0}}\dot{q}^b(\tau )\dot{q}^b(\tau ^{}),`$ (76)
$`\dot{Q}(\tau )\dot{Q}(\tau ^{})_\mathrm{a}^{\mathrm{cl}}`$ $`=`$ $`\dot{\overline{Q}}(\tau )\dot{\overline{Q}}(\tau ^{}),`$ (77)
and their coincidence limit,
$$\dot{Q}(\eta )^2_\mathrm{a}^{\mathrm{qm}}=\frac{\mathrm{}\mathrm{\Omega }_r}{2\mathrm{\Omega }^2m_0}\theta (\eta )e^{2\gamma \eta }\left[\mathrm{\Omega }_r^2\gamma ^2\mathrm{cos}2\mathrm{\Omega }\eta \gamma \mathrm{\Omega }\mathrm{sin}2\mathrm{\Omega }\eta \right],$$
(78)
and $`\dot{Q}(\eta )^2_\mathrm{a}^{\mathrm{cl}}=\dot{\overline{Q}}(\eta )^2`$. Also one has $`Q(\tau )\dot{Q}(\tau )_\mathrm{a}+\dot{Q}(\tau )Q(\tau )_\mathrm{a}=_\tau Q(\tau )^2_\mathrm{a}`$.
Note that the above two-point functions with respect to the coherent state are independent of the proper acceleration $`a`$. $`Q(\eta )^2_\mathrm{a}^{\mathrm{qm}}`$ and the variance (squared uncertainty) of $`Q`$,
$$\mathrm{\Delta }Q(\eta )^2[Q(\eta )\overline{Q}(\eta )]^2=Q(\eta )^2_\mathrm{v}+Q(\eta )^2_\mathrm{a}^{\mathrm{qm}}$$
(79)
have been shown in FIG. 1. $`Q(\eta )^2_\mathrm{a}^{\mathrm{qm}}`$ decays exponentially due to the dissipation of the zero-point energy to the field. As $`Q(\eta )^2_\mathrm{a}^{\mathrm{qm}}`$ decays, $`Q(\eta )^2_\mathrm{v}`$ grows and compensates the decrease, then saturates asymptotically. Similar behavior can be found in FIG. 2, in which
$$\mathrm{\Delta }\dot{Q}(\eta )^2[\dot{Q}(\eta )\overline{\dot{Q}}(\eta )]^2=\dot{Q}(\eta )^2_\mathrm{v}+\dot{Q}(\eta )^2_\mathrm{a}^{\mathrm{qm}}$$
(80)
is illustrated. In FIG. 3 we show the semiclassical two-point funciton $`Q^2_\mathrm{a}^{\mathrm{cl}}`$. Its behavior is quite different from the quantum part shown in the previous figures.
### IV.3 Late-time variances and the proper acceleration
The saturated value $`Q(\mathrm{})^2_\mathrm{v}`$ in Eq.$`(\text{64})`$ is the late-time variance of $`Q`$, namely, $`\mathrm{\Delta }Q(\mathrm{})^2=Q(\mathrm{})^2_\mathrm{v}`$. Its dependence on the proper acceleration $`a`$ is shown in FIG. 4.
One can see that, when $`a`$ is large, $`Q(\mathrm{})^2_\mathrm{v}`$ is nearly proportional to $`a`$, while in the zero-acceleration limit $`a0`$ with $`\eta (2\gamma )^1\mathrm{ln}|\mathrm{ln}a|`$, the saturated value goes to a positive number. From $`(\text{64})`$ and $`(\text{75})`$, one finds that
$$\underset{a0}{lim}\frac{Q(\mathrm{})^2_\mathrm{v}}{Q(0)^2_\mathrm{a}^{\mathrm{qm}}}=\frac{i\mathrm{\Omega }_r}{\pi \mathrm{\Omega }}\mathrm{ln}\frac{\gamma i\mathrm{\Omega }}{\gamma +i\mathrm{\Omega }},$$
(81)
thus $`\mathrm{\Delta }Q(\mathrm{})^2=Q(\mathrm{})^2_\mathrm{v}`$ is smaller than $`\mathrm{\Delta }Q(0)^2=Q(0)^2_\mathrm{a}^{\mathrm{qm}}`$ for every $`\gamma >0`$ when $`a0`$. In other words, for a non-accelerated detector, whose Unruh temperature is zero, the variance of $`Q`$ in the detector-field coupled system is still finite and smaller than the one for the ground state in the free theory.
Actually, $`\mathrm{\Delta }Q(\mathrm{})^2`$ will become smaller than $`\mathrm{\Delta }Q(0)^2`$ whenever $`a`$ is small enough. Observing FIG. 4, there is a critical value of $`a`$ that gives the late-time variance identical to the initial one ($`a=a_{cr}3.0447`$ in FIG. 4). Does this mean that the $`Q`$-component of the final wave-packet with $`a_{cr}`$ is in the original ground state of the free theory? The answer is no. What happens is that the quantum state of $`Q`$ has been highly entangled with the quantum state of $`\mathrm{\Phi }`$ at late times, and the value of $`\sqrt{\mathrm{\Delta }Q(\mathrm{})^2}`$ simply represents the width of the projection of the whole wave-packet (in the $`Q`$-$`\mathrm{\Phi }`$ representation of the state) onto the $`Q`$-axis. There is actually no factorizable $`Q`$-component of the wave-packet, and the final configuration of the wave-packet in $`Q`$-$`\mathrm{\Phi }`$ space looks totally different from the initial one. Indeed, with the same critical value of $`a`$, $`\mathrm{\Delta }\dot{Q}(\mathrm{})^2=\dot{Q}(\mathrm{})^2_\mathrm{v}`$ is not equal to $`\mathrm{\Delta }\dot{Q}(0)^2=\dot{Q}(0)^2_\mathrm{a}^{\mathrm{qm}}`$ for every $`\gamma 0`$.
But one can still imagine that, at $`\eta =0`$, the coherent state for the free detector is an ensemble of particles with a distribution function like $`|Q|q|^2`$ in $`Q`$-space. $`\sqrt{Q(\eta )^2_\mathrm{a}^{\mathrm{qm}}}`$ is the width of this distribution function. When $`\eta >0`$, due to the dissipation which comes with the coupling, all particles in the ensemble are going to fall into the bottom of the potential of $`Q`$, so $`Q(\eta )^2_\mathrm{a}^{\mathrm{qm}}`$ shrinks to zero. On the other hand, the vacuum fluctuations of the field act like a pressure which can push the ensemble of particles outwards, so that the width of the projection of the wave-packet in $`Q`$-space, $`\sqrt{\mathrm{\Delta }Q^2}`$, remains finite. A larger $`a`$ gives a higher Unruh temperature, and a higher outward pressure, so eventually the wave-packet reaches equilibrium with a wider projection in the potential well of $`Q`$.
### IV.4 Shift of the ground state energy
A natural definition of the energy of the dressed detector (a similar concept is that of a “dressed atom”, see e.g., Ref. CPP ; scully ) is
$$E(\eta )\frac{m_0}{2}\left[\dot{Q}^2(\eta )+\mathrm{\Omega }_r^2Q^2(\eta )\right],$$
(82)
with $`Q^2(\eta )=Q^2(\eta )_\mathrm{v}+Q^2(\eta )_\mathrm{a}`$ and $`\dot{Q}^2(\eta )=\dot{Q}^2(\eta )_\mathrm{v}+\dot{Q}^2(\eta )_\mathrm{a}`$ according to $`(\text{51})`$. In FIGs. 1-3, one can see that $`\overline{Q}`$, $`\dot{Q}^2(\eta )_a^{\mathrm{qm}}`$ and $`Q^2(\eta )_a^{\mathrm{qm}}`$ eventually die out. So the late-time energy of the dressed detector is
$`E(\mathrm{})`$ $`=`$ $`{\displaystyle \frac{m_0}{2}}\left[\dot{Q}(\mathrm{})^2_\mathrm{v}+\mathrm{\Omega }_r^2Q(\mathrm{})^2_\mathrm{v}\right]`$ (83)
$`=`$ $`{\displaystyle \frac{\mathrm{}}{2\pi }}\left\{a2\mathrm{R}\mathrm{e}\left[(\gamma +i\mathrm{\Omega })\psi _{\gamma +i\mathrm{\Omega }}\right]2\gamma \mathrm{ln}a\right\}`$
from $`(\text{64})`$ and $`(\text{69})`$. This is actually the true ground-state energy of the dressed detector, with the vacuum fluctuations of the field incorporated. The first term in $`E(\mathrm{})`$ could be interpreted as the total energy of a harmonic oscillator in thermal bath, $`k_BT_U`$, with the Unruh temperature $`T_U=\mathrm{}a/2\pi k_B`$.
The ground-state energy of the dressed detector is not identical to the one for the free detector, $`E_0=\mathrm{}\mathrm{\Omega }_r/2`$. In particular, if $`a`$ is small enough, the subtracted $`E(\mathrm{})`$ is lower than $`E_0`$, though there is an ambiguity of a constant in determining the value of the energy. This is analogous to the Lamb shift in atomic physics Milonni ; scully ; MPB93 .
## V Two-point functions of the quantum field
Similar to the two-point functions of the detector, for the initial quantum state $`(\text{50})`$, the two-point function of $`\mathrm{\Phi }`$ could be split into two parts,
$`\widehat{\mathrm{\Phi }}(x)\widehat{\mathrm{\Phi }}(x^{})`$ $`=`$ $`0_M|q\left|\left[\mathrm{\Phi }_a(x)+\mathrm{\Phi }_b(x)\right]\left[\mathrm{\Phi }_a(x^{})+\mathrm{\Phi }_b(x^{})\right]\right|q|0_M`$ (84)
$`=`$ $`G_\mathrm{v}(x,x^{})+G_\mathrm{a}(x,x^{}),`$
where, from $`(\text{12})`$ and $`(\text{18})`$,
$`G_\mathrm{v}(x,x^{})`$ $``$ $`q|q0_M|\mathrm{\Phi }_a(x)\mathrm{\Phi }_a(x^{})|0_M={\displaystyle \frac{\mathrm{}d^3k}{(2\pi )^32\omega }f^{(+)}(x;𝐤)f^{()}(x^{};𝐤)},`$ (85)
$`G_\mathrm{a}(x,x^{})`$ $``$ $`0_M|0_Mq\left|\mathrm{\Phi }_b(x)\mathrm{\Phi }_b(x^{})\right|q={\displaystyle \frac{\mathrm{}}{2\mathrm{\Omega }_rm_0}}f^a(x)f^a(x^{}).`$ (86)
Eqs.$`(\text{28})`$-$`(\text{30})`$ and $`(\text{47})`$ suggest that $`G_\mathrm{v}`$ accounts for the back reaction of the vacuum fluctuations of the scalar field on the field itself, while $`G_\mathrm{a}`$ corresponds to the dissipation of the zero-point energy of the internal degree of freedom of the detector.
Substituting $`(\text{28})`$ into $`(\text{85})`$, $`G_\mathrm{v}`$ can be decomposed into four pieces,
$$G_\mathrm{v}(x,x^{})=G_\mathrm{v}^{00}(x,x^{})+G_\mathrm{v}^{01}(x,x^{})+G_\mathrm{v}^{10}(x,x^{})+G_\mathrm{v}^{11}(x,x^{}),$$
(87)
in which $`G_\mathrm{v}^{ij}`$ are defined by
$$G_\mathrm{v}^{ij}(x,x^{})\frac{\mathrm{}d^3k}{(2\pi )^32\omega }f_i^{(+)}(x;𝐤)f_j^{()}(x^{},𝐤),$$
(88)
with $`i,j=0,1`$. $`G_\mathrm{v}^{00}`$ is actually the Green’s function for free fields, which should be subtracted to obtain the renormalized Green’s function for the interacting theory, namely,
$$G_{\mathrm{ren}}(x,x^{})\widehat{\mathrm{\Phi }}(x)\widehat{\mathrm{\Phi }}(x^{})G_\mathrm{v}^{00}(x,x^{}).$$
(89)
Since $`G_\mathrm{v}^{01}(x,x^{})=[G_\mathrm{v}^{10}(x^{},x)]^{}`$ by definition, it is sufficient to calculate $`G_\mathrm{v}^{11}(x,x^{})`$ and $`G_\mathrm{v}^{10}(x,x^{})`$ in the following.
The structure of $`G_\mathrm{v}^{11}`$ is quite simple. Comparing $`(\text{32})`$, $`(\text{60})`$ and the definition $`(\text{88})`$, one concludes that
$$G_\mathrm{v}^{11}(x,x^{})=\frac{\lambda _0^2}{(2\pi )^2a^2XX^{}}Q(\tau _{})Q(\tau _{}^{})_\mathrm{v}.$$
(90)
The result $`(\text{67})`$ can be substituted directly to get the coincidence limit of $`G_\mathrm{v}^{11}`$.
By definition, $`G_\mathrm{v}^{10}`$ accounts for the interference between the retarded solution $`f_1^{(+)}`$ and the free solution $`f_0^{(+)}`$. Since we are interested in the coincidence limit of $`G_\mathrm{v}^{10}`$, and $`f_1^{(+)}`$ vanishes in the L-wedge ($`U>0`$, $`V<0`$) and P-wedge ($`U,V<0`$) of Minkowski space, below only the $`G_\mathrm{v}^{10}(x,x^{})`$ with $`x`$ and $`x^{}`$ in the F-wedge ($`U,V>0`$) and R-wedge would be calculated.
It has been given in Ref.lin03b that
$`{\displaystyle \frac{\mathrm{}d^3k}{(2\pi )^32\omega }\phi (\kappa ,𝐤)e^{i\omega ti𝐤𝐱}}`$ $`=`$ $`{\displaystyle \frac{e^{i\kappa \tau }d\tau /(2\pi )^3}{\left(x^1z^1(\tau )\right)^2+\rho ^2\left(tz^0(\tau )+iϵ\right)^2}}`$ (91)
$`=`$ $`{\displaystyle \frac{i}{2\pi aX}}{\displaystyle \frac{\mathrm{}}{(1e^{2\pi \kappa /a})}}\left[e^{i\kappa \tau _{}}Z(\kappa )e^{i\kappa \tau _+}\right],`$
where $`ϵ>0`$, $`Z(\kappa )=1`$ and $`e^{\pi \kappa /a}`$ for $`x`$ in R and F-wedges <sup>1</sup><sup>1</sup>1Note that there is a transcription oversight in Eq.(54) of Ref.lin03b . An overall factor $`1`$ should have been on the right hand side of $`\varphi ^F`$., respectively, $`X`$ and $`\tau _{}`$ were defined in $`(\text{33})`$ and $`(\text{34})`$, and
$$\tau _+\frac{1}{a}\mathrm{ln}\frac{a}{2|U|}\left(XUV+\rho ^2+a^2\right).$$
(92)
Hence, from $`(\text{32})`$ and $`(\text{29})`$, one has
$$G_\mathrm{v}^{10}(x,x^{})=\frac{\mathrm{}\lambda _0^2\theta (\eta _{})}{(2\pi )^3m_0a^2XX^{}}\frac{d\kappa }{1e^{2\pi \kappa /a}}\underset{j}{}\frac{c_je^{i\kappa (\tau _0\tau _0^{})}}{\kappa iw_j}\left[e^{w_j\eta _{}}e^{i\kappa \eta _{}}\right]\left[e^{i\kappa \eta _{}^{}}Z(\kappa )e^{i\kappa \eta _+^{}}\right]$$
(93)
with $`\eta _\pm (x)\tau _\pm (x)\tau _0`$. The coincidence limit of $`G_\mathrm{v}^{10}`$ reads
$`G_\mathrm{v}^{10}(x,x)`$ $``$ $`\underset{x^{}x}{lim}{\displaystyle \frac{1}{2}}\left(G_\mathrm{v}^{10}(x,x^{})+G_\mathrm{v}^{10}(x^{},x)\right)`$ (94)
$`=`$ $`{\displaystyle \frac{\mathrm{}\lambda _0^2\theta (\eta _{})}{(2\pi )^3m_0\mathrm{\Omega }a^2X^2}}\mathrm{Re}\{i\psi _{\gamma +i\mathrm{\Omega }}+{\displaystyle \frac{ia}{\gamma +i\mathrm{\Omega }+a}}[e^{(\gamma +i\mathrm{\Omega }+a)\eta _{}}F_{\gamma +i\mathrm{\Omega }}(e^{a\eta _{}})`$
$`(\pm )e^{(\gamma +i\mathrm{\Omega })\eta _{}a\eta _+}F_{\gamma +i\mathrm{\Omega }}(\pm e^{a\eta _+})\pm e^{a(\eta _+\eta _{})}F_{\gamma +i\mathrm{\Omega }}(\pm e^{a(\eta _+\eta _{})})]\},`$
with $`+`$ and $``$ for $`x`$ in R and F-wedges, respectively. Near the event horizon $`U0`$, $`\eta _+`$ diverges, and the last two terms in $`(\text{94})`$ vanish.
As for $`G_\mathrm{a}(x,x^{})`$, since $`f_0^\mathrm{a}=0`$, one has $`G_\mathrm{a}^{01}=G_\mathrm{a}^{10}=0`$, and only $`G_\mathrm{a}^{11}`$ contributes to $`G_\mathrm{a}`$. Inserting $`(\text{47})`$ into $`(\text{86})`$ and comparing with $`(\text{71})`$, one finds that
$$G_\mathrm{a}(x,x^{})=\frac{\lambda _0^2}{(2\pi )^2a^2XX^{}}Q(\tau _{})Q(\tau _{}^{})_\mathrm{a}.$$
(95)
It can also be divided into a quantum part $`G_\mathrm{a}^{\mathrm{qm}}(x,x^{})`$ and a semiclassical part $`G_\mathrm{a}^{\mathrm{cl}}(x,x^{})`$ according to $`(\text{71})`$ and below.
### V.1 Effects due to the interfering term
In our (3+1) dimensional UD detector theory, the coincidence limit of the quantum part of $`G_{\mathrm{ren}}`$ reads
$`G_{\mathrm{ren}}^{\mathrm{qm}}(x,x)`$ $``$ $`G_{\mathrm{ren}}(x,x)G_\mathrm{a}^{\mathrm{cl}}(x,x)`$ (96)
$`=`$ $`G_\mathrm{a}^{\mathrm{qm}}(x,x)+G_\mathrm{v}^{11}(x,x)+G_\mathrm{v}^{10}(x,x)+G_\mathrm{v}^{01}(x,x),`$
owing to $`(\text{84})`$, $`(\text{87})`$ and $`(\text{89})`$. Collecting the results in $`(\text{90})`$, $`(\text{94})`$ and $`(\text{95})`$, it is found that $`G_{\mathrm{ren}}^{\mathrm{qm}}(x,x)`$ is singular at $`xy(\tau )`$, and one has to be more cautious.
As can be seen from $`(\text{90})`$ and $`(\text{95})`$, $`G_\mathrm{v}^{11}(x,x)`$ and $`G_\mathrm{a}^{\mathrm{qm}}(x,x)`$ look like the squares of the retarded field with effective squared scalar charge $`Q(\tau )^2_\mathrm{v}`$ and $`Q(\tau )^2_\mathrm{a}^{\mathrm{qm}}`$, respectively. Since the detector is accelerating, these two terms do carry radiated energy (this will be shown explicitly later). The interfering term $`G_\mathrm{v}^{10}(x,x)+G_\mathrm{v}^{01}(x,x)`$, is more intriguing: At first glance, it acts like a polarization in the medium, which screens the radiation field carried by $`G_\mathrm{v}^{11}(x,x)`$ and $`G_\mathrm{a}^{\mathrm{qm}}(x,x)`$. However, the interfering term $`G_\mathrm{v}^{10}+G_\mathrm{v}^{01}`$ does not respond to $`G_\mathrm{a}^{\mathrm{qm}}`$ at all – it is independent of $`f^a`$ and impervious to any information about the quantum state of $`Q`$. Hence the interfering term cannot be interpreted as the polarization in the medium. The total effect is simply a destructive interference between the field induced by the vacuum fluctuations, and the vacuum fluctuations themselves. For physical interpretations one should group $`G_\mathrm{v}^{10}+G_\mathrm{v}^{01}`$ and $`G_\mathrm{v}^{11}`$ together and leave $`G_\mathrm{a}^{\mathrm{qm}}`$ alone.
These quantities, together with their sum, are illustrated in FIGs. 5 and 6. In FIG. 5, one can see that, soon after the coupling is turned on at $`V=1/a`$, $`G_\mathrm{v}^{10}+G_\mathrm{v}^{01}`$ build up and pull the solid curve down. Observing $`(\text{94})`$, the time scale (in proper time of the detector) of this pull-down is about $`1/(\gamma +a)`$, which is shorter than the time scale $`1/2\gamma `$ for $`Q(\tau )^2_\mathrm{v}`$ and $`Q(\tau )^2_\mathrm{a}^{\mathrm{qm}}`$, since we take $`\gamma =0.1<a=1`$ here. In FIG. 6, one can also see that $`G_\mathrm{a}^{\mathrm{qm}}+G_\mathrm{v}^{11}`$ diverge as $`X^2`$ around the trajectory, while the divergence of $`G_\mathrm{v}^{10}+G_\mathrm{v}^{01}`$ as $`X0`$ is a bit weaker than $`X^2`$, such that $`X^2(G_\mathrm{v}^{10}+G_\mathrm{v}^{01})`$ goes to zero on the trajectory of the detector.
### V.2 What exactly is the “vacuum polarization cloud” around the detector?
In prior work for (1+1)D spacetime RSG the counterpart of $`G_{\mathrm{ren}}(x,x)`$ has been considered as evidence for the existence of a “vacuum polarization cloud” around the detector Unr92 ; MPB93 ; RHA . This is because $`G_{\mathrm{ren}}(x,x)`$ around the detector does not vanish even after the system reaches equilibrium, it exchanges particles with the detector, and the mean energy it carries is zero. Nevertheless, vacuum polarization is a concept pertinent to field-field quantum interacting systems. In quantum electrodynamics, electrons are described in terms of a field, which distributes in the whole spacetime, so vacuum polarization is pictured as the creation and annihilation of virtual electron-positron pairs everywhere in spacetime. These virtual electron-positron pairs do modify the field strength around the location of a point charge, yielding a non-vanishing variance of the electromagnetic (EM) field. But in the UD detector theory, at the level of precision explored here, the detector-field interaction (hence the virtual processes) only occurs on the trajectory of the detector. There is no virtual detector or scalar charge at any spatial point off the location of the UD detector.
Hence in UD detector theory “vacuum polarization cloud” is not a precise description of $`G_{\mathrm{ren}}(x,x)`$ in steady state. At late times $`G_{\mathrm{ren}}(x,x)`$ simply shows the characteristics of the field in the true vacuum state, in contrast to $`Q(\mathrm{})^2`$ for the true ground state of the detector.
## VI Radiated Power
In classical theory, the modified stress-energy tensor for a massless scalar field $`\mathrm{\Phi }`$ in Minkowski space is BD
$$T_{\mu \nu }[\mathrm{\Phi }(x)]=\left(12\xi \right)\mathrm{\Phi }_{,\mu }\mathrm{\Phi }_{,\nu }2\xi \mathrm{\Phi }\mathrm{\Phi }_{;\mu \nu }+\left(2\xi \frac{1}{2}\right)g_{\mu \nu }\mathrm{\Phi }^{,\rho }\mathrm{\Phi }_{,\rho }+\frac{\xi }{2}g_{\mu \nu }\mathrm{\Phi }\mathrm{}\mathrm{\Phi },$$
(97)
where $`\xi `$ is a field coupling parameter, set to zero here. Denote $`v^\mu =dz^\mu /d\tau `$ as the four velocity of the detector, and define the null distance $`r`$ and the spacelike unit vector $`u^\mu `$ by $`x^\mu z^\mu (\tau _{})r(u^\mu +v^\mu (\tau _{}))`$ rohr with normalization $`u_\mu u^\mu =1`$ and $`v_\mu u^\mu =0`$ (see FIG. 7). Then the stress-energy tensor for the classical retarded field $`\mathrm{\Phi }_{\mathrm{ret}}`$ induced by the UD detector moving along the trajectory $`z^\mu (\tau _{})`$ can be written as lin03b
$`T_{\mu \nu }[\mathrm{\Phi }_{\mathrm{ret}}(x)]_{\xi =0}={\displaystyle \frac{\lambda _0^2}{(4\pi )^2}}\theta (\eta _{})\{{\displaystyle \frac{1}{r^4}}Q^2(\tau _{})({\displaystyle \frac{1}{2}}g_{\mu \nu }+u_\mu u_\nu )`$
$`+{\displaystyle \frac{1}{r^3}}Q(\tau _{})\left[\dot{Q}(\tau _{})+Q(\tau _{})a_\rho u^\rho \right]\left(g_{\mu \nu }+2u_\mu u_\nu +u_\mu v_\nu +v_\mu u_\nu \right)`$
$`+{\displaystyle \frac{1}{r^2}}[\dot{Q}(\tau _{})+Q(\tau _{})a_\rho u^\rho ]^2(u_\mu +v_\mu )(u_\nu +v_\nu )\}.`$ (98)
The $`O(r^2)`$ term in the above expression corresponds to the radiation field, which carries radiated power given by rohr
$$\frac{dW^{\mathrm{rad}}}{d\tau _{}}=\underset{r\mathrm{}}{lim}r^2𝑑\mathrm{\Omega }_{\mathrm{II}}u^\mu T_{\mu \nu }v^\nu (\tau _{})=\frac{\lambda _0^2}{(4\pi )^2}𝑑\mathrm{\Omega }_{\mathrm{II}}\left[\dot{Q}(\tau _{})+Q(\tau _{})a_\rho u^\rho \right]^2$$
(99)
to the null infinity of Minkowski space. Here the $`Q`$-term corresponds to dipole radiation ($`l=1`$, $`m=0`$ in multipole expansion of the radiation field) with the angular distribution $`a_\rho u^\rho =a\mathrm{cos}\theta `$ <sup>2</sup><sup>2</sup>2The angular distribution of the dipole radiation emitted by a scalar charge depends on the choice of $`\xi `$ in the modified stress-energy tensor. For $`\xi =1/6`$, the angular distribution is $`a_\rho u^\rho =a\mathrm{sin}\theta `$, which is the same as the one for the EM radiation emitted by a electric charge in electrodynamics., while the $`\dot{Q}`$-term corresponds to monopole radiation ($`l=0`$) isotropic in the rest frame instantaneously for the UD detector at $`\tau _{}`$. The solid angle $`d\mathrm{\Omega }_{\mathrm{II}}`$ could be further integrated out, then one obtains the classical radiation formula
$$\frac{dW^{\mathrm{rad}}}{d\tau _{}}=\frac{\lambda _0^2}{4\pi }\left[\dot{Q}^2(\tau _{})+\frac{a^2}{3}Q^2(\tau _{})\right],$$
(100)
which is the counterpart of the Larmor formula for EM radiation. The second term is the usual radiation formula for the massless scalar field emitted by a constant, point-like scalar charge in acceleration RenWein .
Naively, one may expect that the quantum version of the radiation formula could look like $`(\lambda _0^2/4\pi )[\dot{Q}^2(\tau _{})+(a^2/3)Q^2(\tau _{})]`$. In the following, we shall calculate the quantum expectation value of the flux $`T_{\mu \nu }`$, from which we will see that the quantum radiation formula is more complicated than expected.
### VI.1 expectation value of the stress-energy tensor
The expectation value of the renormalized stress-energy tensor $`T_{\mu \nu }_{\mathrm{ren}}`$ is obtained by calculating
$$T_{\mu \nu }[\mathrm{\Phi }(x)]_{\mathrm{ren}}=\underset{x^{}x}{lim}\left[\frac{}{x^\mu }\frac{}{x^\nu }\frac{1}{2}g_{\mu \nu }g^{\rho \sigma }\frac{}{x^\rho }\frac{}{x^\sigma }\right]G_{\mathrm{ren}}(x,x^{}),$$
(101)
according to $`(\text{97})`$ with $`\xi =0`$. With the results in the previous section, it is straightforward to obtain $`T_{\mu \nu }_{\mathrm{ren}}`$ induced by the UAD:
$`T_{\mu \nu }[\mathrm{\Phi }(x)]_{\mathrm{ren}}`$ $`=`$ $`{\displaystyle \frac{\lambda _0^2\theta (\eta _{})}{(2\pi )^2a^2X^2}}\left[g_\mu {}_{}{}^{\rho }g_{\nu }^{}{}_{}{}^{\sigma }{\displaystyle \frac{1}{2}}g_{\mu \nu }g^{\rho \sigma }\right]\times `$ (102)
$`[\eta _{,\rho }\eta _{,\sigma }\dot{Q}(\tau _{})^2_{\mathrm{tot}}+{\displaystyle \frac{X_{,\rho }X_{,\sigma }}{X^2}}Q(\tau _{})^2_{\mathrm{tot}}`$
$`{\displaystyle \frac{X_{,\rho }}{X}}\eta _{,\sigma }Q(\tau _{})\dot{Q}(\tau _{})_{\mathrm{tot}}\eta _{,\rho }{\displaystyle \frac{X_{,\sigma }}{X}}\dot{Q}(\tau _{})Q(\tau _{})_{\mathrm{tot}}`$
$`+(\eta _{,\rho }\eta _{+,\sigma }+\eta _{+,\rho }\eta _{,\sigma }){\displaystyle \frac{\mathrm{}\mathrm{\Theta }_+}{2\pi m_0}}({\displaystyle \frac{X_{,\rho }}{X}}\eta _{+,\sigma }+\eta _{+,\rho }{\displaystyle \frac{X_{,\sigma }}{X}}){\displaystyle \frac{\mathrm{}\mathrm{\Theta }_{+X}}{2\pi m_0}}].`$
Upon collecting $`(\text{117})`$ and $`(\text{118})`$ as well as those from $`G^\mathrm{a}`$. Here $`\dot{Q}^2_{\mathrm{tot}}\dot{Q}^2+(\mathrm{}/2\pi m_0)\mathrm{\Theta }_{}`$, $`Q^2_{\mathrm{tot}}Q^2+(\mathrm{}/2\pi m_0)\mathrm{\Theta }_{XX}`$ and $`Q\dot{Q}_{\mathrm{tot}}Q\dot{Q}+(\mathrm{}/2\pi m_0)\mathrm{\Theta }_X`$ with $`\mathrm{\Theta }_{ij}`$ defined in $`(\text{119})`$-$`(\text{123})`$. To see the properties of quantum nature, we define the total variances by subtracting the semiclassical part from $`\mathrm{}_{\mathrm{tot}}`$ as
$`\mathrm{\Delta }Q^2(\tau _{})_{\mathrm{tot}}`$ $``$ $`Q^2(\tau _{})_{\mathrm{tot}}\overline{Q}^2(\tau _{})=\mathrm{\Delta }Q^2(\tau _{})+{\displaystyle \frac{\mathrm{}\mathrm{\Theta }_{XX}(\tau _{})}{2\pi m_0}}.`$ (103)
$`\mathrm{\Delta }\dot{Q}^2(\tau _{})_{\mathrm{tot}}`$ $``$ $`\dot{Q}^2(\tau _{})_{\mathrm{tot}}\dot{\overline{Q}}^2(\tau _{})=\mathrm{\Delta }\dot{Q}^2(\tau _{})+{\displaystyle \frac{\mathrm{}\mathrm{\Theta }_{}(\tau _{})}{2\pi m_0}},`$ (104)
Their evolution against $`\eta _{}`$ are illustrated in FIGs. 8 and 9.
In our case, the Minkowski coordinate $`(U,V,\rho )`$ of a spacetime point in F and R-wedge can be transformed to the coordinate $`(r,\tau _{},\theta )`$ by
$`\rho `$ $`=`$ $`r\mathrm{sin}\theta ,`$ (105)
$`V`$ $`=`$ $`re^{a\tau _{}}\left[1+\mathrm{cos}\theta +(ar)^1\right],`$ (106)
$`U`$ $`=`$ $`re^{a\tau _{}}\left[1\mathrm{cos}\theta (ar)^1\right],`$ (107)
so that $`X=2r/a`$. Also one has
$$u^\mu (\tau _{})=(r\mathrm{cos}\theta \mathrm{sinh}a\tau _{},r\mathrm{cos}\theta \mathrm{cosh}a\tau _{},\frac{x^2}{r},\frac{x^3}{r})$$
(108)
with $`\rho =\sqrt{(x^2)^2+(x^3)^2}`$. Now Eq.$`(\text{102})`$ can be directly compared with $`(\text{98})`$ and $`(\text{99})`$. One can see clearly that the $`\dot{Q}(\tau _{})^2_{\mathrm{tot}}`$ term has the same angular distribution as the one for the $`\dot{Q}^2`$ term in $`(\text{98})`$, hence would be recognized as a monopole radiation by the Minkowski observer. The angular distributions of the remaining terms in $`(\text{102})`$ are, however, much more complicated because of their dependence on $`\eta _+(r,\tau _{},\theta )`$.
### VI.2 Screening
We have mentioned in the previous section that $`G_\mathrm{a}^{\mathrm{qm}}`$ and $`G_\mathrm{v}^{11}`$ in $`(\text{95})`$ and $`(\text{90})`$ carry radiated energy, now this becomes clear. Observing that what correspond to $`_\mu _\nu ^{}[G_\mathrm{a}^{\mathrm{qm}}(x,x^{})+G_\mathrm{v}^{11}(x,x^{})]`$ are those proportional to $`\mathrm{}`$ in $`\mathrm{}_{\mathrm{tot}}`$ terms of $`(\text{102})`$. These terms contribute a positive flux. Nevertheless, due to the presence of the interfering terms $`\mathrm{\Theta }_{ij}`$, most of this positive flux of quantum nature will be screened when the system reaches steady state as $`\eta _{}\mathrm{}`$.
As shown in FIG. 8, the total variance $`\mathrm{\Delta }Q^2_{\mathrm{tot}}`$ near the event horizon $`U=0`$ drops exponentially in proper time (power-law in the Minkowski time) after the coupling is turned on. Note that $`\mathrm{\Delta }Q^2(\eta _{}(x))_{\mathrm{tot}}`$ is proportional to $`G_{\mathrm{ren}}^{\mathrm{qm}}(x,x)`$ defined in $`(\text{96})`$, and $`X`$ is independent of $`V`$ on the event horizon, so FIG. 8 is virtually the same plot as FIG. 5 except that the time variable here is $`\eta _{}`$. Thus, similar to the behavior of $`G_{\mathrm{rem}}^{\mathrm{qm}}(x,x)`$ near the event horizon, $`\mathrm{\Theta }_{XX}`$ ($`G_\mathrm{v}^{10}+G_\mathrm{v}^{01}`$) builds up and the total variance $`\mathrm{\Delta }Q(\tau _{})^2_{\mathrm{tot}}`$ is pulled down during the time scale $`1/(\gamma +a)`$ (for $`\gamma <a`$) according to $`(\text{61})`$, $`(\text{75})`$ and $`(\text{120})`$. Then $`\mathrm{\Delta }Q(\tau _{})^2_{\mathrm{tot}}`$ turns into a tail ($`\eta _{}>1`$ in FIG. 8) which exponentially approaches the saturated value $`\mathrm{}a/2\pi m_0\mathrm{\Omega }_r^2`$ with the time scale $`1/2\gamma `$.
For $`\mathrm{\Delta }\dot{Q}(\tau _{})^2_{\mathrm{tot}}`$ and $`\mathrm{\Delta }Q(\tau _{})\mathrm{\Delta }\dot{Q}(\tau _{})_{\mathrm{tot}}`$, their behaviors are similar (see FIG. 9). In particular, $`\mathrm{\Delta }\dot{Q}(\tau _{})^2_{\mathrm{tot}}=\mathrm{\Delta }\dot{Q}(\tau _{})^2+(\mathrm{}/2\pi m_0)\mathrm{\Theta }_{}`$ goes to zero at late times from $`(\text{68})`$, $`(\text{78})`$ and $`(\text{119})`$ with $`\gamma \eta _{}1`$, so the corresponding monopole radiation vanishes after the transient.
From the calculations of Ref.lin03b based on perturbation theory, it was suggested that the existence of a monopole radiation could be an experimentally distinguishable evidence of the Unruh effect. Here we find from a non-perturbative calculation that, in fact, only the transient of it could be observed. (A comparison of both results will be given in Sec.VII.) This appears to agree with the claim that for a UAD in (1+1)D, emitted radiation is only associated with nonequilibrium process CapHJ . The negative tail of $`\mathrm{\Delta }\dot{Q}(\tau _{})^2_{\mathrm{tot}}`$ in FIG. 9 and the corresponding quantum radiation could last for a long time with respect to the Minkowski observer ($`V^{2\gamma /a}`$), but this is essentially a transient. The interference between the quantum radiation induced by the vacuum fluctuations and the vacuum fluctuations themselves totally screens the information about the Unruh effect in this part of the radiation.
### VI.3 Conservation between detector energy and radiation
What is the physics behind the interfering term in $`\dot{Q}^2(\eta )_{\mathrm{tot}}`$? By inserting our results into $`(\text{82})`$ and $`(\text{104})`$, one can show that,
$`E(\eta _i)E(\eta _f)`$ $`=`$ $`{\displaystyle \frac{\lambda _0^2}{4\pi }}{\displaystyle _{\eta _i}^{\eta _f}}𝑑\eta \dot{Q}^2(\eta )_{\mathrm{tot}}`$ (109)
$`=`$ $`{\displaystyle \frac{\lambda _0^2}{4\pi }}{\displaystyle _{\eta _i}^{\eta _f}}𝑑\eta \left[\dot{\overline{Q}}^2(\eta )+\mathrm{\Delta }\dot{Q}^2(\eta )+{\displaystyle \frac{\mathrm{}\mathrm{\Theta }_{}(\eta )}{2\pi m_0}}\right],`$
for all proper time interval after the interaction is turned on $`(\eta _f>\eta _i>0)`$. The left hand side of this equality is the energy-loss of the dressed detector from $`\eta _i`$ to $`\eta _f`$, while the right hand side is the radiated energy via the monopole radiation corresponding to $`\dot{Q}^2(\eta )_{\mathrm{tot}}`$ during the same period. Therefore $`(\text{109})`$ is simply a statement of energy conservation between the detector and the field in this channel, and the interfering terms $`\mathrm{\Theta }_{}`$ must be included so that $`\dot{Q}^2(\eta )_{\mathrm{tot}}`$ is present on the right hand side instead of the naively expected $`\dot{Q}^2(\eta )`$. A simpler but more general derivation of this relation is given in Appendix B. Eq.$`(\text{109})`$ also justifies that $`(\text{82})`$ is indeed the correct form of the internal energy of the dressed detector.
With the relation $`(\text{109})`$ we can make two observations pertaining to results and procedures given before. First, while the $`\mathrm{\Lambda }_0`$-terms in $`Q^2(\tau )_\mathrm{v}`$ (Eq.(61)) and $`\dot{Q}^2(\tau )_\mathrm{v}`$ (Eq.(68)) are not included in any figure of this paper, they are consistent with the conservation law $`(\text{109})`$. Actually the $`\mathrm{\Lambda }_0`$-term in $`(\text{61})`$ satisfy the driving-force-free equation of motion $`(\text{48})`$, just like the semiclassical $`\overline{Q}`$ does.
Second, Eq.$`(\text{109})`$ implies that all the internal energy of the dressed detector dissipates via a monopole radiation, and the external agent which drives the detector along the trajectory $`(\text{5})`$ has no additional influence on this channel.
### VI.4 Quantum radiation formula
Transforming $`(\text{102})`$ to the form of $`(\text{98})`$ by applying $`(\text{105})`$-$`(\text{107})`$, one can calculate the radiation power
$$\frac{dW^{\mathrm{rad}}}{d\tau _{}}=\underset{r\mathrm{}}{lim}r^2𝑑\mathrm{\Omega }_{\mathrm{II}}u^\mu T_{\mu \nu }_{\mathrm{ren}}v^\nu (\tau _{})$$
(110)
following a similar argument in classical theory. Before calculating, let us observe the behavior of the steady-state $`r^2u^\mu T_{\mu \nu }_{\mathrm{ren}}v^\nu `$ in the forward light cone. As $`r`$ increses, the developments of two terms in late-time $`r^2u^\mu T_{\mu \nu }_{\mathrm{ren}}v^\nu `$ are illustrated in FIGs. 10 and 11. It turns out that both are regular and non-vanishing at the null infinity of Minkowski space ($`r\mathrm{}`$) even in steady state ($`\gamma \eta _{}1`$). FIGs. 10 and 11 also indicate that, near the the null infinity of Minkowski space, almost all the equal-$`r`$ surface lies in the F-wedge, except the region around $`\theta =0`$ is still in the R-wedge. The contribution to the integral around $`\theta =0`$ can be totally neglected because the value of $`r^2u^\mu T_{\mu \nu }_{\mathrm{ren}}v^\nu `$ is regular there while the measure for this portion in the angular integral is zero when $`r\mathrm{}`$. So the radiation power can be written as
$$\frac{dW^{\mathrm{rad}}}{d\tau _{}}=\frac{\lambda _0^2}{8\pi }_0^\pi 𝑑\theta \mathrm{sin}\theta \left\{\dot{Q}^2_{\mathrm{tot}}\frac{\mathrm{}\mathrm{\Theta }_+}{\pi m_0}+a^2\mathrm{cos}^2\theta Q^2_{\mathrm{tot}}+a\mathrm{cos}\theta \left[\{Q,\dot{Q}\}_{\mathrm{tot}}\frac{\mathrm{}\mathrm{\Theta }_{+X}}{\pi m_0}\right]\right\},$$
(111)
by inserting $`(\text{102})`$, $`(\text{108})`$ and $`v^\mu (\tau _{})=(\mathrm{cosh}a\tau _{},\mathrm{sinh}a\tau _{},0,0)`$ into $`(\text{110})`$. This is the quantum radiation formula for the massless scalar field emitted by the UAD in (3+1)D spacetime.
At late times, while $`\dot{Q}^2_{\mathrm{tot}}`$ ceases, it still remains a positive radiated power flow
$`{\displaystyle \frac{dW^{\mathrm{rad}}}{d\tau _{}}}\stackrel{\gamma \tau _{}\mathrm{}}{}`$
$`{\displaystyle \frac{\mathrm{}\lambda _0^2}{8\pi ^2m_0}}{\displaystyle _0^\pi }𝑑\theta \mathrm{sin}\theta \left\{{\displaystyle \frac{a}{2\mathrm{\Omega }_r^2}}a^2\mathrm{cos}^2\theta a\mathrm{sin}^2{\displaystyle \frac{\theta }{2}}{\displaystyle \frac{a}{\mathrm{\Omega }}}\mathrm{tan}^2{\displaystyle \frac{\theta }{2}}\mathrm{Re}\left[{\displaystyle \frac{i(\gamma +i\mathrm{\Omega }a\mathrm{cos}\theta )^2}{\gamma +i\mathrm{\Omega }+a}}F_{\gamma +i\mathrm{\Omega }}\left(\mathrm{tan}^2{\displaystyle \frac{\theta }{2}}\right)\right]\right\}`$
$`={\displaystyle \frac{\mathrm{}\lambda _0^2}{8\pi ^2m_0}}\left\{{\displaystyle \frac{a^3}{3\mathrm{\Omega }_r^2}}a{\displaystyle \frac{2}{3}}\left[{\displaystyle \frac{a^3}{\mathrm{\Omega }_r^2}}a+2\gamma +\mathrm{Re}\left[{\displaystyle \frac{i(\gamma +i\mathrm{\Omega })}{a\mathrm{\Omega }}}\left[(\gamma +i\mathrm{\Omega })^2a^2\right]\psi ^{(1)}\left({\displaystyle \frac{\gamma +i\mathrm{\Omega }}{a}}\right)\right]\right]\right\}`$ (112)
to the null infinity of Minkowski space. Thus we conclude that there exists a steady, positive radiated power of quantum nature emitted by the detector even when the detector is in steady state.
For large $`a`$, the first term in $`(\text{112})`$ dominates, and the radiated power is approximately
$$\frac{dW^{\mathrm{rad}}}{d\tau _{}}\frac{\lambda _0^2}{4\pi }\frac{a^2}{3}\frac{\mathrm{}a}{2\pi m_0\mathrm{\Omega }_r^2}a^2T_U,$$
(113)
where $`T_U`$ is the Unruh temperature. This could be interpreted as a hint of the Unruh effect. Note that it does not originate from the energy flux that the detector experiences in Unruh effect, since the internal energy of the dressed detector is conserved only in relation to the radiated energy of a monopole radiation corresponding to $`\dot{Q}^2_{\mathrm{tot}}`$. Learning from the EM radiation emitted by a uniformly accelerated charge rohr ; Boulware , we expect that the above non-vanishing radiated energy of quantum origin could be supplied by the external agent we introduced in the beginning to drive the motion of the detector. Further analysis on the quantum radiations of the detector involving the dynamics of the trajectory is still on-going.
## VII Comparison with Earlier Results
We can recover the results in Ref.lin03b , which is obtained by perturbation theory, as follows. In Ref.lin03b , the first order approximation of the flux $`T^{tU}_{\mathrm{ren}}`$ through the event horizon $`U=0`$ has been calculated. Here, the expectation value of $`T^{tU}`$ near the event horizon reads, from $`(\text{102})`$,
$`T^{tU}_{\mathrm{ren}}|_{U0}\underset{x^{}x}{lim}\left[2_V_V^{}+{\displaystyle \frac{1}{2}}_\rho _\rho ^{}\right]G_{\mathrm{ren}}(x,x^{})|_{U0}`$ (114)
$`=`$ $`{\displaystyle \frac{2\lambda _0^2}{(2\pi )^2a^4\left(\rho ^2+a^2\right)^2}}\theta [{\displaystyle \frac{1}{a}}\mathrm{ln}{\displaystyle \frac{a}{V}}(\rho ^2+a^2)\tau _0]\{{\displaystyle \frac{1}{V^2}}\dot{Q}(\tau _{})^2_{\mathrm{tot}}+`$
$`{\displaystyle \frac{\rho ^2}{(\rho ^2+a^2)^2}}[a^2Q(\tau _{})^2_{\mathrm{tot}}+a\dot{Q}(\tau _{})Q(\tau _{})+Q(\tau _{})\dot{Q}(\tau _{})_{\mathrm{tot}}+\dot{Q}(\tau _{})^2_{\mathrm{tot}}]\left\}\right|_{U0},`$
in which $`\mathrm{\Theta }_+`$ and $`\mathrm{\Theta }_{+X}`$ terms vanish. By letting $`\gamma 0`$ with $`\eta _{}`$ finite, then taking $`\eta _{}\mathrm{}`$, the total variance $`(\text{104})`$ becomes
$`\mathrm{\Delta }\dot{Q}(\tau _{})^2_{\mathrm{tot}}`$ $`\stackrel{\gamma 0}{}`$ $`{\displaystyle \frac{\mathrm{}\mathrm{\Omega }_r}{2m_0}}{\displaystyle \frac{\mathrm{}}{2\pi m_0}}\left\{a+{\displaystyle \frac{2a\mathrm{cos}\mathrm{\Omega }_r\eta _{}}{e^{a\eta _{}}1}}2\mathrm{\Omega }_r\mathrm{Re}\left[{\displaystyle \frac{ae^{(a+i\mathrm{\Omega }_r)\eta _{}}}{\mathrm{\Omega }_r+ia}}F_{i\mathrm{\Omega }_r}(e^{a\eta _{}})i\psi _{i\mathrm{\Omega }_r}\right]\right\}`$ (115)
$`\stackrel{\eta _{}\mathrm{}}{}`$ $`{\displaystyle \frac{\mathrm{}\mathrm{\Omega }_r}{m_0(1e^{2\pi \mathrm{\Omega }_r/a})}},`$
owing to $`(\text{68})`$, $`(\text{78})`$ and $`(\text{119})`$. This is identical to the corresponding part of Eq.(66) in Ref.lin03b ,
$$2\underset{E}{}|E_0|Q(0)|E|^2\frac{\epsilon ^2}{1e^{2\pi \epsilon }},$$
(116)
by noting that there, $`m_0=1`$, $`_E|E_0|Q(0)|E|^2=E_0|Q(0)^2|E_0=\mathrm{}/2\epsilon `$, and $`\epsilon `$ there is equal to $`\mathrm{\Omega }_r/a`$ here.
The monopole radiation corresponding to $`(\text{116})`$ looks like a constant negative flux since $`\epsilon >0`$. Accordingly it was concluded in Ref.lin03b that such a quantum monopole radiation could be experimentally distinguishable from the bremsstrahlung of the detector. At first glance this constant monopole radiation seems to contradict the knowledge gained from (1+1)D results. But actually similar results for (1+1)D cases were also obtained by Massar and Parentani (MP) MP , who found that a detector initially prepared in the ground state and coupled to a field under a smooth switching function does emit radiation during thermalization. They pointed out that the radiated flux in what they refered to as the “golden rule limit” ($`\eta \mathrm{}`$ with $`\gamma \eta `$ small, while the switching function becomes nearly constant) is approximately a constant negative flux for all $`V>0`$ <sup>3</sup><sup>3</sup>3A word of caution in terminology: Note that the golden rule yields Markovian dynamics which is the prevailing case under equilibrium conditions, as during uniform acceleration. To avoid confusion in its connotation, it is perhaps simpler and more precise just to state the condition explicitly. . In spite of the long interaction time and the nearly constant radiated flux, the detector will remain in dis-equilibrium.
Note that the initial conditions in lin03b are similar to those in Sec.II of MP, and the limiting condition for obtaining $`(\text{115})`$ is exactly what MP assumed there. Hence, the constant negative flux in lin03b is essentially a transient effect, which exists only in the period that the above stated condition holds. When the interaction time $`\eta `$ exceeds $`O(\gamma ^1)`$, this approximation breaks down. To obtain the correct late-time behavior, one should take the limit $`\eta _{}\mathrm{}`$ before $`\gamma 0`$. Then $`\mathrm{\Delta }\dot{Q}(\tau _{})^2_{\mathrm{tot}}`$ goes to zero.
## VIII Summary
In this paper, we consider the Unruh-DeWitt detector theory in (3+1) dimensional spacetime. A uniformly accelerated detector is modeled by a harmonic oscillator $`Q`$ linearly coupled with a massless scalar field $`\mathrm{\Phi }`$. The cases with the coupling constant $`|\lambda _0|`$ less than the renormalized natural frequency $`|\mathrm{\Omega }_r|`$ of the detector are considered.
We solved exactly the evolution equations for the combined system of a moving detector coupled to a quantum field in the Heisenberg picture, and from the evolution of the operators we can obtain complete information on the combined system. For the case that the initial state is a direct product of a coherent state for the detector and the Minkowski vacuum for the field, we worked out the exact two-point functions of the detector and similar functions of the field. By applying the coherent state for the detector, we can distinguish the classical behaviors from others. The quantum part of the coincidence limit of two-point functions, namely, the variances of $`Q`$ and $`\mathrm{\Phi }`$, are determined by the detector and the field together.
From the exact solutions, we were able to study the complete process from the initial transient to the final steady state. In particular, we can identify the time scales of transient behaviors analytically. When the coupling is turned on, the zero-point fluctuations of the free detector dissipates exponentially, then the vacuum fluctuations take over. The time scales for both processes are the same. Eventually the variance of $`Q`$ saturates at a finite value, where the dissipation of the detector is balanced by the input from the vacuum fluctuations of the field. Even in the zero-acceleration limit, the variances of $`Q`$ and $`\dot{Q}`$, thus the ground state energy of the interacting detector, shifted from the ones for the free detector. This fluctuations-induced effect share a similar origin with that of the Lamb shift.
The variance of $`Q`$ yields an effective squared scalar charge, which induce a positive variance in the scalar field. This variance of the field contributes a positive radiated energy at the quantum level. However, the interference between the vacuum fluctuations and the retarded solution induced by the vacuum fluctuations screens part of the emission of quantum radiation. The time scale of the screening is proportional to $`1/(\gamma +a)`$ for $`a>\gamma `$, where $`a`$ is the proper acceleration and $`\gamma `$ is the damping constant proportional to $`\lambda ^2`$. After the screening the renormalized Green’s functions of the field are still non-zero in steady state.
A quantum radiation formula determined at the null infinity of Minkowski space has been derived. We found that even in steady state there exists a positive radiated power of quantum nature emitted by the uniformly accelerated UD detector. For large $`a`$ the radiated power is proportional to $`a^2T_U`$, where $`T_U`$ is the Unruh temperature. This could be interpreted as a hint of the Unruh effect. However, the nearly constant negative flux obtained in Ref.lin03b for (3+1)D case is essentially a transient effect.
Only part of the radiation is connected to the internal energy of the detector. The total energy of the dressed detector and the radiated energy of a monopole radiation from the detector is conserved for every proper time interval after the coupling is turned on. The external agent which drives the detector’s motion has no additional influence on this channel. Since the corresponding monopole radiation of quantum nature ceases in steady state, the hint of the Unruh effect in the late-time radiated power is not directly from the energy flux experienced by the detector in Unruh effect. This extends the result in Ref.CapHR ; RSG that there is no emitted radiation of quantum origin in Unruh effect in (1+1) dimensional spacetime.
Since all the relevant quantum and statistical information about the detector (atom) and the field can be obtained from the results presented here, when appropriately generalized, they are expected to be useful for addressing issues in atomic and optical schemes of quantum information processing, such as quantum decoherence, entanglement and teleportation. These investigations are in progress.
Acknowledgments BLH thanks Alpan Raval and Mei-Ling Tseng for discussions in an earlier preliminary attempt on the 3+1 problem. We appreciate the referees’ queries and suggestions for improvements on the presentation of this paper. This work is supported in part by the NSC Taiwan under grant NSC93-2112-M-001-014 and by the US NSF under grant PHY03-00710 and PHY-0426696.
## Appendix A Derivatives of Two-Point Functions of the Field
From $`G_\mathrm{v}^{11}`$ in $`(\text{90})`$, it is easy to see that
$`_\mu _\nu ^{}G_\mathrm{v}^{11}(x,x^{})={\displaystyle \frac{\mathrm{}d^3k}{(2\pi )^32\omega }_\mu f_1^{(+)}(x)_\nu ^{}f_1^{()}(x^{})}`$ (117)
$`=`$ $`{\displaystyle \frac{\lambda _0^2}{(2\pi )^2a^2XX^{}}}\theta (\eta _{})\theta (\eta _{}^{})[{\displaystyle \frac{X_{,\mu }X_{,\nu }^{}}{XX^{}}}Q(\tau _{})Q(\tau _{}^{})_\mathrm{v}+\eta _{,\mu }\eta _{,\nu }^{}\dot{Q}(\tau _{})\dot{Q}(\tau _{}^{})_\mathrm{v}`$
$`{\displaystyle \frac{X_{,\mu }}{X}}\eta _{,\nu }^{}Q(\tau _{})\dot{Q}(\tau _{}^{})_\mathrm{v}\eta _{,\mu }{\displaystyle \frac{X_{,\nu }^{}}{X^{}}}\dot{Q}(\tau _{})Q(\tau _{}^{})_\mathrm{v}].`$
Note that the $`\delta `$-functions at $`\eta _{}=0`$, coming from the derivative of the step functions, have been neglected here.
With $`G_\mathrm{v}^{10}+G_\mathrm{v}^{01}`$, one can write down in closed form of the interfering terms in the R-wedge of the Rindler space. Under the coincidence limit, it looks like,
$`\underset{x^{}x}{lim}{\displaystyle \frac{1}{2}}\{_\mu _\nu ^{}[G_\mathrm{v}^{10}(x,x^{})+G_\mathrm{v}^{01}(x,x^{})]+(xx^{})\}`$ (118)
$`=`$ $`{\displaystyle \frac{\mathrm{}\lambda _0^2\theta (\eta _{})}{(2\pi )^3m_0a^2X^2}}[\eta _{,\mu }\eta _{,\nu }\mathrm{\Theta }_{}+{\displaystyle \frac{X_{,\mu }X_{,\nu }}{X^2}}\mathrm{\Theta }_{XX}{\displaystyle \frac{X_{,\mu }}{X}}\eta _{,\nu }\mathrm{\Theta }_X\eta _{,\mu }{\displaystyle \frac{X_{,\nu }}{X}}\mathrm{\Theta }_X`$
$`+\eta _{,\mu }\eta _{+,\nu }\mathrm{\Theta }_++\eta _{+,\mu }\eta _{,\nu }\mathrm{\Theta }_+{\displaystyle \frac{X_{,\mu }}{X}}\eta _{+,\nu }\mathrm{\Theta }_{X+}\eta _{+,\mu }{\displaystyle \frac{X_{,\nu }}{X}}\mathrm{\Theta }_{+X}].`$
where
$`\mathrm{\Theta }_{}`$ $`=`$ $`4\gamma (\mathrm{\Lambda }_1\mathrm{ln}a)a+{\displaystyle \frac{2a}{\mathrm{\Omega }}}{\displaystyle \frac{e^{\gamma \eta _{}}}{e^{a\eta _{}}1}}\left(\gamma \mathrm{sin}\mathrm{\Omega }\eta _{}\mathrm{\Omega }\mathrm{cos}\mathrm{\Omega }\eta _{}\right)`$ (119)
$`{\displaystyle \frac{2}{\mathrm{\Omega }}}\mathrm{Re}\left[i(\gamma +i\mathrm{\Omega })^2\psi _{\gamma +i\mathrm{\Omega }}+{\displaystyle \frac{ia(\gamma +i\mathrm{\Omega })^2}{\gamma +i\mathrm{\Omega }+a}}e^{(\gamma +i\mathrm{\Omega }+a)\eta _{}}F_{\gamma +i\mathrm{\Omega }}(e^{a\eta _{}})\right],`$
$`\mathrm{\Theta }_{XX}`$ $`=`$ $`{\displaystyle \frac{2}{\mathrm{\Omega }}}\mathrm{Re}\left[i\psi _{\gamma +i\mathrm{\Omega }}+{\displaystyle \frac{ia}{\gamma +i\mathrm{\Omega }+a}}e^{(\gamma +i\mathrm{\Omega }+a)\eta _{}}F_{\gamma +i\mathrm{\Omega }}(e^{a\eta _{}})\right]+2_0(x),`$ (120)
$`\mathrm{\Theta }_X`$ $`=`$ $`\mathrm{\Theta }_X={\displaystyle \frac{a}{\mathrm{\Omega }}}{\displaystyle \frac{e^{\gamma \eta _{}}}{e^{a\eta _{}}1}}\mathrm{sin}\mathrm{\Omega }\eta _{}+_1(x),`$ (121)
$`\mathrm{\Theta }_{+X}`$ $`=`$ $`\mathrm{\Theta }_{X+}={\displaystyle \frac{a}{\mathrm{\Omega }}}{\displaystyle \frac{e^{\gamma \eta _{}}}{\pm e^{a\eta _+}1}}\mathrm{sin}\mathrm{\Omega }\eta _{}_1(x),`$ (122)
$`\mathrm{\Theta }_+`$ $`=`$ $`\mathrm{\Theta }_+={\displaystyle \frac{a}{\mathrm{\Omega }}}{\displaystyle \frac{e^{\gamma \eta _{}}}{\pm e^{a\eta _+}1}}\left(\gamma \mathrm{sin}\mathrm{\Omega }\eta _{}\mathrm{\Omega }\mathrm{cos}\mathrm{\Omega }\eta _{}\right){\displaystyle \frac{a}{\pm e^{a(\eta _+\eta _{})}1}}+_2(x),`$ (123)
with
$$_n(x)\pm \frac{a}{\mathrm{\Omega }}\mathrm{Re}\left\{\frac{i(\gamma +i\mathrm{\Omega })^n}{\gamma +i\mathrm{\Omega }+a}\left[e^{a\eta _+(\gamma +i\mathrm{\Omega })\eta _{}}F_{\gamma +i\mathrm{\Omega }}(\pm e^{a\eta _+})e^{a(\eta _+\eta _{})}F_{\gamma +i\mathrm{\Omega }}(\pm e^{a(\eta _+\eta _{})})\right]\right\},$$
(124)
with $`+e^{a\eta _+}`$ and $`e^{a\eta _+}`$ for $`x`$ in R and F-wedge, respectively. Note that $`\mathrm{\Theta }_{}`$ is independent of $`\eta _+`$. $`\mathrm{\Theta }_{XX}`$ is actually proportional to $`G_\mathrm{v}^{10}`$ in $`(\text{94})`$. Another observation is that, combining $`(\text{117})`$ and $`(\text{118})`$, one finds that the divergent $`\mathrm{\Lambda }_1`$ term in $`\mathrm{\Theta }_{}`$ is canceled by the one in $`\dot{Q}^2(\eta _{})_\mathrm{v}`$.
As for $`G^\mathrm{a}`$, the result is similar to $`(\text{117})`$ with $`\mathrm{}_\mathrm{v}`$ being replaced by $`\mathrm{}_\mathrm{a}`$. Again it splits into the quantum and semiclassical parts as in Sec.IV B.
## Appendix B A simpler derivation of the conservation law
The conservation law $`(\text{109})`$ is directly obtained by arranging our somewhat complicated results. It can also be derived in a simpler way as follows.
From the definition of the ground-state energy $`(\text{82})`$ of the dressed detector, its first derivative of $`\tau `$ is
$$\dot{E}(\tau )=m_0\mathrm{Re}\left[\dot{Q}(\tau )\ddot{Q}(\tau )+\mathrm{\Omega }_r^2Q(\tau )\dot{Q}(\tau )\right]$$
(125)
Introducing the equations of motion $`(\text{38})`$ and $`(\text{45})`$ to eliminate $`\ddot{Q}(\tau )`$, one has
$`\dot{E}(\tau )`$ $``$ $`m_0\mathrm{Re}\{\dot{Q}(\tau )[2\gamma \dot{Q}(\tau )\mathrm{\Omega }_r^2Q(\tau )+{\displaystyle \frac{\lambda _0}{m_0}}\mathrm{\Phi }_0(y(\tau ))]_\mathrm{v}+\mathrm{\Omega }_r^2Q(\tau )\dot{Q}(\tau )_\mathrm{v}`$ (126)
$`+\dot{Q}(\tau )[2\gamma \dot{Q}(\tau )\mathrm{\Omega }_r^2Q(\tau )]_\mathrm{a}+\mathrm{\Omega }_r^2Q(\tau )\dot{Q}(\tau )_\mathrm{a}\}`$
$`=`$ $`2\gamma m_0\dot{Q}^2(\tau )+\lambda _0\mathrm{Re}\dot{Q}(\tau )\mathrm{\Phi }_0(y(\tau ))_\mathrm{v},`$
where the last term is a short-hand for
$$\lambda _0\mathrm{Re}\frac{\mathrm{}d^3k}{(2\pi )^32\omega }\dot{q}^{(+)}(\tau ;𝐤)f_0^{()}(z(\tau );𝐤).$$
(127)
(There is an additional term $`\lambda _0\dot{\overline{Q}}(\tau )\mathrm{\Phi }_{\mathrm{in}}(y(\tau ))`$ if a background field $`\mathrm{\Phi }_{\mathrm{in}}(x)=\mathrm{\Phi }_0(x)_\mathrm{v}`$ is present.) Substituting $`(\text{41})`$ and $`(\text{56})`$ into $`(\text{127})`$, integrating out $`d^3k`$ with the help of Eq.$`(\text{58})`$, then comparing with $`_\mu _\nu ^{}G^{10}(x,x^{})`$ in $`(\text{118})`$ where $`G^{10}`$ has the formal expression $`(\text{93})`$, one finds that $`(\text{127})`$ is actually identical to $`\mathrm{}\gamma \mathrm{\Theta }_{}(\tau )/\pi `$. Hence
$$\dot{E}(\tau )=2\gamma m_0\dot{Q}^2(\tau )_{\mathrm{tot}}.$$
(128)
Integrating both side from $`\tau =\tau _i`$ to $`\tau _f`$, one ends up with Eq.$`(\text{109})`$. |
warning/0507/quant-ph0507026.html | ar5iv | text | # Quantum Entanglement and fixed point Hopf bifurcation
## I Introduction
Entanglement is a fundamental characteristic of quantum systems which has lately provided for impressive progress in several areas such as quantum information information , quantum cryptography crypto and teleportation teleport . In this context, several measures of this property have been proposed, such as entropy-like quantities entropy , negativity negativ , concurrence concurrence and so forth. In particular, one expects strong correlations to be present in quantum phase transitions (QPT) qpt . Therefore, in this context it seems particularly interesting to study the interplay between quantum correlations and the appearance of a QPT. Since the quantification of the degree of entanglement (or correlations) in a system is not unique, in recent years several examples of specific systems using different measures have been put forth. More specifically, ground state correlations have been investigated as a function of a coupling parameter. In references schneider02 ; osborne02 ; Gvidal03 ; costi03 ; Jvidal04a ; Jvidal04b ; lambert04 ; buzek05 the ground state two atom concurrence has been studied and shown to exhibit a maximum at the transition point, in the context of some spin-spin or spin-boson models note1 . As pointed out recently by Hines-McKenzie-Milburn (HMM) hines05 , in the classical regime corresponding to the qualitative change in the quantum ground state it corresponds a change in minimum energy stable fixed point in phase space. As the parameter value of the Hamiltonian is changed, a quantum instability of the ground state is followed (at $`\lambda =\lambda _c`$), and a qualitative change in phase space structure of the system. In their work, the pitchfork (emergence of two new stable fixed point) type bifurcation has been focused and associated to a spike of the entanglement in the mean field limit ($`N\mathrm{}`$).
The purpose of the present work is firstly to add to these previous contributions a study of an integrable version of Dicke’s model dicke ; tavis , and to show that the phase transition in this case is radically different from the one in the non-integrable situation as revealed by the linear entropy (adopted here as a measure of quantum correlations)note2 . Reference hines05 proposes, based on specific model studies how the entanglement in a nonlinear bipartite system can be associated with a fixed point bifurcation in the classical dynamics. This conjecture thus contemplates pitchfork (one-to-two) bifurcation. Our results here suggest that first order transitions can present a different type of bifurcation. We show that in the integrable version of the model the abrupt change in entanglement content of the ground state (GS) is associated to a Hopf type bifurcation (one-to- infinitely many degenerate fixed points). In fact this also occurs for the Jahn-Teller hines04 and dimer dimer models. So, we suggest that when the QPT is first order the corresponding classical instability is of a Hopf type bifurcation. This idea is supported by the analysis of the GS Wigner function in the integrable and non-integrable situations. The maxima of these functions exactly follow the classical fixed points.
## II the model and the behavior of the entropy
The quantum Hamiltonian we use is written in the form
$`\widehat{H}=\widehat{H}_o+\widehat{H}_G+\widehat{H}_G^{^{}}`$ (1)
with
$`\widehat{H}_o`$ $`=`$ $`\mathrm{}\omega \widehat{a}^{}\widehat{a}+\mathrm{}ϵ\widehat{J}_z`$
$`\widehat{H}_G`$ $`=`$ $`{\displaystyle \frac{G}{\sqrt{2J}}}\left(\widehat{a}\widehat{J}_++\widehat{a}^{}\widehat{J}_{}\right)`$
$`\widehat{H}_G^{^{}}`$ $`=`$ $`{\displaystyle \frac{G^{}}{\sqrt{2J}}}\left(\widehat{a}^{}\widehat{J}_++\widehat{a}\widehat{J}_{}\right).`$ (2)
where we consider in all numerical calculations shown $`\omega =ϵ=1`$ (resonant case). When $`G=G^{}`$ we recover the usual single mode Dicke Hamiltonian. The advantage of working with Hamiltonian (1) is that the integrable regime can be obtained just by setting $`G^{}=0`$ (or $`G=0`$), and it becomes easier to explore various mixed regimes. The Super-radiant phase transition is present in all these situations at $`G+G^{}=ϵ`$ superradiance .
The non-integrable GS features of this model have been studied in ref. lambert04 using the concurrence as a measure of GS correlations. It is shown that concurrence attains a maximum when $`\lambda =\lambda _m(N)`$. This maximum is a function of the number of atoms and tends to the critical value $`\lambda _c=\lambda _m(\mathrm{})`$ as $`N\mathrm{}`$. In the present work we use the linear entropy of the atomic subsystem as a measure of the quantum correlations. We observe the same qualitative behavior which is therefore not shown here. However, for the integrable situation $`G^{}=0`$ shown in Fig.1, the atomic linear entropy (ALE) is plotted as a function of $`\lambda =G/ϵ`$ for $`N=2J=\mathrm{3,9,15}`$ atoms. The ALE presents an abrupt jump from zero to $`0.5`$ at $`\lambda _c=G/ϵ=1`$ for all values of $`J`$. Moreover, the entropy has a monotonic increase in steps of finite size: for example, before the first jump ($`\lambda <\lambda _c`$) only one basis state contributes to the ground state. It is a well known fact that below $`\lambda =\lambda _c=1`$ the ground state of the integrable version of the model is an isolated basis state $`|n=0,m=J`$. It is not connected to any other basis state via the above interaction for $`G^{}=0`$ tavis , until the interaction term of the Hamiltonian becomes dominant and the ground state becomes a state of the form $`\frac{1}{\sqrt{2}}\left(|1,J+|0,J+1\right)`$, which is orthogonal to the previous fundamental state. The larger the value of $`\lambda `$, the smaller the steps become, and finally a plateau is attained for large values of the coupling. The plateau values of the entropy increases with increasing value of $`J`$, indicating a larger number of basis states participating in the entanglement of the two subsystems. At the limit $`J>>1`$ we expect the height of the steps to remain of the same size, but the width to become shorter and shorter, whereas the plateau tends to $`1`$. In this case, the classical analog of this behavior stems from the fact that in the classical limit the equilibrium points present bifurcations at the same parameter values as in the quantum case. The bifurcations as shown in qoptics91 , in both - integrable and non-integrable - cases are also very distinct. The classical limit is discussed below.
## III The classical analog and bifurcation of equilibria
The classical Hamiltonian corresponding to eq. (1) can be obtained by a standard procedure using the coherent states of spin and boson klauder85 :
$$|w=\left(1+w\overline{w}\right)^Je^{wJ_+}|J,J$$
(3)
$$|v=e^{v\overline{v}/2}e^{vb^+}|0$$
(4)
with
$$w=\frac{p_1+iq_1}{\sqrt{4J\left(p_1^2+q_1^2\right)}},$$
(5)
$$v=\frac{1}{\sqrt{2}}\left(p_2+iq_2\right).$$
(6)
The classical Hamiltonian is defined as $`wv\left|H\right|wv`$ annals , and in this case there results a nonlinearly coupled two degrees of freedom Hamiltonian function in terms of the parametrization (5) and (6):
$`(q_1,p_1,q_2,p_2)`$ $`=`$ $`{\displaystyle \frac{\omega }{2}}\left(p_2^2+q_2^2\right)+{\displaystyle \frac{\epsilon }{2}}\left(p_1^2+q_1^2\right)\epsilon J+`$
$`{\displaystyle \frac{\sqrt{4J\left(p_1^2+q_1^2\right)}}{\sqrt{4J}}}\left(G_+p_1p_2+G_{}q_1q_2\right),`$
where $`G_\pm =G\pm G^{}`$. The integrable situation corresponds to $`G^{}=0`$ and $`G_\pm =G`$.
The stable fixed points corresponding to the equilibrium positions in phase space are defined via Hamilton’s equations:
$`\dot{q_1}=ϵp_1{\displaystyle \frac{G_+p_2}{\sqrt{2J}}}(2JH_1)^{1/2}+`$
$`+{\displaystyle \frac{p_1}{\sqrt{2J}(2JH_1)^{1/2}}}(G_+p_1p_2+G_{}q_1q_2)=0`$ (8)
$`\dot{p_1}=ϵq_1+{\displaystyle \frac{G_{}q_2}{\sqrt{2J}}}(2JH_1)^{1/2}+`$
$`+{\displaystyle \frac{q_1}{\sqrt{2J}(2JH_1)^{1/2}}}(G_+p_1p_2+G_{}q_1q_2)=0`$ (9)
$`\dot{q_2}=\omega p_2{\displaystyle \frac{G_+p_1}{\sqrt{2J}}}(2JH_1)^{1/2}=0`$ (10)
$`\dot{p_2}=\omega q_2+{\displaystyle \frac{G_{}q_1}{\sqrt{2J}}}(2JH_1)^{1/2}=0`$ (11)
where $`H_1=\frac{q_1^2+p_1^2}{2}`$.
In ref. qoptics91 it has been shown that there are two solutions for:
A. Integrable case:
i) trivial solution (stable for $`G/ϵ<1`$) :
$`q_1=p_1=q_2=p_2=0\text{, (the origin)}`$ (12)
ii) non-trivial solution (stable for $`G/ϵ>1`$):
$`R_1^2=q_1^2+p_1^2=2J\left(1{\displaystyle \frac{ϵ\omega }{G^2}}\right)`$ (13)
$`R_2^2=q_2^2+p_2^2=J\left({\displaystyle \frac{G^4ϵ^2\omega ^2}{G^2\omega ^2}}\right)`$ (14)
The above equations (12,13,14) show in each phase space a Hopf bifurcation of classical equilibria: we have a single point for the minimum energy $`E_{GS}`$ before the transition and an infinitely degenerate state represented in phase space by a circle.
B. Chaotic regime: the equilibrium point below $`G_+/ϵ=1`$ bifurcates to two points qoptics91 :
i) trivial solution, stable for $`G_+/ϵ<1`$ (and therefore $`G_{}/ϵ<1`$ ):
$`q_1=p_1=q_2=p_2=0\text{, (the origin)}`$ (15)
ii) non-trivial solution I, stable for $`G_+/ϵ>1`$ and $`G_{}/ϵ<1`$ corresponding to the pitchfork bifurcation in each phase space:
$`q_1=q_2=0\text{}p_1=\pm \left({\displaystyle \frac{2J(G_+^2ϵ\omega )}{G_+^2}}\right)^{1/2}\text{ ,}`$
$`p_2=\left({\displaystyle \frac{J(G_+^4ϵ^2\omega ^2)}{\omega ^2G_+^2}}\right)^{1/2}.`$ (16)
iii) non-trivial solution II, stable for $`G_+/ϵ>1`$ and $`G_{}/ϵ>1`$:
$`p_1=p_2=0\text{}q_1=\pm \left({\displaystyle \frac{2J(G_{}^2ϵ\omega )}{G_{}^2}}\right)^{1/2}\text{ ,}`$
$`q_2=\left({\displaystyle \frac{J(G_{}^4ϵ^2\omega ^2)}{\omega ^2G_{}^2}}\right)^{1/2}.`$ (17)
## IV Atomic Wigner Functions
The maxima of the atomic Wigner distributions of the quantum ground state displays amazing similarity with the classical bifurcations:
1) for $`\lambda <\lambda _c`$ the atomic Wigner function (AWF) in both cases are a Gaussian-like state centered at the classical equilibrium position (the origin - not shown).
2) for $`\lambda >\lambda _c`$, we separate the two cases:
(a) Hopf bifurcation: for the integrable case, there is only one non-trivial regime ($`\lambda =G/ϵ>\lambda _c`$), and we show in Fig.2a a contour plot of the AWF on the scaled atomic phase space $`(q_1/\sqrt{4J},p_1/\sqrt{4J})`$. These figures correspond to the ground state of the integrable system for $`J=4.5`$ and $`10.5`$ respectively at the same value of $`\lambda =G/ϵ=1.5>\lambda _c`$. Note that the projection of the region covered by the maxima is an annular region for both cases, and is shown in black continuous line the circle corresponding to the maxima of the AWF. In gray continuous line we show the region corresponding to the bifurcated classical stable fixed points. This shows that although qualitatively the mean field predicts correctly the change in the GS, the result does not give the position of the radius correctly. It is well known that mean field type calculations predict very well the energy (as shown for this model in qoptics91 ), but the states are not so well described. In Fig.2b a 3-dimensional plot is shown, with its negative part clearly visualizable for the case $`J=10.5`$.
(b) pitchfork bifurcation: for the non-integrable case, one regime is shown ($`\lambda _+=G_+/ϵ>\lambda _c`$ and $`\lambda _{}=G_{}/ϵ<\lambda _c`$) the non-trivial solution I is shown in Fig.3. Again, the maxima of the AWF clearly splits into a pair of peaks, connecting to the pitchfork type bifurcation shown in the classical case, and discussed in hines05 .
In order to estimate the number of basis states participating in the entanglement for different values of $`N`$, we calculated the phase space area (in units of $`\mathrm{}`$) at the height of $`50\%`$ of the AWF shown in Fig.2a for $`N=10`$ and $`N=21`$. It becomes clear that it increases with $`N`$: $`a_{10}=\frac{A(J=4.5)}{10\mathrm{}})=3.58`$ and $`a_{21}=\frac{A(J=10.5)}{21\mathrm{}})=5.32`$, thus $`a_{21}>a_{10}`$.
As for the non-integrable regime, Fig.3a shows a very distinct AWF with two bumps centered at the bifurcated classical equilibrium points. We also plot the classical trajectories projected onto the atomic phase space for an energy slightly above the value of the GS. Note again the close analogy: the trajectories are concentrated in two distinct regions whose center correspond to the center of the two peaks in the quantum AWF. Fig.3b shows quantum interference effects, in the negative part of the AWF, but it is very small, not allowing to confirm the suspicion of ref. emary03 that this GS is a Shrödinger-cat like state. It is also worth noting that there are quantitative as well as qualitative differences in what concerns the negative parts of the Wigner functions. In the integrable situation, its negative part is distributed in narrower region of phase space as compared with the non-integrable one, where negative contributions seem to be spread over a larger area. However, from the quantitative point of view, the integrable AWF exhibits a deeper negative contribution note3 .
## V Conclusions
In summary we have shown that the phase transition in Dicke’s model can be qualitatively different in the integrable regime and in the non-integrable one, the last case already discussed by other authors, although using the concurrence as a measure of entanglement. The linear entropy produces the same qualitative results.
The relationship between entanglement in the ground state at transition point, and the phase space bifurcations in the classical limit has already been noted in the literature. We have found here an example of how the type of the quantum phase transition may qualitatively alter the corresponding classical instability. In the case of the Dicke model, the integrable and non-integrable situations are shown to be qualitatively very different. In quantum terms the GS exhibits a first order QPT reflected in the classical limit as a Hopf bifurcation which is very different from the already discussed non-integrable situation (second order QPT and pitchfork bifurcation). The close quantum and classical analogy is also strongly reflected in the corresponding GS atomic Wigner functions.
###### Acknowledgements.
It is a pleasure to acknowledge W. D. Heiss for encouraging the present work. We thank financial support from CAPES and CNPq (Conselho Nacional de Pesquisa, Brazil) and FAPESP. |
warning/0507/cond-mat0507343.html | ar5iv | text | # Superstructure in nano-crystalline 𝐴𝑙₅₀𝐶𝑢₂₈𝐹𝑒₂₂ alloy
## I Introduction
Recently there has been a considerable scientific and technological interest in the formation of nano-crystalline/quasicrystalline phase in the $`AlCuFe`$ alloys by mechanical milling . Quasicrystals have many properties which make them interesting for industrial applications like light weight, large strength to weight ratio and high hardness with a low frictional coefficient . Nanostructured material, which can be defined as a material with the crystallite size less than $`100nm`$ are synthesized by either bottom-up or top-down processes . The bottom up approach starts with atoms, ions or molecules as building blocks and assembles nanoscale clusters or bulk material from them. The top down approach for processing of nanostructured materials starts with bulk solid and ends in obtaining a nano ostructured phase through special processing routes e.g. mechanical milling, re-solidification through chemical methods etc. Nano-phase materials have significantly different behaviour from their macroscopic counterparts because their sizes are smaller than the characteristic length scales of physical phenomena occurring in bulk materials . The nanostructure materials are produced by using various methods, among which high-energy ball milling (BM) which is also known as mechanical milling (MM) has attracted much attention . The advantages of high-energy ball milling for the synthesis of nano-structured materials are the formation of a more homogeneous product and good reproducibility . The mechanical milling technique has been used to obtain amorphous alloys , high coercively permanent magnetic metallic compounds and quasicrystals \[9-10\]. The formation of a quasicrystalline phase by BM/MM has been reported in a number of $`Al`$ and $`Ti`$ based systems . Recently Mukhopadhyay et al have studied the effect of mechanical milling on the stability of $`Al`$-rich, $`AlCuFe`$ and $`AlCuCo`$ quasicrystalline alloys. They have reported that the icosahedral quasicrystalline phase in $`AlCuFe`$ system undergoes transformation to a bcc (B2 type) crystalline phase as a result of ball milling . In this case, B2 phase does not transform into any other crystalline/quasicrystalline phase during isothermal annealing at 850$`{}_{}{}^{0}C`$ up to 20h. It has been concluded that the B2 phase is more stable than the icosahedral quasicrystalline phase at those compositions. It should be mentioned that $`Al`$-rich $`Al_{65}Cu_{20}Fe_{15}`$ system is of significant interest due to the high temperature structural stability of icosahedral quasicrystalline phase. The available phase equilibria data indicate that the B2 phase is a major phase on the $`Al`$-deficient side of stoichiometry . The B2 structure can be understood in term of ordering in a bcc lattice and converting it to be a simple cubic lattice. Therefore, unlike a bcc lattice, one type of atom occupies the body-centered position and another type occupies the cube corners in the ordered lattice. When the composition deviates from the stoichiometry, the compositional defects must be introduced to preserve the crystal structure. Its unit cell contains two different atoms located respectively at the vertex and at the center of the cube. It is one of the basic simple structures that can transform into more complex structures via twinning at the atomic level, termed as chemical twinning . The B2 type phase is often present together with the quasicrystals and has fixed coherent orientation relationship with the latter \[16-17\]. The detailed investigation of the B2 phase is also important due to its practical applications \[18-19\]. Though quasicrystals have many curious properties, they are also extremely brittle, porous and composition-sensitive. It is therefore interesting to substitute them by approximant materials, particularly B2 based ones, which are more easily prepared and have similar performance characteristics . The purpose of the present study is to investigate the influence of high-energy ball milling on the phase stability, crystallite size, lattice strain and lattice parameter of B2 phase formed in the pre-alloyed $`Al_{50}Cu_{28}Fe_{22}`$ sample. Present investigation clearly shows the evolution of ordered simple cubic phase ($`a_{sc}=4.12\AA `$) and as well as fcc ($`a_{fcc}=5.8\AA `$) $`\tau 2`$ phase after milling followed by annealing. The evolution of the nano-structure at different stages of ball milling has also been investigated.
## II Experimental Details:
An alloy of composition $`Al_{50}Cu_{28}Fe_{22}`$ was prepared by melting the high purity $`Al,Cu`$ and $`Fe`$ metals in an induction furnace, in the presence of dry argon atmosphere. The ingot formed was re-melted several times to ensure better homogeneity. The as-cast ingot was crushed to particles less than $`0.5`$ mm in size and placed in an attritor ball mill (Szegvari Attritor) with a ball to powder weight ratio of $`40:1`$. The attritor has a cylindrical stainless steel tank of inner diameter $`13cm`$ and the angular speed of mill was maintained at 400 rpm. The milling operation was conducted from 5 to 80h using hexane as a process control agent. The powder obtained after $`10h`$ and $`80h`$ of milling were annealed isothermally at $`500^0C`$ for 5 to $`20h`$ in the evacuated quartz capsules (with vacuum of $`10^6`$ $`torr`$). The milled and heat-treated powders were characterized by powder X-ray diffraction (XRD) using a Philips 1710 X-ray diffractometer with $`CuK_\alpha `$ radiation. The effective crystallite size and relative strain of mechanically milled powders as well as heat-treated products were calculated based on line broadening of XRD peaks. The use of the Voigt function for the analysis of the integral breadths of broadened X-ray diffraction line profiles forms the basis of a rapid and powerful single line method of crystallite-size and strain determination. In this case the constituent Couchy and Gaussian components can be obtained from the ratio of full width at half maximum intensity($`2\omega `$) and integral breadth ($`\beta `$) . In a single line analysis the apparent crystallite size ’D’ and strain ’e’ can be related to Couchy ($`\beta _c`$) and Gaussian ($`\beta _G`$) widths of the diffraction peak at the Bragg angle ;
$$D=\frac{\lambda }{\beta _ccos\theta }$$
(1)
and
$$e=\frac{\beta _G}{4tan\theta }$$
(2)
The constituent Couchy and Gassian components can be given as
$$\beta _c=(a_0+a_1\psi +a_2\psi ^2)\beta $$
(3)
$$\beta _G=[b_0+b_{1/2}(\psi 2/\pi )^{1/2}+b_1\psi +b_2\psi ^2)]\beta $$
(4)
where $`a_0,a_1`$ and $`a_2`$ are Couchy constants $`b_0,b_{1/2},b_1`$ and $`b_2`$ are Gassian constants and $`\psi `$ is $`2\omega /\beta `$ where $`\beta `$ is the integral breadth obtained from XRD peak. The value of Couchy and Gassian constant have taken from the table of Langford
$`a_0=2.0207,a_1=0.4803,a_2=1.7756`$
$`b_0=0.6420,b_{1/2}=1.4187,b_1=2.2043,b_2=1.8706`$
From these, we have calculated the crystallite size D and the lattice strain ’e’ for the milled $`Al_{50}Cu_{28}Fe_{22}`$ powders.
## III Results and discussion
The X-ray diffraction (XRD) patterns for the $`Al_{50}Cu_{28}Fe_{22}`$ alloy obtained after various milling durations has been shown in figure 1. Curve (a) shows the B2 phase obtained from the as-cast ingot material and curve (b), (c), (d), and (e) are the XRD patterns from the powder milled for $`5h,10h,20h,40h`$ and $`80h`$ respectively. It can be seen from Fig.1 that the peak corresponding to the (110) profiles of the B2 phase becomes broader and the intensity gets reduced with increasing milling time. These two effects are mainly attributed to the increase of the internal lattice strain and reduction of the grain size. The evolution of the nano crystalline phase in $`Al_{50}Cu_{28}Fe_{22}`$ can be easily noticed from the increase in the broadening of x-ray diffraction lines (Fig.1) during different period of ball milling. It should be noted that the shift of the peaks towards lower $`\theta `$ angle side with the increase in milling time indicates the increase in lattice parameter. Intensity of (110) peak goes on decreasing with increasing milling time. After $`80h`$ of milling, a diffuse broad peak appears indicating the transformation of the B2 phase to an amorphous phase. (see Fig. 1e). XRD pattern clearly indicates that the initial sharp diffraction lines get considerably broadened after ball milling, suggesting that the nano-crystalline phase appears as a result of milling.
To study the effect of annealing on the MM powders, the samples milled for 10 hrs and 80hrs were subjected to annealing for various time periods. Fig. 2 (a), (b),(c) and (d) show the XRD pattern obtained after $`10h`$ of ball milling followed by annealing at $`500^0C`$ for $`0h,5h,10h`$ and $`20h`$ respectively. The XRD patterns corresponding to $`10h`$ and $`20h`$ , the annealed samples (Fig.2 (b,c)) have been indexed using simple a cubic structure with $`a_{sc}=4.1\AA `$ . The most interesting feature to be noted is that these samples show cubic structure, which is a superstructure of the B2 phase. The lattice parameters of the superstructure phase and B2 phase are interrelated as $`a_{sc}=\sqrt{2}a_p`$, where $`a_p`$ is the lattice parameter of the B2 phase which is a parent phase. The formation of the superstructure due to the ball milling and subsequent annealing has been observed for the first time in $`AlCuFe`$ alloy. We propose a structural model for this new superstructure of B2 phase. Fig. 3 shows the structural model of superstructure with lattice parameter $`\sqrt{2}a`$ times of B2 phase ($`a_p=2.911\AA `$), which is obtained from the powder after ball milling for 10 h and subsequent annealing at 500$`{}_{}{}^{0}C`$ for $`20h`$. Figure 3(a) shows two-dimensional model of superstructure, which clearly indicates that the face diagonal of a cube is playing a key role for the formation of the superstructure of B2 phase. The three dimensional model (fig. 3(b)) clearly depicts the unit cell of the superstructure of the B2 phase. The edge of the unit cell is the diagonal of a cube face ($`\sqrt{2}a_p=4.18\AA `$). The lattice parameter of the superstructure phase, which is calculated from the model, exactly matches with the lattice parameter obtained from XRD of 10h ball milled and 20h-annealed powders.
Fig. 4 (a) (b) (c) and (d) show the XRD patterns corresponding to $`80h`$ ball milled $`Al_{50}Cu_{28}Fe_{22}`$ powders annealed at 500$`{}_{}{}^{0}C`$ for 0, 5, 10 and $`20h`$ respectively. In the case of $`20h`$ annealing, the sample has been cooled slowly to avoid the quenching effects and detect any transformation during cooling. Unlike the sample milled for $`10h`$ and annealed at 500$`{}_{}{}^{0}C`$ for 10/20 h, the XRD pattern shown in fig.4 (d) could be indexed only in terms of a fcc structure with $`a=5.84\AA `$. However, the lattice parameter of this fcc phase is also related to B2 phase as $`a_{fcc}=2a_p`$. The evolution of $`\tau 2`$ phases in 80 h ball milled and 20h annealed powder can easily be explained by the structural model shown in Fig.5. We propose two-dimensional geometrical model as outlined for the formation of $`\tau 2`$ phases from the present B2 phase. Figure 5(b) shows the relationship between quasicrystllaine and related crystalline ($`\tau 2`$) phases. The model is based on the concept of a polytope and an eight dimensional root lattice. Its basic atomic cluster forms a cuboctahedron with 12 vertices formed by intersection of three perpendicular squares with edge length of $`\sqrt{2}.\sqrt{2}a`$. This polyhedron is transformed into an icosahedron when the squares are changed into rectangles with edge length ratio of $`\tau :1`$. The properties of the eight dimensional root lattice give foundation to the possibility of mapping a quasicrystllaine structure on a crystalline structure . The proposed geometrical model can be applied also to the polymorphoic bcc-fcc transformation
## IV Conclusions
On the basis of our present investigation it may be conclude that, the formation of the nano B2 phase starts after 5h of milling and gets completed after 40h of milling. Beyond 40h of milling the amorphous phase starts forming and the sample shows the coexistence of nano-crystalline and amorphous phases. After 80h of milling nano B2 phase transforms to amorphous phase completely. The $`Al_{50}Cu_{28}Fe_{22}`$ samples ball milled for 10h and 80h and annealed subsequently at 500$`{}_{}{}^{0}C`$ for 20h, transform to the simple cubic ( $`a_{sc}=\sqrt{2}a_p`$) and the fcc ($`a_{fcc}=2a_p`$) phases respectively.
Acknowledgement
The authors would like to thank Prof. A. R. Verma, Prof. S. Ranganathan, Prof. S. Lele and Prof. B.S.Murty for many stimulating discussions. The financial support from Ministry of Non-Conventional Energy Sources (MNES), New Delhi, India is gratefully acknowledged. One of the authors (TPY) acknowledges the support of CSIR for award of senior research fellowship. |
warning/0507/nlin0507028.html | ar5iv | text | # DAMTP-2005-63 Angularly localized Skyrmions
## 1 Introduction
The connection between the quantum and classical descriptions of a many–body system is an important but rather tricky one. In nuclei, the existence of a rotational band, a sequence of states whose energy increases with angular momentum $`j`$ approximately as $`\frac{\mathrm{}^2}{2I}j(j+1)`$, where $`I`$ represents a moment of inertia, suggests the existence of a static intrinsic classical shape to the nucleus which is not spherically symmetric . It is not obvious how this classical shape arises, and it is hard to predict the shape, but one can partially reconstruct it from the spectrum.
For a rigid body, the quantum states of various angular momenta are given by Wigner functions $`D_{sm}^j(\alpha ,\beta ,\gamma )`$, where $`\alpha ,\beta ,\gamma `$ are the Euler angles parametrizing the orientation, $`j`$ is the total angular momentum and $`s,m`$ its components with respect to the body–fixed and space–fixed third axis. Symmetries of the body constrain the possible $`s`$–values or combinations of $`s`$–values that can occur. A classically oriented state is a $`\delta `$–function in the Euler angles. This can be obtained by taking an infinite linear combination of Wigner functions. For a body with symmetry, one would take a sum of $`\delta `$–functions on a set of orientations related by symmetry (which are not distinguishable). Even if there is no fundamental rigid body to start with, one can consider these linear combinations. Thus, given a rotational band of states, one can construct a classically oriented state by taking an infinite linear combination of true quantum states of definite angular momentum. The properties of this oriented state (e.g. the particle density) would define the nature of the intrinsic state.
Something like this has been done in certain condensed matter situations. One may construct a classically oriented state when all that is rigorously available is quantum states labelled by angular momentum. Cooper et al. have studied a model of rotating states of a Bose condensate trapped in a harmonic well (whose shape essentially makes the condensate two–dimensional). By numerically combining precise states over a range of angular momenta, they have shown that a condensate with vortices can be obtained. In the rotating frame, these vortices form a static array, and so are angularly localized despite the rotational invariance of the problem. The vortices do not really exist in any of the states of definite angular momentum, but they do in the combined state, and they can also be physically observed.
This localization depends on the system being large. Ideally, the moment of inertia should be almost infinite. In that case, the angular momentum states of different $`j`$ are almost degenerate, and the angular localization may be achieved at almost no energy cost. (Similarly, an object with large mass can be spatially localized by taking a superposition of momentum states.)
Unfortunately, for nuclei, this is not always a realistic way to proceed. For large nuclei, like $`\mathrm{Hf}^{170}`$, there are many states in a rotational band, and it is pretty clear that an intrinsic nuclear shape exists . For smaller nuclei, however, at most a few low–lying states can be identified as forming a rotational band, and their energy separation is quite large because the moments of inertia are smaller. Not much is known about the wavefunctions of the states in the band, so it is hard to consider linear combinations. Instead it is better to postulate some intrinsic shape and fit its parameters to data. In this way, it is found, for example, that the $`\mathrm{Ne}^{20}`$ nucleus has a prolate deformation , but one cannot say it is exactly a prolate ellipsoid.
An alternative treatment of many–body systems can give angularly (and spatially) localized states much more easily. This is the approach based on an effective field theory, for example a Ginsburg–Landau description of a Bose condensate. Here, classical solutions of the field equation can naturally exhibit spatial order, for example an array of vortices. Because of the underlying symmetries, the classical solution is not unique, but is parametrized by collective coordinates describing, say, the center of mass position and angular orientation. Comparison with the previous discussion suggests that effective field theory can only be valid for large systems of many particles. To reconstruct quantum states of definite angular momentum, one may quantize the collective coordinates; this makes sense if the mass and the moment of inertia are of finite, but not infinite magnitude. A critical comparison of exact quantum states and classical solutions of an effective field theory has been carried out for quantum Hall ferromagnets by Abolfath et al. .
In this paper, we shall consider the Skyrme model and its connection with nuclei and their various angular momentum states. The Skyrme model is an effective field theory of pions, with a topological quantum number that can be identified with baryon number . Skyrme’s original idea was that the model is justified because nuclei can be thought of as made up of a condensate of many light pions (with a topological winding). Recently, the justification is based on the idea that each nucleon is made of $`N_c`$ quarks, where $`SU(N_c)`$ is the gauge group of QCD, so a nucleus of baryon number $`B`$ is made of a large number, $`N_cB`$, of quarks . The Skyrme model has a semi–rigorous standing if $`N_c`$ is large, but it is a controversial matter whether the physical value $`N_c=3`$ is sufficiently large.
The classical Skyrme field equation, like that of the Ginsburg–Landau model, can be solved numerically and much is known about its minimal energy solutions (especially for pion mass equal to zero) . Most importantly, the classical shapes of the solutions, and their symmetries, are known for values of $`B`$ up to and beyond 20 (and work is underway to take account of the finite pion mass, which could have a qualitatively significant effect for $`B10`$). These classical shapes could represent the intrinsic shapes of nuclei of modest size.
The shapes obtained have no obvious relation to shapes of nuclei as understood using other models, in particular, models based on point nucleons. For example, four–nucleon potential models are used to describe the $`\alpha `$–particle, and the classical minimum occurs for a tetrahedral configuration of the nucleons . In the Skyrme model, the solution of minimal energy with $`B=4`$ has cubic symmetry.
Our aim in this paper is to bridge the gap between the classical Skyrmion shapes and the quantum states of nuclei. The traditional approach has been to quantize the collective coordinates of Skyrmions, seek the lowest energy states consistent with the allowed values of the angular momentum, and compare with the ground state properties of nuclei. This approach has some success in reproducing the known spins of nuclei, especially for even baryon number. More recently, a table of allowed angular momenta for the ground and first excited states of rotationally quantized Skyrmions has been constructed . However, in these quantum states of definite angular momentum, the original Skyrmion shape information is sometimes completely lost.
We cannot consider an infinite linear combination of angular momentum states, as we expect that large angular momenta will lead to Skyrmion deformations, or if these are suppressed, then to infinite energy. Instead, here we shall consider an intermediate picture. By taking a small combination of low–lying angular momentum states, we partially reconstruct the shape of the classical Skyrmion solution. We shall optimize the angular localization of the Skyrmion within the limited combinations of states at our disposal. Such a finite combination of states has finite energy (not necessarily very much higher than the ground state). If one could investigate theoretically (or experimentally, although this could be difficult) the same combination of angular momentum states in another nuclear model, one might see better the connection with the Skyrme model picture.
For the $`B=1`$ case we apparently do not have the problem of orientation because a single Skyrmion has a spherically symmetric density. However, the Skyrmion still has rotational collective coordinates, and we will show that a particular combination of $`j=\frac{1}{2}`$ and $`j=\frac{3}{2}`$ states gives the most localized wave function. We also show that the ground state of the deuteron (the $`j=1`$ state of the $`B=2`$ Skyrmion), without an admixture of higher angular momentum states, retains the toroidal symmetry of the classical solution. Forest et al. have argued that not only the pure deuteron state, but also deuteron clusters within larger nuclei, show toroidal structure .
Finally, we shall show that a combination of $`j=0`$ and $`j=4`$ collective states of the $`B=4`$ cubic Skyrmion gives a state close to the classically oriented Skyrmion. The same combination of $`j=0`$ and $`j=4`$ states in a four–nucleon potential model could be compared. Now it is inevitable that in the potential model, the state will have cubic symmetry because of the Wigner functions involved. In that sense our discussion is purely kinematic. However, some detailed properties of the state (density, currents) might show a close similarity with the Skyrme picture.
This paper is restricted to the $`B=1,2,4`$ cases, and is organized as follows. Section 2 contains a review of the Skyrme model (for more details see ). In Section 3 we give an outline of the rational map approximation for Skyrmions, and its consequences, and recall the rational maps for the $`B=1`$, $`B=2`$ and $`B=4`$ Skyrmions. In Section 4, we use the $`j=\frac{1}{2}`$ and $`j=\frac{3}{2}`$ quantum states of a $`B=1`$ Skyrmion introduced in , and find the combination which gives the best localized wave function. In Sections 5 and 6 we use the rational maps introduced in Section 3 to find the “best” wavefunctions for $`B=2`$ and $`B=4`$ Skyrmions, respectively, and calculate the quantum baryon density in these states. In Section 7 we briefly discuss the implications of adding vibrational modes, and summarize our conclusions.
## 2 The Skyrme model
The Skyrme model is an effective low energy theory of QCD attempting to treat pions, nucleons and nuclei. The topological soliton solutions arising from this model can be interpreted as baryons.
The model is defined by the Lagrangian
$`L`$ $`=`$ $`{\displaystyle }\{{\displaystyle \frac{F_\pi ^2}{16}}\text{Tr}(_\mu U^\mu U^{})`$
$`+{\displaystyle \frac{1}{32e^2}}\text{Tr}([_\mu UU^{},_\nu UU^{}][^\mu UU^{},^\nu UU^{}])\}d^3x,`$
where $`U(t,\text{x})`$ is an $`SU(2)`$–valued scalar field. $`F_\pi `$ and $`e`$ are parameters which can be scaled away by using energy and length units of $`F_\pi /4e`$ and $`2/eF_\pi `$, respectively. Thus, with the values of $`F_\pi `$ and $`e`$ as in our units are related to conventional units via
$$\frac{F_\pi }{4e}=5.58\text{MeV},\frac{2}{eF_\pi }=0.755\text{fm}.$$
Introducing the $`su(2)`$–valued right current $`R_\mu =(_\mu U)U^{}`$ and using geometrical units, the Lagrangian (1) may be rewritten in the concise form
$$L=\left\{\frac{1}{2}\text{Tr}(R_\mu R^\mu )+\frac{1}{16}\text{Tr}([R_\mu ,R_\nu ][R^\mu ,R^\nu ])\right\}d^3x.$$
(2)
The Euler–Lagrange equation which follows from (2) is the Skyrme equation
$$_\mu \left(R^\mu +\frac{1}{4}[R_\nu ,[R^\nu ,R^\mu ]]\right)=0.$$
(3)
Static solutions are the stationary points (either minima or saddle points) of the energy function
$$E=\frac{1}{12\pi ^2}\left\{\frac{1}{2}\text{Tr}(R_iR_i)\frac{1}{16}\text{Tr}([R_i,R_j][R_i,R_j])\right\}d^3x,$$
(4)
where we have introduced the additional factor $`1/12\pi ^2`$ for convenience.
$`U`$, at fixed time, is a map from $`^3`$ into $`S^3`$, the group manifold of $`SU(2)`$. However, the boundary condition $`U1`$ implies a one–point compactification of space, so that topologically $`U`$: $`S^3S^3`$, where the domain $`S^3`$ is identified with $`^3\{\mathrm{}\}`$. As the homotopy group $`\pi _3(S^3)`$ is $``$, maps between 3–spheres are indexed by an integer, which is denoted by $`B`$. This integer is also the degree of the map $`U`$ and has the explicit representation
$$B=\frac{1}{24\pi ^2}\epsilon _{ijk}\text{Tr}(R_iR_jR_k)d^3x.$$
(5)
As $`B`$ is a topological invariant, it is conserved under continuous deformations of the field, including time evolution. This conserved topological charge Skyrme identified with baryon number.
Static fields of minimal energy, solving the Skyrme equation, are called multi–Skyrmions (Skyrmions, for short). They have been constructed numerically for $`B`$ up to 22 , and the symmetries of these solutions have been identified. For $`B=1`$ the Skyrmion has spherical symmetry, and for $`B=2`$ toroidal symmetry. It turns out that Skyrmions have non–trivial discrete symmetries for $`B>2`$. Solutions for negative $`B`$ are obtained by the transformation $`UU^{}`$, which preserves the energy.
## 3 Rational map ansatz
In what follows, we will be using the rational map approximation to Skyrmions . Rational maps were first introduced into the theory of three–dimensional solitons by Jarvis , in the context of monopoles, but they prove to be very useful for Skyrmions as well.
Rational maps are maps from $`S^2S^2`$, whereas Skyrmions are maps from $`^3S^3`$. The idea in is to identify the domain $`S^2`$ of the rational map with concentric spheres in $`^3`$, and the target of the rational map $`S^2`$ with spheres of latitude on $`S^3`$. A point in $`^3`$ can be parametrized by $`(r,z)`$; $`r`$ denotes radial distance and the complex variable $`z`$ specifies the direction. Via stereographic projection $`z`$ can be written in terms of usual polar coordinates $`\theta `$ and $`\varphi `$ as $`z=\mathrm{tan}(\theta /2)e^{i\varphi }`$. A rational map may be written as $`R(z)=p(z)/q(z)`$, where $`p(z)`$ and $`q(z)`$ are polynomials in $`z`$. The degree of the rational map, $`N`$, is the greater of the algebraic degrees of the polynomials $`p`$ and $`q`$. $`N`$ is also the topological degree of the map (its homotopy class) as a map from $`S^2S^2`$.
The point $`z`$ on $`S^2`$ corresponds to the unit vector
$$\widehat{𝐧}_z=\frac{1}{1+|z|^2}(z+\overline{z},i(\overline{z}z),1|z|^2).$$
(6)
Similarly, the value of the rational map $`R`$ is associated with the unit vector
$$\widehat{𝐧}_R=\frac{1}{1+|R|^2}(R+\overline{R},i(\overline{R}R),1|R|^2).$$
(7)
The ansatz for the Skyrme field, depending on a rational map $`R(z)`$ and a radial profile function $`f(r)`$, is
$$U(r,z)=\mathrm{exp}(if(r)\widehat{𝐧}_{R(z)}𝝉),$$
(8)
where $`𝝉=(\tau _1,\tau _2,\tau _3)`$ denotes the triplet of Pauli matrices, and $`f(r)`$ satisfies $`f(0)=\pi `$, $`f(\mathrm{})=0`$.
An $`SU(2)`$ Möbius transformation of $`z`$ corresponds to a rotation in physical space; an $`SU(2)`$ Möbius transformation of $`R`$ (i.e. on the target $`S^2`$) corresponds to an isospin rotation. Both these transformations of a rational map are symmetries of the Skyrme model, and both preserve $`N`$.
It can be verified that the baryon number for the ansatz (8) is given by
$$B=\frac{f^{}}{2\pi ^2}\left(\frac{\mathrm{sin}f}{r}\right)^2\left(\frac{1+|z|^2}{1+|R|^2}\left|\frac{dR}{dz}\right|\right)^2\frac{2idzd\overline{z}}{(1+|z|^2)^2}r^2𝑑r.$$
(9)
$`2idzd\overline{z}/(1+|z|^2)^2`$ is equivalent to the usual 2–sphere area element $`\mathrm{sin}\theta d\theta d\varphi `$. The angular part of the integrand,
$$\left(\frac{1+|z|^2}{1+|R|^2}\left|\frac{dR}{dz}\right|\right)^2\frac{2idzd\overline{z}}{(1+|z|^2)^2},$$
(10)
is precisely the pull–back of the area form $`2idRd\overline{R}/(1+|R|^2)^2`$ on the target sphere of the rational map $`R`$, so its integral is $`4\pi `$ times the degree $`N`$ of the map. Therefore (9) simplifies to
$$B=\frac{2N}{\pi }_0^{\mathrm{}}f^{}\mathrm{sin}^2fdr=N.$$
(11)
An attractive feature of the rational map ansatz is that it leads to a simple energy expression which can be separately minimized with respect to the rational map $`R`$ and the profile function $`f`$ to obtain close approximations to the numerical, exact Skyrmion solutions, and having the correct symmetries. (The numerical solutions are in fact best found by starting from the optimal rational map approximations.)
Indeed, using (8) we get the following expression for the energy (4):
$$E=\frac{1}{3\pi }_0^{\mathrm{}}\left(r^2f^2+2N\mathrm{sin}^2f(f^2+1)+\frac{\mathrm{sin}^4f}{r^2}\right)𝑑r.$$
(12)
Here $``$ denotes the purely angular integral
$$=\frac{1}{4\pi }\left(\frac{1+|z|^2}{1+|R|^2}\left|\frac{dR}{dz}\right|\right)^4\frac{2idzd\overline{z}}{(1+|z|^2)^2},$$
(13)
which only depends on the rational map $`R`$. To minimize $`E`$, for maps of given degree $`N`$, one should first minimize $``$ over all maps of degree $`N`$. Then the profile function $`f`$, minimizing the energy (12), may be found by numerically solving a second order, ordinary differential equation with $`N`$ and $``$ as parameters.
For $`B=1`$, the rational map is $`R(z)=z`$, and this reproduces Skyrme’s hedgehog ansatz , which is exactly satisfied by the $`B=1`$ Skyrmion. For $`B=2`$ and $`B=4`$ the symmetries of the computed Skyrmions are $`D_\mathrm{}h`$ and $`O_h`$ respectively, and in each case there is a unique rational map of the desired degree with the given symmetry, which also minimizes $``$. They are, respectively,
$$R(z)=z^2,R(z)=\frac{z^4+2\sqrt{3}iz^2+1}{z^42\sqrt{3}iz^2+1}.$$
(14)
In all these cases, we have made a convenient choice of orientation in presenting the maps.
When quantizing the Skyrme field, we will be interested in the behavior of the wavefunction with respect to different orientations of the Skyrmion configurations. Consequently, all the information we need will be encoded in the angular dependence of the baryon density (10), which only depends on the rational map, and the profile function $`f`$ will not be of much interest for our purposes.
## 4 $`B=1`$ case
The $`B=1`$ Skyrmion is spherically symmetric and takes the hedgehog form
$$U_0(𝐱)=\mathrm{exp}\{if(r)\widehat{𝐱}𝝉)\},$$
(15)
where $`f(0)=\pi `$ and $`f(\mathrm{})=0`$. If $`U_0`$ is the soliton solution, then $`U=AU_0A^1`$, where $`A`$ is an arbitrary constant $`SU(2)`$ matrix, is a static solution as well. But, in order to get solitons which are eigenstates of spin and isospin one needs to treat $`A`$ as a collective coordinate. So substitute
$$U=A(t)U_0A^1(t)$$
in the Lagrangian (1), where $`A(t)`$ is an arbitrary time–dependent $`SU(2)`$ matrix. The Lagrangian for $`A`$ is
$$L=M+\lambda \text{Tr}(_0A_0A^1),$$
(16)
where $`M`$ is the soliton mass and $`\lambda `$ is an inertia constant which may be found numerically.
The $`SU(2)`$ matrix $`A`$ can be written as $`A=a_0+i𝐚𝝉`$, with $`a_0^2+|𝐚|^2=1`$. In terms of $`a_\xi (\xi =0,1,2,3)`$ the Lagrangian (16) becomes
$$L=M+2\lambda \underset{\xi =0}{\overset{3}{}}(\dot{a}_\xi )^2,$$
(17)
and after the usual quantization procedure one gets the Hamiltonian
$$H=M+\frac{1}{8\lambda }\underset{\xi =0}{\overset{3}{}}\left(\frac{^2}{a_\xi ^2}\right).$$
(18)
Because of the constraint $`a_0^2+|𝐚|^2=1`$, the operator $`_{\xi =0}^3\left(^2/a_\xi ^2\right)`$ is to be interpreted as the laplacian $`^2`$ on the 3–sphere. The wavefunctions can be expressed as traceless, symmetric, homogeneous polynomials in the $`a_\xi `$.
Using the isospin and spin operators
$`I_k={\displaystyle \frac{1}{2}}i\left(a_0{\displaystyle \frac{}{a_k}}a_k{\displaystyle \frac{}{a_0}}ϵ_{klm}a_l{\displaystyle \frac{}{a_m}}\right),`$
$`J_k={\displaystyle \frac{1}{2}}i\left(a_k{\displaystyle \frac{}{a_0}}a_0{\displaystyle \frac{}{a_k}}ϵ_{klm}a_l{\displaystyle \frac{}{a_m}}\right),`$ (19)
Adkins, Nappi and Witten have found the normalized wavefunctions for neutron, proton and $`\mathrm{\Delta }`$–resonances . The wavefunctions we require, having opposite $`I_3`$ and $`J_3`$ eigenvalues, are
$`|n,s_z={\displaystyle \frac{1}{2}}`$ $`=`$ $`{\displaystyle \frac{i}{\pi }}(a_0+ia_3),`$
$`|p,s_z={\displaystyle \frac{1}{2}}`$ $`=`$ $`{\displaystyle \frac{i}{\pi }}(a_0ia_3),`$
$`|\mathrm{\Delta }^{},s_z={\displaystyle \frac{3}{2}}`$ $`=`$ $`{\displaystyle \frac{\sqrt{2}}{\pi }}(a_0+ia_3)^3,`$
$`|\mathrm{\Delta }^0,s_z={\displaystyle \frac{1}{2}}`$ $`=`$ $`{\displaystyle \frac{\sqrt{2}}{\pi }}(a_0+ia_3)(13(a_1^2+a_2^2)),`$
$`|\mathrm{\Delta }^+,s_z={\displaystyle \frac{1}{2}}`$ $`=`$ $`{\displaystyle \frac{\sqrt{2}}{\pi }}(a_0ia_3)(13(a_1^2+a_2^2)),`$
$`|\mathrm{\Delta }^{++},s_z={\displaystyle \frac{3}{2}}`$ $`=`$ $`{\displaystyle \frac{\sqrt{2}}{\pi }}(a_0ia_3)^3.`$ (20)
But none of these ”pure” $`j=1/2`$ and $`j=3/2`$ states is the best localized wavefunction. This will instead be given by a superposition of the above states.
If we could take into account an infinite number of angular momentum states the most localized wavefunction would be the Dirac delta function, which may be expressed in the following form
$$\delta (\mu )=\underset{j}{}(2j+1)\chi ^j(\mu ),j=0,\frac{1}{2},1,\frac{3}{2},\mathrm{},$$
(21)
where $`\chi ^j(\mu )`$ is the character of the representation of dimension $`(2j+1)`$, and $`a_0=\mathrm{cos}\mu `$. However, this wavefunction does not respect the Finkelstein–Rubinstein (FR) constraints , which in the case of one Skyrmion requires that the wavefunction is antisymmetric under $`AA`$, thus ensuring that the quantized Skyrmion is a fermion. The sum in (21) must therefore be restricted to half–integer values of $`j`$, giving the total $`\frac{1}{2}(\delta (\mu )\delta (\mu \pi ))`$.
For the $`SU(2)`$ group, the representation matrices are matrices of Wigner functions, i.e. for each $`j`$
$$𝐃^j=\left(\begin{array}{cccc}D_{jj}^j& \mathrm{}& D_{jj}^j& \\ \mathrm{}& \mathrm{}& \mathrm{}& \\ D_{jj}^j& \mathrm{}& D_{jj}^j& \end{array}\right).$$
In what follows we will be interested in the $`j=1/2`$ and $`j=3/2`$ cases, for which the Wigner functions take a concise form in terms of $`a_0,\mathrm{},a_3`$. The character $`\chi ^j`$ is the trace of the above matrix, consequently let us write down the diagonal elements:
$`D_{1/2,1/2}^{1/2}`$ $`=`$ $`a_0+ia_3,`$
$`D_{1/2,1/2}^{1/2}`$ $`=`$ $`a_0ia_3,`$
$`D_{3/2,3/2}^{3/2}`$ $`=`$ $`(a_0+ia_3)^3,`$
$`D_{1/2,1/2}^{3/2}`$ $`=`$ $`(a_0+ia_3)(13(a_1^2+a_2^2)),`$
$`D_{1/2,1/2}^{3/2}`$ $`=`$ $`(a_0ia_3)(13(a_1^2+a_2^2)),`$
$`D_{3/2,3/2}^{3/2}`$ $`=`$ $`(a_0ia_3)^3.`$ (22)
If we truncate the sum (21) at $`j=3/2`$ we get the following candidate for a well localized (normalized) wavefunction,
$$\mathrm{\Psi }(a_0,a_1,a_2,a_3)=\frac{8}{\pi }\sqrt{\frac{2}{5}}\left(a_0^3\frac{3}{8}a_0\right).$$
(23)
In terms of nucleon and $`\mathrm{\Delta }`$–resonance states this can be written as
$$\frac{1}{\sqrt{5}}\left(|\mathrm{\Delta }^{}|\mathrm{\Delta }^0|\mathrm{\Delta }^++|\mathrm{\Delta }^{++}\frac{i}{\sqrt{2}}(|n|p)\right),$$
(24)
with spins as in (20). A more general wavefunction of this type is
$$\mathrm{\Psi }(a_0,a_1,a_2,a_3)=\frac{\sqrt{2}}{\pi }\left(\frac{5}{16}+\kappa +\kappa ^2\right)^{1/2}\left(a_0^3+\kappa a_0\right).$$
(25)
The maximum magnitude of $`\mathrm{\Psi }`$ at $`a_0=\pm 1`$ occurs when $`\kappa =3/8`$, confirming that this is the best localized wavefunction.
Another measure of how well the wavefunction is localized around $`a_0=\pm 1`$ is given by the integral
$$\frac{2}{\pi ^2}\left(\frac{5}{16}+\kappa +\kappa ^2\right)^1_0^\pi a_0^2|a_0^3+\kappa a_0|^2𝑑\mathrm{\Omega }.$$
(26)
Here $`a_0=\mathrm{cos}\mu `$ and $`d\mathrm{\Omega }=4\pi \mathrm{sin}^2\mu d\mu `$ is the measure of integration. After an easy calculation, we find that this integral is maximal when $`\kappa =1/4`$, which is close to the result we got before. One more wavefunction worth considering is
$$\mathrm{\Psi }=\frac{4}{\pi }\sqrt{\frac{2}{5}}a_0^3,$$
which is as well localized as the one with $`\kappa =3/8`$ according to criterion (26), and rather simpler. It is the following combination of nucleon and $`\mathrm{\Delta }`$ states:
$$\frac{1}{2\sqrt{5}}\left(|\mathrm{\Delta }^{}|\mathrm{\Delta }^0|\mathrm{\Delta }^++|\mathrm{\Delta }^{++}2\sqrt{2}i(|n|p)\right).$$
(27)
These localized states are not physically important for isolated nucleons; however, they could be useful for modelling nucleons in interaction. Recent developments have shown that, for example, the deuteron is not formed from a proton and neutron only, but probably also contains some amount of $`\mathrm{\Delta }`$–resonances . Therefore, considering a superposition of states with different angular momenta is definitely physically meaningful. In the deuteron was modelled by a bound state of Skyrmions in the attractive channel, where the relative orientation of the Skyrmions was chosen to maximize the attraction at short range. Such states could be approximated by the combined $`j=1/2`$ and $`j=3/2`$ states we have discussed here. The dependence of the force between two Skyrmions on their relative orientation is the classical analogue of the tensor force between nucleons, and it appears to automatically lead to an admixture of a $`\mathrm{\Delta }`$–resonance component to each nucleon.
## 5 $`B=2`$ case
The $`B=2`$ Skyrmion has $`D_\mathrm{}h`$ symmetry and a toroidal shape , and is used to describe the deuteron. We take the symmetry axis to be the third body–fixed axis, and the Skyrmion to be in its standard orientation if this coincides with the third Cartesian axis in space. In the rigid body approximation to quantization, the wavefunction is a function only of the rotational and isospin collective coordinates. (We ignore the translational collective coordinates, and set the momentum to zero.) To make an appropriate quantization we have to impose FR constraints, which tell us that the ground state has the quantum numbers $`(i,j)=(0,1)`$, where $`i`$ is the total isospin and $`j`$ is the total spin. The wavefunction describing this deuteron state was obtained in . Since $`i=0`$, there is no dependence on the isospin collective coordinates, and the (normalized) state is
$$\mathrm{\Psi }=\sqrt{\frac{3}{8\pi ^2}}D_{0m}^1(\alpha ,\beta ,\gamma ).$$
(28)
Here $`\alpha `$, $`\beta `$ and $`\gamma `$ are the rotational Euler angles, $`D_{0m}^1(\alpha ,\beta ,\gamma )`$ is a Wigner function, and $`m`$ is the third component of the space–fixed spin.
In , the analysis was extended to include one vibrational mode of the system, allowing the $`B=2`$ toroidal Skyrmion to separate into two $`B=1`$ Skyrmions. The wavefunction therefore includes a factor $`u(\rho )`$, the radial part of the deuteron wavefunction, which satisfies a radial Schrödinger equation on the interval $`[\rho _0,\mathrm{})`$, where $`\rho _0`$ corresponds to the toroidal configuration. Here, however, we consider only the rigid body rotational states, and their angular dependence.
Since we are particularly interested in the spatial orientation, we treat states differing in $`m`$ as different. The state we are looking for has to have the same symmetry properties as the classical solution. Consequently, the desired wavefunction is
$$\mathrm{\Psi }=\sqrt{\frac{3}{8\pi ^2}}D_{00}^1(\alpha ,\beta ,\gamma )=\sqrt{\frac{3}{8\pi ^2}}\mathrm{cos}\beta ,$$
(29)
which is axially symmetric both on the left and on the right (i.e. with respect to the body–fixed symmetry axis, and the $`x_3`$-axis in space). $`\mathrm{\Psi }`$ has its maximum magnitude at $`\beta =0`$ and $`\beta =\pi `$, corresponding to the Skyrmion in its standard orientation, and turned up-side down, which is classically indistinguishable after an isospin rotation.
Now, given the orientational quantum state (29) we may calculate the nucleon density, and find how quantum effects change the density of the classical configuration. We find the expression for the baryon density distribution $`\rho _\mathrm{\Psi }(𝐱)`$ in physical space (which is interpreted as nucleon density) by averaging the classical baryon density over orientations weighted with $`|\mathrm{\Psi }|^2`$.
The density in the quantum state is therefore
$$\rho _\mathrm{\Psi }(𝐱)=(D(A)^1𝐱)|\mathrm{\Psi }(A)|^2\mathrm{sin}\beta d\alpha d\beta d\gamma .$$
(30)
Here $`A`$ stands for the $`SU(2)`$ matrix parametrized by Euler angles $`\alpha `$, $`\beta `$, $`\gamma `$, and $`D(A)`$ for the $`SO(3)`$ matrix associated to $`A`$ via
$$D(A)_{ab}=\frac{1}{2}\mathrm{Tr}(\tau _aA\tau _bA^{}).$$
(31)
As was already mentioned, the $`B=2`$ rational map is $`R(z)=z^2`$, and this gives a good approximation to the $`B=2`$ Skyrmion solution. It leads, using (9), to the following expression for the classical baryon density:
$$(r,z)=\frac{1}{\pi }\left(\frac{1+|z|^2}{1+|z|^4}|z|\right)^2g(r),$$
(32)
where $`g(r)`$ is a radial function. $`g(r)`$ is unaffected by the quantum averaging, so we ignore it from now on. In terms of polar angles, the angular dependence of $``$ is given by
$$=\frac{1}{\pi }\frac{(1+\mathrm{tan}^2(\frac{\theta }{2}))^2\mathrm{tan}^2(\frac{\theta }{2})}{(1+\mathrm{tan}^4(\frac{\theta }{2}))^2},$$
(33)
where this is normalized to have angular integral equal to 2, the degree of the rational map.
To evaluate $`\rho _\mathrm{\Psi }(𝐱)`$ we first expand $``$ in terms of spherical harmonics $`Y_{lm}(\theta ,\varphi )`$:
$$=\underset{l,m}{}c_{lm}Y_{lm}(\theta ,\varphi ),$$
(34)
where, because of axial symmetry, there are only terms with $`m=0`$. Although (34) is an infinite series it is a good approximation to take just the first two non–zero terms of the sum,
$$=c_{00}Y_{00}(\theta )+c_{20}Y_{20}(\theta ),$$
(35)
as all the other terms contribute less than a $`5\%`$ correction. Because the map $`R`$ has degree 2, $`c_{00}=1/\sqrt{\pi }`$; also, we find numerically that $`c_{20}=0.36`$. Then $`(D(A)^1𝐱)`$ can be written as
$$(\stackrel{~}{𝐱})=c_{00}Y_{00}(\stackrel{~}{\theta })+c_{20}Y_{20}(\stackrel{~}{\theta }),$$
(36)
where $`\stackrel{~}{𝐱}=D(A)^1𝐱`$ and similarly for $`\stackrel{~}{\theta }`$, $`\stackrel{~}{\varphi }`$. Using the transformation properties of spherical harmonics under rotations,
$$Y_{lm}(\stackrel{~}{\theta },\stackrel{~}{\varphi })=\underset{k}{}D_{mk}^l(A)^{}Y_{lk}(\theta ,\varphi ),(\mathrm{no}\mathrm{sum}\mathrm{on}l),$$
(37)
the fact that $`|\mathrm{\Psi }|^2=(3/8\pi ^2)D_{00}^1(A)D_{00}^1(A)^{}`$, the orthogonality properties of the Wigner functions
$$D_{ab}^j(A)D_{cd}^j^{}(A)^{}\mathrm{sin}\beta d\alpha d\beta d\gamma =\frac{8\pi ^2}{2j+1}\delta ^{jj^{}}\delta _{ac}\delta _{bd},$$
(38)
and (in terms of the Wigner $`3j`$ symbols)
$$D_{ab}^j(A)D_{cd}^j^{}(A)D_{ef}^{j^{\prime \prime }}(A)\mathrm{sin}\beta d\alpha d\beta d\gamma =8\pi ^2\left(\begin{array}{cccc}j& j^{}& j^{\prime \prime }& \\ a& c& e& \end{array}\right)\left(\begin{array}{cccc}j& j^{}& j^{\prime \prime }& \\ b& d& f& \end{array}\right),$$
(39)
we find directly from (30) that the quantum probability distribution is
$$\rho _\mathrm{\Psi }=c_{00}Y_{00}+\frac{2}{5}c_{20}Y_{20}.$$
(40)
This is an exact expression – no higher terms are present. We see that it resembles the classical distribution (35), but is more dominated by the first term. Thus, when quantum effects are included, the classical toroidal density remains, but is smoothed out to become more spherically symmetric.
## 6 $`B=4`$ case
The minimal energy $`B=4`$ solution has cubic symmetry; the region of high baryon density resembles a rounded cube with holes in the faces and at the centre . We define the orthogonal body–fixed axes to be those passing through the face centres, and the standard orientation of the cube to be where these axes are aligned with the Cartesian axes in space. We shall again consider the Skyrmion as a rigid body, which means the configuration is not allowed to vibrate. It was shown in that in this case the ground state, representing the $`\alpha `$–particle, has quantum numbers $`i=0`$ and $`j=0`$, with the (unnormalized) wavefunction $`\mathrm{\Psi }^{(0)}=1`$ being independent of the rotational and isospin collective coordinates. The first excited state has $`i=0`$ and $`j=4`$ and is
$$\mathrm{\Psi }_m^{(4)}=D_{4m}^4(\alpha ,\beta ,\gamma )+\sqrt{\frac{14}{5}}D_{0m}^4(\alpha ,\beta ,\gamma )+D_{4m}^4(\alpha ,\beta ,\gamma ).$$
(41)
In the third component of the space–fixed spin, $`m`$, was arbitrary. The structure of (41) is required by the cubic symmetry with respect to body–fixed axes.
But just fixing $`m`$ is not enough to make the wavefunction cubically symmetric both on the left and on the right, i.e. also with respect to space–fixed axes. To achieve this we need to take the following linear combination of the above wavefunctions:
$$\mathrm{\Psi }^{(4)}=\mathrm{\Psi }_4^{(4)}+\sqrt{\frac{14}{5}}\mathrm{\Psi }_0^{(4)}+\mathrm{\Psi }_4^{(4)}.$$
(42)
The cubic symmetry in space is fairly obvious by analogy with (41), and can be verified as follows. First note that symmetry under $`90^{}`$ rotations about the $`x^3`$-axis implies that all possible terms in (42) with $`m`$ other than $`\pm 4,0`$ vanish. To simplify the calculations a bit further we then introduce new variables
$$a=\mathrm{cos}\left(\frac{\beta }{2}\right)e^{\frac{1}{2}i\gamma }e^{\frac{1}{2}i\alpha },b=\mathrm{sin}\left(\frac{\beta }{2}\right)e^{\frac{1}{2}i\gamma }e^{\frac{1}{2}i\alpha }.$$
(43)
Obviously they satisfy $`|a|^2+|b|^2=1`$. In terms of $`a`$ and $`b`$, the $`SU(2)`$ orientation matrix parametrized by Euler angles $`\alpha ,\beta ,\gamma `$ is
$$A=\left(\begin{array}{cccc}a& b& & \\ \overline{b}& \overline{a}& & \end{array}\right).$$
In this notation the wavefunctions for different $`m`$ take the following compact form:
$`\mathrm{\Psi }_4^{(4)}`$ $`=`$ $`a^8+14a^4b^4+b^8`$
$`\mathrm{\Psi }_0^{(4)}`$ $`=`$ $`\sqrt{70}\left(a^4\overline{b}^4+\overline{a}^4b^4+{\displaystyle \frac{1}{40}}(330(|a^2||b^2|)^2+35(|a^2||b^2|)^4)\right)`$
$`\mathrm{\Psi }_4^{(4)}`$ $`=`$ $`\overline{a}^8+14\overline{a}^4\overline{b}^4+\overline{b}^8.`$ (44)
Therefore the wavefunction (42) in terms of $`a`$ and $`b`$ is
$`\mathrm{\Psi }^{(4)}`$ $`=`$ $`2\text{Re}(a^8+14a^4b^4+b^8)+14(a^4\overline{b}^4+\overline{a}^4b^4)`$ (45)
$`+{\displaystyle \frac{7}{20}}\left(330(|a^2||b^2|)^2+35(|a^2||b^2|)^4\right).`$
As expected, it is real. By acting on $`A`$ with the generators of the cubic group:
$`\left(\begin{array}{cccc}\frac{1+i}{\sqrt{2}}& 0& & \\ 0& \frac{1i}{\sqrt{2}}& & \end{array}\right),\left(\begin{array}{cccc}\frac{1+i}{2}& \frac{1i}{2}& & \\ \frac{1+i}{2}& \frac{1i}{2}& & \end{array}\right),`$ (50)
corresponding to a $`90^{}`$ rotation around a face of the cube, and a $`120^{}`$ rotation around a diagonal of the cube, we find the resulting transformations of $`(a,b)`$, and it is easy to check that $`\mathrm{\Psi }^{(4)}`$ is cubically symmetric both on the left and on the right.
The wavefunction $`\mathrm{\Psi }^{(4)}`$ has a positive maximum of $`\frac{24}{5}`$ at the identity, $`(a,b)=(1,0)`$, and at all other elements of the (double cover of the) cubic group. This is as desired, as it corresponds to the Skyrmion having a high probability to be in its standard orientation. But $`\mathrm{\Psi }^{(4)}`$ also has a negative minimum of $`\frac{104}{45}`$, which gives a further local maximum of $`|\mathrm{\Psi }^{(4)}|^2`$, at an orientation obtained by a $`60^{}`$ rotation around a diagonal of the cube, which is far from the standard orientation. We wish to suppress this.
We can do this by being a bit more sophisticated than in the $`B=2`$ case. We still have the freedom of adding an arbitrary constant to the wavefunction. This means taking a superposition of the ground and first excited states, $`\mathrm{\Psi }^{(0)}`$ and $`\mathrm{\Psi }^{(4)}`$:
$$\mathrm{\Psi }=\mathrm{\Psi }^{(4)}+\kappa \mathrm{\Psi }^{(0)}.$$
(51)
Here again we are interested in the nucleon density of the configuration. Our goal will be to adjust the constant $`\kappa `$ to get a quantum distribution as close as possible to the classical one. As in the $`B=2`$ case we define the quantum nuclear density via
$$\rho _\mathrm{\Psi }(𝐱)=(D(A)^1𝐱)|\mathrm{\Psi }(A)|^2\mathrm{sin}\beta d\alpha d\beta d\gamma ,$$
(52)
where $`(𝐱)`$ is the classical baryon density of the $`B=4`$ Skyrmion in its standard orientation. Using the rational map (14), we find that $``$ has the angular dependence
$$=\frac{12}{\pi }|z|^2(1+|z|^2)^2\frac{(z^4\overline{z}^4z^4\overline{z}^4+1)}{(z^4\overline{z}^4+z^4+12z^2\overline{z}^2+\overline{z}^4+1)^2}.$$
(53)
Expressed in terms of polar angles,
$``$ $`=`$ $`{\displaystyle \frac{12}{\pi }}\mathrm{tan}^2\left({\displaystyle \frac{\theta }{2}}\right)\left(1+\mathrm{tan}^2\left({\displaystyle \frac{\theta }{2}}\right)\right)^2`$
$`\times {\displaystyle \frac{(\mathrm{tan}^8(\frac{\theta }{2})2\mathrm{tan}^4(\frac{\theta }{2})\mathrm{cos}4\varphi +1)}{\left(\mathrm{tan}^8(\frac{\theta }{2})+2\mathrm{tan}^4(\frac{\theta }{2})\mathrm{cos}4\varphi +12\mathrm{tan}^4(\frac{\theta }{2})+1\right)^2}},`$
which may be expanded in the following form:
$$=d_0Y_{00}+d_4Z_4(\theta ,\varphi )+d_6Z_6(\theta ,\varphi )+d_8Z_8(\theta ,\varphi )+\mathrm{}.$$
(55)
Here $`Z_4,Z_6`$ and $`Z_8`$ are the unique cubically symmetric combinations of spherical harmonics with, respectively $`l=4,6`$ and 8:<sup>1</sup><sup>1</sup>1These can be derived by combining the generating, cubically symmetric Cartesian polynomials $`x^2+y^2+z^2`$, $`x^4+y^4+z^4`$, $`x^6+y^6+z^6`$, and finding the combinations which satisfy Laplace’s equation .
$`Z_4`$ $`=`$ $`Y_{44}+\sqrt{{\displaystyle \frac{14}{5}}}Y_{40}+Y_{44},`$
$`Z_6`$ $`=`$ $`Y_{64}\sqrt{{\displaystyle \frac{2}{7}}}Y_{60}+Y_{64},`$
$`Z_8`$ $`=`$ $`Y_{88}+\sqrt{{\displaystyle \frac{28}{65}}}Y_{84}+\sqrt{{\displaystyle \frac{198}{65}}}Y_{80}+\sqrt{{\displaystyle \frac{28}{65}}}Y_{84}+Y_{88}.`$ (56)
The leading coefficient is $`d_0=2/\sqrt{\pi }`$ because the rational map has degree $`4`$, and by numerical calculation we find that $`d_4=0.28`$, $`d_6=0.032`$ and $`d_8=0.024`$. Then, by a similar calculation as in the $`B=2`$ case, normalizing the wavefunction and using the orthogonality properties of the Wigner functions, we find the following numerical result for the angular dependence of the quantum baryon density:
$`\rho _\mathrm{\Psi }=d_0Y_{00}+{\displaystyle \frac{4}{2.56+\kappa ^2}}\left\{(0.038+0.075\kappa )Z_40.006Z_6+0.002Z_8\right\},`$ (57)
which is again a finite sum, all the further terms being zero.
In (57), $`\kappa `$ is not yet specified. Let us adjust it in such a way that the above distribution looks as close as possible to the classical one, i.e. let us maximize the coefficient of the $`l=4`$ terms:
$$\frac{4}{2.56+\kappa ^2}(0.038+0.075\kappa ).$$
The maximum is at $`\kappa 1.17`$, which leads to the following expression for the quantum baryon density:
$`\rho _\mathrm{\Psi }`$ $``$ $`1.13Y_{00}0.13Z_40.006Z_6+0.002Z_8`$ (58)
$``$ $`d_0Y_{00}+0.46d_4Z_4+0.2d_6Z_6+0.1d_8Z_8.`$
Thus in the $`B=4`$ case, as in the $`B=2`$ case, one can find a quantum state which localizes the Skyrmion close to its standard orientation, and which preserves the symmetry of the classical solution. However, the inclusion of quantum effects smoothes the classical baryon density, making it rather closer to spherically symmetric. Again, the effect is to approximately halve the leading non-constant harmonics, here with $`l=4`$. If we considered the pure $`j=4`$ state $`\mathrm{\Psi }^{(4)}`$, we would get
$$\rho _{\mathrm{\Psi }^{(4)}}d_0Y_{00}+0.2d_4Z_4+0.3d_6Z_6+0.15d_8Z_8,$$
(59)
which is much closer to spherically symmetric.
We can also find the energy of our state $`\mathrm{\Psi }`$; it is
$$E=\frac{1}{2.56+\kappa ^2}\left(2.56E_{j=4}+\kappa ^2E_{j=0}\right)0.65E_{j=4}+0.35E_{j=0},$$
(60)
so it is not as highly excited as a pure $`j=4`$ state.
The combination of $`j=0`$ and $`j=4`$ states, $`\mathrm{\Psi }`$, is a bit artificial as the quantum state of a free $`B=4`$ Skyrmion, but would make sense if we were dealing with interacting Skyrmions (for example, when describing compound nuclei such as $`\mathrm{Be}^8,\mathrm{C}^{12}`$ in the Skyrme model equivalent of the $`\alpha `$–particle model). Here we expect the relative orientations of the $`B=4`$ subclusters to be rather precisely fixed when they are close together, so as to minimize their potential energy.
## 7 Conclusions
Three well–localized wavefunctions of the $`B=1`$ Skyrmion have been considered and some of their advantages and physical implications have been discussed. The $`B=2`$ and $`B=4`$ minimal energy Skyrmion solutions have been quantized in such a way that the wavefunctions have the same symmetry properties as the classical Skyrmions (respectively, axial and cubic symmetry both on the left and on the right), and angularly localized quantum states with shapes closest to the classical solutions have been found. In the $`B=4`$ case, a superposition of two low–lying states of definite angular momentum needed to be considered. It is impossible to completely reproduce the shape of the classical solution this way. The quantum state necessarily smoothes out the classical baryon density, making it closer to being spherically symmetric.
All our results were obtained in the rigid body approximation, i.e. we did not allow the Skyrmions to vibrate. Considering the vibrational modes may be an interesting topic for future work. The vibrational modes for the $`B=2`$ and $`B=4`$ Skyrmions were calculated in and a qualitative analysis has been given for $`B=7`$ . The vibration frequencies obtained can be separated into those below and those above the breather mode, which is the oscillation corresponding to a change in scale size of the Skyrmion. In the modes below the breather, the Skyrmion tends to splits up into individual $`B=1`$ Skyrmions or small $`B`$ Skyrmion clusters.
But treating the vibration modes as harmonic oscillators is not very accurate, since, as the minimal energy configuration separates into individual Skyrmions the potential flattens out. A more accurate treatment would involve estimating the inter–Skyrmion potential at intermediate and large separations. Thus it should not be expected that the inclusion of the zero point energy of harmonic vibrational modes will yield accurate results for masses, binding energies of states, etc.
The vibrational modes also couple to the rotational degrees of freedom, which complicates the analysis of the rotational and isospin wavefunctions .
## Acknowledgements
O.M. thanks EPSRC for the award of a Dorothy Hodgkin Scholarship. N.S.M. thanks Nigel Cooper for a helpful discussion, and ECT\*, Trento for hospitality. |
warning/0507/cond-mat0507527.html | ar5iv | text | # Full Counting Statistics in Strongly Interacting Systems: Non-Markovian Effects
## Abstract
We present a theory of full counting statistics for electron transport through interacting electron systems with non-Markovian dynamics. We illustrate our approach for transport through a single-level quantum dot and a metallic single-electron transistor to second order in the tunnel-coupling strength, and discuss under which circumstances non-Markovian effects appear in the transport properties.
The study of current fluctuations in mesoscopic systems has become an intense field of research, since it allows to access information about electron correlations that is not contained in the average current. The phenomenon of shot noise Kogan96 ; BlanterButtiker99 ; NazarovQN03 , that dominates the current noise at low temperatures, has been investigated theoretically and experimentally in various contexts. Further understanding in electron transport can be gained from the study of higher moments Reulet03 which can be conveniently extracted from the study of full counting statistics (FCS) Levitov93-96 ; Nazarov99 . Up to date FCS has been studied in a variety of cases. Examples include normal-superconductor hybrid structures Muzykantskii94 , superconducting weak links Belzig01SS , tunnel junctions ShelankovRammer03 , chaotic cavities Jong96 , entangled electrons Taddei02-04 and spin-correlated systems DiLorenzo03 . Schemes for an experimental measurement of FCS has been put forward in Refs. FCSmeasure, .
Electron-electron interaction may strongly influence quantum transport. The effect of electron correlation on FCS has been considered so far in the case of a weakly interacting mesoscopic conductor Kindermann03prl , almost open dots Andreev01 ; BagrestNazarov03O ; KindermannTrauzettel04 , charge pumping Andreev01 and charge shuttles Pistolesi04 . The FCS for Coulomb-blockade devices has been analyzed BagretsNazarov03 ; Choi01 ; Makhlin00 in the framework of a Born-Markov master equation approach. The aim of the present paper is to extend this idea to a obtain a general theory of full counting statistics for strongly interacting systems with non-Markovian behavior. In particular, we formulate a perturbative non-Markovian expansion that allows for a systematic study of the relative importance of non-Markovian corrections. We demonstrate for the example of a single-level quantum dot with strong Coulomb interaction that non-Markovian effects become increasingly important for higher moments of the current fluctuations.
Full counting statistics. – Full information about all transport properties of a given system is contained in the probability distribution $`P(N,t)`$ that $`N`$ charges have passed through the system after time $`t`$. This distribution function is related to the cumulant generating function $`S(\chi )`$ by
$$S(\chi )=\mathrm{ln}\left[\underset{N=\mathrm{}}{\overset{\mathrm{}}{}}e^{iN\chi }P(N,t)\right],$$
(1)
where $`\chi `$ is the counting field. All moments of the current can be obtained from the cumulant generating function by performing derivatives with respect to the counting field $`I_n=(i)^n(e^n/t)_\chi ^nS(\chi )|_{\chi =0}`$. The first four moments are the average current, the (zero-frequency) current noise, the skewness, and the kurtosis.
In this work we consider systems with strong local interactions, such as electrons in a quantum dot, that are coupled to a reservoir of noninteracting degrees of freedom. In these situations transport properties can often be described in terms of few (local) degrees of freedom (the charge of the quantum dot in the previous example). It is then convenient to integrate out the noninteracting degrees of freedom to arrive at an effective description of the reduced system only. Let $`𝐩^{\mathrm{in}}`$ be the vector of probabilities to find the system in the corresponding state at the initial time $`t=0`$. The time evolution of the system is described by a generalized master equation,
$$\frac{d}{dt}𝐩(N,t)=\underset{N^{}=\mathrm{}}{\overset{\mathrm{}}{}}\underset{0}{\overset{t}{}}𝑑t^{}𝐖(NN^{},t,t^{})𝐩(N^{},t^{}),$$
(2)
where $`𝐩(N,t)`$ is the vector of dot occupation probabilities under the condition that $`N`$ electrons have passed the system. The cumulant generating function is given by Eq. (1) where $`P(N,t)=𝐞^T𝐩(N,t)`$ with $`𝐞^T=(1,1,\mathrm{},1)`$. The matrix $`𝐖(NN^{},t,t^{})`$ describes transitions during which $`NN^{}`$ electrons are transferred. The counting field $`\chi `$ is introduced by Fourier transforming the master equation $`𝐩(\chi ,t)=_N\mathrm{exp}(iN\chi )𝐩(N,t)`$ and $`𝐖(\chi ,t,t^{})=_N\mathrm{exp}(iN\chi )𝐖(N,t,t^{})`$.
In general, the kernels $`𝐖(\chi ,t,t^{})`$ are nonlocal in time as a consequence of having integrated out the reservoir degrees of freedom. We consider the case in which there is no explicit time dependence of the systems parameters, so that $`𝐖(\chi ,tt^{})`$ can be Laplace transformed, $`𝐖(\chi ,z)=_0^{\mathrm{}}𝑑t\mathrm{exp}(zt)𝐖(\chi ,t)`$. Usually, the dynamics of the system is characterized by a time scale of the individual transitions. We assume that $`𝐖(\chi ,t)`$ decays faster than any power of $`t`$, such that the derivatives $`_z^n𝐖(\chi ,z)`$ exist for all $`n`$.
In the Markovian limit, $`𝐖(\chi ,t)\delta (t)`$, the system at time $`t`$ depends only on the state of the system at the same time $`t`$. The master equation Eq. (2) is, then, solved by $`𝐩(\chi ,t)=\mathrm{exp}[𝐖(\chi ,z=0)t]𝐩^{\mathrm{in}}`$, i.e., only the $`z=0`$ component of the transitions $`𝐖(\chi ,z)`$ are taken into account.
Our goal is to describe stationary transport properties in the presence of a memory of the system, described by the full $`z`$-dependence of $`𝐖(\chi ,z)`$. To solve the master equation without making use of the Markovian approximation, we switch to the Laplace representation,
$$z𝐩(\chi ,z)𝐩^{\mathrm{in}}=𝐖(\chi ,z)𝐩(\chi ,z),$$
(3)
which is solved by $`𝐩(\chi ,z)=_{n=0}^{\mathrm{}}[𝐖(\chi ,z)]^n/z^{n+1}𝐩^{\mathrm{in}}`$. By assuming that the kernel $`𝐖(\chi ,t)`$ decays in time faster than any power law, we can define the Taylor series $`[𝐖(\chi ,z)]^n=_{m=0}^{\mathrm{}}_z^m[𝐖(\chi ,z)]^n|_{z=0}z^m/m!`$ and substitute it in the previous solution for $`𝐩(\chi ,z)`$. The long-time behavior of $`𝐩(\chi ,z)`$ is determined by its poles in $`z`$. Performing the inverse Laplace transformation we get
$$𝐩(\chi ,t)=\underset{n=0}{\overset{\mathrm{}}{}}\frac{_z^n\left([𝐖(\chi ,z)]^n\mathrm{e}^{𝐖(\chi ,z)t}\right)}{n!}|_{z=0^+}𝐩^{\mathrm{in}},$$
(4)
for large $`t`$. To proceed, we perform a spectral decomposition of the matrix $`𝐖(\chi ,z)`$. For physical reasonable systems, all eigenvalues have a negative real part. As a consequence of the exponential function in Eq. (4), the long-time behavior will be dominated by the eigenvalue $`\lambda (\chi ,z)`$ with the smallest absolute value of the real part. Let $`𝐪_0`$ and $`𝐩_0`$ be the corresponding left and right eigenvectors, $`𝐪_0^T𝐖(\chi ,z)=\lambda (\chi ,z)𝐪_0^T`$, and $`𝐖(\chi ,z)𝐩_0=\lambda (\chi ,z)𝐩_0`$. Unitarity in the absence of counting fields implies $`\lambda (0,z)=0`$ for all $`z`$.
The cumulant generating function becomes $`S(\chi )=\mathrm{ln}\left[_{n=0}^{\mathrm{}}\frac{1}{n!}_z^n\left(\lambda ^n\mathrm{e}^{\lambda t+\mu }\right)\right]_{z=0^+},`$ with $`\mu (\chi ,z)=\mathrm{ln}[(𝐞^T𝐩_0)(𝐪_0^T𝐩_{\mathrm{in}})]`$. By performing the time derivative and making use of the relation $`_{n=0}^{\mathrm{}}\frac{1}{n!}^n\left(ab^{n+1}\right)/_{n=0}^{\mathrm{}}\frac{1}{n!}^n\left(ab^n\right)=_{n=0}^{\mathrm{}}\frac{1}{n!}^n\left(b^{n+1}\right)/_{n=0}^{\mathrm{}}\frac{1}{n!}^n\left(b^n\right)`$ that holds for arbitrary functions $`a`$ and $`b`$, we arrive at the central result of this Letter,
$$S(\chi )=\frac{\underset{n=0}{\overset{\mathrm{}}{}}\frac{1}{n!}_z^n\left[\lambda ^{n+1}(\chi ,z)\right]}{_{n=0}^{\mathrm{}}\frac{1}{n!}_z^n\left[\lambda ^n(\chi ,z)\right]}|_{z=0^+}t.$$
(5)
The cumulant generating function depends only on the eigenvalue $`\lambda (\chi ,z)`$. This result can be used as a starting point for a non-Markovian expansion, $`S(\chi )=_{n=0}^{\mathrm{}}S_n(\chi )`$ where $`S_n(\chi )`$ contains $`n`$ derivatives with respect to $`z`$ applied to $`n+1`$ factors of $`\lambda `$. While $`S_0(\chi )`$ describes the Markovian limit, $`S_n(\chi )`$ is the $`n`$-th non-Markovian correction.
Perturbative non-Markovian expansion. – For many systems there is a small parameter which allows for a perturbative analysis of all transport properties. In the examples to be discussed below this will be the tunnel-coupling strength between the leads and the interacting region (quantum dot or metallic island). Then, $`\lambda (\chi ,z)=_{i=1}^{\mathrm{}}\lambda ^{(i)}(\chi ,z)`$, where the superscript $`(i)`$ indicates the order in the small parameter. The lowest-order transport properties are derived from the lowest-order cumulant generating function $`S^{(1)}(\chi )=t\lambda ^{(1)}(\chi ,z)|_{z=0^+}`$, as found in Ref. BagretsNazarov03 . Non-Markovian corrections, signaled by derivatives $`_z^k\lambda (\chi ,z)`$, do not enter in this limit. The highest derivative that enters in the evaluation of the $`n`$-th moment in $`m`$-th order perturbation theory is $`_z^k\lambda (\chi ,z)|_{z=0}`$ with $`k=\mathrm{min}\{n,m\}1`$. As a consequence, non-Markovian behavior is probed only in the second or higher moment combined with second or higher order in perturbation theory. The second-order contribution for example reads $`S^{(2)}(\chi )=t[\lambda ^{(2)}(\chi ,z)+\lambda ^{(1)}(\chi ,z)_z\lambda ^{(1)}(\chi ,z)]_{z=0^+}`$. The appearance of these derivatives in the noise of second-order transport through quantum dots has been also found in Ref. Thielmann04, . In the remaining part of this Letter, we illustrate our approach with two examples. We calculate the cumulant generating function for second-order transport through a single-level quantum dot and through a metallic single-electron transistor in the presence of strong Coulomb interaction.
Single level QD.– The single-level quantum dot is described by the Hamiltonian, $`H=H_\mathrm{L}+H_\mathrm{R}+H_\mathrm{D}+H_\mathrm{T}`$. The electrons in the noninteracting left and right leads are represented by $`H_\mathrm{L}`$ and $`H_\mathrm{R}`$, respectively, $`H_\mathrm{D}=ϵ_\sigma c_\sigma ^{}c_\sigma +Un_{}n_{}`$ describes the dot with level energy $`ϵ`$ and charging energy $`U`$ for double occupation. Tunneling is modeled by $`H_{\mathrm{T},r}=_\sigma t_ra_{rk\sigma }^{}c_\sigma +\text{h.c.}`$ with $`r=\mathrm{R},\mathrm{L}`$, where we assume the tunnel matrix element $`t_r`$ to be independent of momentum $`k`$ and spin $`\sigma `$. The tunnel-coupling strength is characterized by the intrinsic linewidth $`\mathrm{\Gamma }=\mathrm{\Gamma }_\mathrm{L}+\mathrm{\Gamma }_\mathrm{R}`$ with $`\mathrm{\Gamma }_r=2\pi \rho _r|t_r|^2`$ where $`\rho _r`$ is the density of states in the leads. An asymmetry of the tunnel couplings is parametrised by $`\gamma =4\mathrm{\Gamma }_\mathrm{L}\mathrm{\Gamma }_\mathrm{R}/\mathrm{\Gamma }^2`$.
To derive the kernels $`𝐖`$ of the generalized master equation, we make use of a diagrammatic real-time technique konigRTT96 for the time evolution of the reduced density matrix formulated on a Keldysh contour. We introduce counting fields $`\chi _r`$ for tunneling through barrier $`r`$ into lead $`r`$ by the replacement $`t_rt_r\mathrm{exp}(i\chi _r)`$ for tunnel vertices on the upper and $`t_rt_r\mathrm{exp}(i\chi _r)`$ on the lower branch of the Keldysh contour with $`\chi _\mathrm{L}=\chi _\mathrm{R}=\chi /2`$.
We consider the limit $`U\mathrm{}`$, in which double occupancy of the dot is prohibited rem , and obtain
$$S^{(1)}(\chi )=\frac{t\mathrm{\Gamma }\stackrel{~}{f}(ϵ)}{2\mathrm{}}\left[1\sqrt{1+\frac{2\gamma _kf_{(k)}(ϵ)(\mathrm{e}^{ik\chi }1)}{[\stackrel{~}{f}(ϵ)]^2}}\right]$$
(6)
with $`f_{(+)}(\omega )=[1f_\mathrm{L}(\omega )]f_\mathrm{R}(\omega )`$, $`f_{()}(\omega )=f_\mathrm{L}(\omega )[1f_\mathrm{R}(\omega )]`$, and $`\stackrel{~}{f}(\omega )=_r\mathrm{\Gamma }_r[1+f_r(\omega )]/\mathrm{\Gamma }`$, where $`f_r(\omega )`$ is the Fermi function for lead $`r`$. This result was previously obtained in BagretsNazarov03 . The second-order contribution, $`S^{(2)}(\chi )=S_{\mathrm{cot}}^{(2)}(\chi )+S_{\mathrm{ren}}^{(2)}(\chi )`$, consists of two terms. The first one,
$$S_{\mathrm{cot}}^{(2)}(\chi )=\frac{t\gamma \mathrm{\Gamma }^2}{4\pi \mathrm{}}\underset{k=\pm }{}(\mathrm{e}^{ik\chi }1)𝑑\omega f_{(k)}(\omega )R(\omega ϵ),$$
(7)
with $`R(\omega )=\mathrm{Re}[1/(\omega +i0^+)^2]`$, describes cotunneling processes cotunneling , and is in agreement with previous work about noise in cotunneling regime SukhorukovBukardLoss . The counting-field dependence corresponds to a bidirectional Poisson statistics of a single barrier where the transition rates are substituted by the cotunneling rates of the quantum dot. However, $`S^{(2)}(\chi )`$ contains a second contribution,
$$S_{\mathrm{ren}}^{(2)}(\chi )=_ϵ\left[S^{(1)}(\chi )\mathrm{Re}[\sigma (ϵ)]\right]$$
(8)
with $`\sigma (ϵ)=_r(\mathrm{\Gamma }_r/2\pi )𝑑\omega f_r(\omega )/(\omega ϵ+i0^+)`$. In the previous formulas an high energy cut-off $`E_\mathrm{c}`$, of the order of charging energy, has to be introduced in order to cure spurious divergences related to the fact that we restricted the charge states to 0,1 SET . The contribution $`S_{\mathrm{ren}}^{(2)}(\chi )`$ is also of second order in the tunnel-coupling strength but obeys the same statistics as the first-order (sequential-tunneling) result, Eq. (6). This suggests that there are two different types of second-order contributions to transport. In addition to the usual cotunneling processes, there are corrections to sequential tunneling due to quantum-fluctuation induced renormalization of the system parameters. From the form of Eq. (8) we deduce a renormalization of level position and coupling strength given by $`\stackrel{~}{ϵ}=ϵ+\mathrm{Re}\sigma (ϵ)`$ and $`\stackrel{~}{\mathrm{\Gamma }}=\mathrm{\Gamma }[1+_ϵ\mathrm{Re}\sigma (ϵ)]`$.
In Fig. 1 we plot the first four moments (current, noise, skewness, and kurtosis) as a function of level position $`ϵ`$. The solid lines represent the full first- plus second-order result, as compared to the first-order contribution (dashed line).
The relative importance of the non-Markovian contributions is illustrated in Fig. 2. While for the current only Markovian contributions enter (see discussion above), non-Markovian corrections become increasingly important for higher moments.
Metallic QD.– A similar analysis can be performed for a metallic single-electron transistor, which accommodates a continuum of states on the dot and includes a large number of transverse channels. Following the notation of Refs. SET, , we characterize the tunnel-coupling strength by the dimensionless conductance $`\alpha _0^r=h/(4\pi e^2R_r)`$ where $`R_r`$ is the resistance of barrier $`r=\mathrm{L},\mathrm{R}`$. We concentrate on the low-temperature regime, in which only two charge states of the metallic island have to be taken into account. This requires, again, the introduction of an high-energy cut-off to regularize the integrals. The difference of the charging energy between them is denoted by $`\mathrm{\Delta }`$. We obtain for the first order of cumulant generating function
$$S^{(1)}(\chi )=\frac{t\pi \alpha (\mathrm{\Delta })}{\mathrm{}}\left[1\sqrt{1+4\underset{k=\pm }{}\frac{\alpha _{(k)}(\mathrm{\Delta })(\mathrm{e}^{ik\chi }1)}{\alpha (\mathrm{\Delta })}}\right]$$
(9)
where we have used the definitions $`\alpha (\omega )=_{r=\mathrm{L},\mathrm{R}}\alpha _r(\omega )`$ and $`\alpha _{(\pm )}(\omega )=\alpha _\mathrm{L}^\pm (\omega )\alpha _\mathrm{R}^{}(\omega )`$ with $`\alpha _r^\pm (\omega )=\alpha _r(\omega )f_r^\pm (\omega )`$ where $`f_r^\pm (\omega )`$ is the Fermi functions of lead $`r`$ and $`\alpha _r(\omega )=\alpha _0^r\mathrm{coth}[\beta (\omega \mu _r)/2]`$. The second-order term is given by $`S^{(2)}(\chi )=S_{\mathrm{cot}}^{(2)}(\chi )+S_{\mathrm{ren}}^{(2)}(\chi )`$, as the case of the single-level quantum dot. Again, there is a contribution due to cotunneling,
$$S_{\mathrm{cot}}^{(2)}(\chi )=\frac{2\pi t}{\mathrm{}}\underset{k=\pm }{}(\mathrm{e}^{ik\chi }1)𝑑\omega \alpha _{(k)}(\omega )R(\omega \mathrm{\Delta }),$$
(10)
and a term
$$S_{\mathrm{ren}}^{(2)}(\chi )=_\mathrm{\Delta }\left[S^{(1)}(\chi )\mathrm{Re}[\sigma (\mathrm{\Delta })]\right]$$
(11)
associated with sequential-tunneling processes with renormalized system parameters, where $`\sigma (\mathrm{\Delta })=_r𝑑\omega \alpha _r(\omega )/(\omega \mathrm{\Delta }+i0^+)`$. This interpretation of the different types of second-order contributions is consistent with the analysis of the second-order current koenig\_cot and of the FCS within a drone Majorana fermion representation Utsumi05 .
Conclusions. – We present a theory of FCS for interacting systems with non-Markovian dynamics. A general expression for the cumulant generating function is derived that provides the starting point for a perturbative non-Markovian expansion. As examples we study transport through a single-level quantum dot and a metallic single-electron transistor to second order in the tunnel-coupling strength. From our formulation we could identify two different types of contributions to second-order transport, namely cotunneling and corrections to sequential tunneling due to renormalization of the system parameters. Furthermore, we demonstrate the increasing importance of non-Markovian effects for higher moments and higher orders in the tunnel-coupling strength.
We thank D. Bagrets, W. Belzig, Y. Gefen, M. Hettler, G. Johansson, F. Plastina, A. Romito, M. Sassetti, and A. Thielmann, for useful discussions. Financial support from and DFG via SFB 491 and GRK 726, EU-RTN-RTNNANO, EU-RTN-Spintronics, MIUR-Firb is gratefully acknowledged. |
warning/0507/quant-ph0507134.html | ar5iv | text | # Standard forms of noisy quantum operations via depolarization
## I Introduction
Quantum systems evolve via unitary operations $`U(t)`$, as they are governed by the Schrödinger equation. This also holds for composite systems, e.g. a small quantum system $`S`$ which is surrounded by some environment $`E`$, where the evolution of the total system is described by a unitary operation $`U_{SE}(t)`$. The dynamics of the system $`S`$ alone —which can be obtained by tracing out the (uncontrollable) degrees of freedom of the environment— is in general no longer unitary. In fact, the system interacts with degrees of freedom of the environment, leading to entanglement between system $`S`$ and environment $`E`$ reflected in $`U_{SE}(t)U_S(t)U_E(t)`$. The system–environment interaction leads to decoherence and the dynamics of the system can be described either by a (time dependent) completely positive map (CPM) $`(t)`$ or —under certain assumptions on the nature of interaction— by a master equation of Lindblad form Lidar01 . From the perspective of quantum information processing, such an interaction with environmental degrees of freedom is undesirable and leads to errors and noise in the system. As discussed below, an arbitrary noise process acting on a $`d`$–dimensional system $`S`$ is, at fixed time $`t_0`$, determined by $`O(d^4)`$ real parameters. Even for small system sizes, e.g. when $`S`$ consists of three qubits (i.e. $`d=8`$), this leads to a huge number of independent parameters (e.g. around 4000 for the three–qubit system), which makes an analytical treatment of the influence of such general noise processes on the properties of the system rather difficult. This is particularly hindering when considering either large systems or sequences of several noisy evolutions (or gates), as is e.g. required in the analysis of quantum circuits or processes such as entanglement purification.
When considering the influence of noise in quantum information processing, one hence often restricts the analysis to certain (ad hoc) noise models, such as Pauli channels or depolarizing (white) noise models. This is usually the case in the analysis of entanglement purification protocols in the presence of noisy operations as well as in the theory of fault–tolerant quantum computation. On the other hand, having a specific physical set–up in mind, one can sometimes justify these (or other) noise models by a microscopic description of the underlying system–environment interactions, where only the dominant part of noisy interactions is considered. However, when considering (abstract) quantum processes that deal with the manipulation of quantum information, one does not want to restrict oneself to a specific physical set–up, but rather would like to keep the analysis at an abstract level and as general as possible. To this aim, it would be very useful to justify the usage of simple noise models in a general context, or to provide a method to bring any noise process to a simple standard form described by a few parameters.
In this article, we provide a method which allows one to achieve this aim. We show that indeed any noise process can be brought to a simple standard form by means of depolarization. That is, by applying appropriate (local) unitary control operations on a system before and after the noisy evolution in a correlated way, one can depolarize the noise process. This depolarization process of the corresponding CPM $``$ or the Liouvillian $``$ can be viewed as an analogue of the depolarization of mixed states. For bipartite states, for instance, it was shown that one can bring any mixed state $`\rho `$ of two $`d`$–level systems to a standard form specified by a single parameter using an appropriate (random) sequence of local unitary operations. Depolarization takes place in such a way that the fidelity of the state, i.e. the overlap with a maximally entangled state $`|\mathrm{\Phi }=1/\sqrt{d}_{k=1}^d|k|k`$, remains invariant. The resulting states (isotropic states) are equivalent to Werner states and are given by $`\rho (x)=x|\mathrm{\Phi }\mathrm{\Phi }|+(1x)\frac{1}{d^2}\mathrm{𝟏}`$. Werner states played an important role in the investigation of the relations between entanglement and local hidden variable theories We89 , as well as in the development of entanglement purification protocols, schemes which are becoming increasingly important since it was realized that entanglement can serve as a valuable resource not only in quantum communication but also in quantum information processing. The development of these important issues was triggered by the simplified description of Werner states (still covering essential entanglement properties), and allowed at the same time to obtain necessary or sufficient conditions for separability or distillability for arbitrary bipartite states.
We are confident that also the depolarization of noisy evolutions will prove to be a fruitful tool in the analysis of noise processes. In direct analogy to the depolarization of states, the depolarization of the noise maps takes place in such a way that the fidelity of the ideal (unitary or Hamiltonian) part of the evolution is not altered. In fact, we make use of the isomorphism between completely positive maps and mixed states Jam72 ; Ci00 and connect the problem of depolarization of maps to the depolarization of the corresponding states, while respecting certain locality restrictions. For decoherence processes (e.g. storage errors of a system due to its interaction with the local environment, or errors resulting due to sending a system through a noisy quantum channel), where the ideal operation is the identity or, equivalently, the Hamiltonian part in the corresponding master equation is zero, we find that one can depolarize the corresponding map or master equation to a standard form which is described by correlated and uncorrelated white noise processes. In the case of two $`d`$–level systems, for instance, the corresponding depolarized map is described by three real parameters (the weights of ideal operation, single particle (uncorrelated) white noise processes and two–particle (correlated) white noise) as compared to $`O(d^8)`$ parameters of an arbitrary map. We also consider noisy interactions (i.e. the ideal evolution is given by some non–local unitary operation $`U`$ or some non–trivial system interaction Hamiltonian $`H`$), where we concentrate on two–system interactions. We find that for certain unitary operations, in particular for SWAP gates, CNOT gates, as well as phase gates with arbitrary phase $`\alpha `$, a depolarization is still possible. The required number of parameters to describe the (depolarized) noise process depends on the unitary operation (interaction) that has to be kept invariant, and is given by 17 in the case of arbitrary phase gate, 8 in the case of the CNOT gate and 3 for the SWAP gate.
Knowledge of the exact form of the noise process (which can e.g. be acquired by means of gate tomography) or additional control of interactions (e.g. the ability to switch a noisy interaction on and off at will) allows to further tailor the noise process. In this case, the fidelity of the ideal operation is decreased by a certain (small) amount, while the description of the noise process is simplified and the number of relevant parameters is further reduced. In many relevant cases (e.g. noisy SWAP or CNOT gate, switchable noisy phase gate), one finds that one can indeed simplify the noise process in such a way that the corresponding CPM is described by a single parameter and the noise process corresponds to correlated white noise. The total amount of noise is —in the worst case— increased by about an order of magnitude, as weight of the ideal operation is transferred to the noise part in an appropriate way to achieve this further simplification.
While in the case of maps a depolarization with a significant reduction of the associated parameters is only possible for certain unitary operations, one finds that, in the case of master equations, sequential application of fast intermediate local unitary control operations allows one to depolarize any master equation (of two systems) to a standard form described by at most 17 parameters. In return this depolarization protocol generally increases the noise level of the decoherence process. Under certain circumstances, one may even achieve a standard form described by a single parameter for arbitrary two–qubit interactions by accepting a further increase of the noise level.
Such standard forms for noisy evolutions may have wide spread applications in the analysis of quantum information processes under realistic conditions. For instance, our approach allows one to obtain lower bounds on the capacity of arbitrary multipartite quantum channels by considering the corresponding depolarized channels. The depolarized noise process also gives rise to a simplified process tomography. The tomography has to reveal fewer parameters (the parameters characterizing the standard form) than those necessary to describe the original decoherence process or the noisy gate. This can lead to a significant reduction of the experimental effort to sufficiently characterize the influence of noise in a given set–up. Also processes involving sequences of noisy gates, e.g. entanglement purification or some quantum circuits, can be analyzed by considering the standard forms for the corresponding gates. The resulting threshold values do no longer refer to specific error models but are valid in general, as any noise process can be brought to the corresponding standard form. When applying this method to derive generally valid error thresholds, e.g. in the context of fault tolerant quantum computation, some care is required. An implicit assumption in order to allow the application of such a local depolarization procedure is that the corresponding (local) control operations can in fact be (noiselessly) applied to the system. When dealing for instance with decoherence processes due to channel noise or local interaction of the system with some environment, such an assumption is perfectly reasonable. Also for two–system interaction gates (such as the CNOT), one may assume that local, single system gates are noiseless (or introduce a negligible amount of noise as compared to the two–system gate). However, when dealing also with noisy single system operations (as is e.g. required in the analysis of fault tolerant quantum computation), it is no longer straightforward to apply our results. One might argue that for sequences of gates the required (random) operations for depolarization can be incorporated in previous/subsequent noisy gates, although it is not entirely clear whether this argument justifies the assumption that any gate within a quantum circuit is already of standard form. However, whenever local, single system control operations can be assumed to be noiseless, our results are applicable and one can indeed bring an arbitrary (non–local) noise process to a simplified standard form.
This paper is organized as follows: In Sec. II we review basic properties of the Jamiołkowski isomorphism between completely positive maps and states, which will be the main tool for the derivation of standard forms for CPM in the following sections. We will then apply this isomorphism in Sec. III in order to provide standard forms for an arbitrary decoherence process in the case where the corresponding control operation to achieve this standard form does not have to obey any locality requirements. In Sec. IV we derive standard forms for maps describing arbitrary decoherence processes and some noisy unitary operations. These standard forms are achieved by control operations that are local with respect to (w.r.t) some given partitioning. In Sec. V we suggest a protocol to bring an arbitrary noisy evolution described by a master equation into some standard form, for which the accompanying noise process is described by a reduced number of parameters. Finally we summarize our results in Sec. VI. Some technicalities can be found in the appendices.
## II The Jamiołkowski correspondence between completely positive maps and states
In this section we review some properties of the Jamiołkowski isomorphism Jam72 between completely positive maps (CPM) and states. In Sec. II.1 we state and discuss this isomorphism first on an abstract level as a correspondence between matrices and the endomorphisms of the corresponding matrix algebras. We will then restrict this general isomorphism to the physical setting of quantum states and quantum operations in Sec. II.2, where the isomorphism has a clear interpretation in terms of a teleportation protocol. In Sec. II.3 we review some applications of the isomorphism GLN . Known distance measures for quantum states can be used to provide distance measures for (trace preserving) CPM, which we will use in the following. Finally we extend the Jamiołkowski isomorphism in Sec. II.4 to the multi–party setting and discuss some implications for the entanglement capabilities of CPMs. For sake of completeness the reader can find a reviewCi00 ; Ar03 about the relation between the spectral decomposition of states and the Kraus representation for CPM in Appendix A and about the relation between the purification for quantum states and quantum operation in Appendix B. Note that the main properties of the isomorphism are stated in the form of short propositions with a consecutive numbering (No. 113) , that is continued in the Appendix.
### II.1 The Isomorphism in the general setting
Let $`𝐇_A`$ and $`𝐇_A^{}`$ be two Hilbert spaces of finite dimensions $`d_A=\text{dim}_{}(𝐇_A)`$ and $`d_A^{}=\text{dim}_{}(𝐇_A^{})`$. With $`_A=(𝐇_A)`$ and $`_A^{}=(𝐇_A^{})`$ we denote the corresponding matrix algebras over $`𝐇_A`$ and $`𝐇_A^{}`$ respectively, which contain the set of physical states (density matrices) $`𝒟_A_A`$ and $`𝒟_A^{}_A^{}`$ as (proper) convex subsets. Similarly, we will write $`(𝐇_A,𝐇_A^{})`$ for the algebra of $`d_A^{}\times d_A`$ matrices representing the linear maps from $`𝐇_A`$ to $`𝐇_A^{}`$. Moreover let $`\text{End}\left(_A_A^{}\right)`$ be the set of linear maps (endomorphisms) between the algebras $`_A`$ and $`_A^{}`$ , which contain the physical operations $`\text{CPM}\left(𝒟_A𝒟_A^{}\right)`$ between the two quantum systems, i.e. completely positive maps (CPM), as a proper subset. In the following we will frequently consider a copy $`\overline{A}`$ of system $`A`$ and use – after identifying and fixing a basis in $`𝐇_A`$ and $`𝐇_A^{}`$ – the maximally entangled state
$`|\mathrm{\Phi }={\displaystyle \frac{1}{\sqrt{d_A}}}{\displaystyle \underset{i=1}{\overset{d_A}{}}}|i^{\overline{A}}|i^A,P_\mathrm{\Phi }=|\mathrm{\Phi }\mathrm{\Phi }|`$ (1)
on the composite system $`𝐇_{\overline{A}}𝐇_A`$.
The Jamiołkowski Isomorphism
The map $`𝒥:\left(𝐇_A^{}𝐇_A\right)\text{End}\left(_A_A^{}\right)`$, that maps a matrix $`E`$ of the matrix algebra over the composite system $`𝐇_A^{}𝐇_A`$ to the linear map $``$ given by
$`(M):=d_A^2\text{tr}_{A\overline{A}}\left[E^{A^{}A}P_\mathrm{\Phi }^{A\overline{A}}M^{\overline{A}}\right]`$ (2)
(for any $`M_A_{\overline{A}}`$) is an isomorphism Jam72 ; Ci00 , i.e. it is linear and bijective. For the inverse of $`𝒥`$ the matrix $`E\left(𝐇_A^{}𝐇_A\right)`$, that corresponds to the linear map $``$, is given by
$`E:=^{\overline{A}}\text{Id}^A\left(P_\mathrm{\Phi }^{\overline{A}A}\right).`$ (3)
If the matrix $`E`$ and map $``$ in correspondence are decomposed with respect to the chosen basis $`|j`$ ($`j_{d_A}`$) on $`𝐇_A`$ (or $`𝐇_{\overline{A}}`$) and $`|i`$ ($`i_{d_A^{}}`$) on $`𝐇_A^{}`$
$`E^{A^{}A}`$ $`=`$ $`{\displaystyle \underset{\genfrac{}{}{0.0pt}{}{i,k_{d_A^{}}}{j,l_{d_A}}}{}}E_{ij|kl}|i^A^{}k||j^Al|`$ (4)
$`(M)`$ $`=`$ $`{\displaystyle \underset{\genfrac{}{}{0.0pt}{}{i,k_{d_A^{}}}{j,l_{d_A}}}{}}_{ik|jl}j|M|l|i^A^{}k|`$ (5)
the Jamiołkowski isomorphism simply is Ar03
$`_{ik|jl}=d_AE_{ij|kl}.`$ (6)
From this relation between the coefficients the bijectivity immediately follows from the linearity of $`𝒥`$ together with the fact, that $`\left(𝐇_A^{}𝐇_A\right)`$ and $`\text{End}\left(_A_A^{}\right)`$ both are linear spaces of dimension $`d_A\times d_A^{}`$. The above result therefore can be shown by deriving relation (6) separately from Eq. (2) and from Eq. (3) using the fact that
$`d_A\text{tr}_{\overline{A}}\left[P_\mathrm{\Phi }^{A\overline{A}}M^{\overline{A}}\right]=(M^A)^t`$ (7)
holds for any $`M_A`$.
The isomorphism also turns out to be an isometry Ar03 : If one uses the inner products
$`E,F`$ $`:=`$ $`\text{tr}E^{}F`$ (8)
$`,`$ $`:=`$ $`{\displaystyle \underset{j,l_{d_A}}{}}\text{tr}\left[\left((|jl|)\right)^{}(|jl|)\right]`$ (9)
for $`E,F\left(𝐇_A^{}𝐇_A\right)`$ and the corresponding maps $`,\text{End}\left(_A_A^{}\right)`$, it can readily be seen that
$`E,F=,`$ (10)
holds. Note that the corresponding norms $`M`$ on $`_A`$ ($`_A^{}`$) and $``$ on $`\text{End}\left(_A_A^{}\right)`$ are Euclidean ($`l_2`$ norms).
Eq. (6) shows, that the isomorphy does not extend to the respective compositions in $`\left(𝐇_A^{}𝐇_A\right)`$ and $`\text{End}\left(_A_A^{}\right)`$. For example, the composition $``$ of to maps would correspond to a matrix composition law $`(E\mathrm{}F)_{ij|kl}=_{mn}E_{im|kn}F_{mj|nl}`$ which differs from the usual matrix multiplication. Nevertheless it also provides $`\left(𝐇_A^{}𝐇_A\right)`$ with a semi-group structure Ar03 .
The effect of matrix multiplication by local local matrices is given by the following formula: Given a map $``$ and its corresponding matrix $`E`$ and the matrices $`B_1,C_1_A^{}`$, $`B_2,C_2_A`$, then the transformed matrix
$`E^{}=B_1^A^{}B_2^AE^{}C_1^A^{}C_2^A`$ (11)
corresponds to a map $`^{}`$ with
$`^{}(M):=B_1\left(B_2^TMC_2^T\right)C_1.`$ (12)
### II.2 The isomorphism for quantum states and quantum operations
Under which conditions on the matrix $`E\left(𝐇_A^{}𝐇_A\right)`$ does the linear map $`\text{End}\left(_A_A^{}\right)`$ correspond to a physical operation, i.e. is a (trace-preserving) CPM ? This can be answered by the following results Ar03 :
* $``$ is Hermiticity preservingNotation , iff $`E`$ is Hermitian.
* $``$ is positivity preserving Notation , iff $`E`$ is Hermitian and for all separable states $`F𝒟\left(𝐇_A^{}𝐇_A\right)`$ $`\text{tr}\left(EF\right)0`$ holds.
* $``$ is completely positive Notation , iff $`E`$ is positive.
* $``$ is a trace-preserving CPM Notation , iff $`E`$ is positive and $`\text{tr}_A^{}E^{A^{}A}=\frac{1}{d_A}\mathrm{𝟏}_A`$ holds.
These results imply that the Jamiołkowski isomorphism can be restricted to $`𝒥:𝒟\left(𝐇_A^{}𝐇_A\right)\text{End}\left(𝒟_A𝒟_A^{}\right)`$ yielding a correspondence between trace-preserving CPM and states on the composite system of $`A`$ and $`A^{}`$. In this case the isomorphism can be given a natural interpretation in terms of a teleportation protocol (without classical communication): In order to obtain the state $`E`$ corresponding to a CPM $``$ according to Eq. (3) the CPM $``$ simply has to be applied at the system $`\overline{A}`$ of the composite system in the maximally entangled state $`|\mathrm{\Phi }`$ (see Fig. 1).
Conversely, given the state $`E`$, the CPM $``$ can be evaluated for an arbitrary input state $`\rho `$ according to Eq. (2) as follows (see Fig 2). Considering the composite system consisting of parties $`A^{}`$ and $`A`$ in the state $`E`$ together with the input state $`\rho `$ at system $`\overline{A}`$, i.e. the total state $`E^{A^{}A}\rho ^{\overline{A}}`$, the joint system $`A\overline{A}`$ is measured in a Bell basis containing the maximally entangled state $`P_\mathrm{\Phi }^{A\overline{A}}`$. With probability $`\frac{1}{d_A^2}`$ the desired output state $`(\rho )`$ is then obtained at system $`A^{}`$.
According to Eq. (12) any operation $`𝒩`$ on $`E`$, which is separable w.r.t. the partitioning $`(A^{},A)`$, i.e.
$`E^{}={\displaystyle \underset{j}{}}B_j^A^{}C_j^AE(B_j^A^{}C_j^A)^{},`$ (13)
translates to a probabilistic application of combined operations before and after the CPM $``$:
$`^{}(M):={\displaystyle \underset{j}{}}B_j\left(C_j^TMC_j^{}\right)B_j^{}.`$ (14)
In particular, the application of local unitaries or measurements to $`E`$ on party $`A`$ \[$`A^{}`$\] corresponds to the application of local unitaries or measurements before \[after\] the CPM $``$. On the other hand, not all separable operations can be implemented by local quantum operations and classical communication (LOCC) Be98 . Since only the measurement results before the CPM $``$ can influence operations performed afterward, we have to restrict the separable operations on side $`A`$ and $`A^{}`$ even to be local quantum operations and one-way classical communication (1-LOCC) from party $`A`$ to party $`A^{}`$. The separable operations in question thus correspond to the state
$`E^{}={\displaystyle \underset{ij}{}}B_{ij}^A^{}C_j^AE(B_{ij}^A^{}C_j^A)^{},`$
where
* $`_jC_j^{}C_j=\mathrm{𝟏}`$, i.e. the quantum operation $`𝒞(\rho )=_jC_j\rho C_j^{}`$ on party $`A`$ is a trace preserving CPM;
* $`𝒞`$ is bi-stochastic $`𝒞(\mathrm{𝟏})=\mathrm{𝟏}`$ and hence the corresponding CPM $`\stackrel{~}{𝒞}(\rho )=_jC_j^T\rho C_j^{}`$ before the application of $``$ is also a trace preserving CPM, i.e. $`_jC_j^{}C_j^T=\mathrm{𝟏}`$;
* for each measurement outcome $`j`$ on party $`A`$ \[before the application of $``$\] the corresponding operation $`_j(\rho )=_iB_{ij}\rho B_{ij}^{}`$, that is performed on party $`A^{}`$ according to the classical information sent by $`A`$, is a trace preserving CPM, i.e. $`_iB_{ij}^{}B_{ij}=\mathrm{𝟏}`$.
Condition (i) and (iii) specify the notion of a general 1-LOCC protocol, that we consider in the following. In many cases such as for local projective measurements or for probabilistic applications of local unitaries, property (ii) follows from (i), but in general (ii) provides an separate condition, which reflects the fact that before $``$ not $`𝒞`$ but $`\stackrel{~}{𝒞}`$ with transposed Kraus operators is applied. To simplify notations we will therefore consider those 1-LOCC protocols, that satisfy all three conditions. The above discussion indicates the two directions, in which one can try to manipulate $``$ with the help of the corresponding state $`E`$:
* If one really has the above teleportation protocol available in practice, any (non-local) operation on $`E`$ can be considered in order to manipulate the corresponding CPM.
* If the isomorphism is only a helpful theoretical tool, then one should only consider 1-LOCC operations on $`E`$ in order to manipulate a given CPM, since these operations can be implemented by a coordinated application of operations before and after the evaluation of the CPM.
We emphasize that only one direction of the isomorphism protocol can be implemented with unit probability of success. This implies that the case (A) corresponds to a probabilistic modification of the CPM $``$ whereas case (B) gives rise to a deterministic manipulation protocol. Case (A) is also equivalent to all protocols, in which one does not only allow arbitrary local operations before and after the application of $``$ (and therefore the use of independent ancillary systems to perform these operations) but also to make an (arbitrary) ancillary system available to store quantum information. This information is obtained during the operations before the CPM $``$ and later used in the operations performed after $``$ ancEquiv .
### II.3 Distance measures for quantum states and quantum operations
In the remainder of this paper we derive standard forms $`^{}`$ for some noisy CPM $``$, that approximates some ideal operation $``$. A reasonable requirement for such a standard form is that it is also a considerably good approximation to the ideal operation. In order to assess this requirement some kind of distance measure between quantum operation is needed. As a first application of the Jamiołkowski correspondence we thus review the derivation GLN of distance measures for quantum operations from those for quantum states.
Concerning the isometry properties discussed in Sec. II.1 note that the Euclidean norm does not provide a proper distance measure $`d`$ for quantum states, since it does not obey the contractivity property, that is
$`d((\rho ),(\sigma ))d(\rho ,\sigma )`$ (15)
for all states $`\rho ,\sigma `$ and trace-preserving quantum operations $``$ Oz00 . This property expresses the physical condition that no quantum process should allow to increase the distinguishability of two quantum states. In the literature (see e.g. GLN ; Ni00 ; Ki02 ) there are mainly two metrics metric considered that also obey the contractivity property, namely:
* trace distance: $`d_1(\rho ,\sigma ):=\frac{1}{2}|\rho \sigma |_{\text{tr}}`$, where $`|M|_{\text{tr}}:=\text{tr}\left(\sqrt{M^{}M}\right)`$ is the trace norm;
* fidelity-based distances, that are monotonically decreasing functions of the fidelity $`F(\rho ,\sigma ):=\text{tr}\left(\sqrt{\sqrt{\rho }\sigma \sqrt{\rho }}\right)^2`$ such as $`d_2(\rho ,\sigma ):=\sqrt{1F(\rho ,\sigma )}`$; note that $`F(|\psi ,\sigma )=\psi |\sigma |\psi `$ if $`\rho =|\psi \psi |`$ is pure.
By again using the Jamiołkowski isomorphism both distance measures for quantum states have a natural counterpart as a distance measure for quantum operation. Given two trace-preserving quantum operations $``$ and $``$ one can define the distance $`\mathrm{\Delta }(,)=d(E,F)`$ as the distance $`d`$ between the corresponding states $`E`$ and $`F`$, which is easily shown to yield a metric $`\mathrm{\Delta }`$ on the set of trace-preserving quantum operations as long as $`d`$ is a metric on the corresponding set of quantum states. Choosing $`d=d_1`$ or $`d=d_2`$ the corresponding distance measures $`\mathrm{\Delta }_1`$ and $`\mathrm{\Delta }_2`$ also have the following two properties, which seem to be reasonable requirements for any distance measure for quantum operations GLN :
* Stability Ki02 : $`\mathrm{\Delta }(\text{Id},\text{Id})=\mathrm{\Delta }(,)`$, i.e. the distance measure of two quantum processes should not depend on whether they are considered to occur in an environment together with some unrelated ancillary quantum system;
* Chaining chainingCond : $`\mathrm{\Delta }(_1_2,_1_2)\mathrm{\Delta }(_1,_1)+\mathrm{\Delta }(_2,_2)`$, i.e. for a composed process, the total error will be less than the sum of the errors in each individual step.
Note that the evaluation of the above distance measures in practice requires some quantum process tomography. Moreover both measures can be shown to have some physical interpretation in the sense of a bound to the average–case–error in function computation and sampling computation GLN . But in the following we will consider a slightly different application. Unfortunately the natural approach for defining error measures for quantum operations by averaging the distances between the output states, i.e. $`\mathrm{\Delta }(,):=𝑑\psi d((\psi ),(\psi ))`$ so far could not be modified in such a way that it would also fulfill the stability property. Nevertheless, for the case that one operation $`=U`$ is a unitary operation, the average fidelity
$`\overline{F}(,U):={\displaystyle 𝑑\psi \psi |U^{}(|\psi \psi |)U|\psi }`$ (16)
has at least a plausible interpretation in terms of the average ’overlap’ between the two outputs $`U|\psi `$ and $`(|\psi \psi |)`$, although it does not even define a metric. It was shown in HoNi that this average fidelity $`\overline{F}(,U)`$ is linearly related to the ’Jamiołkowski’ fidelity $`F(,U)=F(E,|\psi _U)`$ by:
$`\overline{F}(,U)={\displaystyle \frac{F(,U)d+1}{d+1}},`$ (17)
where $`|\psi _U`$ denotes the pure state corresponding to the unitary operation $`U`$.
In the following we will be interested in standard forms for noisy operations $``$, which approximate some ideal operation, that will be either the identity Id or some unitary $`U`$. These standard forms $`^{}`$ are obtained by different protocols, which might introduce additional noise to the operation $``$. Apart from simplicity of the obtained standard form $`^{}`$, it should only differ in the same order of magnitude from the ideal operation $`U`$ or Id as the original imperfect operation $``$. The above argument shows that the ’Jamiołkowski’ fidelity $`F(,\text{Id})`$ or $`F(,U)`$ can in both cases be used to measure this distance: On the one hand the fidelity is related to a decent distance measure for quantum operation by a monotonic decreasing function. On the other hand for our applications the fidelity has a physical interpretation in terms of the average error in approximating an ideal (unitary) quantum operation $`U`$. In the following we therefore try to provide standard forms $`^{}`$ of noisy operations $``$, that have either the same or a slightly decreased fidelity $`F(^{},U)`$ with the ideal operation as the original one ($`F(,U)`$).
### II.4 The Isomorphism in the multi-party setting
The Jamiołkowski isomorphism has a natural extension to multi-party scenarios, which are of special interest in quantum information theory. For this let the system $`A=(A_1,\mathrm{},A_N)`$ consist of $`N`$ parties, each representing Hilbert spaces $`𝐇^{A_i}`$ of different dimensions $`d_{A_i}`$, such that $`𝐇^A=𝐇^{A_1}\mathrm{}𝐇^{A_N}`$ and $`d_A=_{i=1}^Nd_{A_i}`$. In order to keep the argumentation simple we consider only CPM $``$, whose input and output Hilbert spaces are of the same type, i.e. $`𝐇^A^{}=𝐇^{A_1^{}}\mathrm{}𝐇^{A_N^{}}`$ with $`𝐇^{A_i^{}}𝐇^{A_i}`$.
The main point in extending the isomorphism to the multi-party setting is to choose the maximally entangled state $`|\mathrm{\Omega }_{\overline{A}A}`$ to be the tensor product of the respective maximally entangled states
$`|\mathrm{\Phi }^{\overline{A}_iA_i}={\displaystyle \frac{1}{\sqrt{d_{A_i}}}}{\displaystyle \underset{k=1}{\overset{d_{A_i}}{}}}|k^{\overline{A}_i}|k^{A_i}`$ (18)
between the subsystem $`A_i`$ and its copy $`\overline{A}_i`$ at each individual party $`i=1,\mathrm{},N`$. The maximally entangled state $`\omega `$ is therefore
$`P_\mathrm{\Phi }^{\overline{A}A}=P_\mathrm{\Phi }^{\overline{A}_1A_1}\mathrm{}P_\mathrm{\Phi }^{\overline{A}_NA_N}`$ (19)
with $`P_\mathrm{\Phi }^{\overline{A}_iA_i}=|\mathrm{\Phi }^{\overline{A}_iA_i}\mathrm{\Phi }|`$. In this notation the isomorphism will have exactly the same form as stated above with the only difference that the maximally entangled state $`\mathrm{\Phi }`$ used in both directions now also respects the partitioning $`A=(A_1,\mathrm{},A_N)`$:
$`(M)`$ $`:=`$ $`d_A^2\text{tr}_{A\overline{A}}\left[E^{A^{}A}P_\mathrm{\Phi }^{A\overline{A}}M^{\overline{A}}\right]`$ (20)
$`E`$ $`:=`$ $`^{\overline{A}}\text{Id}_A\left(P_\mathrm{\Phi }^{\overline{A}A}\right).`$ (21)
For the interpretation in terms of a teleportation protocol the preparation of the maximally entangled state $`P_\mathrm{\Phi }^{\overline{A}A}`$ and the corresponding Bell measurement can be performed locally at each party separately, since the entanglement present in $`P_\mathrm{\Phi }`$ is only with respect to the systems $`A_i`$ and their copies $`\overline{A}_i`$ but not with respect to the partitioning itself (see Fig. 3 and Fig. 4). For the index notation it is convenient to take the same formula as in Eq. (6)
$`_{\mathrm{𝐢𝐤}|\mathrm{𝐣𝐥}}=d_AE_{\mathrm{𝐢𝐣}|\mathrm{𝐤𝐥}},`$ (22)
but to consider the indices $`𝐢,𝐣,𝐤`$ and $`𝐥`$ as multi-indices, e.g. $`𝐢=(i_1,\mathrm{},i_N)_{d_{A_1}}\times \mathrm{}\times _{d_{A_N}}`$. It turns out, that many of the entangling capabilities of the CPM $``$ are directly related to the entanglement properties of the corresponding state $`E`$ (see Ci00 ; Du0102 ):
* $``$ is separable w.r.t. parties $`A_k`$ and $`A_l`$ (and therefore not capable to create entanglement between them), iff $`E`$ is separable w.r.t. parties $`A_k`$ and $`A_l`$. In particular, we find that the CPM corresponding to the tensor product of states $`EF`$ simply is the tensor product of the corresponding CPMs $``$.
* For the partial transposition $`^{T_{A_k}}`$ w.r.t. party $`A_k`$ we have:
$`\left[\left(M\right)\right]^{T_{A_k^{}}}=^{}\left(M^{T_{A_k}}\right)`$ (23)
with $`E^{}=E^{T_{A_k^{}A_k}}`$.
In particular, $``$ is PPT preserving w.r.t. party $`A_k`$ PPT\_preserving , iff $`E`$ is PPT w.r.t. the joint transposition of $`A_k^{}`$ and $`A_k`$.
* The CPM $``$ can simulate another CPM $``$ under SLOCC slocc , iff the corresponding positive operator $`E`$ can be converted into $`F`$ by means of SLOCC.
* Two CPMs $``$ and $``$ are equivalent under local unitaries (LU), iff the corresponding positive operators $`E`$ and $`F`$ are LU-equivalent w.r.t. the finest partitioning $`(A_1,A_1^{},\mathrm{},A_N,A_N^{})`$ .
* The CPM $``$ can generate a state $`\rho `$ of the composite system $`𝐇^A^{}𝐇^A`$ with non-zero probability of success, iff the corresponding positive operator $`E`$ can be converted into $`\rho `$ by means of SLOCC.
Since the classification of pure states in bipartite and three-qubit systems under SLOCC is known in detail, the results can be transferred to the corresponding maps via No. 7 and 9 (see Du0102 ). For further applications to purification, storage, compression, tomography and probabilistic implementation of non-local operations and its use in quantum computation we refer the reader to Ref. Du00 and Du03 .
In the following we will mainly consider two-qubit gates, which are of special interest in quantum computation and quantum information. Note that in Ref. Ci00 and Du00 (see Sec.III) it was shown for the case of two-qubit unitary operations, how to modify the teleportation protocol to implement an arbitrary two-qubit unitary with unit probability of success. We will illustrate the results No. 7, No. 8 and No. 9 for these gates, namely for the
* CNOT-gate: $`U^{AB}=|0^A0|+|1^A1|\sigma _x^B`$
$`|\psi _{\text{CNOT}}={\displaystyle \frac{1}{\sqrt{2}}}\left(|00^{AA^{}}|\psi _0^{BB^{}}+|11^{AA^{}}|\psi _1^{BB^{}}\right)`$ (24)
* Phase-gate: $`U^{AB}(\alpha )=e^{i\alpha \sigma _y^A\sigma _y^B}`$
$`|\psi \left(\alpha \right)=\mathrm{cos}\left(\alpha \right)|\psi _0^{AA^{}}|\psi _0^{BB^{}}i\mathrm{sin}\left(\alpha \right)|\psi _2^{AA^{}}|\psi _2^{BB^{}}`$ (25)
* SWAP-gate: $`U^{AB}=|00^{AB}00|+|01^{AB}10|+|10^{AB}01|+|11^{AB}11|`$
$`|\psi _{\text{SWAP}}=|\psi _0^{AB^{}}|\psi _0^{BA^{}}`$ (26)
Note that $`|\psi _{\text{SWAP}}`$ is a product state w.r.t. the partitioning $`AB^{}`$ versus $`A^{}B`$ but not w.r.t. the partitioning $`AA^{}`$ versus $`BB^{}`$. In fact $`|\psi _{\text{SWAP}}`$ has a Schmidt decomposition into $`4`$ $`(AA^{},BB^{})`$-product terms, whereas $`|\psi _{\text{CNOT}}`$ and $`|\psi (\alpha )`$ can be decomposed into $`2`$ $`(AA^{},BB^{})`$-product states. From basic facts Conversion about bi-partite entanglement for pure states it follows from No. 7 that the SWAP-gate can simulate the phase gate and the CNOT-gate by means of SLOCC operation to be performed before and after the SWAP operation (but not vice versa). Moreover the CNOT-gate and the phase gate can simulate each other under SLOCC for arbitrary $`\alpha ]0,2\pi [`$. For the case of $`\alpha =\frac{\pi }{4}`$ they actually coincide up to some local unitaries Kr00 ; Du0102 :
$`U_{\text{CNOT}}^{AB}=U_1^AU_2^BU^{AB}({\displaystyle \frac{\pi }{4}})V_1^AV_2^B\text{with}`$ (27)
$`U_1={\displaystyle \frac{1}{\sqrt{2}}}\left(\begin{array}{cc}1& i\\ 1& i\end{array}\right)`$ $`U_2=\left(\begin{array}{cc}1& 0\\ 0& i\end{array}\right)`$
$`V_1={\displaystyle \frac{1}{\sqrt{2}}}\left(\begin{array}{cc}1& i\\ i& 1\end{array}\right)`$ $`V_2={\displaystyle \frac{1}{\sqrt{2}}}\left(\begin{array}{cc}1& i\\ 1& i\end{array}\right)`$
This corresponds to the fact that the corresponding states $`|\psi _{\text{CNOT}}`$ and $`|\psi (\frac{\pi }{4})`$ are LU-equivalent w.r.t. to the partitioning $`(A,A^{},B,B^{})`$ (see No. 8), i.e.
$`|\psi _{\text{CNOT}}=U_1^A^{}U_2^B^{}\left(V_1^A\right)^T\left(V_2^B\right)^T|\psi ({\displaystyle \frac{\pi }{4}}).`$ (28)
According to No. 9 the SWAP-gate will moreover be capable to create more entanglement than the CNOT-gate and the phase-gate, which can – up to SLOCC – create the same type of entanglement.
## III Standard form for decoherence in the single-party-setting
In this section we apply the Jamiołkowski isomorphism in order to derive a standard form for an arbitrary decoherence process, that is described by some CPM $``$. We show that this standard form can be achieved by randomly choosing appropriate unitaries to be performed before and after the actual CPM occurs. We first consider the case of a qubit system and then discuss a generalization to $`d`$-level systems.
Let denote $`\sigma _0=\mathrm{𝟏}`$, $`\sigma _1=\sigma _x`$, $`\sigma _2=\sigma _y`$ and $`\sigma _3=\sigma _z`$ the Pauli matrices. Note that starting with the maximally entangled state $`|\mathrm{\Phi }^{A\overline{A}}=\frac{1}{\sqrt{2}}\left(|00^{A\overline{A}}+|11^{A\overline{A}}\right)`$ we obtain a complete Bell basis $`(|\psi _0,|\psi _1,|\psi _2,|\psi _3)`$ simply by applying $`\sigma _i`$ locally on system $`A`$, i.e. $`|\psi _i^{A\overline{A}}=\sigma _i^A\text{Id}^{\overline{A}}\left(|\psi _0^{A\overline{A}}\right)`$. Thus any decomposition of a state $`E=_{ij}E_{ij}|\psi _i\psi _j|`$ in terms of the Bell basis corresponds to a canonical representation of the CPM $``$ in terms of Pauli matrices
$$(\rho )=\underset{i,j=0}{\overset{3}{}}E_{ij}\sigma _i\rho \sigma _j,$$
(29)
where the conditions on the matrix $`𝐄=(E_{ij})`$ can directly be read off from the isomorphism, i.e. $`𝐄`$ must be density matrix. A case of particular interest in quantum information theory, especially in the study of fault-tolerance of quantum computation, is when $`E`$ is a diagonal matrix. In many applications the corresponding CPM, the so called Pauli channel
$$(\rho )=\underset{i=0}{\overset{3}{}}E_i\sigma _i\rho \sigma _i\text{with}(\underset{i=0}{\overset{3}{}}E_i=1),$$
(30)
describes some underlying noise model or decoherence process. This class contains for $`E_0=\frac{1+3p}{4}`$ and $`E_1=E_2=E_3=\frac{1p}{4}`$ the depolarizing channel (white noise) $`\rho =p\rho +(1p)\frac{1}{2}\mathrm{𝟏}`$, for $`E_0=\frac{1+p}{2}`$, $`E_1=E_2=0`$ and $`E_3=\frac{1p}{2}`$ the dephasing channel $`\rho =p\rho +\frac{1p}{2}\left(\rho +\sigma _z\rho \sigma _z\right)`$ and for $`E_0=\frac{1+p}{2}`$, $`E_2=E_3=0`$ and $`E_1=\frac{1p}{2}`$ the bit-flip channel $`\rho =p\rho +\frac{1p}{2}\left(\rho +\sigma _x\rho \sigma _x\right)`$.
We show now that the decoherence process specified by an arbitrary CPM $``$ as in Eq. (29) can be transformed into a Pauli channel $`^{}`$ (see Eq. (30)) with the same diagonal elements $`E_i=E_{ii}`$. This can be achieved by a probabilistic but correlated application of one of the four Pauli matrices $`\sigma _i`$ before and after the actual noise occurs:
$`^{}(\rho )={\displaystyle \frac{1}{4}}{\displaystyle \underset{i=0}{\overset{3}{}}}\sigma _i\left(\sigma _i\rho \sigma _i\right)\sigma _i.`$ (31)
In other words, by randomly choosing one of the four Pauli matrices with probability $`\frac{1}{4}`$ to apply to a system before and after the noise process affects the system (e.g. some memory device) and ignoring the information about which Pauli matrix has been applied, an experimenter will actually (only) have to deal with noise of the form of a Pauli channel. The fact that the CPM $``$ can be brought to this form $`^{}`$ follows from the Jamiołkowski isomorphism used as in case (B) and the fact, that the corresponding state $`E`$ can be diagonalized to $`E^{}`$ by a mixing procedure, in which each of the local Pauli operators $`\sigma _i^A\sigma _i^A^{}`$ is applied with probability $`\frac{1}{4}`$:
$`E^{AA^{}}={\displaystyle \frac{1}{4}}{\displaystyle \underset{i=0}{\overset{3}{}}}\sigma _i^A\sigma _i^A^{}E^{AA^{}}\sigma _i^A\sigma _i^A^{}.`$ (32)
The achieved standard form in Eq. (30) can be further depolarized, by considering the following three Clifford operations $`Q_k=e^{i\frac{\pi }{4}\sigma _k}=\sqrt{i\sigma _k}`$ with $`k=1,2,3`$. Starting with a state of standard form Eq. (30) one can in fact compute that
$`{\displaystyle \frac{1}{3}}{\displaystyle \underset{k=1}{\overset{3}{}}}Q_kQ_k^{}EQ_k^{}Q_k^T`$
$`=`$ $`E_0|\psi _0\psi _0|+(E_1+E_2+E_3)\times `$
$`\times `$ $`\left(|\psi _1\psi _1|+|\psi _2\psi _2|+|\psi _3\psi _3|\right).`$
This means that by uniformly choosing one of the $`12`$ unitaries $`U_{ki}=Q_k\sigma _i`$ for $`k=1,2,3`$ and $`i=0,1,2,3`$ and applying $`U_{ki}^{}`$ before and $`U_{ki}`$ after the application of an arbitrary CPM $``$ (see Eq. (29)) the resulting CPM $`^{}`$:
$`^{}(\rho )={\displaystyle \frac{1}{12}}{\displaystyle \underset{ki}{}}U_{ki}\left(U_{ki}^{}\rho U_{ki}\right)U_{ki}^{}`$ (34)
will be of the form of depolarizing channel
$`^{}(\rho )=\alpha (f)\rho +(1\alpha (f))\text{tr}\rho {\displaystyle \frac{1}{d}}\mathrm{𝟏}`$ (35)
with $`f=E_{00}`$ and $`\alpha (f)=\frac{4f1}{3}`$. Note that a similar twirling procedure is also used in the recurrence protocol IBM for entanglement purification. Both depolarization procedures Eq. (31) and Eq. (III) leave the state $`|\mathrm{\Phi }`$ and thus the identity operation Id invariant. Hence the Jamiołkowski fidelity remains the same, i.e. $`F(,\text{Id})=E_{00}=F(^{},\text{Id})`$. Since the Jamiołkowski fidelity represents the noise level of the respective operations $``$ and $`^{}`$, both standard forms can be achieved without introducing additional noise to the system.
Let us now turn to the case of general qudit systems with $`d=\text{dim}_{}(𝐇^A)=\text{dim}_{}(𝐇^A^{})`$. Here the following complete basis of maximally entangled states can be chosen:
$`|\psi _{kl}^{AA^{}}:={\displaystyle \frac{1}{\sqrt{d}}}{\displaystyle \underset{m=0}{\overset{d1}{}}}e^{i\frac{2\pi }{d}km}|m+l^A|m^A^{},`$ (36)
where addition $`m+l`$ and multiplication $`km`$ is meant modulo $`d`$. Note that the Bell basis can be generated by acting on only one of the systems by means of unitaries $`U_{kl}`$ (generalized Pauli group)
$`U_{kl}|m:=e^{i\frac{2\pi }{d}km}|m+l`$ (37)
out of the maximally entangled state $`|\psi _{00}=|\mathrm{\Phi }`$, e.g.
$`|\psi _{kl}^{AA^{}}=U_{kl}^A\text{Id}_A^{}\left(|\psi _{00}^{AA^{}}\right).`$ (38)
Similar to Eq. (29) the canonical form for an arbitrary CPM in terms of the generalized Pauli operators is
$`(\rho )={\displaystyle \underset{kl,k^{}l^{}}{}}E_{kl,k^{}l^{}}U_{kl}\rho U_{k^{}l^{}}^{}.`$ (39)
With respect to this Bell basis the corresponding state has the decomposition
$`E={\displaystyle \underset{kl,k^{}l^{}}{}}E_{kl,k^{}l^{}}|\psi _{kl}\psi _{k^{}l^{}}|.`$ (40)
By generalizing the depolarization procedure in Eq. (31) we can again diagonalize the state $`E`$ and thus bring the corresponding CPM $``$ to the form of a (generalized) Pauli channel.
Standard form: Pauli Channel
By uniformly choosing one of the $`d^2`$ Pauli operators $`U_{kl}`$ and applying $`U_{kl}^{}`$ before and $`U_{kl}`$ after the application of an arbitrary CPM $``$ (see Eq. (39)) the resulting CPM $`^{}`$:
$`^{}(\rho )={\displaystyle \frac{1}{d^2}}{\displaystyle \underset{k,l=0}{\overset{d1}{}}}U_{kl}\left(U_{kl}^{}\rho U_{kl}\right)U_{kl}^{}`$ (41)
will be of the form
$`^{}(\rho )={\displaystyle \underset{kl}{}}E_{kl}^{}U_{kl}\rho U_{kl}`$ (42)
with $`E_{kl}^{}=E_{kl,kl}`$.
A proof of this statement can be found in the Appendix C. Whereas an arbitrary hermitian matrix $`E`$ is described by $`\frac{1}{2}d^2(d^21)`$ real parameters, which in addition have to fulfill the constraints No. 4 (Sec. II.2) in order to correspond to a trace preserving CPM $``$, an arbitrary Pauli channels $`^{}`$ can be described by only $`d^21`$ positive parameters $`E_{kl}`$ parametercount .
The number of parameters can even be decreased by considering for the symmetrization procedure not only the Pauli group $`S:=\{\pm 1,\pm i\}\times \{\sigma _0,\sigma _1,\sigma _2,\sigma _3\}`$ but the larger group
$`S^{}=\{U^AU_{}^{}{}_{}{}^{A^{}}|U𝐔(d)\}`$ (43)
of all local unitaries of the form $`U^AU_{}^{}{}_{}{}^{A^{}}`$. Since the group $`S^{}`$ contains $`S`$, it has at most a smaller commutant. Whereas the commutant of $`S`$ is the set of all Bell diagonal states, the commutant of $`S^{}`$ is indeed Hor99 only generated by the (orthogonal) states $`P_\mathrm{\Phi }`$ and
$`\gamma :={\displaystyle \frac{1}{d^21}}{\displaystyle \underset{\genfrac{}{}{0.0pt}{}{k,l}{(k,l)(0,0)}}{}}|\psi _{kl}\psi _{kl}|={\displaystyle \frac{1}{d^21}}\left(\mathrm{𝟏}P_\mathrm{\Phi }\right).`$ (44)
In other words the set of states, that is invariant under $`S^{}`$, is determined by a fewer set of parameters. In fact, the states $`E^{}=𝒟(E)`$ obtained by this ’twirling’ operation
$`𝒟(E)={\displaystyle (UU^{})E(U^{}U^T)𝑑U},`$ (45)
where $`dU`$ denotes the uniform probability distribution on the unitary group $`𝐔(d)`$ proportional to the Haar measure, is determined by a single real parameter :
$`E^{}`$ $`=`$ $`fP_\mathrm{\Phi }+(1f)\gamma `$ (46)
$`=`$ $`\alpha (f)P_\mathrm{\Phi }+(1\alpha (f)){\displaystyle \frac{1}{d^2}}\mathrm{𝟏},`$ (47)
with $`\alpha (f)=\frac{d^2f1}{d^21}`$. Note that the fidelity $`f=\mathrm{\Phi }|E^{}|\mathrm{\Phi }=\mathrm{\Phi }|E|\mathrm{\Phi }`$ ($`0f1`$) is left unchanged under the twirling procedure $`𝒟`$, since $`𝒟`$ simply is a projection onto the subset of states invariant under this ’isotropic symmetry’:
$`𝒟(E)=\text{tr}(P_\mathrm{\Phi }E)P_\mathrm{\Phi }+(d^21)\text{tr}(\gamma E)\gamma .`$ (48)
We remark that by partial transposition these isotropic states $`E^{}`$ are in one-to-one correspondence Wo03 to the set of Werner states We89 . Since $`E^{}`$ is a mixture of the maximally entangled state $`P_\mathrm{\Phi }`$ and a maximally mixed state, we find that the normal form of the corresponding CPM $`^{}`$ is the (generalized) depolarizing channel IsoOf1 :
Standard form: Depolarizing Channel
By uniformly choosing a unitary $`U𝐔(d)`$ and applying $`U^{}`$ before and $`U`$ after the application of an arbitrary CPM $``$ (see Eq. (39)) the resulting CPM $`^{}`$:
$`^{}(\rho )={\displaystyle U\left(U^{}\rho U\right)U^{}𝑑U}`$ (49)
will be of the form
$`^{}(\rho )`$ $`=`$ $`f\rho +{\displaystyle \frac{1f}{d^21}}{\displaystyle \underset{\genfrac{}{}{0.0pt}{}{k,l}{(k,l)(0,0)}}{}}U_{kl}\rho U_{kl}^{}`$ (50)
$`=`$ $`\alpha (f)\rho +(1\alpha (f))\text{tr}\rho {\displaystyle \frac{1}{d}}\mathrm{𝟏}`$
with $`f=E_{00}`$ and $`\alpha (f)=\frac{d^2f1}{d^21}`$.
Since an isotropic state $`E^{}`$ as in Eq. (47) is separable iff $`f\frac{1}{d}`$ Hor99 , we note that, according to No. 13. in Appendix A, the corresponding depolarizing channel becomes entanglement breaking at this point.
Let us briefly address the question of possible practical implementations of the twirling protocol described above. As it is shown in the Appendix C it is actually sufficient for the depolarization protocol to uniformly choose some unitaries from a finite set of Clifford unitaries.
To summarize we have shown that both standard forms $`^{}`$, the Pauli channel and the depolarizing channel, can be obtained by a random application of quantum operations applied before and after the actual CPM $``$. These operations are chosen uniformly at random from a finite set of unitaries. Moreover we have seen that these depolarization protocols do not introduce additional noise to the system.
## IV Standard forms for CPM in the multi-party setting
In this section we will continue the discussion of standard forms for noisy quantum operations. We will consider a partitioning of the system $`A=(A_1,\mathrm{},A_N)`$ into $`N`$ parties, which might be located at distant places. Any depolarization protocol that brings a given (non-local) CPM into its standard form therefore should be local w.r.t. this partitioning. In the following we will consider an ideal operation $``$, that can only be realized imperfectly as a CPM $``$. We are now interested in the possible normal forms $`^{}`$, into which we can transform $``$ by means of LOCC operation (w.r.t. to the given partitioning), that are carried out before and after $``$ dir/indir is actually applied. If one is interested in the standard form for a map describing a given decoherence process itself, the ideal operation is the identity $`=\text{Id}`$. Apart from the identity we will in the multi-party setting also consider the case, where the ideal operation is some 2-qubit unitary operation $`=𝒰`$ ($`𝒰(\rho )=U\rho U^{}`$), which can only be realized in form of some noisy quantum operation $``$. In contrast to the case discussed in the previous section, the locality requirements now impose rather severe constraints on the allowed operations to manipulate a given CPM. Note that the state $`I`$ corresponding to the ideal operation (identity or unitary) is pure. That is for $`=𝒰`$ \[$`=\text{Id}`$\] we have $`I=|\psi _U\psi _U|`$ \[$`I=|\mathrm{\Phi }\mathrm{\Phi }|`$\] respectively. Before we give an outline of this section we mention several aspects of the problem of finding such a standard form.
* One can distinguish the two cases where only deterministic or also probabilistic transformations are considered, i.e. whether it is possible to transform $``$ into the respective normal form $`^{}`$ in all of the possible measurement branches of the LOCC protocol or in at least one.
* Closely related to this distinction is the question whether one uses the teleportation protocol directly as in (A) or indirectly as in (B), since a direct use of the isomorphism protocol in general has only a certain probability of success.
* Firstly, one would like the transformation protocol $`𝒟`$ (on state level) to leave the ideal operation invariant, i.e. $`𝒟(I)=I`$. In this case the fidelity $`F(^{},)=\text{tr}IE^{}`$ of the ideal operation with the transformed noisy operation $`^{}`$ will be the same as the fidelity $`F(,)=\text{tr}IE`$ of the ideal operation with the noisy operation $``$. Since the Jamiołkowski fidelity with the ideal operation can be regarded as some kind of distance measure, the transformation will keep $``$ as close to the ideal operation as before. For the case of the ideal operation being the identity, the protocol $`𝒟`$ simply should be unital. On the other hand one might as well be allowed to sacrifice some fidelity with the ideal operation in order derive simpler standard forms.
* The transformation protocol might bring any CPM onto its respective standard form (universal protocol) or it might be designed to transform a specific CPM into standard form .
Note that most of the differences in these versions of the problem only become important in the multi-party setting. This is mainly due to the fact that the depolarizing channel already provides a standard form for an arbitrary noise process, which can be achieved deterministically by a unital transformation and which is already specified by a single noise parameter $`f`$.
We generalize the results of Sec. III in Sec. IV.1 and derive standard form for decoherence processes (i.e. for the case $`=\text{Id}`$) under the constraint that the underlying control operation have to be local w.r.t. the given partitioning. In Sec. IV.2 we discuss the case where the ideal operation is one of the unitary gates SWAP, CNOT or a phase gate with some arbitrary angle. In this section we restrict first to those depolarization procedures that are universal, deterministic and leave these unitary gates invariant. In Sec. IV.3 we also discuss the case where the fidelity $`f`$ of the operation is decreased by a certain amount (i.e. additional noise is introduced) in order to obtain a simpler standard form describing the noise process, that is to reduce the number of required parameters. For gates locally equivalent to SWAP and CNOT, this leads to noise processes described by global white noise. A similar result is obtained for all phase gates, provided that one has control over switching the noisy operation on and off at will. The noise is – in the worst case – increased by an order of magnitude. Finally we briefly mention the problems in Sec. IV.4 that occur when trying to transfer the techniques developed in Sec. IV.2 and Sec. IV.3 to the more general case of an arbitrary unitary operation as the ideal operation.
### IV.1 Standard forms for decoherence in the multi–party setting
#### IV.1.1 Depolarization without sacrificing
Let us now consider possible standard forms for noise operations (i.e. ideal operation is the identity) in the multi-partite setting. Note that the twirling operation $`𝒟`$ used in Sec. III corresponds to a projection into the space of states $`E`$ over $`𝐇^A𝐇^A^{}`$ (recall that we chose $`d_A=\text{dim}_{}(𝐇^A)=\text{dim}_{}(𝐇^A^{})=_{i=1}^Nd_{A_i}`$ ). Thus it is straightforward to derive the corresponding standard form for the multi-party case, which is obtained after sequential application of the twirling operation $`𝒟`$ locally at each party. Since the invariant group in question are $`_{i=1}^NS_{A_i}`$ and $`_{i=1}^NS_{A_i}^{}`$ respectively, the commutants of these groups (i.e. the subspace of invariant matrices in $`\left(𝐇^A^{}𝐇^A\right)`$, onto which the projection $`_{i=1}^N𝒟_{A_i}`$ projects) are just given by the tensor products of the commutants (subspaces) for each party. Hence, a probabilistic application of local unitaries $`g_{\mathrm{𝐤𝐥}}=g_{k_1l_1}^{A_1A_1^{}}\mathrm{}g_{k_Nl_N}^{A_NA_N^{}}`$MultiIndex with probability $`\frac{1}{d_A}=\frac{1}{d_{A_1}}\mathrm{}\frac{1}{d_{A_N}}`$ diagonalizes $`E`$ and gives the generalized multi-partite Pauli channel
$`^{}(\rho )={\displaystyle \underset{𝐤,𝐥_{d_{A_1}}\times \mathrm{}\times _{d_{A_N}}}{}}E_{\mathrm{𝐤𝐥}}U_{\mathrm{𝐤𝐥}}\rho U_{\mathrm{𝐤𝐥}},`$ (51)
which is again specified by $`d_A^21`$ positive parameters ($`_{d_i}:=\{0,\mathrm{},d_i1\}`$). In the case of equal dimensions $`d`$ the channel is determined by $`d_A=d^{2N}1`$ parameters. For a noise operation on two qubits, for example, the corresponding standard form is given by
$`^{}(\rho )={\displaystyle \underset{i,j=0}{\overset{3}{}}}E_{ij}\sigma _i^{A_1}\sigma _j^{A_2}\rho \sigma _i^{A_1}\sigma _j^{A_2}.`$ (52)
As in the case of a single system, further depolarization is possible and hence a simpler standard form can be achieved. To this aims one performs a complete twirl over the larger group $`S^{}`$. The result of this twirl is that one projects $`E`$ into the set of states of the form
$`E={\displaystyle \underset{𝐤_2\times \mathrm{}\times _2}{}}E_𝐤\gamma _𝐤,`$ (53)
where $`\gamma _𝐤=\gamma _{k_1}^{A_1}\mathrm{}\gamma _{k_N}^{A_N}`$ and at each party $`A_i`$ $`\gamma _{k_i}^{A_i}`$ denotes one of the two orthogonal states $`\gamma _0=P_\mathrm{\Phi }`$ or $`\gamma _1=\gamma `$ spanning the respective isotropic subspace. If $`E^{}`$ is decomposed w.r.t. the (non-orthogonal) basis $`(\gamma _0,\gamma _1)=(P_\mathrm{\Phi },\frac{1}{d_{A_i}^2}\mathrm{𝟏})`$, the corresponding CPM $`^{}`$ has a natural interpretation in terms of different white noise factors. To be more precise let us consider the example of two qudits with $`d=d_{A_1}=d_{A_2}`$. In this case the state $`E^{}`$ is of the form
$`E^{}`$ $`=`$ $`\alpha _{00}P_\mathrm{\Phi }^{A_1}P_\mathrm{\Phi }^{A_2}+\alpha _{01}P_\mathrm{\Phi }^{A_1}{\displaystyle \frac{1}{d^2}}\mathrm{𝟏}_{A_2}`$ (54)
$`+\alpha _{10}{\displaystyle \frac{1}{d^2}}\mathrm{𝟏}_{A_1}P_\mathrm{\Phi }^{A_2}+\alpha _{11}{\displaystyle \frac{1}{d^2}}\mathrm{𝟏}_{A_1}{\displaystyle \frac{1}{d^2}}\mathrm{𝟏}_{A_2}.`$
Since $`(\delta _0,\delta _1)=(P_\mathrm{\Phi }\gamma ,d^2\gamma )`$ is a dual basis for $`(P_\mathrm{\Phi },\frac{1}{d^2}\mathrm{𝟏})`$, we have that
$`\alpha _{00}`$ $`=`$ $`\text{tr}\left[\delta _0^{A_1}\delta _0^{A_2}E^{A_1A_2}\right]`$
$`=`$ $`E_{00}{\displaystyle \frac{E_{01}}{d^21}}{\displaystyle \frac{E_{10}}{d^21}}+{\displaystyle \frac{E_{11}}{(d^21)^2}}`$
$`\alpha _{01}`$ $`=`$ $`\text{tr}\left[\delta _0^{A_1}\delta _1^{A_2}E^{A_1A_2}\right]`$
$`=`$ $`{\displaystyle \frac{d^2E_{01}}{d^21}}{\displaystyle \frac{d^2E_{11}}{(d^21)^2}}`$
$`\alpha _{10}`$ $`=`$ $`\text{tr}\left[\delta _1^{A_1}\delta _0^{A_2}E^{A_1A_2}\right]`$
$`=`$ $`{\displaystyle \frac{d^2E_{10}}{d^21}}{\displaystyle \frac{d^2E_{11}}{(d^21)^2}}`$
$`\alpha _{11}`$ $`=`$ $`\text{tr}\left[\delta _1^{A_1}\delta _1^{A_2}E^{A_1A_2}\right]={\displaystyle \frac{d^4}{(d^21)^2}}E_{11}.`$ (55)
For the corresponding normal form $`^{}`$ of the CPM $``$ we obtain
$`^{}(\rho )`$ $`=`$ $`\alpha _{00}\rho +\alpha _{01}\rho _{A_1}{\displaystyle \frac{1}{d}}\mathrm{𝟏}_{A_2}+\alpha _{10}{\displaystyle \frac{1}{d}}\mathrm{𝟏}_{A_1}\rho _{A_2}`$ (56)
$`+\alpha _{11}{\displaystyle \frac{1}{d^2}}\text{tr}(\rho )\mathrm{𝟏}_{A_1A_2},`$
where $`\rho _{A_1}=\text{tr}_{A_2}(\rho )`$ and $`\rho _{A_2}=\text{tr}_{A_1}(\rho )`$. The second and third term correspond to white noise introduced locally at each party, whereas the last summand introduces global white noise. Note that some of the coefficients $`\alpha _{00}`$, $`\alpha _{01}`$ or $`\alpha _{10}`$ can be negative. In the case of equal dimensions at each party ($`d_{A_i}=d`$) we have reduced the number of parameters from initially $`O(d^{4N})`$ for a general CPM $``$ over $`d^{2N}1`$ for a general multi-party Pauli channel to $`2^N1`$ parameters (independent of the dimension $`d`$) to describe different types of multi-party white noise.
So far we have only considered twirling protocols, that were deterministic, made direct use of the isomorphism and left the ideal operation invariant, namely the identity operation. But can we further reduce the number of parameters for a different standard form, which is achieved by a LOCC protocol, that has only a certain probability of success or that makes indirect use of the isomorphism or allows to sacrifice some of the initial fidelity $`E_{00}`$ with the ideal operation ? To be more precise, is it possible by weakening one of these conditions to achieve e.g. just a global white noise channel
$`^{}(\rho )=p\rho +(1p){\displaystyle \frac{1}{d^2}}\mathrm{𝟏},`$ (57)
i.e. the case, for which also all the diagonal elements vanish except $`\alpha _{00}`$ and $`\alpha _{11}`$?
If one allows for a direct use of the isomorphism protocol and thus to perform local Bell measurements in the basis $`|\psi _i`$, this is certainly possible, since the ideal operation Id is local w.r.t. any partitioning and the coefficients $`E_{\mathrm{𝐤𝐥}}`$ in Eq. (51) can deliberately be adapted without increasing the noise level by simply ’twirling’ all components $`E_{\mathrm{𝐤𝐥}}`$ but $`E_{\mathrm{𝟎𝟎}}`$ into $`\frac{1}{d}\mathrm{𝟏}`$. Note that this procedure gives rise to a probabilistic LOCC protocol, since a direct use of the Jamiołkowski isomorphism can in general not be achieved with unit probability of success (see Sec. II.2). As discussed in Sec. II.2 this use of the isomorphism will therefore in general not be of great interest for all applications, in which one would like to work with a standard form of a given CPM rather than with the CPM itself.
Using the isomorphism only indirectly, we have to restrict the transformations to the class of SLOCC operation that are also local w.r.t. to $`(A_i,\overline{A}_i)`$. Note that one cannot increase the fidelity $`f`$ with the ideal operation $`=\text{Id}`$ by means of any physical protocol, whenever $`f`$ corresponds to the largest eigenvalue in $`E`$ noIncrease , e.g. Eq. (51) with $`E_{\mathrm{𝟎𝟎}}>E_{\mathrm{𝐤𝐥}}`$ ($`\neg 𝐤=𝐥=0`$). In all these cases any SLOCC operation $`C_1^AC_2^{\overline{A}}`$, that is contained in the transformation protocol and that does not leave $`|\mathrm{\Phi }`$ invariant, will cause a decrease in the fidelity $`f^{}<f`$ of the respective standard form $`^{}`$. Thus any universal transformation protocol, that yields a respective standard form for all CPM without decreasing the fidelity, consists in a probabilistic application of operation $`C_1^AC_2^{\overline{A}}`$ with $`C_1^AC_2^{\overline{A}}|\mathrm{\Phi }=|\mathrm{\Phi }`$, i.e. $`C_1C_2=C(C^1)^T`$ for some invertible matrix $`C`$ AllInvTrafo . Since all such transformation will also leave all $`\gamma _𝐤`$ in Eq. (53) invariant, the respective standard form cannot contain fewer terms than the multi-party white noise channel. In this sense the above twirling procedure already yields a standard form, that is optimal among all forms achieved by some universal protocol, that only use the isomorphism indirectly and does not sacrifice any fidelity with the ideal operation.
#### IV.1.2 Depolarization by means of sacrificing
In the remainder of this subsection we will show how to design twirling protocols that bring a specific CPM into the standard form of global white noise by introducing additional noise. The procedure described below does therefore satisfy neither the universality property nor the no–sacrificing condition, but it will - especially for the many party case - significantly reduce the number of parameters of the standard form to a single one. By applying the above universal depolarization procedures, we can start with considering only states $`E`$, that are already in isotropic form (see Eq. (53)).
We will illustrate the procedure for the case of two qubits. Generalization to the multipartite case and higher dimensions are straightforward. For such two–qubit maps, the corresponding state $`E`$ (after applying the universal depolarization protocol described in the previous section) is given by
$`E`$ $`=`$ $`E_{00}P_{\psi _0}^{A_1}P_{\psi _0}^{A_2}+E_{01}{\displaystyle \underset{j=1}{\overset{3}{}}}P_{\psi _0}^{A_1}P_{\psi _j}^{A_2}+`$ (58)
$`E_{10}{\displaystyle \underset{i=1}{\overset{3}{}}}P_{\psi _i}^{A_1}P_{\psi _0}^{A_2}+E_{11}{\displaystyle \underset{i,j=1}{\overset{3}{}}}P_{\psi _i}^{A_1}P_{\psi _j}^{A_2},`$
where $`P_{\psi _i}=|\psi _i\psi _i|`$ is the projector onto one of the Bell states $`|\psi _i`$. In the following we will collect the parameters in a vector $`𝐄=(E_{00},E_{01},E_{10},E_{11})^T`$. The fact, that $`E`$ corresponds to a trace preserving CPM then simply reads (i) $`𝐄0`$ component-wise (i.e. $`E_{ij}0`$ for $`i,j=0,1`$) and (ii) $`N(𝐄):=E_{00}+3\left(E_{01}+E_{10}+3E_{11}\right)=1`$. We now consider the following type of depolarization
$`𝒟(E)`$ $`=`$ $`p_{00}E+p_{01}{\displaystyle \underset{j=1}{\overset{3}{}}}\sigma _j^{A_2}E\sigma _j^{A_2}+p_{10}{\displaystyle \underset{i=1}{\overset{3}{}}}\sigma _i^{A_1}E\sigma _i^{A_1}`$ (59)
$`+p_{11}{\displaystyle \underset{i,j=1}{\overset{3}{}}}\sigma _i^{A_1}\sigma _j^{A_2}E\sigma _i^{A_1}\sigma _j^{A_2}.`$
In a similar notation as before $`𝒟`$ corresponds to a trace preserving CPM iff $`N(𝐩)=1`$ and $`𝐩0`$. It is straight forward to calculate, that the resulting state $`E^{}=𝒟(E)`$ is again in isotropic form Eq. (58) with new parameters $`E_{ij}^{}`$, that are given by the linear transformation
$`𝐄^{}=𝐃[𝐄]𝐩,`$ (60)
where the matrix $`𝐃[𝐄]`$ depends on the initial state $`E`$:
$`\left(\begin{array}{cccc}E_{00}& 3E_{01}& 3E_{10}& 9E_{11}\\ E_{01}& E_{00}+2E_{01}& 3E_{11}& 3\left(E_{10}+2E_{11}\right)\\ E_{10}& 3E_{11}& E_{00}+2E_{10}& 3\left(E_{01}+2E_{11}\right)\\ E_{11}& E_{10}+2E_{11}& E_{01}+2E_{11}& E_{00}+2\left(E_{01}+E_{10}+2E_{11}\right)\end{array}\right)`$ (61)
Our goal is to bring $`E`$ into a form, that corresponds only to global white noise (see Eq. (57)), i.e. $`E_{01}^{}=E_{10}^{}=\frac{1}{3}E_{11}^{}`$. Since we require $`𝐄^{}0`$ and $`N(𝐄^{})=1`$, $`E^{}`$ should be of the form $`𝐄^{}(f^{})=(f^{},\frac{1f^{}}{9},\frac{1f^{}}{9},\frac{1f^{}}{27})^T`$ with the new fidelity $`f^{}[0,1]`$. Thus we look for solutions $`𝐩`$ to the linear system Eq. (60) for the specific choice of $`𝐄^{}(f^{})`$, that additionally fulfills the constraints $`N(𝐩)=1`$ and $`𝐩0`$. One computes that $`N(𝐄^{})=N(𝐩)N(𝐄)`$ and we can therefore omit the trace preservation condition $`N(𝐩)=1`$, since we already require (chose) $`E`$ and $`E^{}`$ to be trace preserving. Using $`N(𝐄)=1`$ one can compute for the determinant $`\text{det}(𝐃[𝐄])=rst`$, where $`r=14(E_{01}+3E_{11})`$ , $`s=14(E_{10}+3E_{11})`$ and $`t=14(E_{01}+E_{10}+2E_{11})`$. Thus the linear system Eq. (60) will definitely have a unique solution, whenever $`\text{max}(E_{01},E_{10},E_{11})<\frac{1}{16}`$ (or alternatively the initial fidelity $`f=E_{00}>4\text{max}(E_{01},E_{10},E_{11})`$). Let us consider a fixed vector $`𝐄`$ with $`\text{det}(𝐃[𝐄])0`$ first. For the corresponding CPM $``$ we want to design a standard form $`^{}`$ with maximal fidelity $`f^{}`$. In contrast to the standard forms discussed so far the depolarization process $`𝒟`$, which translates into applying Pauli operators before and after the actual CPM $``$ occurs, will be specifically designed for the given initial CPM $``$, since the corresponding probabilities are given by the unique solution $`𝐩(f^{})=𝐃[𝐄]^1𝐄^{}(f^{})`$. Note that each of the inequalities $`p_{ij}0`$ is linear in $`f^{}`$, i.e. of the form $`a_{ij}f^{}+b_{ij}0`$, where the coefficients are
$`a_{00}={\displaystyle \frac{1}{6}}\left({\displaystyle \frac{1}{r}}+{\displaystyle \frac{1}{s}}+{\displaystyle \frac{4}{t}}\right)`$ $`b_{00}={\displaystyle \frac{1}{48}}\left(3+{\displaystyle \frac{1}{r}}+{\displaystyle \frac{1}{s}}{\displaystyle \frac{5}{t}}\right)`$
$`a_{01}={\displaystyle \frac{1}{18}}\left({\displaystyle \frac{3}{s}}{\displaystyle \frac{1}{r}}{\displaystyle \frac{4}{t}}\right)`$ $`b_{01}={\displaystyle \frac{1}{144}}\left(9+{\displaystyle \frac{3}{s}}+{\displaystyle \frac{5}{t}}{\displaystyle \frac{1}{r}}\right)`$
$`a_{10}={\displaystyle \frac{1}{18}}\left({\displaystyle \frac{3}{r}}{\displaystyle \frac{1}{s}}{\displaystyle \frac{4}{t}}\right)`$ $`b_{10}={\displaystyle \frac{1}{144}}\left(9+{\displaystyle \frac{3}{r}}+{\displaystyle \frac{5}{t}}{\displaystyle \frac{1}{s}}\right)`$
$`a_{11}={\displaystyle \frac{1}{54}}\left({\displaystyle \frac{4}{t}}{\displaystyle \frac{3}{r}}{\displaystyle \frac{3}{s}}\right)`$ $`b_{11}={\displaystyle \frac{1}{432}}\left(27{\displaystyle \frac{3}{r}}{\displaystyle \frac{3}{s}}{\displaystyle \frac{5}{t}}\right).`$
Depending on the signs of $`a_{ij}0`$ the constraints are thus represented by intervals starting or ending at $`f_{ij}^{}=\frac{b_{ij}}{a_{ij}}`$ (if $`a_{ij}=0`$ the corresponding condition is either always or never satisfied). In the following we will discuss the (complete) positivity condition $`𝐩0`$ in terms of the parameter $`(r,s,t)`$ instead of $`(E_{01},E_{10},E_{11})`$ since they are linearly related. Because of $`0E_{ij}1`$ we generally have $`15r,s,t1`$. The back transformation is given by $`E_{01}=\frac{1}{16}\left(1+3sr3t\right)`$, $`E_{10}=\frac{1}{16}\left(1+3rs3t\right)`$ and $`E_{11}=\frac{1}{16}\left(1+trs\right)`$, which implies $`E_{00}=\frac{1}{16}\left(1+3(r+s+3t)\right)`$.
Let us consider the situation, in which the initial CPM $``$ is close enough to the ideal operation, such that $`\text{max}(E_{01},E_{10},E_{11})<\frac{1}{16}`$. Then there exists a unique solution with $`1r,s,t>0`$. Moreover a further restriction of $`r,s,t`$ to the interval $`]\frac{2}{3},1]`$ will ensure $`a_{01},a_{10},a_{11}<0`$ and $`f_{00}^{}f_{01}^{},f_{10}^{},f_{11}^{}`$. Thus for a fixed initial vector $`𝐄`$ we are left with the three constraints $`f^{}f_{01}^{},f_{10}^{},f_{11}^{}`$ (and $`f^{}1`$) and the maximal achievable fidelity is $`f_{\text{max}}^{}=\text{min}(1,f_{01}^{},f_{10}^{},f_{11}^{})`$. Although restricting $`E_{10},E_{01},E_{11}`$ to the interval $`[0,\frac{1}{48}]`$ will be sufficient to guarantee that $`r,s,t]\frac{2}{3},1]`$, not all $`r,s,t`$ in the interval $`]\frac{2}{3},1]`$ correspond to $`E_{01},E_{01},E_{01}[0,1]`$. A minimization of $`f_{\text{max}}^{}`$ for $`r,s,t]\frac{2}{3},1]`$ therefore yields only an upper bound to the fidelity decrease $`f_{\text{max}}^{}`$ (versus $`f`$) or the increase of $`1f_{\text{max}}^{}`$ (versus $`1f`$), which represents the noise level of the CPM. A numerical minimization in this region shows that the relative fidelity $`\frac{f_{\text{max}}^{}}{f}`$ is at least $`37.14\%`$ and the relative noise level $`\frac{1f_{\text{max}}^{}}{1f}`$ is at most increased by a factor $`5.5`$.
Note that for an initial fidelity $`f>\frac{15}{16}`$ we have $`0E_{10},E_{01},E_{11}<\frac{1}{48}`$. In other words any decoherence process on two qubits, that introduces only little noise ( i.e. $`f>\frac{15}{16}`$), can be brought into the form of global white noise by increasing the noise level by a factor less than $`5.5`$. This standard form is achieved by application of local Pauli operations, that are chosen randomly according to some probability distribution specified by the parameters $`p_{ij}(f_{\text{max}}^{})`$. Thus the protocol is specifically designed for the initial form of the decoherence process (more precisely it depends on the vector $`𝐄`$, that is obtained after the decoherence is brought into the form Eq. (58) by the methods described above).
If the initial CPM does not belong to the region with $`0E_{10},E_{01},E_{11}<\frac{1}{48}`$, a similar derivation can be applied. The constraints $`a_{ij}f^{}+b_{ij}0`$ again determine, whether a standard form can be obtained in this way and how much fidelity has to be sacrificed in order to achieve the normal form with $`𝐄^{}(f^{})`$. Moreover a generalization to the case of $`d`$-level systems with $`d>2`$ and to the multi-party setting with $`N>2`$ can be developed along the lines of the previous discussion. Note that for increasing $`N`$, although the achieved standard forms will also be global white noise and thus be specified by one parameter only, the derivation and the transformation protocol itself will become more involved, since the number of parameters $`p_{i_1\mathrm{}i_N}`$ will be $`2^N`$ and thus increase exponentially.
### IV.2 Standard forms for noisy unitary operations
We now turn to standard forms of noisy operations where the ideal operation is given by some unitary operation $`U`$. We concentrate on two–particle operations and will illustrate our approach with help of several examples, including gates which are up to local unitary operations equivalent to the SWAP gate, the CNOT gate and a general phase gate with arbitrary phase $`\alpha `$. We will show that one can depolarize these noisy gates to standard forms with a reduced number of parameters, without changing the fidelity of the ideal operation. To this aim, we decomposes a CPM $``$ into a unitary part $`U`$ and some remaining (orthogonal) part $`^{}`$ (where $`\stackrel{~}{}`$ is in general no longer a CPM), i.e.
$`\rho =fU\rho U^{}+(1f)\stackrel{~}{}\rho ,`$ (62)
and both, $`f`$ and $`\stackrel{~}{}`$, are determined by the Isomorphism. We have that
$`f`$ $`=`$ $`\mathrm{\Psi }_U|E|\psi _U`$ (63)
$`\stackrel{~}{E}`$ $`=`$ $`{\displaystyle \frac{Ef|\mathrm{\Psi }_U\mathrm{\Psi }_U|}{1f}},`$ (64)
where $`\stackrel{~}{E}`$ is the operator corresponding to the map $`\stackrel{~}{}`$ and $`f`$ specifies the initial fidelity of the operation $`U`$. It is not necessary to make such a decomposition, however in this notation it is immediately evident that only the noise part, namely $`\stackrel{~}{}`$ is altered by the depolarization procedure. We remark that $`\text{tr}(\stackrel{~}{E})=1`$, however $`\stackrel{~}{E}`$ may have negative eigenvalues and is hence not a density operator. Nevertheless, we can formally decompose $`E`$ (and thus $``$) into these two parts. We will show that one can depolarize the map $``$ to
$`^{}\rho =fU\rho U^{}+(1f)\stackrel{~}{}^{}\rho ,`$ (65)
where $`\stackrel{~}{}^{}`$ is a (generally non–positive) map of certain standard form, specified by a few parameters. Clearly, the total map $`^{}`$ remains completely positive. Note that the depolarization of $``$ takes place in such a way that the weight of the ideal operation is not altered. In particular, if the operation is initially noiseless (i.e. $`f=1`$), it will remain noiseless after the depolarization. This is achieved by considering depolarization processes that leave the unitary operation $`U`$ (or equivalently the state $`|\mathrm{\Psi }_U`$ when considering the operator $`E`$ corresponding to the operation $``$) invariant. The number of required parameters and the explicit form of $`\stackrel{~}{}`$ depends on the ideal operation $`U`$, as the group of local operations that leave $`U`$ invariant is determined by the structure of the state $`|\mathrm{\Psi }_U`$.
#### IV.2.1 The noisy SWAP gate
In this section we determine a standard form for noisy SWAP operations. The ideal $`d`$–level SWAP operation is defined via its action on product basis states, namely $`U_{\mathrm{SWAP}}|i^A|j^B=|j^A|i^B`$, where $`\{|k\}_{k=0,1,\mathrm{},d1}`$ is a basis of $`𝐇=^d`$. The state $`E_{\mathrm{SWAP}}=|\mathrm{\Psi }_{\mathrm{SWAP}}\mathrm{\Psi }_{\mathrm{SWAP}}|`$ corresponding to $`U_{\mathrm{SWAP}}`$ is specified by (see Eq. (26))
$`|\mathrm{\Psi }_{\mathrm{SWAP}}=|\mathrm{\Phi }^{AB^{}}|\mathrm{\Phi }^{BA^{}}.`$ (66)
Consider the mixed state $`E`$ describing —via the isomorphism— a noisy SWAP gate. We have that all operations of the form $`U^AU_{}^{}{}_{}{}^{B^{}}V^BV_{}^{}{}_{}{}^{A^{}}`$ leave $`|\mathrm{\Psi }_{\mathrm{SWAP}}`$ invariant and hence can be used to depolarize $`E`$. This implies that we can essentially use the same depolarization procedure as in the case where the ideal operation is given by the identity (see Sec. IV.1), only the role of particles $`A^{}`$ and $`B^{}`$ is exchanged. This implies that the resulting standard form can again be interpreted as a local and global white noise processes with three independent parameters,that occur before the application of an ideal SWAP operation, i.e. $`^{}(\rho )=U_{\mathrm{SWAP}}𝒟(\rho )U_{\mathrm{SWAP}}^{}`$ with
$`𝒟(\rho )`$ $`=`$ $`\alpha _{00}\rho +\alpha _{01}\rho _A{\displaystyle \frac{1}{d}}\mathrm{𝟏}_B+\alpha _{10}{\displaystyle \frac{1}{d}}\mathrm{𝟏}_A\rho _B`$ (67)
$`+\alpha _{11}{\displaystyle \frac{1}{d^2}}\text{tr}(\rho )\mathrm{𝟏}_{AB}.`$
Note that the parameters $`\alpha _{kl}`$ are again given by Eq. (IV.1.1), where $`E_{kl}`$ are the coefficients in a decomposition Eq. (53) of $`E`$ according to the basis
$`\gamma _𝐤\{P_\mathrm{\Phi }^{AB^{}}P_\mathrm{\Phi }^{BA^{}},P_\mathrm{\Phi }^{AB^{}}\gamma ^{BA^{}},\gamma ^{AB^{}}P_\mathrm{\Phi }^{BA^{}},\gamma ^{AB^{}}\gamma ^{BA^{}}\}`$
with $`\gamma =\frac{1}{d^21}\left(\mathrm{𝟏}P_\mathrm{\Phi }\right)`$. In particular by the twirling procedure the Jamiołkowski fidelity remains the same, i.e. $`F(,U_{\text{SWAP}})=F(^{},U_{\text{SWAP}})`$.
#### IV.2.2 The noisy phase gate and CNOT gate
In this section we consider the unitary operation
$`U_{AB}(\alpha )=e^{i\alpha \sigma _y^A\sigma _y^B},`$ (68)
for arbitrary angles $`\alpha `$. Up to the local unitary operations, $`U(\alpha )`$ is equivalent to a controlled phase gate, while for $`\alpha =\pi /4`$, $`U(\alpha )`$ is equivalent to the CNOT gate (see Eq. (27)), i.e. $`U_{\mathrm{CNOT}}=U_1^AU_2^BU(\pi /4)V_1^AV_2^B`$.
We will obtain a standard form for noisy operations, given in the ideal case by $`U(\alpha )`$, by depolarizing the corresponding CPM $``$. The depolarization takes place by applying appropriate random local unitary operations that leave the state
$`|\mathrm{\Psi }_\alpha =\mathrm{cos}(\alpha )|\psi _0_𝑨|\psi _0_𝑩i\mathrm{sin}(\alpha )|\psi _2_𝑨|\psi _2_𝑩,`$ (69)
invariant (up to an irrelevant phase), where $`|\mathrm{\Psi }_\alpha \mathrm{\Psi }_\alpha |`$ is the state corresponding to $`U(\alpha )`$ via the Isomorphism, and
$`|\psi _j=\mathrm{𝟏}\sigma _j|\mathrm{\Phi }`$ (70)
are Bell states. Note that such a depolarization procedure for $`U(\alpha )`$ automatically provides a depolarization procedure for all operations that are local unitary equivalent to $`U(\alpha )`$, leading to a standard form with the same number of parameters for these noisy gates. The depolarization procedure simply has to be adopted according to the local unitary operations. To be specific, consider for instance the noisy $`U(\pi /4)`$ gate and the noisy CNOT gate. If $`W^{𝑨𝑩}`$ is a local unitary operation that keeps $`|\mathrm{\Psi }_{\pi /4}`$ invariant , then the operation
$`W^{𝑨𝑩}`$ $`=`$ $`U_1^A^{}U_2^B^{}(V_1^A)^T(V_2^B)^TW^{𝑨𝑩}`$
$`(U_1^A^{})^{}(U_2^B^{})^{}V_{}^{}{}_{1}{}^{A}V_{}^{}{}_{2}{}^{B}`$
keeps $`|\mathrm{\Psi }_{\mathrm{CNOT}}`$ invariant. That is, one obtains a depolarization procedure for the noisy CNOT gate from the depolarization procedure for the $`U(\pi /4)`$ gate by replacing each unitary operation $`W`$ by $`W^{}`$.
We now present an explicit depolarization procedure for the noisy $`U(\alpha )`$ gate, described by the CPM $``$, with arbitrary $`\alpha `$. We will consider the depolarization of the corresponding state by means of 4–local operations. We remark that any sequence of depolarization steps can be translated into a single step with multiple possibilities by considering all possible combinations. Such a single step procedure can then be translated to appropriate random operations applied to the system before and after the application of $``$ and hence to depolarize the corresponding map. For notational convenience, we define the four–qubit states
$`|\mathrm{\Psi }_{ij}^{AA^{}BB^{}}|\psi _i^{AA^{}}|\psi _j^{BB^{}}.`$ (71)
Given an arbitrary CPM $``$ specified by
$`\rho ={\displaystyle \underset{i,j,k,l=0}{\overset{3}{}}}\lambda _{ij,kl}\sigma _i\sigma _j\rho \sigma _k\sigma _l,`$ (72)
the corresponding state $`E`$ is given by
$`E={\displaystyle \underset{i,j,k,l=0}{\overset{3}{}}}E_{ij,kl}|\mathrm{\Psi }_{ij}\mathrm{\Psi }_{kl}|,`$ (73)
where $`\lambda _{ij,kl}=\lambda _{kl,ij}^{}`$. We define two–qubit unitary operations $`𝒰,\stackrel{~}{𝒰},𝒱`$ by
$`𝒰`$ $``$ $`(i\sigma _y)(i\sigma _y),`$
$`\stackrel{~}{𝒰}`$ $``$ $`\sigma _x\sigma _x,`$ (74)
$`𝒱`$ $``$ $`e^{i\pi /4\sigma _y}e^{i\pi /4\sigma _y}.`$
The action of these operations on Bell-basis states $`\{|\psi _j\}`$ can be readily obtained and one finds that $`𝒰,\stackrel{~}{𝒰}`$ introduce relative phases between the Bell states, while $`𝒱`$ exchanges two of them. To be specific, we have ,
$`𝒰\{|\psi _0,|\psi _1,|\psi _2,|\psi _3\}`$ $`=`$ $`\{|\psi _0,|\psi _1,|\psi _2,|\psi _3\},`$
$`\stackrel{~}{𝒰}\{|\psi _0,|\psi _1,|\psi _2,|\psi _3\}`$ $`=`$ $`\{|\psi _0,|\psi _1,|\psi _2,|\psi _3\},`$
$`𝒱\{|\psi _0,|\psi _1,|\psi _2,|\psi _3\}`$ $`=`$ $`\{|\psi _0,|\psi _3,|\psi _2,|\psi _1\}.`$ (75)
All local operations of the form $`\mathrm{𝟏}_𝑨\mathrm{𝟏}_𝑩`$, $`\mathrm{𝟏}_𝑨𝒰_𝑩`$, $`𝒰_𝑨\mathrm{𝟏}_𝑩`$, $`𝒰_𝑨𝒰_𝑩`$, $`\stackrel{~}{𝒰}_𝑨\stackrel{~}{𝒰}_𝑩`$, $`\mathrm{𝟏}_𝑨𝒱_𝑩`$, $`𝒱_𝑨\mathrm{𝟏}_𝑩`$, $`𝒱_𝑨𝒱_𝑩`$ keep the state $`|\mathrm{\Psi }_\alpha `$ (and the fidelity $`f=\mathrm{\Psi }_\alpha |E|\mathrm{\Psi }_\alpha `$ of the ideal operation) invariant and can thus be used for depolarization.
We decompose $``$ into the unitary part $`U(\alpha )`$ and the remaining noise part $`\stackrel{~}{}`$ (see Eq. (62)) and consider the corresponding (non–positive) operator $`\stackrel{~}{E}`$ (see Eq. (64)) in the following,
$`\stackrel{~}{E}={\displaystyle \underset{i,j,k,l=0}{\overset{3}{}}}\lambda _{ij,kl}|\mathrm{\Psi }_{ij}\mathrm{\Psi }_{kl}|.`$ (76)
We randomly apply $`\mathrm{𝟏}_𝑨\mathrm{𝟏}_𝑩`$, $`\mathrm{𝟏}_𝑨𝒰_𝑩`$, $`𝒰_𝑨\mathrm{𝟏}_𝑩`$ or $`𝒰_𝑨𝒰_𝑩`$, each with probability $`1/4`$, which leads to an operator
$`\stackrel{~}{E}^{}={\displaystyle \frac{1}{4}}\left(\stackrel{~}{E}+U^𝑩\stackrel{~}{E}\left(U^𝑩\right)^{}+U^𝑨\stackrel{~}{E}\left(U^𝑨\right)^{}+U^𝑨U^𝑩\stackrel{~}{E}\left(U^𝑨U^𝑩\right)^{}\right).`$
One finds that $`\stackrel{~}{E}^{}`$ is of block–diagonal form with coefficients $`\lambda _{ij,kl}^{}`$, that fulfill $`\lambda _{ij,kl}^{}=0`$ whenever $`(imod2)(kmod2)`$ or $`(jmod2)(lmod2)`$ and remain invariant otherwise. This follows from Eq. (75), as $`𝒰`$ introduces a phase $`(1)`$ for Bell states $`|\psi _i`$ with $`(imod2=1)`$ while states with even parity ($`imod2=0`$) remain invariant, which results in the cancellation of the corresponding off–diagonal elements. Thus only elements $`\lambda _{ij,kl}^{}`$ with $`(imod2)=(kmod2)`$ and $`(jmod2)=(lmod2)`$ remain, which can be grouped into four $`4\times 4`$ blocks $`\mathrm{\Gamma }_{ab}`$ with $`a=(imod2)=(kmod2),b=(jmod2)=(lmod2)`$. For instance, $`\mathrm{\Gamma }_{01}=_{i,k\{0,2\};j,l\{1,3\}}\lambda _{ij,kl}|\mathrm{\Psi }_{ij}\mathrm{\Psi }_{kl}|`$.
In the following, we consider the depolarization of the subspaces $`\mathrm{\Gamma }_{ab}`$ separately. We start with $`\mathrm{\Gamma }_{00}`$, which is spanned by the states $`\{|\mathrm{\Psi }_{00},|\mathrm{\Psi }_{02},|\mathrm{\Psi }_{20},|\mathrm{\Psi }_{22}\}`$. By randomly applying $`\mathrm{𝟏}_𝑨\mathrm{𝟏}_𝑩`$ or $`\stackrel{~}{𝒰}_𝑨\stackrel{~}{𝒰}_𝑩`$ with probability $`1/2`$, we find that the resulting operator $`\mathrm{\Gamma }_{00}^{}`$ has coefficients $`\lambda _{00,02}^{}=\lambda _{00,20}^{}=\lambda _{02,22}^{}=0`$, while the other coefficients remain invariant, i.e.
$`\lambda _{00,00}^{}=\lambda _{00,00};\lambda _{02,02}^{}=\lambda _{02,02};\lambda _{20,20}^{}=\lambda _{20,20};`$
$`\lambda _{22,22}^{}=\lambda _{22,22};\lambda _{00,22}^{}=\lambda _{00,22};\lambda _{02,20}^{}=\lambda _{02,20}.`$ (77)
We thus find that $`\mathrm{\Gamma }_{00}`$ is of the form
$`\mathrm{\Gamma }_{00}^{}=\left(\begin{array}{cccc}\lambda _{00,00}^{}& 0& 0& \lambda _{00,22}^{}\\ 0& \lambda _{02,02}^{}& \lambda _{02,20}^{}& 0\\ 0& \lambda _{02,20}^{}& \lambda _{20,20}^{}& 0\\ \lambda _{}^{}{}_{00,22}{}^{}& 0& 0& \lambda _{22,22}^{}\end{array}\right),`$ (82)
which are 8 independent real parameters as $`\lambda _{ij,kl}=\lambda _{kl,ij}^{}`$.
The effect of these (random) operations on the other subspaces $`\mathrm{\Gamma }_{01},\mathrm{\Gamma }_{10},\mathrm{\Gamma }_{11}`$ is similar, i.e. the corresponding off–diagonal term vanish. However, in these subspaces further depolarization is possible. Consider $`\mathrm{\Gamma }_{01}`$ which is spanned by the states $`\{|\mathrm{\Psi }_{01},|\mathrm{\Psi }_{03},|\mathrm{\Psi }_{21},|\mathrm{\Psi }_{23}\}`$. Applying randomly $`\mathrm{𝟏}_𝑨\mathrm{𝟏}_𝑩`$ or $`\mathrm{𝟏}_𝑨𝒱_𝑩`$ with probability $`1/2`$ leads to coefficients
$`\lambda _{01,01}^{}`$ $`=`$ $`\lambda _{03,03}^{}={\displaystyle \frac{1}{2}}(\lambda _{01,01}+\lambda _{03,03})`$
$`\lambda _{21,21}^{}`$ $`=`$ $`\lambda _{23,23}^{}={\displaystyle \frac{1}{2}}(\lambda _{21,21}+\lambda _{23,23})`$ (83)
$`\lambda _{01,23}^{}`$ $`=`$ $`\lambda _{03,21}^{}={\displaystyle \frac{1}{2}}(\lambda _{01,23}\lambda _{03,21})`$
This can readily be seen by using that
$`\mathrm{𝟏}_𝑨𝒱_𝑩\{|\mathrm{\Psi }_{01},|\mathrm{\Psi }_{03},|\mathrm{\Psi }_{21},|\mathrm{\Psi }_{23}\}`$
$`=\{|\mathrm{\Psi }_{03},|\mathrm{\Psi }_{01},|\mathrm{\Psi }_{23},|\mathrm{\Psi }_{21}\}.`$ (84)
Thus we find that $`\mathrm{\Gamma }_{01}`$ is of the form
$`\mathrm{\Gamma }_{01}^{}=\left(\begin{array}{cccc}\lambda _{01,01}^{}& 0& 0& \lambda _{01,23}^{}\\ 0& \lambda _{01,01}^{}& \lambda _{01,23}^{}& 0\\ 0& \lambda _{}^{}{}_{01,23}{}^{}& \lambda _{21,21}^{}& 0\\ \lambda _{}^{}{}_{01,23}{}^{}& 0& 0& \lambda _{21,21}^{}\end{array}\right),`$ (89)
and is thus described by 4 independent, real parameters ($`\lambda _{01,01}^{}`$ and $`\lambda _{21,21}^{}`$ are real, $`\lambda _{01,23}^{}`$ is complex).
Similarly, by randomly applying $`\mathrm{𝟏}_𝑨\mathrm{𝟏}_𝑩`$ or $`𝒱_𝑨\mathrm{𝟏}_𝑩`$ with probability $`1/2`$, one depolarizes the subspace $`\mathrm{\Gamma }_{10}`$ —spanned by the states $`\{|\mathrm{\Psi }_{10},|\mathrm{\Psi }_{30},|\mathrm{\Psi }_{12},|\mathrm{\Psi }_{32}\}`$— to the form
$`\mathrm{\Gamma }_{10}^{}=\left(\begin{array}{cccc}\lambda _{10,10}^{}& 0& 0& \lambda _{10,32}^{}\\ 0& \lambda _{10,10}^{}& \lambda _{10,32}^{}& 0\\ 0& \lambda _{}^{}{}_{10,32}{}^{}& \lambda _{12,12}^{}& 0\\ \lambda _{}^{}{}_{10,32}{}^{}& 0& 0& \lambda _{12,12}^{}\end{array}\right),`$ (94)
where
$`\lambda _{10,10}^{}`$ $`=`$ $`\lambda _{30,30}^{}={\displaystyle \frac{1}{2}}(\lambda _{10,10}+\lambda _{30,30})`$
$`\lambda _{12,12}^{}`$ $`=`$ $`\lambda _{32,32}^{}={\displaystyle \frac{1}{2}}(\lambda _{12,12}+\lambda _{32,32})`$ (95)
$`\lambda _{10,32}^{}`$ $`=`$ $`\lambda _{30,12}^{}={\displaystyle \frac{1}{2}}(\lambda _{10,32}\lambda _{30,12}),`$
which is again described by 4 independent, real parameters.
Finally, the subspace $`\mathrm{\Gamma }_{11}`$ —spanned by the states $`\{|\mathrm{\Psi }_{11},|\mathrm{\Psi }_{13},|\mathrm{\Psi }_{31},|\mathrm{\Psi }_{33}\}`$— can be further depolarized by randomly applying one of the operations $`\mathrm{𝟏}_𝑨\mathrm{𝟏}_𝑩`$, $`\mathrm{𝟏}_𝑨𝒱_𝑩`$, $`𝒱_𝑨\mathrm{𝟏}_𝑩`$ or $`𝒱_𝑨𝒱_𝑩`$ with probability $`1/4`$. One finds that
$`\lambda _{11,11}^{}`$ $`=`$ $`\lambda _{11,11}^{}=\lambda _{13,13}^{}=\lambda _{31,31}^{}=\lambda _{33,33}^{}=`$
$`=`$ $`{\displaystyle \frac{1}{4}}(\lambda _{11,11}+\lambda _{13,13}+\lambda _{31,31}+\lambda _{33,33})`$
$`\lambda _{11,33}^{}`$ $`=`$ $`\lambda _{13,31}^{}={\displaystyle \frac{1}{2}}\left(\mathrm{}(\lambda _{11,33})\mathrm{}(\lambda _{13,31})\right)`$ (96)
where $`\mathrm{}(x)`$ denotes the real part of $`x`$. Thus $`\mathrm{\Gamma }_{11}`$ is described by 2 independent, real parameters and is of the form
$`\mathrm{\Gamma }_{11}^{}=\left(\begin{array}{cccc}\lambda _{11,11}^{}& 0& 0& \lambda _{11,33}^{}\\ 0& \lambda _{11,11}^{}& \lambda _{11,33}^{}& 0\\ 0& \lambda _{11,33}^{}& \lambda _{11,11}^{}& 0\\ \lambda _{11,33}^{}& 0& 0& \lambda _{11,11}^{}\end{array}\right),`$ (101)
We remark that the depolarization process described in this final step leaves the subspaces $`\mathrm{\Gamma }_{00},\mathrm{\Gamma }_{01},\mathrm{\Gamma }_{10}`$ —which were already depolarized earlier— invariant. The final depolarized CPM $`\stackrel{~}{}_S`$ is specified by $`(8+4+4+21)=17`$ real parameters (where the (-1) results from the normalization condition $`\text{tr}(\stackrel{~}{E})=1`$) and is of Block–diagonal form. The coefficients $`\lambda _{ij,kl}^{}`$ are given by Eqs. IV.2.2,IV.2.2,IV.2.2,96 and are zero otherwise. This leads to the standard form,
$`\rho =fU(\alpha )\rho U(\alpha )^{}+(1f){\displaystyle \underset{ij,kl}{}}\lambda _{ij,kl}^{}\sigma _i\sigma _j\rho \sigma _k\sigma _l,`$ (102)
where $`f^{}=\mathrm{\Psi }_\alpha |E|\mathrm{\Psi }_\alpha =\mathrm{\Psi }_\alpha |E|\mathrm{\Psi }_\alpha `$, i.e. the fidelity of the ideal operation remains invariant. To summarize, we can achieve the following standard form:
Standard form for the Phase Gate
By uniformly choosing one of the unitaries $`U_k`$ from $`𝒰_1𝒰_2𝒰_3`$, where
$`𝒰_1`$ $`=`$ $`\{\mathrm{𝟏}_A\mathrm{𝟏}_B,e^{i\frac{\pi }{4}\sigma _y^A}\mathrm{𝟏}_B,\mathbf{\hspace{0.17em}1}_Ae^{i\frac{\pi }{4}\sigma _y^B},e^{i\frac{\pi }{4}\sigma _y^A}e^{i\frac{\pi }{4}\sigma _y^B}\}`$
$`𝒰_2`$ $`=`$ $`\{\mathrm{𝟏}_A\mathrm{𝟏}_B,\sigma _x^A\sigma _x^B\}`$ (103)
$`𝒰_3`$ $`=`$ $`\{\mathrm{𝟏}_A\mathrm{𝟏}_B,\sigma _y^A\mathrm{𝟏}_B,\mathbf{\hspace{0.17em}1}_A\sigma _y^B,\sigma _y^A\sigma _y^B\},`$
and applying $`U_k^{}`$ before and $`U_k`$ after the application of the noisy phase gate $``$ the resulting CPM $`^{}`$ is of the standard form $`^{}(\rho )=_{i,j,k,l=0}^3E_{ij,kl}^{}\sigma _i\sigma _j\rho \sigma _k\sigma _l`$ with
$`E^{}=\left(\begin{array}{cccc}\mathrm{\Gamma }_{00}^{}& 0& 0& 0\\ 0& \mathrm{\Gamma }_{01}^{}& 0& 0\\ 0& 0& \mathrm{\Gamma }_{10}^{}& 0\\ 0& 0& 0& \mathrm{\Gamma }_{11}^{}\end{array}\right)\mathrm{where}`$ (108)
$`\mathrm{\Gamma }_{00}^{}=\left(\begin{array}{cccc}a& 0& 0& u\\ 0& b& v& 0\\ 0& v^{}& \stackrel{~}{b}& 0\\ u^{}& 0& 0& \stackrel{~}{a}\end{array}\right)`$ , $`\mathrm{\Gamma }_{01}^{}=\left(\begin{array}{cccc}c& 0& 0& w\\ 0& c& w& 0\\ 0& w^{}& \stackrel{~}{c}& 0\\ w^{}& 0& 0& \stackrel{~}{c}\end{array}\right),`$ (117)
$`\mathrm{\Gamma }_{10}^{}=\left(\begin{array}{cccc}d& 0& 0& x\\ 0& d& x& 0\\ 0& x^{}& \stackrel{~}{d}& 0\\ x^{}& 0& 0& \stackrel{~}{d}\end{array}\right)`$ , $`\mathrm{\Gamma }_{11}^{}=\left(\begin{array}{cccc}e& 0& 0& \stackrel{~}{e}\\ 0& e& \stackrel{~}{e}& 0\\ 0& \stackrel{~}{e}& e& 0\\ \stackrel{~}{e}& 0& 0& e\end{array}\right)`$ (126)
with the following choice of basis
$``$ $`=`$ $`\{|\mathrm{\Psi }_{00},|\mathrm{\Psi }_{02},|\mathrm{\Psi }_{20},|\mathrm{\Psi }_{22},`$ (127)
$`|\mathrm{\Psi }_{01},|\mathrm{\Psi }_{03},|\mathrm{\Psi }_{21},|\mathrm{\Psi }_{23},`$
$`|\mathrm{\Psi }_{10},|\mathrm{\Psi }_{30},|\mathrm{\Psi }_{12},|\mathrm{\Psi }_{32},`$
$`|\mathrm{\Psi }_{11},|\mathrm{\Psi }_{13},|\mathrm{\Psi }_{31},|\mathrm{\Psi }_{33}\}`$
and the parameters $`a,\stackrel{~}{a},b,\stackrel{~}{b},c,\stackrel{~}{c},d,\stackrel{~}{d},e,\stackrel{~}{e}`$ and $`u,v,w,x`$. This depolarization does not increase the noise level, i.e. $`f^{}=f`$.
#### IV.2.3 The CNOT–type gate
For certain values of $`\alpha `$, further depolarization is possible. In particular, we consider $`\alpha =\pi /4`$, i.e. the operations $`U(\pi /4)`$ which is local unitary equivalent to the CNOT gate. In this case, the state $`|\mathrm{\Psi }_{\pi /4}`$ is a maximally entangled state (with respect to systems $`𝑨,𝑩`$), which remains invariant under a larger set of local unitary operations than a non–maximally entangled state $`|\mathrm{\Psi }_\alpha `$. In particular, we consider the unitary operations
$`𝒲`$ $``$ $`\mathrm{𝟏}(i\sigma _y),`$ (128)
$`\stackrel{~}{𝒲}`$ $``$ $`\sigma _z\sigma _x.`$ (129)
which act on Bell states as follows
$`𝒲\{|\psi _0,|\psi _1,|\psi _2,|\psi _3\}`$ $`=`$ $`\{i|\psi _2,|\psi _3,i|\psi _0,|\psi _1\},`$
$`\stackrel{~}{𝒲}\{|\psi _0,|\psi _1,|\psi _2,|\psi _3\}`$ $`=`$ $`\{i|\psi _2,|\psi _3,i|\psi _0,|\psi _1\}.`$
The operation $`𝒲_𝑨\stackrel{~}{𝒲}_𝑩`$ leaves the state $`|\mathrm{\Psi }_{\pi /4}`$ —up to an irrelevant global phase factor $`(i)`$— invariant. Note that this is not true for $`|\mathrm{\Psi }_\alpha `$ with $`\alpha \pi /4`$. We take the standard form Eq. (102) as initial map, and apply randomly either $`\mathrm{𝟏}_𝑨\mathrm{𝟏}_𝑩`$ or $`𝒲_𝑨\stackrel{~}{𝒲}_𝑩`$. One finds that the resulting operator $`\stackrel{~}{E}^{\prime \prime }`$ is significantly simplified and is described by 8 independent, real parameters. We denote the coefficients of $`\stackrel{~}{E}^{\prime \prime }`$ by $`\mu _{ij,kl}`$.
To be specific, for $`\mathrm{\Gamma }_{00}^{\prime \prime }`$ we find
$`\mu _{00,00}=\mu _{22,22}={\displaystyle \frac{1}{2}}(\lambda _{00,00}^{}+\lambda _{22,22}^{});`$
$`\mu _{02,02}=\mu _{20,20}={\displaystyle \frac{1}{2}}(\lambda _{02,02}^{}+\lambda _{20,20}^{});`$ (130)
$`\mu _{00,22}=i\mathrm{}(\lambda _{00,22}^{});\mu _{02,20}=i\mathrm{}(\lambda _{02,20}^{}),`$
where $`\mathrm{}(x)`$ denotes the imaginary part of $`x`$, i.e. $`i\mathrm{}(x)=(xx^{})/2`$ and we thus have 4 real parameters. This follows from $`𝒲_𝑨\stackrel{~}{𝒲}_𝑩\{|\mathrm{\Psi }_{00},|\mathrm{\Psi }_{02},|\mathrm{\Psi }_{20},|\mathrm{\Psi }_{22}\}=\{|\mathrm{\Psi }_{22},|\mathrm{\Psi }_{20},|\mathrm{\Psi }_{02},|\mathrm{\Psi }_{00}\}`$.
Similarly, for $`\mathrm{\Gamma }_{01}^{\prime \prime }`$ we find
$`\mu _{01,01}`$ $`=`$ $`\mu _{03,03}=\mu _{21,21}=\mu _{23,23}={\displaystyle \frac{1}{2}}(\lambda _{01,01}^{}+\lambda _{21,21}^{});`$
$`\mu _{01,23}`$ $`=`$ $`\mu _{03,21}=\mathrm{}(\lambda _{01,23}^{}),`$ (131)
while $`\mathrm{\Gamma }_{10}^{\prime \prime }`$ simplifies to
$`\mu _{10,10}`$ $`=`$ $`\mu _{30,30}=\mu _{12,12}=\mu _{32,32}={\displaystyle \frac{1}{2}}(\lambda _{10,10}^{}+\lambda _{12,12}^{});`$
$`\mu _{10,32}`$ $`=`$ $`\mu _{30,12}=\mathrm{}(\lambda _{10,32}^{}),`$ (132)
where we have 2 real parameters in each case.
Finally, for $`\mathrm{\Gamma }_{11}^{\prime \prime }`$ we have
$`\mu _{11,11}`$ $`=`$ $`\mu _{13,13}=\mu _{31,31}=\mu _{33,33}=\lambda _{11,11}^{};`$
$`\mu _{11,33}`$ $`=`$ $`\mu _{13,31}=0,`$ (133)
which is a single, real parameter. It follows that the standard form for the depolarized gate $`U(\pi /4)`$ is given by
$`^{\prime \prime }\rho =fU({\displaystyle \frac{\pi }{4}})\rho U({\displaystyle \frac{\pi }{4}})^{}+(1f){\displaystyle \underset{ij,kl}{}}\stackrel{~}{\mu }_{ij,kl}\sigma _i\sigma _j\rho \sigma _k\sigma _l,`$ (134)
where the coefficients $`\stackrel{~}{\mu }_{ij,kl}`$ are defined in Eqs. (IV.2.3), (IV.2.3), (IV.2.3), (IV.2.3) and are zero otherwise. Note that the fidelity of the ideal operation $`U(\pi /4)`$ remains invariant.
Standard form for the CNOT–type Gate
The total state $`\stackrel{~}{E}`$ is thus of the form
$`\left(\begin{array}{cccccccccccccccc}a& 0& 0& iu& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0\\ 0& b& iv& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0\\ 0& iv& b& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0\\ iu& 0& 0& a& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& c& 0& 0& w& 0& 0& 0& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& c& w& 0& 0& 0& 0& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& w& c& 0& 0& 0& 0& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& w& 0& 0& c& 0& 0& 0& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& 0& 0& d& 0& 0& x& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& 0& 0& 0& d& x& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& 0& 0& 0& x& d& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& 0& 0& x& 0& 0& d& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& e& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& e& 0& 0\\ 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& e& 0\\ 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& e\end{array}\right)`$ (151)
or equivalently
$`\stackrel{~}{E}=\left(\begin{array}{cccc}\mathrm{\Gamma }_{00}& 0& 0& 0\\ 0& \mathrm{\Gamma }_{01}& 0& 0\\ 0& 0& \mathrm{\Gamma }_{10}& 0\\ 0& 0& 0& \mathrm{\Gamma }_{11}\end{array}\right)\mathrm{with}`$ (156)
$`\mathrm{\Gamma }_{00}=\left(\begin{array}{cccc}a& 0& 0& iu\\ 0& b& iv& 0\\ 0& iv& b& 0\\ iu& 0& 0& a\end{array}\right)`$ , $`\mathrm{\Gamma }_{01}=\left(\begin{array}{cccc}c& 0& 0& w\\ 0& c& w& 0\\ 0& w& c& 0\\ w& 0& 0& c\end{array}\right),`$ (165)
$`\mathrm{\Gamma }_{10}=\left(\begin{array}{cccc}d& 0& 0& x\\ 0& d& x& 0\\ 0& x& d& 0\\ x& 0& 0& d\end{array}\right)`$ , $`\mathrm{\Gamma }_{11}=\left(\begin{array}{cccc}e& 0& 0& 0\\ 0& e& 0& 0\\ 0& 0& e& 0\\ 0& 0& 0& e\end{array}\right),`$ (174)
where we use the basis $``$ in Eq. (127) and $`a=\mu _{00,00}`$, $`b=\mu _{02,02}`$ etc. are all real parameters.
### IV.3 Standard forms by means of sacrificing
While the standard forms for the general $`U(\alpha )`$ gate or the CNOT–type gate $`U(\pi /4)`$ are already relatively simple (as the number of relevant parameters is significantly reduced, namely from 255 to 17 or 8 respectively), for many practical applications a further simplification might still be desirable. If, for instance, one would like to analyze error thresholds for processes involving several particles and/or operations (as is e.g. the case in fault tolerant quantum computation or entanglement purification with imperfect means) where noisy operations are described by these standard forms, the corresponding CPMs are still rather complex.
In this section we will provide such a further simplification of the corresponding noise process, where we find that in many relevant cases a single parameter is sufficient and the noise process can be described by (global) white noise. In contrast to the previous depolarization procedure, here the exact form of the noise process (equivalently the corresponding state $`E`$) has to be known. This may e.g. achieved by performing a process tomography of the CPM resulting after the universal depolarization protocol describe in the previous section (note that only the knowledge of the depolarized map is required). In addition, the fidelity of the ideal operation is no longer conserved but decreased by a certain amount. That is, by ”sacrificing“ a (small) amount of the fidelity of the operation, one can modify the resulting noise process in such a way that one obtains a very simple standard form. This is done by transferring weight from the ideal operation to the noisy part in an appropriate way and hence tailor the noise process.
#### IV.3.1 The noisy SWAP gate
Let us consider a noisy SWAP gate in the two qubit case. For a specific noisy SWAP operation with sufficiently large Jamiołkowski fidelity $`f=F(,U_{\text{SWAP}})=\psi _{\text{SWAP}}|E|\psi _{\text{SWAP}}>\frac{15}{16}`$ a depolarization procedure can be designed that brings the noisy operation $``$ to the standard form
$`^{}(\rho )=f^{}U_{\text{SWAP}}\rho U_{\text{SWAP}}^{}+(1f^{}){\displaystyle \frac{1}{16}}\mathrm{𝟏}_{AB}`$ (175)
with $`f^{}>f/3`$. Thus the noise in the standard form corresponds to white noise, where the noise level $`(1f^{})`$ is at most increased by a factor of $`5.5`$. As in Sec. IV.2.1 this immediately follows from the results for the case, where the ideal operation is the identity, by simply applying the designed depolarization procedure from the end of the Sec. IV.1 with the role of particles $`A^{}`$ and $`B^{}`$ exchanged. Note that the corresponding twirling procedure remains local w.r.t. the physical partitioning $`(AA^{},BB^{})`$.
#### IV.3.2 The noisy CNOT–type gate
We consider now the noisy CNOT–type gate $`U(\pi /4)`$, described by the standard form given in Eq. (134). We will further depolarize the corresponding noise process in such a way that the fidelity of the ideal operation is decreased (as few as possible) and the noise is global white noise, i.e. the simplified standard form is given by
$`^{}\rho `$ $`=`$ $`\stackrel{~}{q}U(\pi /4)\rho U(\pi /4)^{}+(1\stackrel{~}{q}){\displaystyle \frac{1}{16}}{\displaystyle \underset{ij}{}}\sigma _i\sigma _j\rho \sigma _i\sigma _j`$ (176)
$`=`$ $`\stackrel{~}{q}U(\pi /4)\rho U(\pi /4)^{}+{\displaystyle \frac{1\stackrel{~}{q}}{16}}\mathrm{𝟏},`$
where the fidelity of the ideal operation $`\stackrel{~}{f}=\stackrel{~}{q}+(1\stackrel{~}{q})/16`$. We find that the amount of noise is increased at most by (approximately) an order of magnitude, i.e. $`(1\stackrel{~}{f})/(1f)20`$. Clearly, such a further depolarization is only useful if the fidelity of the ideal operation is initially sufficiently large, i.e. $`f0.96`$, as otherwise the completely depolarizing operation would be obtained.
We will first demonstrate that a depolarization to global white noise is possible , and will then discuss the resulting decrease of fidelity. The state $`E_S`$ can be written as
$`E_S=fE_{\pi /4}+(1f)\stackrel{~}{E},`$ (177)
where $`\stackrel{~}{E}`$ is the operator corresponding to the noise part of the CPM $`_S`$ (see Eqs. (134), (151)) and
$`E_{\pi /4}`$ $`=`$ $`|\mathrm{\Psi }_{\pi /4}\mathrm{\Psi }_{\pi /4}|={\displaystyle \frac{1}{2}}(|\mathrm{\Psi }_{00}\mathrm{\Psi }_{00}|+|\mathrm{\Psi }_{22}\mathrm{\Psi }_{22}|`$ (178)
$`+i|\mathrm{\Psi }_{00}\mathrm{\Psi }_{22}|i|\mathrm{\Psi }_{22}\mathrm{\Psi }_{00}|),`$
is the operator corresponding to $`U(\pi /4)`$.
We will first show that by means of local unitaries, one can change the off–diagonal elements of $`E_{\pi /4}`$ in such a way that each off–diagonal element in $`\stackrel{~}{E}`$ (or equivalently $`E`$) can be erased by probabilistically applying either the corresponding unitary operation or the identity with appropriate probability. Here, we are no longer restricted to operations that keep $`|\mathrm{\Psi }_{\pi /4}`$ invariant, but can use arbitrary local unitaries. We first note that by applying $`\mathrm{𝟏}_{AA^{}}\sigma _z^B\sigma _z^B^{}`$, one can change the sign of the off–diagonal elements $`|\mathrm{\Psi }_{00}\mathrm{\Psi }_{22}|`$ and $`|\mathrm{\Psi }_{02}\mathrm{\Psi }_{20}|`$, which implies that in the following discussion the sign of the off–diagonal elements do not play a role.
By the depolarization procedure in Sec. IV.2.3 we can assume the sub–block $`\mathrm{\Gamma }_{00}`$ —spanned by the states $`\{|\mathrm{\Psi }_{00},|\mathrm{\Psi }_{02},|\mathrm{\Psi }_{20},|\mathrm{\Psi }_{22}\}`$— of the (total) state $`E`$ to be in the form
$`\mathrm{\Gamma }_{00}=\left(\begin{array}{cccc}A& 0& 0& iY\\ 0& B& iX& 0\\ 0& iX& B& 0\\ iY& 0& 0& A\end{array}\right),`$ (183)
with
$`A=f/2+(1f)a,`$ $`B=(1f)b,`$ (184)
$`Y=f/2+(1f)u,`$ $`X=(1f)v,`$ (185)
where $`a=\mu _{00,00},b=\mu _{02,02},iu=\mu _{00,22},iv=\mu _{02,20}`$ (see Eqs. (134), (151)).
We consider the operations $`𝒰_x=(\mathrm{𝟏}\sigma _x)`$ and $`𝒰_z=(\mathrm{𝟏}\sigma _z)`$. We have that the action of these operations on Bell states is given by
$`𝒰_x\{|\psi _0,|\psi _1,|\psi _2,|\psi _3\}`$ $`=`$ $`\{|\psi _1,|\psi _0,i|\psi _3,i|\psi _2\},`$
$`𝒰_z\{|\psi _0,|\psi _1,|\psi _2,|\psi _3\}`$ $`=`$ $`\{|\psi _3,i|\psi _2,i|\psi _1,|\psi _0\},`$
It follows that applying with probability $`1/2`$ the operation $`\mathrm{𝟏}_𝑨𝒰_x^𝑩`$ or $`\mathrm{𝟏}_𝑨𝒰_z^𝑩`$ transforms the state $`E`$ to the state $`E^1`$. In particular, the subspace $`\mathrm{\Gamma }_{00}`$ is transformed to the subspace $`\mathrm{\Gamma }_{01}^1`$ —spanned by $`\{|\mathrm{\Psi }_{01},|\mathrm{\Psi }_{03},|\mathrm{\Psi }_{21},|\mathrm{\Psi }_{23}\}`$— and some coefficients are aligned. One finds that the resulting elements are given by
$`\mathrm{\Gamma }_{01}^1=\left(\begin{array}{cccc}C& 0& 0& Z\\ 0& C& Z& 0\\ 0& Z& C& 0\\ Z& 0& 0& C\end{array}\right),`$ (190)
where
$`C=(A+B)/2,`$ $`Z=(X+Y)/2,`$ (191)
Note that the element $`Z`$ is real. At the same time, the subspace $`\mathrm{\Gamma }_{01}`$ is transformed to $`\mathrm{\Gamma }_{00}^1`$ (where the off diagonal elements are given by $`i(1f)w`$ after the transformation, while the diagonal elements are still given by $`(1f)c`$). Also the subspaces $`\mathrm{\Gamma }_{10}`$ and $`\mathrm{\Gamma }_{11}`$ are transformed into each other, where one finds that $`\mathrm{\Gamma }_{10}^1`$ is diagonal with elements $`(1f)e`$, and also $`\mathrm{\Gamma }_{11}^1`$ is diagonal with elements $`(1f)d`$.
Similarly, if one applies the operations $`𝒰_x`$, $`𝒰_z`$ in $`𝑨`$ instead of $`𝑩`$, the state $`E`$ is transformed to a state $`E^2`$. In particular, the subspace $`\mathrm{\Gamma }_{00}`$ is transformed to the subspace $`\mathrm{\Gamma }_{10}^2`$ —spanned by $`\{|\mathrm{\Psi }_{10},|\mathrm{\Psi }_{30},|\mathrm{\Psi }_{12},|\mathrm{\Psi }_{32}\}`$—, where the elements of $`\mathrm{\Gamma }_{10}^2`$ are the same as of $`\mathrm{\Gamma }_{01}`$ (see Eq. (190)). The transformation of the other subspaces follows accordingly, only the role of systems $`𝑨`$ and $`𝑩`$ is exchanged (e.g. $`\mathrm{\Gamma }_{01}\mathrm{\Gamma }_{11}^2`$ etc.). Note that one can simultaneously change the sign of all off–diagonal elements of the resulting state $`E^2`$ by applying $`\mathrm{𝟏}_{AA^{}}\sigma _z^B\sigma _z^B^{}`$.
If one thus mixes the resulting states $`E`$ (with probability $`p_0`$), $`\stackrel{~}{E}^1`$ (with probability $`p_1`$) and $`E^2`$ (with probability $`p_2`$), —and by appropriately choosing the phases of the corresponding off diagonal elements— one can achieve that the final state $`E^{\prime \prime }`$ has no off diagonal elements in the subspaces $`\mathrm{\Gamma }_{01}^{\prime \prime }`$ and $`\mathrm{\Gamma }_{10}^{\prime \prime }`$. This is guaranteed by choosing
$`p_0={\displaystyle \frac{Z}{Z+(|w|+|x|)(1f)}},`$
$`p_1={\displaystyle \frac{|w|(1f)}{Z+(|w|+|x|)(1f)}},`$ (192)
$`p_2={\displaystyle \frac{|x|(1f)}{Z+(|w|+|x|)(1f)}},`$
The other coefficients of the resulting state can be readily calculated, taking into account whether a change of sign was necessary for $`E^1`$ or $`E^2`$, where we denote $`\sigma _1=\mathrm{sign}(w)+1,\sigma _2=\mathrm{sign}(x)+1`$ with $`(1)^{\sigma _1}=\mathrm{sign}(w)`$. One finds that each of the subspaces $`\mathrm{\Gamma }_{01}^{\prime \prime },\mathrm{\Gamma }_{10}^{\prime \prime },\mathrm{\Gamma }_{11}^{\prime \prime }`$ is diagonal (with all coefficients equal) and described by a coefficient, $`\gamma _{01},\gamma _{10},\gamma _{11}`$ respectively, where
$`\gamma _{01}`$ $`=`$ $`p_0\left(1f\right)c+\left(1\right)^{\sigma _1}p_1C+\left(1\right)^{\sigma _2}p_2\left(1f\right)e,`$
$`\gamma _{10}`$ $`=`$ $`p_0\left(1f\right)d+\left(1\right)^{\sigma _1}p_1\left(1f\right)e+\left(1\right)^{\sigma _2}p_2C,`$ (193)
$`\gamma _{11}`$ $`=`$ $`p_0\left(1f\right)e+\left(1\right)^{\sigma _1}p_1\left(1f\right)d+\left(1\right)^{\sigma _2}p_2\left(1f\right)c.`$
The subspace $`\mathrm{\Gamma }_{00}^{\prime \prime }`$ is given by
$`p_0\mathrm{\Gamma }_{00}+(1)^{\sigma _1}p_1\mathrm{\Gamma }_{00}^1+(1)^{\sigma _2}p_2\mathrm{\Gamma }_{00}^2,`$ (194)
where we find that resulting diagonal elements are
$`A^{\prime \prime }=p_0A+\left(1\right)^{\sigma _1}p_1\left(1f\right)c+\left(1\right)^{\sigma _2}p_2\left(1f\right)d,`$
$`B^{\prime \prime }=p_0B+\left(1\right)^{\sigma _1}p_1\left(1f\right)c+\left(1\right)^{\sigma _2}p_2\left(1f\right)d,`$ (195)
while the off diagonal elements are given by
$`iY^{\prime \prime }=p_0iY+\left(1\right)^{\sigma _1}ip_1\left(1f\right)w+\left(1\right)^{\sigma _2}ip_2\left(1f\right)x,`$
$`iX^{\prime \prime }=p_0iX+\left(1\right)^{\sigma _1}ip_1\left(1f\right)w+\left(1\right)^{\sigma _2}ip_2\left(1f\right)x.`$ (196)
It remains to show that one can erase the off–diagonal element $`|\mathrm{\Psi }_{02}\mathrm{\Psi }_{20}|`$, $`i\stackrel{~}{X}`$. This can be accomplished by randomly applying the operation $`\mathrm{𝟏}_𝑨\mathrm{𝟏}_𝑩`$ or $`𝒲^𝑨\mathrm{𝟏}_𝑩`$ (see Eq. (128)) with probabilities $`p`$ and $`(1p)`$. If $`\mathrm{sign}(\stackrel{~}{X})=\mathrm{sign}(\stackrel{~}{Y})`$, one applies in a addition $`\sigma _z^B\sigma _z^B^{}`$ in the second case in order to change the sign of the corresponding off–diagonal elements. Choosing
$`p=(|X^{\prime \prime }|+|Y^{\prime \prime }|)^1,`$ (197)
ensures that the off–diagonal element $`|\mathrm{\Psi }_{02}\mathrm{\Psi }_{20}|`$ vanishes, while $`|\mathrm{\Psi }_{00}\mathrm{\Psi }_{22}|`$ becomes
$`Y^{}=ipY^{\prime \prime }+(1)^{\mathrm{sign}X^{\prime \prime }+1}i(1p)X^{\prime \prime }`$ (198)
and the diagonal elements become
$`A^{}=pA^{\prime \prime }+(1p)B^{\prime \prime },`$ (199)
$`B^{}=pB^{\prime \prime }+(1p)A^{\prime \prime }.`$ (200)
Note that all other elements of $`\stackrel{~}{E}^{}`$ remain invariant. Finally, the remaining off–diagonal element $`|\mathrm{\Psi }_{00}\mathrm{\Psi }_{22}|`$, $`i\stackrel{~}{Y}^{}`$, which corresponds in part to the ideal operation and in part to the noise part, can be formally incorporated into the ideal part of the evolution, i.e. the resulting state can be formally rewritten as
$`E_f=f^{}|\mathrm{\Psi }_{\pi /4}\mathrm{\Psi }_{\pi /4}|+(1f^{})D,`$ (201)
where $`D`$ is diagonal in the basis $`\{|\mathrm{\Psi }_{ij}\}`$ with elements $`d_{ij}`$ and $`f^{}=2Y^{}`$. The elements $`d_{ij}`$ of $`D`$ in the blocks $`\mathrm{\Gamma }_{01},\mathrm{\Gamma }_{10},\mathrm{\Gamma }_{11}`$ are given by Eq. (IV.3.2) (i.e. $`d_{01}=d_{03}=d_{21}=d_{23}=\gamma _{01}`$ etc.), while in the block $`\mathrm{\Gamma }_{00}`$ we have
$`d_{02}`$ $`=`$ $`d_{20}=B^{},`$ (202)
$`d_{00}`$ $`=`$ $`d_{22}=A^{}Y^{}.`$ (203)
The coefficients $`d_{ij}`$ can even be made equal by further reducing the fidelity of the ideal operation. In this case, $`D`$ corresponds to the completely mixed state, and the corresponding map is given by global white noise. This is done as follows: Using that the probabilistic operation $`\stackrel{~}{𝒰}^𝑨\mathrm{𝟏}_𝑩`$ or $`\mathrm{𝟏}_𝑨\mathrm{𝟏}_𝑩`$, one produces a state diagonal in the basis $`\{|\mathrm{\Psi }_{ij}\}`$ with the same diagonal coefficients as $`E_f`$, where we consider the situation where $`d_{00}=d_{22}d_{ij}`$, i.e. the fidelity of the ideal operation is sufficiently large. In this sense, weight from the ideal operation can be transferred to the other states. In particular, one uses $`\mathrm{𝟏}_A𝒲^B`$ to increase weight in $`|\mathrm{\Psi }_{02}\mathrm{\Psi }_{02}|`$ and $`|\mathrm{\Psi }_{20}\mathrm{\Psi }_{20}|`$; $`\mathrm{𝟏}_𝑨𝒰_x^𝑩`$ and $`\mathrm{𝟏}_𝑨𝒰_z^𝑩`$ to increase weight of $`\mathrm{\Gamma }_{01}`$; $`𝒰_x^𝑨\mathrm{𝟏}_𝑩`$ and $`𝒰_z^𝑨\mathrm{𝟏}_𝑩`$ to increase weight of $`\mathrm{\Gamma }_{10}`$; $`𝒰_x^𝑨𝒰_x^𝑩`$, $`𝒰_z^𝑨𝒰_z^𝑨`$, $`𝒰_x^𝑨𝒰_z^𝑩`$ and $`𝒰_z^𝑨𝒰_x^𝑩`$ to increase weight in $`\mathrm{\Gamma }_{11}`$.
One thus ends up with a standard form described by global white noise as announced (see Eq. (176)). We will evaluate the loss factor for the fidelity for an alternative protocol discussed in the next section. This protocol is capable of depolarizing also noisy phase gates to a one–parameter standard form.
#### IV.3.3 The noisy phase gate
In principle, it may be possible to obtain a simplified standard form for the gate $`U(\alpha )`$ with arbitrary $`\alpha `$ by similar means as in the case of the noisy CNOT–type gate $`U(\pi /4)`$, i.e. by manipulating the noisy evolution in such a way that weight is transferred from the ideal evolution to the appropriate noise parts. However, for small $`\alpha `$ one encounters a difficulty which may be hard to circumvent. When using parts of the operator corresponding to the ideal evolution to eliminate off diagonal elements in other parts of the density matrix corresponding to the noisy evolution, we have that automatically also the diagonal elements are transferred and hence the noise part is further increased. While in the case of $`U(\pi /4)`$, the increase of diagonal elements of the noise part is of the same order of magnitude as the off–diagonal elements, for small $`\alpha `$ this is no longer the case. We have that the off diagonal element of the ideal operations, $`E_\alpha `$, is given by $`\lambda _{00,22}=i\mathrm{cos}(\alpha )\mathrm{sin}(\alpha )`$, while the larger diagonal element, $`\lambda _{00,00}`$, is given by $`\mathrm{cos}^2(\alpha )`$. Imagine we have elements in the noise part of order $`ϵ1`$, where both diagonal and off diagonal terms of order $`ϵ`$ appear. If one wants to eliminate an off–diagonal element in the noise part which is of order $`ϵ`$, we need $`(1p)\mathrm{cos}(\alpha )\mathrm{sin}(\alpha )=|ϵ|`$, i.e. with probability $`(1p)`$ the off–diagonal element of the ideal evolution is transferred to the off–diagonal element of the noise part with the sign chosen in such a way that the total off–diagonal element in the noise part vanishes. However, by doing such a transformation, one of the diagonal elements of the noise part is automatically increased by $`(1p)\mathrm{cos}^2(\alpha )`$ which is of the order $`|ϵ|/\mathrm{tan}(\alpha )`$. Note that for small $`\alpha `$, $`1/\mathrm{tan}(\alpha )1`$, i.e. the amount of noise may be increased by orders of magnitude. This is clearly not acceptable, as our goal was to obtain a simplified standard form by sacrificing a relatively small amount of the fidelity and not to decrease the fidelity by (several) orders of magnitude.
However, under certain conditions one may use an alternative method which still allows one to obtain a standard form corresponding to global white noise, specified by a single parameter. In particular, if one can switch the noisy operation on and off at will, i.e. one can decide whether one wants to apply the noisy operation or does not want to apply it (and instead apply something else), then such a depolarization is possible. If one considers the case where $`U(\alpha )`$ is some non–local gate, then it is natural to assume such a controllability. In this case, one can either apply the noisy evolution with certain probability $`p`$ or apply some other operation with probability $`(1p)`$. In particular, one can apply any separable operation. This may, however, involve the application of arbitrary local operations (including measurements), rather than the application of local unitaries as we have assumed so far. Considering the corresponding states, this amounts to mixing of the state $`E`$ with some separable (in the sense $`𝑨𝑩`$) state $`D`$. The separable operation $`𝒟`$ associated to $`D`$ can be implemented by some random application of local operations, $`𝒟=_ip_iA_iB_i^T\rho (A_iB_i^T)^{}`$. The corresponding Kraus operators can be obtained from the spectral decomposition of $`D`$ as indicated in Appendix A. Let us now consider a density matrix $`A`$ of a separable two–qubit state written in the standard basis. Then all states of the block diagonal form (see e.g. Eq. (108) with separable block matrices $`\mathrm{\Gamma }_{ij}=A`$ for $`i,j\{0,1\}`$ are again separable (recall that $`\mathrm{\Gamma }_{ij}`$ denotes subspaces spanned by $`|\mathrm{\Psi }_{kl}`$ with $`kmod2=i,lmod2=j)`$. In particular, any matrix $`A`$ of the form
$`A={\displaystyle \frac{1}{4}}\left(\begin{array}{cccc}1& 0& 0& \beta \\ 0& 1& \alpha & 0\\ 0& \alpha ^{}& 1& 0\\ \beta ^{}& 0& 0& 1\end{array}\right),`$ (208)
with $`\alpha ,\beta \{0,1,1,i,i\}`$ as well as any diagonal matrix $`A`$ (with positive coefficients summing up to one) is separable. This can be checked by calculating the partial transposition of these states, where one finds that the partial transposition is positive in all cases which is (for two–qubit states) sufficient to ensure separability Pe96 ; Ho96 . The corresponding separable maps can be implemented by a simple sequence of random local unitary operations (in the case of diagonal $`A`$), or measurements (in the case of matrices of the form 208). The Kraus representation of the state can be obtained as shown in Appendix A.
It is now straightforward to see that mixing $`E`$ with separable operators of the form $`\mathrm{\Gamma }_{ij}=A`$ with $`A`$ given by Eq. (208) or an appropriate diagonal matrix, allows one to eliminate all (unwanted) off–diagonal elements as well as to align all diagonal coefficients of the noise part. Thus the resulting simplified normal form of the noisy operation is given by
$`_S\rho `$ $`=`$ $`\stackrel{~}{q}U(\alpha )\rho U(\alpha )^{}+(1\stackrel{~}{q}){\displaystyle \frac{1}{16}}{\displaystyle \underset{ij}{}}\sigma _i\sigma _j\rho \sigma _i\sigma _j`$ (209)
$`=`$ $`\stackrel{~}{q}U(\alpha )\rho U(\alpha )^{}+{\displaystyle \frac{1\stackrel{~}{q}}{16}}\mathrm{𝟏},`$
The fidelity of the ideal operation $`\stackrel{~}{f}=\stackrel{~}{q}+(1\stackrel{~}{q})/16`$ is reduced, where we find e.g. for $`\alpha =\pi /4`$ that if $`f=1ϵ`$, then
$`\stackrel{~}{f}117ϵ.`$ (210)
That is, the fidelity of the operation is reduced by about an order of magnitude. This can be seen as follows. Consider for example the CNOT like gate $`U(\pi /4)`$ with corresponding standard form of noise $`\stackrel{~}{E}`$ given by Eq. (151). We have $`\text{tr}(\stackrel{~}{E})=2a+2b+4c+4d+4e=(1f)`$ and denote $`y=\mathrm{max}(a,b,c/2,d/2,e/2)`$, i.e. $`f12y`$. One can make $`\stackrel{~}{E}`$ diagonal by mixing with matrices $`\mathrm{\Gamma }_{ij}^{}=A`$. Consider for instance the case where all noise is concentrated in $`b,v`$ (this in fact turns out to correspond to a (non unique) worst case scenario), i.e. $`y=b`$. One mixes $`\stackrel{~}{E}`$ (with probability $`p`$) with a matrix of the form $`\mathrm{\Gamma }_{00}^{}=A`$ with $`\beta =0,\alpha =i`$ (with probability $`(1p)`$). The resulting matrix is diagonal for $`p=(4v+1)^1`$ and has diagonal elements $`b^{}=bp+(1p)/42y(4y+1)^1`$, $`a^{}=(1p)/4`$ and $`c^{}=d^{}=e^{}=0`$. Note that the worst case corresponds to $`v=b`$. By mixing the resulting matrix (with probability $`q`$) with a diagonal matrix (with probability $`1q`$), one can make all diagonal elements equal (which corresponds to white noise). We have $`q(4y+1)/(32y+1)`$. This leads to a total final fidelity $`\stackrel{~}{f}=pqff/(32y+1)f/(1716f)`$. For $`f=1ϵ`$, we thus have
$`\stackrel{~}{f}f/(1716f)117ϵ,`$ (211)
i.e. the fidelity is decreased by about an order of magnitude. Note that this formula also holds in the general scenario with arbitrary coefficients $`a,b,c,d,e,u,v,w,x`$. In the first step, the worst case is given when all off diagonal elements are maximal, $`u=a,v=b,w=c,x=d`$. In the second step (making all diagonal elements equal), it is clearly the worst case if all weight is concentrated in one element (e.g. $`b`$) and all others are zero, as one has to fill up the weights of the other (14) diagonal elements. Non–zero diagonal elements would require less mixing, leading to a larger final fidelity. Thus $`\stackrel{~}{f}117(1f)`$ is a conservative bound on the final fidelity, where in many situations one will end up with a much larger final fidelity of the depolarized noisy operation.
### IV.4 Standard form for arbitrary two–qubit unitary operation
Let us briefly discuss the possibility to generalize the results in Sec. IV.2 and Sec. IV.3 to the more general case of an arbitrary unitary operation as the ideal operation. Consider an arbitrary unitary $`U`$ (or even a class unitaries $`U_\alpha `$). Note that similar to Eq. (27) any unitary two-qubit gate can be represented uniquely as Kr00
$`U^{AB}=U_1^AU_2^Be^{i_{i=1}^3\mu _i\sigma _i^A\sigma _i^B}V_1^AV_2^B`$ (212)
with $`\frac{\pi }{4}\mu _1\mu _2|\mu _3|0`$ and some single-qubit unitaries $`U_i,V_i`$ ($`i=1,2`$). The main block of the depolarization procedures in Sec. IV.2 was to find a set of $`(A,A^{})`$–local unitaries, that leave the corresponding pure state $`|\mathrm{\Psi }_U`$ (or a class pure states $`|\mathrm{\Psi }_{U_\alpha }`$ ) invariant. However, in general, —apart from some special cases where e.g. all $`\mu _i`$ equal or $`\mu _2=\mu _3=0`$— the only local unitary operation that keeps the state invariant is given by the identity operation. Hence, in generic cases, a depolarization of the operation —under the condition that the fidelity of the operation remains invariant— is impossible following this approach. Thus no universal standard form for arbitrary two–qubit unitary operations can be obtained along these lines.
It seems more appropriate to follow the ideas of Sec. IV.3 and to specifically design such a standard form for a noisy unitary taking the given form of decoherence into account, and allow for an increase of noise. We do not deepen such a discussion at this point but rather refer to an alternative approach. Instead of regarding the noisy unitary gate as a CPM and allowing for manipulation before and after the application of this operation one might as well consider the dynamical evolution realizing this gate and allow for a manipulation of this evolution by several short intermediate pulses of local unitary control operation. Inspired by the results in HamSim it is shown in the following section that by this procedure it is possible to depolarize an arbitrary master equation (of two systems) to a standard form described by at most 17 parameters.
## V Standard form for noisy evolutions described by a master equation
In this section, we will consider the evolution of two qubits described by a master equation of Lindblad form, where the ideal evolution is given by an arbitrary two–qubit Hamiltonian. The results can be readily generalized to multi–qubit systems whose (noiseless) interaction is described by Hamiltonians that are sums of two–body Hamiltonians.
We will consider a continuous evolution $`\rho (t)=_t\rho (0)`$ of the system due to Markovian quantum dynamics starting at $`t=0`$ in the state $`\rho (0)`$. Thus the family of quantum operations $`_t`$ forms a Markovian semi-group Markovian , determined by some generator $`𝒵`$, that in the case of two qubits $`A`$ and $`B`$ and the convention $`\mathrm{}1`$ satisfies the following differential equation (master equation):
$`{\displaystyle \frac{}{t}}\rho `$ $`=𝒵\rho :=i\rho +i_\text{l}\rho +\rho \text{with}`$ (213)
$`\rho `$ $`:=`$ $`[H^{AB},\rho ],`$
$`_\text{l}\rho `$ $`:=`$ $`[H_\text{l}^{AB},\rho ]\text{and}`$
$`\rho `$ $`:=`$ $`{\displaystyle \underset{\genfrac{}{}{0.0pt}{}{𝐤,𝐥}{𝐤\mathrm{𝟎}𝐥\mathrm{𝟎}}}{}}L_{\mathrm{𝐤𝐥}}\left([\sigma _𝐤^{AB}\rho ,\sigma _𝐥^{AB}]+[\sigma _𝐤^{AB},\rho \sigma _𝐥^{AB}]\right).`$
We have separated the unitary evolution of the dynamics into one part $`H`$, which corresponds to the ideal unitary process in question, and into a ’Lamb shift’ $`H_\text{l}`$, which is some unitary dynamics, that is induced by the coupling between the system and its environment and therefore corresponds to noise. The more relevant influence of the decoherence process $``$ is however incorporated in the ’Liouvillian’, which is determined by the positive ’GKS’-matrix $`𝐋=(L_{\mathrm{𝐤𝐥}})`$ GKS . Note that the corresponding sum is over all multi–indices $`𝐤=(k_1,k_2)`$ and $`𝐥=(l_1,l_2)`$ with $`k_i,l_j=0,1,2,3`$ except $`𝐤=\mathrm{𝟎}:=(0,0)`$ or $`𝐥=\mathrm{𝟎}`$. Thus the totally mixed state $`\sigma _\mathrm{𝟎}^{AB}=\frac{1}{4}\mathrm{𝟏}_{AB}`$ does not occur in the sum and $`𝐋`$ is a positive $`15\times 15`$–matrix.
Our goal is now to bring some dynamical evolution $`_t=e^{𝒵t}`$ into an appropriate standard form $`_t^{}`$. Here, $`_t`$ approximates the ideal unitary evolution $``$ given by $``$, where
$`𝒵=ii_\text{l}+,`$ (214)
while the standard form $`_t^{}`$ is specified by
$`𝒵^{}=ii_\text{l}^{}+^{}.`$ (215)
That is, the decoherence process corresponding to the standard form is described by $`H_\text{l}^{}`$ and $`𝐋^{}`$, rather than $`H_\text{l}`$ and $`𝐋`$ in the original evolution. As in the case of CPMs, our goal is to obtain a simplified standards form in the sense that the number of relevant parameters describing the decoherence process are decreased, while the desired Hamiltonian evolution is not altered. To achieve this we will make use of the following facts MarkovSim :
* Let $`U`$ be some unitary matrix and $`𝒰\rho :=U\rho U^{}`$ be the corresponding operation. By unitary conjugation the Markovian evolution $`_t=e^{𝒵t}`$ can be transformed into the Markovian evolution $`_t^{}=𝒰_t𝒰^{}=e^{𝒵^{}t}`$ described by
$`H^{}`$ $`=`$ $`UHU^{},`$ (216)
$`H_\text{l}^{}`$ $`=`$ $`UH_\text{l}U^{},`$ (217)
$`𝐋^{}`$ $`=`$ $`O𝐋O^T,`$ (218)
where $`O`$ is the orthogonal matrix corresponding to $`U`$, that describes the basis change $`\sigma _𝐤U\sigma _𝐤U^{}`$ for the linear basis of hermitian traceless operators $`\sigma _𝐤`$.
* A Markovian evolution $`e^{𝒵^{}t}`$ according to a linear combination $`𝒵^{}=_{i=1}^Rp_i𝒵_i`$ ($`_ip_i=1`$) can be simulated by repeatedly applying a sequence $`e^{𝒵_1\mathrm{\Delta }t}\mathrm{}e^{𝒵_R\mathrm{\Delta }t}`$ for some small time intervals $`\mathrm{\Delta }t=\frac{t}{M}`$ ($`M`$ is the number of repetitions), i.e.
$`\begin{array}{ccc}\left(_{i=1}^Re^{p_i𝒵_i\frac{t}{M}}\right)^M& \stackrel{M\mathrm{}}{}& e^{_{i=1}^Rp_i𝒵_it}.\end{array}`$ (219)
Note that the approximated GKS matrix is $`𝐋^{}=_ip_i𝐋_i`$ and the error in approximation is of order $`O(\mathrm{\Delta }t^2)`$ error .
An alternative method consists in the random application of the time evolutions $`e^{𝒵_i\mathrm{\Delta }t}`$ with probability $`p_i`$ in each of the time intervals $`\mathrm{\Delta }t=t/M`$. That is, the evolution in the time interval $`\mathrm{\Delta }t`$ is given by $`_ip_ie^{𝒵_i\mathrm{\Delta }t}`$, which accurately approximates the desired operation in first order $`\mathrm{\Delta }t`$. A sequential application of these random operations $`M`$ times reproduces the desired operation $`e^{_{i=1}^Rp_i𝒵_it}`$ in the limit $`M\mathrm{}`$, i.e. note\_randomops
$`\begin{array}{ccc}\left(_{i=1}^Rp_ie^{𝒵_i\frac{t}{M}}\right)^M& \stackrel{M\mathrm{}}{}& e^{_{i=1}^Rp_i𝒵_it}\end{array}.`$ (220)
Following the structure of the previous section we first consider the case of decoherence itself, i.e. we set $`H=0`$. We propose a depolarization protocol that, after integration of the corresponding master equation, yields the same standard forms as obtained in Sec. III and Sec. IV.1. Second, we consider the case where $`H`$ corresponds to some Ising-type interaction, e.g. $`H=\sigma _y^A\sigma _y^B`$. We make use of the results obtained in Sec. IV.2.2 for the corresponding unitary $`e^{iHt}`$. We show that a depolarization procedure exists for which —in the limit of infinite intermediate local control operations— the system evolves according to some standard form which has the standard form Eq. (108), when regarded as a CPM $`_t^{}`$. The depolarization procedures we describe in the following (Sec. V.1 and Sec. V.2) are only relevant in cases where one is interested in the standard form for the complete dynamics and not only after some given time (the latter corresponding to the case of CPMs discussed in the previous Sections). Otherwise, one may use the conceptually simpler depolarization procedure for CPMs. Nevertheless subsections V.1 and V.2 provide the necessary tools for the procedure proposed in the subsequent Sec. V.3, where we show how to achieve a standard form for some arbitrary unitary dynamics. Although this procedure overcomes the restrictions to the area of applications in Sec. IV.2, this depolarization protocol generally increases the noise level of the decoherence process. In the following we assume that local unitary control operations can be performed on time scales negligible compared to the interaction time for the dynamics. We will thus refer to these operations as being instantaneous.
### V.1 Standard form for decoherence processes
We first consider maps describing pure decoherence processes ($`H=0`$) for a single qubit. For the depolarization we consider the same twirling procedures as in Sec. III, but now we intend to bring the Markovian generator $`𝒵`$ of the initial dynamics into the standard form
$`𝒵^{}={\displaystyle \underset{k}{}}u_kU_k𝒵U_k^{},`$ (221)
where $`U_k`$ denote the unitaries which were applied in Sec. III with equal probability $`u_k`$ to achieve the standard form of a Pauli channel, i.e. $`U_k=\sigma _k`$ is one of the Pauli matrices and $`u_k=\frac{1}{4}`$, or the simpler standard form of a depolarizing channel, i.e. $`U_k`$ is of the form $`Q_l\sigma _i`$ ($`Q_l=e^{i\frac{\pi }{4}\sigma _l}`$, $`l=1,2,3`$) and $`u_k=\frac{1}{12}`$. In Sec. III the number $`u_k`$ represented the probability to apply the different twirling unitaries. According to (ii), a similar procedure still works, where intermediate random applications of the corresponding unitaries lead to a standard form of the Markovian generator $`𝒵`$. Alternatively, one can consider a further splitting of the time interval $`Deltat`$ (see Eq. 219), where all possible unitary operations corresponding to the depolarization process are applied sequentially. We will consider the second approach in the following.
More precisely, we will consider the following depolarization protocol: Let the actual dynamics of the system (i.e. the decoherence process) evolve for some time $`t`$ and choose a split of the total time $`t`$ into $`M`$ sufficiently small time intervals $`\mathrm{\Delta }t`$. During each of these small time intervals the system dynamics is accompanied by the sequence of instantaneous local control operations $`U_{k+1}U_k`$ ($`U_0=\mathrm{𝟏}`$) applied in arbitrary order but with equal distance $`u_k\mathrm{\Delta }t`$. From fact (i) it follows that instead of the original dynamics $`e^{i𝒵u_k\mathrm{\Delta }t}`$ during each of the time intervals $`u_k\mathrm{\Delta }t`$ the system evolves according to the Markovian generator $`U_k𝒵U_k^{}`$. In the time interval $`\mathrm{\Delta }t`$ the evolution is thus given by
$`{\displaystyle \underset{k}{}}e^{u_k\mathrm{\Delta }t𝒰_k𝒵𝒰_k^{}}.`$ (222)
If these intervals are chosen sufficiently small (i.e. $`M\mathrm{}`$) fact (ii) implies that the overall system dynamics can effectively be approximated by the Markovian generator in Eq. (221).
Let us consider the effect of this depolarization on $`𝐋`$ and $`H_\text{l}`$ more closely. According to (i) the GKS matrix is brought into the standard form Eq. (30) or Eq. (35), except that the first row and column of Eq. (29) is disregarded in both equations. Thus we obtain
$`^{}\rho =2{\displaystyle \underset{k=1}{\overset{3}{}}}L_k\left(\rho \sigma _k\rho \sigma _k\right)`$ (223)
with $`L_k=L_{kk}`$ in the case of twirling with Pauli operators $`U_k=\sigma _k`$ and
$`^{}\rho =2L\left(3\rho {\displaystyle \underset{k=1}{\overset{3}{}}}\sigma _k\rho \sigma _k\right)=4L\left(2\rho \mathrm{𝟏}\right)`$ (224)
with $`L=\frac{1}{3}\left(L_{11}+L_{22}+L_{33}\right)`$ in the case of complete depolarization. Similarly for the Hamiltonian $`H_\text{l}`$ the twirling by means of the Pauli matrices $`U_k=\sigma _k`$ yields (see (i)):
$`H_\text{l}^{}={\displaystyle \frac{1}{4}}{\displaystyle \underset{k=0}{\overset{3}{}}}\sigma _kH_\text{l}\sigma _k={\displaystyle \frac{1}{2}}\text{tr}(H_\text{l})\mathbf{\hspace{0.17em}1}.`$ (225)
Since this twirling is also performed before each of the Clifford unitaries $`Q_l`$ applied, we obtain the same result in the case of complete depolarization. Thus in both cases the Lamb shift Hamiltonian in the standard form gives only rise to some overall phase factor $`e^{\frac{i}{2}\text{tr}(H_\text{l})t}`$, which can be neglected.
We briefly examine the dynamics $`_t^{}`$ due to the Standard forms, i.e. the solutions of the master equation
$`\dot{\rho }=^{}\rho ,`$ (226)
where the Liouvillian $`^{}`$ is given by Eq. (223) or Eq. (224). It is straightforward to see that in case of Eq. (223) the Pauli matrices diagonalize the Liouvillian $`^{}`$ BriegelEnglert , i.e.
$`^{}\sigma _0=`$ $`0`$ $`^{}\sigma _1=4(L_2+L_3)\sigma _1`$
$`^{}\sigma _2=`$ $`4(L_1+L_3)\sigma _2`$ $`^{}\sigma _3=4(L_1+L_2)\sigma _3.`$
For an arbitrary initial state $`\rho (0)=\frac{1}{2}\left[\mathrm{𝟏}+𝐧𝝈\right]`$ ($`|𝐧|=1`$, $`𝝈(\sigma _1,\sigma _2,\sigma _3)^T`$) we thus obtain
$`\rho (t)=e^^{}t\rho (0)={\displaystyle \frac{1}{2}}\left[\mathrm{𝟏}+𝐧(t)𝝈\right],`$ (227)
where $`𝐧(t)=(n_1e^{4(L_2+L_3)t},n_2e^{4(L_1+L_3)t},n_3e^{4(L_1+L_2)t})^T`$. The action of $`_t^{}`$ in terms of Pauli matrices as in Eq. (29) is actually that of a Pauli channel
$`_t\rho ={\displaystyle \underset{k=0}{\overset{3}{}}}E_k(t)\sigma _k\rho \sigma _k,`$ (228)
where
$`E_0\left(t\right)`$ $`=`$ $`{\displaystyle \frac{1}{4}}\left(1+e^{4\left(L_2+L_3\right)t}+e^{4\left(L_1+L_3\right)t}+e^{4\left(L_1+L_2\right)t}\right)`$
$`E_1\left(t\right)`$ $`=`$ $`{\displaystyle \frac{1}{4}}\left(1+e^{4\left(L_2+L_3\right)t}e^{4\left(L_1+L_3\right)t}e^{4\left(L_1+L_2\right)t}\right)`$
$`E_2\left(t\right)`$ $`=`$ $`{\displaystyle \frac{1}{4}}\left(1e^{4\left(L_2+L_3\right)t}+e^{4\left(L_1+L_3\right)t}e^{4\left(L_1+L_2\right)t}\right)`$
$`E_3\left(t\right)`$ $`=`$ $`{\displaystyle \frac{1}{4}}\left(1e^{4\left(L_2+L_3\right)t}e^{4\left(L_1+L_3\right)t}+e^{4\left(L_1+L_2\right)t}\right).`$
Similarly a specialization to complete depolarization reveals the depolarizing channel
$`_t(\rho )=p(t)\rho +(1p(t))\text{tr}(\rho ){\displaystyle \frac{1}{2}}\mathrm{𝟏}`$ (229)
with $`p(t)=e^{8Lt}`$.
The fact that the above depolarization procedure for continuous dynamics rediscover the standard forms already obtained in Sec. III for CPM can also be understood by the simple fact, that the evolution $`e^{𝒵^{}t}`$ is invariant under unitary conjugation, i.e. $`𝒰e^{𝒵^{}t}𝒰^{}=e^{𝒵^{}t}`$, iff the generator $`𝒵^{}`$ is invariant under this conjugation. By construction of the protocol above, this definitely holds for the standard form $`𝒵^{}`$ w.r.t. any of the corresponding twirling unitaries $`U_k`$. With $`u_k`$ being the twirling constant we thus obtain
$`_t^{}=e^{𝒵^{}t}={\displaystyle \underset{k}{}}u_k𝒰_ke^{𝒵^{}t}𝒰_k^{}.`$ (230)
Since this Eq. precisely reflects the corresponding depolarization procedure for CPM of Sec. III, the resulting time dependent CPM $`_t^{}`$ has to be of standard form for CPMs.
The depolarization protocols for decoherence processes (described in this subsection) can be readily generalized to the multi-party setting. The Lamb shift $`H_\text{l}^{}`$ in the standard form can be neglected again whereas the Liouvillian is of a standard form that corresponds to the multi-party Pauli channel Eq. (51) or the multi-party depolarizing channel Eq. (53). Note, however, that each additional party causes a finer splitting of the time interval $`\mathrm{\Delta }t`$, yielding an exponential increase of the number of intermediate control operations. More precisely for $`N`$ parties the system dynamics in each time interval $`\mathrm{\Delta }`$ has to be interrupted by $`4^N`$ \[$`12^N`$\] control operations in order to achieve a dynamics corresponding to a Pauli channel \[depolarizing channel\]. In this case, the alternative method of applying random local unitary operations rather than the complete sequence of unitaries is certainly privileged.
### V.2 Standard forms for noisy Ising-type interactions
Let us now move to the case where the ideal operation is given by some Ising-type interaction
$`H^{AB}=g\sigma _y^A\sigma _y^B.`$ (231)
For depolarization we can essentially consider the same protocol as in the previous section, except that we now take the twirling unitaries $`U_k`$, that were used in Sec. IV.2.2. More precisely $`u_k=\frac{1}{32}`$ and the $`U_k`$ are given by the $`32`$ unitaries from the product set $`𝒰_1𝒰_2𝒰_3`$, where
$`𝒰_1`$ $`=`$ $`\{\mathrm{𝟏}_A\mathrm{𝟏}_B,e^{i\frac{\pi }{4}\sigma _y^A}\mathrm{𝟏}_B,\mathbf{\hspace{0.17em}1}_Ae^{i\frac{\pi }{4}\sigma _y^B},e^{i\frac{\pi }{4}\sigma _y^A}e^{i\frac{\pi }{4}\sigma _y^B}\}`$
$`𝒰_2`$ $`=`$ $`\{\mathrm{𝟏}_A\mathrm{𝟏}_B,\sigma _x^A\sigma _x^B\}`$ (232)
$`𝒰_3`$ $`=`$ $`\{\mathrm{𝟏}_A\mathrm{𝟏}_B,\sigma _y^A\mathrm{𝟏}_B,\mathbf{\hspace{0.17em}1}_A\sigma _y^B,\sigma _y^A\sigma _y^B\},`$
e.g. $`U_k=e^{i\frac{\pi }{4}\sigma _y}\sigma _x\sigma _x\sigma _y`$. Recall that a twirling with these unitaries brings the phase gate $`U(\alpha )=e^{iH\alpha }`$ with an arbitrary angle $`\alpha `$ into a standard form Eq. (108) described by only $`17`$ independent parameters.
Apart from the different choice of twirling unitaries the depolarization reads exactly as in Sec. V.1: The overall interaction time $`t`$ is divided into sufficiently small time intervals $`\mathrm{\Delta }t`$, in which the system dynamics is interrupted for short local unitary control operations $`U_k`$. The resulting dynamics approximates an evolution $`_t^{}=e^{𝒵^{}t}`$ with new Markovian generator
$`𝒵^{}={\displaystyle \underset{k=1}{\overset{32}{}}}u_k𝒰_k𝒵𝒰_k^{}=ii_\text{l}^{}+^{}.`$ (233)
Note that due to the choice of the twirling unitaries the ideal Ising-type interaction Hamiltonian $`H^{}=H`$ is in fact not changed by this protocol. The GKS matrix $`𝐋^{}`$ is of standard form Eq. (108), except that in Eq. (108) the first row and column are disregarded. Since no normalization constraints are involved, the smaller $`15\times 15`$-matrix is still specified by $`17`$ real parameters. For the new Hamiltonian of the lamb shift one can compute that the twirling yields $`H_\text{l}^{}=H_{00}^\text{l}\sigma _0^A\sigma _0^B+H_{22}^\text{l}\sigma _2^A\sigma _2^B`$. Omitting the term $`H_{00}^\text{l}\mathbf{\hspace{0.17em}1}_{AB}`$, which would contribute only as global phase $`e^{iH_{00}^\text{l}t}`$ to the system dynamics, the Lamb shift in the standard form thus can be specified by a single real parameter $`H_{22}^\text{l}`$. Moreover we will include this term into the ideal interaction term rewriting $`H^{}=g^{}\sigma _y^A\sigma _y^B`$ with $`g^{}=g+H_{22}^\text{l}`$ and will disregard any Lamb shift in the following. The change in coupling strength demands a reinterpretation of the interaction time. In order to simulate the actual dynamics $`e^{𝒵t}`$ running for some time $`t`$ by a dynamics of the standard form $`e^{𝒵^{}t}`$ the simulation actually has to run for the time $`t_s=ct`$, where $`c=\frac{g}{g^{}}`$ represents the time cost of the simulation.
A similar argument as in the previous subsection shows that the obtained standard form $`e^{𝒵^{}t}`$ seen as a CPM is actually in the standard form considered in Sec. IV.2.2.
### V.3 Standard forms for arbitrary noisy evolutions by means of sacrificing
Let us now consider standard forms for arbitrary ideal unitary evolutions $``$. We make use of the fundamental fact HamSim that by a stroboscopic application of a sequence of local unitaries any (entangling) two–qubit Hamiltonian $`H`$ can simulate the Hamiltonian $`H_y=\sigma _y\sigma _y`$ of the phase gate operation $`U(\alpha )=e^{iH_y\alpha }`$ in Sec. IV.2.2 to arbitrary good approximation (and vice versa). Before going into detail we shortly sketch the procedure of deriving a standard form for any noisy unitary evolution. Along each single stroboscopic step $`\mathrm{\Delta }t=\frac{t}{M}`$ ($`M`$ number of steps) of the simulation protocols we propose to
* first apply the sequence of unitary operations in order to obtain the evolution according to $`i_yi_{\text{l}y}+_y`$,
* depolarize the CPM $`_t`$ described by $`i_yi_{\text{l}y}+_y`$ according to the procedure derived in Sec. V.2 yielding a standard form $`_t^{}`$ given by $`i_yi_{\text{l}y}^{}+_y^{}`$ and finally
* transform it back to the original Hamiltonian $`H`$ accompanied by some decoherence process of desired standard form $`_\text{l}^{}`$ and $`^{}`$.
$$\begin{array}{ccccc}& \stackrel{\text{approximates}}{}& 𝒵& \stackrel{\text{simulate}}{}& 𝒵_y\\ & & \text{standard form}& & \text{standard form}& & \\ & \underset{\text{approximates}}{}& 𝒵^{}& \underset{\text{simulate}}{}& 𝒵_y^{}\end{array}$$
We remark that steps (1) and (3) require in general a ’time cost’, i.e. the simulation of the action of a desired Hamiltonian for some time $`ct`$ requires a time $`t`$. This time cost translates into a smaller pre–factor for the interaction Hamiltonian in the corresponding generator $`𝒵^{}`$, ultimately leading to an increased noise. That is, the ratio of the strength of desired interaction (described by $``$) to strength of noise (described by $``$) decreases, leading to a reduction of the fidelity. We note that for two–qubit systems the time cost is at most 3.
In order to simulate the Hamiltonian $`H_y`$ by the (entangling) two–qubit Hamiltonian $`H`$ and fast local unitary transformations (see HamSim ) one considers the decomposition of $`H_y`$ in terms of $`H`$ (term isolation):
$`H_y^{AB}=c_v{\displaystyle \underset{i=1}{\overset{R_v}{}}}v_iV_iH^{AB}V_i^{}+Q_1^A+Q_2^B,`$ (234)
where $`V_i`$ are the local unitaries with probabilities $`v_i>0`$ ($`_iv_i=1`$), $`Q_1`$ and $`Q_2`$ are some local Hamiltonian on qubit $`A`$ and $`B`$ respectively and $`c_v>0`$ is some factor to adjust the coupling ’strength’ of the Hamiltonians $`H`$ and $`H_y`$. If the unitary evolution $`e^{iH_yt}`$ is supposed to be simulated for the time $`t`$, the simulation has to be carried out for the time $`t_s=c_vt`$. Since the local unitary control operation can be performed on negligible time scales, the factor $`c_v`$ thus determines the time cost for the following simulation: One chooses a split of the time $`t_s`$ into $`M`$ time intervals $`\mathrm{\Delta }t`$, such that the sequence $`\left(_ie^{iv_i𝒱_i𝒱_i^{}\mathrm{\Delta }t}e^{i𝒬_1\mathrm{\Delta }t}e^{i𝒬_2\mathrm{\Delta }t}\right)^M`$ is a sufficient approximation for $`e^{i_yt_s}`$ as discussed in (ii). Note that in each time step $`\mathrm{\Delta }t`$ the original dynamics according to $`H`$ is simply interrupted after the time $`v_i\mathrm{\Delta }t`$ in order to apply the local unitaries $`V_{i+1}^{}V_i`$ ($`V_0=\mathrm{𝟏}`$). This corresponds to the system evolving according to a sequence of Hamiltonians $`V_iHV_i^{}`$ for the time intervals $`v_i\mathrm{\Delta }t`$. At the end of each simulation step $`\mathrm{\Delta }t`$ the local unitary $`e^{i𝒬_1\mathrm{\Delta }t}e^{i𝒬_2\mathrm{\Delta }t}`$ has to be applied in order to cope with the single qubit dynamics in the Hamiltonian simulation.
Similarly, one can consider a Hamiltonian simulation for step (3) according to the decomposition
$`H^{AB}=c_w{\displaystyle \underset{j=1}{\overset{R_w}{}}}w_jW_jH_y^{AB}W_j^{}+\stackrel{~}{Q}_1^A+\stackrel{~}{Q}_2^B`$ (235)
with local unitaries $`W_j`$, single qubit Hamiltonians $`\stackrel{~}{Q}_1`$, $`\stackrel{~}{Q}_2`$ and time cost $`c_w`$ for the ’backward’ simulation. For step (2) we use the twirling protocol derived in Sec. V.2 providing a standard form Eq. (108) described by $`17`$ independent parameters. With $`c_y`$ we denote the corresponding time cost of this depolarization procedure.
With these notations at hand we can now specify the protocol to achieve the standard form $`_t^{}=e^{𝒵^{}t}`$ for an arbitrary noisy two–qubit evolution $`_t=e^{𝒵t}`$. Let $`_t=e^{𝒵t}`$ be a Markovian evolution with the generator $`𝒵=ii_\text{l}+`$, where $`\rho =i[H,\rho ]`$ corresponds to the ideal evolution with Hamiltonian $`H`$, $`_\text{l}\rho =i[H_\text{l},\rho ]`$ represents the Lamb shift with Hamiltonian $`H_\text{l}`$ and
$`\rho ={\displaystyle \underset{𝐤,𝐥\mathrm{𝟎}}{}}L_{\mathrm{𝐤𝐥}}\left([\sigma _𝐤\rho ,\sigma _𝐥]+[\sigma _𝐤,\rho \sigma _𝐥]\right)`$ (236)
corresponds to the Liouvillian with GKS-matrix $`𝐋`$. For notational simplicity we will in the following restrict to the case where in both steps (1) and (3) no single qubit dynamics has to be corrected, i.e. the terms $`Q_1`$, $`Q_2`$, $`\stackrel{~}{Q}_1`$ and $`\stackrel{~}{Q}_2`$ in the decompositions Eq. (234) and Eq. (235) vanish. If the system evolves for some time $`t`$, the following protocol requires the time $`t_s:=c_vc_yc_wt`$ and thus has time cost $`c_vc_yc_w`$. The ’simulation’ $`t_s`$ again has to be divided into sufficiently small time steps $`\mathrm{\Delta }t=\frac{t_s}{M}`$. In these time intervals we consider the following sequence of $`R=32R_vR_w`$ operations:
$`{\displaystyle \underset{i,j,k=1}{\overset{R}{}}}𝒲_j𝒰_k𝒱_ie^{𝒵w_ju_kv_i\mathrm{\Delta }t}𝒱_i^{}𝒰_k^{}𝒲_j^{}.`$ (237)
This sequence of operations corresponds to a splitting of the time interval $`\mathrm{\Delta }t`$ into smaller intervals of length $`w_ju_kv_i\mathrm{\Delta }t`$, in which at the beginning the (fast) local unitary $`W_j^{}U_k^{}V_i^{}`$ is performed, the system then evolves according to the given dynamics $`𝒵`$ and finally the inverse unitary ’pulse’ $`W_jU_kV_i`$ is applied at the end of the interval $`w_ju_kv_i\mathrm{\Delta }t`$. In the limit $`\mathrm{\Delta }t0`$ we obtain the Markovian dynamics $`_t^{}=e^{𝒵^{}t}`$ with
$`𝒵^{}={\displaystyle \underset{i,j,k=1}{\overset{R}{}}}w_ju_kv_i𝒲_j𝒰_k𝒱_i𝒵𝒱_i^{}𝒰_k^{}𝒲_j^{}.`$ (238)
It is straightforward to show that the ideal operation $``$ in the generator $`𝒵^{}`$ remains the same, since the twirling over $`U_k`$ leaves the Hamiltonian $`H_y`$ invariant. Moreover $`𝒵^{}`$ again has a decomposition of the form $`𝒵^{}=ii_\text{l}^{}+^{}`$ described by the new lamb shift Hamiltonian $`H_\text{l}^{}`$ and the GKS matrix $`𝐋^{}`$, that are obtained as follows:
$`\begin{array}{ccccccc}𝒵& \underset{(1)}{\overset{v_i,V_i}{}}& 𝒵_y& \underset{(2)}{\overset{u_k,U_k}{}}& 𝒵_y^{}& \underset{(3)}{\overset{w_j,W_j}{}}& 𝒵^{}\\ H_\text{l}& & H_{\text{yl}}=_iv_iV_iH_\text{l}V_i^{}& & H_{\text{yl}}^{}=_ku_kU_kH_{\text{yl}}U_k^{}& & H_\text{l}^{}=_jw_jW_jH_{\text{yl}}^{}WU_j^{}\\ 𝐋& & 𝐋_y=_iv_iO_{V_i}𝐋O_{V_i}^T& & 𝐋_y^{}=_ku_kO_{U_k}𝐋_yO_{U_k}^T& & 𝐋^{}=_jw_jO_{W_j}𝐋_y^{}O_{W_j}^T\end{array}`$ (239)
As discussed in Sec. V.2 the effect of step (2) on $`𝐋_y`$ is to bring the matrix into the standard form Eq. (108). The final standard form $`𝐋^{}`$ of the GKS matrix is obtained from $`𝐋_y^{}`$ by mixing according to $`(w_j,W_j)`$ and is thus specified by $`17`$ independent parameters only, although $`𝐋^{}`$ in general is not of the form Eq. (108). As seen in Sec. V.2 we can neglect the lamb shift by introducing some time cost $`c_y`$.
To summarize, we have shown how to achieve a standard form for arbitrary two–qubit interactions, where the noise process (described by the GKS matrix) is specified by 17 parameters. The above protocol can be affected by different sources of errors. For this, one can again compare the noise level of the standard form dynamics $`𝒵^{}`$ with the noise level of the original dynamics $`𝒵`$ in terms of the distance $`d(_t,_t)`$ and $`d(_t^{},_t)`$ to the ideal unitary evolution $`_t=e^{it}`$ for different times $`t`$, where $`d(,)`$ is a suitable distance measure (see Sec. II.3). Although we have yet not performed a detailed error analysis in this sense, a non–unit time cost (at most a factor of $`c_vc_w3`$ from simulating corresponding Hamiltonians - steps (1) and (3), plus the time cost $`c_y`$ from ’Lamb’ shift), in general, corresponds to an increase of the noise level for the evolution.
As it holds for the depolarization of CPMs in previous chapters and as opposed to the assumptions made in this paper, in practise, the depolarization protocol has to face imperfections in the local control operations, whose extent depends on the physical realization. Additionally, for the depolarization of master equations by means of stroboscopic control operations one also encounters errors of order $`O(\mathrm{\Delta }t^2)`$ due to the finite approximation (see fact (ii)) error . Note that, in practise, there will be a trade-of between errors in approximation and errors due to imperfect local control operations.
### V.4 Simplified standard forms for arbitrary noisy evolutions
A further reduction of the number of relevant noise parameters and thus a simpler standard form may be achieved following the ideas developed in Sec. IV.3 for CPMs. There, by increasing the noise level and hence reducing the fidelity of the operation, we have shown that one can in fact achieve that the noise part of the evolution is described by only a single parameter (white noise). The procedure outlined in Sec. IV.3.3 is based on probabilistically mixing the (already depolarized) noisy CPM $``$ with a certain separable map $`𝒟`$, i.e. a map which can be obtained without interactions between particles. That is, one chooses randomly whether one wishes to apply the map $``$ corresponding to the noisy operation, or the separable map $`𝒟`$. For a proper choice of $`𝒟`$ the resulting map $`_S`$ is of the form Eq. (209).
In the case of master equations, one may adopt this procedure in such a way that for each time interval $`\mathrm{\Delta }t`$, one applies the (already depolarized) noisy evolution described by the $`𝒵^{}`$, together with an appropriate separable evolution (that may e.g. be generated with the help of available local unitary control operations and additional measurements) with corresponding generator $`𝒵_D`$. Both evolutions now have to be applied either sequentially or chosen randomly. This implies that either one has the ability to switch off the evolution $`𝒵^{}`$, or one can produce a separable evolution of arbitrary strength $`𝒵_D`$. Note that in this case, fast local unitary operations are in general not sufficient, but arbitrary local control operations (including measurements) are required to generate the desired separable operations. As in the case of CPMs this depolarization procedure requires moreover the knowledge of the exact form for $`𝒵^{}`$ in order to choose an appropriate, separable $`𝒵_D`$. The total evolution is finally described by a Liouvillian with GKS matrix proportional to the identity, which corresponds to global white noise at the level of the respective CPM.
## VI Summary
In this article, we have introduced the concept of depolarization of noisy evolutions. We have shown how to reduce the relevant number of parameters describing an arbitrary, unknown noise process described by a CPM in such a way that the ideal (unitary) part of the evolution is not altered. For decoherence processes we have explicitly calculated the corresponding standard forms for multipartite systems of arbitrary number $`N`$ of parties and arbitrary dimension $`d`$. We find a reduction of an arbitrary noise process described by $`O(d^{4N})`$ to local and global white noise processes described by only $`2^N`$ parameters. For specific two–qubit unitary operations (e.g. Phase gate with arbitrary phase), we obtain a standard form described by at most 17 parameters. For other gates, the standard forms can be further simplified. In particular we find standard forms described by 8 parameters for the CNOT–gate and 3 parameters for the SWAP gate. The depolarization procedures used to obtain these standard forms are universal in the sense that the exact form of the noise process need not be known. With knowledge of the exact form of the noise process, and by allowing for a (small) reduction of the fidelity, one can further simplify the standard forms. In fact, we have derived a depolarization protocol that yields a reduction to global white noise, which is described by only a single noise parameter and where, in the worst case, the noise level is increased by about an order of magnitude.
We have generalized our results to evolutions described by a master equation of Lindblad form. Standard forms for decoherence processes and interaction Hamiltonians proportional to the Ising Hamiltonian can be derived using similar methods as for CPMs, leading to standard forms with same number of parameters. We have also obtained a standard form described by 17 parameters for arbitrary two–qubit interaction Hamiltonians, which, in general, goes along with an increase of the noise level. As the basic tool we have used the possibility to simulate the Ising Hamiltonian by an arbitrary Hamiltonian (and vice versa), together with depolarization of the Ising type interaction. Again, a further simplification to a single parameter leading to a GKS matrix proportional to identity is possible under certain circumstances.
We are confident that such simplified standard forms for noise processes will provide a useful tool to investigate various problems in quantum information processing involving noisy apparatus and interactions with environment. Straightforward applications include the possibility to calculate lower bounds on the channel capacity of arbitrary noise channels (by investigating the corresponding depolarized channels), and a simplified process tomography where only a reduced number of parameters of the noise process needs to be determined. Further conceivable applications include the determination of lower bounds to the lifetime of entangled states, and strict error thresholds for quantum computation that are valid for arbitrary noise processes and are not restricted to certain noise models.
## Acknowledgements
We thank F. Verstraete for interesting discussions. This work was supported by the Austrian Science Foundation (FWF), the European Union (IST-2001-38877,-39227,OLAQUI,SCALA), the Österreichische Akademie der Wissenschaften through project APART (W.D.) and the Deutsche Forschungsgemeinschaft (DFG).
## Appendix A: Spectral decomposition versus Kraus representation
For sake of completeness we will shortly discuss the relation of the Jamiołkowski state $`E`$ with two common representations of the corresponding CPM $``$ in terms of its Kraus representation and its purification. In Appendix A we consider a CPM $``$ described by its Kraus representation. More precisely any CPM $``$ acting on $`𝐇^A`$ allows for a decomposition of the form
$`(M)={\displaystyle \underset{i=1}{\overset{r}{}}}K_iMK_i^{}`$ (240)
with $`rd_A\times d_A^{}`$ ($`r=`$ Choi rank) Kraus operators $`K_i(𝐇^A,𝐇^A^{})`$, where $`K_i`$ can be chosen to be orthogonal, i.e. $`\text{tr}\left[K_i^{}K_j\right]=\delta _{ij}`$. $``$ corresponds to SLOCC operation iff
$`{\displaystyle \underset{\alpha =1}{\overset{r}{}}}K_i^{}K_i\mathrm{𝟏},`$ (241)
where equality holds iff $``$ is trace preserving. Eq. (240) is an immediate consequence of the corresponding fact, that any positive operator $`E0`$ (see No. 3 in Sec. II.2) allows a spectral decomposition
$`E={\displaystyle \underset{i=1}{\overset{r}{}}}|v_iv_i|,`$ (242)
where $`|v_i`$ are some unnormalized vectors in $`𝐇^A^{}𝐇^A`$, $`r=\text{rank}(E)d_A\times d_A^{}`$, which can be chosen to be orthogonal. Using the decomposition
$`|v_i={\displaystyle \underset{\genfrac{}{}{0.0pt}{}{\alpha _{d_A^{}}}{\beta _{d_A}}}{}}v_i^{\alpha \beta }|\alpha ^A^{}|\beta ^A`$ (243)
this correspondence is simply given by
$`K_i=\sqrt{d_A}{\displaystyle \underset{\alpha \beta }{}}v_i^{\alpha \beta }|\alpha ^A^{}^A\beta |`$ (244)
or $`K_i^{\alpha |\beta }=\sqrt{d_A}v_i^{\alpha \beta }`$.
Tracing out either system $`A^{}`$ or $`A`$ yields:
$`\text{tr}_A^{}E^{A^{}A}`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{r}{}}}\left[K_i^{}K_i\right]^T`$ (245)
$`\text{tr}_AE^{A^{}A}`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{r}{}}}K_iK_i^{}`$ (246)
Note that No. 4 in Sec. II.2 now follows directly from Eq. (241) using Eq. (245). The unitary freedom in the choice of decompositions for $`E`$ translates to the corresponding CPM as follows. Two decomposition of $``$ \[$`E`$\] with Kraus operators $`K_i`$ and $`L_j`$ \[vectors $`|v_i`$ and $`|w_j`$\] correspond to the same CPM \[positive operator\] iff there exists a unitary matrix $`U_{ij}`$ such that $`K_i=_{ij}U_{ij}L_j`$ \[$`|v_i=_{ij}U_{ij}|w_j`$\]. From these decompositions one can easily obtain a few more results about the relation between positive operators and CPM under the Jamiołkowski isomorphism:
* $``$ is factorizable, i.e. $`(M)=KMK^{}`$, iff $`E`$ is pure, i.e. $`E=|vv|`$.
* $``$ is an isometry, i.e. $`(M)=VMV^{}`$ with $`V^{}V=\mathrm{𝟏}_A`$, iff $`d_Ad_A^{}`$ and $`E`$ is maximally entangled, i.e. pure and $`\text{tr}_A^{}E^{A^{}A}=\mathrm{𝟏}_A`$. (For a unitary we have $`d_A=d_A^{}`$.)
* $``$ is a projection, i.e. $`(M)=d_A\text{tr}\left[\tau _2^tM\right]\tau _1`$, iff $`E`$ is a product state, i.e. $`E=\tau _1\tau _2`$.
* $``$ can be decomposed into a sum of projections, iff $`E`$ is separable. In this case $``$ is entanglement breaking Shor .
## Appendix B: Purification for quantum states and operations
Frequently a CPM $``$ is also regarded as a description of the (non-unitary) evolution of the system alone (i.e. by tracing the environmental degrees of freedom), where the system $`A`$ together with its environment $`C`$ is believed to evolve according to some unitary operation $`U_{}`$. Any such system-environment model $`U_{}`$ then is said to be a purification of the CPM $``$, if it yields $``$ as the evolution of the system alone after tracing out the environment:
$`(\rho )=\text{tr}_C\left[U_{}^{AC}\rho \rho _C(U_{}^{AC})^{}\right].`$ (247)
Here we assume that the system $`A`$ and its environment $`A^{}`$ are initially decoupled $`\rho \rho _C`$, where $`\rho _C=\frac{1}{d_C}\mathrm{𝟏}_C`$ is the maximally mixed state of the environment. Like the operator sum decomposition, a purification of a quantum operation $``$ in terms of unitary evolution $`U_{}`$ can simply be derived from the corresponding purification of the quantum state $`E`$ in terms of a pure state on the joint system $`A^{}A`$ and $`C`$. For this let $`|\psi ^{A^{}AC}`$ be the pure state, such that $`E^{A^{}A}=\text{tr}_C\left[|\psi ^{A^{}AC}\psi |\right]`$, then the corresponding purification $`U_{}`$ is given similarly to Eq. (244) by
$`U_{}^{AC}=\sqrt{d_A\times d_C}{\displaystyle \underset{\genfrac{}{}{0.0pt}{}{i_{d_A^{}}}{\genfrac{}{}{0.0pt}{}{j_{d_A}}{k_{d_C}}}}{}}\psi _{ijk}|i^A^{}^Aj|{}_{}{}^{C}k|,`$ (248)
using the decomposition $`|\psi =_{i,j,k}\psi _{ijk}|i^A^{}|j^A|k^C`$. Note that in both cases purifications can be chosen such that $`d_C=dim_{}(𝐇^C)d_A\times d_A^{}`$.
## Appendix C: Details for the extension to $`d`$–level systems
In this appendix we present some details about the generalization of the standard forms for decoherence to $`d`$–level systems as it was introduced in Sec. III. We proof that
* the twirling over the Pauli group as in Eq. (41) depolarizes any CPM to the standard form of a generalized Pauli channel Eq. (42),
* the standard form of a generalized depolarizing channel Eq. (50) can be achieved by twirling over a finite set of generalized Clifford operations.
But let us briefly consider the measurements in the generalized Bell basis Eq. (36) of the isomorphism protocol. It is straight forward to compute, that the other Bell measurement results $`|\psi _{kl}^{A\overline{A}}`$ yield $`\left(U_{kl}^{}\rho U_{kl}^T\right)`$ instead of $`(\rho )`$ for the outcome of the protocol on system $`A^{}`$ but with the same probability $`\frac{1}{d^2}`$.
Before coming to the proofs, we summarize some useful facts (see also Wo03 ) about the generalized Pauli operators $`U_{kl}`$ (see Eq. (37)), that are straightforward to proof:
$`U_{k^{}l^{}}U_{kl}`$ $`=`$ $`e^{i\frac{2\pi }{d}k^{}l}U_{(k+k^{})(l+l^{})}`$ (249)
$`U_{kl}^{}`$ $`=`$ $`e^{i\frac{2\pi }{d}kl}U_{(k)(l)}`$ (250)
$`U_{kl}^{}`$ $`=`$ $`U_{(k)(l)}`$ (251)
$`U_{kl}^T`$ $`=`$ $`e^{i\frac{2\pi }{d}kl}U_{(k)(l)}`$ (252)
$`U_{k^{}l^{}}^A|\psi _{kl}`$ $`=`$ $`e^{i\frac{2\pi }{d}k^{}l}|\psi _{(k+k^{})(l+l^{})}`$ (253)
$`U_{k^{}l^{}}^A^{}|\psi _{kl}`$ $`=`$ $`e^{i\frac{2\pi }{d}(k+k^{})l^{}}|\psi _{(k+k^{})(ll^{})}`$ (254)
With these relations at hand it is easy to verify (i).
First note that the set of generalized local Pauli operators
$`S=\{g_{kl}|g_{kl}=U_{kl}^AU_{(k)(l)}^A^{};k,l_d\}`$ (255)
is a commutative subgroup (of the generalized local Pauli group) that stabilizes $`|\mathrm{\Omega }`$ and is generated by the two elements $`g_{10}`$ and $`g_{01}`$. Moreover a simple calculation shows that
$`g_{k^{}l^{}}|\psi _{kl}=e^{i\frac{2\pi }{d}\left(k^{}lkl^{}\right)}|\psi _{kl}.`$ (256)
For a general state $`E=_{\alpha \beta ,\alpha ^{}\beta ^{}}E_{\alpha \beta ,\alpha ^{}\beta ^{}}|\psi _{\alpha \beta }\psi _{\alpha ^{}\beta ^{}}|`$ we therefore find, that it can be diagonalized by a probabilistic application of the local unitaries $`g_{kl}`$ with uniform probability $`\frac{1}{d^2}`$:
$`𝒟(E)`$ $`:=`$ $`{\displaystyle \frac{1}{d^2}}{\displaystyle \underset{k,l=0}{\overset{d1}{}}}g_{kl}Eg_{kl}^{}`$ (257)
$`=`$ $`{\displaystyle \frac{1}{d^2}}{\displaystyle \underset{\alpha \beta ,\alpha ^{}\beta ^{}}{}}E_{\alpha \beta ,\alpha ^{}\beta ^{}}|\psi _{\alpha \beta }\psi _{\alpha ^{}\beta ^{}}|\times `$
$`\times \left({\displaystyle \underset{k=0}{\overset{d1}{}}}e^{i\frac{2\pi }{d}k(\beta \beta ^{})}\right)\times \left({\displaystyle \underset{l=0}{\overset{d1}{}}}e^{i\frac{2\pi }{d}l(\alpha \alpha ^{})}\right)`$
$`=`$ $`{\displaystyle \underset{\alpha \beta }{}}E_{\alpha \beta ,\alpha \beta }|\psi _{\alpha \beta }\psi _{\alpha \beta }|`$
This mixing operation $`𝒟=𝒟_1𝒟_2`$ can also be decomposed into a mixing of shift operation $`𝒟_1(E)=\frac{1}{d}_{l=0}^{d1}g_{0l}Eg_{0l}^{}`$ and a mixing of phase multiplication operation $`𝒟_2(E)=\frac{1}{d}_{k=0}^{d1}g_{k0}Eg_{k0}^{}`$.
Let us now consider statement (ii).
For this we assume that $``$ is already brought into the form of a (generalized) Pauli channel of Eq. (42) by a random application of one of the Pauli operators $`U_{kl}`$. Now the generalized Clifford operation is a unitary operation $`Q`$ that maps the generalized Pauli group $`𝒫=\left\{e^{i\frac{2\pi }{d}\delta }U_{kl}\right|k,l,\delta _d\}`$ to itself under conjugation, i.e. $`Q𝒫Q^{}=𝒫`$. It is totally specified by its action on the two generators $`U_{10}`$ and $`U_{01}`$ of the Pauli group, since
$`QU_{kl}Q^{}`$ $`=`$ $`\left(QU_{01}Q^{}\right)^l\left(QU_{10}Q^{}\right)^k`$ (258)
$`=`$ $`U_{(ka+lb)(kc+ld)}.`$ (259)
Here we have chosen $`QU_{10}Q^{}=U_{ac}`$ and $`QU_{01}Q^{}=U_{bd}`$. In addition we disregard appropriate phase factors $`e^{i\frac{\pi }{d}\delta }`$ with $`\delta _{2d}`$, since they will be irrelevant for our purposes. Up to these phase factors the Clifford unitary $`Q`$ permutes a Pauli operator $`U_{kl}`$ to a new element $`U_{k^{}l^{}}`$, which in modular arithmetic (modulo $`d`$) is related to $`U_{kl}`$ by the linear transformation
$`\left(\begin{array}{c}k^{}\\ l^{}\end{array}\right)=𝐂_Q\left(\begin{array}{c}k\\ l\end{array}\right):=\left(\begin{array}{cc}a& b\\ c& d\end{array}\right)\left(\begin{array}{c}k\\ l\end{array}\right).`$ (260)
This linear transformation needs to be symplectic symplectic in order to truly correspond to a Clifford operation Ho04 . Symplecticity of $`𝐂_Q`$ in our single-party case simply reduces to the condition $`\text{det}𝐂_Q=1`$ (modulo $`d`$). Applying one of these Clifford unitaries $`Q^AQ_{}^{}{}_{}{}^{A^{}}`$ to the state $`E^{AA^{}}`$ gives the state
$`Q^AQ_{}^{}{}_{}{}^{A^{}}E(Q^AQ_{}^{}{}_{}{}^{A^{}})^{}`$ (261)
$`=`$ $`{\displaystyle \underset{kl}{}}E_{kl}Q^AU_{kl}^A(Q^A)^{}Q^AQ_{}^{}{}_{}{}^{A^{}}\times `$
$`\times `$ $`|\mathrm{\Phi }^{AA^{}}\mathrm{\Phi }|\left(Q^AQ_{}^{}{}_{}{}^{A^{}}\right)^{}\left(Q^AU_{kl}^A(Q^A)^{}\right)^{}`$
$`=`$ $`{\displaystyle \underset{k^{}l^{}}{}}E_{kl}|\psi _{k^{}l^{}}^{AA^{}}\psi _{k^{}l^{}}|,`$
where as claimed above any phase factor would cancel out. Note that the component $`E_{00}`$ will remain ’untouched’, since any symplectic matrix $`𝐂_Q`$ (even over $`𝔽_d`$ with $`d`$ non-prime) is invertible. In the following the Clifford unitary $`Q`$ will be chosen uniformly at random from a set of all Clifford unitaries, where each $`Q`$ corresponds only to a single $`𝐂_Q`$ (i.e.: fix a choice of phase factor for each $`𝐂_Q`$).
By elementary results from group theory it follows, that by application of the different $`𝐂`$ the set of all vectors $`(k,l)^T`$ with $`k0`$ or $`l0`$ will be mapped onto itself in such a way, that all vectors will occur equally often. For this let $`G`$ denote the group of symplectic matrices over $`𝔽_d^2`$, that act on the set $`X=𝔽_d^2`$. Furthermore for $`xX`$ let $`Gx:=\{gx|gG\}`$ denote the orbit of $`x`$ under the group action $`G`$ and let $`G_x:=\{gG|gx=x\}`$ denote the stabilizer of $`x`$. From the stabilizer orbit and Lagrange theorem it follows that for any finite group $`G`$ acting on a set $`X`$ we have $`|G|=|G_x||Gx|`$. This result can be used to show that each non–trivial element $`yX\{0\}`$ ($`0:=(00)^T𝔽_d^2`$) is obtained $`|G|`$ times, if the complete group $`G`$ is applied to all elements in $`X\{0\}`$. Since $`G`$ consists of invertible matrices, it maps the set $`X\{0\}`$ onto itself. Moreover any $`yX\{0\}`$ is only obtained from elements of its orbit $`Gy`$. For a fixed element $`xGy`$ the set $`G_{xy}:=\{gG|gx=y\}`$ can be rewritten in terms of only one of its elements $`g^{}`$ (i.e. $`g^{}x=y`$) and the stabilizer $`G_x`$ as $`G_{xy}=g^{}G_x`$. Thus $`y`$ is obtained form each of the $`|Gy|`$ elements (of its orbit) by $`|G_x|`$ different matrices. Since for two elements in the same orbit we have $`Gx=Gy`$, any element $`y`$ is obtained $`|G_x||Gx|=|G|`$ times. Note that in the case of prime dimension $`d`$, the set $`X`$ is (not only a module but also) a vector space over the field $`𝔽_d`$ and one can easily show that for all $`x0`$ the orbits are the same $`Gx=X\{0\}`$. This is due to the fact, that for each non–trivial vector $`x`$ one can find a symplectic (i.e. $`𝔽_d`$–invertible) matrix $`g`$ with the first column being $`x`$ (and the second the orthonormal vector $`x^{}`$).
A random application of the corresponding Clifford operations therefore provides a mixing of all components $`E_{kl}`$ with $`k0`$ or $`l0`$. Thus starting with a CPM $``$ in the form of a Pauli channel ( Eq. (42)) we can achieve the standard form in Eq. (50) by uniformly choosing a unitary $`Q`$ form the set of Clifford operations and applying $`Q^{}`$ before and $`Q`$ after the application of the CPM $``$. In fact the actual set, which the Clifford operations have to be chosen from in order to achieve a complete mixing of all the components $`E_{kl}`$ with $`k0`$ or $`l0`$, might even be decreased, as it is illustrated for the qubit case in Eq. (III). Note that the Clifford operation $`Q_1`$, $`Q_2`$ and $`Q_3`$ in Eq. (III) correspond to the three symplectic matrices $`C_1=\left(\begin{array}{cc}1& 1\\ 0& 1\end{array}\right)`$, $`C_2=\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right)`$ and $`C_2=\left(\begin{array}{cc}1& 0\\ 1& 1\end{array}\right)`$. |
warning/0507/hep-ph0507299.html | ar5iv | text | # References
Perturbative approach to $`U_A(1)`$ breaking<sup>1</sup><sup>1</sup>1Prepared for the Proceedings of The International Conference on High Energy and Mathematical Physics, Marrakech, Morrocco, 3-7 April 2005
Brigitte Hiller<sup>(1)</sup>, Alexander A. Osipov<sup>(1),(2)</sup>, Alex H. Blin<sup>(1)</sup>
(1) Centro de Física Teórica, Departamento de Física da Universidade de Coimbra, 3004-516 Coimbra, Portugal
(2) Joint Institute for Nuclear Research, Laboratory of Nuclear Problems, 141980 Dubna, Moscow Region, Russia
Abstract
The six-quark instanton induced ’t Hooft interaction is considered in combination with the Nambu-Jona-Lasinio (NJL) type $`U_L(3)\times U_R(3)`$ chiral symmetric Lagrangian. We discuss the bosonization of this multi-quark interaction, taking the $`U_A(1)`$ breaking as a perturbation. We discuss its relation with the usual approach.
Introduction
In low-energy QCD phenomena one considers basic powerful classification schemes, as symmetries and symmetry breaking patterns, large $`N_c`$ counting rules \- and scales at which some type of interactions set in . A well known example is $`\eta ^{}`$ meson physics, where all these concepts are needed to explain its large mass $`m_\eta ^{}1GeV`$. If the singlet $`SU(3)`$ axial current of QCD, $`U_A(1)`$, were conserved as $`m_{curr}0`$, the $`\eta ^{}`$ would occur as the ninth pseudoscalar Goldstone. As the global $`U_L(3)\times U_R(3)`$ chiral symmetry of QCD is broken by the $`U_A(1)`$ Adler-Bell-Jackiw anomaly , it was realized through the study of instantons - that effective $`2N_f`$ quark vertices arise, the ’t Hooft interactions , responsible for the non-conservation of the singlet axial current and absence of the related Goldstone boson. Typically, the mass squared of an approximate Goldstone boson is linear in the symmetry breaking parameter. The anomaly is of order $`1/N_c`$, one expects $`m_\eta ^{}^2`$ to be of order $`(1/N_c)`$. Although the $`U_L(3)\times U_R(3)`$ symmetry is recovered in the limit $`N_c\mathrm{}`$, the empirical mass of the $`\eta ^{}`$ is abnormaly large to conform alone with the large $`N_c`$ counting rules, which is Witten’s puzzle . The solution was given in , where with the help of QCD sum rules it was shown that a large scale emerges in the $`0^{},0^+`$ gluonic channels, compared to which the $`\eta ^{}`$ mass is small, and characterizing the breaking of asymptotic feedom in these channels.
Our plan consists in addressing the $`\eta ^{}`$ physics and lowest lying ($`0^{},0^+`$) meson nonets in a many-fermion vertices model without explicit gluon degrees of freedom. In this way one takes instead the quark structure of the mesons explicitly into account, and may obtain relevant information about the vacuum structure of QCD, albeit in an effective field theoretical approach. We use the successful Nambu - Jona-Lasinio model extended to the $`U(3)_L\times U(3)_R`$ chiral symmetry of massless QCD and the six quark ’t Hooft interactions for the $`U_A(1)`$ symmetry breaking. We review some known bosonization schemes with functional integral methods and show their restrictions. Within these schemes the leading order $`\eta ^{}`$ mass is obtained close to its empirical value and a good description of the remaining members of the lowest pseudoscalar nonet and related weak decay constants and quark condensates. The $`\eta ^{}`$ puzzle is also present in this case. Finally, using the spectral decomposition approach, we indicate how to obtain systematically all $`U_A(1)`$ corrections in a perturbative way and its relation to the previous approach.
The model Lagrangian
The starting Lagrangian is
$$=\overline{q}(i\gamma ^\mu _\mu \widehat{m})q+_{NJL}+_6,$$
(1)
with NJL four-fermion vertices of the scalar and pseudoscalar types
$$_{NJL}=\frac{G}{2}\left[(\overline{q}\lambda _aq)^2+(\overline{q}i\gamma _5\lambda _aq)^2\right]$$
(2)
and the six-quark ’t Hooft interaction
$$_6=\kappa (\text{det}\overline{q}P_Lq+\text{det}\overline{q}P_Rq),$$
(3)
Here the positive coupling $`G,[G]=\text{GeV}^2`$ has order $`G1/N_c`$, the negative coupling $`\kappa `$, $`[\kappa ]=\text{GeV}^5`$ with large $`N_c`$ asymptotics $`\kappa 1/N_c^{N_f}`$. The $`P_{L,R}=(1\gamma _5)/2`$ are projectors and the determinant is over flavor indices. At large $`N_c`$, $`_{NJL}`$ dominates over $`_6`$.
This Lagrangian has been studied at mean field level in terms of quark degrees of freedom in and in a bosonized form in and has been widely used since then . .
Bosonization
We use functional integral methods to bosonize the multi-quark Lagrangian, for which the vacuum persistance amplitude is
$$Z=𝒟q𝒟\overline{q}\mathrm{exp}\left(id^4x\right).$$
(4)
The fermion vertices of the original Lagrangian are linearized with help of the functional identity
$`1`$ $`=`$ $`{\displaystyle \underset{a}{}𝒟s_a𝒟p_a\delta (s_a\overline{q}\lambda _aq)\delta (p_a\overline{q}i\gamma _5\lambda _aq)}`$
$`=`$ $`{\displaystyle \underset{a}{}𝒟s_a𝒟p_a𝒟\sigma _a𝒟\varphi _a}`$
$`\times `$ $`\mathrm{exp}\{i{\displaystyle }\mathrm{d}^4x[\sigma _as_a\overline{q}\lambda _aq)+\varphi _a(p_a\overline{q}i\gamma _5\lambda _aq)]\},`$
$`Z`$ $`=`$ $`{\displaystyle \underset{a}{}𝒟\sigma _a𝒟\varphi _a𝒟q𝒟\overline{q}\mathrm{exp}\left(id^4x_q(\overline{q},q,\sigma ,\varphi )\right)}`$ (5)
$`\times `$ $`{\displaystyle \underset{a}{}𝒟s_a𝒟p_a\mathrm{exp}\left(id^4x_r(\sigma ,\varphi ,s,p)\right)}`$
with
$`_q(\overline{q},q,\sigma ,\varphi )`$ $`=`$ $`\overline{q}(i\gamma ^\mu _\mu \sigma i\gamma _5\varphi )q,`$ (6)
$`_r(\sigma ,\varphi ,s,p)`$ $`=`$ $`{\displaystyle \frac{G}{2}}\left(s_a^2+p_a^2\right)+s_a(\sigma _a\widehat{m}_a)+p_a\varphi _a`$ (7)
$`+`$ $`{\displaystyle \frac{\kappa }{32}}A_{abc}s_a\left(s_bs_c3\textcolor[rgb]{1,0,0}{p}_\textcolor[rgb]{1,0,0}{b}\textcolor[rgb]{1,0,0}{p}_\textcolor[rgb]{1,0,0}{c}\right).`$
The totally symmetric constants $`A_{abc}`$ are defined through the flavor determinant
$$detU=A_{abc}U_aU_bU_c,$$
(8)
where $`U=U_a\lambda _a`$. The scalar fields $`\sigma _a`$, $`a=0,3,8`$ have non-zero vacuum expectation values related to spontaneous breaking of global chiral symmetry, which give rise to the constituent masses $`m_a`$ of the quarks. New scalar fields with zero vacuum expectation values $`<0|\sigma _a|0>=0`$ must be defined through the shift $`\sigma _a\sigma _a+m_a`$. We seek the final bosonized Lagrangian expressed in terms of tree level mesonic fields $`\sigma _a`$ and $`\varphi _a`$ and must integrate out all remaining fields.
Methods used for the functional integration
The Gaussian integration over the quarks is done using the generalized heat kernel technique developed in . It accounts for a chiral covariant treatment of the one-loop determinant of the Dirac operator with an arbitrary non-degenerate quark mass matrix at each order of an expansion with generalized Seeley-DeWitt coefficients.
The integral over the auxiliary bosonic fields $`s_a,p_a`$
$$I[\sigma _a,\varphi _a]=\underset{a}{}𝒟s_a𝒟p_a\mathrm{exp}\left(id^4x_r(\sigma ,\varphi ,s,p)\right).$$
(9)
is done perturbatively. For that let us devide the lagrangian $`_r`$ in two parts. The free part, $`_0`$, given by
$$_0(\sigma ,\varphi ,s,p)=\frac{G}{2}\left(s_a^2+p_a^2\right)+s_a\sigma _a+p_a\varphi _a.$$
(10)
The ’t Hooft interaction is considered as perturbation $`_I`$
$$_I(s_a,p_a)=\frac{\kappa }{32}A_{abc}s_a\left(s_bs_c3p_bp_c\right).$$
(11)
For simplicity, a one-dimensional field theory approximation for the integral (9) is considered. The extension to the real case is straightforward. We put our system in the interval of size $`L`$, i.e. $`L/2xL/2`$, assuming the limit $`L\mathrm{}`$ in the end of our calculations.
The Fourier decomposition of the fields $`f_a(x)=\{s_a,p_a,\sigma _a,\varphi _a\}`$ inside the interval is
$$f_a(x)=\underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}f_n^a\mathrm{exp}\left(i2\pi \frac{nx}{L}\right).$$
(12)
This corresponds to the periodic boundary conditions $`f_a(L/2)=f_a(L/2)`$. Then we have
$$𝑑x_r(x)_{L/2}^{L/2}𝑑x_r(x)=L_r,$$
(13)
where $`_r=_0+_I`$, and
$$_0=\underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}\left[\frac{G}{2}\left(s_n^as_n^a+p_n^ap_n^a\right)+\sigma _n^as_n^a+\varphi _n^ap_n^a\right],$$
(14)
$$_I=\frac{\kappa }{32}A_{abc}\underset{n,m,k}{}s_n^a\left(s_m^bs_k^c3p_m^bp_k^c\right)\delta _{n+m+k,0}.$$
(15)
The functional integral can be understood as the product of integrals over the Fourier coefficients
$$\underset{\mathrm{}}{\overset{+\mathrm{}}{}}𝒟s_a(x)𝒟p_a(x)\underset{n}{}\underset{\mathrm{}}{\overset{+\mathrm{}}{}}𝑑s_n^a𝑑p_n^a.$$
(16)
Thus, in the perturbative approximation we have
$$I[\sigma ,\varphi ]\mathrm{exp}\left\{iL\widehat{\textcolor[rgb]{1,0,0}{}}_\textcolor[rgb]{1,0,0}{I}(S_n^a,P_n^a)\right\}\underset{n}{}\underset{\mathrm{}}{\overset{+\mathrm{}}{}}𝑑s_n^a𝑑p_n^a\mathrm{exp}\left(iL\textcolor[rgb]{1,0,0}{}_\textcolor[rgb]{1,0,0}{0}\right)$$
(17)
where
$$S_n^a=\frac{i}{L}\frac{}{\sigma _n^a},P_n^a=\frac{i}{L}\frac{}{\varphi _n^a}.$$
(18)
It is convinient to use the normalized functional $`I_N[\sigma ,\varphi ]`$ defined as
$$I_N[\sigma ,\varphi ]=\frac{I[\sigma ,\varphi ]}{I[0,0]}.$$
(19)
The Gaussian functional integrals can be simply evaluated, for instance, one has
$`I[\sigma ]`$ $`=`$ $`{\displaystyle \underset{n}{}}{\displaystyle \underset{\mathrm{}}{\overset{+\mathrm{}}{}}}𝑑s_n\mathrm{exp}\left\{iL{\displaystyle \underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}}\left({\displaystyle \frac{G}{2}}s_ns_n+\sigma _ns_n\right)\right\}`$ (20)
$`=`$ $`I[0]\mathrm{exp}\left\{{\displaystyle \frac{iL}{2G}}{\displaystyle \underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}}\sigma _n\sigma _n\right\}.`$
Thus the functional integration yields
$`I_N[\sigma ,\varphi ]`$ $``$ $`\mathrm{exp}\left\{iL\widehat{}_I(S_n^a,P_n^a)\right\}`$ (21)
$`\times `$ $`\mathrm{exp}\left\{{\displaystyle \frac{iL}{2G}}{\displaystyle \underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}}\left(\sigma _n\sigma _n+\varphi _n\varphi _n\right)\right\}.`$
One should calculate partial derivatives to obtain the effective action $`\mathrm{\Gamma }_{eff}`$
$$I_N[\sigma ,\varphi ]A[\sigma ,\varphi ]e^{i\mathrm{\Gamma }_{eff}}.$$
(22)
We have
$$\mathrm{\Gamma }_{eff}=i\mathrm{ln}A[\sigma ,\varphi ]+\mathrm{\Gamma }_0i\mathrm{ln}\left(1+e^{i\mathrm{\Gamma }_0}\left(e^{i\widehat{\mathrm{\Gamma }}_I}1\right)e^{i\mathrm{\Gamma }_0}\right),$$
(23)
where $`\widehat{\mathrm{\Gamma }}_I=L\widehat{}_I(S_n^a,P_n^a)`$, and $`A[\sigma ,\varphi ]`$ is fixed by the requirement that the effective action $`\mathrm{\Gamma }_{eff}`$ is real. Here $`\mathrm{\Gamma }_0`$ represents the leading order action.
$$\mathrm{\Gamma }_0=\frac{L}{2G}\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}\left(\sigma _n^a\sigma _n^a+\varphi _n^a\varphi _n^a\right).$$
(24)
The logarithm in eq.(23) is a source of $`U(1)`$ breaking corrections which arise as a series in powers of the partial derivatives. Expanding
$$\delta =e^{i\mathrm{\Gamma }_0}\left(e^{i\widehat{\mathrm{\Gamma }}_I}1\right)e^{i\mathrm{\Gamma }_0}=\underset{n=1}{\overset{\mathrm{}}{}}\kappa ^n\delta _n,$$
(25)
one can determine
$$\delta _1=\frac{iL}{32G^3}A_{abc}\underset{n,m,k}{}\sigma _n^a(\sigma _m^b\sigma _k^c3\varphi _m^b\varphi _k^c)\delta _{n+m+k,0},$$
(26)
$`\delta _2`$ $`=`$ $`{\displaystyle \frac{\delta _1^2}{2}}i{\displaystyle \frac{9L}{64(2G)^5}}A_{abc}A_{a\overline{b}\overline{c}}{\displaystyle \underset{m,k,\overline{m},\overline{k}}{}}\delta _{m+k+\overline{m}+\overline{k},0}`$ (27)
$`\times `$ $`\left[\mathrm{\hspace{0.17em}4}\sigma _m^b\sigma _{\overline{m}}^{\overline{b}}\varphi _k^c\varphi _{\overline{k}}^{\overline{c}}+\left(\sigma _m^b\sigma _k^c\varphi _m^b\varphi _k^c\right)\left(\sigma _{\overline{m}}^{\overline{b}}\sigma _{\overline{k}}^{\overline{c}}\varphi _{\overline{m}}^{\overline{b}}\varphi _{\overline{k}}^{\overline{c}}\right)\right]`$
$`+`$ $`{\displaystyle \frac{1}{(8G^2)^2}}{\displaystyle \underset{n}{}}\left(\sigma _n^a\sigma _n^a+\varphi _n^a\varphi _n^a\right){\displaystyle \underset{m}{}}1`$
$`+`$ $`{\displaystyle \frac{3i}{32LG^3}}\left({\displaystyle \underset{m}{}}1\right)^2,`$
and so forth. The factor $`A[\sigma ,\varphi ]`$ must be also expanded
$$A[\sigma ,\varphi ]=1+\kappa \beta _1+\kappa ^2\beta _2+𝒪(\kappa ^3),$$
and, up to the considered order in $`\kappa `$, we have
$$\beta _1=0,\beta _2=\frac{1}{(8G^2)^2}\underset{n}{}\left(\sigma _n^a\sigma _n^a+\varphi _n^a\varphi _n^a\right)\underset{m}{}1.$$
Therefore the perturbative action systematically obtains $`U(1)`$ breaking corrections in $`\kappa ^n`$, which we collect in the corresponding part of the action, $`\mathrm{\Gamma }_n`$,
$$\mathrm{\Gamma }_{eff}=\underset{n=0}{\overset{+\mathrm{}}{}}\mathrm{\Gamma }_n,$$
(28)
where $`\mathrm{\Gamma }_0`$ is given by eq.(24), and
$`\mathrm{\Gamma }_1`$ $`=`$ $`{\displaystyle \frac{\kappa L}{32G^3}}A_{abc}{\displaystyle \underset{n,m,k}{}}\sigma _n^a(\sigma _m^b\sigma _k^c3\varphi _m^b\varphi _k^c)\delta _{n+m+k,0},`$
$`\mathrm{\Gamma }_2`$ $`=`$ $`{\displaystyle \frac{9\kappa ^2L}{64(2G)^5}}A_{abc}A_{a\overline{b}\overline{c}}{\displaystyle \underset{m,k,\overline{m},\overline{k}}{}}\delta _{m+k+\overline{m}+\overline{k},0}[\mathrm{\hspace{0.17em}4}\sigma _m^b\sigma _{\overline{m}}^{\overline{b}}\varphi _k^c\varphi _{\overline{k}}^{\overline{c}}`$ (29)
$`+`$ $`(\sigma _m^b\sigma _k^c\varphi _m^b\varphi _k^c)(\sigma _{\overline{m}}^{\overline{b}}\sigma _{\overline{k}}^{\overline{c}}\varphi _{\overline{m}}^{\overline{b}}\varphi _{\overline{k}}^{\overline{c}})].`$
Here an unessential constant has been omitted in $`\mathrm{\Gamma }_2`$. Taking the infinite – volume limit $`L\mathrm{}`$, one finally obtains
$$\mathrm{\Gamma }_0=\frac{1}{2G}\underset{\mathrm{}}{\overset{+\mathrm{}}{}}𝑑x\left(\sigma _a^2(x)+\varphi _a^2(x)\right),$$
$$\mathrm{\Gamma }_1=\frac{\kappa }{32G^3}A_{abc}\underset{\mathrm{}}{\overset{+\mathrm{}}{}}𝑑x\sigma _a(x)\left[\sigma _b(x)\sigma _c(x)3\varphi _b(x)\varphi _c(x)\right],$$
$$\mathrm{\Gamma }_2=\frac{9\kappa ^2}{64(2G)^5}A_{abc}A_{a\overline{b}\overline{c}}\underset{\mathrm{}}{\overset{+\mathrm{}}{}}dx\{\mathrm{\hspace{0.17em}4}\sigma _b(x)\sigma _{\overline{b}}(x)\varphi _c(x)\varphi _{\overline{c}}(x)$$
$$+[\sigma _b(x)\sigma _c(x)\varphi _b(x)\varphi _c(x)][\sigma _{\overline{b}}(x)\sigma _{\overline{c}}(x)\varphi _{\overline{b}}(x)\varphi _{\overline{c}}(x)]\},$$
and so forth. The infinite volume limit for the coefficient $`\beta _2`$ yields
$$\beta _2=\frac{1}{(8G^2)^2}_{L/2}^{L/2}𝑑x\left(\sigma _a(x)\sigma _a(x)+\varphi _a(x)\varphi _a(x)\right)\frac{1}{L}\underset{m=\mathrm{}}{\overset{+\mathrm{}}{}}1.$$
The integral has a smooth $`L\mathrm{}`$ limit, but contains a local factor, which can be understood as $`\delta `$-function singularity, $`\delta (0)`$, see also , . To show this one needs the following result
$$\underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}\mathrm{exp}\left(i\frac{2\pi x}{L}n\right)=L\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}\delta (xLn)=L\delta (x).$$
(30)
On the last step we took into account that in the considered problem all $`x`$-dependent functions are integrated only in the interval $`L/2xL/2`$ and, therefore, only the term with $`n=0`$ contributes. The infinite value
$$\frac{1}{L}\underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}1=\delta (0)$$
(31)
represents the density of Fourier harmonics in the interval. It must be regularized by cutting an upper part of the spectrum, e.g.
$$\delta (0)_{\text{reg}}=\frac{1}{L}\underset{n=N}{\overset{N}{}}1=\frac{2N+1}{L}$$
(32)
where $`N`$ is large enough. One can fix $`N1`$ without any relation to the size of the box. In this case the limit $`L\mathrm{}`$ leads to the vanishing of the $`\delta `$-function and to $`\beta _2=0`$. Alternatively, one can relate $`N`$ with $`L`$ by introducing the momentum space cutoff $`\mathrm{\Lambda }_E`$: $`N(L)=L\mathrm{\Lambda }_E/(4\pi )1`$. Unlike $`L`$ the cutoff $`\mathrm{\Lambda }_E`$ has an obvious physical meaning giving the scale of momenta relevant for the problem. Indeed, the $`n`$th harmonic has a momentum $`p_n=2\pi n/L`$. The size of the considered box is the difference $`p_Np_N=4\pi N/L=\mathrm{\Lambda }_E`$. This scale cannot be eliminated by taking the limit $`L\mathrm{}`$. One has instead
$$\delta (0)_{\text{reg}}=\frac{\mathrm{\Lambda }_E}{2\pi }+\frac{1}{L},$$
(33)
where only the second term does not contribute in the limit $`L\mathrm{}`$. By introducing the cutoff $`\mathrm{\Lambda }_E`$, we suppose that the density of Fourier harmonics has a finite value which can be fixed phenomenologically. In this case $`\beta _20`$. It is a pure physical matter of argument to choose among these two alternatives.
The Reinhardt and Alkofer estimate takes the integral (9) at one stationary phase point without any integration over the auxiliary variables $`s_a,p_a`$.
The contributions at the stationary phase condition are obtained as
$`_r(r_{\text{st}})`$ $`=`$ $`{\displaystyle \frac{G}{12}}\text{tr}(U_{st}U_{st}^{})+{\displaystyle \frac{1}{6}}\text{tr}(WU_{st}^{}+W^{}U_{st})`$ (34)
$`=`$ $`h_a\sigma _a+{\displaystyle \frac{1}{2}}h_{ab}^{(1)}\sigma _a\sigma _b+{\displaystyle \frac{1}{2}}h_{ab}^{(2)}\varphi _a\varphi _b+\mathrm{}.`$
fulfilling
$$GU_a+W_a+\frac{3\kappa }{32}A_{abc}U_b^{}U_c^{}=0$$
(35)
with covariant combinations
$$W_a=\sigma _a+\mathrm{\Delta }_ai\varphi _a,$$
and definitions
$$\mathrm{\Delta }_a=m_a\widehat{m}_a,U_a=s_aip_a.$$
The field $`U_{st}`$ represents the exact solution of condition (35), obtained by expanding $`s_a,p_a`$ in parity even and odd combinations respectively, of increasing powers of bosonic fields $`\varphi _a,\sigma _a`$
$`s_{st}^a`$ $`=`$ $`h_a+h_{ab}^{(1)}\sigma _b+h_{abc}^{(1)}\sigma _b\sigma _c+h_{abc}^{(2)}\varphi _b\varphi _c`$
$`+`$ $`h_{abcd}^{(1)}\sigma _b\sigma _c\sigma _d+h_{abcd}^{(2)}\sigma _b\varphi _c\varphi _d+\mathrm{}`$
$`p_{st}^a`$ $`=`$ $`h_{ab}^{(2)}\varphi _b+h_{abc}^{(3)}\varphi _b\sigma _c+h_{abcd}^{(3)}\sigma _b\sigma _c\varphi _d`$ (36)
$`+`$ $`h_{abcd}^{(4)}\varphi _b\varphi _c\varphi _d+\mathrm{}.`$
The coefficients $`h_{ab\mathrm{}}^{(k)}`$ depend explicitly on the quark masses and coupling constants $`G,\kappa `$ and are fixed by a series of coupled equations following from (35) and obtained by equating to zero the factors before independent combinations of mesonic fields. Due to recurrency of the considered equations all coefficients are determined once the first one, $`h_a`$, has been obtained. Note that the stationary phase condition (35) has several solutions . If one wants to apply the stationary phase (SP) method consistently, one must take into account the contributions of all critical points (even at leading order). The $`1/N_c`$ counting indicates that the last term in the stationary phase eq.(35) is $`1/N_c`$ suppressed. This argument can be used for the systematic $`1/N_c`$ expansion of the effective action, what actually has been done in our perturbative approach.
We have found strong indications that the estimate corresponds to a resummation of the perturbative series in $`\kappa `$, collecting all tree (non-singular) contributions related with the auxiliary fields $`s_a,p_a`$ of the perturbative approach. Further terms of the ressummed series yield systematically ”loop”-corrections of these fields .
Results and Discussion
We refer to for a detailed derivation and calculation of the leading order low lying pseudoscalar and scalar meson spectra and related observables. Further sets are also presented in . Already in its simplest form, the Lagrangian yields overall quite good fits, even the scalars are reasonable. They have been obtained with the following approximations.
a) Related with the evaluation of the Gausiian functional integral over quarks: knowing that the lack of confinement in the NJL model introduces serious difficulties with the crossing of non-physical thresholds associated with the production of free quark - antiquark pairs, which one may encounter by formally continuing the full Euclidean action to Minkowski space, the heat kernel series has been truncated at second order. We admittedly deviate from the original NJL Lagrangian, however in a way which relies heavily on its symmetries and asymptotic dynamics, which are fully taken into account.
b) Related with the functional integral calculation over the auxiliary variables $`s_a,p_a`$, of cubic order: we have used the SPA with one critical point. We know however that there exist more critical points which should be taken into account even in leading order in SPA.
Finally we have suggested a different way of bosonization, considering the t Hooft determinant as a perturbation to the four-quark intercation. One can relate the perturbative expansion to the SPA with one critical point through a resummation. This is discussed in full detail in . Corrections to the leading order can be obtained in a systematic way. Due to the local character of the interactions these corrections involve a divergent density of harmonics. We discuss how a cutoff can be introduced, and together with the ultraviolet cutoff related with the quark loops, we obtain an effective bosonized Lagrangian. Calculation of the related mass spectra is in progress .
Acknowledgements We are very grateful to the organizers for the invitation and for hosting a very interesting and enjoyable meeting. This work has been supported by grants provided by Fundação para a Ciência e a Tecnologia, POCTI/35304/FIS/2000, POCTI/FNU/50336/2003 and Centro de Física Teórica unit 535/98. This research is part of the EU integrated infrastructure initiative HadronPhysics project under contract No.RII3-CT-2004-506078. A.A. Osipov also gratefully acknowledges the Fundação Calouste Gulbenkian for financial support. |
warning/0507/cond-mat0507180.html | ar5iv | text | # Random multi-index matching problems
## I Introduction
The statistical properties of random combinatorial optimization problems can be studied from a number of angles, with tools depending on the discipline. Recent years have however witnessed a convergence of interests and techniques across mathematics, computer science and statistical physics. An archetype example is the matching problem with random edge weights, defined as follows: suppose one has $`M`$ different jobs and $`M`$ people to perform them, one person per job, and let $`c_{ij}`$ be the cost when job $`i`$ is executed by person $`j`$; the 2-index matching problem consists in assigning jobs to people in such a way as to minimize the total cost. The statistical properties of the optimal matching when the cost $`c_{ij}`$ are drawn independently from a common distribution were found two decades ago using the replica Mézard and Parisi (1985) and the cavity Mézard and Parisi (1986) methods. These two non-rigorous statistical physics approaches have recently been used to tackle a number of computationally more difficult problems such as satisfiability or graph coloring, but the 2-index matching problem sets apart for belonging to one of the very few problems where such predictions have been rigorously confirmed Aldous (2001).
In this work, we take the statistical physics approach and study the properties of a generalization of the 2-index to multi-index matching problems (MIMPs) where the elementary costs are now associated with $`d`$-uplets, representing for example persons, jobs and machines when $`d=3`$. At variance with the 2-index matching, $`d`$-index matching problems with $`d3`$ are NP-hard. We show here that their low lying configurations also have a different, glassy, structure whose description requires replica symmetry to be broken. Remarkably, the replica symmetry breaking scheme differs from the common picture that has emerged from the study of other optimization problems such as the coloring Mulet et al. (2002) and satisfiability problems Mézard and Zecchina (2002). In particular, a naïve application of the 1-RSB cavity method at zero temperature Mézard et al. (2002), which successfully solves these two problems, is here doomed to fail. The reason for this will be traced back to the presence of “hard constraints”. By unraveling this specificity, we put forward arguments whose relevance goes beyond matching problems; they indicate when a similar scenario can be expected on other constrained systems. The particularly simple glassy structure that we find is also of interest from the interdisciplinary point of view: in conjunction with the rigorous formalism available for the 2-index case, it places MIMPs in a choice place for working out a most awaited mathematical understanding of replica symmetry breaking.
The present paper provides an extensive account of our results on the MIMPs, some of which have already been mentioned in Martin et al. (2004). The paper is organized as follows. We first define precisely multi-index matching problems, and briefly review the past approaches from physics, mathematics and computer science that were developed mainly to address the 2-index case. Then we start our statistical study by establishing the scaling of the minimal cost as a function of the number of variables and by providing a lower bound from an annealed calculation. A large part of the paper is then devoted to present our implementation of the cavity method to matching problems, including a detailed discussion of its relations with the rigorous formalism proposed by Aldous; we explain why and how replica symmetry must be broken when $`d3`$, in order to account for the presence of a frozen glassy phase. Finally, the last section is dedicated to a numerical analysis of small samples that provides support to the proposed scenario.
## II Multi-index matchings
### II.1 Definitions
Two classes of MIMPs can be distinguished, $`d`$-partite matching problems and simple $`d`$-matching problems, whose asymptotic properties will be shown to be related. We first start with the $`d`$-partite matching problem that corresponds to the version alluded to in the introduction. An instance consists of $`d`$ sets, $`A_1`$,…, $`A_d`$, of $`M`$ nodes each, and a cost $`c_a`$ is associated with every $`d`$-uplet $`a=\{i_1,\mathrm{},i_d\}A_1\times \mathrm{}\times A_d`$. Graphically, it is represented by a factor graph as shown in Fig. 2 with hyperedges (factor nodes) joining exactly one node from each ensemble. A matching $``$ is a maximal set of disjoint hyperedges, such that each node is associated to one and only one hyperedge of the matching; it can be described by introducing an occupation number $`n_a\{0,1\}`$ on each hyperedge $`a`$, with the correspondence
$$an_a=1.$$
(1)
The condition for a set of hyperedges to be a matching can then be written
$$r=1,\mathrm{},d,i_rA_r,\underset{a:i_ra}{}n_a=1.$$
(2)
The $`d`$-partite matching problem consists in finding the matching with minimal total cost,
$$C_M^{(d)}=\underset{\{n_a\}}{\mathrm{min}}\underset{a}{}c_an_a$$
(3)
with the $`\{n_a\}`$ subject to the constraints (2). We consider here the random version of the problem, where the costs $`c_a`$ are independent identically distributed random variables taken from a distribution $`\rho (c)`$, and we are interested in the typical value of an optimal matching in the $`M\mathrm{}`$ limit. For definiteness, we take for $`\rho `$ the uniform distribution in $`[0,1]`$, but the asymptotic properties of $`d`$-matchings depend only on the behavior of $`\rho `$ close to $`c=0`$, and are identical for all distributions $`\rho `$ with $`\rho (c)1`$ as $`c0`$, such as the exponential distribution, $`\rho (c)=e^c`$. The case $`\rho (c)c^r`$, $`r>0`$, can be treated along the same lines, but gives different quantitative results.
A variant of this setup is the simple $`d`$-matching problem, where a unique set of $`N`$ nodes, with $`N`$ being a multiple of $`d`$, is considered and a cost is associated to each $`d`$-uplet of nodes (see Fig. 2). The $`d`$-partite case can be seen as a particular instance of a simple $`d`$-matching problem where the hyperedges joining more than one node of any $`A_i`$ are given an infinite cost. Simple $`d`$-matchings problems are formulated as finding
$$L_N^{(d)}=\underset{\{n_a\}}{\mathrm{min}}\underset{a}{}c_an_a$$
(4)
under the constraints
$$i=1,\mathrm{},N,\underset{a:ia}{}n_a=1.$$
(5)
Before presenting our analysis of random matching problems by means of an adaptation of the cavity method for finite connectivity statistical physics models, we briefly review past approaches to the subject, with an emphasis on open questions that motivated the present study.
### II.2 Physical approach
The 2-index matching problem was the first combinatorial optimization problem to be tackled with the replica method, an analytical method initially developed in the context of spin glasses Mézard et al. (1987). In the paper Mézard and Parisi (1985), Mézard and Parisi analyzed both the simple and bipartite matching problems for cost distributions $`\rho `$ with $`\rho (c)c^r`$ as $`c0`$. Using replica theory within a replica symmetric Ansatz, they derived the minimal total cost; thus, for the bipartite matching with $`r=0`$, they predicted $`lim_M\mathrm{}𝒞_M^{(2)}=\pi ^2/6`$; moreover, they obtained the distribution of cost in the optimal matching. Support in favor of their prediction has first come from numerical results and from an analytical study of the stability of the replica symmetric solution Mézard and Parisi (1987); Parisi and Ratiéville (2002). This last analysis further yields the leading corrections of order $`1/N`$ for the value of the minimum matching.
Interestingly, the same results can be reobtained using a variant of the cavity method based on a representation of self-avoiding walks using $`m`$-component spins Mézard and Parisi (1986). This alternative formulation, avoiding the bold prescriptions of replica theory, furthermore suggests that, if the cost of the hyperedges connected to a given node are ordered from the lowest to the highest, the probability for the $`k`$-th hyperedge to be included in the optimal matching is $`2^k`$ Parisi and Ratiéville (2001), as first conjectured from a numerical study Houdayer et al. (1998).
### II.3 Mathematical approach
Replica theory, while a powerful tool to obtain analytical formulae, is not a rigorously controlled method, and its predictions have only the status of conjectures within the usual mathematical standards. For the 2-index matching problem with $`r=0`$ however, the results mentioned above (value of the optimal matching, distribution of costs, and probability of inclusion of $`k`$-th hyperedge) have all been confirmed by a rigorous derivation, due to Aldous Aldous (2001). His contribution also includes the proof an asymptotic essential uniqueness property that mathematically expresses the fact that replica symmetry indeed holds for this problem. The weak convergence approach Aldous and Steele (2003) on which the proof is built is closely related to the cavity method we will employ, and the relations between the two formalisms will be discussed in Sec. IV.4. Confirmation of the $`\zeta (2)=\pi ^2/6`$ value for the bipartite assignment problem also comes from the recent proofs Linusson and Wastlund (2003); C. Nair and Sharma (2003) of a more general conjecture formulated by Parisi Parisi (1998); this conjecture states that, for the bipartite matching with exponential distribution of the costs, $`\rho (c)=e^c`$, the mean optimal matching for finite $`M`$ is $`_{k=1}^Mk^2`$.
These mathematical contributions are part of a more ambitious program aiming at developing rigorous proofs and possibly a rigorous framework of the replica and cavity methods. Interestingly, Talagrand, one of the prominent advocate of this program, devotes the last chapter of his book on the subject Talagrand (2003) to the 2-index matching problem, stressing that, in spite of the major advances mentioned, it stays a particularly challenging issue. Indeed, finite temperature properties have so far resisted to mathematical investigations, even in the limit of high temperature, that has been successfully addressed in other spin-glass like models Talagrand (2003). We shall comment on the peculiarities of matchings with respect to other constrained systems in Sec. V.2. It is our hope that our work not only provides new challenging conjectures, but also suggests some hints for solving unanswered preexisting mathematical questions.
### II.4 Computer science approach
If analytical studies of random $`d`$-matchings by statistical physicists and mathematicians have been restricted up to now to the $`d=2`$ case, $`d`$-index matching problems with $`d>2`$ have a longer history in the computer science community. $`d`$-partite extensions of the bipartite matching problem were introduced in 1968 under the name of multidimensional assignment problems Pierskalla (1968); they are also referred in the literature as multi-index assignment problems, and, more specifically, as multi-index axial assignment problems (to distinguish them from the so-called planar versions Spieksma (2000); Burkard (2002)). MIMPs, as we call them (for multi-index matching problems), have a number of practical applications. The most commonly cited one is for data association in connection with multi-target tracking Poore (1994). Besides a major interest for real-time air traffic control, such approaches are for instance helpful for tracking elementary particles in high energy physics experiments Pusztaszeri et al. (1995).
From the algorithmic complexity point of view, matching problems have also a pioneering role since the 3-index matching problem was among the first 21 problems to be proved NP-complete Karp (1972). In contrast, polynomial algorithms are known that solve 2-index matching problems Papadimitriou and Steiglitz (1982). Note that being based on a worst case analysis, NP-hardness is however only a necessary condition for hard typical complexity, which is the issue which interests us here. Due to their intrinsic algorithmic difficulty and to the broad range of their applications, generalized assignment problems are the subject of numerous studies in the computer science community; we refer to the reviews Spieksma (2000); Burkard (2002) for additional information and references.
## III Scaling and a lower bound
The first task in studying random optimization problems is to determine the scaling of the optimal cost with the number of variables Vannimenus and Mézard (1984). Here, we address this issue for the two variants of MIMPs, the multi-partite and simple multi-index matching problems. The scaling is inferred from an heuristic argument, and confirmed by an annealed calculation (first moment method) yielding a lower bound. This leads us to a statistical physics formulation that encompasses the two versions of MIMPs.
### III.1 Scaling
The statistical physics approach of combinatorial optimization problems consists in defining the energy $`E()`$ of each admissible solution, here a $`d`$-matching $``$, as its total cost, $`E()=_ac_a`$, and in determining the minimal total cost, identified with the ground-state energy, by looking at the zero temperature properties of the system. For $`d`$-matchings, the corresponding Hamiltonian
$$[\{n_a\}]=\underset{a}{}c_an_a$$
(6)
defines a lattice gas model, where the particles are occupying the hyperedges. The constraints (2) or (5) implement a hard-core interaction between the particles: two “neighboring” hyperedges are not allowed to be occupied simultaneously. To have a sensible statistical physics model, the ground state has to be extensive, i.e., proportional to $`M`$ in the $`d`$-partite case, and to $`N`$ in the simple case. We propose here a heuristic argument to determine how $`𝔼[C_M^{(d)}]`$ and $`𝔼[L_N^{(d)}]`$ scale with $`M`$ and $`N`$ respectively, where $`𝔼[]`$ represents the average over the different realizations of the costs. The central (local) quantity that monitors the scaling behavior is the number of hyperedges to which a given node belongs, noted $`W_\mathrm{\Lambda }^{(d)}`$ ($`\mathrm{\Lambda }=M`$ or $`N`$). Indeed, with the costs uniformly distributed in , the lowest costs to which a node can be associated are of order $`1/W_\mathrm{\Lambda }^{(d)}`$ and the optimal matching is expected to scale like $`\mathrm{\Lambda }/W_\mathrm{\Lambda }^{(d)}`$. Thus, for $`d`$-partite matchings, $`W_M^{(d)}=M^{d1}`$ and $`𝔼[C_M^{(d)}]M^{2d}`$, while for simple $`d`$-matchings, $`W_N^{(d)}=\left(\genfrac{}{}{0pt}{}{N1}{d1}\right)`$ and $`𝔼[L_N^{(d)}](d1)!N^{2d}`$. We will therefore be interested in computing the (finite) quantities
$$\begin{array}{cc}\hfill 𝒞^{(d)}& =\underset{M\mathrm{}}{lim}M^{d2}𝔼[C_M^{(d)}],\hfill \\ \hfill ^{(d)}& =\underset{N\mathrm{}}{lim}\frac{N^{d2}}{(d1)!}𝔼[L_N^{(d)}].\hfill \end{array}$$
(7)
The factor $`(d1)!`$ in the second definition is meant to reflect the different number of hyperedges to which a given node can connect, in the $`d`$-partite and simple versions (this difference is absent when $`d=2`$). With this convention we will find the equality $`𝒞^{(d)}=d^{(d)}`$, where the remaining $`d`$ factor merely comes from the fact that the total number of nodes is $`N`$ for simple $`d`$-matchings, but is $`dM`$ for $`d`$-partite matchings.
### III.2 Annealed approximation
When energies are extensive in the size $`N`$ of the system, the equilibrium properties of a statistical physics model are entirely encoded in the partition function, $`Z_N(\beta )=_{}e^{\beta E()}`$, or, equivalently, in its logarithm, the free energy $`F_N(\beta )\mathrm{log}[Z_N(\beta )]/\beta `$. The free energy depends on the realization of the elementary costs, but it is expected to be a self-averaging quantity, i.e., such that the free-energy density $`f(\beta )=lim_N\mathrm{}F_N(\beta )/N`$ exists and is independent of the sample. The self-averaging property is proved for $`d=2`$ Aldous (1990), and we assume here that it holds for $`d3`$ as well. The value of the optimal matching is given by the ground state energy, obtained as $`lim_\beta \mathrm{}f(\beta )`$, where the free energy is calculated by performing a quenched average of the partition function, $`𝔼[\mathrm{ln}Z]`$, with $`𝔼[]`$ referring to the average with respect to the realization of the elementary costs.
A much simpler calculation is the annealed average, $`\mathrm{ln}𝔼[Z]`$. Due to the concavity of the logarithm, it yields a lower bound on the correct quenched free energy, $`f_{\mathrm{an}}(\beta )\mathrm{ln}𝔼[Z]/(N\beta )𝔼[\mathrm{ln}Z]/(N\beta )f(\beta )`$. In fact, since the entropy $`s(\beta )=\beta ^2_\beta f(\beta )`$ is necessarily positive for a system with discrete degrees of freedom, the free energy $`f(\beta )`$ must be an increasing function, and a tighter lower bound can be inferred for the ground-state energy Vannimenus and Mézard (1984),
$$\underset{\beta \mathrm{}}{lim}f(\beta )\underset{\beta >0}{sup}f_{\mathrm{an}}(\beta ).$$
(8)
These considerations are made under the hypothesis that the energies, or equivalently the temperature $`\beta `$, are correctly scaled with $`N`$, so that $`lim_\beta \mathrm{}f(\beta )`$ is indeed finite. Reciprocally, requiring the annealed free energy to be extensive provides us with the appropriate scaling of $`\beta `$. For $`d`$-index matching problems, we have
$$𝔼[Z]=𝔼\left[\underset{\{n_a\}}{}e^{\beta _ac_an_a}\right]=(\mathrm{\#})𝔼[e^{\beta c_a}]^{\mathrm{\#}\{a\}}$$
(9)
where $`\mathrm{\#}`$ denotes the total number of possible matchings and $`\mathrm{\#}\{a\}`$ the number of hyperedges contained in a given matching. For $`d`$-partite matchings, $`\mathrm{\#}=(M!)^{d1}`$ and $`\mathrm{\#}\{a\}=M`$. To enforce the correct scaling of the free energy, we anticipate a rescaling in temperature of the form $`\beta =M^\alpha \widehat{\beta }`$, yielding
$$\mathrm{ln}𝔼[Z]=[d1\alpha ]M\mathrm{ln}M[\mathrm{ln}\widehat{\beta }+d1]M+o(M).$$
(10)
An extensive annealed free energy is therefore obtained by taking $`\alpha =d1`$, in which case
$$f_{\mathrm{an}}^{(d\mathrm{part})}(\widehat{\beta })=\frac{\mathrm{ln}\widehat{\beta }+d1}{\widehat{\beta }}.$$
(11)
The scaling $`1/(M\widehat{\beta })M^{2d}`$ we obtain corresponds to one introduced in Eq. (7). For simple $`d`$-index matchings, $`\mathrm{\#}=N!/[(N/d)!(d!)^{N/d}]`$ and $`\mathrm{\#}\{a\}=N/d`$. Rescaling the temperature as $`\beta =N^{d1}\widehat{\beta }`$, we get
$$\mathrm{ln}𝔼[Z]=[\mathrm{ln}\widehat{\beta }+d1\mathrm{ln}(d1)!]N/d+o(N).$$
(12)
To make contact with the $`d`$-partite case however, we adopt a slightly different scaling, $`\beta =N^{d1}\stackrel{~}{\beta }/(d1)!`$, so that
$$f_{\mathrm{an}}^{(\mathrm{simple})}(\stackrel{~}{\beta })=\frac{\mathrm{ln}\stackrel{~}{\beta }+d1}{d\stackrel{~}{\beta }}=\frac{1}{d}f_{\mathrm{an}}^{(d\mathrm{part})}(\widehat{\beta }=\stackrel{~}{\beta }).$$
(13)
This annealed calculation illustrates the correspondence between the $`d`$-partite and simple $`d`$-matchings stated in the previous section. Apart for the trivial factor $`d`$, corresponding to the relation $`N=dM`$, the equality is obtained by normalizing differently $`\widehat{\beta }`$ and $`\stackrel{~}{\beta }`$, thereby accounting for the difference in the number of hyperedges a given node locally sees \[extra factor $`(d1)!`$ in Eq. (7)\]. The annealed free energy is a concave function with maximum for $`\widehat{\beta }_d^{}=e^{2d}`$ so that we get lower bounds $`𝒞^{(d)}e^{d2}`$ and $`^{(d)}e^{d2}/d`$.
### III.3 Statistical physics reformulation
From now on, we will cease distinguishing between $`d`$-partite and simple $`d`$-matchings, and consider a unique statistical physics model that describes both problems in a common framework. Our approach is indeed based on the cavity method Mézard and Parisi (2001) for which only the local properties at the level of each node are relevant, and we have seen that by making the appropriate scalings of $`\beta `$, we can match the local properties of both models. The Hamiltonian we consider is
$$[\{n_a\}]=\underset{a}{}\xi _an_a,$$
(14)
with the $`\xi _aM^{d1}c_a`$ uniformly distributed in $`[0,M^{d1}]`$ for the $`d`$-partite case, and $`\xi _aN^{d1}c_a/(d1)!`$ in $`[0,N^{d1}/(d1)!]`$ for the simple case. The (inverse) temperature, denoted by $`\beta `$ to simplify, will correspond to $`\widehat{\beta }`$ for the $`d`$-partite case and $`\stackrel{~}{\beta }`$ for the simple case. The only remaining difference kept is the factor $`d`$ between the two free energies, accounting for the relation $`N=dM`$. Unless explicitly stated, the formulae to be given hold for the simple version ; to get the $`d`$-partite counterparts, one has consequently to multiply by $`d`$ the intensive quantities.
## IV Replica symmetric solution
The approach we adopt to treat the $`d`$-matching problems is the cavity method recently developed to solve statistical physics models defined on finite connectivity graphs Mézard and Parisi (2001). This section explains the formalism of the replica symmetric solution for general $`d`$. While the correctness of the replica symmetric approach is a mathematical fact when $`d=2`$, we show that it leads to some inconsistency when $`d=3`$, requiring replica symmetry to be broken.
### IV.1 From complete to dilute graphs
The hypergraph on which an instance of the simple $`d`$-matching problem is defined is complete, in the sense that every possible hyperedge arises once and is given a random cost. However, the factor nodes with the smallest elementary costs are more likely to belong to the optimal matching; for instance the probability that the $`k`$-th most costly hyperedge originating from a given node will be included in the optimal 2-matching is $`2^k`$ Aldous (2001). This suggests that hyperedges with large costs can be ignored while retaining most of the structure relevant to the determination of the optimal matching. Eliminating hyperedges results in a diluted hypergraph, where each node is connected to only a restricted number of hyperedges. From this point of view, in spite of being defined on a complete graph, random matchings are effectively closer to statistical physics models defined on finite connectivity random graphs. In fact, such a feature already transpired from the initial replica treatment Mézard and Parisi (1985) of $`2`$-index matchings where all the multioverlaps $`Q_{a_1\mathrm{}a_p}`$ were required, and not only the two replica overlaps $`Q_{a_1a_2}`$, like in usual Curie-Weiss mean field models of disordered systems Mézard et al. (1987).
To exploit the underlying diluted structure, one possible method is to introduce a cut-off $`C`$, suppress all nodes with rescaled cost $`\xi _a>C`$, solve the matching problem on the diluted hypergraph, and finally send $`C\mathrm{}`$. The hypergraph obtained by this procedure is Poissonian: if $`\xi _1,\xi _2,\mathrm{}`$ are the costs ordered in increasing sequence of the hyperedges connected to a given node, the probability for the connectivity to be $`k`$ is
$$p_k=\mathrm{Prob}[\xi _1<\mathrm{}<\xi _k<C<\xi _{k+1}<\mathrm{}]=\left(\genfrac{}{}{0pt}{}{W_\mathrm{\Lambda }^{(d)}}{k}\right)\left(\frac{C}{W_\mathrm{\Lambda }^{(d)}}\right)^k\left(1\frac{C}{W_\mathrm{\Lambda }^{(d)}}\right)^{W_\mathrm{\Lambda }^{(d)}k}\frac{C^k}{k!}e^C,$$
(15)
with $`W_\mathrm{\Lambda }^{(d)}`$ giving the number of hyperedges to which a node is connected, as in Sec. III.2. Diluting the complete graph has a major drawback however: the diluted hypergraph typically does not allow any matching at all, since for instance there is always a finite probability $`e^C`$ that a given node is isolated.
To circumvent this problem, we come back to the model on the complete graph and start by weakening the constraints, allowing a node not to belong to a matching, at the expense of paying an extra cost. In more physical terms, we view a matching as the close-packing limit of a lattice gas model whose particles are subject to hard-core interactions: particles can occupy the hyperedges but two hyperedges connected through a node can not both admit a particle. We introduce a grand-canonical Hamiltonian
$$_\mu [\{n_a\}]=\underset{a}{}\xi _an_ad\mu \underset{a}{}n_a=\underset{a}{}(\xi _ad\mu )n_a.$$
(16)
where $`d\mu `$ is a chemical potential per hyperedge ($`\mu `$ per node). In the limit $`\mu \mathrm{}`$, the maximum number of hyperedges is occupied by a particle and we recover the matching problem. For finite $`\mu `$ however, the constraints reflecting the hard-core repulsion are
$$i,\underset{a:ia}{}n_a1,$$
(17)
to be compared with the hard constraints of Eq. (5), recovered only in the $`\mu \mathrm{}`$ limit. Each value of $`\mu `$ defines an optimization problem whose minimum energy $`E_\mu `$ corresponds to a zero temperature limit $`\beta \mathrm{}`$. The solution of the matching problem thus appears as the result of a double limit, $`\beta \mathrm{}`$ and $`\mu \mathrm{}`$. The point is that a diluted structure is now naturally associated with the system at finite $`\mu `$. Indeed since $`_\mu `$ is minimized by taking $`n_a=0`$ whenever $`\xi _a>d\mu `$, the ground state is unaffected if all hyperedges $`a`$ with $`\xi _a>d\mu `$ are suppressed, yielding the Poissonian hypergraph considered above with $`C=d\mu `$. This construction allows us to formulate the initial MIMP as the limit $`\mu \mathrm{}`$ of optimization problems defined on Poissonian graphs with increasing mean connectivity $`d\mu `$. We will give in Sec. IV.4 an alternative construction based on regular graphs.
### IV.2 Cavity method
The problem at finite $`\mu `$ defined on a Poissonian hypergraph can be studied by means of the cavity method as developed for finite connectivity graphs Mézard and Parisi (2001); one of the main advantages of this method over the replica method Mézard and Parisi (1985) or previous versions of the cavity method Mézard and Parisi (1986) is that it allows for a practical investigation of replica symmetry breaking (RSB). Since we are interested in the ground-state properties, the cavity method directly at zero temperature seems particularly well suited Mézard and Parisi (2003). However, it will turn out to be necessary to get the finite temperature equations as well, and we therefore work at finite $`\beta `$, postponing the discussion of the $`\beta \mathrm{}`$ limit to the next section.
In strong analogy with Aldous’ framework (see Sec. IV.4), the RS cavity method associates the diluted hypergraph with an infinite tree or, stated differently, a tree with self-consistent boundary conditions. The starting point is however finite rooted trees, that is trees with a singularized node $`i`$ called the root. Consider for instance the part of a tree represented in Fig. 4: given a hyperedge $`a`$ and one of its connected nodes $`i`$ (relation noted $`ia`$), we call $`Z^{(ai)}`$ the partition function of the system defined on the rooted-tree with root $`a`$ resulting from the removal of $`i`$. To express it in terms of the partition functions $`Z^{(bj)}`$ where $`j`$ refers to the nodes, connected to $`a`$, but distinct from $`i`$ (noted $`jai`$), we decompose $`Z^{(ai)}`$ as $`Z^{(ai)}=Z_0^{(ai)}+Z_1^{(ai)}`$, where $`Z_0^{(ai)}`$ and $`Z_1^{(ai)}`$ are the conditional partition functions where the root $`a`$ is either constrained to be empty or occupied by a particle. As an intermediate stage in the recursion, we also introduce $`Y_0^{(ja)}`$ and $`Y_1^{(ja)}`$, which are defined similarly to $`Z_0^{(ai)}`$ and $`Z_1^{(ai)}`$, but for rooted-trees whose root is the node $`j`$, in absence of the hyperedge $`a`$: the index 1 means that $`j`$ is already matched and the index 0 that it is not. With the notations of Fig. 4, we have the relation
$$\begin{array}{cc}\hfill Z_0^{(ai)}& =\underset{jai}{}\left(Y_0^{(ja)}+Y_1^{(ja)}\right),\hfill \\ \hfill Z_1^{(ai)}& =e^{\beta (\xi _ad\mu )}\underset{jai}{}Y_0^{(ja)},\hfill \\ \hfill Y_0^{(ja)}& =\underset{bja}{}Z_0^{(bj)},\hfill \\ \hfill Y_1^{(ja)}& =\underset{bja}{}Z_1^{(bj)}\underset{cj\{a,b\}}{}Z_0^{(cj)}.\hfill \end{array}$$
(18)
These formulae have simple interpretations: for instance, the first line means that when $`a`$ is empty, the neighboring nodes $`jai`$ can be equally matched or not with upstream hyperedges, while the second line means that when $`a`$ is occupied, it generates a cost $`\xi _ad\mu `$ and requires the nodes $`jai`$ to not be matched.
From the conditional partition functions, we define the cavity fields
$$\begin{array}{cc}\hfill e^{\beta x^{(ja)}}& e^{\beta \mu }\frac{Y_0^{(ja)}}{Y_0^{(ja)}+Y_1^{(ja)}},\hfill \\ \hfill e^{\beta u^{(ai)}}& e^{\beta (\xi _a\mu )}\frac{Z_1^{(ai)}}{Z_0^{(ai)}}.\hfill \end{array}$$
(19)
These definitions are made to insure a proper scaling when $`\mu \mathrm{}`$ and recover quantities used in previous studies for $`d=2`$. Note however that for finite $`\beta `$, it is more natural to introduce $`\psi ^{(ja)}\mathrm{exp}[\beta (x^{(ja)}\mu )]`$ interpreted as the probability that node $`j`$ not matched in the absence of $`a`$ (or equivalently that $`j`$ is associated to $`a`$ in a matching); this alternative notation will turn out to be particularly convenient when discussing the freezing phenomenon, in Sec. V.2. On a given rooted tree, it follows from Eq. (18) that the fields attached to the different oriented edges are related by the following message-passing rules,
$$\begin{array}{cc}\hfill u^{(ai)}& =\underset{jai}{}x^{(ja)},\hfill \\ \hfill x^{(ja)}& =\frac{1}{\beta }\mathrm{ln}\left(e^{\beta \mu }+\underset{bja}{}e^{\beta (\xi _au^{(bj)})}\right).\hfill \end{array}$$
(20)
The limit of infinite rooted trees is taken implicitly by considering the stationary distribution $`𝒫(x)`$ that is assumed to result from the repeated iteration of the message passing relations. By definition $`𝒫(x)`$ is a distribution of cavity fields over the different oriented edges that satisfies the following self-consistent equation, called the RS cavity equation,
$$𝒫(x^{(0)})=𝔼_{k,\xi }\underset{a=1}{\overset{k}{}}\underset{j_a=1}{\overset{d1}{}}dx^{(j_a)}𝒫(x^{(j_a)})\delta \left(x^{(0)}\widehat{x}^{(k,\xi )}[\{x^{(j_a)}\}]\right)$$
(21)
where the function $`\widehat{x}^{(k,\xi )}`$ is defined according to Eq. (20) as
$$\widehat{x}^{(k,\xi )}[\{x^{(j_a)}\}]\frac{1}{\beta }\mathrm{ln}\left(e^{\beta \mu }+\underset{a=1}{\overset{k}{}}e^{\beta (\xi _{j_a=1}^{d1}x^{(j_a)})}\right),$$
(22)
and the expectation $`𝔼_{k,\xi }`$ expresses the average over the disorder, which includes both an average over the connectivity $`k`$ and over the rescaled costs $`\xi `$,
$$𝔼_{k,\xi }[F^{(k)}(\{\xi _a\})]\underset{k=0}{\overset{\mathrm{}}{}}\frac{(d\mu )^ke^{d\mu }}{k!}\underset{a=1}{\overset{k}{}}\left(\frac{1}{d\mu }_0^{d\mu }𝑑\xi _a\right)F^{(k)}(\xi _1,\mathrm{},\xi _k).$$
(23)
The RS cavity equation (21) can be solved by a population dynamics algorithm, whose principle is presented in Appendix A; the resulting distribution $`𝒫(x)`$ for $`d=3`$ and different $`\beta `$ is shown in Fig. 4. $`𝒫(x)`$ contains all the information on the equilibrium properties and, in particular, allows one to compute the free-energy density. It can be derived from the Bethe approximation which produces on a given hypergraph the formula
$$f(\beta )=\frac{1}{N}\left[\underset{i}{}\mathrm{\Delta }F^{(i+ai)}(\beta )\underset{a}{}(\mathrm{}_a1)\mathrm{\Delta }F^{(a)}(\beta )\right]$$
(24)
where $`\mathrm{}_a`$ is the degree of hyperedge $`a`$, which here is $`\mathrm{}_a=d`$ independently of $`a`$. The shifts $`\mathrm{\Delta }F^{(i+ai)}(\beta )`$ and $`\mathrm{\Delta }F^{(a)}(\beta )`$ correspond respectively to the free-energy shift induced by the addition of a node $`i`$ together with its connected hyperedges $`ai`$, and to the free-energy shift induced by the addition of hyperedge $`a`$. They are given by
$$\begin{array}{cc}\hfill e^{\beta \mathrm{\Delta }F^{(i+ai)}(\beta )}& =\frac{Y_0^{(i)}+Y_1^{(i)}}{_{ai}_{jai}\left(Y_0^{(j)}+Y_1^{(j)}\right)}=e^{\beta \mu }+\underset{ai}{}e^{\beta (\xi _a_{jai}x^{(ja)})},\hfill \\ \hfill e^{\beta \mathrm{\Delta }F^{(a)}(\beta )}& =\frac{Z_0^{(a)}+Z_1^{(a)}}{_{jai}\left(Y_0^{(j)}+Y_1^{(j)}\right)}=1+e^{\beta (\xi _a_{ja}x^{(ja)})},\hfill \end{array}$$
(25)
where we introduced the analogs of the partitions functions for rooted tree, but on the complete trees:
$$\begin{array}{cc}\hfill Z_0^{(a)}& =\underset{ja}{}\left(Y_0^{(ja)}+Y_1^{(ja)}\right),\hfill \\ \hfill Z_1^{(a)}& =e^{\beta (\xi _ad\mu )}\underset{ja}{}Y_0^{(ja)},\hfill \\ \hfill Y_0^{(j)}& =\underset{bj}{}Z_0^{(bj)},\hfill \\ \hfill Y_1^{(j)}& =\underset{bj}{}Z_1^{(bj)}\underset{cjb}{}Z_0^{(cj)}.\hfill \end{array}$$
(26)
Physically, $`Z_1^{(a)}/(Z_0^{(a)}+Z_1^{(a)})`$ gives the probability for the hyperedge $`a`$ to be included in the matching. By averaging over the realizations of the disorder, since the mean number of hyperedges per nodes is $`\mu `$, we get
$$f_{RS}(\beta )=𝔼[\mathrm{\Delta }F^{(i+ai)}(\beta )](d1)\mu 𝔼[\mathrm{\Delta }F^{(a)}(\beta )]$$
(27)
with explicitly
$$\begin{array}{cc}& 𝔼[\mathrm{\Delta }F^{(i+ai)}(\beta )]=\frac{1}{\beta }𝔼_{k,\xi }\underset{a=1}{\overset{k}{}}\underset{j_a=1}{\overset{d1}{}}dx^{(j_a)}𝒫(x^{(j_a)})\mathrm{ln}\left(e^{\beta \mu }+\underset{a=1}{\overset{k}{}}e^{\beta (\xi _a_{j_a=1}^{d1}x^{(j_a)})}\right),\hfill \\ & 𝔼[\mathrm{\Delta }F^{(a)}(\beta )]=\frac{1}{\beta }𝔼_\xi \underset{j=1}{\overset{d}{}}dx^{(j)}𝒫(x^{(j)})\mathrm{ln}\left(1+e^{\beta (\xi _a_{j=1}^dx_j)}\right).\hfill \end{array}$$
(28)
### IV.3 Integral relations
The $`\mu \mathrm{}`$ limit can be taken explicitly. The corresponding equations generalize the formulae established by Mézard and Parisi in their first treatment of the 2-index matching problem Mézard and Parisi (1985). For general $`d`$, they are
$$\begin{array}{cc}\hfill G(l)& =\frac{1}{\beta }_{\mathrm{}}^+\mathrm{}\underset{j=1}{\overset{d1}{}}dy_je^{G(y_j)}B_d\left(l+\underset{j=1}{\overset{d1}{}}y_j\right),\hfill \\ \hfill B_d(x)& \underset{p=1}{\overset{\mathrm{}}{}}\frac{(1)^{p1}p^{d2}e^{px}}{(p!)^d}.\hfill \end{array}$$
(29)
Given $`G(\mathrm{})`$, the energy $`ϵ(\beta )`$ and entropy $`s(\beta )`$ are
$$\begin{array}{cc}\hfill ϵ(\beta )& =\frac{1}{\beta d}_{\mathrm{}}^+\mathrm{}𝑑lG(l)e^{G(l)},\hfill \\ \hfill s(\beta )& =_{\mathrm{}}^+\mathrm{}𝑑l\left[e^{e^l}e^{G(l)}\right]\frac{d2}{d}_{\mathrm{}}^+\mathrm{}𝑑lG(l)e^{G(l)},\hfill \end{array}$$
(30)
and the free energy is obtained as $`f(\beta )=ϵ(\beta )s(\beta )/\beta `$. The relation between the function $`G(l)`$ and the order parameter $`𝒫(x)`$ is, up to a change of variable, a Laplace transform,
$$e^{G(l)}=_{\mathrm{}}^+\mathrm{}𝑑x𝒫(x)e^{e^{l\beta x}}.$$
(31)
From the practical point of view of numerically solving the cavity equations, the finite $`\mu `$ cavity equations are however easier to handle than these compact formulae.
### IV.4 Zero temperature limit
In view of an extension of the mathematical approach from 2-index to $`d`$-index matchings with $`d>2`$, it is interesting to discuss in some details the relations between our equations and those used by Aldous in his rigorous study of the 2-index matching problem Aldous (2001). Aldous’ formalism is obtained from our RS cavity equations by taking the zero temperature, $`\beta \mathrm{}`$. When $`\beta \mathrm{}`$, Eqs. (20) become
$$\begin{array}{cc}\hfill u^{(ai)}& =\underset{jai}{}x^{(ja)},\hfill \\ \hfill x^{(ja)}& =\underset{bja}{\mathrm{min}}(\xi _bu^{(bj)}).\hfill \end{array}$$
(32)
Taking $`\mu =\mathrm{}`$ and $`d=2`$ leads to the recursive distributional equation Aldous and Bandyopadhyay (2004),
$$x^{(a)}=\underset{b}{\mathrm{min}}\left(\xi _bx^{(b)}\right)$$
(33)
on which Aldous’ work is based Aldous (2001). A difference is however that its costs $`\xi _b`$ derive from a Poisson point process (the uniform distribution does not make sense when $`\mu =\mathrm{}`$). This Poisson process can nonetheless be related to our formalism by implementing a variant of the cut-off procedure. Consider selecting at each step of the cavity recursion the $`k`$ parents of smallest costs, $`k`$ being now fixed. Then the successive $`k`$ costs are distributed according to a Poisson process with rate one. Nonetheless, while the cavity recursion is perfectly well defined, the corresponding system on a given hypergraph does not make sense: a hyperedge may belong to the the list of the hyperedges with the $`k`$-th smallest costs for one of its node but not for an other one. This is why we introduced the version with a cut-off on the costs, which constitutes for finite $`\mu `$ a perfectly sensible statistical physics model. From a purely formal point of view the version with cut-off on the number of connected clauses works as well, and provides an alternative formulation for numerically solving the cavity equations (see Appendix A for the details and Fig. 10 for an illustration).
The cavity fields at zero temperature have an interpretation in terms of differences in ground-state energies. The cavity field $`x^{(ja)}`$ corresponds to the extra cost of a particle on node $`j`$ with respect to no particle, in the absence of hyperedge $`a`$, and the cavity bias $`u^{(ai)}`$ to the cost of connecting the node $`i`$ to the hyperedge $`a`$. Note that these quantities are actually well defined only if $`\mu `$ is kept finite, otherwise a particle cannot be removed or added without destroying the perfect matching, i.e., without leaving the space of admissible configurations. Similarly, the total fields on the complete graph are
$$\begin{array}{cc}\hfill U^{(a)}& =\underset{ja}{}x^{(ja)},\hfill \\ \hfill X^{(i)}& =\underset{bi}{\mathrm{min}}(\xi _bu^{(bi)}).\hfill \end{array}$$
(34)
From the interpretation given, it appears that the hyperedges which indeed participate to the optimal matching are those which achieve the minima, i.e., the solution is given by
$$n_a=\delta _{a,a^{}},a^{}=\mathrm{arg}\underset{a}{\mathrm{min}}(\xi _au^{(ai)}).$$
(35)
Since this has to hold for all $`ia`$, the question arises whether this prescription effectively defines a matching, i.e., whether $`\mathrm{arg}\mathrm{min}_a(\xi _au^{(ai)})=\mathrm{arg}\mathrm{min}_a(\xi _au^{(aj)})`$ for all $`i,ja`$. A positive answer is obtained by generalizing to $`d>2`$ the inclusion criterion invoked by Aldous when $`d=2`$, which states
$$a^{}=\mathrm{arg}\underset{a}{\mathrm{min}}(\xi _au^{(ai)})=1\xi _au^{(ai)}+x^{(ia)}.$$
(36)
The independence on $`i`$ is then a consequence of the dentity $`u^{(ai)}+x^{(ia)}=U^{(a)}`$. The proof of the inclusion criterion itself is straightforward with the present notations: if $`a^{}=\mathrm{arg}\mathrm{min}_a(\xi _au^{(ai)})`$,
$$\xi _a^{}u^{(a^{}i)}=\underset{bi}{\mathrm{min}}(\xi _bu^{(bi)})\underset{bia^{}}{\mathrm{min}}(\xi _bu^{(bi)})=x^{(ia^{})}.$$
(37)
Reciprocally, if $`a\mathrm{arg}\mathrm{min}_b(\xi _bu^{(bi)})`$,
$$\xi _au^{(ai)}\underset{bi}{\mathrm{min}}(\xi _bu^{(bi)})=\underset{bia}{\mathrm{min}}(\xi _bu^{(bi)})=x^{(ia)}.$$
(38)
As an alternative to the Bethe formula, the value of the optimal matching can be obtained by inferring $`\xi _a^{}`$ from the distribution of the fields $`𝒫(x)`$. Thanks to the inclusion criterion, we have
$$\begin{array}{cc}\hfill _{RS}^{(d)}=\frac{1}{d}\xi _a^{}& =\frac{1}{d}_0^{\mathrm{}}𝑑\xi \xi \mathrm{Prob}(U>\xi )\hfill \\ & =\frac{1}{d}_0^{\mathrm{}}𝑑\xi \underset{j=1}{\overset{d}{}}dx_j𝒫(x_j)\xi \theta \left(\underset{j=1}{\overset{d}{}}x_j\xi \right)\hfill \end{array}$$
(39)
where the factor $`d`$ corresponds to the number of nodes per hyperedge and $`\theta `$ represents the Heaviside function, $`\theta (x)=1`$ if $`x>0`$ and 0 otherwise. The RS cavity equations at zero temperature can also be written in terms of closed integral relations that generalize known equalities for the $`d=2`$ case,
$$\stackrel{~}{G}(x)=_{_jt_j>x}\underset{j=1}{\overset{d1}{}}dt_j\stackrel{~}{G}^{}(t_j)e^{\stackrel{~}{G}(t_j)}\left(x+\underset{j=1}{\overset{d1}{}}t_j\right),$$
(40)
$$_{RS}^{(d)}=\frac{1}{2d}_{_jx_j>0}\underset{j=1}{\overset{d}{}}dx_j\stackrel{~}{G}^{}(x_j)e^{\stackrel{~}{G}(x_j)}\left(\underset{j=1}{\overset{d}{}}x_j\right)^2.$$
(41)
The distribution $`\stackrel{~}{G}(x)`$ is related to the RS distribution $`𝒫`$ of the cavity fields by
$$\stackrel{~}{G}(x)=\mathrm{ln}_x^{\mathrm{}}𝑑t𝒫(t),$$
(42)
and can be obtained from the finite temperature order parameter $`G(l)=G_\beta (l)`$ given in Eq. (29) by
$$\stackrel{~}{G}(x)=\underset{\beta \mathrm{}}{lim}G_\beta (\beta ^{1/(d1)}x).$$
(43)
Comparing with the predictions of the cavity method, $`_{RS}^{(d)}=𝔼[\mathrm{\Delta }ϵ^{(i+ai)}](d1)\mu 𝔼[\mathrm{\Delta }ϵ^{(a)}]`$, we obtain the consistency condition that the RS distribution must satisfy
$$𝔼[x]=\frac{2d}{2d}𝔼\left[\left(\underset{j=1}{\overset{d}{}}x_j\right)^2\theta \left(\underset{j=1}{\overset{d}{}}x_j\right)\right]$$
(44)
where the average $`𝔼[]`$ is here taken with respect to $`𝒫`$. This formula is indeed numerically verified with a good precision, in agreement with the equivalence between the two approaches. As a corollary, it shows that $`𝔼[x]<0`$ unless $`d=2`$ where $`𝔼[x]=0`$ (we recall that in this case one has in fact an explicit formula Mézard and Parisi (1985), $`𝒫(x)=1/[4\mathrm{cosh}^2(x/2)]`$). Finally, we note that for $`d>2`$, the RS energy at zero temperature is only dependent on the mean of $`𝒫(x)`$, $`_{RS}^{(d)}=𝔼[x]/(d2)`$. However as shown in the following, the RS approach yields incorrect predictions when $`d3`$.
### IV.5 Entropy crisis
Using the population dynamics algorithm described in Appendix A, we obtain for the RS free energy $`f_{RS}(\beta )`$ the curves displayed in Fig. 6 ($`d=2`$) and 6 ($`d=3`$). For $`d=2`$, the free energy is an increasing function of $`\beta `$ with limit $`f_{RS}(\beta =\mathrm{})=\pi ^2/120.82`$ corresponding to the cost of a minimal 2-index matching. The free energy obtained for $`d=3`$ is qualitatively different, as it displays a maximum at a finite temperature $`\beta _s0.41`$ (see Fig.8). This entropy crisis reflects an inconsistency of the RS approach fnV . If one assumes the RS approximation holds at high temperature in some range of temperature (a non-trivial statement), a phase transition must occur at some $`\beta _c\beta _s`$.
### IV.6 Stability of the replica symmetric Ansatz
Replica symmetry fails to correctly describe the low temperature properties of many frustrated systems Mézard et al. (1987). A necessary requirement for its validity is that it be stable. Here we show that when $`d=3`$ the RS solution is unstable below a strictly posisive temperature, that is for $`\beta >\beta _i`$. Even if the breakdown of the RS hypothesis was already inferred above from the negative value of the RS entropy, studying the stability is instructive since the relative positions of $`\beta _i`$ and $`\beta _s`$ will establish the discontinuous nature of the phase transition. In Mézard and Parisi (1987), Mézard and Parisi used the replica method to prove that the RS Ansatz is stable when $`d=2`$ Mézard and Parisi (1985); their approach is however quite complicated (see Parisi and Ratiéville (2001) for a recent reexamination of their analysis), and to tackle the $`d=3`$ case, we adopt a simpler approach based on the cavity method Montanari and Ricci-Tersenghi (2003). Physically, it amounts to computing the non-linear susceptibility $`\chi _2`$ and checking that it does not diverge Rivoire et al. (2004). Picking a hyperedge labeled 0 at random, this susceptibility is written
$$\chi _2=\underset{a}{}n_0n_a_c^2\underset{r=0}{\overset{\mathrm{}}{}}[C(d1)]^r𝔼[n_0n_r_c^2]$$
(45)
where $`𝔼[]`$ denotes the thermal average and $`𝔼[]`$ the spatial average over the disorder. Using the fluctuation-dissipation relation, the averaged squared correlation function $`𝔼[n_0n_r_c^2]`$ between two hyperedges separated by distance $`r`$ can be expressed in terms of the cavity fields as Rivoire et al. (2004)
$$𝔼[n_0n_r_c^2]𝔼\left[\underset{i=1}{\overset{r}{}}\left(\frac{\widehat{x}^{(k,\xi )}(x_{i_1},\mathrm{},x_{i_{(d1)k}})}{x_{i_1}}\right)^2\right](r\mathrm{}),$$
(46)
where the average $`𝔼[]`$ is performed with respect to the distribution of the disorder $`(k,\xi )`$ and to the distribution $`𝒫(x)`$ of the cavity fields, except for the $`x_{i_1}`$ with $`i>1`$ which are fixed by $`x_{(i+1)_1}=\widehat{x}^{(k,\xi )}(x_{i_1},\mathrm{},x_{i_{(d1)k}})`$. To determine whether the series in Eq. (45) converges or not, we compute
$$\mathrm{ln}\mu _r=r\mathrm{ln}[C(d1)]+\mathrm{ln}𝔼\left[\underset{i=1}{\overset{r}{}}\left(\frac{\widehat{x}^{(k,\xi )}(x_{i_1},\mathrm{},x_{i_{(d1)k}})}{x_{i_1}}\right)^2\right]$$
(47)
by using cavity fields from the population dynamics, and check whether $`lim_r\mathrm{}(\mathrm{ln}\mu _r)/r<0`$ or not. The numerical results are limited to small values of $`r`$, but as shown in Fig. 8 they are sufficient to conclude unambiguously that an instability shows up for 3-index matchings at $`\beta _i0.6`$, thus confirming the incorrectness of the RS Ansatz for describing the $`\beta =\mathrm{}`$ limit (the same procedure with $`d=2`$ consistently finds no instability). In addition, since the instability takes place only after the entropy crisis, $`\beta _i>\beta _s`$, we conclude from this analysis that the phase transition, located at $`\beta _c\beta _s`$, must be discontinuous as a function of the order parameter.
## V Replica symmetry breaking
The inconsistencies of the RS Ansatz indicate that replica symmetry must be broken in the low temperature phase. This feature is present in many other NP-hard combinatorial optimization problems and is commonly overcome by adopting a one-step replica symmetry breaking (1RSB), which, in most favorable cases, turns out to be exact.
### V.1 General 1RSB Ansatz
As formulated by Aldous with the essential uniqueness property Aldous (2001), replica symmetry in matching problems means that quasi-solutions, that is low energy configurations (LECs), all share most of their hyperedges. In contrast, replica symmetry breaking (RSB) refers to a situation where LECs arise, which, while being close in cost to the optimal solution, are far apart in the configurational space (the measure of distances is the overlap between two matchings, i.e., the fraction of common hyperedges, see Sec. V.3). One-step replica symmetry breaking (1RSB) is a particular scheme of RSB where the structure of the set of LECs can be described with only two characteristic distances, $`d_0`$ and $`d_1<d_0`$. For it to be correct, two LECs taken at random (with the Gibbs probability measure when working at finite $`\beta `$) must be typically found either at distance $`d_0`$ or $`d_1`$. In the replica jargon, close by LECs (at the short distance $`d_1`$) are said to belong to the same state (or cluster). At the level of 1RSB, it is assumed that the number $`𝒩_N(f)`$ of states with a given free energy $`f`$ grows exponentially with $`N`$ and is characterized by a complexity $`\mathrm{\Sigma }(f)`$ defined by $`\mathrm{\Sigma }(f)=lim_N\mathrm{}[\mathrm{ln}𝒩_N(f)]/N`$.
The 1RSB cavity method derives this “entropy of states” by a Legendre transformation method mimicking the derivation of entropy from the free energy in canonical statistical mechanics Monasson (1995). The object generalizing the free energy is the replica potential $`\varphi (\beta ,m)`$; the parameter $`m`$ is the Lagrange multiplier fixing the free energy of the relevant states, in the same way that the temperature $`\beta `$ selects the energy of equilibrium configurations in the canonical ensemble. The replica potential is defined as
$$e^{N\beta m\varphi (\beta ,m)}\underset{\alpha }{}e^{N\beta mf_\alpha },$$
(48)
where the sum is over the states $`\alpha `$, and $`f_\alpha `$ denotes the free energy of a system whose configurations are restricted to $`\alpha `$. To obtain the relevant states for the equilibrium properties, replica theory prescribes to choose the $`m`$ in $`[0,1]`$ that maximizes $`\varphi (\beta ,m)`$ Mézard et al. (1987), so that the equilibrium free energy is given by
$$f_{1\mathrm{R}\mathrm{S}\mathrm{B}}(\beta )=\underset{0m1}{\mathrm{max}}\varphi (\beta ,m).$$
(49)
Calculating $`\varphi (\beta ,m)`$ requires introducing as order parameter a distribution $`𝒬[Q^{(ja)}]`$ over the oriented edges $`(ja)`$ of distributions $`Q^{(ja)}(x)`$ of the cavity fields, taken over the different states $`\alpha `$ Mézard and Parisi (2001). The 1RSB cavity equations for the order parameter read
$$\begin{array}{cc}\hfill 𝒬[Q^{(0)}]& =𝔼_{k,\xi }\underset{a=1}{\overset{k}{}}\underset{j_a=1}{\overset{d1}{}}𝒟Q^{(j_a)}𝒬[Q^{(j_a)}]\delta \left[Q^{(0)}\widehat{Q}^{(k,\xi )}[\{Q^{(j_a)}\}]\right],\hfill \\ \hfill \widehat{Q}^{(k,\xi )}[\{Q^{(j_a)}\}](x^{(0)})& =\frac{1}{Z}\underset{a=1}{\overset{k}{}}\underset{j_a=1}{\overset{d1}{}}dx^{(j_a)}Q^{(j_a)}(x^{(j_a)})\delta \left(x^{(0)}\widehat{x}^{(k,\xi )}(\{x^{(j_a)}\})\right)e^{\beta m\mathrm{\Delta }\widehat{F}_n^{(k,\xi )}(\{x^{(j_a)}\})},\hfill \end{array}$$
(50)
where $`\widehat{x}^{(k,\xi )}`$ is given by Eq. (22) and the reweighting term is
$$e^{\beta \mathrm{\Delta }\widehat{F}_n^{(k,\xi )}(\{x^{(j_a)}\})}=e^{\beta \mu }+\underset{a=1}{\overset{k}{}}e^{\beta (\xi _a_{j_a=1}^{d1}x^{(j_a)})}.$$
(51)
The latter corresponds to the shift of free energy due to the addition of the new node. Its presence insures that the different states described by the $`Q^{(ja)}(x)`$ have indeed all the same free energy, in spite of the fact that the addition of a node inevitably introduces a free-energy shift. The distribution $`𝒬[Q]`$ determines the replica potential $`\varphi (\beta ,m)`$ whose explicit expression is
$$\varphi (\beta ,m)=𝔼[\mathrm{\Phi }^{(i+ai)}(\beta ,m)](d1)\mu 𝔼[\mathrm{\Phi }^{(a)}(\beta ,m)]$$
(52)
with
$$𝔼[\mathrm{\Phi }^{(i+ai)}(\beta ,m)]=\frac{1}{\beta }𝔼_{(k,\xi )}\underset{a=1}{\overset{k}{}}\underset{j_a=1}{\overset{d1}{}}𝒟Q^{(j_a)}𝒬[Q^{(j_a)}]\mathrm{ln}\left[\underset{a=1}{\overset{k}{}}\underset{j_a=1}{\overset{d1}{}}dx^{(j_a)}Q^{(j_a)}(x^{(j_a)})e^{m\beta \mathrm{\Delta }\widehat{F}_n^{(k,\xi )}(\{x^{(j_a)}\})}\right]$$
(53)
and
$$𝔼[\mathrm{\Phi }^{(a)}(\beta ,m)]=\frac{1}{\beta }𝔼_\xi \underset{j=1}{\overset{d}{}}𝒟Q^{(j)}𝒬[Q^{(j)}]\mathrm{ln}\left[\underset{j=1}{\overset{d}{}}dx^{(j)}Q^{(j)}(x^{(j)})\left(1+e^{\beta (\xi _{j=1}^dx^{(j)})}\right)^m\right].$$
(54)
The 1RSB equations can in principle be numerically solved via a population dynamics algorithm Mézard and Parisi (2001). However, our efforts in this direction failed to yield a sensible order parameter because the fields were found to diverge as $`\mu `$ was increased : the reason for this behavior is elucidated below.
### V.2 Frozen 1RSB Ansatz
Although rarely explicitly mentioned, there exists a replica symmetry breaking Ansatz somewhat intermediate between the RS and general 1RSB as just described. The frozen 1RSB Ansatz, which will be argued to apply to matchings, is a particular realization of the 1RSB scheme where states are made of single configurations (or, more generally, of a non-exponential number of configurations). In such a case, all the information can be extracted from the RS quantities, provided they are adequately reinterpreted. Consider for instance the definition given by Eq. (48) in the special case where states $`\alpha `$ have no internal entropy, i.e., $`f_\alpha =ϵ_\alpha `$. We thus have
$$e^{N\beta m\varphi (\beta ,m)}\underset{\alpha }{}e^{N\beta mf_\alpha }=\underset{\alpha }{}e^{N\beta mϵ_\alpha }e^{N\beta mf_{RS}(\beta m)}$$
(55)
where the last equality holds because of the very definition of a RS free energy. The replica potential $`\varphi `$ can therefore be expressed in term of the RS free energy only,
$$\varphi (\beta ,m)=f_{RS}(\beta m).$$
(56)
Following the prescriptions of replica theory, the quenched free energy is obtained by maximizing $`\varphi (\beta ,m)`$ over $`m[0,1]`$. Being a concave function function, the RS free energy can have at most one maximum. If $`\beta _s`$ denotes the location of this maximum (with maybe $`\beta _s=\mathrm{}`$, like for 2-index matchings), we obtain that $`f_{1RSB}(\beta )=\varphi (1,\beta )=f_{RS}(\beta )`$ for $`\beta <\beta _s`$ and $`f_{1RSB}(\beta )=\varphi (\beta _s/\beta ,\beta )=f_{RS}(\beta _s)`$ for $`\beta >\beta _s`$. In other words, starting from the assumption that the content of states is trivial, the frozen Ansatz predicts a complete freezing of the system at the point $`\beta _s`$ where the RS entropy becomes zero : for $`\beta >\beta _s`$, the system is trapped in a single configuration and its free energy stays constant when the temperature is further decreased ($`\beta `$ increased).
This scenario is already known to apply to a few models of disordered systems, including the random energy model (REM) Derrida (1980), the directed polymer on disordered trees Derrida and Spohn (1988), the binary perceptron Krauth and Mézard (1989) and the XOR-SAT problem on its core Mézard et al. (2003) (with a particular case being error-correcting codes of the Gallager type Montanari (2001)). Our intention is here both to add the matchings to this list, and to clarify the conditions under which such a scenario may apply. At this stage, we can already state the following necessary conditions (all satisfied by $`d`$-index matchings with $`d3`$):
$`(i)`$ the RS entropy must become negative at a finite $`\beta _s`$;
$`(ii)`$ the RS solution must be stable up to (at least) $`\beta _s`$;
$`(iii)`$ no discontinuous 1RSB transition must be detected before $`\beta _s`$.
In addition to these properties, the consistency of the frozen Ansatz requires the model to have particular kinds of constraints, called hard constraints. Elucidating this point requires a more refined description of the relation between the frozen 1RSB order parameter and the RS order parameter. First remember that in the RS picture at finite temperature $`\beta `$, one has a spatial distribution $`𝒫(x^{(ja)})`$ of cavity fields, where following Eq. (19), $`\psi _{RS}^{(ja)}\mathrm{exp}[\beta (x^{(ja)}\mu )]`$ is interpreted as giving the probability under the Boltzmann measure that node $`j`$ is not matched given that the hyperedge $`a`$ is absent. For a general 1RSB problem, the order parameter is instead $`𝒬[Q^{(ja)}(x^{(ja)})]`$ where $`\psi ^{(ja)}\mathrm{exp}[\beta (x^{(ja)}\mu )]`$ is again a thermal probability, but now restricted to a particular state taken from the distribution over states $`Q^{(ja)}`$. In this context, a RS system, characterized by a single state, has $`Q^{(ja)}(\psi ^{(ja)})=\delta (\psi ^{(ja)}\psi _{RS}^{(ja)})`$. For a system in a frozen glassy phase instead, the thermal averages inside each state are trivial since there is a single frozen configuration, $`\psi ^{(ja)}=0`$ or 1 meaning that a particle is present or absent with probability one. Therefore, the relation with the RS order parameter has the form
$$Q^{(ja)}(\psi ^{(ja)})=\psi _{RS}^{(ja)}\delta (\psi ^{(ja)})+(1\psi _{RS}^{(ja)})\delta (\psi ^{(ja)}1).$$
(57)
Plugging this expression into the general 1RSB cavity equation, it is found that such an Ansatz is consistent only if the system satisfies the condition that in the cavity recursion, the variable on a node is completely determined by the values of the variables on the neighboring nodes. Such is the case with matchings when $`\mu =\mathrm{}`$ where a particle is to be assigned to a hyperedge if and only if none of the neighboring edges are occupied. This is however not the case in all constraint problems. Consider for instance the 3-coloring problem where each node is assigned one of three colors with the constraint that its color must differ from its neighbors : in the case where all the neighbors have the same color, the choice is left for the node between the two other colors. When a variable is fixed by the value of its neighbors in the cavity recursion, we say that the system has hard constraints ; hard constraints can be shown Rivoire (2005) to indeed be present in the binary perceptron and in the XOR-SAT model on its core, models where the frozen Ansatz applies too. Finally, we note that in the presence of hard constraints, the cavity fields $`\psi ^{(ja)}`$ take at the 1RSB level values 0 and 1 only, which are associated with $`x^{(ja)}=\mu `$ and $`\mathrm{}`$. This explains the divergences observed when trying to implement the 1RSB population algorithm at zero temperature with $`\mu \mathrm{}`$.
### V.3 Distances
As mentioned in Sec. V.1, a 1RSB glassy system is generally described by two distances, $`d_0`$ corresponding to the typical distance between two states, and $`d_1`$ corresponding to the typical distance between two configurations inside a common state. In the case of a frozen 1RSB glassy phase, one has however $`d_1=0`$ and the structure of low-energy configurations (LECs) is characterized by only one distance, $`d_0`$. If $`n_a`$ denotes the mean occupancy of a particular hyperedge $`a`$, with the average $``$ taken over the LECs, the probability for $`a`$ to belong to two different LECs is given by $`n_a^2`$. Averaging over the different hyperedges, it defines the overlap
$$q=𝔼[n_a^2],$$
(58)
which is directly related to the typical distance between LECs through $`d_0=1q`$. As argued before, for a system in a frozen glassy phase the distribution of energies of the LECs is described by the thermal average at $`\beta _c`$ in the RS approximation, so that
$$n_a=\frac{Y_1^{(a)}}{Y_0^{(a)}+Y_1^{(a)}}=\frac{1}{1+e^{\beta _c(\xi _a_{ia}x^{(ia)})}}.$$
(59)
Averaging over the disorder therefore yields
$$q=𝔼[n_a_{\beta _c}^2]=𝔼_{\xi _a}\underset{j=1}{\overset{d}{}}dx^{(j)}𝒫(x^{(j)})\left(1+e^{\beta _c(\xi _a_{ia}x^{(j)})}\right)^2.$$
(60)
The overlap $`q(\beta )`$ is represented for all values of $`\beta `$ in Fig. 10 when $`d=3`$; given the value of $`\beta _c`$ obtained before, we get $`q=q(\beta _c)=0.321\pm 0.002`$.
## VI Numerical analysis of finite size systems
The theoretical analysis provided concerned the $`M\mathrm{}`$ limit. How is that limit reached, and in particular is the convergence exponentially fast in $`M`$ or is it algebraic? To answer such questions, we consider in this section the properties of $`d`$-partite matchings when $`M`$ is finite; in the absence of other tools, we do this numerically. It should be clear that the most challenging questions concern the low temperature phase of our system; because of that, we will focus on the optimum matching and low lying excitations. Even though such a numerical approach requires sampling the disorder (random instances) and extracting for instance distributions with inevitable statistical uncertainties, it will give evidence that our frozen 1RSB Ansatz is correct; it will also provide some statistical properties of finite size systems that are of interest on their own.
### VI.1 The branch and bound procedure
When $`M`$ is very small, it is possible to enumerate all $`\left[M!\right]^{d1}`$ $`d`$-partite matchings of a given sample. Not surprisingly, this becomes unwieldy even when $`M`$ reaches 10, forcing us to choose an alternate approach. Since it is the *low energy* matchings that are of greatest interest, we have developed a branch and bound algorithm that computes the $`p`$ lowest energy matchings, for any given $`p`$. Some technical aspects of the algorithm are presented in Appendix B, but the essential elements are as follows.
We represent a matching via a list of $`M`$ hyperedges, one for each of the $`M`$ sites of the first set (recall that there are $`d`$ sets, each of $`M`$ sites). Such a representation includes also some non-legal matchings as some of the sites in the second or higher sets could belong to more than one hyperedge; if a matching is not legal, it is discarded. This representation can be mapped onto a rooted tree: each level of the tree is associated with one of the sites of the first set, while a segment (branch) emerging from a node corresponds to a choice of hyperedge that contains the site of that node’s level. The root node is associated with the first site, the nodes of the next level are associated with the second site, etc… This tree is regular, each node having $`M^{d1}`$ outgoing segments as there are that many hyper-edges containing a given site of the first set. Furthermore, it has $`M+1`$ levels: there is one level for each site of the first set while the last level consists of leaves rather than of nodes; each leaf corresponds to a candidate matching specified by the list of hyperedges obtained when going from the tree’s root to that leaf. This list may correspond to a legal matching or not, but each matching appears exactly once as a leaf. (In fact, there are $`M^{M(d1)}`$ leaves while there are only $`\left[M!\right]^{d1}`$ legal matchings.)
The principle of the branch and bound algorithm is to find those leaves which satisfy the desired criterion (the energy must be less or equal to that of the $`p`$th lowest energy matching) by exploiting a pruning procedure, thereby avoiding having to explore all leaves. To begin our pruned search, we produce $`p`$ distinct legal matchings and put them into a list $``$; the largest energy of the matchings in this list is an upper bound $`E_{UB}`$ on the $`p`$th energy level for our system. Then we start at the level of the tree’s root and consider all of its segments; for each choice of segment, the search problem corresponds to finding matchings on a smaller system with one less site in each of the $`d`$ sets; the search can thus be implemented recursively. Suppose we have done $`k`$ recursions; the sub-problem is associated to the node on our tree that is obtained by following the choices of hyperedges in the recursive construction. This node corresponds to a partial matching in which the first $`k`$ sites of the first set have each been assigned a hyperedge. An important property is that all hyperedges have positive energies; then we know that any matching that is compatible with the current partial matching has an energy greater than it, thereby providing a lower bound on all the leaf energies obtainable from the current node. If that lower bound is greater than $`E_{UB}`$, then the subtree rooted on the current node can be pruned (discarded from the search); otherwise, one iterates the recursion (that is one performs branching on the different choices of the hyperedge to include at the present level) and $`k`$ goes to $`k+1`$. When this process leads to a leaf that corresponds to a legal matching, we compute the energy $`E`$ of this matching. If $`E<E_{UB}`$, we insert that matching into our list $``$ and remove its worst element so that it always has $`p`$ elements; we also update $`E_{UB}`$ which by definition is the largest energy of the matchings in $``$; on the contrary, if $`E>E_{UB}`$, we discard the matching (leaf). After a finite number of branchings and prunings, the algorithm has explored all choices for the segments emerging from the tree’s root and one is done. The best $`p`$ matchings are then in the list $``$.
The algorithm without pruning requires $`O(M^{M(d1)})`$ operations; with pruning and the different optimizations sketched in Appendix B, the number of operations grows roughly by a constant factor when $`M`$ is increased by $`1`$; in particular, for the random instances studied here and $`d=3`$, this factor is about $`2.2`$.
### VI.2 Ground state energies
We generated a large number of random samples (disorder instances with the hyperedge costs taken to be independent uniformly distributed random variables in $`[0,1]`$) and for each sample determined its ground state. We used several random number generators to check that our results were robust. Because of the exponential growth of the computation time with $`M`$, in practice we were limited to relatively modest values of $`M`$. For the results presented here and involving only ground states, at $`d=3`$ we used $`10000`$ samples for $`M=20`$ and $`M=22`$, while for the smaller values of $`M`$ we used $`20000`$ samples. We also performed runs at $`d=4`$ but with lower statistics because the algorithm becomes less efficient as $`d`$ increases; in fact, we were limited to $`M14`$ for that case and had only $`5000`$ samples for each $`M`$.
Let’s first focus on the behavior of ground-state energy. For each sample, we determine with our Branch & Bound algorithm the ground-state energy density $`e_0E_0/MM^{d2}C_M^{(d)}`$ \[cf. Eq. (7)\]; then we can analyse its mean in our ensemble or consider other properties of its distribution.
In Fig. 12 we show how the mean ground-state energy density $`𝔼[e_0]`$ changes as one increases $`M`$. The behavior is roughly linear in $`1/M`$, but by eye one can definitely see some curvature. Because of this, linear fits do not give good values of $`\chi ^2`$ unless the $`M<10`$ data are ignored; for instance, keeping only the $`M10`$ data, the linear fit gives $`3.040(3)`$ as the limiting value with $`\chi ^2=3.6`$ for $`9`$ degrees of freedom, while if we use all the data we obtain $`3.021(3)`$ with $`\chi ^2=32`$ for $`14`$ degrees of freedom. We have also tried corrections of the type $`\mathrm{ln}(M)/M`$ but this did not work well. Thus we proceed by considering quadratic fits. In that case, the resulting $`M=\mathrm{}`$ intercept does not depend much on whether one uses all or just the highest values of $`M`$. In particular, for all the data, we get the limiting value $`3.046(5)`$ with $`\chi ^2=9.6`$ for $`13`$ degrees of freedom, while using the $`M10`$ data only one has $`3.06(1)`$ with $`\chi ^2=2.3`$ for $`8`$ degrees of freedom. (In all these estimates, the error bars quoted are statistical only, as obtained from the statistical fluctuations.) We have also considered power fits, namely $`𝔼[e_0]=a+b/M^c`$. Fitting all the data gives the limiting value $`3.08(1)`$ with $`\chi ^2=7.2`$ for $`13`$ degrees of freedom while keeping only the $`M10`$ data leads to $`3.09(3)`$ with $`\chi ^2=2.3`$ for $`8`$ degrees of freedom (in both cases, the exponent $`c`$ is close to $`0.88`$). Since these $`\chi ^2`$ are similar to those of the quadratic fits, we see that the systematic errors are not negligible and are at least of the same order as the statistical errors; because of these effects, the agreement with the theoretical value of $`3.126`$ can be considered rather good.
We studied similarly the case $`d=4`$. The data again has positive curvature when plotted as a function of $`1/M`$, but since we have less statistics and a much smaller range of $`M`$, much less precision can be obtained for the large $`M`$ limit. For the linear fit ($`M9`$) we get a limiting value of $`6.75(3)`$ with $`\chi ^2=4.7`$ for $`4`$ degrees of freedom. For the quadratic fit ($`M9`$ again), we get $`7.22(8)`$ with $`\chi ^2=0.37`$ for $`3`$ degrees of freedom. Finally, for the power fit we get $`10.2(9)`$ with $`\chi ^2=1.0`$ for $`5`$ degrees of freedom; the exponent is $`c=0.3`$ which is small and leads to a large upturn for $`M>100`$; clearly that regime is far beyond our reach and suggests that the power fit is probably inappropriate as non robust (note for instance that the uncertainty on the limiting value is far higher here than for the other fits). The different estimates show that uncertainties arising from systematic effects ($`M`$ too small) are severe; instead of the $`1\%`$ precision we had at $`d=3`$, we have a precision of at best $`10\%`$ at $`d=4`$ (compare to the theoretical prediction of $`7.703`$). The conclusion is that numerics do not teach us much for the case $`d=4`$ and so hereafter we shall concentrate on the different properties arising when $`d=3`$.
One of the expectations for the $`d`$-index matching problem is that the free energy is self-averaging. Although at present there is no proof of such a property, there is no reason to expect otherwise; here we are limited by the numerical approach to ground states, but in that framework we can determine empirically the *distribution* of energies in the ensemble of random instances. Fig. 12 displays the probability distribution of the (extensive) ground-state energy $`E_0`$ for several values of $`M`$ ($`d=3`$). If as expected, the ground-state energy is self-averaging, the relative width of these distributions should go to zero. We have thus measured the first few moments of these distributions. In the inset of Fig. 12, we have plotted the standard deviation $`\sigma `$ of the ground-state energy divided by $`\sqrt{M}`$ as a function of $`1/M`$. Self-averaging corresponds to having $`\sigma /M0`$; from the inset we see that $`\sigma /M^{1/2}`$ goes to a constant at large $`M`$ so self-averaging holds and the convergence of the distribution is compatible with a central limit theorem type behavior; such a scaling arises from sums of not too dependent random variables and leads to a Gaussian limiting shape. To confirm this, we have looked at higher moments: we find that indeed the skewness and kurtosis of the distributions decrease, in line with a central limit theorem type convergence.
Having a limiting Gaussian distribution for $`E_0`$ is not a consequence of the frozen 1RSB pattern of replica symmetry breaking since in the random energy model the distribution of $`E_0`$ follows a Gumbel distribution; furthermore, in that case the fluctuations in $`E_0`$ are $`O(1)`$ whereas in the matching problem they are $`O(\sqrt{M})`$. To see why such large fluctuations are “natural”, consider instead of $`E_0`$ the quantity $`_0`$ obtained by adding the lengths $`\mathrm{}_i`$ of the shortest hyperedges containing each site $`i`$ of the first set. This quantity arises in a greedy algorithm (but which does not necessarily generate a legal matching) and clearly one has $`E_0_0`$. The central limit theorem applies to $`_0`$, so it will have a standard deviation that grows as $`\sqrt{M}`$ and its distribution will become Gaussian at large $`M`$. The actual ground-state energy $`E_0`$ is obtained by allowing hyperedge lengths that are slightly larger than the $`\mathrm{}_i`$, but this should not suppress the large fluctuations nor prevent the central limit theorem scaling.
### VI.3 Other ground state properties
As discussed at the beginning of this paper, one expects the hyperedge containing a given site in the ground state matching to be one of the shortest possible ones. To investigate this issue quantitatively, let us order all the hyperedges containing a given site, going from the shortest to the longest hyperedge. The “order” of a hyperedge is then $`1`$ if it is the shortest, $`2`$ if it is the next shortest, etc… The orders arising in the ground state should be dominated by the lowest ones, $`1`$, $`2`$, $`3`$… Consider thus the frequencies with which these orders arise; in Fig.14 we show the behavior of these frequencies for increasing $`M`$ in the case $`d=3`$. We see that there is a limiting histogram at large $`M`$, and that indeed the lowest orders dominate. Furthermore, we see that for large $`k`$ the probability of occupation of an edge tends to decrease exponentially with $`k`$ (the data are displayed on a semi log plot). Note that in the standard matching ($`d=2`$) problem, the decrease goes as $`1/2^k`$ exactly, while for our $`d=3`$ case, the exponential decay is only asymptotic; furthermore, we have found no simple expression giving the decay rate of this exponential.
### VI.4 Excited states
Let us consider now states above the ground state. Define the excitation energy or “gap” as $`E_1E_0`$ where $`E_0`$ is the extensive ground-state energy and $`E_1`$ that of the next lowest energy state. In Fig. 14 we show that this random variable has a limiting distribution so that $`E_1E_0=O(1)`$ in the large $`M`$ limit, just as happens in the random energy model. Furthermore, the distribution is very well fit by an exponential (cf. the curve shown in the figure).
Following our theoretical conclusions obtained earlier, consider now the overlap between the ground state and the first excited state. In our frozen 1RSB picture, these matchings are expected to have a fixed (self-averaging) overlap when $`M`$ grows. In Fig. 16 we show the probability distribution of such overlaps for increasing $`M`$. We see that there is a local peak at large overlap that shifts toward $`q=1`$ but which simultaneously decays. The bulk of the overlaps however arise around $`q=0.3`$ and when $`M`$ increases we see that the corresponding peak both gets higher and more narrow. Overall, the behavior is compatible with a convergence toward a Dirac peak near $`q=0.32`$, to be compared with the theoretical prediction $`q_c=0.321`$.
### VI.5 Low energy entropy
Finally, consider the *density* of energy levels. In the case of the random energy model, this density becomes self-averaging when the excitation energy grows. We have thus computed the disorder averaged density of levels as a function of the excitation energy, $`EE_0`$. That is a measure of the exponential of the microcanonical entropy; within the frozen 1RSB scenario, it gives the critical temperature via $`\rho (EE_0)\mathrm{exp}[\beta _c(EE_0)]`$. In Fig. 16 we display our numerical estimate of $`\rho `$ and see that it is very nearly a pure exponential. From the slope on the semi-log plot we extract $`\beta _c0.405`$; this value should be compared to the theoretical prediction of $`0.412`$; the agreement is reasonable but not perfect. To get better agreement, we believe it would be necessary to go to larger $`M`$ and also to go further in the self-averaging regime, i.e., to consider larger $`EE_0`$ which numerically is an arduous task.
## VII Conclusion
We presented an analysis of multi-index matching problems (MIMPs) based on an adaptation of the cavity method for finite connectivity systems. For the well-known two-index matching problem, our approach provides an alternative derivation of results previously obtained using the replica and cavity methods. With respect to these older studies, the present one has the advantages of being closer to the mathematical framework developed by Aldous, and of allowing replica symmetry breaking effects to be incorporated in a tractable manner. Exploiting this latter possibility, we predict the value of the asymptotic minimal cost to be given for $`d`$-index matching problems by $`^{(d)}=ϵ_{RS}(\beta _s)`$ with $`ϵ_{RS}`$ obtained from Eq. (29) and (30) and $`\beta _s`$ satisfying $`s_{RS}(\beta _s)=0`$. Formally, this $`d3`$ conjecture differs from the case $`d=2`$ (where it is a theorem) in that $`\beta _s=\mathrm{}`$ when $`d=2`$, while $`\beta _s<\mathrm{}`$ when $`d3`$. The distinction between 2-index and $`d`$-index matching problems with $`d3`$ arises clearly from our analytical and numerical analysis: in the first case all low cost matchings share most of their hyperedges, while in the second case they differ from each other by a finite fraction of their hyperedges. In mathematical terms, the essential uniqueness property does not hold when $`d3`$ or in physical terms replica symmetry must be broken. Extending Aldous’ framework to rigorously account for this fact and providing a proof of our conjecture for $`d3`$ seems to us a particularly interesting mathematical challenge.
From a physical perspective, the qualitative difference between 2-index and $`d`$-index matchings problems with $`d3`$ hinges on the presence at low temperature of a glassy phase. This is similar to the difference that has been found between the 2-SAT and 2-coloring problems, which are polynomial, and the $`K`$-SAT and $`q`$-coloring problems with $`K3`$ and $`q3`$, which are NP-complete. For MIMPs, the nature of the glassy phase is however simpler, as it is made of isolated configurations instead of separate clusters of many configurations. We termed this phase a “frozen 1RSB glassy phase” and attributed it to the nature of the constraints, called hard constraints. As a technical consequence of this distinctive feature, a particular frozen 1RSB Ansatz has to be implemented. Such an Ansatz has repetitively been used in the literature as a convenient (but rarely justified) substitute for the more complicated general 1RSB Ansatz; our discussion on the role of hard constraints provides a clarification of its conditions of validity which we believe is of general interest for the investigation of glassy phases in other systems.
###### Acknowledgements.
This work was supported in part by the European Community’s Human Potential Programme under contracts HPRN-CT-2002-00307 (DYGLAGEMEM) and HPRN-CT-2002-00319 (STIPCO) as well as by the Community’s EVERGROW Integrated Project.
## Appendix A Population dynamics algorithm
Here we give a short description of the population dynamics algorithm we used to solve the RS cavity equations. We implemented two different versions, corresponding to the two different cut-off procedures mentioned in the text, associated either with Poissonian (algorithm P) or regular graphs (algorithm R). In addition to the inputs $`d`$ and $`\beta `$, the algorithm has essentially 3 parameters: the mean degree of the nodes, $`C`$ (algorithm P) or $`K`$ (algorithm R), the size of the population, $`𝒩_{\mathrm{pop}}`$, and the number of iterations $`𝒩_{\mathrm{iter}}`$. The common structure of the two algorithms is the following:
* Initialize with random values a population of cavity fields $`x[i]`$, $`i=1,\mathrm{},𝒩_{\mathrm{pop}}`$;
* Do $`𝒩_{\mathrm{trans}}=100`$ times: $`Update()`$;
* Do $`𝒩_{\mathrm{iter}}`$ times: $`Update()`$ and $`Measure()`$.
The first loop allows the system to equilibrate toward the stationary distribution. The subroutine $`Update()`$ depends on the cut-off procedure and can schematically described as follows:
Do $`𝒩_{\mathrm{pop}}`$ times:
* Draw $`k`$ either at random with Poissonian distribution of mean $`C`$ (algorithm P), or take $`k=K`$ (algorithm R);
* Draw costs $`\{\xi _a\}_{a=1,\mathrm{},k}`$ either independently with the uniform distribution in $`[0,C]`$ (algorithm P), or according to a Poisson process with rate 1 (algorithm R);
* Draw at random $`k(d1)`$ members of the population and use them together with the $`\xi _a`$ to compute a new field $`x_0`$ according to Eq. (22);
* Draw at random one member of the population and replace its cavity field value with $`x_0`$.
The subroutine $`Measure()`$ is implemented similarly and computes the free energy according to Eqs. (27)-(28). The final output for the free energy is obtained by averaging over the $`𝒩_{\mathrm{iter}}`$ iterations, while the fluctuations across iterations are used to check convergence. The algorithm must be run for increasing values of $`C`$ or $`K`$ to extrapolate the $`C\mathrm{}`$ (algorithm P) or $`K\mathrm{}`$ (algorithm R) limit, requiring one to consider larger and larger population sizes $`𝒩_{\mathrm{pop}}`$ to obtain reliable results. Taking this limit is however facilitated by the numerical observation that the Poissonian approximation (algorithm P) approaches the solution from below while the regular approximation (algorithm R) approaches it from above; this is illustrated in Fig. 10 with $`d=4`$. We refer to the captions of the various figures for typical choices of the parameters $`C`$, $`𝒩_{\mathrm{pop}}`$ and $`𝒩_{\mathrm{iter}}`$. The numerical results we obtained for $`d=2`$ are consistent with the exact solution, $`\beta _c=\mathrm{}`$ and $`f_{RS}(\beta _c)=\pi ^2/12`$, and are the following for $`d=3,4`$:
$$\begin{array}{c}\hfill d=3:\beta _c=0.412\pm 0.001,f_{RS}(\beta _c)=1.042\pm 0.0003,\\ \hfill d=4:\beta _c=0.135\pm 0.002,f_{RS}(\beta _c)=1.925\pm 0.0006.\end{array}$$
(61)
The free-energy densities are given here for simple matching problems and their counterpart for $`d`$-partite matchings are obtained by multiplying the values by $`d`$.
We have also implemented the generalization of this algorithm to solve the 1RSB cavity equations (50) and used it to check that no discontinuous transition occurs prior to the entropy crisis (see Mézard and Parisi (2001) for algorithmic details).
## Appendix B Aspects of the branch and bound algorithm
Our objective is to solve $`d`$-partite matching problems at sufficiently large $`M`$ so that an extrapolation to the $`M\mathrm{}`$ limit can be performed without too much uncertainty. For many problems (satisfiability, coloring, etc…), one prefers an easily implementable algorithm such as one in the class of “heuristic” algorithms; in such approaches one performs a fast search for the ground state but no guarantee is provided that the global optimum will be found. Examples of these algorithms are simulated annealing and variable depth local search. Heuristic algorithms typically attempt to move towards regions of lower energy by searching in the neighborhood of a current configuration. However, since the search is local, such an approach is bound to break down for problems in which the frozen 1RSB scenario applies. This fact pushed us towards the development of an “exact” algorithm capable of delivering a certificate of optimality of the proposed ground state. Amongst exact algorithms, enumeration can be discarded because it is much too slow; Branch & Bound gets around this problem through pruning of the enumeration/search. There are also other possible methods such as Branch & Cut, but these require an in depth understanding of polytopes and rely on separation procedures which have not yet been developped for MIMP. Note that in all exact methods, the key to efficiency is to have good bounds; fortunately MIMPs are relatively well adapted to such a strategy.
We already discussed in the main text our choice of representation of matchings and partial matchings. Given a partial matching of the first $`k`$ sites of the first set, we have to solve a MIMP with $`Mk`$ sites and so the algorithm can be implemented recursively. Since at each node we need to consider all of its possible branchings (naively, there are $`(Mk)^{d1}`$ of these), it is useful to order these branchings according to the length of the corresponding hyperedges, going from short to long. Rather than recompute these orderings dynamically every time the partial matching changes, we do it once and for all at the initialization of the program. This allows for speed but it must be compensated by a rapid determination of whether a given hyperedge is allowed; for that we use a data structure which tells us for each site of each set whether it is matched (belongs to one of the occupied hyperedges). This structure is updated whenever a partial matching is extended or reduced.
The pruning of the search must be as stringent as possible, and this depends on the quality of the bounds. Our simplest bound $`B_1`$ is just the current partial matching’s energy $`E_k`$: if that energy is higher than $`E_{UB}`$ (the upper bound $`E_{UB}`$ as defined in section VI.1), then the whole sub-tree below the current node can be pruned. A better bound is $`B_2`$, obtained by adding to $`E_k`$ the sum over each remaining unmatched site of the first set of the shortest hyperedge containing that site. This sum can be precomputed and tabulated. A still better bound is $`B_3`$ obtained as $`B_2`$ but where now one takes for each site the shortest hyperedge that is compatible with the current partial matching. This bound cannot be predefined once and for all and is slow to compute. Since we have found it to be useful for pruning, we have optimized its determination by noticing that it can be tabulated and modified incrementally: every time the partial matching is extended (a hyperedge is added), we perform the search for the compatible hyperedges of each unmatched site starting from the index (order) previously found to be compatible. When backtracking, one has to remove a hyperedge and there we simply go back to the tables we had at that level: in effect, we maintain efficiency if we assign tables at each level and follow their updating one step at a time.
The rate of pruning is very different for the three bounds, and we found that a good strategy (for balancing pruning rate and computation time) was to apply the three bounds successively: if the first one does not prune, one goes on to the second one and so forth. To speed up the computation further, we found it useful to implement the recursivity of the program in a limited mode only: the data structures are set up once and for all at initialization time, the hyperedges are ordered once and for all too, and then the recursion is used mainly to go through the branchings and to maintain the tables. Efficiency is gained as no reorganization of the instance (hyperedge weights) is performed, and in particular no “smaller matching problem” is ever defined explicitly. |
warning/0507/astro-ph0507607.html | ar5iv | text | # The Potential of Multi-Colour Photometry for Pulsating Subdwarf B Stars1footnote 11footnote 1Predicted colour-amplitude ratios for a series of representative EC 14026 and PG 1716 stars are available upon request. Interested collaborators please contact S.K. Randall or G. Fontaine.
## 1 INTRODUCTION
Subdwarf B (sdB) stars are evolved extreme horizontal branch stars with effective temperatures in the range 20,000$``$42,000 K and low masses of around 0.5 $`M_{}`$ (e.g., Saffer et al. 1994). They are believed to be composed of a helium-burning core surrounded by a hydrogen envelope too thin for them to ascend the asymptotic giant branch after core helium exhaustion (Heber 1986; Dorman 1995). Instead, they evolve off and along the horizontal branch and eventually end their lives as low-mass white dwarfs (Bergeron et al. 1994). The discovery of pulsators among subdwarf B stars (Kilkenny et al. 1997) has opened them up to asteroseismological probing, an important tool for understanding the internal structure of these interesting objects. Pulsating subdwarfs can be divided into two categories: the rapidly pulsating EC 14026 stars and the slowly oscillating long-period variable subdwarf B stars (or PG 1716 stars for short).
The first EC 14026 stars were discovered to be pulsating multi-periodically with typical periods in the range 100-200 s in 1997 (Kilkenny et al. 1997; Koen et al. 1997; Stobie et al. 1997; O’Donoghue et al. 1997). At about the same time and completely independently, Charpinet et al. (1996, 1997) predicted the existence of pressure mode ($`p`$-mode) instabilities in these stars caused by a classical kappa mechanism associated with the iron opacity peak in the subdwarfs’ envelopes. This opacity peak in turn is dependent on a local overabundance of iron, which is achieved by the competitive action of gravitational settling and radiative levitation. As the relative contributions of these two processes are determined by the surface gravity and the effective temperature of the star, all pulsating subdwarfs should lie in a designated instability strip on the log $`gT_{\mathrm{eff}}`$ diagram. To date, the number of known EC 14026 stars has risen to 33 (see Kilkenny 2002, Silvotti et al. 2002, Billères et al. 2002, and Fontaine et al. 2004 for a recent census), all of which fall into the theoretically predicted instability strip, clustering around $`\mathrm{log}g`$5.75 and $`T_{\mathrm{eff}}`$33,000. Since theoretical and observed pulsational properties are in good agreement, it is believed that a qualitative understanding of the pulsation mechanism has been reached. Beyond this, detailed quantitative interpretations of observed period spectra have been possible in a few instances (Brassard et al. 2001; Charpinet et al. 2003, 2005), allowing the identification of the modes excited through the so-called forward method. Being the ultimate goal in asteroseismology, this led to the determination of the stars’ fundamental parameters, including the mass fraction of the thin hydrogen shell which cannot be deduced otherwise. It goes without saying that independent observational tests of these mode idendifications would be most welcome. In particular, the dense period spectra detected in certain EC 14026 pulsators force the controversial inclusion of modes with $`l`$ = 3 and 4 in the asteroseismological process, since there are not enough theoretical modes with $`l`$ = 0, 1, and 2 to account for the mode density observed.
Long-period variables constitute a newer, less extensively studied class of subdwarf B pulsator. Their variability was first announced by Betsy Green at the “Asteroseismology across the H-R diagram” conference held in Porto in July 2002 (Green et al. 2003a; see also Green et al. 2003b). Slowly pulsating sdB stars are distinctly cooler than their short period counterparts and show multi-periodic, low-amplitude ($``$ 1 milli-magnitude) luminosity variations with typical periods between 0.8 and 1.6 hours. These are about a factor of 30 longer than those of the EC 14026 stars, and automatically imply high radial order gravity modes ($`g`$-modes). Early ideas on a possible excitation mechanism for this new class of star included tidal excitation in a close binary system (Fontaine et al. 2003a) and a mechanism involving a slow reviving of the hydrogen shell (Green et al. 2003a), but proved unfruitful. A much more promising mechanism capable of qualitatively explaining the observed periods was brought forward more recently (Fontaine et al. 2003b) and involves the same driving process that so successfully explains the instabilities in the EC 14026 stars. This is analogous to the case of the $`\beta `$ Cep/slowly pulsating B stars on the main sequence (Dziembowski & Pamyatnykh 1993; Dziembowski, Moskalik, & Pamyatnykh 1993; Gautschy & Saio 1993).
In their paper describing the excitation mechanism in the slow pulsators, Fontaine et al. (2003b) built a series of representative models of subdwarf B stars along the extreme horizontal branch and computed their nonadiabatic pulsation properties in a wide range of periods for modes with $`l`$ = 0 to $`l`$ = 8. This uncovered two distinct “islands” of instability in the pulsational period vs $`T_{\mathrm{eff}}`$ diagram (as can be seen in their Figure 4). The first of these is clustered around $`T_{\mathrm{eff}}31,000`$ K and corresponds to the short period, low-order $`p`$-modes typical of the EC 14026 stars, where modes with degree values of $`l`$ = 0 and upward are driven. The second region of instability, attributed to the long period pulsators, lies at lower temperatures between 22,000 and 26,000 K and features $`g`$-modes with long periods similar to those observed in the slow oscillators. However, in the majority of models only modes with degree indices of $`l`$ = 3 and higher can currently be excited, which would imply that the long-period luminosity variations observed in the cooler subdwarfs correspond to modes with $`l`$ = 3, 4, and 5. This is in conflict with canonical wisdom, which suggests that modes with $`l3`$ should generally not be observable due to cancellation effects when integrating over the visible disk of the star. While Fontaine et al. (2003b) argue that the low amplitudes of the variations observed point to modes of relatively high degree indices, they concede that the instability calculations are subject to a real blue-edge problem. In particular, we now know of long-period pulsators with effective temperatures as high as $`28,000`$ K, which according to current theory should excite only modes with $`l8`$. As it does not seem feasible that these could be observed even in the best of circumstances, it is clear that we face major challenges as far as mode identification for the slow pulsators is concerned. However, there is scope for improvement on the theoretical side. Following a suggestion of Hideyuki Saio, we are currently developing more realistic subdwarf B star models that incorporate the presence of helium in the iron radiative levitation calculations, rather than adopting the pure hydrogen background assumed by Fontaine et al. (2003b). While this may well lead to a better description of the PG 1716 stars’ blue edge, we nevertheless feel it would be highly beneficial if the degree indices of the periods observed could be determined using a method completely independent of the instability calculations.
The best way of achieving this is to exploit the wavelength dependence of a mode’s pulsational amplitude, which depends on $`l`$ as well as on other parameters such as the viewing angle and the intrinsic amplitude of the oscillation in question. By calculating the ratio of amplitudes in different wavebands, the latter unknown quantities can be elimitated, leaving (to a first approximation) a dependence only on $`l`$ and the atmospheric parameters of the star. The theoretical colour-amplitude ratios can thus be computed for a given target and compared to pulsational amplitudes from multi-colour photometry in order to determine the modes’ degree indices. Likeways, phase shifts between oscillations in different wavebands may be exploited in certain cases. This method is by no means revolutionary and has been applied to many different types of pulsating star, such as $`\delta `$ Scuti stars (Garrido, García-Lobo, & Rodriguez 1990), $`\beta `$ Cepheids (Cugier, Dziembowski, & Pamyatnykh 1994), ZZ Ceti white dwarfs (Robinson et al. 1995; Fontaine et al. 1996), and $`\gamma `$ Doradus stars (Breger et al. 1997), to name just a few. Attempts to understand the observed period spectrum on the basis of multi-colour photometry have also been made for EC 14026 stars, first by Koen (1998) and more recently by Jeffery et al. (2004). The former study gives a qualitative interpretation of the periods observed for KPD 2109+4401 based on the theory of Watson (1988), while the latter asserts to have provisionally identified the modes detected for the fast oscillators KPD 2109+4401 and HS 0039+4302. In this case, the identification is based on amplitude ratios computed by Ramachandran, Jeffery, & Townsend (2004) as well as on comparisons with the pulsational properties of the evolutionary sequences published by Charpinet et al. (2002).
In the present study, we assess the potential of multi-colour photometry for pulsating sdB stars in some detail. Using a grid of specially designed model atmospheres, we are able to compute vital quantities such as the emergent specific intensities and their derivatives to high accuracies. Moreover, we carry out full nonadiabatic pulsation calculations in order to obtain eigenfunctions that are as realistic as possible, an aspect that was not studied by Ramachandran et al. (2004). In the next section, we briefly review the theory of colour-amplitude variations in pulsating stars. We then describe our computations for subdwarf B stars, and present results for a representative EC 14026 and PG 1716 model before examining the influence of the atmospheric parameters of the model in question. We end with a discussion of the practical applications of the tools developed.
## 2 BASIC THEORY
The theoretical foundations for the modelling of lightcurves of a star undergoing non-radial pulsations in the linear regime were laid by the pioneering calculations of Osaki (1971), Dziembowski (1977), Balona & Stobie (1979), and Buta & Smith (1979). Based on the Baade-Wesselink technique, their work exploited the wavelength dependence of a pulsational mode’s amplitude and phase and enabled the inference of its degree index $`l`$ based on light and radial velocity observations. The equations were reformulated by Stamford & Watson (1981) and Watson (1988) for use with multi-colour photometric data alone. Their approach proved very convenient for practical purposes, and has been applied to many types of non-radially pulsating stars by a host of different authors. More recent versions, comparable in scope to the Watson method, have been presented by Cugier, Dziembowski, & Pamyatnykh (1994), Heynderickx, Waelkens, & Smeyers (1994), Balona & Evers (1999), Cugier & Daszyńska (2001), and Townsend (2002). The latest advancement was proposed by Dupret et al. (2003) and consists of including a detailed discussion of non-adiabatic effects in the optically-thin atmospheric layers. All of the above treatments model the lightcurves of non-radially pulsating stars in terms of perturbations to the photospheric pressure or surface gravity, the effective temperature and the stellar radius. For cases where temperature effects completely dominate the brightness variations $``$ in pulsating white dwarfs for instance $``$ a simpler approach is possible, as was first discussed by Robinson, Kepler and Nather (1982) and later exploited for ZZ Ceti white dwarfs by Brassard, Fontaine, & Wesemael (1995, hereafter BFW95). For our application to pulsating sdB stars below, we mostly follow the presentations of Cugier and Daszyńska (2001) and Townsend (2002), but adopt a notation closer to that of BFW95.
The Lagrangian perturbations to the stellar radius $`R`$, the effective temperature $`T_{\mathrm{eff}}`$, and surface gravity $`g_s`$ caused by a single non-radial pulsation may be expressed as
$`{\displaystyle \frac{\delta R}{R}}(\theta ,\varphi ,t)`$ $`=`$ $`\mathrm{}\left[{\displaystyle \frac{\delta r}{r^0}}Y_l^m(\theta ,\varphi )e^{i\omega t}\right],`$ (1)
$`{\displaystyle \frac{\delta T_{\mathrm{eff}}}{T_{\mathrm{eff}}}}(\theta ,\varphi ,t)`$ $`=`$ $`\mathrm{}\left[{\displaystyle \frac{\delta T}{T^0}}Y_l^m(\theta ,\varphi )e^{i\omega t}\right],`$ (2)
$`{\displaystyle \frac{\delta g_s}{g_s}}(\theta ,\varphi ,t)`$ $`=`$ $`\mathrm{}\left[{\displaystyle \frac{\delta g}{g^0}}Y_l^m(\theta ,\varphi )e^{i\omega t}\right],`$ (3)
where $`\mathrm{}[\mathrm{}]`$ denotes the real part of a complex quantity, $`\omega `$ is the complex angular eigenfrequency of the mode (we define $`\omega _{nlm}\mathrm{}[\omega ]=2\pi /P_{nlm}`$, where $`P_{nlm}`$ is the pulsation period of a mode of radial order $`n`$, harmonic degree $`l`$ and azimuthal order $`m`$), $`t`$ is the time coordinate, $`Y_l^m(\theta ,\varphi )`$ is the usual spherical harmonic function describing the angular dependence of the mode, and $`\frac{\delta r}{r^0}`$, $`\frac{\delta T}{T^0}`$ and $`\frac{\delta g}{g^0}`$ are the radial components of the complex eigenfunctions. We define the spherical harmonics in terms of the polar angle $`\theta `$ and the azimuthal angle $`\varphi `$ using the nomenclature of Jackson (1975) as
$$Y_l^m(\theta ,\varphi )=\sqrt{\frac{2l+1}{4\pi }\frac{(lm)!}{(l+m)!}}P_l^m(\mathrm{cos}\theta )e^{im\varphi },$$
(4)
where the associated Legendre functions $`P_l^m`$ can be generated by the equation
$$P_l^m(x)=\frac{(1)^m}{2^ll!}(1x^2)^{m/2}\frac{d^{l+m}}{dx^{l+m}}(x^21)^l.$$
(5)
With this definition, we may introduce the real quantity
$$\overline{Y}_l^m(\theta )Y_l^m(\theta ,\varphi )e^{im\varphi },$$
(6)
which will be useful below and is described in more detail by BFW95. The radial components of the complex eigenfunctions may be written as
$`{\displaystyle \frac{\delta r}{r^0}}=`$ $`\left|{\displaystyle \frac{\delta r}{r^0}}\right|e^{i\varphi _r}`$ $`ϵ_re^{i\varphi _r},`$ (7)
$`{\displaystyle \frac{\delta T}{T^0}}=`$ $`\left|{\displaystyle \frac{\delta T}{T^0}}\right|e^{i\varphi _T}`$ $`ϵ_Te^{i\varphi _T},`$ (8)
$`{\displaystyle \frac{\delta g}{g^0}}=`$ $`\left|{\displaystyle \frac{\delta g}{g^0}}\right|e^{i\varphi _g}`$ $`ϵ_ge^{i\varphi _g},`$ (9)
where we have introduced the dimensionless amplitudes (moduli) $`ϵ_r`$, $`ϵ_T`$ and $`ϵ_g`$ of the complex radial eigenfunctions, as well as their phases $`\varphi _r`$, $`\varphi _T`$ and $`\varphi _g`$ with respect to some arbitrary value, which by convention is set to $`\varphi _r`$ = 0. Note that the upperscript “ <sup>0</sup> ” indicates the unperturbed value of the variable of interest in the atmospheric layers.
In the Watson approach, the brightness variation of a non-radially pulsating star is calculated by assuming a dependence of the local flux on the local instantaneous effective temperature and surface gravity, and then integrating over the observed disk taking into account the geometry of the mode and the radial distortions involved. As implemented by Townsend (2002), the first order perturbation to the emergent Eddington flux of a star undergoing a single non-radial pulsation can thus be written in terms of the radial components of the complex eigenfunctions detailed in equations 7$``$9 as
$`{\displaystyle \frac{H_\nu ^1}{H_\nu ^0}}`$ $`=`$ $`\mathrm{}[(\left\{(2+l)(1l){\displaystyle \frac{\delta r}{r^0}}{\displaystyle \frac{I_{l\nu }^0}{I_{0\nu }^0}}\right\}+\left\{{\displaystyle \frac{\delta T}{T^0}}{\displaystyle \frac{1}{I_{0\nu }^0}}{\displaystyle \frac{I_{l\nu }^0}{\mathrm{ln}T^0}}\right\}+`$ (10)
$`+\left\{{\displaystyle \frac{\delta g}{g^0}}{\displaystyle \frac{1}{I_{0\nu }^0}}{\displaystyle \frac{I_{l\nu }^0}{\mathrm{ln}g^0}}\right\})Y_l^m(\theta _0,\varphi _0)e^{i\omega t}].`$
Keeping with the notation of BFW95, $`H_\nu ^0`$ is the unperturbed emergent monochromatic Eddington flux, $`H_\nu ^1`$ is its first-order perturbation, and $`I_{l\nu }^0`$ is the angle-integrated unperturbed emergent monochromatic specific intensity
$$I_{l\nu }^0=_0^1I_\nu ^0(\mu )P_l(\mu )\mu 𝑑\mu ,$$
(11)
with a weight function given by a Legendre polynomial $`P_l(\mu )`$. The angles $`(\theta _0,\varphi _0)`$ are the angular coordinates of the observer in the spherical coordinate system of the star. The quantity $`I_{0\nu }^0`$ refers to the specific case of $`I_{l\nu }^0`$ where the degree index $`l=0`$. It is also understood that the derivative of $`I_{l\nu }^0`$ with respect to the effective temperature $`T^0`$ (surface gravity $`g^0`$) is taken at constant surface gravity (effective temperature).
In order to simplify equation (10) and to enable the eventual elimination of the intrinsic amplitude of the oscillation by taking flux ratios obtained at different wavelengths, it is necessary to link the radial components of the three eigenfunctions. By convention, we express the surface gravity and temperature perturbations in terms of the radius eigenfunction. Regarding the former, Cugier and Daszyńska (2001; see also Dupret et al. 2003) have argued that one may approximate
$$\frac{\delta g}{g^0}(2+\sigma _{nlm}^2)\frac{\delta r}{r^0}D_{nlm}\frac{\delta r}{r^0}$$
(12)
in the limit where the radial gradient of the amplitude of the pressure perturbation ($`|\delta P/P^0|`$) is small. The real dimensionless quantity $`\sigma _{nlm}`$ is given by
$$\sigma _{nlm}\omega _{nlm}\sqrt{\frac{R}{g_s}},$$
(13)
where $`R`$ is the stellar radius and $`g_s`$ is the surface gravity. Given the definition of $`\delta r/r^0`$ and $`\delta g/g^0`$ in equations (7) and (9), the relationship implies that the radius and surface gravity variations occur in phase, i.e., $`\varphi _g=\varphi _r`$ (=0, by convention). It must be mentioned that equation (12) differs from the more standard expression employed in the Watson approach (see, e.g., equation (11) of Cugier et al. (1994)), which suggests
$$\frac{\delta g}{g^0}\left(4+\sigma _{nlm}^2\frac{l(l+1)}{\sigma _{nlm}^2}\right)\frac{\delta r}{r^0}C_{nlm}\frac{\delta r}{r^0}.$$
(14)
In the original theory outlined by Stamford and Watson (1981) and Watson (1988), the coefficient $`C_{nlm}`$ is multiplied by an additional term of order unity denoted by $`P^{}(\mathrm{log}g^0/\mathrm{log}P^0|_{\tau =1})`$ when used in the context of equation (14). There has been some debate as to the proper value to adopt for $`P^{}`$, with some authors (e.g., Cugier et al. 1994, Balona & Evers 1999, Townsend 2002) arguing that it should be taken strictly at unity, while most others employ grids of model atmospheres to compute precise values. Regardless of whether or not the $`P^{}`$ quantity is included in equation (14), the expression remains considerably different from that recommended by Cugier and Daszyńska (2001; our equation (12)), particularly when considering long period modes. The latter is deemed more physical, as it is more consistent with the outer boundary conditions employed during pulsation calculations for a stellar model. We thus adopt the result of Cugier and Daszyńska (2001) after verifying that we do indeed find $`|\delta P/P^0|`$ to be mostly flat in the atmospheric layers of our subdwarf B star models (as discussed below).
The radial component of the temperature eigenfunction for its part may be related to the radius perturbation by the expression
$$\frac{\delta T}{T^0}=\frac{\left|\frac{\delta T}{T^0}\right|}{\left|\frac{\delta r}{r^0}\right|}e^{i(\varphi _T\varphi _r)}\frac{\delta r}{r^0}\frac{ϵ_T}{ϵ_r}e^{i\psi _T}\frac{\delta r}{r^0},$$
(15)
where we have introduced $`\psi _T`$ as the phase lag between the temperature and the radius perturbation. In the adiabatic approximation, we set $`\psi _T=\pi `$, since maximum temperature occurs at minimum radius for $`p`$-modes. In the non-adiabatic case, the phase lag can be evaluated using
$$\psi _T=\mathrm{tan}^1\left(\frac{\mathrm{}\left[\frac{\delta T}{T^0}\right]}{\mathrm{}\left[\frac{\delta T}{T^0}\right]}\right)\mathrm{tan}^1\left(\frac{\mathrm{}\left[\frac{\delta r}{r^0}\right]}{\mathrm{}\left[\frac{\delta r}{r^0}\right]}\right),$$
(16)
where $`\mathrm{}[\mathrm{}]`$ indicates the imaginary part, and $`\mathrm{}[\mathrm{}]`$ denotes the real part of a complex quantity. We will apply this relation to detailed non-adiabatic calculations with the aim of modelling $`\psi _T`$ as a function of depth, period, and degree index $`l`$ below.
In the standard Watson approach, it is customary to introduce a second dimensionless real parameter $`R`$, which measures the departure from adiabacity of the amplitude factor in equation (15), and can hence be defined as
$$R\frac{ϵ_T/ϵ_r}{\left|\left(\frac{\delta T}{T^0}\right)_{ad}\right|/\left|\left(\frac{\delta r}{r^0}\right)_{ad}\right|}.$$
(17)
It is obvious that, in the adiabatic limit, $`R`$=1. As for the non-adiabatic case, many authors seem to believe that physically acceptable values of $`R`$ are confined to the range $`0R1`$. We beg to differ, since we see no physical reason for $`R`$ not to exceed 1 and, indeed, find no justification for the constraint in the literature. As far as we can see, it arose primarily from the fact that observations of certain types of pulsating stars (such as the $`\beta `$ Cephei, $`\delta `$ Scuti and short-period Cepheids mentioned by Stamford & Watson 1981) indicated $`0.25R1`$. While similar values have been recovered computationally for other classes of pulsator (see, e.g., Townsend 2002 describing slowly pulsating B stars), predictions for white dwarfs have yielded $`R`$ values greater than 1 in some cases (Robinson, Kepler, & Nather, 1982). This is also what our pulsation calculations indicate for the slowly pulsating subdwarf B stars, as will be decribed in the next section.
The quantities appearing in the denominator of equation (17) can be described by recalling a well-known expression relating the moduli of the temperature perturbation and the radius perturbation under the assumption of adiabacity and the Cowling approximation, namely
$$\left|\left(\frac{\delta T}{T^0}\right)_{ad}\right|=_{ad}C_{nlm}\left|\left(\frac{\delta r}{r^0}\right)_{ad}\right|,$$
(18)
where $`_{ad}(=1\mathrm{\Gamma }_2^1)`$ is the usual adiabatic temperature gradient and $`C_{nlm}`$ is the same coefficient as introduced in equation (14). Here the eigenfunctions $`(\delta T/T^0)_{ad}`$ and $`(\delta r/r^0)_{ad}`$ are real quantities, and the minus sign corresponds to a phase shift of $`\psi _T=\pi `$. Using this relation together with the definition of $`R`$, we can reformulate equation (15) to yield
$$\frac{\delta T}{T^0}=R_{ad}C_{nlm}e^{i\psi _T}\frac{\delta r}{r^0}.$$
(19)
Note that, in practice, the most direct way of computing the relationship between $`\delta T/T^0`$ and $`\delta r/r^0`$ is to use equation (15), calculating their relative phase $`\psi _T`$ from equation (16), and evaluating their amplitude ratio using
$$\frac{ϵ_T}{ϵ_r}=\frac{\left\{\left(\mathrm{}\left[\frac{\delta T}{T^0}\right]\right)^2+\left(\mathrm{}\left[\frac{\delta T}{T^0}\right]\right)^2\right\}^{1/2}}{\left\{\left(\mathrm{}\left[\frac{\delta r}{r^0}\right]\right)^2+\left(\mathrm{}\left[\frac{\delta r}{r^0}\right]\right)^2\right\}^{1/2}}.$$
(20)
In addition, we then explicitly evaluate the adiabacity parameter $`R`$ using equation (19). This will prove useful in numerical experiments aimed at assessing the impact of non-adiabatic effects on the predicted pulsational amplitudes and phases in different filters, as $`\psi _T`$ and $`R`$ can be input directly and set to the adiabatic values for instance.
Returning to equation (10), the expression describing the perturbation to the emergent Eddington flux, we now seek to reformulate the intensity terms and link them to both the BFW95 and the standard Watson (1988) notation. It can be shown that the specific intensity term appearing in the temperature perturbation can be related to the monochromatic quantity $`A_{l\nu }`$ defined by BFW95 by the expression
$$\frac{1}{I_{0\nu }^0}\frac{I_{l\nu }^0}{\mathrm{ln}T^0}=\frac{T^0A_{l\nu }}{H_\nu ^0},$$
(21)
where, according to equation (17) of BFW95,
$$A_{l\nu }=\frac{1}{2}_0^1\frac{I_\nu ^0}{T^0}P_l(\mu )\mu 𝑑\mu .$$
(22)
By following the steps described in Appendix B of BFW95, we can re-write equation (21) using the notion of the limb-darkening law $`h_\nu (\mu )`$ as
$$\frac{1}{I_{0\nu }^0}\frac{I_{l\nu }^0}{\mathrm{ln}T^0}=\alpha _{T\nu }b_{l\nu }+\frac{b_{l\nu }}{\mathrm{ln}T^0},$$
(23)
where
$$\alpha _{T\nu }\frac{\mathrm{ln}H_\nu ^0}{\mathrm{ln}T^0},$$
(24)
and
$$b_{l\nu }\frac{_0^1h_\nu (\mu )P_l(\mu )\mu 𝑑\mu }{_0^1h_\nu (\mu )\mu 𝑑\mu }.$$
(25)
The two last quantities, the logarithmic derivative of the emergent flux with respect to the effective temperature, and the weighted monochromatic limb darkening integral, are familiar notions in the Watson model. It can be seen that, when expressing the intensity term of the temperature perturbation, one has the choice of using a single expression such as the left-hand side of equation (23) employed by BFW95 or Townsend (2002), or the splitted terms on the right hand side of the equation favoured in the more traditional implementations of the Watson model. We choose to adopt the latter here in order to facilitate comparisons with the Watson approach.
Clearly, the specific intensity component of the gravity perturbation in equation (10) can be expressed in an equivalent way, yielding
$$\frac{1}{I_{0\nu }^0}\frac{I_{l\nu }^0}{\mathrm{ln}g^0}=\alpha _{g\nu }b_{l\nu }+\frac{b_{l\nu }}{\mathrm{ln}g^0},$$
(26)
with
$$\alpha _{g\nu }\frac{\mathrm{ln}H_\nu ^0}{\mathrm{ln}g^0}.$$
(27)
Finally, we use the fact that
$$I_\nu ^0(\mu )=I_\nu ^0(0)h_\nu (\mu )$$
(28)
together with the definition of $`I_{l\nu }^0`$ (equation (11)) and equation (25) to reformulate the specific intensity coefficient in the radius term of equation (10) as
$$\frac{I_{l\nu }^0}{I_{0\nu }^0}=b_{l\nu }.$$
(29)
We can now re-write the perturbation to the relative instantaneous emergent monochromatic Eddington flux of equation (10), using equations (12), (19), (23), (26) and (29), as
$`{\displaystyle \frac{H_\nu ^1}{H_\nu ^0}}`$ $`=`$ $`ϵ_r\overline{Y}_l^m(i)[\{(2+l)(1l)b_{l\nu }D_{nlm}b_{l\nu }\alpha _{g\nu }D_{nlm}{\displaystyle \frac{b_{l\nu }}{\mathrm{ln}g^0}}\}`$ (30)
$`\times \mathrm{cos}(m\varphi _0+\omega _{nlm}t)+`$
$`+\left\{R_{ad}C_{nlm}b_{l\nu }\alpha _{T\nu }+R_{ad}C_{nlm}{\displaystyle \frac{b_{l\nu }}{\mathrm{ln}T^0}}\right\}`$
$`\times \mathrm{cos}(m\varphi _0+\omega _{nlm}t+\psi _T)],`$
where we used the fact that $`\theta _0=i`$, the inclination angle, in $`\overline{Y}_l^m(i)`$, the real function giving the viewing aspect (see BFW95 for details).
In practice, the last equation will be applied to broadband, rather than monochromatic photometry, so it is necessary to express it in terms of frequency-integrated quantities. We thus introduce
$`b_{lx}`$ $``$ $`{\displaystyle \frac{_0^{\mathrm{}}W_\nu ^xb_{l\nu }𝑑\nu }{_0^{\mathrm{}}W_\nu ^x𝑑\nu }},`$ (31)
$`\alpha _{Tx}`$ $``$ $`{\displaystyle \frac{_0^{\mathrm{}}W_\nu ^x\alpha _{T\nu }𝑑\nu }{_0^{\mathrm{}}W_\nu ^x𝑑\nu }},`$ (32)
$`\alpha _{gx}`$ $``$ $`{\displaystyle \frac{_0^{\mathrm{}}W_\nu ^x\alpha _{g\nu }𝑑\nu }{_0^{\mathrm{}}W_\nu ^x𝑑\nu }},`$ (33)
where $`W_\nu ^x`$ represents the transmission function for filter $`x`$, convolved, in principle, with the response of the telescope/detector combination and the atmospheric extinction curve at a given site.
We can now regroup the various components of equation (30) into five terms analogous to those employed in the traditional Watson model. We thus introduce the following short-hand notation:
$`T_1`$ $``$ $`R_{ad}C_{nlm}b_{lx}\alpha _{Tx},`$ (34)
$`T_2`$ $``$ $`R_{ad}C_{nlm}{\displaystyle \frac{b_{lx}}{\mathrm{ln}T^0}},`$ (35)
$`T_3`$ $``$ $`(2+l)(1l)b_{lx},`$ (36)
$`T_4`$ $``$ $`D_{nlm}b_{lx}\alpha _{gx},`$ (37)
$`T_5`$ $``$ $`D_{nlm}{\displaystyle \frac{b_{lx}}{\mathrm{ln}g^0}}.`$ (38)
We finally follow Koen (1998) and define
$`\gamma _1`$ $``$ $`T_1+T_2,`$ (39)
$`\gamma _2`$ $``$ $`T_3+T_4+T_5,`$ (40)
which represent the effects on the brightness variation during a pulsation cycle due to effective temperature ($`\gamma _1`$) and radius/gravity perturbations ($`\gamma _2`$) respectively. Using a trigonometric identity we may then obtain our final expression for the relative flux in a photometric bandpass $`x`$ arising from the excitation of a single pulsation mode
$$\frac{H_x^1}{H_x^0}=ϵ_r\overline{Y}_l^m(i)A_{nlm}^x\mathrm{cos}(m\varphi _0+\omega _{nlm}t+\varphi _{nlm}^x),$$
(41)
with the wavelength-dependent amplitude given by
$$A_{nlm}^x=(\gamma _1^2+\gamma _2^2+2\gamma _1\gamma _2\mathrm{cos}\psi _T)^{1/2},$$
(42)
and the wavelength-dependent phase given by
$$\varphi _{nlm}^x=\mathrm{tan}^1\left(\frac{\gamma _1\mathrm{sin}\psi _T}{\gamma _1\mathrm{cos}\psi _T+\gamma _2}\right).$$
(43)
Strictly speaking, this expression is valid only for non-rotating stars where the amplitude and phase depend on the radial order $`n`$ and the degree index $`l`$, but not the azimuthal order $`m`$. However, Cugier & Daszyńska (2001) have argued that it may also be applied to slowly rotating stars, loosely defined as those with a spin parameter $`S=2\mathrm{\Omega }/\omega _{nlm}<0.5`$ (where $`\mathrm{\Omega }`$ is the rotation frequency). In this case, the $`\omega _{nlm}t`$ term in the cosine function of equation (41) is replaced by $`(\omega _{nlm}m\mathrm{\Omega })t`$. The amplitude and phase also become sensitive to the azimuthal index $`m`$, but only through the period dependence of an $`m`$ component in a rotationally split $`(2l+1)`$ multiplet. For fast rotators with higher values of $`S`$, things become more complicated and a rigorous treatment such as the one developed by Townsend (2003) becomes necessary.
In practice, equation (41) is not used directly since it depends on the unknown factor $`ϵ_r\overline{Y}_l^m(i)`$. Instead, we take advantage of the wavelength independence of the latter and eliminate it by calculating the amplitude ratios $`A_{nlm}^x/A_{nlm}^y`$ and phase differences $`\varphi _{nlm}^x\varphi _{nlm}^y`$ arising from the lightcurves for two different bandpasses $`x`$ and $`y`$. Mode discrimination can then proceed by exploiting the fact that the wavelength dependence of an oscillation’s amplitude and phase may depend strongly on its degree index $`l`$, and that observational amplitude ratios and phase differences are readily obtainable from multi-colour photometry.
We note that if non-adiabatic effects are neglected in the calculations of the theoretical quantities, then $`\psi _T=\pi `$, the phase $`\varphi _{nlm}`$ calculated from equation (43) will always be zero, and no phase shifts will be predicted between different bandpasses. However, a small or negligible observed phase shift does not necessarily imply the absence of non-adiabatic effects, and one may well encounter situations where the expected phase shifts remain quite small despite important deviations of the adiabacity parameter $`R`$ from its adiabatic value of $`R=1`$. This is the case for our model of a typical EC 14026 star discussed below. On the other hand, the observed phase shifts will be expected to be negligible if effective temperature perturbations completely dominate the brightness variation and radius/surface gravity effects can safely be ignored ($`\gamma _2=0`$ in the limiting case). In that instance, equation (43) immediately infers $`\varphi _{nlm}^x=\psi _T`$ irrespective of the bandpass in question. Moreover, the period dependence of the $`T_1`$ and $`T_2`$ terms through their common factor $`RC_{nlm}`$ cancels out in the calculation of the amplitude ratios, with the result that the latter bear the signature only of the degree index $`l`$ and not the period dependent radial order $`n`$. By the same logic, the amplitude ratios are no longer affected by non-adiabatic effects, a point on which we concur with Ramachandran et al. (2004). In retrospect, this alleviates the worry expressed by Robinson et al. (1982) about the legitimacy of using the adiabatic relationship between $`\delta T/T^0`$ and $`\delta r/r^0`$ in their discussion of colour variations in pulsating white dwarfs (see also BFW95). When considering the pulsations of white dwarfs in the linear regime, the brightness variations are completely dominated by temperature effects, which implies an absence of phase shifts between the light curves of different colours, as well as amplitude ratios insensitive to non-adiabatic effects and the period of the mode in question. The latter then depend only on the degree index $`l`$, a situation also encountered for high-order $`g`$-modes in our typical PG 1716 star model discussed below.
## 3 MODEL ATMOSPHERES AND MONOCHROMATIC QUANTITIES
The quantities required in the theoretical framework discussed above can broadly be divided into three groups: those that can be inferred observationally, those that must be computed on the basis of full stellar models, and finally those that are derived from model atmospheres. In this section we focus on the latter group, which includes the monochromatic quantities $`\alpha _{T\nu }`$, $`\alpha _{g\nu }`$, $`b_{l\nu }`$, $`b_{l\nu ,T}b_{l\nu }/\mathrm{ln}T^0`$, and $`b_{l\nu ,g}b_{l\nu }/\mathrm{ln}g^0`$. Since their computation involves not only the standard specific intensities, but also the corresponding derivatives with respect to effective temperature and surface gravity across the visible disk, we needed to modify our model atmosphere code for subdwarf B stars in order to carry out the task efficiently and accurately. We constructed a grid of model atmospheres defined at 9 gravity points (log $`g`$ = 4.8 to 6.4 in steps of 0.2 dex) and 11 temperature points ($`T_{\mathrm{eff}}`$ = 20,000 to 40,000 K in steps of 2000 K) representative of the distribution of sdB stars in the $`\mathrm{log}gT_{\mathrm{eff}}`$ plane. Detailed interpolation within this grid enables the calculation of the desired quantities for any given $`\mathrm{log}gT_{\mathrm{eff}}`$ combination. The model atmospheres are computed under the assumption of LTE and uniform composition specified by log $`N(He)/N(H)`$ =$``$2.0, a typical value for sdB’s. Metals were not included, since subdwarf B stars are known to be chemically peculiar, and metal abundances vary from one target to the next. While it could prove interesting to incorporate representative metal abundances and thus assess the importance of metals in this kind of calculation in the future, our H/He LTE model grid is quite sufficient for the purposes of this study.
As mentioned above, the colour-amplitude technique is usually applied to broadband photometry, but we find it instructive to first examine the behavior of the key monochromatic quantities. We begin with the unperturbed emergent Eddington flux, which is shown in the top panel of Figure 1 in the optical domain for a model with $`T_{\mathrm{eff}}`$ = 33,000 K and log $`g`$ = 5.75. These are also the atmospheric parameters we adopt for our representative EC 14026 star model in the next section, where more details are provided. As is typical for the observed optical spectra of sdB stars, the Eddington flux is characterized by the presence of broad hydrogen Balmer lines and several weak and narrow helium lines. For comparison, the middle panel illustrates the predicted first-order perturbation to the emergent Eddington flux assuming a nonradial pulsation with a period of 150 s (a typical low-order $`p`$-mode in an EC 14026 star) and six values of the degree index from $`l`$ = 0 (top curve) to $`l`$ = 5 (bottom curve). Note that here, $`H_\nu ^1`$ is divided by the unknown factor $`ϵ_r\overline{Y}_l^m(i)`$. It is evident that the pulsational amplitude rapidly decreases with increasing $`l`$, which is a direct manifestation of the well-known geometric cancellation effects associated with an increasing number of nodal lines crisscrossing the visible disk. For values of $`l`$ $`>`$ 2, the decrease of the amplitude with increasing $`l`$ at a given frequency is no longer monotonic, but also depends on the limb-darkening law of the model atmosphere in question in quite a complex way. The amplitude of the $`l`$ = 4 curve for instance is higher than that of the $`l`$ = 3 curve in the optical domain shown, and the latter dips below the $`l`$ = 5 curve above $``$ 4000 Å. By dividing the curves by the unperturbed flux $`H_\nu ^0`$, one obtains the $`relative`$ monochromatic amplitude of the assumed mode (again to within the unknown parameter $`ϵ_r\overline{Y}_l^m(i)`$), as indicated in the bottom panel of the figure. It is the latter quantity that forms the basis of the colour-amplitude technique, which relies on comparing the relative amplitude at two wavelengths in order to eliminate the unknown factor.
The result of such an operation is illustrated in Figure 2, where we have divided the relative monochromatic amplitude curves for each $`l`$ by the corresponding relative amplitude at an (arbitrary) frequency point in the continuum with $`\lambda `$ = 3650 $`\mathrm{\AA }`$. In this particular example, there is little difference between curves with $`l`$ = 0, 1, and 2, however those with $`l`$ = 3, 4, and 5 bear a stronger and more dictinct signature of their degree index. Along with the amplitude ratios, the phase differences between the brightness variations at different wavelengths may also be used to infer the degree $`l`$ of a pulsation mode. This is shown in Figure 3, where we plot the monochromatic phase difference with respect to the spectral point at 3650 $`\mathrm{\AA }`$ for the same assumed pulsation mode with a period of 150 s and $`l`$ values from 0 to 5 as indicated. In our example, the phase shifts remain relatively small (less than a few degrees) over the optical domain. This implies that, in practice, mode discrimation will be difficult to achieve on the basis of phase differences alone.
The amplitude ratios and phase shifts discussed in the following sections are simply frequency-integrated counterparts to the monochromatic curves pictured in the last two figures. In this context it should be noted that the spikes associated with the central cores of the absorption lines, and in particularly those related to the narrow helium lines, do not significantly contribute to the bandpass integrated quantities. We would like also to recall that the behaviour of the amplitude ratios and phase shifts with wavelength depends not only on the degree index $`l`$, but also on the period of the mode and, of course, on the atmospheric parameters of the model in question.
## 4 NONADIABATIC EFFECTS IN REPRESENTATIVE MODELS OF PULSATING SDB STARS
We now evaluate the quantities $`R`$ and $`\psi _T`$ through the use of full nonadiabatic pulsation calculations. To do this, we use the same numerical tools employed earlier by Charpinet et al. (1997; see also Fontaine et al. 1998, Charpinet et al. 2001, and Fontaine et al. 2003b) to construct their second-generation stellar models. These are characterized by four free parameters, the effective temperature $`T_{\mathrm{eff}}`$, the surface gravity log $`g`$, the fractional mass contained in the H-rich envelope $`M(H)/M_{}`$, and the total mass $`M_{}`$. The models feature an opacity profile that largely depends on the nonuniform distribution of iron as a function of depth. This distribution results from the competition between gravitational settling and radiative levitation and has been shown to be responsible for the excitation of low-order $`p`$-modes in models of EC 14026 stars as well as high-order $`g`$-modes in models of PG 1716 stars (Fontaine et al. 2003b) through the $`\kappa `$-mechanism. Note that, according to the same authors, standard models of sdB stars with uniform metallicity are unable to excite pulsation modes.
In this section we will focus on a representative model of an EC 14016 star and a PG 1716 pulsator respectively. According to Figure 1 of Fontaine et al. (2004), which summarizes the location of subdwarf B stars on the H-R diagram, a typical EC 14026 star has log $`g`$ $``$ 5.75 and $`T_{\mathrm{eff}}`$ $``$ 33,000 K, which implies that it is significantly denser and hotter than its typical PG 1716 counterpart at log $`g`$ $``$ 5.40 and $`T_{\mathrm{eff}}`$ $``$ 27,000 K. We adopt these values, and list some of the models’ other characteristics, including the two defining variables $`M(H)/M_{}`$ and $`M_{}`$ in Table 1. We explicitly give the value of the total radius as well as that of the adiabatic temperature gradient averaged over the atmospheric layers (see below), as both of these quantities enter into our equations. For completeness, we also provide the value of the quantity $`P^{}`$ evaluated from our model atmosphere grid at the appropriate values of ($`T_{\mathrm{eff}}`$, log $`g`$), despite the fact that we do not use that variable in our calculations, since we choose to relate the surface gravity and radius perturbation via the expression developed by Cugier $`\&`$ Daszyńska (2001).
Taking into account the range of periods observed in typical EC 14026 stars, we compute all modes with periods in the range 80$``$300 s and with values of $`l`$ from 0 to 5 for our representative short-period pulsator model. For the case of our PG 1716 star model, we calculate modes with periods in the range 2000$``$6000 s and with values of $`l`$ = 1, 2, and 3. The first results are illustrated in Figures 4 and 5, where we show the modulus of the radial component of each mode’s nonadiabatic pressure eigenfunction as a function of depth in the outermost layers for the EC 14026 and PG 1716 model respectively. We define the “atmospheric layers of interest” as those lying between the optical depths $`\tau `$ = 0.1 and $`\tau `$ = 10.0. It can be seen that about half of the modes considered in the EC 14026 model show small gradients across these layers as is required in the Cugier & Daszyńska (2001) expression connecting the surface gravity perturbation to the total radius perturbation. The other modes, which systematically correspond to those with shorter periods, do not strictly pass this test but, lacking a more generally applicable relation, there is little we can do to remedy this shortcoming. The situation is better for the PG 1716 model, where the vast majority of the modes considered show very small $`|\delta P/P|`$ gradients across the atmospheric layers, with the exception of the few modes with the longest periods. As it turns out, we will find out below that surface gravity perturbations contribute very little to the brightness variations in pulsating sdB stars, which is lucky in the context of our present “problems”, as any inaccuracies in the surface gravity perturbations will have little impact on the final results.
Figures 6a and 6b show the behaviour of the adiabaticity parameter $`R`$ in the atmospheric layers of our EC 14026 model for all modes of interest. Figures 7a and 7b illustrate similar results for the phase lag $`\psi _T`$. The figures give the distinct impression that both $`R`$ and $`\psi _T`$ depend primarily on the period and very little on the degree index $`l`$. This can, in fact, be confirmed quantitatively by computing an “average” atmospheric value of both quantities for each mode, and plotting them as functions of the mode’s period as shown in Figure 8. The averaging process used is a simple unweighted integration over all atmospheric layers between $`\tau `$ = 0.1 and $`\tau `$ =10.0. We believe that this approach is slightly more rigorous than simply taking the local values at the photosphere itself ($`\tau `$ = 2/3). Figure 8 highlights the almost perfect one-to-one relationship that exists between the value of $`<R>`$ and the period of the mode for the range of interest. A similar situation is encountered for the averaged quantity $`<\psi _T>`$. In both cases there is very little, if any, dependence on the degree index $`l`$, allowing us to model $`<R>`$ and $`<\psi _T>`$ as functions of the period $`P`$ in a simple and accurate way (we drop the subscript “$`nlm`$” in what follows for a more concise notation). While this is not necessary when examining a specific equilibrium model for which individual values can be obtained for each mode (e.g., the small circles in Figure 8), the procedure will prove useful in our applications below, where we want to treat the pulsation period as a free continuous variable rather than a discrete eigenvalue. Values of $`R`$ and $`\psi _T`$ sufficiently accurate for our needs were obtained by $`\chi ^2`$-fitting the data points in Figure 8 to cubic curves as illustrated. The cubic solutions are given by
$$<R>=0.34593.031\times 10^4P+1.784\times 10^5P^23.040\times 10^8P^3,$$
(44)
and
$$<\psi _T>=\pi +1.10101.415\times 10^2P+3.702\times 10^5P^22.122\times 10^8P^3,$$
(45)
which are formally valid in the period range 80$``$300 s and for values of the degree index in the range $`l`$ = 0$``$5. Figure 8 reaffirms the well-known fact that non-adiabatic effects are never negligible in stellar atmospheres, but it is the particular dependence of $`<R>`$ (and $`<\psi _T>`$) on the period that is of central interest here and that cannot be ignored in the computations. In future studies, the accuracy of the estimates for $`R`$ and $`\psi _T`$ could be further improved by examining the nonadiabatic pulsation equations in the presence of optically thin layers more closely, for example by following the theory of Dupret (2001) and Dupret et al. (2003).
The situation is slightly more complicated for the high-order $`g`$-modes in our representative PG 1716 model, where $`R`$ and $`\psi _T`$ depend on both the period $`P`$ and the index $`l`$ as is illustrated in Figures 9 and 10. The presence of a few trapped modes (trapped above the H-rich envelope/He core interface) renders things somewhat more challenging in terms of defining an “average” behavior for a pulsation mode. Furthermore, while $`<\psi _T>`$ depends monotonically on the period for a given $`l`$ (at least in the period range of interest), this is not the case for $`<R>`$, making it ever more tedious to acceptably fit the data points. As illustrated in Figure 11, we were finally able to find fits to $`<R>`$ and $`<\psi _T>`$ sufficiently accurate for our present needs by using the following analytic relationships
$$<R>l^{0.13}=0.5817+6.399\times 10^4(Pl^{0.38})1.164\times 10^7(Pl^{0.38})^2+6.082\times 10^{12}(Pl^{0.38})^3,$$
(46)
and
$$<\psi _T>=\pi 2.30e^{(Pl^{0.38}/1800)}.$$
(47)
Figure 11 clearly indicates that nonadiabatic effects are important in the atmospheres of PG 1716 stars, and should not be neglected in the computations. As mentioned in the theory section, the adiabaticity parameter $`<R>`$ may become larger than 1 for the long period $`g`$-modes characteristic of these objects. However, as we will see below, temperature variations increasingly dominate the brightness variations in the limit of long periods, causing the $`observable`$ effects of a departure from adiabaticity to disappear.
## 5 APPLICATION TO UBVRI PHOTOMETRY
As a representative application, we consider multi-colour photometry in the standard Johnson-Cousins system $`UBVRI`$. For each reference model (specified by a value of $`T_{\mathrm{eff}}`$ and a value of log $`g`$ for the model atmosphere calculations), we computed the frequency-integrated quantities $`\alpha _{Tx}`$, $`\alpha _{gx}`$, $`b_{lx}`$, $`b_{lx,T}b_{lx}/\mathrm{ln}T^0`$, and $`b_{lx,g}b_{lx}/\mathrm{ln}g^0`$, where the subscript “$`x`$” denotes one of the wavebands. In our modelling of the effective response curves $`W_\nu ^x`$, we convolved the standard transmission functions of the $`UBVRI`$ filters with the extinction curve of Kitt Peak National Observatory (representative of sites with 2000$``$2500 m altitudes). We also assumed a grey response for the instrument/telescope combination. Our values for the frequency-integrated quantities are given in Table 2 (Table 3) for our reference EC 14026 (PG 1716) star model. For the former model, we list the values for modes with degree indices in the range $`l`$ = 0$``$5, while for the latter they are provided for modes with $`l`$ between 1 and 6 since the case $`l`$ = 0 is of no interest for the long-period variables (as a rule, the $`p`$-branch period spectra of sdB stars do not reach into the range of the long periods observed in PG 1716 stars).
We draw attention to the fact that in both Table 2 and Table 3 $`\alpha _{gx}<<\alpha _{Tx}`$ and $`b_{lx,g}<<b_{lx,T}`$ with only very few exceptions. This means that the $`T_4`$ and $`T_5`$ terms will generally be very small compared to the $`T_1`$ and $`T_2`$ terms in the equations discussed in Section 2, implying that in sdB stars the contributions to the brightness variations due to surface gravity perturbations are small compared to the effects of effective temperature perturbations ($`T_1`$ and $`T_2`$) and radius changes ($`T_3`$). Thus, the exact expression used to relate the surface gravity perturbation to the radius perturbation (see our discussion of the Cugier & Daszyńska 2001 method versus the more traditional one in Section 2) is of minor importance for sdB stars.
Employing the data listed in Tables 1, 2, and 3 as well as our polynomial models for $`R`$ and $`\psi _T`$ given by equations (44) through (47), we computed amplitudes and phases for a number of modes according to our equations (42) and (43). The period was treated as a free parameter, which enabled a detailed examination of its influence on the final results. Keeping with tradition, the amplitudes ratios and phases shifts were calculated with respect to the bluest filter, in this case the $`U`$ bandpass. The results for our representative EC 14026 star model and those for our PG 1716 model are discussed separately in the following subsections.
### 5.1 Results for our representative EC 14026 star model
We first show, in Figure 12, the ratio $`|\gamma _1/\gamma _2|`$ as a function of effective wavelength in the five filters considered for three representative periods spanning the range of those observed in a typical EC 14026 pulsator. All the modes illustrated are associated with degree indices $`l`$ = 0$``$5 and correspond to low-order $`p`$-modes. Excepting the modes with $`l`$ = 1, for which $`T_3`$ is always equal to 0 (see our equation (36)), as well as the $`l`$ = 3 modes in the $`I`$ bandpass, for which the contributions of the effective temperature perturbations are particularly small, the figure infers that both effects (i.e., temperature and radius changes) contribute to the brightness variations in a non-negligible way. Unlike for white dwarfs, it is thus not appropriate to assume that temperature effects completely dominate in EC 14026 stars. While we already mentioned that the $`T_4`$ and $`T_5`$ terms usually remain quite small, it is the $`T_3`$ term, whose numerical value dominates $`\gamma _2`$, that becomes appreciable compared to $`\gamma _1`$ (= $`T_1+T_2`$).
In order to assess the importance of non-adiabatic effects, we repeated the calculation of $`|\gamma _1/\gamma _2|`$, this time imposing the adiabatic values $`R`$ = 1 and $`\psi _T`$ = $`\pi `$ for all modes considered. The results are represented by the dotted lines in Figure 12. Keeping in mind the logarithmic scale of the ordinate axis, it is obvious that non-adiabatic effects are significant in the composition of brightness variations in EC 14026 pulsators.
Figure 13 shows the expected phase shifts for the same set of modes. It can be seen that, in the optical domain, the phase shifts depend on the period quite sensitively, while remaining relatively small and reaching a maximum of a few degrees at the most in the period range of interest. In the adiabatic approximation we expect no phase shifts at all. This means that, in practice, it will be very difficult to exploit the (weak) $`l`$ dependence of such phase shifts. The situation seems more promising for the amplitude ratios illustrated in Figure 14. However, while the modes with $`l`$ = 3, 4, and 5 bear distinct signatures of their degree index, the capacity to discriminate between modes with $`l`$ = 0, 1, and 2 on the basis of optical photometry appears much more limited. Observational amplitude measurements for such modes will have to be unusually precise if they are to be used as mode discriminators in EC 14026 stars. We would like to point out that non-adiabatic effects on the amplitude ratios are generally non-negligible, as can be deduced by comparing the continuous lines (non-adiabatic results) to the dotted lines (adiabatic values) in Figure 14. It may also be worth mentioning that the results depicted in the middle panels of Figures 13 and 14 (the case with $`P`$ = 150 s) are consistent with the monochromatic results illustrated in Figures 2 and 3, as is to be expected.
In Figures 15a and 15b (separated for visualization purposes), we plot the expected amplitude ratios for modes with $`P`$ = 100 s (dotted lines), 150 s (dashed lines), and 200 s (long-dashed lines), with the aim of emphasizing their strong period dependence. In order to compare the results to corresponding period-independent values, we also include amplitude ratios (solid lines) obtained by postulating that effective temperature perturbations completely dominate the brightness variations (i.e. by setting $`\gamma _2`$ = 0 in the calculations). As suggested above, it becomes evident that this not a good working assumption for the EC 14026 stars.
Finally, in Figure 16, we show the results of a numerical experiment in which we explicitly adopt $`T_3`$ = 0 in order to assess the relative impact of the $`T_4`$ and $`T_5`$ terms on the amplitude ratios. The resulting period-dependent ratios are indicated by the dotted lines and contrast with the continuous lines corresponding to the period-independent limiting case in which $`\gamma _2`$ = 0. While we previously argued that the contributions of the $`T_4`$ and $`T_5`$ terms are generally be quite small in pulsating sdB stars, the figure reveals that non-negligible effects remain for certain cases, especially in the $`B`$ band. We thus recommend that these terms be kept by in the calculations, particularly since they are readily computable.
### 5.2 Results for our representative PG 1716 star model
The computations for our reference PG 1716 model uncover two distinct regimes as far as the relative importance of the temperature perturbation is concerned. At the lower end of the period range of interest, contributions from both the $`\gamma _1`$ and $`\gamma _2`$ terms are significant, whereas the brightness variations are dominated by temperature changes for the longer periods involved. This can be seen from Figures 17a and 17b, where we show the ratio $`|\gamma _1/\gamma _2|`$ as a function of effective wavelength in a format similar to that of Figure 12. In these and the following figures, we illustrate our results for six different periods between $`P`$= 2400 s and $`P`$ = 6000 s. Comparing Figures 17a and 17b clearly reveals that the ratio $`|\gamma _1/\gamma _2|`$ monotonically increases with period for all filters and degree indices $`l`$. This is a direct consequence of the the $`\gamma _1`$ term increasing in absolute value with period via the period-dependent coefficient $`C_{nlm}`$ for the long periods considered, while $`T_3`$ remains constant for a given filter and degree index. We recall that modes with $`l`$ = 1 represent a special case, since $`T_3`$ = 0 in that instance.
This transition from one regime to the other strongly influences the phase shifts and amplitude ratios calculated. The former are depicted in Figures 18a and 18b for the same set of modes as considered above. For the shorter periods we predict significant phases shifts, which reach more than 10 degrees for the 2400 s period. Unlike the much smaller phase shifts expected for the EC 14026 stars, this is well within achievable measurement accuracy. At the same time, we find the magnitude of the expected phase shifts to drop very rapidly with increasing period in PG 1716 stars (note, in particular, the change of the ordinate scale between Fig, 18a and Fig. 18b), to the point where they would not be detected with any confidence for the longer period modes.
The expected amplitude ratios are shown in Figures 19a and 19b. Once again, the transition from the period-dependent regime, where $`\gamma _1`$ and $`\gamma _2`$ are of similar importance, to the period-independent case dominated by effective temperature perturbations is striking. The plots also emphasize the diminishing importance of nonadiabatic effects with increasing period, as can be seen from the convergence of the dotted (adiabatic) and continuous (non-adiabatic) lines. Thus, in our PG 1716 model nonadiabatic effects range from extremely significant for $`P`$ = 2400 s to practically negligible for $`P`$ = 6000 s.
In Figures 20a and 20b (separated for visualization purposes), we compare the expected amplitude ratios for modes with $`P`$ = 3000 s (dotted lines), 4000 s (dashed lines), 5000 s (long-dashed lines), 6000 s (dot-dashed lines), 7000 s (dot-long-dashed lines), and 8000 s (dashed-long-dashed lines), again with the aim of highlighting the period dependence of the results. The convergence of the ratios to the period-independent values (solid lines) in the limiting case of long periods is recovered nicely. As before, the period-independent ratios are obtained by assuming that effective temperature perturbations completely dominate the brightness variations (i.e., by setting $`\gamma _2`$ = 0 in the calculations).
Analogously to the case presented in Figure 16 for our EC 14026 model, we carry out supplementary calculations where we explicitly set $`T_3`$ = 0 in order to assess the relative impact of the $`T_4`$ and $`T_5`$ terms on the amplitude ratios. The results of this numerical experiment are shown in Figure 21. As in Figure 16, the resulting period-dependent ratios are indicated by dotted lines and are compared to the solid lines representing the period-independent limiting case for which $`\gamma _2`$ = 0. In contrast to the previous plot, the contributions of the $`T_4`$ and $`T_5`$ terms are essentially negligible in PG 1716 stars, apart for in the very shortest periods. For the three longest periods (not illustrated) the two curves are not distinguishable.
Having discussed the signature of the period and the degree index, as well as that of the different terms contributing to the brightness variations on bandpass-integrated wavelength, we now focus specifically on results expected from the $`UI`$ filter combination (which corresponds to the longest wavelength base in the system used here). Figure 22 shows the $`I`$ to $`U`$ amplitude ratio as a function of the degree index $`l`$ on the basis of calculations extending to $`l`$ = 8. From bottom to top, the various curves correspond to periods with $`P`$ = 2400, 3000, 4000, 5000, 6000, 7000 and 8000 s respectively. The top curve represents an infinite period, to which the finite period curves converge in the limiting case. While evident from previous figures, the peculiar behaviour of the $`l`$=3 and 5 modes is particularly striking here. In practice, these are the modes most likely to be identified from multi-colour photometry, since the amplitude ratios of the remaining modes may not differ enough to the extent where they can be resolved observationally.
Finally, Figure 23 and 24 respectively show the $`I/U`$ amplitude ratio and the $`IU`$ phase shift expected for our PG 1716 model as a function of period for modes with $`l`$=1 (black), 2 (grey), 3 (blue), 4 (cyan), 5 (green) and 6 (red). Note that here, the periods illustrated were not considered as free parameters, but are the solution of the eigenvalue problem. Once again, the convergence of values to the period-independent value in the limiting case of long periods is evident. Interestingly, in Figure 23 the results for the very shortest periods break with the systematic decrease in amplitude ratio with period, and slightly increase instead. These same periods are associated with particularly large (potentially detectable) phase shifts in Figure 24, however it should be noted that these are shorter than the periods so far observed in PG 1716 stars and of primarily theoretical interest.
### 5.3 Results for other models: Dependence on $`T_{\mathrm{eff}}`$ and log $`g`$
In this section, we go beyond the two representative subdwarf B star models explored above and discuss the dependence of amplitude ratios and phase differences on the atmospheric parameters $`\mathrm{log}g`$ and $`T_{\mathrm{eff}}`$. To this end, we constructed a sequence of subdwarf B star models parallel to the zero-age extreme horizontal branch, the vital parameters of which are detailed in Table 4. Note that the surface gravity increases with effective temperature, while the thickness of the hydrogen-rich outer layer decreases. The division between the PG 1716 and EC 14026 regime is based on spectroscopically determined values of $`\mathrm{log}g`$ and $`T_{\mathrm{eff}}`$ for the two types of pulsator, such as those illustrated in Figure 1 of Fontaine et al. (2004). In addition to the fundamental parameters, Table 4 also gives the approximate period range in which modes are believed to be excited for each model. For the EC 14026 stars, these correspond to modes that are found to be unstable from our non-adiabatic pulsation calculations, since the latter have been shown to predict the observed period ranges very accurately. Note that we have uniformly decreased the iron abundance in our models by a factor of 3 compared to the original “second-generation” models, as suggested by Fontaine et al. (2003b). In the case of the PG 1716 stars the situation is more complicated due to the discrepancies that still exist between modelled and observed instability regions (see our discussion in the Introduction). We can however constrain the unstable period ranges for the PG 1716 models in our sequence on the basis of the frequencies extracted from long-period variable subdwarfs observed. To date, a quantitative analysis of the period spectrum has been possible for three targets spanning the range of PG 1716 stars in effective temperature: PG 1627+017 at $`T_{\mathrm{eff}}`$ 23,000 K (Randall et al. 2004a), PG 1338+481 at $`T_{\mathrm{eff}}`$ 26,000 K (Randall et al. 2004b) and PG 0101+039 at $`T_{\mathrm{eff}}`$ 28,000 K (Randall et al. 2005). The study of these objects revealed that the width and mode density of the excited period range, as well as the numerical values of the periods themselves, decrease systematically with increasing temperature, enabling us to infer approximate instability bands for all models in our sequence by interpolation of the observed values. It is these estimated period ranges, rather than precisely modelled quantities, that are given in Table 4.
For each model in the sequence, we computed the necessary model atmosphere parameters in the same bandpasses as used for the representative models for modes with $`l=06`$ according to their specific $`\mathrm{log}gT_{\mathrm{eff}}`$ combination. The stellar radius was likewise determined for every model on the basis of its mass and surface gravity, and its characteristic periods were evaluated from pulsation theory in the range of interest. On the other hand, the remaining three quantities derived from the full stellar models and non-adiabatic pulsation calculations were approximated to the values determined for the appropriate representative model. Thus the values of $`<_{ad}>`$ were taken from Table 1, while $`<R>`$ and $`<\psi _T>`$ were computed using equations (44) and (45) or expressions (46) and (47) for the EC 14026 and PG 1716 models respectively. The resulting accuracy of $`<R>`$ and $`<\psi _T>`$ is deemed sufficient for the illustrative purposes sought here, however it is understood that for quantitative analyses of observed multi-colour photometry the fitting process outlined in section 4 will have to be repeated for the designated object.
Figures 25 and 26 respectively show the phase shifts and amplitude ratios predicted for our sequence of subdwarf B star models from the $`IU`$ bandpass combination. They are illustrated for modes with $`l`$ = 0 (yellow, EC 14026 stars only), $`l`$ = 1 (black), $`l`$ = 2 (grey), $`l`$ = 3 (blue), $`l`$ = 4 (cyan), $`l`$ = 5 (green) and $`l`$ = 6 (red). Points of different colour are shifted slightly in effective temperature for a given model in order to facilitate viewing and the PG 1716 and EC 14026 domains are separated by the vertical dotted line. From Figure 25 we find that the phase shifts predicted are small (less than 5 degrees) for the majority of modes in both EC 14026 and PG 1716 stars, and would most likely lie below the detection threshold of multi-colour observations. Larger phase shifts of up to $``$ 10 degrees are expected for the very shortest periods in the hotter PG 1716 variables, as well as for the $`l`$ = 3 modes at the higher end of the period range excited in the hotter EC 14026 stars. While it seems plausible that the latter could be detected observationally, the potential for mode discrimination in subdwarf B stars clearly lies with the amplitude ratios depicted in Figure 26. In the case of the PG 1716 stars, the values for modes with $`l`$ = 1, 2, 4 or 6 are very similar, but can easily be distinguished from those of $`l`$ = 3 or $`l`$ = 5 modes. Even though the three regimes slowly approach each other with increasing temperature, they remain well separated irrespective of the model in question. In the case of the EC 14026 stars, we identify four distinct groups of modes with $`l`$ = 0, 1 or 2, $`l`$ = 3, $`l`$ = 5 and $`l`$ = 4 or 6. Again, the influence of the model parameters is relatively small compared to the separation between the domains and becomes vital only when attempting to discriminate between the modes of a given group.
It should be kept in mind that the results depicted in Figures 25 and 26 depend not only on the atmospheric parameters of the models in question, but also on the period range excited in each case. As such, they represent the phase shifts and amplitude ratios predicted in real subdwarf B stars and give a good indication of what would be expected from multi-colour photometry. However, in terms of a purely theoretical exploration we find it instructive to examine the impact of both the surface gravity and the effective temperature on the amplitude ratios individually. To this end, we repeated the computation process detailed above for both EC 14026 and PG 1716 stars, keeping the effective temperature (surface gravity) constant at the representative value from Table 1, and varying the surface gravity (effective temperature) within the ranges found in the sequences of Table 4. We imposed a representative period of $`P`$ = 150 s for the EC 14026 star domain and $`P`$ = 4500 s for the PG 1716 star regime. The variation of the amplitude ratio with effective temperature is illustrated in Figures 27a and 27b for short- and long-period variable subdwarf B stars respectively (see the figure captions for more detail). For both types of oscillator, the curves are reminiscent of the amplitude ratios found for our sequence of subdwarf B stars presented in Figure 26, and change gradually and monotonically with temperature. The situation is more interesting for the variation with surface gravity, depicted in Figure 28a (28b) for EC 14026 (PG 1716) models. In the case of the fast oscillators, the amplitude ratios for most of the modes move in the opposite direction compared to Figure 26, and the closely spaced $`l`$ = 0, 1, 2 graphs seem to diverge from the (constant) $`l`$ = 1 value as the surface gravity increases. For the PG 1716 models, curves representing modes of different degree indices overlap in several instances, giving rise to a behaviour similar to that encountered as a function of period (see Figure 23).
## 6 FEASIBILITY OF APPLICATION TO MULTI-COLOUR DATA
In this section we demonstrate the potential of the method developed by applying it to published multi-colour photometry. To date, the only sets of multi-colour observations that exist for pulsating subdwarf B stars are those of Koen (1998), Falter et al. (2003), Jeffery et al. (2004), and Oreiro et al. (2005). Among these, the Jeffery et al. (2004) data for the fast pulsator KPD 2109+4401 yielded the most accurate amplitude estimates. Based on three nights of Sloan filter u’g’r’ photometry with ULTRACAM (Dhillon et al., in preparation) mounted on the 4.2-m William Herschel Telescope, the authors were able to extract seven periodicities in the 180$``$200 s range. For the purposes of our brief feasibility study, we attempt to determine the degree index of the highest amplitude mode at 182.4 s only.
The first step towards finding the theoretical amplitudes in different wavebands is the calculation of the monochromatic model atmosphere quantities $`\alpha _{T\nu }`$, $`\alpha _{g\nu }`$, $`b_{l\nu }`$, $`b_{l\nu ,T}`$ and $`b_{l\nu ,g}`$ described in Section 3. Since these depend quite sensitively on the atmospheric parameters of the target, they need to be computed for KPD 2109+4401 specifically. We adopt $`T_{\mathrm{eff}}`$=31,380 K and $`\mathrm{log}g`$=5.65 as derived from our model atmosphere fit to the hydrogen Balmer and helium lines present in the time-averaged high-resolution MMT spectrum obtained by Betsy Green (private communication). These values are part of an ongoing program designed to provide homogeneous estimates of the atmospheric parameters of a large sample of subdwarf B stars, and we refer the interested reader to the forthcoming Paper by Green, Fontaine, $`\&`$ Chayer (in preparation) for more information. The monochromatic quantities are then integrated over the effective u’g’r’ wavebands, computed by convolving the Sloan bandpasses with the quantum efficiency curves of the ULTRACAM CCD chips (Vik Dhillon, private communication) and the atmospheric transparency curve of a representative observing site at 2000$``$3000 m altitude (in this case Kitt Peak National Observatory). Next, we derive the adiabacity parameters $`R`$ and $`\mathrm{\Psi }_T`$ (Section 4) for a period of 182.4 s from the non-adiabatic eigenfunctions of an envelope model characterised by the atmospheric parameters given above as well as representative values of $`M_{}`$=0.48 $`M_{}`$ and $`\mathrm{log}q(H)`$=$``$4.0. We finally calculate the pulsational amplitudes expected from the u’, g’, and r’ photometry for degree indices from $`l`$=0 to $`l`$=5 using the equations given in Section 2.
The predicted multi-colour amplitudes are fit to those observed using a $`\chi ^2`$ minimisation routine following Fontaine et al. (1986). Compared to the standard normalisation of all amplitudes to one particular waveband this is a more objective way of determining the quality of a match, since the data from all bandpasses are weighted evenly. For every degree index $`l`$ the theoretical amplitudes $`a_{theo}`$ in each of the three bandpasses i are multiplied by a scale factor $`f_l`$, chosen in such a way as to minimise
$$\chi ^2=\underset{i=1}{\overset{3}{}}\left(\frac{f_la_{theo}^ia_{obs}^i}{\sigma ^i}\right)^2,$$
(48)
where $`a_{obs}^i`$ is the amplitude observed in a given waveband and $`\sigma ^i`$ is the error on the measurement. The results of this operation for the 182.4-s mode of KPD 2109+4401 detected by Jeffery et al. (2004) are illustrated in Figure 29, and the corresponding $`\chi ^2`$ and quality-of-fit (Q) values are listed in Table 5. It is immediately obvious that the data are matched well by the predictions for an $`l`$=0 mode, the theoretical values falling within the (very small) error bars in all bandpasses. The next best fit is that of the $`l`$=1 mode, however the associated $`\chi ^2`$ residuals are a factor of 30 larger than those for $`l`$=0. In fact, if we adopt the canonical notion that a fit cannot be considered convincing unless the quality-of-fit $`Q>0.001`$ (see, e.g., Press et al. 1986), the theoretical $`l`$=0 mode is the $`only`$ one that can reproduce the observed amplitudes in a satisfactory manner. While this implies an unambiguous identification of the mode’s degree index, it should be kept in mind that the estimated $`\chi ^2`$ and $`Q`$ values are sensitive to the formal uncertainties on the observed amplitudes, which may well have been underestimated. Assuming, in an extreme case scenario, that the true errors are twice as large as those calculated by Jeffery et al. (2004), both $`l`$=0 and $`l`$=1 would provide acceptable fits. Nevertheless, the match for $`l`$=0 remains far superior, and we believe that our identification of the degree index is sound.
We would like to point out that this is the first partial mode identification in a subdwarf B star on the basis of its amplitude-wavelength dependence alone. While the net result – the mode has a degree index of $`l`$=0 – is the same as that of Jeffery et al. (2004), we were not forced to invoke additional constraints to distinguish modes with $`l`$=0, 1, and 2. The greater accuracy achieved is undoubtedly due to the fact that we were able to compute a detailed model atmosphere characteristic of KPD 2109+4401 specifically, and could incorporate non-adiabatic effects from realistic envelope models.
## 7 CONCLUSION
We have modelled the brightness variations expected across the visible disk during a pulsation cycle for both short- and long-period variable subdwarf B stars taking into account the effects of temperature, radius, and surface gravity perturbations. The quantities related to the emergent intensity and its derivatives with respect to effective temperature and surface gravity were computed with the aid of a full model atmosphere code specifically modified for this purpose. Employing full model atmospheres has led to a degree of self-consistency not often achieved in this kind of calculation for other types of pulsating stars. For instance, in the case of pulsators near the main sequence most researchers use the Kurucz (1993) model atmospheres (or their equivalent) to compute $`\alpha _{Tx}`$ and $`\alpha _{gx}`$. However, since these data do not treat limb darkening, the same authors are forced to turn elsewhere, in particular to the popular tables of Wade & Rucinski (1985), for limb darkening coefficients that allow them to approximately estimate the quantities $`b_{lx}`$, $`b_{lx,T}`$, and $`b_{lx,g}`$. Notwithstanding the fact that sdB stars are not main sequence stars, our approach alleviates the uncertainties associated with this mixed procedure.
In contrast to previous studies concerning the potential of multi-colour photometry for mode identification in sdB stars, we were able to model non-adiabatic effects in some detail. Applying full non-adiabatic pulsation calculations to representative EC 14026 and PG 1716 star models showed that these are by no means negligible. In EC 14026 stars, the two parameters measuring the departure from adiabacity of the eigenfunctions in the atmospheric layers of interest ($`<R>`$ and $`<\psi _T>`$) vary systematically with period, but are independent of the degree index $`l`$. This is fortunate as it implies that $`<R>`$ and $`<\psi _T>`$ can be accurately computed for observed periodicities without assuming any prior knowledge of $`l`$. We should mention here that, while we use the fits to $`<R>`$ and $`<\psi _T>`$ obtained from our respresentative model throughout this explorative study, the latter are sensitive to the atmospheric parameters of the model in question. As such, they must be computed individually according to the specifications of each target if a quantitative interpretation of the observational data is to be achieved. For the PG 1716 stars, the situation is more complicated as the departure from adiabacity depends on $`l`$ as well as on the period. The values of $`<R>`$ and $`<\psi _T>`$ attributed to oscillations observed will therefore constitute only rough estimates, introducing inaccuracies into the amplitude ratios and phase shifts computed. Fortunately, measurements of the longer periods excited in these stars are insensitive to non-adiabatic effects since they are dominated by temperature perturbations and $`<R>,<\psi _T>`$ cancel out in the amplitude ratios and phase shifts. It is of interest to note that our computations return $`R>1`$ for the majority of modes believed to be excited in long-period variable subdwarf B stars. Although this is in conflict with the prevailing sentiment that $`R`$ must lie in the range $`0<R<1`$, we find no physical justification for this and believe our results to be accurate.
According to our computations, the brightness variations observed in subdwarf B stars are caused primarily by temperature and radius perturbations, the contribution of surface gravity changes being small in EC 14026 stars and negligible in PG 1716 stars. For the latter, temperature effects alone dominate the flux changes in the limit of long periods. In this regime, non-adiabatic effects loose their influence on the amplitude ratios and phase shifts, and the period of the mode is no longer an issue. The adiabatic approximation is thus valid in this particular case, which immediately implies that oscillations should occur in phase at all wavelengths. Note however that conversely a lack of observed phase shifts does not automatically justify use of the adiabatic approximation. Even outside the temperature dominated regime, phase shifts are generally predicted to be small, although they may reach up to $``$ 10 degrees for the shortest periods in PG 1716 stars and certain $`l`$=3 modes in EC 14026 stars. Whereas this may be large enough for an observational detection, it is clear that mode discrimination will occur primarily on the basis of the amplitude ratios. We note that for main sequence $`g`$-mode oscillators (e.g., Aerts et al. 2004 or De Cat et al. 2005) as well as for white dwarfs (e.g., Robinson et al. 1982) the measured phase shifts are negligible.
In the case of the EC 14026 stars, it should be relatively straightforward to distinguish modes with $`l`$ = 0, 1, or 2 from those with $`l`$ = 3, $`l`$ = 4 or 6 and $`l`$ = 5. This could well prove invaluable as a consistency check for the “forward approach” in asteroseismology, which has been used to claim mode identification in a number of short period variables through the inference of modes with $`l`$ = 0, 1, 2 and 3/4 (e.g., Charpinet et al. 2005). It is interesting to note, in this connection, that the visibility of the $`l`$ = 3 modes in the optical domain is less than that of the $`l`$ = 4 modes (see Fig. 1), very much like the situation in main sequence $`p`$-mode pulsators (e.g., Heyndericks et al. 1994), and this should be taken into account in future asteroseismological exercises of the sort. Discrimination between modes with $`l`$ = 0, 1, or 2 is much more challenging and requires multi-colour photometry of unprecedented quality as well as accurate spectroscopic estimates of the atmospheric parameters for the target observed. Nevertheless, we have demonstrated that it is feasible using the highest amplitude mode detected for KPD 2109+4401 (Jeffery et al. 2004) as an example. It is not yet clear whether similar results can be achieved on the basis of other published datasets, or even for the lower-amplitude modes detected in KPD 2109+4401. This can only be answered through detailed quantitative analyses, which we plan to carry out in the near future. It will be particularly interesting to see whether we can confirm the tentative mode identifications reported by Jeffery et al. (2004) for the remaining modes detected for KPD 2109+4401 and HS 0039+4302. Regardless of the outcome of this project, we feel confident that colour-amplitude ratios and phase shifts could be measured to sufficient accuracy from future observations provided that the targets are bright enough and light curves with well resolved frequency peaks in the Fourier domain are obtained. Several consecutive nights on a 4 m-class telescope, or even a single night on an 8 m telescope for a well-chosen “simple” pulsator such as PG 1219+534 (see Charpinet et al. 2005) would likely be adequate. Alternatively, simultaneous ground- and space-based observations would be useful insofar as the frequency baseline could be extented to the UV, where the signature of the degree index on the pulsational amplitude is greater than in the visible domain. Given the observing time on the appropriate instrument, it would be interesting to monitor a target for which an asteroseismological analysis has already been completed, and compare the degree indices inferred from the two independent methods.
For the PG 1716 stars, discrimination on the basis of colour-amplitude ratios seems feasible between modes with degree indices $`l`$ = 1, 2, 4, 6 and those with $`l`$ = 3 and $`l`$ = 5. To date, the only substantial set of multi-colour photometry for a long-period variable subdwarf consists of the $``$ 250 hours of simultaneous (Johnson-Cousins) U/R data obtained for PG 1338+481. While detailed results will be presented elsewhere, a preliminary analysis of the photometry indicates amplitude ratios consistent with those predicted for $`l`$ = 1, 2, 4 or 6 rather than $`l`$ = 3 or $`l`$ = 5.<sup>2</sup><sup>2</sup>2We point out in this context that main sequence $`g`$-mode pulsators with good empirical mode identification all have $`l`$ = 1. Compared to the study of EC 14026 stars, that of the PG 1716 stars is still in its infancy, which is partly due to deficiencies in the models, and partly a result of the considerable observational challenges presented by the low amplitudes and long periods of the pulsations. Unambiguous mode identification in these objects will likely be possible only by using a combination of the “forward approach” employed for the EC 14026 stars, and inference of the degree index from multi-colour photometry, which we have developed the tools for. In the immediate future we hope that restricting, if not identifying, the degree index will clarify the current discrepancies between predicted and observed instabilities and pave the way for a more mature understanding of these exciting objects.
This work was supported in part by the Natural Sciences and Engineering Research Council of Canada and by the Fonds de recherche sur la nature et les technologies (Québec). G.F. also acknowledges the contribution of the Canada Research Chair Program.
FIGURE CAPTIONS
Fig. 1 — Behavior of some key monochromatic quantities for our representative EC 14026 star model. The latter has log $`g`$ = 5.75 and $`T_{\mathrm{eff}}`$ = 33,000 K. Top panel: Unperturbed emergent Eddington flux in the optical domain. Middle panel: First order perturbation to the flux caused by a nonradial pulsation with a period of 150 s ($`p`$-mode); at 3500 $`\mathrm{\AA }`$, from top to bottom, each curve is characterized by a value of the degree index $`l`$ = 0, 1, 2, 4, 3, and 5. Bottom panel: Similar to the middle panel, but, this time, illustrating the $`relative`$ amplitude of the perturbation.
Fig. 2 — Logarithm of monochromatic amplitude ratios with respect to an arbitrary spectral point at 3650 $`\mathrm{\AA }`$. This again refers to our representative EC 14026 star model and for a mode with a period of 150 s and degree index $`l`$ = 0 (cyan), 1 (red), 2 (blue), 3 (green), 4 (magenta), and 5 (black).
Fig. 3 — Similar to Figure 2, but for monochromatic phase differences.
Fig. 4 — Behaviour of the radial component of the pressure eigenfunction in the atmospheric layers of interest for our representative EC 14026 star model on the basis of modes with degree indices from $`l`$ = 0 to 5 for all periods in the range 80$``$300 s. The vertical dotted lines correspond, from right to left, to Rosseland optical depths of $`\tau `$ = 0.1, 1.0, and 10.0, respectively.
Fig. 5 — Similar to Figure 4, but for our representative PG 1716 model and considering modes with $`l`$= 1, 2 and 3 in the period range 2000$``$6000 s.
Fig. 6 — $`R`$ values for our representative EC 14026 model in the atmospheric layers of interest for periods in the range 80$``$300 s. The curves for different degree indices are illustrated separately as indicated.
Fig. 7 — Similar to Figure 6, but for the quantity $`\psi _T\pi `$.
Fig. 8 — Cubic fit to computed $`<R>`$ (continuous line) and $`<\psi _T>`$ (dashed line) values as a function of period for our representative EC 14026 star model.
Fig. 9 — $`R`$ values for our representative PG 1716 model in the atmospheric layers of interest for periods in the range 2000$``$6000 s. The curves for different degree indices are illustrated separately as indicated.
Fig. 10 — Similar to Figure 9, but for the quantity $`\psi _T\pi `$.
Fig. 11 — Cubic fit to computed $`<R>`$ (continuous line) and $`<\psi _T>`$ (dashed line) values as a function of period and of degree index $`l`$ for our representative PG 1716 star model.
Fig. 12 — Relative importance of the temperature terms ($`\gamma _1`$) compared to the radius and surface gravity terms ($`\gamma _2`$) for our representative EC 14026 star model in the $`UBVRI`$ bandpasses. Continuous lines indicate the results obtained by fitting $`<R>`$ and $`<\psi _T>`$ according to equations (44) and (45) while dotted curves represent results obtained by setting the two parameters to their adiabatic values of $`<R>=1`$ and $`<\psi _T>=\pi `$. Three typical periods (low-order $`p`$-modes) of 100, 150, and 200 s, as well as six values of the degree index, $`l`$ = 0$``$5, are considered. The effective wavelengths of the various bandpasses are 3650 Å($`U`$), 4400 Å($`B`$), 5500 Å($`V`$), 6500 Å($`R`$), and 8000 Å($`I`$).
Fig. 13 — Phase shifts relative to the $`U`$ filter calculated for our EC 14026 star model for the $`UBVRI`$ bandpasses. Continuous lines represent full non-adiabatic results while the dashed curve indicates the adiabatic values (which are always equal to zero).
Fig. 14 — Amplitude ratios relative to the $`U`$ filter calculated for our EC 14026 star model for the $`UBVRI`$ bandpasses. Continuous lines represent full non-adiabatic results while the dashed curves indicates the adiabatic values obtained by forcing $`<R>=1`$ and $`<\psi _T>=\pi `$.
Fig. 15 — Amplitude ratios relative to the $`U`$ filter calculated for our EC 14026 star model for the $`UBVRI`$ bandpasses. The $`l`$ indices of the modes are indicated. Dotted, dashed and long-dashed lines refer to periods with 100, 150 and 200 s respectively. They are compared to the amplitude ratios obtained in the limit where $`\gamma _2`$ = 0 (continuous curves).
Fig. 16 — Results of a numerical experiment in which the amplitude ratios relative to the $`U`$ band were obtained by adopting $`T_3=0`$ (dotted lines) and by setting $`\gamma _2=0`$ (continuous lines).
Fig. 17 — Similar to Figure 12, but referring to our representative PG 1716 model. Six periods from 2400 s to 6000 s as well as six values of $`l`$ from 1 to 6 are considered. For the continuous lines, $`<R>`$ and $`<\psi _T>`$ were derived from equations (46) and (47) while adiabatic values were imposed for the dotted lines.
Fig. 18 — Similar to Figure 13, but referring to our representative PG 1716 model. Note the change of ordinate scale between panel a) and panel b).
Fig. 19 — Similar to Figure 14, but referring to our representative PG 1716 model.
Fig. 20 — Similar to Figure 15, but referring to our representative PG 1716 model. Amplitude ratios are illustrated for representative periods of $`P`$ = 3000 s (dotted lines), 4000 s (dashed lines), 5000 s (long-dashed lines), 6000 s (dot-dashed lines), 7000 s (dot-long-dashed lines) and 8000 s (dashed-long-dashed lines). The continuous line refers to the case of $`\gamma _2`$ = 0.
Fig. 21 — Similar to Figure 16, but referring to our representative PG 1716 model.
Fig. 22 — $`I`$ to $`U`$ amplitude ratio as a function of degree index $`l`$ on the basis of modes extending up to $`l`$ = 8. From bottom to top, the various curves correspond to periods with $`P`$ = 2400, 3000, 4000, 5000, 6000, 7000, 8000 and an infinite period respectively.
Fig. 23 — Variation of the $`I/U`$ amplitude ratio with period for our representative PG 1716 model. The values for different modes are indicated by points in black ($`l`$ = 1), grey ($`l`$ = 2), blue ($`l`$ = 3), cyan ($`l`$ = 4), green ($`l`$ = 5) and red ($`l`$ = 6).
Fig. 24 — Similar to Figure 23, but for phase differences between oscillations in the $`I`$ and $`U`$ bandpasses.
Fig. 25 — Phase differences between oscillations in the $`I`$ and $`U`$ bandpasses for the sequence of subdwarf B star models listed in Table 4. The dotted vertical line divides the PG 1716 star models on the left from the EC 14026 models on the right. We illustrate modes with $`l`$ = 0 (yellow, EC 14026 models only), $`l`$ = 1 (black), $`l`$ = 2 (grey), $`l`$ = 3 (blue), $`l`$ = 4 (cyan), $`l`$ = 5 (green) and $`l`$ = 6 (red).
Fig. 26 — Similar to Figure 25, but for $`I/U`$ amplitude ratios.
Fig. 27 — Behaviour of the $`I/U`$ amplitude ratio with effective temperature of a model. (a) EC 14026 regime: the surface gravity was kept constant at $`\mathrm{log}g=5.75`$, while $`T_{\mathrm{eff}}`$ was varied from 29,000 K $``$ 35,000 K in increments of 1000 K. From top to bottom, the curves refer to modes with degree indices $`l`$ = 5, 6, 4, 2, 1, 0 and 3. (b) PG 1716 regime: we adopted the representative value of $`\mathrm{log}g=5.40`$ and varied the effective temperaure from 22,000 K $``$ 28,000 K, again in steps of 1000 K. From top to bottom, the modes in question correspond to $`l`$ = 5, 4, 1, 6, 2 and 3.
Fig. 28 — Behaviour of the $`I/U`$ amplitude ratio with surface gravity. (a) EC 14026 regime: the effective temperature was kept constant at $`T_{\mathrm{eff}}=33,000`$ K, while $`\mathrm{log}g`$ was varied from 5.5 $``$ 5.9 in increments of 0.1 dex. From top to bottom on the right hand side, the curves refer to modes with degree indices $`l`$ = 5, 6, 4, 2, 1, 0 and 3. (b) PG 1716 regime: we adopted the representative value of $`T_{\mathrm{eff}}=27,000`$ K and varied the surface gravity from 5.1 to 5.5, again in steps of 0.1 dex. From top to bottom on the right hand side, the modes in question correspond to $`l`$ = 5, 4, 1, 6, 2 and 3.
Fig. 29 — Fit to the u’, g’ and r’ pulsational amplitudes observed for the 182.4 s mode of KPD 2109+4401 by Jeffery et al. (2004). The predicted amplitude-wavelength behaviours of modes with $`l`$=0 to $`l`$=5 have been fit to the observed values using a least-squares procedure. Only the $`l`$=0 curve provides an acceptable fit. |
warning/0507/hep-ph0507287.html | ar5iv | text | # First Order QED Corrections to the Parity-Violating Asymmetry in Møller Scattering
## I Introduction
Measurements of the parity-violating observables in electron scattering provide information about low-energy structure of weak neutral current processes. Such observables arise from the interference between the weak and electromagnetic amplitudes ref:Zeldovich , and are sensitive to the electroweak couplings. In the Standard Model, the couplings of the $`Z^0`$ boson to the fermions are determined by the weak mixing angle $`\theta _\mathrm{W}`$, which has been measured with high precision at the $`Z^0`$ resonance ref:PDG2004 . Precision measurements of the low-energy parity-violating observables provide an independent determination of the weak mixing angle, directly test higher order electroweak corrections, and provide strong independent constraints on the new physics contributions at the TeV scales ref:MarcianoRosner .
The experiment E158 at the Stanford Linear Accelerator Center (SLAC) has measured the parity-violating left-right asymmetry $`A_{LR}`$ in Møller scattering of polarized 50 GeV electrons off unpolarized atomic electrons in a liquid hydrogen target ref:E158 . The final uncertainty on $`A_{LR}`$ is about 10%. The measurement of $`A_{LR}`$ translates into the measurement of $`\mathrm{sin}^2\theta _W`$ with a precision of $`\sigma (\mathrm{sin}^2\theta _W)0.001`$ at low momentum transfer $`Q^20.03\mathrm{GeV}^2/c^2`$, and is sensitive to both electroweak one-loop radiative corrections and new physics phenomena at the TeV scales.
A precise comparison of the experimental results with the Standard Model predictions requires detailed understanding of the radiative corrections, including effects of the QED bremsstrahlung. The leading-order electroweak corrections to the Møller scattering have been computed a number of years ago cz-marc ; denn . The authors of Ref. cz-marc factorized out the soft bremsstrahlung contribution, but did not include the effects of hard bremsstrahlung, arguing that their effects are small, and would require the knowledge of experimental kinematics and acceptance. In this paper, improving on the calculation of Ref. zyk , we present a complete calculation of the first order QED corrections. The detailed analysis shows that the QED corrections are indeed small but not insignificant, compared to the systematic uncertainties of E158 and the projected uncertainties of the future parity violation experiments.
Experimentally, Møller scattering is often used to measure polarization of electron beams ref:polarimetry . In such measurements, both beam and target electrons are polarized, and electroweak effects can typically be neglected. QED corrections to the parity-conserving polarized Møller scattering have been computed in Ref. suarez ; IlyichevZykunov and are relatively important, compared to the typical precision of Møller polarimeters. Similar to Ref. suarez ; IlyichevZykunov , we perform our calculations in the covariant framework of Bardin and Shumeiko covar , which allows us to cancel out infrared divergences without introducing unphysical parameters (such as a frame-dependent cutoff $`\mathrm{\Delta }E`$ that separates the soft bremsstrahlung region from the hard bremsstrahlung contributions).
This paper is organized as follows. Section II introduces the kinematics of the Møller scattering and Born cross section and parity-violating asymmetry. Section III explains the regularization of infrared divergences. Section VI presents numerical results applied to the kinematics of SLAC E158 experiment.
## II Definitions, Born Cross Section, and Kinematics
The lowest-order (Born) cross section for Møller scattering, defined by the Feynman diagrams in Fig. 1, can be written as
$$\sigma ^0=\frac{2\pi \alpha ^2}{s}\underset{i,j=\gamma ,Z}{}[\lambda _{}^{ij}(u^2D^i(t)D^j(t)+t^2D^i(u)D^j(u))+\lambda _+^{ij}s^2(D^i(t)+D^i(t))(D^j(t)+D^j(u))],$$
(1)
where the four-momenta of the initial and final electrons $`k_1,p_1`$ and $`k_2,p_2`$ (see Fig.1) are combined to form Mandelstam invariants
$$s=(k_1+p_1)^2,t=(k_1k_2)^2,u=(k_2p_1)^2.$$
(2)
In Eq. (1) and hereafter, we use a short-hand notation $`\sigma d\sigma /dy`$.
The Born cross section in Eq. (1) is written in terms of the photon and $`Z^0`$ propagators
$$D^i(k)D^{ik}=\frac{1}{km_i^2}(i=\gamma ,Z)$$
(3)
and the coupling factors
$$\lambda _\pm ^{ij}=\lambda _{1}^{}{}_{B}{}^{ij}\lambda _{1}^{}{}_{T}{}^{ij}\pm \lambda _{2}^{}{}_{B}{}^{ij}\lambda _{2}^{}{}_{T}{}^{ij}.$$
(4)
The latter in turn depend on the polarizations of the beam ($`p_B`$) and target ($`p_T`$) electrons:
$$\lambda _{1}^{}{}_{B(T)}{}^{ij}=\lambda _V^{ij}p_{B(T)}\lambda _A^{ij},\lambda _{2}^{}{}_{B(T)}{}^{ij}=\lambda _A^{ij}p_{B(T)}\lambda _V^{ij},$$
(5)
$$\lambda _V^{ij}=v^iv^j+a^ia^j,\lambda _A^{ij}=v^ia^j+a^iv^j,$$
(6)
where $`v^i,a^i`$ are the vector and axial-vector coupling constants for the photon and $`Z^0`$:
$$v^\gamma =1,a^\gamma =0,$$
(7)
$`v^Z`$ $`=`$ $`(1+4s_W^2)/(4s_Wc_W)`$
$`a^Z`$ $`=`$ $`1/(4s_Wc_W)`$ (8)
and $`s_W`$ ($`c_W`$) is sine (cosine) of the Weinberg angle.
It is convenient to rewrite the cross section in terms of four Born-level matrix elements $`M_l^0`$:
$$\sigma ^0=\frac{\pi \alpha ^2}{s}\mathrm{Re}\underset{l=1}{\overset{4}{}}(M_l^0+\widehat{M}_l^0),$$
(9)
where matrix elements $`\widehat{M}_l^0`$ are obtained from $`M_l^0`$ by crossing symmetry $`tu`$. The matrix elements $`M_l^0`$ are expressed through the “even” and “odd” functions $`M_e`$ and $`M_o`$:
$`M_1^0`$ $`=`$ $`D^{\gamma t}(D^{\gamma t}M_e^{\gamma \gamma \gamma \gamma }D^{\gamma u}M_o^{\gamma \gamma \gamma \gamma }),`$
$`M_2^0`$ $`=`$ $`D^{\gamma t}(D^{Zt}M_e^{\gamma Z\gamma Z}D^{Zu}M_o^{\gamma Z\gamma Z}),`$
$`M_3^0`$ $`=`$ $`D^{Zt}(D^{\gamma t}M_e^{Z\gamma Z\gamma }D^{\gamma u}M_o^{Z\gamma Z\gamma }),`$
$`M_4^0`$ $`=`$ $`D^{Zt}(D^{Zt}M_e^{ZZZZ}D^{Zu}M_o^{ZZZZ}).`$ (10)
The matrix elements $`M_e`$ and $`M_o`$ are defined in such a way that they can be used, with minimal modifications, in both Born and first-order matrix elements. They are defined in terms of the couplings $`\lambda `$
$`M_e^{ijkl}`$ $`=`$ $`2(s^2+u^2)\lambda _{1}^{}{}_{B}{}^{ij}\lambda _{1}^{}{}_{T}{}^{kl}+2(s^2u^2)\lambda _{2}^{}{}_{B}{}^{ij}\lambda _{2}^{}{}_{T}{}^{kl},`$ (11)
$`M_o^{ijkl}`$ $`=`$ $`2s^2(\lambda _{1}^{}{}_{B}{}^{ij}\lambda _{1}^{}{}_{T}{}^{kl}+\lambda _{2}^{}{}_{B}{}^{ij}\lambda _{2}^{}{}_{T}{}^{kl}).`$ (12)
The kinematic variable $`y`$ is defined as
$$y=\frac{t}{s}\frac{1\mathrm{cos}\mathrm{\Theta }}{2}\frac{E^{^{}}}{E^{}},$$
(13)
where $`\mathrm{\Theta }`$ is the center of mass (CMS) scattering angle of the detected electron with momentum $`k_2`$. $`E^{}(E^{^{}})`$ is the energy of the initial (detected) electron in CMS, respectively. In the Born approximation, the Møller scattering is elastic and $`E^{}=E^{^{}}`$, therefore
$$y_{\mathrm{Born}}=\frac{1\mathrm{cos}\mathrm{\Theta }}{2}=1\frac{E^{^{}}}{E}$$
(14)
where $`E`$ and $`E^{^{}}`$ are the initial and scattered electron energies in the Lab frame, respectively. Whenever possible, we ignore the electron mass $`m`$ (which cannot be done in the collinear singularity regions).
At the Born level, the unpolarized (averaged over helicity states) cross section is given analytically by Moller
$$\sigma ^0=\frac{2\pi \alpha ^2}{sy^2(1y)^2}\left(1+y^4+(1y)^4\right).$$
(15)
Polarization asymmetry $`A_{LR}`$ is conventionally defined as
$$A_{LR}\frac{\sigma _{LL}+\sigma _{LR}\sigma _{RL}\sigma _{RR}}{\sigma _{LL}+\sigma _{LR}+\sigma _{RL}+\sigma _{RR}},$$
(16)
where the first helicity index refers to the beam electrons, and the second helicity index corresponds to the target electrons. Since the target helicity is summed over, $`A_{LR}`$ defined by Eq. (16) is a parity-violating observable<sup>1</sup><sup>1</sup>1Notice that $`A_{LR}`$ has the opposite sign compared to the asymmetry $`A_{\mathrm{PV}}`$ defined in many low-energy experiments, e.g. Ref. ref:E158 . In our definition, $`A_{LR}`$ in Møller scattering is positive.. The Born-level asymmetry is given by cz-marc
$`A_{LR}^0`$ $`=`$ $`𝒜^0(Q^2,y)\left(14\mathrm{sin}^2\theta _W\right)={\displaystyle \frac{G_Fs}{\sqrt{2}\pi \alpha }}{\displaystyle \frac{y(1y)}{1+y^4+(1y)^4}}\left(14\mathrm{sin}^2\theta _W\right),`$ (17)
where $`𝒜^0(Q^2,y)`$ is an experimental acceptance-dependent analyzing power. Kinematics of the E158 experiment correspond to the laboratory beam energies of $`E=45(48)`$ GeV and CMS scattering angles $`0.5<\mathrm{cos}\mathrm{\Theta }0`$, or average invariants $`s=2mE0.048\mathrm{GeV}^2/c^2`$, $`Q^2=t=0.026\mathrm{GeV}^2/c^2`$ and $`y0.6`$ ref:E158 .
We perform our calculation in the on-shell (OS) renormalization scheme, defining the weak mixing angle to all orders in perturbation theory as $`c_Wm_W/m_Z`$, where $`m_W`$ and $`m_Z`$ are the physical masses of the $`W^\pm `$ and $`Z^0`$ bosons, respectively. For consistency with the precision electroweak measurements ref:PDG2004 ; ref:EWWG , we use
$`m_W`$ $`=`$ $`80.390\mathrm{GeV}`$ (18)
$`m_Z`$ $`=`$ $`91.188\mathrm{GeV}`$ (19)
which implies
$$\mathrm{sin}^2\theta _W=0.2228$$
(20)
in the on-shell scheme. As we will note in Section VI, while the absolute value of the Born asymmetry depends on the electron neutral current coupling $`14\mathrm{sin}^2\theta _W`$, the relative corrections to the experimentally measured asymmetries computed here are quite insensitive to the choice of couplings or the renormalization scheme.
## III Radiative Corrections
It is well known that effects of “internal” bremsstrahlung (real photon emission) need to be combined with the contributions from the other one-loop (leading order) electroweak radiative corrections (so-called virtual, or V-contributions). The virtual contributions to Møller scattering have been studied extensively cz-marc ; denn and are not repeated here. However, we need the infrared-divergent part of the V-contributions, which cancels out the corresponding divergences in the bremsstrahlung cross section. Naturally, the extraction of the IR-divergent parts is a somewhat subjective procedure, which may lead to ambiguities similar to the concept of scheme dependence in the ultraviolet renormalization. We follow the framework of Bardin and Shumeiko covar and extract the infrared-divergent contributions that are strictly proportional to the Born cross section. Such contributions cancel in the parity-violating asymmetry, and remain small even after other corrections are taken into account. We take the infrared-divergent parts of the virtual corrections from BSH86 ; Hol90 , which include the vacuum polarization and vertex correction contributions. We also compute the IR-divergent $`\gamma \gamma `$ and $`\gamma Z`$ box diagrams.
The virtual contributions to the Møller scattering cross section and asymmetry can be classified into three categories: the vacuum polarizations of the gauge boson propagators, vertex corrections, and box diagrams (see Fig. 2):
$$\sigma ^V=\sigma ^S+\sigma ^{Ver}+\sigma ^B.$$
(21)
The vacuum polarizations of $`\gamma `$ and $`Z^0`$ bosons are infrared-convergent, and are not considered in this paper. The infrared-divergent parts of the vertex corrections (Figures 2b and 2c) are obtained from form-factors $`\delta F_{V,A}^{je}`$ given in BSH86 (for $`k^2=t,u`$). Substituting the coupling constants for the vertex form-factors (e.g. $`v^\gamma \delta F_V^{\gamma e}`$) in the expressions for the Born functions $`M_{e,o}`$, we get the vertex part of the cross section
$$\sigma ^{Ver}=\frac{2\pi \alpha ^2}{s}\mathrm{Re}\underset{l=1}{\overset{4}{}}(M_l^V+\widehat{M}_l^V),$$
(22)
where
$`M_1^V`$ $`=`$ $`D^{\gamma t}(D^{\gamma t}(M_e^{F^\gamma \gamma \gamma \gamma }+M_e^{\gamma \gamma F^\gamma \gamma })D^{\gamma u}(M_o^{F^\gamma \gamma \gamma \gamma }+M_o^{\gamma \gamma F^\gamma \gamma })),`$
$`M_2^V`$ $`=`$ $`D^{\gamma t}(D^{Zt}(M_e^{F^\gamma Z\gamma Z}+M_e^{\gamma ZF^\gamma Z})D^{Zu}(M_o^{F^\gamma Z\gamma Z}+M_o^{\gamma ZF^\gamma Z})),`$
$`M_3^V`$ $`=`$ $`D^{Zt}(D^{\gamma t}(M_e^{F^Z\gamma Z\gamma }+M_e^{Z\gamma F^Z\gamma })D^{\gamma u}(M_o^{F^Z\gamma Z\gamma }+M_o^{Z\gamma F^Z\gamma })),`$
$`M_4^V`$ $`=`$ $`D^{Zt}(D^{Zt}(M_e^{F^ZZZZ}+M_e^{ZZF^ZZ})D^{Zu}(M_o^{F^ZZZZ}+M_o^{ZZF^ZZ})).`$ (23)
The box diagrams with at least one photon (i.e. Figures 2d and 2e plus $`u`$-channel graphs) also contain infrared divergences. The diagrams with two $`Z^0`$ or two $`W`$ bosons are infrared-convergent and are not considered here. We compute the box diagram contribution as a sum of the infrared-divergent and infrared-finite parts $`\sigma ^B=\sigma _F^B+\sigma _{IR}^B`$. The IR-finite parts of the $`\gamma \gamma `$ and $`\gamma Z`$ boxes are expressed by
$$\sigma _F^B=\frac{2\alpha ^3}{s}\underset{k=\gamma ,Z}{}(B_{\gamma \gamma }^k+B_{\gamma Z}^k)+(tu).$$
(24)
The terms $`B`$ have the following form:
$`B_{(\gamma \gamma )}^k`$ $`=`$ $`D^{kt}\lambda _{}^{\gamma k}\delta _{(\gamma \gamma )}^1+(D^{kt}+D^{ku})\lambda _+^{\gamma k}\delta _{(\gamma \gamma )}^2`$
$`B_{(\gamma Z)}^k`$ $`=`$ $`D^{kt}\lambda _{}^{Zk}\delta _{(\gamma Z)}^1+(D^{kt}+D^{ku})\lambda _+^{Zk}\delta _{(\gamma Z)}^2`$ (25)
At low energies ($`s,|t|,|u|m_Z^2`$), $`\delta _{(ij)}^{1,2}`$ are given by the following expressions:
$`\delta _{(\gamma \gamma )}^1`$ $`=`$ $`L_s^2{\displaystyle \frac{s^2+u^2}{2t}}L_su(L_x^2+\pi ^2){\displaystyle \frac{u^2}{t}}`$
$`\delta _{(\gamma \gamma )}^2`$ $`=`$ $`L_s^2{\displaystyle \frac{s^2}{t}}+L_xs(L_x^2+\pi ^2){\displaystyle \frac{s^2+u^2}{2t}}`$ (26)
$`\delta _{(\gamma Z)}^1`$ $`=`$ $`8u^2(4I_{\gamma Z}\widehat{I}_{\gamma Z})`$
$`\delta _{(\gamma Z)}^2`$ $`=`$ $`8s^2(I_{\gamma Z}4\widehat{I}_{\gamma Z})`$
The logs in the electromagnetic box diagrams are
$$L_s=\mathrm{ln}\frac{s}{|t|},L_x=\mathrm{ln}\frac{u}{t}$$
(27)
and the scalar integrals in $`\gamma Z`$-parts are
$`I_{\gamma Z}`$ $`=`$ $`{\displaystyle \frac{1}{2\sqrt{u}}}{\displaystyle _0^1}z𝑑z{\displaystyle _0^1}𝑑x{\displaystyle \frac{1}{\sqrt{\beta }}}\mathrm{ln}\left|{\displaystyle \frac{xz\sqrt{u}+\sqrt{\beta }}{xz\sqrt{u}\sqrt{\beta }}}\right|`$
$`\widehat{I}_{\gamma Z}`$ $`=`$ $`I_{\gamma Z}|_{us}`$ (28)
$`\beta `$ $`=`$ $`ux^2z^2+4(1z)(tz(x1)+m_Z^2)`$
Over a fairly broad kinematic range of interest $`10^4\mathrm{GeV}^2|u|,|t|1\mathrm{GeV}^2`$, Eq. (28) can be integrated numerically and approximated by the following expression to better than 1% precision:
$$I_{\gamma Z}\frac{0.123}{m_Z^2}\left(1.64+\mathrm{ln}\frac{m_Z^2}{u}\right)+\frac{0.61t}{m_Z^4}\left(0.102+\mathrm{ln}\frac{m_Z^2}{u}\right).$$
(29)
Separating the infrared-divergent and finite virtual contributions, we can write
$$\sigma ^V=\sigma _{IR}^V+\sigma ^V(\lambda ^2s),$$
(30)
where we introduce the finite photon mass $`\lambda `$ to regulate the IR divergence. The IR-divergent part proportional to the Born cross section is
$$\sigma _{IR}^V=\frac{2\alpha }{\pi }\mathrm{ln}\frac{s}{\lambda ^2}(\mathrm{ln}\frac{tu}{m^2s}1)\sigma ^0.$$
(31)
## IV Bremsstrahlung contribution
The complete leading order radiative corrections need to include the inelastic processes that correspond to the real photon emission $`e^{}e^{}e^{}e^{}\gamma `$ ($`R`$-contributions, or real photon bremsstrahlung). The diagrams are shown in Fig. 3 (plus crossed terms for a total of 16 diagrams with $`\gamma `$ and $`Z^0`$ propagators). Let $`k`$ be the four-momentum of the emitted photon. The differential cross section is given by
$`\sigma ^R={\displaystyle \frac{\alpha ^3}{4s\pi }}{\displaystyle _0^{v^{\mathrm{max}}}}𝑑v{\displaystyle \frac{d^3k}{k_0}\delta [(k_1+p_1k_2k)^2m^2]\underset{j,i=1,4}{}M_{ij}^R(1)^{i+j}},`$ (32)
where indices $`i`$ and $`j`$ refer to a particular contribution to the cross section ($`u`$ and $`t`$ channels with $`Z^0`$ or $`\gamma `$ exchange):
$$i,j=(1;2;3;4)=(\gamma t;\gamma u;Zt;Zu).$$
(33)
We also use a somewhat unconventional set of kinematic variables
$`z`$ $`=`$ $`2kk_2`$
$`z_1`$ $`=`$ $`2kk_1`$
$`t_1`$ $`=`$ $`(p_2p_1)^2`$
$`v`$ $`=`$ $`2kp_2=s+u+t4m^2`$ (34)
$`v_1`$ $`=`$ $`2kp_1=s+u+t_14m^2`$
$`z_2`$ $`=`$ $`(k_1p_2)^2=uv+z_1,`$
which satisfy the following relations:
$$v_1v=zz_1=t_1t.$$
(35)
The integration variable $`v`$ in Eq. (32) describes the inelasticity of the reaction, the deviation from the 2-body kinematic constraint $`s+u+t=4m^2`$. The kinematically allowed region for variable $`v`$ is given by chew-low
$$vv^{\mathrm{lim}}=\frac{2(s+t4m^2)}{1+\sqrt{(14m^2/s)(1+4m^2/t)}}s+t$$
(36)
The limit $`v=v^{\mathrm{lim}}`$ corresponds to the collinear singularity $`u=0`$. However, for E158, which only detects scattered electrons with laboratory energy $`E^{}E_{\mathrm{cut}}11`$ GeV, the integration region is further restricted to $`u^{\mathrm{max}}=2m(mE_{\mathrm{cut}})`$ and
$$v^{\mathrm{max}}=s+t+u^{\mathrm{max}}4m^2.$$
(37)
The squares of the matrix elements $`M_{ij}^R`$ are given by
$$M_{ij}^R=(M_{ij}^R)_{zz}+(M_{ij}^R)_{zv}+(M_{ij}^R)_{vz}+(M_{ij}^R)_{vv}$$
(38)
for $`ij=13,31,11,33`$, and
$$M_{ij}^R=(M_{ij}^R)_f+(M_{ij}^R)_l+(M_{ij}^R)_{tu}+(M_{ij}^R)_s$$
(39)
for $`ij=12,14,32,34`$, where the traces of the appropriate $`\gamma `$-matrix combinations are multiplied by the density matrices and the corresponding propagators
$`(M_{ij}^R)_{zz}`$ $`=`$ $`\text{T}r[G_1^{\mu \alpha }\rho ^{ij}(k_1)G_{1}^{\nu \alpha }{}_{}{}^{T}\mathrm{\Lambda }(k_2)]\text{T}r[\gamma _\mu \rho ^{ij}(p_1)\gamma _\nu \mathrm{\Lambda }(p_2)]D^{it_1}D^{jt_1},`$
$`(M_{ij}^R)_{zv}`$ $`=`$ $`\text{T}r[G_1^{\mu \alpha }\rho ^{ij}(k_1)\gamma _\nu \mathrm{\Lambda }(k_2)]\text{T}r[\gamma _\mu \rho ^{ij}(p_1)G_{2}^{\nu \alpha }{}_{}{}^{T}\mathrm{\Lambda }(p_2)]D^{it_1}D^{jt},`$
$`(M_{ij}^R)_{vz}`$ $`=`$ $`\text{T}r[G_2^{\mu \alpha }\rho ^{ij}(p_1)\gamma _\nu \mathrm{\Lambda }(p_2)]\text{T}r[\gamma _\mu \rho ^{ij}(k_1)G_{1}^{\nu \alpha }{}_{}{}^{T}\mathrm{\Lambda }(k_2)]D^{it}D^{jt_1},`$
$`(M_{ij}^R)_{vv}`$ $`=`$ $`\text{T}r[G_2^{\mu \alpha }\rho ^{ij}(p_1)G_{2}^{\nu \alpha }{}_{}{}^{T}\mathrm{\Lambda }(p_2)]\text{T}r[\gamma _\mu \rho ^{ij}(k_1)\gamma _\nu \mathrm{\Lambda }(k_2)]D^{it}D^{jt},`$
$`(M_{ij}^R)_f`$ $`=`$ $`\text{T}r[G_1^{\mu \alpha }\rho ^{ij}(k_1)G_3^{\nu \alpha }\mathrm{\Lambda }(p_2)\gamma _\mu \rho ^{ij}(p_1)\gamma _\nu \mathrm{\Lambda }(k_2)]D^{it_1}D^{ju},`$
$`(M_{ij}^R)_l`$ $`=`$ $`\text{T}r[G_1^{\mu \alpha }\rho ^{ij}(k_1)\gamma _\nu \mathrm{\Lambda }(p_2)\gamma _\mu \rho ^{ij}(p_1)G_4^{\nu \alpha }\mathrm{\Lambda }(k_2)]D^{it_1}D^{jz_2},`$
$`(M_{ij}^R)_{tu}`$ $`=`$ $`\text{T}r[\gamma _\mu \rho ^{ij}(k_1)G_3^{\nu \alpha }\mathrm{\Lambda }(p_2)G_2^{\mu \alpha }\rho ^{ij}(p_1)\gamma _\nu \mathrm{\Lambda }(k_2)]D^{it}D^{ju},`$
$`(M_{ij}^R)_s`$ $`=`$ $`\text{T}r[\gamma _\mu \rho ^{ij}(k_1)\gamma _\nu \mathrm{\Lambda }(p_2)G_2^{\mu \alpha }\rho ^{ij}(p_1)G_4^{\nu \alpha }\mathrm{\Lambda }(k_2)]D^{it}D^{jz_2},`$ (40)
where
$$\mathrm{\Lambda }(p)=\widehat{p}+m,\widehat{p}=\gamma ^\mu p_\mu ,$$
(41)
$$G_1^{\mu \alpha }=\gamma ^\mu \frac{2k_1^\alpha \widehat{k}\gamma ^\alpha }{z_1}+\frac{2k_2^\alpha +\gamma ^\alpha \widehat{k}}{z}\gamma ^\mu ,$$
(42)
$$G_2^{\mu \alpha }=\gamma ^\mu \frac{2p_1^\alpha \widehat{k}\gamma ^\alpha }{v_1}+\frac{2p_2^\alpha +\gamma ^\alpha \widehat{k}}{v}\gamma ^\mu ,$$
(43)
$$G_3^{\nu \alpha }=\frac{2k_1^\alpha \gamma ^\alpha \widehat{k}}{z_1}\gamma ^\nu +\gamma ^\nu \frac{2p_2^\alpha +\widehat{k}\gamma ^\alpha }{v},$$
(44)
$$G_4^{\nu \alpha }=\frac{2p_1^\alpha \gamma ^\alpha \widehat{k}}{v_1}\gamma ^\nu +\gamma ^\nu \frac{2k_2^\alpha +\widehat{k}\gamma ^\alpha }{z}.$$
(45)
Expressions for $`M_{ij}^R`$ for other values of $`i`$ and $`j`$ (e.g. $`\{ij\}=22,44,24,42`$, etc.) can be obtained from the symmetry of expressions in Eq. (40):
$`(M_{24}^R)_{zz}`$ $`=`$ $`(M_{13}^R)_{zz}|_{k_2p_2}=(M_{13}^R)_{zz}|_{k_2p_2,k_1p_1}`$ (46)
$`=`$ $`(M_{13}^R)_{vv}|_{k_1p_1}=(M_{13}^R)_{vv}|_{tu},`$
and
$`(M_{24}^R)_{vv}`$ $`=`$ $`(M_{13}^R)_{zz}|_{tu}.`$ (47)
Cases $`\{ij\}=22,44`$ are analyzed in a similar manner. The symmetry noted above is less apparent in the interference terms (indices $`zv`$ and $`vz`$):
$`(M_{24}^R)_{zv}+(M_{42}^R)_{vz}`$ $`=`$ $`\left[(M_{13}^R)_{vz}+(M_{31}^R)_{zv}\right]|_{tu}`$ (48)
and likewise for other parts in Eq. (32).
### IV.1 Infrared Divergences in Bremsstrahlung Contributions
Next we need to address the issue of the infrared divergences in the bremsstrahlung cross section. According to the prescription of Bardin and Shumeiko covar , we find the infrared-divergent parts in the squares of the matrix elements that are proportional to the corresponding Born contributions:
$`(M_{ij}^R)_{zz}^{IR}`$ $`=`$ $`4({\displaystyle \frac{m^2}{z^2}}+{\displaystyle \frac{m^2}{z_1^2}}+{\displaystyle \frac{t}{zz_1}})M_{ij},`$ (49)
$`(M_{ij}^R)_{zv}^{IR}+(M_{ij}^R)_{vz}^{IR}`$ $`=`$ $`4({\displaystyle \frac{u}{zv_1}}+{\displaystyle \frac{s}{zv}}+{\displaystyle \frac{s}{z_1v_1}}+{\displaystyle \frac{u}{z_1v}})M_{ij},`$
$`(M_{ij}^R)_{vv}^{IR}`$ $`=`$ $`4({\displaystyle \frac{m^2}{v_1^2}}+{\displaystyle \frac{m^2}{v^2}}+{\displaystyle \frac{t}{v_1v}})M_{ij},`$
$`(M_{ij}^R)_f^{IR}`$ $`=`$ $`({\displaystyle \frac{4m^2}{z_1^2}}+{\displaystyle \frac{2u}{z_1v}}+{\displaystyle \frac{2t}{zz_1}}+{\displaystyle \frac{2s}{zv}})M_{ij},`$
$`(M_{ij}^R)_l^{IR}`$ $`=`$ $`({\displaystyle \frac{2s}{z_1v_1}}+{\displaystyle \frac{2t}{z_1z}}+{\displaystyle \frac{2u}{zv_1}}+{\displaystyle \frac{4m^2}{z^2}})M_{ij},`$
$`(M_{ij}^R)_{tu}^{IR}`$ $`=`$ $`({\displaystyle \frac{2s}{z_1v_1}}+{\displaystyle \frac{2u}{z_1v}}+{\displaystyle \frac{2t}{vv_1}}+{\displaystyle \frac{4m^2}{v^2}})M_{ij},`$
$`(M_{ij}^R)_s^{IR}`$ $`=`$ $`({\displaystyle \frac{4m^2}{v_1^2}}+{\displaystyle \frac{2u}{v_1z}}+{\displaystyle \frac{2t}{vv_1}}+{\displaystyle \frac{2s}{vz}})M_{ij},`$
where
$`M_{11}`$ $`=`$ $`D^{\gamma t}D^{\gamma t}M_e^{\gamma \gamma \gamma \gamma }`$
$`M_{13}`$ $`=`$ $`D^{\gamma t}D^{Zt}M_e^{\gamma Z\gamma Z}`$
$`M_{31}`$ $`=`$ $`D^{\gamma t}D^{Zt}M_e^{Z\gamma Z\gamma }`$
$`M_{33}`$ $`=`$ $`D^{Zt}D^{Zt}M_e^{ZZZZ}`$
$`M_{12}`$ $`=`$ $`D^{\gamma t}D^{\gamma u}M_o^{\gamma \gamma \gamma \gamma }`$
$`M_{14}`$ $`=`$ $`D^{\gamma t}D^{Zu}M_o^{\gamma Z\gamma Z}`$
$`M_{32}`$ $`=`$ $`D^{\gamma u}D^{Zt}M_o^{Z\gamma Z\gamma }`$
$`M_{34}`$ $`=`$ $`D^{Zt}D^{Zu}M_o^{ZZZZ}.`$ (50)
The complete $`R`$-contribution to the cross section is infrared-divergent, but can be separated into an infrared-infinite part $`\sigma _{IR}^R`$ and the finite part $`\sigma _F^R`$:
$$\sigma ^R=\sigma _{IR}^R+\sigma _F^R.$$
(51)
The IR-divergent part of the bremsstrahlung matrix elements, proportional to the Born contributions, can be constructed from Eq. (49) according to Eq. (32). The finite contribution to the cross section is then obtained by subtraction
$$(M_{ij}^R)^F=M_{ij}^R(M_{ij}^R)^{IR},$$
(52)
The infrared-divergent part of Eq. (51), integrated over variables $`k`$ and $`v`$ is given in terms of a finite photon mass $`\lambda `$ by my3
$$\sigma _{IR}^R=\frac{2\alpha }{\pi }\mathrm{ln}\frac{(v^{\mathrm{max}})^2}{s\lambda ^2}(\mathrm{ln}\frac{tu}{m^2s}1)\sigma ^0.$$
(53)
The integration over the phase space of the bremsstrahlung photon is performed analytically, and integration over variable $`v`$ is done numerically due to complexity of the integral expressions. The photon phase space integral can be written as bu-ka
$`I[A]`$ $`=`$ $`{\displaystyle \frac{1}{\pi }}{\displaystyle \frac{d^3k}{k_0}\delta [(k_1+p_1k_2k)^2m^2][A]}={\displaystyle \frac{1}{\pi }}{\displaystyle \underset{t_1^{\mathrm{min}}}{\overset{t_1^{\mathrm{max}}}{}}}𝑑t_1{\displaystyle \underset{z^{\mathrm{min}}}{\overset{z^{\mathrm{max}}}{}}}{\displaystyle \frac{dz}{\sqrt{R_z}}}[A],`$ (54)
where $`R_z`$ is the Gramm determinant (modulo $`1`$), and can be parameterized as a second-order polynomial in $`z`$ as
$$R_z=A_zz^2+2B_zz+C_z.$$
(55)
Coefficients $`A_z`$, $`B_z`$, and $`C_z`$ are given by the following expressions:
$`A_z`$ $`=`$ $`4m^2t(vt)^2,`$
$`B_z`$ $`=`$ $`Et_1+F`$
$`E`$ $`=`$ $`v(u2m^2)st,F=t(2m^2v+stsv),`$
$`C_z`$ $`=`$ $`(At_1^2+2Bt_1+C),`$ (56)
$`A`$ $`=`$ $`(sv)^24m^2s,`$
$`B`$ $`=`$ $`st(sv4m^2)2m^2v^2,`$
$`C`$ $`=`$ $`st^2(s4m^2).`$
Integration limits $`z^{\mathrm{min}/\mathrm{max}}`$ and $`t_1^{\mathrm{min}/\mathrm{max}}`$ are solutions of the equations $`R_z=0`$ and $`z^{\mathrm{min}}=z^{\mathrm{max}}`$:
$`z^{\mathrm{min}/\mathrm{max}}`$ $`=`$ $`{\displaystyle \frac{B_z\pm \sqrt{B_z^2A_zC_z}}{A_z}}`$ (57)
$`t_1^{\mathrm{min}/\mathrm{max}}`$ $`=`$ $`{\displaystyle \frac{v(tv)+2m^2tv\sqrt{A_z}}{2(v+m^2)}}`$ (58)
This set of variables makes integration more convenient.
Expressions for $`(M_{ij}^R)^F`$ are computed using symbolic manipulation program REDUCE reduce but are excessively complex to be listed here. They are available as subroutines in the Fortran program rcAPV<sup>2</sup><sup>2</sup>2Fortran program rcAPV is available from the authors upon request.. The relevant integrals, listed in the Appendix of Ref. zyk , were computed both analytically and numerically.
## V Cancellation of Infrared Singularities
Adding infrared-divergent parts of $`V`$\- and $`R`$-contributions given in Eq. (31) and Eq. (53)) together with the IR-finite pieces, we obtain the finite expression for the radiatively-corrected cross section, free of non-physical parameters:
$`\sigma `$ $`=`$ $`\sigma ^0+\sigma ^C`$
$`\sigma ^C`$ $`=`$ $`\sigma _{IR}^V+\sigma _{IR}^R+\sigma _F^R+\sigma _F^{Ver}+\sigma _F^B`$
$`=`$ $`{\displaystyle \frac{\alpha }{\pi }}(4\mathrm{ln}{\displaystyle \frac{v^{max}}{m\sqrt{s}}}(\mathrm{ln}{\displaystyle \frac{tu}{m^2s}}1)+\delta _1^S+\delta _1^H)\sigma ^0+\sigma _F^{Ver}+\sigma _F^B+\sigma _F^R,`$
where
$`\delta _1^S`$ $`=`$ $`\mathrm{ln}{\displaystyle \frac{s(s+t)}{m^4}}{\displaystyle \frac{1}{2}}l_m\mathrm{ln}{\displaystyle \frac{s^2(s+t)^2}{tm^6}}{\displaystyle \frac{1}{2}}l_r^22l_rl_m+l_ml_m^2{\displaystyle \frac{\pi ^3}{3}}+1,`$
$`\delta _1^H`$ $`=`$ $`{\displaystyle \underset{0}{\overset{v^{max}}{}}}dv({\displaystyle \frac{2}{v}}\mathrm{ln}(1{\displaystyle \frac{v}{s}})+{\displaystyle \frac{2}{v}}\mathrm{ln}(1{\displaystyle \frac{v}{t}})+{\displaystyle \frac{2}{v}}\mathrm{ln}(1{\displaystyle \frac{v}{s+t}}){\displaystyle \frac{1}{v}}\mathrm{ln}(1+{\displaystyle \frac{v}{m^2}})`$ (60)
$`+{\displaystyle \frac{2}{s+tv}}\mathrm{ln}{\displaystyle \frac{s+tv}{m^2}}{\displaystyle \frac{1}{sv}}\mathrm{ln}{\displaystyle \frac{(sv)^2}{m^2\tau }}{\displaystyle \frac{1}{vt}}\mathrm{ln}{\displaystyle \frac{(vt)^2}{m^2\tau }}{\displaystyle \frac{1}{\tau }}),`$
$`l_m=\mathrm{ln}{\displaystyle \frac{t}{m^2}},l_r=\mathrm{ln}{\displaystyle \frac{s+t}{s}},\tau v+m^2,`$
## VI Results and Discussion
### VI.1 Numerical Results
In the following, we evaluate the effect of the bremsstrahlung radiative corrections on the parity-violating asymmetry $`A_{LR}`$ in the scattering of the longitudinally polarized electrons off unpolarized target electrons. We consider the kinematic conditions that correspond to the experimental setup of the SLAC E158 experiment E158 , i.e. beam energies of $`45`$ and $`48`$ GeV. E158 setup is such that the radiated photon is not detected. Moreover, the scattered electrons are only detected if their energy is above the threshold $`E^{}11`$ GeV; this restriction limits the range of integration over variable $`v`$ as given in Eq. (37).
The relative corrections to the cross section and asymmetry can be defined as
$$\delta \sigma =\frac{\sigma \sigma ^0}{\sigma ^0},\delta A=\frac{A_{LR}A_{LR}^0}{A_{LR}^0},$$
(61)
where $`A_0`$ is the Born asymmetry, and $`A^{RC}`$ is the radiatively corrected asymmetry.
The Born cross section $`\sigma ^0`$, the radiatively corrected cross sections $`\sigma ^{\mathrm{brem}}`$ (which includes only soft and hard bremsstrahlung corrections and results of IR cancellation) and $`\sigma ^{QED}`$ (full QED corrections), as well as the asymmetry $`A_{LR}`$ are shown as a function of variable $`y`$ in Fig. 4 for beam energy of $`E=45`$ GeV. Fig. 5 shows the double-differential cross section $`d^2\sigma /dE^{}/dy`$ and the asymmetry $`A_{LR}`$ as a function of the scattered electron energy in the lab frame $`E^{}`$ for a fixed $`y=0.5`$. This double-differential cross section is used to properly average the radiative corrections over the experimental acceptance. The corrections to cross section and asymmetry are shown in Fig. 6. The numerical precision of the corrections is about $`0.1\%`$.
We find that at fixed value of $`y=0.5`$ and $`E=45`$ GeV ($`Q^2=0.023\mathrm{GeV}^2`$), hard bremsstrahlung reduces the value of the parity-violating asymmetry by $`1\%`$. Contribution from the $`\gamma \gamma `$ and $`\gamma Z`$ box diagrams is also negative and reduces the asymmetry by $`5\%`$, so that the total QED correction at $`y=0.5`$ and $`E=45`$ GeV is $`6.2\%`$.
At fixed scattering angles, radiative effects move events towards lower values of $`y`$. Therefore, even though at a fixed value of $`y=0.5`$ the hard bremsstrahlung correction is negative, the net change of the asymmetry, integrated over E158 acceptance $`0.5\mathrm{cos}\mathrm{\Theta }0`$ at $`E=45`$ GeV is $`\delta A^{\mathrm{brem}}+1\%`$. The full QED correction, including $`\gamma \gamma `$ and $`\gamma Z`$ boxes, is $`\delta A^{QED}=4.5\%`$.
### VI.2 Factorization of QED Radiative Corrections and NLO Uncertainties
At leading order, corrections to the parity-violating asymmetry from the diagrams involving photons (i.e. soft and hard bremsstrahlung, $`\gamma \gamma `$ and $`\gamma Z`$ boxes, and photonic vertex diagrams) are proportional to the product $`v^Za^Z`$. In the OS scheme, these contributions are strictly proportional to the Born asymmetry. In other words, in the OS scheme, the relative corrections to the asymmetry due to QED diagrams are independent of the weak mixing angle, and can be factorized out. We can write
$$A_{LR}^{QED}=𝒜^0(Q^2,y)(1+\delta A^{QED})\left(14\mathrm{sin}^2\theta _W\right)$$
(62)
where $`𝒜^0(Q^2,y)`$ is the Born analyzing power defined in Eq. (17) and $`\delta A^{QED}`$ is the relative radiative correction. For $`E=45`$ GeV and $`y=0.5`$, full QED correction is $`\delta A^{QED}=0.061`$, and the average over E158 kinematics (beam energies of 45 and 48 GeV and $`0.5\mathrm{cos}\mathrm{\Theta }0`$) is $`\delta A^{QED}_{\mathrm{E158}}=0.043`$.
Virtual corrections, such as vacuum polarization and box and vertex diagrams with heavy gauge bosons, are not in general proportional to $`v^Za^Z`$, and spoil the simple factorization of Eq. (62). Nevertheless, one can relate the leading-order asymmetry to the Born-level formula through the set of multiplicative and additive corrections:
$$A_{LR}(Q^2,y)=𝒜^0(Q^2,y)\rho (Q^2)(1+\delta A(Q^2,y))\left(14\mathrm{sin}^2\theta _W^{\mathrm{eff}}(Q^2)+\mathrm{\Delta }(Q^2)\right).$$
(63)
Here the effective mixing angle relevant for a scattering process at momentum transfer $`Q`$ is defined through the form-factor $`\kappa (Q^2)`$ (in OS scheme) or $`\widehat{\kappa }(Q^2,\mu ^2)`$ (in $`\overline{\text{MS}}`$ scheme):
$$\mathrm{sin}^2\theta _W^{\mathrm{eff}}(Q^2)\kappa (Q^2)\mathrm{sin}^2\theta _W^{OS}=\widehat{\kappa }(Q^2,\mu ^2)\mathrm{sin}^2\widehat{\theta }_W(\mu ^2).$$
(64)
In $`\overline{\text{MS}}`$ scheme, one typically chooses $`\mu =m_Z`$.
All corrections in Eq. (63) are of order $`𝒪(\alpha )`$ but have a different physical meaning. The form-factor $`\kappa =1+𝒪(\alpha )`$ defines the momentum dependence (running) of the effective weak mixing angle. We would like to define it in a process-independent way, such that various experimental measurements could be directly compared in terms of $`\mathrm{sin}^2\theta _W^{\mathrm{eff}}`$. Typically, $`\kappa `$ includes contributions from $`\gamma Z`$ mixing and anapole moment diagrams (Fig. 2a-c) cz-marc , but may include other terms sirlin ; erler-musolf . In definition of Ref. cz-marc , carried out in $`\overline{\text{MS}}`$ scheme with $`\mathrm{sin}^2\widehat{\theta }_W(m_Z^2)\widehat{s}^2=0.23120\pm 0.00015`$ ref:PDG2004 , $`\widehat{\kappa }(Q^2=0,\mu ^2=m_Z^2)=1.0298\pm 0.0026`$
The form-factor $`\rho (Q^2)=1+𝒪(\alpha )`$ is a low-energy ratio of the neutral weak coupling to the charge coupling. It depends on the choice of the Fermi constant $`G_F`$ in Eq. (17); for $`G_F`$ derived from the muon-decay constant $`G_\mu `$, this correction is cz-marc $`\rho =1.0012\pm 0.0005`$. $`\rho `$ contains logarithmic dependence on the Higgs mass (we use $`m_H=113_{40}^{+56}`$ GeV ref:PDG2004 ) and linear dependence on the top quark mass ($`m_t=177\pm 4`$ GeV ref:PDG2004 ).
The remaining first-order corrections are included in Eq. (63) as factors $`\delta A(Q^2,y)`$ and $`\mathrm{\Delta }(Q^2)`$. $`\mathrm{\Delta }(Q^2)`$ typically includes box diagrams with two heavy bosons, and $`\delta A(Q^2,y)`$ contains the kinematics-dependent factorizable QED corrections computed in Section VI.
The only remaining question is evaluating the next-to-leading order correction uncertainties. Normally, NLO corrections would be of order $`𝒪(\alpha )`$ of the LO terms. However, we have to pay special attention to the logarithmically-enhanced contributions in the $`\gamma Z`$ box diagrams, e.g. terms proportional to $`\mathrm{ln}(m_Z^2/s)`$ in Eq. (29).
The effective $`Z^0`$-electron coupling, $`Q_w=\rho (14\mathrm{sin}^2\theta _W)`$ changes by about $`40\%`$ between zero momentum transfer and $`Z^0`$ pole cz-marc . Since the box diagrams in Fig. 2 involve integration over internal momenta of the photon and $`Z^0`$ propagators, a complete calculation of the box diagrams has to take into account the momentum dependence of $`Q_w`$ and $`\alpha `$. Strictly speaking, the momentum dependence of the weak or electromagnetic charges in the box integrals is a next-to-leading order effect, and is beyond the scope of this work. However, a judicious choice of the coupling at leading order could reduce the NLO corrections.
Czarnecki and Marciano cz-marc have argued that the NLO errors are reduced if the $`\gamma Z`$ box diagrams are evaluated in $`\overline{\text{MS}}`$ scheme with the average value of the weak mixing angle
$$\mathrm{sin}^2\theta _W\frac{\mathrm{sin}^2\theta _W(0)+\mathrm{sin}^2\theta _W(m_Z^2)}{2}$$
(65)
since the relevant integrals over the internal momentum $`k`$ are dominated by the poles near $`k^20`$ and $`k^2m_Z^2`$. Similar arguments apply to the value of the fine structure constant in the $`\gamma Z`$ box diagrams. Thus, following the spirit of this argument, we move the leading logarithmic contribution to the $`\gamma Z`$ box diagrams cz-marc
$$\mathrm{\Delta }_{\gamma Z}(Q^2)=\frac{22}{3}\frac{\alpha (Q^2)}{4\pi }\left(14\mathrm{sin}^2\theta _W(Q^2)\right)\mathrm{ln}\frac{m_Z^2}{s}$$
(66)
from the multiplicative correction $`\delta A(Q^2,y)`$ to the additive correction $`\mathrm{\Delta }(Q^2)`$. Moreover, we use the average value $`\mathrm{\Delta }_{\gamma Z}(\mathrm{\Delta }_{\gamma Z}(0)+\mathrm{\Delta }_{\gamma Z}(m_Z))/2`$ in the expression for $`\mathrm{\Delta }(Q^2)`$, and treat the spread $`|\mathrm{\Delta }_{\gamma Z}(0)\mathrm{\Delta }_{\gamma Z}(m_Z)|/2`$ as an estimate of the NLO uncertainties. This choice keeps all of the experimental acceptance dependence in the multiplicative correction $`1+\delta A(Q^2,y)`$, properly propagates the bremsstrahlung corrections, and reduces the overall size of the theoretical uncertainties. For experimental kinematics of the E158 experiment, we find
$$\delta A(Q^2,y)=0.006\pm 0.005,$$
(67)
where the uncertainty is dominated by the possible variations of the experimental acceptance and numerical precision. Employing the calculation of Ref. cz-marc for the terms $`\widehat{\kappa }`$ and $`\rho `$, the corresponding contributions from the $`WW`$ and $`ZZ`$ box diagrams, and our definition of $`\mathrm{\Delta }_{\gamma Z}`$ term, the residual additive correction in Eq. (63) is
$`\mathrm{\Delta }(Q^2)`$ $`=`$ $`\mathrm{\Delta }_{\gamma Z}+{\displaystyle \frac{\alpha (m_Z^2)}{4\pi \widehat{s}^2}}{\displaystyle \frac{3\alpha (m_Z^2)}{32\pi \widehat{s}^2(1\widehat{s}^2)}}(14\widehat{s}^2)\left[1+(14\widehat{s}^2)\right]=0.0007\pm 0.0009`$ (68)
where the uncertainty is dominated by our conservative estimate of the NLO terms.
The fact that both corrections $`\delta A(Q^2,y)`$ and $`\mathrm{\Delta }(Q^2)`$ are small is somewhat accidental. $`\delta A(Q^2,y)`$ is dominated by the bremsstrahlung contributions and the infrared parts of the $`\gamma \gamma `$ and $`\gamma Z`$ diagrams that happen to cancel each other for the asymmetric acceptance of the E158 spectrometer. On the other hand, $`\mathrm{\Delta }(Q^2)`$ is small at the $`Q^2`$ of E158 due to the cancellation between the $`WW`$ box diagrams and the large logarithmic contribution in the $`\gamma Z`$ box. Such precise cancellation is not expected at lower momentum transfers, or for other processes, such as elastic $`ep`$ scattering. For example, at $`y=0.5`$ and $`E=12`$ GeV, which corresponds to the idealized kinematics for a proposed Møller scattering experiments at the Jefferson Lab JLab12 , the corrections would be
$$\delta A(Q^2,y=0.5)_{12\mathrm{GeV}}=0.024\pm 0.005,\mathrm{\Delta }(Q^2)_{12\mathrm{GeV}}=0.0011\pm 0.0010.$$
(69)
For a 250 GeV fixed target Møller experiment, e.g. at a future Linear Collider MollerLC , one would find
$$\delta A(Q^2,y=0.5)_{250\mathrm{GeV}}=0.012\pm 0.005,\mathrm{\Delta }(Q^2)_{12\mathrm{GeV}}=0.0002\pm 0.0007.$$
(70)
## VII Conclusions
In conclusion, we have computed the QED corrections to the parity-violating left-right asymmetry $`A_{LR}`$ in Møller scattering. We used a covariant method for removing infrared divergences without introducing unphysical cutoffs. For the kinematics of SLAC E158 experiment, the overall corrections appear to be small, due to a fortuitous cancellation between electroweak and electromagnetic terms. We reduce the theoretical uncertainties due to higher order logarithmic terms by the appropriate choice of the couplings used to compute the box diagram contributions. Our calculation is applicable to a wide range of fixed target energies, from the proposed Møller scattering experiments at 12 GeV at the Jefferson Lab JLab12 , to the possible fixed target experiments at a future Linear Collider MollerLC .
## VIII Acknowledgments
The authors would like to thank Peter Bosted, Stanley Brodsky, Lance Dixon, Krishna Kumar, William Marciano, Michael Peskin, and Frank Petriello for stimulating discussions. VZ and JS would like to thank SLAC staff for the generous hospitality during their visits. This work has been partially supported by the National Science Foundation under grant PHY-0140366. |
warning/0507/math0507403.html | ar5iv | text | # Exact calculation of Fourier series in nonconforming spectral-element methods
## 1 Usefulness of calculating Fourier series in the SEM
In this note is presented a method, given nodal values on multidimensional nonconforming spectral elements, for calculating global Fourier-series coefficients. This method is “exact” in that given the approximation inherent in the spectral-element method (SEM), no further approximation is introduced that exceeds computer round-off error. The method is very useful when the SEM has yielded an adaptive-mesh representation of a spatial function whose global Fourier spectrum must be examined, e.g., in dynamically adaptive fluid-dynamics simulations such as .
## 2 Derivation of an exact transform
Suppose we have some functional problem in a spatial domain $`𝔻\mathit{:=}[\pi ,\pi ]^d`$ (possibly including toroidal geometry) and use coordinate transformations
$$\stackrel{}{\vartheta }_k\text{from}\stackrel{}{\xi }𝔼_0:=[1,1]^d\text{to}\stackrel{}{x}𝔼_k$$
(1)
to partition $`𝔻=_{k=1}^K𝔼_k`$ by $`K`$ elements $`𝔼_k\mathit{:=}\stackrel{}{\vartheta }_k(𝔼_0)`$ with disjoint<sup>1</sup><sup>1</sup>1$`\stackrel{}{𝔼}_k\stackrel{}{𝔼}_k^{}=\mathrm{}`$ if $`kk^{}`$ interiors. Typically the SEM approximates the exact solution by its piecewise polynomial representation of degree $`P`$:
$$u^{\mathrm{ex}}(\stackrel{}{x})u(\stackrel{}{x})=\underset{k=1}{\overset{K}{}}\underset{\stackrel{}{ȷ}𝕁}{}u_{\stackrel{}{ȷ},k}\varphi _{\stackrel{}{ȷ},k}(\stackrel{}{x}),$$
(2)
where $`𝕁:=\{0,\mathrm{}P\}^d`$ indexes the values $`u_{\stackrel{}{ȷ},k}\mathit{:=}u(\stackrel{}{x}_{\stackrel{}{ȷ},k})`$ and nodes $`\stackrel{}{x}_{\stackrel{}{ȷ},k}\mathit{:=}\stackrel{}{\vartheta }_k(\stackrel{}{\xi }_\stackrel{}{ȷ})`$ mapped from the $`d`$-dimensional Gauss-Lobatto-Legendre (GLL) quadrature nodes $`\xi _\stackrel{}{ȷ}^\alpha \mathit{:=}\xi _{ȷ^\alpha }[1,1]`$,
$$\varphi _{\stackrel{}{ȷ},k}(\stackrel{}{x})\mathit{:=}\{\begin{array}{cc}\varphi _\stackrel{}{ȷ}\stackrel{}{\vartheta }_k^1(\stackrel{}{x}),\hfill & \stackrel{}{x}𝔼_k\hfill \\ 0,\hfill & \stackrel{}{x}𝔼_k\hfill \end{array}$$
(3)
is the $`\stackrel{}{x}_{\stackrel{}{ȷ},k}`$-interpolating piecewise-polynomial,
$$\varphi _\stackrel{}{ȷ}(\stackrel{}{\xi })\mathit{:=}\underset{\alpha =1}{\overset{d}{}}\varphi _{ȷ^\alpha }(\xi ^\alpha )\text{and}\varphi _j(\xi )\mathit{:=}\underset{p=0}{\overset{P}{}}\stackrel{ˇ}{\varphi }_{j,p}\mathrm{L}_p(\xi )$$
(4)
are $`\stackrel{}{\xi }_\stackrel{}{ȷ}`$ \- and $`\xi _j`$-interpolating polynomials, $`\stackrel{ˇ}{\varphi }_{j,p}w_j\mathrm{L}_p(\xi _j)/_{j^{}=0}^Pw_j^{}\mathrm{L}_p(\xi _j^{})^2`$ is a Legendre coefficient \[e.g., 4, (B.3.15)\], $`\sqrt{p+\frac{1}{2}}\mathrm{L}_p(\xi )`$ is the orthonormal Legendre polynomial of degree $`p`$ on $`[1,1]`$ and $`w_j`$ is the GLL quadrature weight. In many cases a physically interesting quantity is the global Fourier-series coefficient $`\widehat{u}_\stackrel{}{q}`$ at integer wavenumber components $`q^\alpha `$, usually approximated by $`M^d`$-point trigonometric $`d`$-cubature in such manner as
$`\widehat{u}_\stackrel{}{q}`$ $`:={\displaystyle \frac{1}{(2\pi )^d}}{\displaystyle _𝔻}u(\stackrel{}{x})\mathrm{}^{\mathrm{i}\stackrel{}{q}\mathbf{}\stackrel{}{x}}v(\stackrel{}{x}){\displaystyle \underset{k=1}{\overset{K}{}}}{\displaystyle \underset{\stackrel{}{ȷ}𝕁}{}}\widehat{\varphi }_{\stackrel{}{ȷ},k,\stackrel{}{q}}u_{\stackrel{}{ȷ},k},`$ (5)
$`\text{where}\widehat{\varphi }_{\stackrel{}{ȷ},k,\stackrel{}{q}}`$ $`={\displaystyle \frac{1}{M^d}}{\displaystyle \underset{\stackrel{}{m}𝕄}{}}\varphi _{\stackrel{}{ȷ},k}(\stackrel{}{x}_\stackrel{}{m})\mathrm{}^{\mathrm{i}\stackrel{}{q}\mathbf{}\stackrel{}{x}_\stackrel{}{m}}_\stackrel{}{q}\varphi _{\stackrel{}{ȷ},k},`$ (6)
$`v(\stackrel{}{x}):=_{\alpha =1}^dx^\alpha `$ is the volume differential and $`𝕄:=\{1,\mathrm{}M\}^d`$ indexes trigonometric nodes $`x_\stackrel{}{m}^\alpha \mathit{:=}(2m^\alpha /M1)\pi `$. Note whenever $`𝔻`$ is adaptively repartitioned there is an additional computation cost of $`𝒪(M^d)`$ per node to use (2) to provide in (6) the values $`\varphi _{\stackrel{}{ȷ},k}(\stackrel{}{x}_\stackrel{}{m})`$, as well as a $`d`$-cubature error \[generalizing 3, theorem 4.7\]
$`_\stackrel{}{q}u{\displaystyle \underset{\stackrel{}{r}\mathrm{}^d\{\stackrel{}{0}\}}{}}\widehat{u}_{\stackrel{}{q}+M\stackrel{}{r}}`$
that in general converges no faster than $`𝒪(M^2)`$, because $`^1`$ discontinuities of (2) across element boundaries cause $`|\widehat{u}_\stackrel{}{q}|`$ to decay only as $`𝒪(|\stackrel{}{q}|^2)`$. We discover a more accurate method by substituting (3) into (5) to yield
$`\widehat{\varphi }_{\stackrel{}{ȷ},k,\stackrel{}{q}}`$ $`={\displaystyle \frac{1}{(2\pi )^d}}{\displaystyle _{𝔼_k}}\mathrm{}^{\mathrm{i}\stackrel{}{q}\mathbf{}\stackrel{}{x}}\varphi _\stackrel{}{ȷ}\stackrel{}{\vartheta }_k^1(\stackrel{}{x})v(\stackrel{}{x})`$
$`\stackrel{(\text{1})}{=}{\displaystyle \frac{1}{(2\pi )^d}}{\displaystyle _{𝔼_0}}\mathrm{}^{\mathrm{i}\stackrel{}{q}\mathbf{}\stackrel{}{\vartheta }_k(\stackrel{}{\xi })}\varphi _\stackrel{}{ȷ}(\stackrel{}{\xi })\left|{\displaystyle \frac{\stackrel{}{\vartheta }_k}{\stackrel{}{\xi }}}\right|v(\stackrel{}{\xi })`$
$`\stackrel{(\text{4})}{=}{\displaystyle \frac{1}{(2\pi )^d}}{\displaystyle _{𝔼_0}}\mathrm{}^{\mathrm{i}\stackrel{}{q}\mathbf{}\stackrel{}{\vartheta }_k(\stackrel{}{\xi })}\left({\displaystyle \underset{\alpha =1}{\overset{d}{}}}{\displaystyle \underset{p=0}{\overset{P}{}}}\stackrel{ˇ}{\varphi }_{ȷ^\alpha ,p}\mathrm{L}_p(\xi ^\alpha )\right)\left|{\displaystyle \frac{\stackrel{}{\vartheta }_k}{\stackrel{}{\xi }}}\right|v(\stackrel{}{\xi }).`$
In many applications, especially when $`u`$-structure rather than domain geometry is guiding the mesh adaption, each $`𝔼_k`$ is a $`d`$-parallelepiped with center $`\stackrel{}{a}_k`$ and $`d`$ legs $`2\stackrel{}{h}_k^\alpha `$, so we have an affinity $`\stackrel{}{\vartheta }_k(\stackrel{}{\xi }):=\stackrel{}{a}_k+\stackrel{}{\stackrel{}{h}}_k\mathbf{}\stackrel{}{\xi }`$, where $`\stackrel{}{h}_k^\alpha `$ make up the columns of $`\stackrel{}{\stackrel{}{h}}_k`$. Then we obtain
$$\widehat{\varphi }_{\stackrel{}{ȷ},k,\stackrel{}{q}}=\frac{1}{(2\pi )^d}\left|\stackrel{}{\stackrel{}{h}}_k\right|\mathrm{}^{\mathrm{i}\stackrel{}{q}\mathbf{}\stackrel{}{a}_k}\underset{\alpha =1}{\overset{d}{}}\underset{p=0}{\overset{P}{}}\stackrel{ˇ}{\varphi }_{ȷ^\alpha ,p}_1^1\mathrm{}^{\mathrm{i}\stackrel{}{q}\mathbf{}\stackrel{}{h}_k^\alpha \xi }\mathrm{L}_p(\xi )\xi .$$
Finally, recalling the classical identity \[e.g., 1, exercise 12.4.9\] for the spherical Bessel function $`\mathrm{B}_p(r)`$ of the first kind,
$`\mathrm{B}_p(r)`$ $`{\displaystyle \frac{\mathrm{i}^p}{2}}{\displaystyle _1^1}\mathrm{}^{\mathrm{i}r\xi }\mathrm{L}_p(\xi )\xi ,`$ (7)
$`\text{we obtain}\widehat{\varphi }_{\stackrel{}{ȷ},k,\stackrel{}{q}}`$ $`={\displaystyle \frac{1}{\pi ^d}}\left|\stackrel{}{\stackrel{}{h}}_k\right|\mathrm{}^{\mathrm{i}\stackrel{}{q}\mathbf{}\stackrel{}{a}_k}{\displaystyle \underset{\alpha =1}{\overset{d}{}}}{\displaystyle \underset{p=0}{\overset{P}{}}}\stackrel{ˇ}{\varphi }_{ȷ^\alpha ,p}\mathrm{i}^p\mathrm{B}_p(\stackrel{}{q}\mathbf{}\stackrel{}{h}_k^\alpha ).`$ (8)
Note that most expressions in (8) can be precomputed; objects that may vary during a dynamically adaptive computation, such as $`\stackrel{}{a}_k`$ or $`\stackrel{}{h}_k^\alpha `$, typically take values from a sparse set, e.g., a collection of powers of 2. The computation of (5) now incurs no additional error beyond that of (2). Also note, to generalize to the case $`P=P_k^\alpha `$ is straightforward.
## 3 Accuracy of transform for 1D & 2D test cases
Equation (8) was implemented in MatLab<sup>®</sup> and tested using known results for (5). The most immediate test follows from (7), namely $`\widehat{u}_q^{\mathrm{ex}}=\widehat{\mathrm{L}_\mathrm{p}(/\pi )}_q=\mathrm{i}^p\mathrm{B}_p(\pi q)`$. In this case (5) was found to reproduce (7) to 12-16 digits for $`K=1`$, $`P18`$, implying similar performance for any polynomial $`u(\stackrel{}{x})`$ in this range. The next test was to put $`u^{\mathrm{ex}}(x)=\mathrm{sin}x`$, or $`\widehat{u}_q^{\mathrm{ex}}=(\delta _{q,1}\delta _{q,1})/2\mathrm{i}`$. Since this is not a polynomial we should expect at best to see algebraic convergence w.r.t. $`K`$ in a uniform meshing $`a_k=(k1)h_k\pi `$, $`h_k=2\pi /K`$ and exponential convergence w.r.t. $`P`$, as verified in Fig. 1. Note there is no need to test $`u^{\mathrm{ex}}(x)=\mathrm{sin}rx`$ for $`r>1`$ because of scaling.
We conclude by examining three 2D tests with adaptive meshing in the fashion of , using MatLab<sup>®</sup>. Fig. 2 confirms (5) in the case \[6, (19)\]
$$u^{\mathrm{ex}}(\stackrel{}{x})\underset{\stackrel{}{q}\mathrm{}^2}{}\mathrm{}^{b^1|q^1|+b^2|q^2|+\mathrm{i}\stackrel{}{q}\mathbf{}\stackrel{}{\stackrel{}{l}}\mathbf{}\stackrel{}{x}},$$
(9)
where $`b^\alpha =\frac{2}{5}`$ and $`\stackrel{}{\stackrel{}{l}}\left(\begin{array}{cc}l^1& l^2\\ l^2& l^1\end{array}\right)=\left(\begin{array}{cc}1& 2\\ 2& 1\end{array}\right)`$ is a biperiodicity-preserving “rotation”. As expected, the red curve (connecting the $`|\widehat{u}_\stackrel{}{q}|`$ peaks) shows a power-law decay in $`\stackrel{}{q}`$-space. Note, in this plot and those below the $`\stackrel{}{\stackrel{}{l}}`$-operation helps instigate mesh adaption but has the consequence of leaving $`\stackrel{}{q}`$ undersampled in $`\mathrm{}^2`$. In Fig. 3 is shown an initial condition \[6, (22)\]
$$\stackrel{}{u}^{\mathrm{ex}}(0,\stackrel{}{x}):=\stackrel{}{l}\mathrm{sin}\stackrel{}{l}\mathbf{}\stackrel{}{x}$$
(10)
for the 2D Burgers eq. As expected, $`\widehat{u}_\stackrel{}{q}`$ almost vanishes for $`\stackrel{}{q}\pm \stackrel{}{l}`$. Finally, at time $`t=1.6037/\pi |\stackrel{}{l}|^2`$ the analytic solution generalizing \[2, (2.5)\] to 2D is shown in Fig. 4. As expected for the *nearly* $`^0`$-discontinuous fronts $`\stackrel{}{l}`$ seen at left, $`|\widehat{u}_\stackrel{}{q}^1|`$ decays slightly faster than $`𝒪(|\stackrel{}{q}|^1)`$ but *only for wavevectors* $`\stackrel{}{q}\stackrel{}{l}`$ (red curve). |
warning/0507/nlin0507019.html | ar5iv | text | # Regularization of moving boundaries in a Laplacian field by a mixed Dirichlet-Neumann boundary condition — exact results
## Abstract
The dynamics of ionization fronts that generate a conducting body, are in simplest approximation equivalent to viscous fingering without regularization. Going beyond this approximation, we suggest that ionization fronts can be modeled by a mixed Dirichlet-Neumann boundary condition. We derive exact uniformly propagating solutions of this problem in 2D and construct a single partial differential equation governing small perturbations of these solutions. For some parameter value, this equation can be solved analytically which shows that the uniformly propagating solution is linearly convectively stable.
Boundaries between two phases that move according to the gradient of a Laplacian or diffusive field, occur in many fields of the natural sciences and have a long and intricate research history Pelce ; well known examples include viscous fingering in Hele-Shaw flow st ; visfing , solidification fronts in under-cooled melts Pelce , migration of steps BCF or electromigration of voids electro0 ; electro on the surface of layered solids or boundaries of bacterial colonies in an external nutrition field bact . Viscous fingering here takes a paradigmatic role as the oldest and most studied problem — determining the long time dynamics up to today leads to mathematical surprises Tanveer ; Tanveer2 ; KL ; Jaume .
A similar moving boundary problem arises in so-called streamer discharges PREUWC ; PRLMan that precede sparks and lightning. Streamer ionization fronts can be understood as moving boundaries separating an ionized phase from a non-conducting phase PRLMan ; Man04 ; Bern1 . The inner front structure can be approximated by a boundary condition of mixed Dirichlet-Neumann-type, as we will sketch below. A similar boundary condition appears in step motion on the surface of layered solids when the Schwoebel barrier is taken into account BCF . Our boundary condition has a similar physical effect as the curvature correction in viscous fingering. We do show here that it indeed stabilizes certain uniformly translating shapes. In our analysis below, we encountered a number of surprises: $`(i)`$ planar fronts are linearly unstable to transversal perturbations of arbitrary wave vector $`0<k<\mathrm{}`$, still we find that sufficiently curved fronts are linearly convectively stable; $`(ii)`$ a simple explicit uniformly translating solution can always be found; $`(iii)`$ linear perturbations of these solutions can be reformulated in terms of a single partial differential equation, $`(iv)`$ if the solution of this perturbation problem is Taylor expanded and truncated at any finite order, the eigensolutions seem to be purely oscillating, $`(v)`$ however, for a particular parameter value, the linear perturbation theory has an explicit analytical solution that shows that there are no oscillations. Rather, perturbations might grow for some time, while they simultaneously are convected to the back where they disappear. Only a shift of the original shape remains for $`t\mathrm{}`$. To our knowledge, this is the first explicit solution showing the convective stabilization of a curved front according to the concept of Zeldovich comb .
In fact, the interfacial dynamics with our boundary condition can be addressed by explicit analysis much further than the classical viscous fingering problem. It therefore might contribute not only to the understanding of ionization fronts, but also shed new light on other moving boundary problems like the classical viscous fingering problem.
A simple moving boundary approximation for a streamer ionization front was suggested by Lozansky and Firsov Firsov : The front penetrates into a non-ionized and electrically neutral region (indicated with a <sup>+</sup>) with a velocity determined by the local electric field $`𝐄^+=\phi ^+`$:
$$^2\phi ^+=0,v_n=\widehat{𝐧}\phi ^+,$$
(1)
where $`\widehat{𝐧}`$ is the local normal on the boundary. Approximating the interior ionized region as ideally conducting
$$\phi ^{}=\mathrm{const}.,$$
(2)
and the electric potential as continuous across the ionization boundary $`\phi ^+=\phi ^{}`$, one arrives at the Lozansky-Firsov interfacial model. This model was suggested in PRLMan to explain streamer branching, and it was explicitly analyzed in Bern1 . Replacing the electric potential $`\phi `$ by the pressure field $`p`$, one finds the non-regularized motion of viscous fingers in a Hele-Shaw cell. The model generically leads within finite time to the formation of cusps, i.e., of locations on the interface with vanishing radius of curvature entov .
We here propose to replace the boundary condition $`\phi ^+\phi ^{}=0`$ by
$$\phi ^+\phi ^{}=ϵ\widehat{𝐧}\phi ^+$$
(3)
to suppress these unphysical cusps. Here the length scale $`ϵ`$ characterizes the width of the ionization front where the ionization increases and the electric field decreases. It determines the jump $`\phi ^+\phi ^{}`$ of the electric potential across the boundary for given field $`\phi ^+`$ ahead of the front.
The classical boundary condition for viscous fingering is $`\phi ^+\phi ^{}=\gamma \kappa `$ where $`\kappa `$ is the local curvature of the moving interface, and $`\gamma `$ is surface tension. In contrast, the boundary condition (3) does not involve front curvature, but can be derived from planar ionization fronts, more precisely from a minimal set of partial differential equations for electron and ion densities and their coupling to the electric field PREUWC . The formal derivation will be given elsewhere. Here we note that ignoring electron diffusion ($`D=0`$) as in Man04 , the planar uniformly translating front solutions of the p.d.e.s always yield a relation $`\phi ^+\phi ^{}=F(\widehat{𝐧}\phi ^+)`$. For large field $`E^+=\widehat{𝐧}\phi ^+`$ ahead of the front, the function $`F`$ becomes linear, and the boundary condition (3) results.
This boundary condition has a similar physical effect as the curvature correction in viscous fingering: high local fields ahead of the front decrease due to the change of $`\phi ^+`$ on the boundary, and the interface moves slower than an equipotential interface (where $`\phi ^+=\mathrm{const}`$). While the boundary condition of viscous fingering suppresses high interfacial curvatures that can lead to high fields, the boundary condition (3) suppresses high fields that frequently are due to high local curvatures. This physical consideration has motivated our present study whether the boundary condition (3) also regularizes the interfacial motion.
The minimal p.d.e. model for streamer fronts with $`D=0`$ leads to a dispersion relation with asymptotes
$$s(k)=\{\begin{array}{cc}vk\hfill & \text{for }k1/ϵ\hfill \\ v/ϵ\hfill & \text{for }k1/ϵ\hfill \end{array}$$
(4)
for linear transversal perturbations $`e^{ikx+st}`$ of planar interfaces Man04 . It is important to check whether the moving boundary approximation (1)–(3) reproduces this behavior. Indeed, analyzing planar interfaces we find $`s(k)=vk/(1+ϵk)`$ in full agreement with (4) as we will show in detail elsewhere. This strongly suggests that the interfacial model captures the correct physics. It shows furthermore, that planar fronts are linearly unstable against any wave-vector $`k`$ for all $`ϵ`$.
We now restrict the analysis to the two-dimensional version of the model and to arbitrary closed streamer shapes in an electric field that becomes homogeneous
$$\phi (x,y)E_0x\text{far from the ionized body}.$$
(5)
The problem is treated with conformal mapping methods Bern1 : The exterior of the streamer where $`^2\phi ^+=0`$ can be mapped onto the interior of the unit circle. Parameterizing the original space with $`z=x+iy`$ and the interior of the unit disk with $`\omega `$, the position of the streamer can be written as
$$z=x+iy=f_t(\omega )=\frac{1}{h_t(\omega )}=\underset{k=1}{\overset{\mathrm{}}{}}a_k(t)\omega ^k,$$
(6)
where $`h_t(\omega )`$ is analytical on the unit disk with a single zero at $`\omega =0`$ and therefore has the Laurent expansion given on the right. The boundary of the ionized body
$$\omega =e^{i\alpha },\alpha [0,2\pi [,$$
(7)
is parametrized by the angle $`\alpha `$.
The potential $`\phi ^+`$ is a harmonic function due to (1), therefore one can find a complex potential $`\mathrm{\Phi }(z)=\phi ^++i\psi `$ that is analytic. Its asymptote is $`\mathrm{\Phi }(z)E_0z`$ for $`|z|\mathrm{}`$ according to (5). For the complex potential $`\widehat{\mathrm{\Phi }}(\omega )`$, this means that
$$\widehat{\mathrm{\Phi }}(\omega )=\mathrm{\Phi }(f_t(\omega ))=E_0a_1(t)\left(\frac{1}{\omega }+\underset{k=0}{\overset{\mathrm{}}{}}c_k(t)\omega ^k\right),$$
(8)
where the pole $`1/\omega `$ stems from the constant far field $`E_0`$, and the remainder is a Taylor expansion that accounts for the analyticity of $`\widehat{\mathrm{\Phi }}`$. The boundary motion $`v_n=\widehat{𝐧}\phi `$ (1) is rewritten as
$$\mathrm{Re}\left[i_\alpha f_t^{}_tf_t\right]=\mathrm{Re}\left[i_\alpha \widehat{\mathrm{\Phi }}(e^{i\alpha })\right].$$
(9)
The boundary condition (3) takes the form
$$\mathrm{Re}\left[\widehat{\mathrm{\Phi }}(e^{i\alpha })\right]=ϵ\mathrm{Re}\left[\frac{i_\alpha \widehat{\mathrm{\Phi }}(e^{i\alpha })}{|_\alpha f_t|}\right].$$
(10)
The moving boundary problem is now reformulated as Eqs. (9) and (10) together with the ansätze (6) and (8) for $`f_t`$ and $`\widehat{\mathrm{\Phi }}`$.
For the unregularized problem (where $`ϵ=0`$), it is well known that all ellipses with a main axis oriented parallel to the external field are uniformly translating solutions: for principal radii $`r_{x,y}=a_1\pm a_1`$, they propagate with velocity $`v=E_0(r_x+r_y)/r_y`$, while the potential is $`\widehat{\mathrm{\Phi }}=E_0a_1(t)(\omega 1/\omega )`$ Bern1 ; entov .
For a moving boundary problem with regularization, there are only rare cases where analytical solutions can be given, and frequently they are given only implicitly SolVF ; SolVF2 ; entov . For the present problem, however, an explicit solution is found for all $`ϵ>0`$:
$`z`$ $`=`$ $`f_t(\omega )={\displaystyle \frac{a_1}{\omega }}+vt,_ta_1=0,`$ (11)
$`\widehat{\mathrm{\Phi }}(\omega )`$ $`=`$ $`E_0a_1\left({\displaystyle \frac{1}{\omega }}{\displaystyle \frac{1ϵ/a_1}{1+ϵ/a_1}}\omega \right).`$ (12)
This solution simply describes a circle $`z=x+iy=a_1e^{i\alpha }+vt`$ with radius $`a_1`$ that according to (9) propagates with velocity $`v=2E_0/(1+ϵ/a_1)`$. Note that $`ϵ`$ changes the velocity, but not the shape of the solution. Note further that the multiplicity of uniformly translating solutions reduces through regularization in a similar way as in viscous fingering, namely from a family of ellipse solutions characterized by two continuous parameters $`a_1`$ and $`a_1`$ to a family of circle solutions characterized by only one radius $`a_1`$ or interface width $`ϵ`$.
The physical problem has two length scales, the interface width $`ϵ`$ and the circle radius $`a_1`$. In the sequel, we set $`a_1=1`$, measuring all lengths relative to the radius of the circle.
Now the question arises whether a uniformly translating circle is stable against small perturbations, in particular, in view of the linear instability of the planar front (4). The basic equations (6)–(10) show a quite complicated structure, and it is a remarkable feature that linear stability analysis of the translating circle (11)–(12) can be reduced to solving a single partial differential equation. We write
$`f_t(\omega )`$ $`=`$ $`{\displaystyle \frac{1}{\omega }}+\tau +\beta (\omega ,\tau ),\tau =vt,v={\displaystyle \frac{2E_0}{1+ϵ}},`$ (13)
$`\widehat{\mathrm{\Phi }}(\omega )`$ $`=`$ $`E_0\left({\displaystyle \frac{1}{\omega }}{\displaystyle \frac{1ϵ}{1+ϵ}}\omega \right)+v\varphi (\omega ,\tau ),`$ (14)
where $`\beta `$ and $`\varphi `$ are analytical in $`\omega `$ and assumed to be small. Eqs. (9) and (10) are expanded to first order in $`\beta `$ and $`\varphi `$ about the uniformly translating circle and read
$`\mathrm{Re}\left[\omega _\tau \beta \omega _\omega \beta \right]=\mathrm{Re}\left[\omega _\omega \varphi \right]\text{for }\omega =e^{i\alpha },`$ (15)
$`{\displaystyle \frac{ϵ}{2}}\left(\omega +{\displaystyle \frac{1}{\omega }}\right)\mathrm{Re}\left[\omega ^2_\omega \beta \right]=\mathrm{Re}\left[ϵ\omega _\omega \varphi +\varphi \right].`$ (16)
By construction, $`F(\omega )=_\tau \beta _\omega \beta +_\omega \varphi `$ is analytical for $`|\omega |<1`$, and Eq. (15) shows that $`\mathrm{Re}[\omega F(\omega )]=0`$ for $`|\omega |=1`$. Furthermore, it is clear that $`\omega F(\omega )`$ vanishes for $`\omega =0`$. Therefore,
$$0=\omega F(\omega )=\omega \left(_\tau \beta _\omega \beta +_\omega \varphi \right)$$
(17)
is valid on the whole closed unit disk. The corresponding analysis of Eq. (16) yields
$$\frac{ϵ}{2}\left(\omega +\frac{1}{\omega }\right)\omega ^2_\omega \beta =ϵ\omega _\omega \varphi +\varphi +\mathrm{const}.$$
(18)
To this equation the operator $`\omega _\omega `$ is applied, and Eq. (17) is used to eliminate terms containing $`\omega _\omega \varphi `$. As a result, we find an equation only for the function $`\beta (\omega ,\tau )`$:
$`_ϵ\beta `$ $`=`$ $`0,`$ (19)
$`_ϵ`$ $`=`$ $`ϵ(1\omega ^2)\omega _\omega ^2(2+ϵ3ϵ\omega ^2)_\omega `$ (20)
$`+2ϵ\omega _\omega _\tau +2(1+ϵ)_\tau .`$
Eq. (19) has to be solved for arbitrary initial conditions $`\beta (\omega ,0)`$ that are analytical in some neighborhood of the unit disk. The operator $`_ϵ`$ conserves analyticity in time. $`ϵ`$ is a singular perturbation that multiplies the highest derivatives $`_\omega ^2`$ and $`_\omega _\tau `$.
The case $`ϵ=0`$ is almost trivial, since $`_ϵ`$ reduces to
$$_0=2(_\tau _\omega ).$$
(21)
Thus all solutions can be written as
$$\beta (\omega ,\tau )=\widehat{\beta }(\omega +\tau ),$$
(22)
where $`\widehat{\beta }(\zeta )`$ is any function analytic in a neighborhood of the unit disk $`|\zeta |1`$. The time evolution just amounts to a translation along the strip $`1\mathrm{Re}\zeta \mathrm{}`$, $`|\mathrm{Im}\zeta |1`$. Any singularity of $`\widehat{\beta }`$ at some finite point $`\zeta `$ on the strip will lead to a breakdown of perturbation theory within finite time; this is the generic behavior as found previously in the full nonlinear analysis of this unregularized problem. Of course, there also exist solutions that stay bounded for all times.
A different perspective on $`ϵ=0`$ is that the Richardson moments are an infinite sequence of conserved quantities entov . A reflection of this property is that any polynomial $`\beta (\omega ,\tau )=_{k=0}^Nb_k(\tau )\omega ^k`$ for any $`N`$ with an appropriate choice of the time dependent functions $`b_k(\tau )`$ is an exact solution for all times $`\tau >0`$, for linear perturbation theory (19) as well as for the full nonlinear problem Bern1 , i.e., any truncation of the Laurent series (6) leads to exact solutions.
This observation suggests that an expansion in powers of $`\omega `$ is a natural ansatz also for nonvanishing (but small) $`ϵ`$. Taking as initial condition some polynomial of order $`N`$, one finds from the form (20) of $`_ϵ`$ that higher modes $`\omega ^k`$, $`k>N`$ are generated dynamically — similarly to the daughter singularities in regularized viscous fingers Tanveer . When the expansion in $`\omega `$ is truncated at some arbitrary $`N^{}`$, it can be shown that the problem for any truncation $`N^{}`$ and for any value $`ϵ>0`$ has purely imaginary temporal eigenvalues. One would therefore expect all eigensolutions for $`ϵ>0`$ to be purely oscillating in time. However, this behavior seems inconsistent with our exact solution for $`ϵ=1`$.
For $`ϵ=1`$ it turns out that the operator factorizes
$`_1`$ $`=`$ $`\left[2_\tau (1\omega ^2)_\omega \right]\left[2+\omega _\omega \right].`$ (23)
which allows us to construct the general solution. We introduce the function
$$g(\omega ,\tau )=[2+\omega _\omega ]\beta (\omega ,\tau ),$$
(24)
that obeys the equation
$$[2_\tau (1\omega ^2)_\omega ]g(\omega ,\tau )=0.$$
(25)
The general solution of this equation reads
$$g(\omega ,\tau )=G\left(\frac{\omega +T}{1+T\omega }\right),T=\mathrm{tanh}\frac{\tau }{2}.$$
(26)
The function $`G`$ is derived from the initial condition as
$$G(\omega )=g(\omega ,0)=[2+\omega _\omega ]\beta (\omega ,0);$$
(27)
hence it is analytical in a neighborhood of the unit disk. Finally, Eq. (24) is solved by
$$\beta (\omega ,\tau )=_0^\omega \frac{xdx}{\omega ^2}G\left(\frac{x+T}{1+Tx}\right).$$
(28)
Now the one parameter family of mappings
$$\omega \zeta _T(\omega )=\frac{\omega +T}{1+T\omega },1<T<1,$$
(29)
forms a subgroup of the automorphisms of the unit disk. Thus on the level of $`G(\zeta )`$, the dynamics amounts to a conformal mapping of the unit disk $`|\omega |1`$ onto itself. This dynamics is somewhat distorted by the additional integration (28) leading to $`\beta (\omega ,\tau )`$, but it is easily seen that $`\beta (\omega ,\tau )`$ and $`_\omega \beta (\omega ,\tau )`$ are bounded uniformly in $`\tau `$ for $`|\omega |1`$. Hence, contrary to the unregularized problem for $`ϵ=0`$, only perturbations contribute that are bounded for all times. Hence an infinitesimal perturbation can never form cusps. Furthermore, the mapping $`\omega \zeta _T(\omega )`$ has fixed points $`\omega =\pm 1`$; and for $`\tau \mathrm{}`$, i.e., $`T1`$, it degenerates to $`\zeta _1(\omega )1`$, provided $`\omega 1`$. We thus find the asymptotic behavior
$$\beta (\omega ,\tau )\stackrel{\tau \mathrm{}}{}\frac{G(1)}{2},$$
(30)
independent of $`\omega `$ for any initial condition. Therefore asymptotically, the perturbation just shifts the basic circular solution without change of shape. Indeed, it is easily checked that any pronounced structure of the initial perturbation that is not located right at the top at $`\omega =1`$, is convected with increasing time toward $`\omega =1`$ where it vanishes. This is an outflow of the simple dynamics of $`G(\zeta )`$ as pointed out above. Fig. 1 illustrates this behavior.
To summarize, we have found that the boundary condition (3) at least for $`ϵ=1`$ regularizes our problem in the sense that an infinitesimal perturbation of a uniformly translating circle stays infinitesimal for all times and vanishes asymptotically for $`\tau \mathrm{}`$ up to an infinitesimal shift of the complete circle. This statement is based on an exact analytical solution for an arbitrary initial perturbation. At the present stage, we have indications that this behavior of infinitesimal perturbations is generic for $`ϵ>0`$, while the solution is unstable for $`ϵ=0`$. Furthermore, we expect that the convection of perturbations to the back of the structure applies similarly for other shapes like fingers. When applying the present calculation to streamers, we in fact have to assume this to be true, since streamers are typically not closed bodies, but rather the tips of ionized channels. Finally, the behavior of finite perturbations and their nonlinear analysis will require future investigations. |
warning/0507/physics0507200.html | ar5iv | text | # On a natural definition of the kilogram and the ampere
## Abstract
We consider a recent proposal to redefine the kilogram in terms of natural constants. In our opinion, the main objective of the redefinition should be to build such a version of the SI system in which the electric measurements are possible with the highest accuracy in SI units, and not in practical units as now. We emphasize that this objective can be achieved only with a simultaneous redefinition of the kilogram and ampere. This redefinition must be in terms of fixed values of the Planck constant $`h`$ and the elementary charge $`e`$. Certain details of the possible redefinition are considered.
This paper considers the recent proposal to redefine the SI kilogram and possibly the ampere in terms of fixed values of fundamental physical constants. This would change the International System of units (the SI) , which is a commonly accepted coherent system for all branches of macroscopic measurements in education, sciences and technology. The redefinition , which has been suggested in terms of fundamental constants, also indirectly involves certain natural quantum phenomena, which should appear due to realizations of the redefinition.
Some time ago the SI was changed in a similar matter by fixing a value of the speed of light $`c`$ . However, the present situation is very different. To our opinion the major problem now is that the present high-accuracy measurements in mechanics and electricity are performed in two different versions of the SI. While macroscopic mass measurements are performed in terms of the SI kilogram, the most accurate electric measurements are performed in terms of the practical units ohm-90 and volt-90. These latter are apparently not consistent with the SI.
Microscopic mass measurements, however, are related to mass determined in unified atomic mass units and in frequency units, i.e., dealing with a value of $`mc^2/h`$ instead of the mass $`m`$. Thus, they are measured in units closely related to the ohm-90 and volt-90. For instance, the SI value of $`h`$ has a larger uncertainty than microscopic mass comparisons, while in the practical units-90 the numerical value of $`h`$ is known exactly.
The proposal and its numerous considerations in various international commissions have been focussed on the desirability of replacing the definition of the unit for mass measurements, now based on the last artefact of the SI, the kilogram prototype kept at the BIPM in Sèvres, by a definition which is stable and independently reproducible. That mainly focuss attention on the kilogram alone, while a redefinition of the ampere is considered as one of a number of unnecessary collateral options.
On contrary, we believe that the gap between the present version of the SI and the system based on the ohm-90 and the volt-90 is a crucial reason to consider such a redefinition. The modern version of the SI was introduced in 1983 by fixing the value of the speed of light $`c`$ by CIPM , while in 1988 CIPM recommended a departure from the SI by introducing the practical electric units which have been in effect since 1990.
The desirability of resolving the inconsistency between units used in precision electric and macroscopic mass measurements and restoring the SI system as the only system of units for precision macroscopic measurements drives us to a possible redefinition of the kilogram and the ampere at the same time. We note that the gap appeared because the requirement for performing the most precise electric and mass measurements in the SI units was partly inconsistent. It still is and may remain for an uncertain period of time.
The present version of the SI is based on the kilogram prototype and a fixed value of the magnetic constant $`\mu _0`$, while the practical units are based on fixed values of the von Klitzing constant $`R_K`$ and the Josephson constant $`K_J`$ (see Table 1 for the values). If we intend to define a version which allows the derivation of fixed values of $`R_K=h/e^2`$ and $`K_J=2e/h`$, we have to fix values of two fundamental constants, e.g., the Planck constant $`h`$ and the elementary charge $`e`$. To fix two values, we must redefine two units at the same time.
Thus, the necessary requirement for the redefinition to resolve the inconsistency is to redefine two units by fixing values of two fundamental constants at the same time. The redefinition of the kilogram alone would be of a reduced importance.
Most of the fundamental constants are related to microscopic physics (atomic, nuclear or particle physics) and their numerical values are of two kinds being a result of
* a pure microscopic comparison (e.g., a value of $`m_e/m_p`$);
* a comparison between microscopic and macroscopic values (e.g., a value of the electron mass in kilograms or eV/c<sup>2</sup>)
The pure microscopic data are more accurate than the data which involve also macroscopic physics and that is mainly a consequence of the limited accuracy of measurements linking macroscopic and atomic physics. Very few numerical values, such as for the gravitation constant $`G`$, comes from pure macroscopic experiments, and these play only a marginal role in precision measurements.
Apparently, without any new experiments we cannot improve the links between the microscopic and macroscopic physics. However, the numerical values of the fundamental constants play an important role as anchor reference data. For instance, it is customary to express results for X-ray transitions rather in units of energy (eV) than in terms of the frequency or wave length. To interpret the frequency as an energy (in eV), one has to apply a value of $`h/e`$. We note that the accuracy of comparisons of two transitions is higher than that of the available numerical value of $`h/e`$ in the SI units . By changing the basis of the definition of the SI units of mass and charge we can improve quality of the reference data, and the characterization of the X-ray transition in terms of electron-volts would be adequate. This could be achieved by defining the units of mass and charge, which are now macroscopic, in microscopic terms, i.e., in terms of $`h`$ and $`e`$.
For the SI, the most questionable link between macroscopic and microscopic physics is related to experiments on determination of the Planck constant $`h`$. There is currently an unresolved discrepancy of 1 ppm between values of the Planck constant derived from the watt-balance experiments and from the X-ray crystal density (XRCD) determination (see, e.g., ). The results of all other measurements together produce a third value that is competitive in accuracy with the XRCD result and is in a perfect agreement with the watt-balance values (see Fig. 1). The importance of this third result is often underplayed.
These experiments determine a link between the macroscopic mass unit, the kilogram, and the electric power unit expressed in terms of volt-90 and ohm-90. This is the crucial link for the realization of the SI ampere in the present version of the SI. In the proposed version of the SI , based on the kilogram unit defined by a fixed value of the Planck constant $`h`$ or the Avogadro constant $`N_A`$, these experiments determine the mass of the kilogram prototype.
Recently a number of international commissions and committees considered this issue. They emphasized the importance of the problem related to this link and its undesirable effect on accuracy in mass measurements in the case of the redefinition. Their concerns are based on an assumption that it is up to those who decide on the redefinition to involve this link into the SI or not. We agree that this link is a great problem. But we unfortunately disagree that this link can be avoided by, e.g., postponing the redefinition of the kilogram. This link, as we mention above, is crucial in present-day realizations of the ampere (and the volt) of the SI. In other words, it has been used at least from 1990 for the realization of the SI ampere and there is no way to avoid this troubled link.
We also raise a question about the conceptual difference between a constant-based unit and an artefact-based unit. In the latter case, the definition can have fundamental problems, but it is very instructive. It is clear in a practical sense what the unit is and, in the most of comparisons, the method of the comparison is also obviously fixed. There is not much room for any variety in realizing the standards. In the former case, when a unit is based on a constant and certain relations to other quantities (i.e. certain physical laws), there are a number of ways to realize the definition and, as in the case of any scientific experiment, the results may disagree. A substantial difference for a constant-based unit and an artefact-base unit is due to possible systematic effects. For the artefact, the systematic effects may take place, but be reproducible. That is an advantage of the artefact from a practical point of view. However, such a systematic effect, being invisible, can produce a drift of the unit or a reproducible systematic shift (if a way of comparison is compromised). Various differences in possible realizations of the constant-based units may produce a discrepancy but that would allow detection of a possible problem. The very opportunity to discover the problem, even accompanied by possible discrepancies, is an advantage.
In the case of a constant-based unit, the systematic effects may be different. For instance, for determination of the Planck constant, which is the realization of the SI ampere (presently) and of the SI kilogram (in the case of the redefinition), these effects are different and particular results disagree with each other. Relative mass measurements and relative electric measurements are more accurate than the link. For the electric units, CIPM has chosen a clear strategy. A conservative result for the Planck constant (and, consequently, for the realization of the SI volt and ampere) has been accepted <sup>1</sup><sup>1</sup>1We have to note that any legal adoption of any scientific result (as, e.g., various mise en pratique) is an introduction of a non-SI unit. The SI is a closed system of definitions. Adoption of anything else as a part of the SI changes the system, while adoption of anything beyond the SI leads to practical units. In particular, once the accuracy for the SI values of $`K_J`$ and $`R_K`$ is adopted, we arrive at a contradiction between the CIPM recommendation and the accuracy determined by scientific means , which is a result of direct application of the original SI definitions ., while the most accurate measurements are to be performed in practical units. The same approach should be used for the kilogram in the case of the redefinition.
In principle, after the redefinition of the kilogram and progress in its realization, another situation could take place. It may happen that uncertainty of best realizations and even their discrepancy (if any) could be smaller than uncertainty related to the prototype. What strategy should be used to deal with a possible discrepancy (which in principle could appear from time to time for any constant-based units)? As long as one particular method could provide us with reproducible results we should accept it to define a practical unit, while the related SI unit would be still defined in a conservative way.
Let us to look now into possible consequences of the redefinition. First, we need to stress that the only reasonable version of such a redefinition is to fix $`h`$ and $`e`$. We can present certain advantages in fixing $`h`$ instead of $`N_A`$ for the redefinition of the kilogram only. In particular, the variety, a relatively low degree of interdependence, and the level of accuracy that has been achieved makes watt-balance experiment more desirable than the XRCD measurement. The watt-balance experiment would be a preferred realization of the kilogram if the Planck constant $`h`$ is fixed. On the other hand, the XRCD technique is the most natural choice for the kilogram based on a fixed value of $`N_A`$.
One should not overestimate importance of these straightforward preferences, which are rather educational and practical. However, they may become of practical importance if the accuracy of mass measurements increases. As shown in Table 1, the uncertainty of the molar Planck constant $`hN_A`$ is below 10 ppb. If we fix one of these two constants, this value would be the uncertainty of the other. At the present level of accuracy in maintaining the kilogram, in mass measurements and in the link between the kilogram and the electric units, this uncertainty is as good as zero. The accuracy of the link (i.e. of the present-day determinations of $`h`$ and $`N_A`$ separately) for a number of experimental reasons should be substantially lower than that of the $`hN_A`$ for a long but uncertain period of time. For a redefinition of the kilogram alone, it does not much matter which of these two constants to fix.
However, we have to redefine the kilogram and the ampere at the same time. We should clearly choose a redefinition that will produce fixed values of $`R_K`$ and $`K_J`$ to preserve the advantages of using the present-day practical units. There is no way to do that unless we fix $`h`$ and $`e`$, and we consider this scenario below. In contrast, if the values of $`N_A`$ and $`e`$ are fixed instead, the electrical community would still need practical units of resistance and potential. The accuracy of the SI measurements of those quantities would increase in comparison with the present situation, but still would be below the one accessible in terms of the practical units.
If the value of $`h`$ and $`e`$ were fixed, then the uncertainties of the Avogadro constant, the mass of the electron, proton, and various atoms and nuclei would be substantially smaller. The uncertainties of the Rydberg constant and various other frequencies expressed in eV would be greatly improved as well. The uncertainties of the many electrical measurements in SI units would be obviously substantially smaller.
Unfortunately, the proposal would have an undesirable effect on the macroscopic mass measurements in SI units. The consequences of such effects should be prevented by using the existing prototype of the kilogram in a way similar to the way the quantum Hall and Josephson standards are used now. Measurements in terms of the unit determined by the existing prototype would be considered measurements in conventional units recommended by CIPM for a transition period. Macroscopic mass measurements in SI units could be performed by comparison to the existing prototype together with its calibration by the measurements which now serve as determinations of the Planck constant.
How large could the undesirable effects be? At one time it was assumed that a 10-ppb uncertainty in experiments relating macroscopic and microscopic masses would be desirable in order to implement a redefinition of the kilogram. However, it is now felt that this is not realistic or necessary, particularly in view of the instability of the mass of the kilogram. In fact, its mass changed by more than 60 ppb when it was last washed in the verification in 1988–1992 (see, e.g., ). However, this level of accuracy is needed only for experiments like determination of $`h`$ and $`N_A`$. Accuracy, needed for practical applications is at the level of few parts in $`10^7`$. Indeed, we should like to have more accurate standards to perform various tests, but such tests on, e.g., consistency of various national standards, may be performed in practical units.
Since the current definition of the SI ampere was essentially abandoned by the electrical community when conventional values of the von Klitzing and Josephson constants were introduced in 1988, there would be no discontinuity in the electrical units if the ampere were redefined to make the elementary charge $`e`$ exact and the kilogram were defined via a fixed value of $`h`$. In fact, there would be a significant gain in precision of SI electrical units, because the SI ohm and volt would be exactly defined in terms of the quantum Hall and Josephson effects. The present CIPM recommendations on the ohm-90 and volt-90 set the uncertainty at the level of one and four parts in $`10^7`$, respectively, while the most accurate measurements in practical units are done at the level of few parts in $`10^9`$ and such measurements are of practical interest.
The accuracy, continuity, and stability of the units and of the fundamental constants is critical to the scientific community. In principle the requirement for the continuity of various units and constants may be controversial. In particular, the suggested units, obtained by fixing $`h`$ and $`e`$, will not necessarily be the same as conventional units, since corrections to the standard expressions for $`R_K`$ and $`K_J`$ in terms $`h`$ and $`e`$ of are possible.
There are basically two options.
* We can set new ohm and new volt to be the same as volt-90 and ohm-90 (except for the necessity of rounding the values in the definitions). In this case all results in practical units will be accepted as the SI results, while value of the kilogram and certain fundamental constants will jump (numerical values of some constants in the SI units and units-90 are presented in Table 1).
* We can choose an option to adopt values of $`h`$ and $`e`$ as they are in the CODATA paper (or the newest available CODATA results at the time of the redefinition). That will reduce a possible jump<sup>2</sup><sup>2</sup>2The very existence of the jump becomes questionable and some believe that there would be no jump at all. We choose here to use a common word ‘jump’ instead of ‘discontinuity’ because in a sense there would be no discontinuity, but there should be a certain jump. The jump would be a result of a two-step action. For instance, we use an exact value of $`\mu _0`$ . With the redefinition, we make it measurable. Once we use the CODATA data (see Table 1), the new result for $`\mu _0=2\alpha R_K/c`$ should have an uncertainty. The new result may be consistent with the presently fixed value of $`\mu _0`$ . Nevertheless, sooner or later with improvement of accuracy, a value of $`\mu _0`$ would depart from the previously fixed numerical value. There is no chance that we can guess the values for $`e`$ and $`h`$ in such a way that the measurable $`\mu _0`$ would be exactly the same as before. That is the same as trying to guess an exact value of the fine structure constant $`\alpha `$. So, eventually a certain jump would take place, but in each step we should have no discontinuity. in the kilogram of SI and values of the fundamental constants.
Technically we can fix first $`R_K`$ and $`K_J`$, calculate $`h`$ and $`e`$, and round them properly at the end. It is more transparent to discuss consequences of fixing different values of $`R_K`$ and $`K_J`$ than of $`h`$ and $`e`$.
At present, CODATA’s $`\{R_K\}_{\mathrm{SI}}`$ differs from CIPM’s $`\{R_K\}_{90}`$ by approximately five standard deviations (see Table 1) and we have to choose between them. On the other hand, CODATA’s $`\{K_J\}_{\mathrm{SI}}`$ differs from CIPM’s $`\{K_J\}_{90}`$ by less than one standard deviations (see Fig. 1) and we can choose either of them without any serious consequences.
A choice between different values of $`R_K`$ would affect a value of $`\mu _0`$ and a possible departure of the new SI ohm from the present SI ohm. If we choose the CIPM’s value, the new ohm SI would be related to ohm-90 and to the present CIPM central value of the SI ohm (we cannot discuss here the SI ampere because its definition involve $`\mu _0`$ and the kilogram, which is also a subject of changes). A value of $`\mu _0`$ would depart from its present value, but it is not particularly important for precision measurements. The only experiments are with calculable capacitors, but there are very few of them around the world and a shift at the level below 20 ppb is not important at their present level of realization.
In principle, impact of choice between different values of $`K_J`$ could be more important. Different $`K_J`$ would lead to different values of the kilogram, the volt and the ampere and different numerical values of various important fundamental constants such as $`h`$, $`e`$, particle masses in kilograms and eV/c<sup>2</sup>, and energy of various atomic and nuclear transitions in electron volts. Fortunately, as we mention above, CODATA’s $`\{K_J\}_{\mathrm{SI}}`$ and CIPM’s $`\{K_J\}_{90}`$ agree to each other within a standard deviation (the difference is approximately half a deviation) and perhaps, because of this agreement, we should choose the CIPM’s value. A change in a value of the fundamental constants within one sigma is not a discontinuity, and we would prefer to set the redefined SI units to be equal to the practical units.
That does not downplay the importance of the CODATA values. The CODATA evaluation will determine a recommended value of the magnetic constant $`\mu _0`$ and a value of the mass of the prototype $`m(𝒦)`$. The situation with $`K_J`$ may change by 2007 and with new watt-balance and XRCD results the CODATA’s $`\{K_J\}_{\mathrm{SI}}`$ could depart from $`\{K_J\}_{90}`$. If the difference would be above one standard deviation we will need to make a real choice between CODATA’s and CIPM’s values.
We believe that the kilogram and the ampere should be redefined and that they should be redefined at the same time by fixing values of the fundamental constants $`h`$ and $`e`$. Two open practical questions are related to choice for the fixed values (discussed above) and to a proper time for the redefinition.
A choice for the timing should consider the following.
* Any decision (positive or negative) on the redefinition will have benefits and expenses. These have to be examined carefully.
* Some disadvantages (discontinuity in values of units and numerical values of the constants, worsening of accuracy of measurements in SI units, etc.) may be unavoidable. It is necessary to take into account that postponing a necessary decision could increase expenses.
Considering the advantages and disadvantages we point that
* from the point of view of mass metrology, the redefinition of the kilogram will be successful once the Planck constant is reliably determined with a standard uncertainty less than about 50 ppb;
* from the point of view of electric measurements, the redefinition will be successful even now because of the immediate improvement in accuracy of precision electric measurements in SI units;
* the final decision on the proper time for the redefinition of the kilogram and the ampere can be made when there is a net gain based on a careful comparison of the advantages and disadvantages. Careful examination should be given to the relative importance to the fields of electric and mass measurements (accuracy, volume of measurements, area of applications etc.) where the changes would take place.
The standards themselves have no value if they are not needed for actual or future applications. For this it is most important for us is to consider the consequences outside of the standards community. Metrologists are trained to deal with different units (the SI units and various practical units), while outside people are not. There is a limited number of scientific experiments were such a level of accuracy is important. Improvement of the reference data is clearly an advantage for outsiders.
We hope that a real study on the practical importance of precision mass and electric measurements with uncertainty below a 100 ppb will be done. Up to now, despite numerous considerations of the redefinition, this question has not been discussed at all, or at least the results of such a discussion have not been made available.
This paper is an extended version of document CCU/05-27, a working document of the 17th meeting of of Consultative Commitee for Units. During discussions there and at the preceding meeting of the CODATA task group on fundamental constants, it became clear for me that the accuracy actually demanded in precision electric measurements are about two orders of magnitude higher than in mass measurements. Still, examination of the problem is necessary, because a question is not only accuracy, but also number of the measurements.
To conclude this paper let us suggest wording for the redefinition of the kilogram: “The kilogram is the mass of a body whose rest energy is equal to the energy of $`\mathrm{299\hspace{0.17em}792\hspace{0.17em}458}10^{27}`$ optical photons in vacuum of wavelength of $`\mathrm{66.606\hspace{0.17em}9311}`$ nanometres.” The explicit indication of the number of the photons is necessary<sup>3</sup><sup>3</sup>3I discussed the issue on the number of the photons with Peter Mohr and it resulted in footnote 2 in . I need to mention that I am impressed by elegancy of the solution there of another problem of the definition, which is avoiding of a multiplication and division of numerical values of constants $`c`$ and $`h`$ in such a combination as $`\{c\}^2/\{h\}`$. The version presented here is somewhat different from that in . From a point of view of relativistic physics it is preferable to speak in terms of the rest energy rather than the rest mass (see, e.g., ).. The kilogram is a macroscopic quantity, while we try to link it to a microscopic object. With a single microscopic object we arrive at the situation when the energy is bigger than the Planck energy or the wave length is shorter than the Planck length.
Concerning the ampere, a redefinition in terms of the elementary charge is rather trivial. Still, it should be mentioned, that, when the definition of the ampere was adopted, its direct realization was possible. Now, we see that there are two units (the ohm and the volt), which we can realize directly and two units (the ampere and the coulomb) which are related to more fundamental quantities but cannot be realized directly. In former time, the ampere was a good choice. Now, if we like to make a practical choice it should favors the volt (potential is more fundamental than resistance), and if we like to make a physical choice we should prefer the coulomb (in particular, because of its educational advantages). There are no advantages for the ampere anymore.
The author is grateful to R. Davis, J. Flowers, P. J. Mohr, L. B. Okun, L. Pendrill, B. N. Taylor and B. M. Wood for useful and stimulating discussions. |
warning/0507/astro-ph0507129.html | ar5iv | text | # Orientation and size of the ‘Z’ in X-shaped radio galaxies
## 1 Introduction
A small subclass of radio galaxies is formed by the X-shaped radio galaxies (XRGs), which all have radio luminosities close to the FR I/II transition of $`10^{25}\mathrm{W}/\mathrm{Hz}`$ at $`178\mathrm{MHz}`$ (Fanaroff & Riley, 1974). These sources show two misaligned pairs of radio lobes of comparable extent (e.g. Ekers et al., 1978; Leahy & Parma, 1992), which have also been referred to as wings and appear as X-shaped structures. After the discovery of this class of sources various mechanisms for their formation have been proposed. Here we give only a short summary, for a more detailed discussion see e.g. Rottmann (2001) or Dennett-Thorpe et al. (2002).
According to the backflow model by Leahy & Williams (1984) jet material is streaming from the hot spots of the primary lobes back towards the host galaxy. It remains collimated until it hits the backflow from the opposite lobe and then expands laterally in a fat disk perpendicular to the radio lobes. It is not clear in this model how the plasma, falling back on the disk, can be diverted to just one side of the primary lobes in order to form the X-shape. Moreover this mechanism can not explain that the secondary lobes extend as far or even farther than the primary lobes and how such an extension can be achieved with subsonic velocities within the life-time of the radio source.
The buoyancy model suggests that the radio lobes have a lower density than the ambient medium, resulting in buoyant forces (Gull & Northover, 1973; Cowie & McKee, 1975), which are thought to bend the lobes towards regions that provide density equilibrium. This model has the same problems as the backflow model to explain the symmetry of XRGs and the extension of the secondary lobes.
Apparently the radio jets, which are assumed to be aligned with the spin of the black hole at their origin (Wilson & Colbert, 1995), have been stable for a long time span before undergoing a short period of reorientation into another stable state (Rottmann, 2001). This can not be explained by a steady precession as has been originally done by Ekers et al. (1978). Gower et al. (1982) applied a precessing jet model to various galaxies, among them NGC 326. Though the wings could be reproduced the inner symmetry is rotated by about $`45^{}`$. Also according to present day deep radio images of NGC 326 a precessing jet seems to be highly unlikely the cause for its shape (Rottmann, 2001). An interesting idea, pointed out by the referee, is whether such a structure could be related to the combination of a precessing jet and the recoil suffered by the merged black hole due to anisotropic emission of gravitational waves (e.g. Blandford 1979). Without going into the details, what would be beyond the scope of this article, it does not seem to be very likely for both the sources we are interested in here. Such a kick would give the black hole, i.e. the source of the post-merger jet, a linear momentum and hence break the symmetry. Though this might explain the apparent curve close to the center in the post-merger jet of NGC 326, generally both jet pairs are seen to be symmetrical about the common center. The symmetry can rather be explained by rapid realignment of the jet due to accretion from a misaligned disk or the coalescence of a supermassive binary black hole (BBH) on time scales less than $`10^7\mathrm{yr}`$ (Rottmann, 2001; Zier & Biermann, 2001, 2002). Such a BBH is formed previously in a merger of two galaxies each of which host one of the supermassive black holes (SMBH) in their center (Begelman et al., 1980). The final stage of the merger is dominated by emission of gravitational radiation which leaves the spin of the resulting SMBH aligned with the orbital angular momentum $`𝐋_{\mathrm{orb}}`$ of the binary (Rottmann, 2001; Zier & Biermann, 2001, 2002; Chirvasa, 2002; Biermann et al., 2002). After a short time a new jet will propagate along the spin axis, i.e. the jet flips from the direction of the spin of the pre-merger SMBH in the direction of $`𝐋_{\mathrm{orb}}`$ (see Fig. 9 in Zier & Biermann 2002). These ideas were also explored by Merritt & Ekers (2002).
In at least two XRGs the ridges of the secondary lobes have been observed to be offset from each other laterally by about their width, hence showing a Z-shaped symmetry about the nucleus (Fig. 1). Because a high angular resolution as well as a rather special aspect angle are necessary to see such an offset it is possible that a Z-symmetry is more common than current detections suggest. To explain this symmetry Gopal-Krishna, et al. (2003) (hereafter G-KBW) propose some modifications to the spin-flip model outlined above: As the captured galaxy spirals to the common center it induces a rotational stream-field in the ambient medium on large scales. If the trajectory of the secondary galaxy passes through the polar regions of the primary, the motion of the ISM bends the original jet into Z-shape before the merger is completed. This means that purely Z-symmetric radio sources (i.e. without X-morphology) might be spotted before evolving to an XRG once the SMBHs have coalesced.
In the present article we will deproject the Z-shaped sources in order to understand their geometry and possible orientation to us. Of interest is the distance in which the old jet is bent into Z-shape since it gives us information about the strength of the jet and the properties of the gas stream in the wake of the secondary galaxy. Ultimately this contributes to our knowledge of the history of a merger between two galaxies.
In the next section we will explain the geometry of the jets and lobes and deduce limits for the involved angles before deriving the expressions for deprojecting the jets. In Sect. 3 we apply our model to the two Z-shaped sources NGC 326 and 3C 52. The results are discussed and compared with other observations in Sect. 4. A summary and conclusions are presented in Sect. 5.
## 2 Jet orientations and Z-shapes
### 2.1 The geometry of ZRGs
In order to observe a source as XRG both, the primary and secondary lobes, i.e. the post- and pre-merger lobes respectively, have to be close to the plane of sky. Both pairs also have to subtend a sufficiently large angle on the sky so that we can distinguish them. Because the pre-merger spin of the primary SMBH and the orbital angular momentum of the merging binary, defining the later post-merger SMBH spin, are not correlated, we expect the angle between them to be large on average. In his thesis Rottmann (2001) used statistical methods to estimate the most likely distribution of the intrinsic angle $`\theta `$ between both pairs of lobes. For this purpose he constructed a theoretical distribution of projected angles $`\theta ^{}`$ for an ensemble of XRGs with intrinsinc angles uniformly distributed between $`\theta _1`$ and $`\theta _2`$. Taking into account various selection effects (the projected angle should not be too small or large; inclination of the jets with respect to the plane of sky should be small; the projected length of the secondary lobe should not be too short) and comparing the theoretical results with the observed distribution he obtains the best fit if $`60^{}\theta 90^{}`$. This is in agreement with our expectation of the intrinsic angle to be large. However, Rottmann points out that the obtained distribution can not reproduce the peak observed at $`50^{}`$ in the distribution of the projected angle. He suggests that a non-uniform distribution of intrinsic angles will improve the fit.
Provided that the direction of the spin of the post-merger SMBH and hence also the direction of the post-merger jet are dominated by the orbital angular momentum of the binary, the directions of the pre- and post-merger jets are uncorrelated. This allows us to imagine both jets as a pair of uncorrelated arrows which we can superpose so that both their centers lie on each other and they enclose an angle in the range $`0^{}\theta 180^{}`$. Without loss of generality we fix one arrow to be aligned with the $`z`$-axis. Asking now for the distribution of the intrinsic angles between both arrows is like looking for the distribution of the pinholes the second arrow pierces through the surface of the unit-sphere. Since both arrows are uncorrelated this is analogue to a uniform distribution of stars projected on the unit-sphere. Therefore the probability to find a star or pinhole in a solid angle element $`d\mathrm{\Omega }=\mathrm{sin}\theta d\theta d\varphi `$ around the coordinates $`(\theta ,\varphi )`$ is
$$p(\theta ,\varphi )=\{\begin{array}{cc}\frac{d\mathrm{\Omega }}{4\pi }\hfill & 0\theta \pi \text{and}0\varphi 2\pi ,\hfill \\ 0\hfill & \text{otherwise}.\hfill \end{array}$$
With the substitution $`u=\mathrm{cos}\theta `$ a uniform distribution over the unit-sphere requires a uniform distribution of both coordinates in the ranges $`1u1`$ and $`0\varphi 2\pi `$. Hence the intrinsic angle $`\theta `$, instead of being uniformly distributed, is distributed according to $`p(\theta )=1/2\mathrm{sin}\theta `$, peaking at $`90^{}`$. The less the orbital angular momentum of the binary dominates the post-merger spin the more the maximum in the distribution of $`\theta `$ will shift to smaller angles and the more the distribution will deviate from a symmetric distribution about the maximum. Qualitatively this seems to be in good agreement with the observed distribution that Rottmann shows in his thesis. A careful comparision of the theoretical with the observed distribution could give a clue about the nature of the formation of XRGs and which component dominates the post-merger spin after the SMBHs have merged due to emission of gravitational radiation. But this requires much more and better data. We just keep the result in mind that the formation mechanism of XRGs is likely to create jet-pairs with large intrinsic angles, while both lobes have to be close to the plane of sky so that we can actually observe the X-shape.
As the secondary galaxy is spiraling into the common center of mass, it will generate a streaming motion in the merger plane due to mass loss and dragging along ISM of the primary galaxy. On large distances the density and velocity of the streaming motion will probably be strong enough to bend the jets into wings. As the the secondary galaxy spirals inwards to smaller distances the power of the jet will become stronger and beyond a certain distance no bending of the jet will be possible. In the following we will refer to the distance where the bending happens as $`r`$. The rotation stream will have some thickness $`2h`$ perpendicular to the merger plane and in the distance $`r`$ be roughly confined to the surface of a cylinder, which is aligned with the orbital angular momentum of the galaxies. The possible orientations of both jet-pairs and the line of sight (LOS) are shown in Fig. 2. The $`z`$-axis is aligned with the orbital angular momentum and hence identified with the post-merger jet. With the bold solid circle we denote the merger plane and with the pair of thin solid and dashed circles we mark the rotation stream with radius $`r`$ and minimum half-height $`h`$ for the three possible secondary pairs of jets. The LOS is perpendicular to the paper plane and pointing straight to the center. It encloses an angle $`\delta _{\mathrm{LOS}}`$ with the merger plane and $`\theta _{\mathrm{LOS}}`$ with the $`z`$-axis so that $`\theta _{\mathrm{LOS}}+\delta _{\mathrm{LOS}}=90^{}`$. If we would move around the $`z`$-axis with constant $`\theta _{\mathrm{LOS}}`$ our LOS would intersect with the cylinder of the rotation stream along the dotted circle. The three possible orientations (a, b, c) of the pre-merger jet relative to the post-merger jet and the LOS are depicted by the bold lines with the wings on the circles following the rotational motion after they have been bent around (here shown in clockwise direction). $`\theta _{\mathrm{jet}}`$ and $`\delta _{\mathrm{jet}}`$ are the intrinsic angle between both jet pairs and the angle from the pre-merger jet to the merger plane respectively. $`\phi `$ denotes the angle between the planes defined by both jet-pairs on the one hand and the LOS and the $`z`$-axis on the other hand and therefore lies in a plane parallel to the merger plane.
If we assume the hight of the cylinder, i.e. the streaming matter, to be of the order of its radius or less, the trajectory of the secondary galaxy necessarily has to pass through the polar regions of the primary in order to bend the jet into Z-shape. Therefore the pre-merger jet and the orbital angular momentum ($`𝐋_{\mathrm{orb}}`$), i.e. the post-merger jet, will be roughly perpendicular to each other. This is consistent with our prediction for the intrinsic angle above and with the explanation of the Z-shape within the framework of the merger model. Hitting the rotation stream the pre-merger jet will be deflected into the wings at some angle which is determined by the relative strengths of the ram pressure of the rotating matter and the jet. For a weak jet the wings will be dragged along with the rotation stream and hence their projected extension would be limited by the radius of the stream. For NGC 326 and 3C 52 extensions of at least $`50`$ and $`100\mathrm{kpc}`$ have been observed respectively. So we can conlude that either the bending happens on such large radii, or a stronger jet is deflected at a smaller angle. Fig. 1 shows that both wings are almost perpendicular to the post-merger jet and because their brightness does not differ very much both wings lie close the plane of sky. Now, a strong pre-merger jet would be deflected at a small angle and thus it would propagate at an angle a little larger than $`\phi `$ in projection on the merger plane. Because we observe both wings close to the plane of sky this means that the initial angle $`\phi `$ must have been quite large and as a consequence we should be able to distinguish the pre-merger jet before the bending from the post-merger jet. But this has not been observed ($`\phi `$ is small) and so we conclude that large radii of the rotating stream are more likely than small deflecting angles. Because $`\phi `$ is small as well as the angles $`\delta _{\mathrm{LOS}}`$ ($`z`$-axis close to plane of sky) and $`\delta _{\mathrm{jet}}`$ ($`\theta _{\mathrm{jet}}`$ intrinsically large) the pre-merger jet will enclose only a small angle with the LOS. This is consistent with the wings being almost perpendicular to the post-merger jet and lying in the plane of sky, see Figs. 1, 2.
Within these limits we have three different possibilities to align both pairs of jets and the LOS relative to each other, see Fig. 2 (angles are positive in clockwise direction):
$`\delta _{\mathrm{LOS}}<\delta _{\mathrm{jet}}`$ The LOS is closer to the merger plane than the pre-merger jet. The upper part of the jet is approaching us, the lower receding (dotted blue/dark-grey lobes).
$`\delta _{\mathrm{LOS}}>\delta _{\mathrm{jet}}`$ Same as a, but with the pre-merger jet closer to the merger plane than the LOS (dashed green/light-grey lobes).
$`\delta _{\mathrm{LOS}}>0`$, $`\delta _{\mathrm{jet}}<0`$ The receding part of the jet and the LOS are in the same hemisphere which is defined by the post-merger jet ($`z`$-axis) as polar axis (solid red/grey lobes). In this case it doesn’t matter whether the LOS or the pre-merger jet is closer to the merger plane.
If we measure the lateral offset of the ridges of the wings we can deproject the jets under assumptions for the bending radius. The thus obtained expressions enable us to put some limits on this radius and to decide wich orientation is most likely. This will be done in the following section.
### 2.2 Deprojection of the Jets
With the observed projected angular extension in the sky of the straight part of the Z-shape, $`\mathrm{\Theta }_\mathrm{p}`$, and the known distance $`D`$ of the galaxy we can determine the angle $`\theta _{\mathrm{jet}}`$ between the pre- and post-merger jets, provided we know the radius $`r`$ where the jet is bent. Fig. 3 shows the situation for orientation a ($`\delta _{\mathrm{LOS}}<\delta _{\mathrm{jet}}`$) of Fig. 2, projected in a plane perpendicular to the merger plane and containing both the LOS and $`z`$-axis. Because $`\phi `$ is assumed to be $`0`$, also the pre-merger jet lies in this plane and the wings are perpendicular to the paper plane. As in Fig. 2 the angles $`\delta `$ and $`\theta `$ in Fig. 3 denote the angle to the merger plane and $`z`$-axis respectively. See Fig. 3 for the meaning of the other quantities. Projecting the sum of the approaching and receding jet $`y_a`$ and $`y_r`$ in the plane of sky yields the wanted relation between $`\mathrm{\Theta }_\mathrm{p}`$ = $`\mathrm{tan}^1y_\mathrm{p}/D`$, the bending radius $`r`$ and the angles $`\theta _{\mathrm{jet}}`$ and $`\theta _{\mathrm{LOS}}`$.
Orientation a: (dotted blue/dark-grey lobes in Fig. 2) With $`a=D\mathrm{tan}\delta _{\mathrm{LOS}}`$ and the definition $`k\mathrm{tan}\delta _{\mathrm{jet}}=h/r`$ we get from Fig. 3
$$\mathrm{tan}\delta _a=\frac{ah}{Dr}=\frac{hy_a}{r},$$
which can be solved for $`y_a`$:
$$y_a=r\left(k\frac{\mathrm{tan}\delta _{\mathrm{LOS}}kr/D}{1r/D}\right).$$
(1)
In the same way we get for the receding part of the jet
$$y_r=r\left(k\frac{\mathrm{tan}\delta _{\mathrm{LOS}}+kr/D}{1+r/D}\right),$$
(2)
and hence for their sum along the $`z`$-axis in the limit that the distance to the galaxy is much larger than the radius of the rotation stream
$$\begin{array}{cc}\hfill y_z& =y_a+y_r=2r\left(k\frac{\mathrm{tan}\delta _{\mathrm{LOS}}k(r/D)^2}{1(r/D)^2}\right)\hfill \\ & \underset{Dr}{\overset{}{}}2r(k\mathrm{tan}\delta _{\mathrm{LOS}}).\hfill \end{array}$$
(3)
Finally, after projecting $`y_z`$ into the plane of sky we obtain for the projected lateral offset of the ridges
$$y_\mathrm{p}=D\mathrm{tan}\mathrm{\Theta }_\mathrm{p}=y_z\mathrm{cos}\delta _{\mathrm{LOS}}=2r\frac{\mathrm{sin}(\delta _{\mathrm{jet}}\delta _{\mathrm{LOS}})}{\mathrm{cos}\delta _{\mathrm{jet}}}.$$
(4)
Proceeding in the same way we obtain for the other possible orientations:
Orientation b: (dashed green/light-grey lobes) Like a, with the difference of the angles $`\delta _{\mathrm{jet}}`$ and $`\delta _{\mathrm{LOS}}`$ changing sign:
$$y_\mathrm{p}=2r\frac{\mathrm{sin}(\delta _{\mathrm{LOS}}\delta _{\mathrm{jet}})}{\mathrm{cos}\delta _{\mathrm{jet}}}.$$
(5)
Orientation c: (solid red/grey lobes) In notation of Fig. 3 like orientation b with the sign of $`\delta _{\mathrm{jet}}`$ changed so that in the following expression the angle varies in the range $`0<\delta _{\mathrm{jet}}<90^{}`$:
$$y_\mathrm{p}=2r\frac{\mathrm{sin}(\delta _{\mathrm{LOS}}+\delta _{\mathrm{jet}})}{\mathrm{cos}\delta _{\mathrm{jet}}}.$$
(6)
As we pointed out in Sect. 2.1 we expect $`\delta _{\mathrm{LOS}}`$ to be small because the post-merger jet is lying close to the plane of sky and $`\delta _{\mathrm{jet}}`$ is small because the intrinsic angle between both jets is expected to be large. The latter is larger than zero though, because otherwise we could not observe the offset of the wings.
## 3 Application on NGC 326 and 3c 52
In this section we apply the results obtained above on the two observed Z-shaped XRGs to derive limits for the bending radius and to find the possible orientations. Afterwards we will check whether the jet can still be bent into the wings at the obtained distances.
### 3.1 Bending radius and orientation
Given the radius $`r`$ for the rotation field we can solve the above Eqs. (4) to (6) for $`\delta _{\mathrm{LOS}}`$ and plot it in dependency of $`\delta _{\mathrm{jet}}`$ in order to find the most likely orientation that minimizes both angles. The expressions we get are
$$\delta _{\mathrm{LOS}}=\{\begin{array}{cc}\delta _{\mathrm{jet}}\mathrm{sin}^1\left(\frac{y_\mathrm{p}}{2r}\mathrm{cos}\delta _{\mathrm{jet}}\right),\hfill & \text{orientation }\text{a}\hfill \\ \delta _{\mathrm{jet}}+\mathrm{sin}^1\left(\frac{y_\mathrm{p}}{2r}\mathrm{cos}\delta _{\mathrm{jet}}\right),\hfill & \text{orientation }\text{ }\text{b}\hfill \\ \delta _{\mathrm{jet}}+\mathrm{sin}^1\left(\frac{y_\mathrm{p}}{2r}\mathrm{cos}\delta _{\mathrm{jet}}\right),\hfill & \text{orientation }\text{ }\text{c}\text{.}\hfill \end{array}$$
(7)
The lateral offset $`y_\mathrm{p}`$ of the ridges we take from observations of NGC 326 and 3C 52.
NGC 326: The distance to this source is about $`160\mathrm{Mpc}`$. Therefore the projected angular size of the middle part of the ‘Z’, $`\mathrm{\Theta }_\mathrm{p}=20^{\prime \prime }`$, translates into a projected length of about $`y_\mathrm{p}=16\mathrm{kpc}`$. Schiminovich et al. (1994) and Charmandaris et al. (2000) have detected dense clouds that contain both H i and molecular gas in Cen A. With an assumed distance of $`3.5\mathrm{Mpc}`$ to Cen A they locate the gas at a radius of about $`10\mathrm{kpc}`$ to the center. If we use this as the radius $`r`$ of the rotation field and plot $`\delta _{\mathrm{LOS}}`$ in dependency of $`\delta _{\mathrm{jet}}`$ (Eq. (7)) we get the thin curves in Fig. 4. Again the dotted blue/dark-grey line represents the solutions for orientation a, the dashed green/light-grey line for b and the solid red/grey line for c. Because the intrinsic angle is expected to be larger than $`60^{}`$, $`\delta _{\mathrm{jet}}`$ is less than $`30^{}`$ (left from the shaded area). This limit excludes orientation a which has a minimum of about $`\delta _{\mathrm{jet}}38^{}`$. The smallest pair of angles in configuration b is $`0^{}`$ for $`\delta _{\mathrm{jet}}`$ and $`\delta _{\mathrm{LOS}}53^{}`$. Such a large angle violates the condition that primary lobes have to be close to the plane of sky in order to detect an X-shape (region below the shaded area). Hence also this orientation is ruled out and the only remaining possibility is c, where the LOS and the approaching part of the jet are in different hemispheres relative to the post-merger jet. To minimize both angles we get $`\delta _{\mathrm{LOS}}=\delta _{\mathrm{jet}}23.6^{}`$, with not too much range left for them on curve c, if we assume the shaded areas outside the inner rectangle as not permitted. Being pushed to the limit for a radius of $`10\mathrm{kpc}`$, the situation is much less restrictive if we allow for a larger radius. The thick curves in Fig. 4 show the results if the jet is bent in a distance of $`r=30\mathrm{kpc}`$. On branch c both angles are much smaller, with an upper limit of about $`15^{}`$ for $`\delta _{\mathrm{jet}}`$. For the larger radius also the other orientations a and b are possible. These two branches appear in the allowed, not-shaded region of Fig. 4 only for $`r14\mathrm{kpc}`$.
3C 52: For 3C 52, in a distance of about $`1\mathrm{Gpc}`$, with $`\mathrm{\Theta }_\mathrm{p}=10^{\prime \prime }`$ the projected offset of the wings is $`y_\mathrm{p}=50\mathrm{kpc}`$. This is five times the length of the assumed radius of $`10\mathrm{kpc}`$. But for Eq. (7) to have a solution, the argument of $`\mathrm{sin}^1`$ has to be less than $`1`$ ($`\delta _{\mathrm{jet}}`$ varies in a range where the cosine is positive), leaving us with the condition
$$\mathrm{cos}\delta _{\mathrm{jet}}\frac{2r}{y_\mathrm{p}}.$$
For $`r=10\mathrm{kpc}`$ this is fulfilled only if $`\delta _{\mathrm{jet}}`$ is larger than $`66^{}`$, in contradiction with the intrinsic angle between the jet-pairs to be large. So for this small radius none of the orientations is within the allowed rectangle of Fig. 5. Only for $`r`$ larger than $`y_\mathrm{p}/2=25\mathrm{kpc}`$ the full range between $`0^{}`$ and $`90^{}`$ becomes mathematically possible for $`\delta _{\mathrm{jet}}`$. For this radius the matching angles of the LOS are too large and none of the orientations provides a satisfying solution, see the thin lines in Fig. 5. A larger bending radius can solve the problem again. If we gradually increase the radius, first branch c offers physically reasonable results for both angles ($`r=24\mathrm{kpc}`$) before the other two branches enter the allowed region for $`r>44\mathrm{kpc}`$. The thick lines in Fig. 5 show the results for $`r=75\mathrm{kpc}`$, with all orientations being possible.
These results show that the bending of the lobes into the wings occurs already much earlier during the merger in distances of more than $`20`$ to maybe even $`100\mathrm{kpc}`$. The smaller the radius is, the more likely the jets have orientation c relative to us, i.e. the receding pre-merger jet and the LOS are in the same hemisphere that is defined by the post-merger jet (axis of orbital angular momentum) as polar axis.
### 3.2 Bending the jet
If the bending of the jet happens at the distances suggested above, we have to verify that the rotating stream is able to exert a large enough pressure on the jet at these radii to bend it into Z-shape as it has been shown by G-KBW for a radius of $`10\mathrm{kpc}`$. The bending of the jet is described by Euler’s equation, which for a steady state flow reads
$$\rho _{\mathrm{jet}}(𝐯_{\mathrm{jet}}\mathbf{})𝐯_{\mathrm{jet}}=\mathbf{}P_{\mathrm{ISM}},$$
(8)
with the gradient of the ram pressure of the ISM being applied transverse to the beam (O’Donoghue et al., 1993). The velocity of the jet is assumed to change by order of itself over the bending scale $`l_{\mathrm{bend}}`$, so that the left hand side of the equation can be written as $`\rho _{\mathrm{jet}}v_{\mathrm{jet}}^2/l_{\mathrm{bend}}`$. As O’Donoghue et al. point out a pressure gradient that provides a centripetal acceleration $`v_{\mathrm{jet}}^2/l_{\mathrm{bend}}`$ around a curve with radius $`l_{\mathrm{bend}}`$ gives the same result for the left hand side of Eq. (8). The pressure gradient due to ram pressure on the right hand side, $`\mathbf{}P_{\mathrm{ISM}}`$, can be approximated by $`\rho _{\mathrm{ISM}}v_{\mathrm{ISM}}^2/l_{\mathrm{press}}`$ if $`v_{\mathrm{ISM}}`$ is the relative velocity between the ISM and the galaxy, i.e. the rotation velocity. $`l_{\mathrm{press}}`$ is the length scale over which the ISM exerts the ram pressure on the jet and is taken to be the radius of the jet ($`R_{\mathrm{jet}}`$) at the bending point ($`r`$). Hence we can rewrite Euler’s equation in the approximation for jet flows as
$$\rho _{\mathrm{jet}}\frac{v_{\mathrm{jet}}^2}{l_{\mathrm{bend}}}=\rho _{\mathrm{ISM}}\frac{v_{\mathrm{ISM}}^2}{l_{\mathrm{press}}}.$$
(9)
Assuming that the clouds in Cen A represent the properties of the ISM reasonably well the following reference values can be used: $`n_{\mathrm{ISM}}=\rho _{\mathrm{ISM}}/(1.4m_\mathrm{p})=0.1\mathrm{cm}^3`$ and $`v_{\mathrm{ISM}}=100\mathrm{km}/\mathrm{s}`$, with $`l_{\mathrm{press}}=R_{\mathrm{jet}}1\mathrm{kpc}`$. The jet is assumed to be semirelativistic ($`v_{\mathrm{jet}}=10^5\mathrm{km}/\mathrm{s}`$) and made of ordinary proton-electron plasma that is bent over a scale of $`l_{\mathrm{bend}}=10\mathrm{kpc}`$. Thus for the density of a jet that is bent into Z-shape by ram pressure a density of $`n_{\mathrm{jet}}=10^6\mathrm{cm}^3`$ is obtained.
Now we have to scale up the values obtained at $`r_1=10\mathrm{kpc}`$ to larger distances $`r_2`$ such that Eq. (9) is still fulfilled. In the following the additional indices $`1`$ and $`2`$ refer to the quantities in distance $`r_1`$ and $`r_2`$ respectively. If the half-opening angle of the jet is $`\vartheta `$, then its spherical surface perpendicular to the direction of propagation in a distance $`r`$ is $`A=2\pi r^2(1\mathrm{cos}\vartheta )`$. Taking the flux of momentum along the jet $`\rho _{\mathrm{jet}}v_{\mathrm{jet}}^2A`$ to be constant we find
$$\rho _{\mathrm{jet}_2}v_{\mathrm{jet}_2}^2=\rho _{\mathrm{jet}_1}v_{\mathrm{jet}_1}^2\left(\frac{r_1}{r_2}\right)^2$$
(10)
With this expression, Eq. (9), the relation $`R_{\mathrm{jet}_2}=R_{\mathrm{jet}_1}r_2/r_1`$ and the assumption that $`l_{\mathrm{bend}}`$ scales linearly with $`r`$ we finally obtain for the density at $`r_2`$
$$\rho _{\mathrm{ISM}_2}=\rho _{\mathrm{ISM}_1}\left(\frac{r_1}{r_2}\right)^2\left(\frac{v_{\mathrm{ISM}_1}}{v_{\mathrm{ISM}_2}}\right)^2.$$
(11)
Hence, to be able to bend the jet in, say $`r_2=50\mathrm{kpc}`$, the density of the ISM can be 25 times less than at $`r_1=10\mathrm{kpc}`$. For an annulus with a width from $`50`$ to $`60\mathrm{kpc}`$ and of $`10\mathrm{kpc}`$ height that density corresponds to a total mass of $`3.4\times 10^9M_{}`$. For a velocity of about $`200\mathrm{km}/\mathrm{s}`$, as has been observed in such distances (see Sect. 4) the mass is reduced by another factor of four, so that it is less than $`10^9M_{}`$. At $`r_1`$ the mass in a ring with inner and outer radius $`7.5`$ and $`12.5\mathrm{kpc}`$ respectively and a hight of $`5\mathrm{kpc}`$ is $`4\times 10^9M_{}`$. Thus about the same mass that is required in a rotating stream with radius $`10\mathrm{kpc}`$ to bend the jet into a Z-shape is sufficient to bend the jet at much larger radii. The required mass and velocity at such distances is in agreement with observations (see next section). Hence our results show that the ram pressure of the rotating gas in a distance of $`50\mathrm{kpc}`$ is indeed strong enough to bend the jet in Z-shape, as is required by the geometrical arguments above.
## 4 Discussion
### 4.1 Geometry and dynamics of the jet
In the previous sections we showed that the bending of the pre-merger jet into Z-shape, as proposed by G-KBW within the merger model, must happen on distances larger than the $`10\mathrm{kpc}`$ that they have suggested. Because the half-thickness of the rotating gas stream will not be larger than its radius, it has to pass through the polar regions of the primary galaxy in order to bend the jet and thus pre- and post-merger jets are approximately perpendicular to each other. This corresponds to the maximum of the distribution of the intrinsic angle ($`\theta _{\mathrm{jet}}=90^{}`$) between both pairs of lobes in XRGs, if the directions of their propagation are uncorrelated. But this is exactly what we expect in the merger model for XRGs, if the spin of the post-merger SMBH is dominated by the orbital angular momentum of the binary, as pointed out in Sect. 2.1. In ZRGs a larger bending radius is in favour of a larger angle $`\theta _{\mathrm{jet}}`$ between the jets. For example $`y_\mathrm{p}=50\mathrm{kpc}`$ has been observed in 3C 52. Trying to minimize the the half-height $`h`$ for $`r=10\mathrm{kpc}`$, i.e. maximizing $`\theta _{\mathrm{jet}}`$, we obtain with orientation a $`h=y_\mathrm{p}/2=2.5r`$ at $`\theta _{\mathrm{jet}}=21.8^{}`$ and $`\delta _{\mathrm{LOS}}=0`$. In case b there is no solution at all for $`y_\mathrm{p}2r`$ and for c we get $`h=2.3r`$ at $`\theta _{\mathrm{jet}}=25.6^{}`$ and $`\delta _{\mathrm{LOS}}=23.6^{}`$. This is in direct contradiction with a slim gas stream and with the assumption of a large angle between both jet pairs and could be solved with a larger bending radius (Sect. 3.1).
As G-KBW estimated, the ram pressure of the rotation field at $`10\mathrm{kpc}`$ radius is strong enough to bend a jet with power close to the FR I/II transition into Z-symmetry. A stronger jet would not be much deflected by the rotating gas-stream. Because the wings are in the plane of sky and extend almost perpendicular from the post-merger jet, the pre-merger jet would also have to be close to the plane of sky and hence distinguishable from the primary jet, as we pointed out at the end of Sect. 2.1. However, this has not been observed and thus also the appearance and morphology of the source argue for weaker jets which can be deflected by a large angle. But the jet should not be too weak as well, because then it would be just dragged along with the circular gas stream and hence have a maximum projected extension of the bending radius. Depending on the angle of the LOS to the merger plane we might see the curvatures of the wings following the circular motion and exhibiting inversion symmetry relative to the center as depicted in Fig. 2 (of course jets deflected at smaller angles also show inversion symmetry, but they can more easily deviate from that due to interaction with the ambient medium). In case of NGC 326 and 3C 52 this means a bending radius of at least $`50`$ and $`100\mathrm{kpc}`$ respectively. It is more likely that a weak jet would be bent at smaller radii by the inspiralling gas stream, where the jet has a larger power and hence would probably be deflected but not dragged along with the rotational motion. In Sect. 3.2 we showed that a gas stream with a similar mass content as that required for bending at $`10\mathrm{kpc}`$ is able to bend the same jet at larger distances of about $`50\mathrm{kpc}`$. After spiralling further inside to smaller radii the secondary galaxy will have suffered more stripping of material and the gas stream becomes weaker while the jet becomes stronger. The requirements for the stream at $`50\mathrm{kpc}`$ are in good agreement with observations (Sect. 4.2). Hence the conclusion by G-KBW that the bending of jets into Z-shape happens at a jet-power close to the FR I/II transition holds also at larger radii.
### 4.2 Evidence for the required streams
In the present model we assume XRGs and ZRGs to be merger products. As such it is expected that the secondary galaxy, while spiralling inwards, induces a stream of gas and dust on scales of tens of $`\mathrm{kpc}`$ in the primary galaxy. This stream is due to matter of the primary galaxy dragged along by the secondary as well as matter stripped off from the secondary galaxy. Now looking for such streams in other sources shows that they have been observed in various objects. These streams are always related to a merger between two galaxies. This is also in very good agreement with numerical simulations of mergers which produce tidal tails and streams with the properties required in our model ($`\rho ,v,r`$) and hence lends strong support to it. In the following we compile some information of these sources. We use $`H_0=70\mathrm{km}/\mathrm{s}\mathrm{Mpc}`$ and scaled the values from the cited papers accordingly. A summary is given in Table 1.
Recent H i observations of M 31 by Thilker et al. (2004) show a circumgalactic cloud population in $`50\mathrm{kpc}`$ distance. These clouds are moving with a velocity component along the LOS of $`v_{\mathrm{syst}}^{+128}{}_{215}{}^{}\mathrm{km}/\mathrm{s}`$, matching the velocity extent of the disk of M 31. Though the H i content of the halo cloud population is estimated to be only $`37\times 10^7M_{}`$, it might trace more substantial amounts of ionized gas and dark matter. As an obvious source of the high-velocity H i gas the authors give tidal stripping from mergers in agreement with Brown et al. (2003), who relate the young halo to a major merger or several minor mergers.
Braun et al. (2003) conducted a H i survey and found significant positional offsets exceeding $`10\mathrm{kpc}`$ in some of the sources, what they attribute to tidal interaction. While the mean observed offset of H i is about $`66\mathrm{kpc}`$, in NGC 1161 H i is observed in $`110\mathrm{kpc}`$ distance to the center at speeds that differ by $`200\mathrm{km}/\mathrm{s}`$ from the systemic velocity. The H i mass is estimated to be about $`1.8\times 10^9M_{}`$.
The H i detected by van Gorkom et al. (1986) in NGC 1052, an active elliptical galaxy, is distributed in a disk that extends $`2025\mathrm{kpc}`$ along the minor axis and is seen almost edge on. The gas has a circular velocity of $`200\mathrm{km}/\mathrm{s}`$ that is roughly constant with radius. The H i mass is about $`5.7\times 10^8M_{}`$ and shows an outer structure that resembles tidal tails, which van Gorkom et al. attribute to a merger about $`10^9\mathrm{yr}`$ ago.
In another early type galaxy, IC 5063 which has a Seyfert 2 nucleus, Morganti et al. (1998) detect H i that to first order is distributed in disk of about $`28\mathrm{kpc}`$ radius. This disk is oriented very similar to a system of dust lanes and in projection rotates at $`240\mathrm{km}/\mathrm{s}`$ with an H i mass of $`4.2\times 10^9M_{}`$. Optical data from previous observations (e.g., Danziger et al., 1981) revealed ionized gas that also lies in a disk, which is extending to $`14.4\mathrm{kpc}`$. The faint structures in the outer regions could be tidal arms and the origin of H i is most likely a merger between spiral galaxies as in the other sources.
Among the five elliptical galaxies that Oosterloo et al. (2002) observed they detected $`2.3\times 10^9M_{}`$ of H i in NGC 3108 that is distributed in a disk-like structure perpendicular to the optical major axis of the galaxy and extends to $`30\mathrm{kpc}`$. They assume an inclination of $`70^{}`$ and thus obtain for the rotation velocity $`290\mathrm{km}/\mathrm{s}`$, which appears to be constant from $`1\mathrm{kpc}`$ to the very outer regions. Within the central $`1^{}`$ the disk seems to have a hole that is filled by a disk seen in emission from ionized gas. While the boxy outer isophotes indicate that also NGC 3108 has undergone a major merger, the regular and settled appearance of the disk suggest that it happened some $`10^9\mathrm{yr}`$ ago.
Cen A is a giant elliptical galaxy with an active nucleus that shows strong radio lobes on both sides of a dust lane which is aligned with the minor axis (Clarke et al., 1992) and a warped gaseous disk which is seen in optical and H i emission (Dufour et al., 1979; van Gorkom et al., 1990). Schiminovich et al. (1994) find the H i morphology to be closely correlated with diffuse shells seen in the optical range (Malin et al., 1983) and estimate the total mass in the shells to be $`1.5\times 10^8M_{}`$. The position-velocity (PV) plot of H i in the shells is well fitted by a single ring with uniform rotation velocity ($`250\mathrm{km}/\mathrm{s}`$) and the rotation axis being roughly perpendicular to that of the inner H i disk. The rotation curve is flat out to $`15^{}`$, what corresponds to a radius of $`34\mathrm{kpc}`$ for $`D=8\mathrm{Mpc}`$, using a redshift of $`z=0.001825`$ (note that Schiminovich et al. used $`D=3.5\mathrm{Mpc}`$, and hence $`r10\mathrm{kpc}`$, what we initially used as bending radius in Sect. 3.1). As possible explanation for the misalignment bewteen the rotation axis of the H i in the shells and the disk they suggest a merger which is not proceeding in the plane of Cen A, and differential precession of the stripped material. Later Charmandaris et al. (2000) suggested that the morphology of the shells is a combination of both, phase wrapping of tidal debris on nearly radial orbits (Quinn, 1984) and spatial wrapping of matter in thin disks for mergers with large angular momentum (Dupraz & Combes, 1987; Hernquist & Quinn, 1989). They also detect CO emission in the shells and associate it with the H i gas which shows the same velocity signatures and deduce a $`\text{H}_2`$-mass of $`4.3\times 10^7M_{}`$.
In NGC 5266, a bright E4 galaxy, Morganti et al. (1997) find H i gas distributed in two perpendicular disks. The inner disk is aligned with the dust lane and fills the hole of $`2^{}`$ diameter of the outer disk, which extends to $`4^{}`$ ($`51\mathrm{kpc}`$). For the rotation velocity of this disk they obtain $`270\mathrm{km}/\mathrm{s}`$, which is constant in radius, and for the total H i mass $`1.2\times 10^{10}M_{}`$. They point out that the H i distribution is similar to that observed in Cen A, with most of H i being associated with the dust lane but a different kinematical behaviour at larger distances and forming a ring that is roughly perpendicular to the dust lane. This is unlike than in most polar-ring and dust-lane galaxies. The outer parts of H i, extending to $`100\mathrm{kpc}`$, could be a settled ring but are rather interpreted as tidal tails that formed during a merger of two gas-rich spiral galaxies. Numerical simulations by Hibbard & Mihos (1995) (see later in this section) have shown that after gas piling up in the center and fuelling a star burst, the fraction that remains at larger distances in tidal tails will settle in a disk or a ring, depending on the initial conditions. Earlier observations in the optical range indicate that the kinematic axes of stars and gas are orthogonal, with the gas in the dust lane rotating about the optical major axis (Caldwell, 1984). Goudfrooij et al. (1994) could show that ionized gas lies in a ring that is clearly associated with the dust ring detected. In CO observations Sage & Galletta (1993) found that the molecular gas is also distributed in a ring that is co-rotating with the ionized gas at velocities of $`270300\mathrm{km}/\mathrm{s}`$ within $`1^{}`$ ($`13\mathrm{kpc}`$) and has a mass in $`\text{H}_2`$ of $`2.7\times 10^9M_{}`$.
Recently Swaters & Rubin (2003) observed the polar ring galaxy NGC 4650A and found the velocities of both, stars and gas in the polar ring component to be closely correlated. The ring is seen close to edge on and rotates close to its outer parts ($`10\mathrm{kpc}`$) with a velocity of $`120\mathrm{km}/\mathrm{s}`$. The flatness of the rotation curve suggests that the gas and stars are rather distributed in a disk than a norrow ring. This is supported by the results of Bekki (1998) who simulated a dissipational polar merger of two disk galaxies of about the same mass. In his numerical experiments he could reproduce polar ring galaxies, with the intruding galaxy being transformed into a S0-like host and the victim into a narrow polar ring. As standard model Bekki used $`10^{10}M_{}`$ and $`10\mathrm{kpc}`$ for the disk mass and radius respectively. These are quite small values and we scaled the model to masses in the range of $`10^{1112}M_{}`$. Then crude estimates of the size, velocity and mass of the polar ring result in the ranges that are required by our model, i.e. $`30100\mathrm{kpc}`$, $`100300\mathrm{km}/\mathrm{s}`$ and $`1/100M_{\mathrm{disk}}`$ respectively. Bekki notes that he seemed to have failed to reproduce annular polar ring galaxies like NGC 4650A, unless the annular ring component with an apparent hole in the center is a part from the galactic disk, in good agreement with the conjecture by Swaters & Rubin (2003) based on their observations.
In other numerical simulations of mergers Bournaud et al. (2004) compared their results with the kinematics of tidal tails in interacting galaxies. Their main goal is to distinguish whether apparent massive condensations close to the tips of the tails are real or caused by projection effects. The PV plots can qualitatively distinguish both possibilities: Either the difference between the systemic velocity and the tidal tail increases with position along the tail, reaching a maximum at its end, or it passes through a maximum before it decreases and even turns back to closer positions at smaller speeds, thus following a loop. We are more interested in the latter case which can give us information about the azimuthal component of the velocity in the tail, that we assume to be edge on. The matter in the tail is not streaming along its spatial extension and also has a velocity component perpendicular to the tangential one. As the velocity component aligned with the LOS is measured along the tail with increasing distance to the center, the velocity increases. Before the tip is reached the velocity assumes a maximum when the velocity in the tail is aligned with the LOS. As the observer moves on to the tip where only the azimuthal component is aligned with the LOS, the velocity decreases. Comparison with observations show that IC 1182 and the northern tail of Arp 105 fall in this category with a circular velocity of $`100\mathrm{km}/\mathrm{s}`$ in about $`60\mathrm{kpc}`$ distance from the center and $`200\mathrm{km}/\mathrm{s}`$ at $`100\mathrm{kpc}`$ respectively. The corresponding H i mass at the end of the tails they estimate to be about $`1.8\times 10^{10}M_{}`$ and $`6.5\times 10^9M_{}`$ respectively.
NGC 7252 is a late stage merger of two gas-rich disk galaxies (e.g., Dupraz et al., 1990; Wang et al., 1992; Hibbard et al., 1994). The morphology and kinematical properties of the tidal tails have been tried to be reproduced with numerical simulations by Hibbard & Mihos (1995). The PV plot in a previous paper by Hibbard et al. (1994) shows that the tails have a maximum velocity along the LOS of $`100\mathrm{km}/\mathrm{s}`$ at $`60\mathrm{kpc}`$ before turning back in a loop in the PV-plane. The H i mass in the tails is estimated to $`4\times 10^9M_{}`$. The best-fit model of Hibbard & Mihos (1995) succeeds in reproducing both the observed spatial morphology and the velocity structure of the tails. By the time the simulated merger fits best the observations ($`8.3\times 10^8\mathrm{yr}`$) about $`4\times 10^9M_{}`$ H i is still in the tails, half of which will fall back to radii $`20\mathrm{kpc}`$ during the next $`4.3\times 10^9\mathrm{yr}`$. We used the angular momentum – radius plot to compute the circular velocity of the matter falling back to $`35\mathrm{kpc}`$ from the range between $`65`$ and $`130\mathrm{kpc}`$ and obtain a velocity range from $`100`$ to $`200\mathrm{km}/\mathrm{s}`$ respectively. Hence for a long time a stream of gas strong enough to bend the jet is maintained in a distance of about $`35\mathrm{kpc}`$. Some of the falling back material will form loops or shells and other structures. In more recent simulations Mihos et al. (1998) examined the effect of dark matter halo potentials on the morphology and kinematics of tidal tails. They basically confirm the previous results from Hibbard & Mihos (1995) which are relevant for our purposes. From the mass of the initial H i disk $`1020\%`$ and about $`1520\%`$ of the initial total disk mass ends as stellar mass in the tidal tails. Scaled to the Milky Way this is about $`6.6\times 10^9`$ and $`7.2\times 10^9M_{}`$ respectively that circulates at velocities of $`220\mathrm{km}/\mathrm{s}`$ at distances more than $`100\mathrm{kpc}`$ around the center.
Both, the observations of galaxies showing signatures of mergers about $`10^9\mathrm{yr}`$ ago, and simulations of merging galaxies are in good agreement with our model. With the properties shown in Table 1 the merger induced gas stream is able to bend a jet with power close to the FR I/II transition into Z-shape on distances larger than $`30\mathrm{kpc}`$.
## 5 Conclusions
In the present article we are investigating in the possible orientation and geometry of Z-shaped radio galaxies (ZRGs). The formation of objects of this class has been explained by G-KBW within the framework of a merger model. As the secondary galaxy spirals in towards the common center of mass it generates a rotational velocity field of gas and dust in its wake that is made up by matter stripped off from the secondary galaxy and matter dragged along from the ambient medium of the primary galaxy. If the trajectory passes through the polar region of the primary galaxy, its jet can be bent into a Z-symmetric shape, depending on the relative pressure between the gas stream and the jet. Thus for these sources the spin of the primary SMBH and the orbital angular momentum $`𝐋_{\mathrm{orb}}`$ of the binary are necesserily roughly perpendicular to each other. Provided that the spin of the post-merger SMBH is dominated by $`𝐋_{\mathrm{orb}}`$ and that jets are aligned with the spin of the SMBH at their base, consequently the pre- and post-merger jets will also be perpendicular to each other. While this should be true for every ZRG, this holds for XRGs only on average since the direction of the $`𝐋_{\mathrm{orb}}`$ and the pre-merger spin of the SMBH are uncorrelated, as explained in Sect. 2.1 (Rottmann, 2001; Zier & Biermann, 2001, 2002). We used this argument for large angles between the jet pairs and the assumption that the post-merger lobes are close to the plane of sky, because they are similarly luminous, to deproject the jets with respect to the LOS. Applied to the ZRGs NGC 326 and 3C 52 our results showed that to fulfil both conditions the bending of the jet must happen on scales between about $`30`$ and $`100\mathrm{kpc}`$. One important result is that the possibility to see the source at one of the possible three orientations, indicated in Fig. 2, depends on the bending radius (Sect. 3.1). To maintain a large angle between the jet pairs and a primary jet close to the plain of sky we used the following limits for the angles: $`\theta _{\mathrm{jet}}=90^{}\delta _{\mathrm{jet}}>60^{}`$ and $`\delta _{\mathrm{LOS}}<35^{}`$, which we think to be quite conservative. For very small bending radii $`r`$ no solution is in the allowed range (white rectangle in Figs. 4 and 5). As we increase $`r`$ first orienation c, where the LOS and the approaching part of the pre-merger jet are in different hemispheres that are defined by the post-merger jet as polar axis, appears in the allowed region for $`ry_\mathrm{p}/2`$ (i.e. $`8`$ and $`25\mathrm{kpc}`$ for NGC 326 and 3C 52 respectively). If we further increase $`r`$ both the other orientations also appear in the allowed rectangle at roughly the same radius $`ry_\mathrm{p}\sqrt{3}/2`$ ($`14`$ and $`44\mathrm{kpc}`$ for NGC 326 and 3C 52 respectively). Knowing the correct orientation we also know the sense of rotation, i.e. $`𝐋_{\mathrm{orb}}`$, and consequently the direction of the spin of the post-merger SMBH (Chirvasa, 2002; Zier & Biermann, 2002; Biermann et al., 2002). In future work this could be compared with the spin inferred from circular polarization measurements at cm-wavelengths, as has been dicussed by Enßlin (2003) and suggested by G-KBW.
The radius of the bending will depend on the relative pressure between the jet and the gas stream. In Sect. 3.2 we showed that for a jet with a power close to the FR I/II transition this happens on scales of $`50\mathrm{kpc}`$, in agreement with the results from the geometrical arguments above. Thus the conclusion by G-KBW that the jet is bent at a power close to the FR I/II transition is also valid at radii in a range of $`30`$ to $`100\mathrm{kpc}`$. In fact in Sect. 4.1 we showed that neither very strong nor weak jets are in agreement with the geometry of ZRGs.
ZRGs can not be explained by the rapid jet reorientation from instabilities in an accretion disk, what is also considered as one possible formation mechanism of XRGs (Dennett-Thorpe et al., 2002), and might not be easily reconciled with the observed distribution of angles between the jet pairs. Since ZRGs are a subset of XRGs their existence strongly supports the merger model in favour of the accretion model as formation mechanism of XRGs. In this context our result that the angles between both jet pairs have to be large in ZRGs and are large on average in XRGs, as has been observed (Rottmann, 2001), further strengthen the merger model.
This in turn will have some impact on the “final parsec problem”, i.e. the conjecture that after a merger of two galaxies the merging of the SMBHs stalls in a distance of about $`0.01`$ to $`1\mathrm{pc}`$ (Begelman et al., 1980). At this distance dynamical friction is inefficient and gravitational radiation still unimportant so that slingshot ejection of individual stars provides the only mechanism to extract energy and angular momentum from the binary (Zier & Biermann, 2001). If there are no stars with small enough pericenters as to interact with the binary (i.e. loss-cone depletion) the SMBHs are prevented from further merging. But contrary to that the existence of XRGs and ZRGs shows that the binary has merged. In ZRGs they probably merge on timescales of some $`10^8\mathrm{yr}`$ after the bending of the jet in a distance of $`50\mathrm{kpc}`$. While in XRGs the binary could have stalled for a long time on scales of $`1\mathrm{pc}`$, in ZRGs the merger must have been completed after the bending in a time short enough to maintain the rotational gas stream and that we are still able to see the bended lobes, which are fading away and undergo spectral ageing (Rottmann, 2001). Thus, in a way, the bending starts a stop watch for the rest of the merger.
## Acknowledgements
I would like to thank Gopal-Krishna for helpful and valuable discussions on the rotational motion of the ISM.
I am happy to have the opportunity to thank the RRI for the generous support and very kind hospitality.
I also would like to thank Wolfram Krülls for his valuable comments to improve this manuscript.
This research has made use of the NASA/IPAC Extragalactic Database (NED) which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration. |
warning/0507/hep-th0507051.html | ar5iv | text | # 1 Introduction
## 1 Introduction
Recently Hitchin proposed to consider the generalized geometry where the tangent bundle $`TM`$ is replaced by the tangent plus cotangent bundle $`TMT^{}M`$. In different context and by different authors it has been pointed out that there is string theory origin of the generalized geometry based on $`TMT^{}M`$. Indeed many concepts of generalized geometry have their string theory counterpart. Insprired by this relation we would like to make one step further and ask about possible relevant geometric concepts for $`p`$-brane theories. In this note we propose that for $`p`$-brane theories the relevant geometry is based on $`TM^pT^{}M`$ bundle.
The paper consists of two results. First of all we generalize Alekseev-Strobl observation to the case of generic $`p`$-brane theory. Namely we associte to anomaly free algebra of $`p`$-brane currents an “isotropic” involutive subbundle $`L`$ of $`T^pT^{}`$. This algebra can be regarded as an algebra of first class constraints for some gauge theory. In particular we consider a few interesting examples of such gauge theories, namely topological $`p`$-brane theories. We study the compatibility condition between $`L`$ and Riemannian geometry and show that it singles out a very interesting subclass of topological $`p`$-brane theories on special class of manifolds. These examples complement the recent discussion of topological M-theory and topological F-theory , however at microscopic level. This is our second result.
The structure of the paper is as follows. In Section 2 we describe the phase space for $`p`$-brane theory which is a simple generalization of the cotangent bundle of loop spaces. In Section 3 we associate currents to the sections of $`T^pT^{}`$ and calculate the Poisson bracket between them. The calculation gives rise to the Vinogradov bracket on $`T^pT^{}`$ (the direct generalization of Courant bracket on $`TT^{}`$) and a specific anomalous term. The anomaly free subalgebras of the currents can be associated with “isotropic” involutive subbundles of $`T^pT^{}`$. We discuss the examples of such subbundles and show that the anomaly free subalgebras of currents can be interpreted as first class constraints of some gauge theory. In Section 4 we consider the class of topological $`p`$-brane theories which are related to the Nambu-Goto $`p`$-branes in a specific way. Actually we obtain the topological strings on symplectic and Kähler manifolds, topological membranes on $`G_2`$-manifolds and topological $`3`$-branes on $`Spin(7)`$-manifolds. Section 5 presents some comments on the open $`p`$-brane theory. In particular we discuss the allowed boundary conditions which preserve the relevant symmetries. In Section 6 we summarize and collect some general comments for the future research.
## 2 Hamiltonian formalism for $`p`$-branes
The phase space of closed strings on a manifold $`M`$ can be identified with the cotangent bundle $`T^{}LM`$ of the loop space $`LM=\{X:S^1M\}`$. Below we present a straightforward generalization of this construction to the case of generic closed $`p`$-brane theory.
Following the logic above for the $`p`$-brane world-volume $`\mathrm{\Sigma }_{p+1}=\mathrm{\Sigma }_p\times `$ the phase space can be identified with the cotangent bundle $`T^{}\mathrm{\Sigma }_pM`$ of the space of maps, $`\mathrm{\Sigma }_pM=\{X:\mathrm{\Sigma }_pM\}`$. Using local coordinates $`X^\mu (\sigma )`$ and their conjugate momenta $`p_\mu (\sigma )`$ the standard symplectic form on $`T^{}\mathrm{\Sigma }_pM`$ is given by
$$\omega =\underset{\mathrm{\Sigma }_p}{}d^p\sigma \delta X^\mu \delta p_\mu ,$$
(2.1)
where $`\delta `$ is de Rham differential on $`T^{}\mathrm{\Sigma }_pM`$. The canonical dimensions of the fields should be chosen such that $`\omega `$ is dimensionless. Namely we choose<sup>3</sup><sup>3</sup>3We work in units where $`p`$-brane tension $`T_p`$ is equal to one. For details see Appendix B. $`dim[X^\mu ]=0`$, $`dim[\sigma ]=1`$ $`dim[]=1`$ and $`dim[p_\mu ]=p`$. The symplectic form (2.1) can be twisted by a closed $`(p+2)`$-form $`H`$, $`H\mathrm{\Omega }^{p+2}(M)`$, $`dH=0`$, as follows
$$\omega =\underset{\mathrm{\Sigma }_p}{}d^p\sigma \left(\delta X^\mu \delta p_\mu +\frac{1}{2}H_{\mu _1\mu _2\mu _3\mathrm{}\mu _{p+2}}\delta X^{\mu _1}\delta X^{\mu _2}ϵ^{\alpha _1\mathrm{}\alpha _p}_{\alpha _1}X^{\mu _3}\mathrm{}_{\alpha _p}X^{\mu _{p+2}}\right),$$
(2.2)
where $`ϵ^{\alpha _1\mathrm{}\alpha _p}`$ is completely antisymmetric tensor on $`\mathrm{\Sigma }_p`$. The symplectic form (2.2) implies the Poisson brackets
$$\{X^\mu (\sigma ),X^\nu (\sigma ^{})\}=0,\{X^\mu (\sigma ),p_\nu (\sigma ^{})\}=\delta _\nu ^\mu \delta (\sigma \sigma ^{}),$$
(2.3)
$$\{p_\mu (\sigma ),p_\nu (\sigma ^{})\}=H_{\mu \nu \rho _1\mathrm{}\rho _p}ϵ^{\alpha _1\mathrm{}\alpha _p}_{\alpha _1}X^{\rho _1}\mathrm{}_{\alpha _p}X^{\rho _p}\delta (\sigma \sigma ^{}).$$
(2.4)
For the symplectic structure (2.2) the transformation
$$X^\mu X^\mu ,p_\mu p_\mu +b_{\mu \nu _1\mathrm{}\nu _p}ϵ^{\alpha _1\mathrm{}\alpha _p}_{\alpha _1}X^{\nu _1}\mathrm{}_{\alpha _p}X^{\nu _p}$$
(2.5)
is canonical if $`b\mathrm{\Omega }^{p+1}(M)`$, $`db=0`$. There are also canonical transformations which correspond to $`Diff(M)`$ when $`X`$ transforms as a coordinate and $`p`$ as a section of cotangent bundle $`T^{}M`$. Indeed the group of local canonical transformations for $`T^{}\mathrm{\Sigma }_pM`$ is a semidirect product of $`Diff(M)`$ and $`\mathrm{\Omega }_{closed}^{p+1}(M)`$ in analogy with the loop space case .
Finally we conclude the discussion of Hamiltonian formalism for $`p`$-brane theory with the following comment. Typically the symplectic form (2.2) arises from the action
$$S(\gamma )=\underset{\gamma }{}(\theta h),$$
(2.6)
where $`\theta `$ is a Liouville form $`\omega =\delta \theta `$, $`h`$ is a Hamiltonian and $`\gamma `$ is a path in $`T^{}\mathrm{\Sigma }_pM`$. In order the exponential of this action, $`e^{iS(\gamma )}`$ to be well-defined we have to impose the intergrality condition on $`H`$. Namely we have to require that $`[H]H^{p+2}(M,)`$.
## 3 Current algebra and generalized Dirac structure
In this section we consider the generalization of the idea proposed in , where the authors established the relation between 2D anomaly free current algebras and Dirac structures.
Let us consider the currents which are linear in momentum $`p_\mu `$. If we assume that the currents do not depend on any dimensionful parameter or world-volume metric then the most general form is given by
$$J_ϵ(v+\omega )=\underset{\mathrm{\Sigma }_p}{}d^p\sigma ϵ\left(v^\mu (X)p_\mu +\omega _{\mu _1\mathrm{}\mu _p}(X)ϵ^{\alpha _1\mathrm{}\alpha _p}_{\alpha _1}X^{\nu _1}\mathrm{}_{\alpha _p}X^{\nu _p}\right),$$
(3.7)
where $`v+\omega `$ is a section of $`T^pT^{}`$ and $`ϵC^{\mathrm{}}(\mathrm{\Sigma }_p)`$ is a test function. Using the symplectic structure (2.1) we calculate the Poisson bracket of two currents associated to $`(v+\omega ),(\lambda +s)C^{\mathrm{}}(T^pT^{})`$,
$$\{J_{ϵ_1}(v+\omega ),J_{ϵ_2}(\lambda +s)\}=J_{ϵ_1ϵ_2}([[v+\omega ,\lambda +s]])$$
$$p\underset{\mathrm{\Sigma }_p}{}d^p\sigma (_{\alpha _1}ϵ_1)ϵ_2(i_vs+i_\lambda \omega )_{\nu _2\mathrm{}\nu _p}ϵ^{\alpha _1\alpha _2\mathrm{}\alpha _p}_{\alpha _2}X^{\nu _2}\mathrm{}_{\alpha _p}X^{\nu _p},$$
(3.8)
where the bracket $`[[,]]`$ is defined as follows
$$[[v+\omega ,\lambda +s]]=[v,\lambda ]+_vs_\lambda \omega +d(i_\lambda \omega ).$$
(3.9)
In (3.9) $`[,]`$ is the standard Lie bracket on $`TM`$ and $``$ is a Lie derivative. Alternatively the result (3.8) can be rewritten as
$$\{J_{ϵ_1}(v+\omega ),J_{ϵ_2}(\lambda +s)\}=J_{ϵ_1ϵ_2}([v+\omega ,\lambda +s]_c)+$$
$$+\frac{p}{2}\underset{\mathrm{\Sigma }_p}{}d^p\sigma (ϵ_1_{\alpha _1}ϵ_2ϵ_2_{\alpha _1}ϵ_1)(i_vs+i_\lambda \omega )_{\nu _2\mathrm{}\nu _p}ϵ^{\alpha _1\alpha _2\mathrm{}\alpha _p}_{\alpha _2}X^{\nu _2}\mathrm{}_{\alpha _p}X^{\nu _p},$$
(3.10)
where the bracket $`[,]_c`$ is given by
$$[v+\omega ,\lambda +s]_c=[v,\lambda ]+_vs_\lambda \omega \frac{1}{2}d(i_vsi_\lambda \omega ).$$
(3.11)
The bracket $`[,]_c`$ is just antisymmetrization of the bracket $`[[,]]`$.
The bracket $`[[,]]`$ is an example of derived bracket (see for a review) and its antisymmetrization $`[,]_c`$ is called Vinogradov bracket. One interesting feature is that the bracket $`[,]_c`$ has non-trivial automorphisms defined by forms . Let $`b\mathrm{\Omega }^{p+1}(M)`$ be a closed $`(p+1)`$-form which defines the vector bundle automorphism $`e^b`$ of $`T^pT^{}`$
$$e^b(v+\omega )v+\omega +i_vb.$$
(3.12)
Then the bracket $`[,]_c`$ satisfies
$$e^b\left([v+\omega ,\lambda +s]_c\right)=[e^b(v+\omega ),e^b(\lambda +s)]_c.$$
(3.13)
This non-trivial automorphism of $`[,]_c`$ corresponds to the canonical transformation (2.5) at the level of Poisson bracket of currents (3.10). If we are interested in the situation when anomalous term is absent in (3.10) and the currents form a closed algebra then we should require the following. Let label the currents by sections of a subbundle $`LT^pT^{}`$. In (3.10) the anomalous term is absent if for any $`(v+\omega ),(\lambda +s)C^{\mathrm{}}(L)`$
$$\frac{1}{2}(i_vs+i_\lambda \omega )v+\omega ,\lambda +s=0,$$
(3.14)
where $`,`$ is “pairing” between two sections of $`T^pT^{}`$ which is a map $`(T^pT^{})\times (T^pT^{})^{p1}T^{}`$ where $`^0T^{}`$ is understood as $``$. The bundle automorphism (3.12) preserves this “pairing”. We call isotropic any subbundle $`L`$ which satisfies (3.14). Moreover if we require that our currents form a closed subalgebra then we have to impose that for any two sections $`(v+\omega ),(\lambda +s)C^{\mathrm{}}(L)`$ the section $`[v+\omega ,\lambda +s]_cC^{\mathrm{}}(L)`$, i.e. the subbundle $`L`$ is involutive. Indeed the bracket $`[,]_c`$ restricted to involutive isotropic subbundle of $`T^pT^{}`$ is a Lie bracket<sup>4</sup><sup>4</sup>4 Indeed $`L`$ has a structure of the Lie algebroid with the anchor being a natural projection to $`TM`$.. Since we could not find the proof of this statement in the literature we present the proof in Appendix A as well as other relevant properties of the brackets. The proof is a direct generalization of the proof for $`TT^{}`$. Thus isotropic involutive subbundle $`L`$, as defined above, corresponds to anomaly free algebra of currents
$$\{J_{ϵ_1}(v+\omega ),J_{ϵ_2}(\lambda +s)\}=J_{ϵ_1ϵ_2}([v+\omega ,\lambda +s]_c|_L).$$
(3.15)
For the case $`p=1`$ if $`L`$ is also maximally isotropic then it is called Dirac structure. In the general situation $`p2`$ it is tempting to define a generalized Dirac structure as a maximally isotropic involutive subbundle of $`T^pT^{}`$. Although we have to admit that the notion of maximality of isotropic condition (3.14) is not very natural, however see some comments in Appendix. For different definitions of generalization of Dirac structure for $`T^pT^{}`$ (also called the Dirac-Nambu structure) see and .
The algebra of currents (3.15) corresponding to involutive isotropic subbundle $`L`$ can be regarded as an algebra of first class constraints for some gauge theory. In next Section we will give a few examples of such theories, namely topological $`p`$-branes.
Let us present some examples of isotropic involutive subbundles of $`T^pT^{}`$.
###### Example 1
Let us fix a $`(p+1)`$-form, $`\varphi \mathrm{\Omega }^{p+1}(M)`$ and consider the subbundle $`L=\{v+i_v\varphi ,vT\}T^pT^{}`$ which is obviously isotropic
$$v+i_v\varphi ,\lambda +i_\lambda \varphi =\frac{1}{2}(i_vi_\lambda \varphi +i_\lambda i_v\varphi )=0.$$
Next calculate the bracket between two sections
$$[v+i_v\varphi ,\lambda +i_\lambda \varphi ]_c=[v,\lambda ]+i_{[v,\lambda ]}\varphi +i_\lambda i_vd\varphi ,$$
(3.16)
where we used the property $`[_v,i_\lambda ]=i_{[v,\lambda ]}`$. The subbundle is involutive if the last term vanishes in (3.16), i.e. $`d\varphi =0`$. In other words $`T`$ is involutive isotropic subbundle and $`L=e^\varphi (T)`$, where $`e^\varphi `$ is the bundle automorphism defined in (3.12) for closed $`(p+1)`$-form.
The next example is related to the complexification of the bundle $`(T^pT^{})`$.
###### Example 2
On complex manifold we can consider the subbundle $`L=T_{(1,0)}(^pT^{})_{(0,p)}`$ of $`(T^pT^{})`$. The sections of $`L`$ are holomorphic vector fields and antiholomorphic forms (i.e., elements of $`\mathrm{\Omega }^{(0,p)}(M)`$). The subbundle $`L`$ is obviously isotropic and the bracket of two sections of $`L`$ is
$$[v+\omega ,\lambda +s]_c=[v,\lambda ]+i_vsi_\lambda \omega $$
which is clearly a section of $`T_{(1,0)}(^pT^{})_{(0,p)}`$. Thus $`L`$ is an isotropic involutive subbundle.
It is not hard to produce other examples of involutive isotropic subbundles of $`T^pT^{}`$, for example based on foliated geometry. In addition we can apply any closed $`(p+1)`$-form $`b`$ which defines automorphism (3.12) to an isotropic involutive subbundle $`L`$ to obtain another isotropic involutive subbundle $`e^b(L)`$.
So far we calculated the Poisson brackets using (2.1) as symplectic structure. More generally we can calculate the Poisson brackets (3.10) using the twisted symplectic structure (2.2) with $`H\mathrm{\Omega }^{p+2}(M)`$, $`dH=0`$. In this case the bracket $`[,]_c`$ in (3.10) gets replaced by its twisted version
$$[v+\omega ,\lambda +s]_H=[v+\omega ,\lambda +s]_c+i_vi_\lambda H.$$
(3.17)
All considerations above can be generalized to this case. Thus in particular Example 1 gives rise to isotropic involutive (with respect to $`[,]_H`$) subbundle if $`d\varphi =H`$.
Finally let us note that the currents (3.7) behave nicely under the diffeomorphisms of $`\mathrm{\Sigma }_p`$. Introduce the generator of of $`Diff(\mathrm{\Sigma }_p)`$
$$_\alpha [N^\alpha ]=\underset{\mathrm{\Sigma }_p}{}d^p\sigma N^\alpha _\alpha X^\mu p_\mu ,$$
where $`N^\alpha `$ is a text function. The Poisson bracket between generator of $`Diff(\mathrm{\Sigma }_p)`$ and the current (3.7) is
$$\{_\alpha [N^\alpha ],J_ϵ(v+\omega )\}=J_{N^\alpha _\alpha ϵ}(v+\omega ),$$
where we assume (2.2) as symplectic structure.
## 4 Vector cross product and topological branes
In this Section we use the construction of involutive isotropic subbundle $`L`$ given in Example 1 from previous Section. For this subbundle we can construct the anomaly free subalgebra of currents (3.15). We interpret these currents as first class constraints for a topological p-brane theory. We impose a specific compatibility of $`\varphi `$ with a Riemannian metric $`g`$ on $`M`$ which leads to a certain relation between topological and physical p-brane theories. Indeed all such theories can be classified and there is a finite number of them.
We start by explaining the compatibility condition between the $`(p+1)`$-form $`\varphi `$ and a Riemannian metric $`g`$ on $`M`$. We all are familiar with the usual vector cross product $`\times `$ of two vectors in $`^3`$, which satisfies
$``$ $`u\times v`$ is bilinear and skew symmetric
$``$ $`u\times vu,v`$; so $`(u\times v)v=0`$ and $`(u\times v)u=0`$
$``$ $`(u\times v)(u\times v)=det\left(\begin{array}{cc}uu\hfill & uv\hfill \\ vu\hfill & vv\hfill \end{array}\right)`$
The generalization of vector cross product to a Riemannian manifold leads to the following definition by Brown and Gray
###### Definition 3
On $`d`$-dimensional Riemannian manifold $`M`$ with a metric $`g`$ an $`p`$-fold vector cross product is a smooth bundle map
$$\chi :^pTMTM$$
satisfying
$$g(\chi (v_1,\mathrm{},v_p),v_i)=0,\mathrm{\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}1}ip$$
$$g(\chi (v_1,\mathrm{},v_p),\chi (v_1,\mathrm{},v_p))=v_1\mathrm{}v_p^2$$
where $`\mathrm{}`$ is the induced metric on $`^pTM`$.
Equivalently the last property can be rewritten in the following form
$$g(\chi (v_1,\mathrm{},v_p),\chi (v_1,\mathrm{},v_p))=det(g(v_i,v_j))=v_1\mathrm{}v_p^2.$$
The first condition in the above definition is equivalent to the following tensor $`\varphi `$
$$\varphi (v_1,\mathrm{},v_p,v_{p+1})=g(\chi (v_1,\mathrm{},v_p),v_{p+1})$$
being a skew symmetric tensor of degree $`p+1`$, i.e. $`\varphi \mathrm{\Omega }^{p+1}(M)`$. Thus in what follows we consider the $`(p+1)`$-form $`\varphi `$ which defines the $`p`$-fold vector cross product.
Cross product on real spaces were classified by Brown and Gray . The global vector cross products on manifolds were first studied by Gray . They fall into four categories:
(1) With $`p=d1`$ and $`\varphi `$ is the volume form of manifold
(2) When $`d`$ is even and $`p=1`$, we can have a one-fold cross product $`J:TMTM`$. Such a map satisfies $`J^2=1`$ and is almost complex structure. The associated $`2`$-form is the Kähler form.
(3) The first of two exceptional cases is a $`2`$-fold cross product ($`p=2`$) on a $`7`$-manifold. Such a structure is called a $`G_2`$-structure and the associated $`3`$-form is called a $`G_2`$-form.
(4) The second exceptional case is $`3`$-fold cross product ($`p=3`$) on $`8`$-manifold. This is called a $`Spin(7)`$-structure and the associated $`4`$-form is called $`Spin(7)`$-form.
Notice that there are similarities of this list of real vector cross products with the list of stable forms . Namely the cases (2) and (3) correspond to stability of $`\varphi `$. The complexified version of the vector cross product which allows to consider Calabi-Yau manifolds, see . However we will not review the complex version of vector cross product.
Following the discussion from previous section, in particular Example 1, there is a set of topological $`p`$-brane theories we can associate to a $`p`$-fold vector cross product characterized by $`(p+1)`$-form $`\varphi `$. Consider a subbundle $`L=\{v+i_v\varphi ,vT\}`$ of $`T^pT^{}`$. To the sections of $`L`$ we can associate the following constraints (currents)
$$J_\mu =p_\mu +\varphi _{\mu \nu _1\mathrm{}\nu _p}ϵ^{\alpha _1\mathrm{}\alpha _p}_{\alpha _1}X^{\nu _1}..._{\alpha _p}X^{\nu _p}=0,$$
(4.18)
where we work in local basis $`_\mu `$. Alternatively we can rewrite the constraints in coordinate free form
$$i_vJ=i_vp+g(\chi (_1X,\mathrm{},_pX),v)=0,$$
(4.19)
where $`v`$ is a section of $`TM`$. The constraints (4.18) are the first class with respect to the symplectic form (2.1) if $`d\varphi =0`$. In the twisted case, when one uses (2.2), the first class condition leads to $`d\varphi =H`$.
Let us now study the compatibility condition between the topological system (4.18) and the Nambu-Goto dynamics. The constraints (4.18) imply the Nambu-Goto costraints (see Appendix B)
$`_\alpha =p_\mu _\alpha X^\mu =0`$ (4.20)
$`=p_\mu g^{\mu \nu }p_\nu det(_\alpha X^\mu g_{\mu \nu }_\beta X^\nu )=0`$ (4.21)
if and only if $`\varphi `$ corresponds to vector cross product<sup>5</sup><sup>5</sup>5Indeed previously the cross vector product has been discussed in the context of $`p`$-brane instantons for the Nambu-Goto theory .. Namely
$$_\alpha =J_\mu _\alpha X^\mu =0$$
and
$$p_\mu g^{\mu \nu }p_\nu =_1X\mathrm{}_pX=det(g(_\alpha X,_\beta X)),$$
where we have used the second property in the definition of vector cross product. Indeed the Nambu-Goto $`p`$-brane theory is decribed by $`(p+1)`$ constraints (4.20) and (4.21), see Appendix B for the details.
We constructed TFTs such that their constraint surface $`J_\mu =0`$ lies inside the constraint surface for the standard $`p`$-brane theory,
$$J_\mu =0_\alpha =0,=0.$$
Classically it means that the BRST cohomology of topological branes is subspace of the BRST cohomology of physical brane theory. At quantum level we may speculate that the correlators of observables of topological brane theory are related to subsector of physical brane theory, in analogy with the relation between topological strings and superstrings. However, at the present level of discussion, we cannot elaborate more on the relation between quantum toopological and physical brane theories.
There is an alternative point of view on the relation between the topological $`p`$-brane theory and standard $`p`$-brane theory (i.e., given by Nambu-Goto (NG) action) on a manifold with a vector cross product structure. Namely the Nambu-Goto action can be thought of as a deformation of the corresponding topological theory. The Hamiltonian of Nambu-Goto theory is given by the following expression
$$h_{NG}=d^p\sigma \left(N+N^\alpha _\alpha \right),$$
where $``$, $`_\alpha `$ are the constraints (4.20)-(4.21) and $`N`$, $`N^\alpha `$ are Lagrangian multipliers. Next assume that $`\varphi `$ defines a vector cross product with respect to $`g`$, so does $`\varphi `$. We define the currents
$$J_\mu ^\pm =p_\mu \pm \varphi _{\mu \nu _1\mathrm{}\nu _p}ϵ^{\alpha _1\mathrm{}\alpha _p}_{\alpha _1}X^{\nu _1}..._{\alpha _p}X^{\nu _p}$$
and rewrite the constraints $``$, $`_\alpha `$ as follows
$$=J_\mu ^+g^{\mu \nu }J_\nu ^{}\mathrm{and}_\alpha =_\alpha X^\mu J_\mu ^+,$$
where we used the fact that $`\varphi `$ is vector cross product with respect to $`g`$. As a further preparatory step, we introduce the auxiliary fields $`B_\pm ^\mu `$ and rewrite the NG action as
$$S_{NG}=𝑑td^p\sigma \left(p_\mu \dot{X}^\mu B_{}^\mu J_\mu ^{}B_+^\mu J_\mu ^++\frac{1}{N}B_+^\mu g_{\mu \nu }B_{}^\nu \frac{1}{N}N^\alpha _\alpha X^\mu g_{\mu \nu }B_{}^\nu \right)$$
(4.22)
Since the fields $`B_+^\mu `$ enter linearly we can integrate them and arrive at the standard Nambu-Goto action in the phase space form. Obviously the action (4.22) is not unique and there are other equivalent ways to rewrite it.
For the topological $`p`$-brane theory we have two possible (equivalent) Hamiltonians
$$h^\pm =d^p\sigma B_\pm ^\mu J_\mu ^\pm ,$$
where $`B_\pm ^\mu `$ are the Lagrange multipliers. In action (4.22) actually both currents $`J_\mu ^\pm `$ enter. However we do not want to introduce two copies of the topological theory and thus one of the two should be fake. This can be easily obtained by considering the action
$$S_{top}=𝑑td^p\sigma \left(p_\mu \dot{X}^\mu B_{}^\mu J_\mu ^{}B_+^\mu J_\mu ^+\chi _\mu B_+^\mu \right),$$
(4.23)
where $`\chi _\mu `$ is the Lagrange multiplier freezing $`B_+^\mu `$. Now combining (4.22) and (4.23) it is straightforward to write the Nambu-Goto action as follows
$$S_{NG}=S_{top}\lambda S_{def}$$
(4.24)
where
$$S_{def}=𝑑td^p\sigma \left(\frac{1}{N}N^\alpha _\alpha X^\mu g_{\mu \nu }B_{}^\mu +\eta _\mu \left(g^{\mu \nu }\chi _\nu +\frac{1}{N}B_{}^\mu \right)\right)$$
where $`\eta `$ is an additional auxiliary field. In (4.24) at $`\lambda =0`$ the theory describes the topological $`p`$-brane theory. If $`\lambda `$ is non zero then the action (4.24) becomes $`S_{NG}`$ upon a rescaling of the Lagrange multipliers $`N^\alpha \lambda ^1N^\alpha `$. This construction (or its versions) exists only if $`\varphi `$ corresponds to a vector cross product structure and $`d\varphi =0`$.
Thus for the list of vector cross product structures given above there is a corresponding list of topological $`p`$-brane theories. The first case with $`\varphi `$ given by the volume structure corresponds to the trivial case when the Nambu-Goto action is itself topological since it describes the embedding of $`(d1)`$-branes into a $`d`$-dimensional manifold, for details see .
We would like to discuss the other three non-trivial cases: topological strings on symplectic manifolds (also on generalized Kähler manifolds), topological membranes on $`G_2`$-manifolds and topological $`3`$-branes on $`Spin(7)`$-manifolds.
### 4.1 Topological strings on symplectic manifolds
Case (2) in the list of real vector cross products corresponds to A-model topological strings. $`1`$-fold cross product $`J:TMTM`$ corresponds to an almost complex structure<sup>6</sup><sup>6</sup>6The vector cross product properties read $`(gJ)^t=gJ`$ and $`J^tgJ=g`$ which imply $`J^2=1`$., $`J^2=1`$. The associated $`2`$-form $`\omega =gJ`$ is the Kähler form. The constraints corresponding to maximally isotropic subbundle $`L=\{v+i_v\omega ,vT\}`$ of $`TT^{}`$ are
$$p_\mu +\omega _{\mu \nu }X^\nu =0.$$
(4.25)
They are first class constraints if $`d\omega =0`$ and thus the manifold $`M`$ is symplectic. Indeed this is nothing but A-model topological string theory.
As far as classical B-model is concern we have to introduce another structure on $`M`$. This would correspond to Example 2 in Section 3 with $`p=1`$. Thus in this case $`M`$ is a complex manifold with the complex structure $`J`$ and the constraints are given by
$$p_i=0,X^{\overline{i}}=0$$
in complex coordinates. To accomodate both A- and B-models on the same $`M`$ we have to restrict ourselves to the case of Kähler manifold $`(J,g,\omega =gJ)`$. In this case we have the following decomposition into holomorphic (antiholomorphic) subbundles
$$(TT^{})=T^{(1,0)}T^{(0,1)}T^{(1,0)}T^{(0,1)}.$$
(4.26)
There are two interesting sets of complex Dirac structures, first one is $`T^{(1,0)}T^{(0,1)}`$ (or complementary $`T^{(0,1)}T^{(1,0)}`$) and second is $`T^{(1,0)}T^{(1,0)}`$ (or complementary $`T^{(0,1)}T^{(0,1)}`$). Indeed they corresponds to two different generalized complex structures
$$𝒥_i:(TT^{})(TT^{}),i=1,2$$
such that $`𝒥_i^2=1`$ and $`\mathrm{\Pi }_\pm ^i=\frac{1}{2}(1\pm i𝒥_i)`$ project maximally isotropic involutive subbundles of $`(TT^{})`$ (for more details see ). In the case of Kähler manifolds the corresponding generalized complex structures are
$$𝒥_1=\left(\begin{array}{cc}J\hfill & 0\hfill \\ 0\hfill & J^t\hfill \end{array}\right),𝒥_2=\left(\begin{array}{cc}0\hfill & \omega ^1\hfill \\ \omega \hfill & 0\hfill \end{array}\right),$$
(4.27)
which commute and give rise to the following positive metric on $`TT^{}`$
$$𝒢=𝒥_1𝒥_2=\left(\begin{array}{cc}0\hfill & g^1\hfill \\ g\hfill & 0\hfill \end{array}\right).$$
Introducing
$$\mathrm{\Lambda }=\left(\begin{array}{c}iX\hfill \\ p\hfill \end{array}\right)$$
as a section of pull-back of tangent and cotangent bundle, $`X^{}((TT^{}))`$ we have four topological string theories given by the set of first class constraints
$$\mathrm{\Pi }_\pm ^i\mathrm{\Lambda }=0.$$
(4.28)
Indeed there are only two distinct theories. For the case $`T^{(1,0)}T^{(0,1)}`$ we have $`\mathrm{\Pi }_{}^2\mathrm{\Lambda }=0`$, i.e.
$$p_iig_{i\overline{j}}X^{\overline{j}}=0,p_{\overline{i}}+ig_{\overline{i}j}X^j=0$$
(4.29)
which is A-model topological strings. For the other case $`T^{(1,0)}T^{(1,0)}`$ the constraints are $`\mathrm{\Pi }_{}^1\mathrm{\Lambda }=0`$, i.e.
$$p_i=0,X^{\overline{i}}=0$$
(4.30)
corresponding to B-model topological strings<sup>7</sup><sup>7</sup>7Using the relation $`p_\mu =g_{\mu \nu }\dot{X}`$ in (4.29) and (4.30) one can recoginize the holomorphic map and constant map conditions over which A- and B-model path integrals are localized correspondently.. Obviously both A- and B-models constraints imply the physical string constraints, $`_1=p_\mu X^\mu =0`$ and $`=p_\mu g^{\mu \nu }p_\nu X^\mu g_{\mu \nu }X^\nu =0`$. Using the natural pairing $`,`$ on $`TT^{}`$ (see (A.5) for $`p=1`$) we can rewrite the string constraints as follows
$$i_1=\mathrm{\Lambda },\mathrm{\Lambda }=0,\mathrm{\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}2}=\mathrm{\Lambda },𝒢\mathrm{\Lambda }=0.$$
(4.31)
Since we have formulated everything in $`TT^{}`$ covariant language it is not hard to generalize above discussion to the case (twisted) generalized Kähler manifolds as defined in . The generalized Kähler structure is given by two generalized complex structures $`𝒥_1`$ and $`𝒥_2`$ which commute and $`𝒢=𝒥_1𝒥_2`$ defines the positive metric on $`TT^{}`$.
### 4.2 Topological membrane on $`G_2`$ manifolds
The first exceptional case, namely (3) in the list of real vector cross product structures, corresponds to $`M`$ being oriented $`7`$-manifold with a global $`2`$-fold cross product structure ($`p=2`$). This cross product is defined by Riemannian metric $`g`$ and $`3`$-form $`\mathrm{\Phi }`$ which gives rise to a $`G_2`$-structure on the manifold<sup>8</sup><sup>8</sup>8In this case the metric $`g`$ can be expressed in terms of $`\mathrm{\Phi }`$, .. The topological membrane theory on $`G_2`$-manifold is defined by the following first class constraints in $`T^{}\mathrm{\Sigma }_2M`$
$$p_\mu +\mathrm{\Phi }_{\mu \nu \rho }ϵ^{\alpha \beta }_\alpha X^\nu _\beta X^\rho =0.$$
(4.32)
The algebraic properties of $`\mathrm{\Phi }`$ are such that the constraints (4.32) imply the membrane constraints (4.20)-(4.21). $`d\mathrm{\Phi }=0`$ is equivalent to the fact that (4.32) are first class constraints with respect to the symplectic structure (2.1). We put forward this as the Hamiltonian description of recently proposed topological M-theory at microscopic level.
Suppose that $`G_2`$-manifold $`M_7`$ is of the form $`M_7=M_6\times S^1`$, where $`M_6`$ is a six-dimensional manifold with $`SU(3)`$ structure. Let $`X^7`$ be a coordinate along $`S^1`$ then $`\mathrm{\Phi }`$ can be written as
$$\mathrm{\Phi }=\omega dX^7+\rho ,$$
(4.33)
where $`\omega `$ is the Kähler $`2`$-form and $`\rho `$ is the $`3`$-form which defines the almost complex structure on $`6`$-manifold. If $`\omega `$ and $`\rho `$ do not depend on $`X^7`$ then $`d\mathrm{\Phi }=0`$ implies that $`d\omega =0`$ and $`d\rho =0`$ on $`M_6`$. Membranes on such $`M_7`$ can be reduced either to strings on $`M_6`$ or to membranes on $`M_6`$ depending on the orientation with respect to $`S^1`$. If the brane is wrapped along $`S^1`$ then we can make a partial gauge fixing $`X^7=\sigma _2/L`$ with $`L`$ being the size of $`S^1`$. Then the constraint (4.32) becomes
$$Lp_n+2L\rho _{nml}_1X^m_2X^l+2\omega _{nm}_1X^m=0,p_7+2\omega _{nm}_1X^n_2X^m=0,$$
(4.34)
where $`\mu =(n,7)`$. If we want to reinterpret this as a constraint in $`M_6`$ we have to redefine the momenta<sup>9</sup><sup>9</sup>9See Section 2 and Appendix for our conventions on the dimensionality of fields. $`p_n|_{M_6}Lp_n`$ and restrict our attention only to $`\sigma _2`$ inedpendent configurations (e.g., by requiring $`_2X^n=0`$). Assuming this we arrive to the constraint
$$p_n+2\omega _{nm}_1X^m=0$$
(4.35)
which is A-model on $`M_6`$. Another possibility corresponds to the case when original membrane does not have excitations along $`X^7`$, e.g. $`X^7`$ chosen to be a constant. Then in this case the theory on $`M_6`$ is membrane theory<sup>10</sup><sup>10</sup>10Using the notion of complex vector cross product we can show that a complex version of the constraints (4.36) implies the membrane Nambu-Goto constraints.,
$$p_n+\rho _{nml}_1X^m_2X^l=0.$$
(4.36)
Since this theory depends on complex moduli it is tempting to call it B-model. Although perturbative B-model is typically defined as a topological string theory there should be a dual formulation in terms of membrane theory. Indeed this option is very natural from geometrical point of view due to the moduli dependence.
### 4.3 Topological 3-brane on $`Spin(7)`$ manifolds
The last case in the list of real vector cross products to an oriented $`8`$-manifold $`M`$ with a global cross product structure with $`p=3`$. This cross product gives rise to an associated Riemannian metric $`g`$ and $`4`$-form $`\mathrm{\Psi }`$. Indeed $`\mathrm{\Psi }`$ is self-dual form $`\mathrm{\Psi }=\mathrm{\Psi }`$, which is called sometime Cayley form and defines $`Spin(7)`$-structure on $`M`$. The theory is described by the following first class constraints in $`T^{}\mathrm{\Sigma }_3M`$
$$p_\mu +\mathrm{\Psi }_{\mu \nu \rho \sigma }ϵ^{\alpha \beta \gamma }_\alpha X^\nu _\beta X^\rho _\gamma X^\sigma =0.$$
(4.37)
The algebraic properties of $`\mathrm{\Psi }`$ would follow from the requirement that above constraints imply the $`3`$-brane constraints (4.20)-(4.21). The closure of $`\mathrm{\Psi }`$ is equivalent to the constraints (4.37) being first class with respect to the symplectic structure (2.1). We propose that this topological $`3`$-brane theory is microscopic description of topological F-theory recently discussed in .
Let us study two possible reductions of $`3`$-brane topological theory on $`Spin(7)`$-manifold down to $`G_2`$\- and $`SU(3)`$-manifolds. As a first case consider $`Spin(7)`$-manifold of the form $`M_8=M_7\times S^1`$ with
$$\mathrm{\Psi }=dX^8\mathrm{\Phi }+\mathrm{\Phi }$$
(4.38)
where $`\mathrm{\Phi }`$ is $`G_2`$-structure on $`M_7`$ independent on $`X^8`$. As result $`d\mathrm{\Psi }=0`$ implies $`d\mathrm{\Phi }=0`$ and $`d\mathrm{\Phi }=0`$. In analogy with the reduction we discussed in previous subsection a reduction of topological $`3`$-brane theory on $`M_8`$ gives a topological membrane theory (with $`\mathrm{\Phi }`$ in constraint) theory and topological $`3`$-brane theory (with $`\mathrm{\Phi }`$ in constraint) on $`M_7`$. However topological $`3`$-brane theory cannot be related to $`3`$-brane Nambu-Goto theory in a way described previously.
Following we can consider $`Spin(7)`$-manifold $`M_8=M_6\times T^2`$ where $`M_6`$ is $`SU(3)`$-manifold. Assuming that $`(X^7,X^8)`$ are coordinates along $`T^2`$ the Cayley form is given by
$$\mathrm{\Psi }=dX^7\rho dX^8\widehat{\rho }+dX^7dX^8\omega +\frac{1}{2}\omega \omega ,$$
(4.39)
where $`(\rho ,\omega )`$ defines $`SU(3)`$-structure on $`M_6`$, such that $`\mathrm{\Omega }=\rho +i\widehat{\rho }`$. We can reduce the topological $`3`$-brane theory given by (4.37) down to $`M_6`$. We get a family of topological theories: topological strings ($`\omega `$), topological $`3`$-branes ($`\omega \omega `$) and two topological membranes (for $`\rho `$ and $`\widehat{\rho }`$). Since on $`M_8`$ topological $`3`$-brane theory is self-dual ($`\mathrm{\Psi }=\mathrm{\Psi }`$), in $`M_6`$ we get the duality between topological string ($`\omega `$) and topological $`3`$-brane ($`\omega \omega `$) and another duality between topological membrane theories ($`\rho `$ and $`\widehat{\rho }`$). Indeed two first theories can be interpreted as A-model and membrane theories as B-model. This would agree with the expected moduli dependence. Presumably the duality we just discussed is related to proposed S-duality .
## 5 Open $`p`$-branes
The open string phase space can be identified with the cotangent bundle $`T^{}PM`$ of the path space $`PM=\{X:[0,1]MX(0)D_0,X(1)D_1\}`$. This construction can be generalized to the case of open $`p`$-branes. Assume for the sake of clarity that $`\mathrm{\Sigma }_p`$ consists of one component. For such open $`p`$-brane the phase space can be identified with the cotangent bundle $`T^{}\mathrm{\Sigma }_pM_D`$ of the space $`\mathrm{\Sigma }_pM_D=\{X:\mathrm{\Sigma }_pM,X(\mathrm{\Sigma }_p)D\}`$ where $`D`$ is a submanifold of $`M`$, $`i:DM`$. To write down the symplectic structure on $`T^{}\mathrm{\Sigma }_pM_D`$ we have to require that there exists $`B\mathrm{\Omega }^{p+1}(D)`$ such that $`dB=i^{}H`$. Hence the symplectic structure is given by
$$\omega =\underset{\mathrm{\Sigma }_p}{}d^p\sigma \left(\delta X^\mu \delta p_\mu +\frac{1}{2}H_{\mu _1\mu _2\mu _3\mathrm{}\mu _{p+2}}\delta X^{\mu _1}\delta X^{\mu _2}ϵ^{\alpha _1\mathrm{}\alpha _p}_{\alpha _1}X^{\mu _3}\mathrm{}_{\alpha _p}X^{\mu _{p+2}}\right)$$
$$\frac{1}{2}\underset{\mathrm{\Sigma }_p}{}d^{p1}\sigma B_{\mu _1\mu _2\mu _3\mathrm{}\mu _{p+1}}\delta X^{\mu _1}\delta X^{\mu _2}ϵ^{\alpha _1\mathrm{}\alpha _{p1}}_{\alpha _1}X^{\mu _3}\mathrm{}_{\alpha _{p1}}X^{\mu _{p+1}},$$
(5.40)
where the boundary contributions are needed in order $`\omega `$ to be closed, $`\delta \omega =0`$. If we require the symplectic form (5.40) to be compatible with the action (2.6) with $`\theta `$ being a Liouville form for $`\omega =\delta \theta `$ then, in order to the exponent of this action to be well-defined, we have to impose $`[(H,B)]H^{p+2}(M,D,)`$, where $`H^{p+2}(M,D,)`$ is an integer relative cohomology group.
Let us introduce a few useful mathematical notions which are generalizations of the ideas from used in the context of $`TT^{}`$.
###### Definition 4
Let $`M`$ be a manifold with a closed $`(p+2)`$-form $`H`$. Then the pair $`(D,B)`$ of a submanifold $`i:DM`$ together with a $`(p+1)`$-form $`B\mathrm{\Omega }^{p+1}(D)`$ is a generalized submanifold of $`(M,H)`$ iff $`dB=i^{}H`$.
A generalized submanifold $`(D,B)`$ is exactly the data we need to construct the phase space $`T^{}\mathrm{\Sigma }_pM_D`$ together with the symplectic structure (5.40).
###### Definition 5
The generalized tangent bundle $`\tau _D^B`$ of the generalized submanifold $`(D,B)`$ is
$$\tau _D^B=\{v+\omega TD^pT^{}M|_D:\omega |_D=i_vB\}$$
isotropic subbundle of $`(TM^pT^{}M)|_D`$.
If we choose $`B=0`$ then $`\tau _D^0=TD^pN^{}D`$, where $`N^{}D`$ is the conormal subbundle of the submanifold $`D`$ (in other word $`N^{}D=AnnTDT^{}M`$). The action of the non-trivial automorphism (3.12) of $`TM^pT^{}M`$ on generalized submanifolds is given as follows
$$e^b(D,B)=(D,B+b)$$
First consider the simple case when $`H=0`$ and $`B=0`$. Introducing the currents (3.7) labelled by the section of subbundle $`L`$ of $`TM^pT^{}M`$ we can calculate their Poisson bracket with respect to the symplectic structure (2.1). Thus in the case of boundary the calculation (3.10) is modified
$$\{J_{ϵ_1}(v+\omega ),J_{ϵ_2}(\lambda +s)\}=J_{ϵ_1ϵ_2}([v+\omega ,\lambda +s]_c)+$$
$$+\frac{p}{2}\underset{\mathrm{\Sigma }_p}{}d^p\sigma (ϵ_1_{\alpha _1}ϵ_2ϵ_2_{\alpha _1}ϵ_1)(i_vs+i_\lambda \omega )_{\nu _2\mathrm{}\nu _p}ϵ^{\alpha _1\alpha _2\mathrm{}\alpha _p}_{\alpha _2}X^{\nu _2}\mathrm{}_{\alpha _p}X^{\nu _p}+$$
$$+\frac{1}{2}\underset{\mathrm{\Sigma }_p}{}d^{p1}\sigma ϵ_1ϵ_2(i_\lambda \omega i_vs)_{\nu _2\mathrm{}\nu _p}ϵ^{\alpha _2\alpha _3\mathrm{}\alpha _p}_{\alpha _2}X^{\nu _2}\mathrm{}_{\alpha _p}X^{\nu _p}.$$
(5.41)
As discussed in Section 3 we have to require that $`L`$ is an isotropic and involutive subbundle of $`TM^pT^{}M`$. However now we have to take care of the boundary term in (5.41) to the anomaly. This can be done by requiring that
$$(i_\lambda \omega i_vs)|_D=0$$
for any $`(v+\omega ),(\lambda +s)C^{\mathrm{}}(L)`$. Moreover we have to insure that the action of the currents (i.e., the transformations they generate) do not change the boundary conditions, $`X(\mathrm{\Sigma }_p)D`$, i.e. $`v`$ and $`\lambda `$ restricted to $`D`$ should be the sections of $`TD`$. We can fulfill these two conditions together with the isotropy condition of $`L`$ by the following
$$L|_DTD^pN^{}D,$$
where $`L|_D`$ is the restriction of subbundle $`L`$ to the submanifold $`D`$. In the general situation if we allow a generalized submanifold $`(D,B)`$ then the correct condition is
$$L|_D\tau _D^B,$$
(5.42)
i.e. $`L|_D`$ is a subbundle of the generalized tangent bundle of the generalized submanifold $`(D,B)`$.
## 6 Conclusions
Let us first of all summarize what we have been finding in the previous sections. We started by studying specific current algebras for extended objects requiring the currents to be linear in the momenta, do not involve any world-volume metric and do not contain any dimensionfull parameter. The current algebras where shown to close under the (twisted or untwisted) Poisson bracket if their structure is parametrized by an ”isotropic” involutive subbundle of $`T^pT^{}`$. We may interpreted then these currents as first class constraints for topological p-branes theories.
In order to link with the usual Nambu-Goto theory, we required the gauge constraints of the topological theory to imply the ones defining the NG theory itself. Equivalently, we required the topological brane theory to be a topological truncation of the NG one. We have shown that the above requirements, namely the algebra closure and the deformability to the NG theory, correspond to the existence of a real cross vector product on the manifold on which the p-brane theory is formulated. This mathematical condition reveals to be quite restrictive leaving with few well defined cases. These, and the induced p-brane topological theories, were listed and analised. One of them was the A-model topological string in six dimensions, which we reconstruct in detail. Through an alternative scheme, we reconstructed the B-model in its usual formulation too. In seven dimensions we encountered membrane theory on $`G_2`$ manifolds which upon reduction to six dimensions gave the A-model and a novel membrane theory naturally coupled to the complex moduli of the six manifold. Analogous phenomena appeared in the last case of 3-branes on eight dimensional manifolds admitting a $`Spin(7)`$ structure.
The reduction of topological F-theory from $`Spin(7)`$-manifold down to $`SU(3)`$-manifold produces a whole set of topological brane theories. Some of them are related to Nambu-Goto theories in the way described above. One is the topological membrane theory which should be a version of the B-model since it couples naturally to the complex moduli. This should be regarded as the nonperturbative completition of the A-model. The whole picture requires further study especially at the quantum level. We believe that the present reduction can be generalized to BV set-up<sup>11</sup><sup>11</sup>11For some discussion of BV formalism applied to open topological membrane see . and we hope to come back to this issue in future.
Acknowledgement: We are grateful to Nigel Hitchin, Simon Lyakhovich, Alessandro Tanzini and Pierre Vanhove for discussions. The research of G.B. is supported by the Marie Curie European Reintegration Grant MERG-CT-2004-516466, the European Commission RTN Program MRTN-CT-2004-005104 and by MIUR. M.Z. thanks SISSA (Trieste) where part of work was carried out. The research of M.Z. was supported by EU-grant MEIF-CT-2004-500267.
Note added in Proof: After we have finished this work we became aware of two interesting works. In the authors discuss the gauging of sigma model with boundary. Motivated by their example they argue that the notion of isotropic subbundle (3.14) can be extended to
$$\frac{1}{2}(i_vs+i_\lambda \omega )v+\omega ,\lambda +s=dq,$$
where $`q\mathrm{\Omega }^{p1}(M)`$. We find this observation interesting. However it is not clear to us the proper interpretation of this condition within our motivating example.
Also after our paper appeared on the net the different proposal for microscopic description of topological M-theory has been given in .
## Appendix A Appendix: brackets on $`C^{\mathrm{}}(T^pT^{})`$
In this Appendix we collect the relevant properties of the brackets $`[[,]]`$ and $`[,]_c`$ defined on the sections of $`T^pT^{}`$. The proofs of these properties are similar to those presented in , in the context of Courant algebroid.
On smooth sections of $`T^pT^{}`$ we can define the bracket
$$[[v+\omega ,\lambda +s]]=[v,\lambda ]+_vs_\lambda \omega +d(i_\lambda \omega ),$$
(A.1)
which is not skew-symmetric. However it satisfies a kind of Leibniz rule
$$[[A,[[B,C]]]]=[[[[A,B]],C]]+[[B,[[A,C]]]],$$
(A.2)
where $`A,B,CC^{\mathrm{}}(T^pT^{})`$. The property (A.2) is easily proved from the definition (A.1). In fact the bracket $`[[,]]`$ makes $`C^{\mathrm{}}(T^pT^{})`$ into a Loday algebra. Next we define a new bracket $`[,]_c`$ as anitsymmetrization of $`[[,]]`$
$$[A,B]_c=\frac{1}{2}\left([[A,B]][[B,A]]\right).$$
(A.3)
The explicite expresion for $`[,]_c`$ is given by
$$[v+\omega ,\lambda +s]_c=[v,\lambda ]+_vs_\lambda \omega \frac{1}{2}d(i_vsi_\lambda \omega ).$$
(A.4)
Let us introduce “pairing” between two sections of $`T^pT^{}`$
$$v+\omega ,\lambda +s=\frac{1}{2}(i_vs+i_\lambda \omega ),$$
(A.5)
which is a map
$$(T^pT^{})\times (T^pT^{})^{p1}T^{},$$
(A.6)
where $`^0T^{}`$. Thus the relation between two brackets (A.1) and (A.4) is as follows
$$[A,B]_c=[[A,B]]dA,B.$$
(A.7)
The bracket $`[,]_c`$ does not satisfies the Jacobi identity. However it is interesting to examine how it fails to satisfy the Jacobi identity. Let us introduce a trilinear operator, Jacobiator, which measures the failure to satisfy the Jacobi identity
$$Jac(A,B,C)=[[A,B]_c,C]_c+[[B,C]_c,A]_c+[[C,A]_c,B]_c.$$
(A.8)
We can prove the following property
$$Jac(A,B,C)=d\left(Nij(A,B,C)\right)$$
(A.9)
where $`Nij`$ is the Nijenhuis operator
$$Nij(A,B,C)=\frac{1}{3}\left([A,B]_c,C+[B,C]_c,A+[C,A]_c,B\right).$$
(A.10)
In order to prove (A.9) we note that
$$[[A,B]_c,C]_c=[[[[A,B]],C]]d[A,B]_c,C$$
(A.11)
where we have used (A.7) and the fact that $`[[\omega ,C]]=0`$ whenever $`\omega `$ is closed form.
As corollary of (A.9) we can establish a few useful theorems. Let us call a subbundle $`LT^pT^{}`$ isotropic if for any $`A,BC^{\mathrm{}}(L)`$, $`A,B=0`$, where $`,`$ is defined by (A.5).
###### Theorem 6
If subbundle $`LT^pT^{}`$ is isotropic and involutive with respect to bracket $`[,]_c`$ then $`Nij|_L=0`$ and $`Jac|_L=0`$.
Thus the bracket $`[,]_c`$ restricted to isotropic involutive subbundle of $`T^pT^{}`$ is a Lie bracket. If we add the requirement of maximality to isotropic condition then there is the following theorem. By maximal isotropic subbundle $`L`$ we mean that if the condition
$$v+\omega ,\lambda +s=0$$
is satisfied for all $`(v+\omega )C^{\mathrm{}}(L)`$ then $`(\lambda +s)C^{\mathrm{}}(L)`$, where $`,`$ is defined by (A.5).
###### Theorem 7
If subbundle $`LT^pT^{}`$ is maximally isotropic then the following statements are equivalent:
$``$ $`L`$ is involutive with respect to $`[,]_c`$
$``$ $`Jac|_L=0`$
$``$ $`Nij|_L=0`$
For $`p=1`$ a maximally isotropic involutive subbundle of $`TT^{}`$ is called a Dirac structure. Thus for the case $`p2`$ we refer to a maximally isotropic involutive subbundle of $`T^pT^{}`$ as a generalized Dirac structure.
## Appendix B Hamiltonian constaints for p-brane
In this Appendix we remind the elements of Hamiltonian analysis of the standard $`p`$-brane theory. The $`p`$-brane theory describes the embedding of a $`(p+1)`$-dimensional world-volume into a $`d`$-dimensional manifold $`M`$. The Nambu-Goto action is given by the volume of the embedded $`(p+1)`$ manifold
$$S=T_p\underset{\mathrm{\Sigma }_{p+1}}{}d^{p+1}\sigma \sqrt{det(g_{\mu \nu }_aX^\mu _bX^\nu )},$$
(B.1)
where $`g_{\mu \nu }`$ is the metric with Euclidean signature on $`M`$ and $`T_p`$ is brane tension. If we put $`T_p=1`$ then we choose that $`dim[X]=0`$. In order to carry the Hamiltonian analysis we assume $`\mathrm{\Sigma }_{p+1}=\mathrm{\Sigma }_p\times `$, i.e. $`\sigma ^a=(\sigma ^\alpha ,\sigma ^0)`$ with $`\sigma ^0`$ being the evolution parameter.
Denoting by $`p_\mu `$ the momenta conjugate to $`X^\mu `$ and starting from the Nambu-Goto action (B.1) the constraints can be worked out as
$`=g^{\mu \nu }p_\mu p_\nu det(q_{\alpha \beta })`$ (B.2)
$`_\alpha =p_\mu _\alpha X^\mu `$ (B.3)
where
$$q_{\alpha \beta }=g_{\mu \nu }_\alpha X^\mu _\beta X^\nu $$
(B.4)
is induced spatial metric on the brane. |
warning/0507/nucl-th0507066.html | ar5iv | text | # Quenching of pairing gap at finite temperature in 184W
## Abstract
We extract pairing gap in <sup>184</sup>W at finite temperature for the first time from the experimental level densities of <sup>183</sup>W, <sup>184</sup>W, and <sup>185</sup>W using “thermal” odd-even mass difference. We found the quenching of pairing gap near the critical temperature $`T_c=0.47`$ MeV in the BCS calculations. It is shown that the monopole pairing model with a deformed Woods-Saxon potential explains the reduction of the pairing correlation using the partition function with the number parity projection in the static path approximation plus random-phase approximation.
Pairing correlations are essential for many-fermion systems such as electrons in the superconducting metal, nucleons in the nucleus, and quarks in the color superconductivity. The Bardeen-Cooper-Schriffer (BCS) theory BCS of superconductivity has succesfully described the pairing correlations. This theory was applied to nuclear problems at zero temperature Bohr ; Belyaev for stable nuclei. The thermodynamical properties of nuclear pairing were investigated by using the BCS theory in the study of hot nuclei Sano ; Goodman . Breaking of the Cooper pairs is expected to occur at a certain critical temperature in the BCS theory.
It has recently been reported Schiller ; Melby that the canonical heat capacities extracted from the observed level densities in <sup>162</sup>Dy, <sup>166</sup>Er and <sup>172</sup>Yb form the S shape with a peak around the temperature $`T0.5`$ MeV. These S-shaped heat capacities were interpreted as the breaking of nucleon Cooper pairs and the pairing transition because this temperature is close to the critical temperature $`T_c=0.57\mathrm{\Delta }0.5`$ MeV in the BCS theory. For the finite Fermi system like a nucleus, however, nucleon number fluctuation and statistical fluctuations beyond the mean field become large. The fluctuations wash out the sharp phase transition, and then the pairing gap $`\mathrm{\Delta }`$ does not become quickly zero at the BCS critical temperature. Several models have taken into account the fluctuations beyond the mean field. The quenching of pairing correlations have been obtained in recent theoretical approaches: the static path approximation (SPA) plus random-phase approximation (RPA) Rossignoli , the shell model Monte Calro (SMMC) calculations Rombouts ; Liu , the finite-temperature Hartree-Fock-Bogoliubov (HFB) theory Egido , and the relativistic mean field theory Agrawal .
The odd-even mass difference observed in nuclear masses is well known as one of signatures of pairing correlations. In solid state physics, difference between the free energies with odd and even numbers of electrons in ultrasmall superconducting grains is found and known as even-odd effect Tuominen . In our previous paper Kaneko , we suggested that the suppression of the pairing correlations due to finite temperature appears in “thermal” odd-even mass difference rather than the S shape of the heat capacity.
In this Rapid Communication, using the thermal odd-even mass difference we extract the pairing gap of <sup>184</sup>W from the experimental level densities of <sup>183</sup>W, <sup>184</sup>W, and <sup>185</sup>W recently observed Bondarenko ; Sukhovoj . It is shown that the reduction of the thermal odd-even mass difference is interpreted as a signature of pairing transition, but not the S shape of the heat capacity.
Figure 1 shows the experimental level densities of <sup>183</sup>W, <sup>184</sup>W, and <sup>185</sup>W extracted from the two-step $`\gamma `$-cascade intensities. To study the thermal properties from the measured level densities, let us start from the partition function in the canonical ensemble with the Laplace transform of the level density $`\rho (E_i)`$
$`Z(T)={\displaystyle \underset{i=0}{\overset{\mathrm{}}{}}}\delta E_i\rho (E_i)\mathrm{e}^{E_i/T},`$ (1)
where $`E_i`$ are the excitation energies and $`\delta E_i`$ are the energy bins. Then any thermodynamical quantities $`O(Z,N,T)`$ can be evaluated by
$`O(Z,N,T)={\displaystyle \underset{i=0}{\overset{\mathrm{}}{}}}\delta E_i\rho (E_i)O_i\mathrm{e}^{E_i/T}/Z(T).`$ (2)
For instance, the thermal energy is expressed as
$`E(Z,N,T)={\displaystyle \underset{i=0}{\overset{\mathrm{}}{}}}\delta E_i\rho (E_i)E_i\mathrm{e}^{E_i/T}/Z(T).`$ (3)
The heat capacity are then given by
$`C(Z,N,T)`$ $`=`$ $`{\displaystyle \frac{E(Z,N,T)}{T}}.`$ (4)
We now introduce the thermal odd-even mass difference for neutrons defined by the following three-point indicator:
$`\mathrm{\Delta }_n^{(3)}(Z,N,T)`$ $`=`$ $`{\displaystyle \frac{(1)^N}{2}}[E_s(Z,N+1,T)`$
$`2E_s(Z,N,T)+E_s(Z,N1,T)],`$
where $`E_s`$ is a shifted thermal energy which is defined by substracting the the Coulomb energy from the binding energy at zero temperature. The odd-even mass difference at zero temperature is known theoretically and experimentally as an important quantity to evaluate the pairing correlations in a nucleus. The thermal odd-even mass difference would be also an indicator of the pairing correlations at finite temperature, and is obtained from the experimental energies and the level density as well as the heat capacity.
We can calculate the canonical partition function $`Z(T)`$ and the thermodynamical quantities from the measured level densities. Formally, the calculations using Eqs. (1)-(3) require infinite summation. However, the experimental level densities of Fig. 1 only cover the excitation energy up to $`68`$ MeV. In the evaluation of Eqs. (1)-(LABEL:eq:5), therefore, we extrapolate the plots of the experimental density to $``$ 40 MeV. Here, we use the level density formula of the back-shifted Fermi gas model in Refs. Gilbert
$`\rho (U)`$ $`=`$ $`f{\displaystyle \frac{\mathrm{exp}[2\sqrt{aU}]}{12\sqrt{2}a^{1/4}U^{5/4}\sigma }},`$ (6)
where the back-shifted energy is $`U=EE_1`$ and the spin cutoff parameter $`\sigma `$ is defined through $`\sigma ^2=0.0888A^{2/3}\sqrt{aU}`$. The level density parameter $`a`$ and the back-shifted parameter $`E_1`$ are defined by $`a=0.21A^{0.87}`$ MeV<sup>-1</sup> and $`E_1=C_1+\mathrm{\Delta }`$, respectively, where the correction factor is given by $`C_1=6.6A^{0.32}`$. The factor $`f`$ is determined so as to adjust the back-shifted level density to experimental one. The factors are, respectively, 0.3, 0.4, and 0.7 for <sup>183</sup>W, <sup>184</sup>W, and <sup>185</sup>W. The parameters $`\mathrm{\Delta }`$ for <sup>183</sup>W and <sup>185</sup>W are taken as the neutron pairing energies 0.61 and 0.72 MeV deduced from mass differences, respectively, and for <sup>184</sup>W it is fixed at 1.60 MeV.
The heat capacities of <sup>183</sup>W, <sup>184</sup>W, and <sup>185</sup>W are shown in Fig. 2. All the heat capacities exhibit the S shape with peaks around $`T=`$0.5 MeV. These heat capacities display characteristic behavior similar to those of <sup>161,162</sup>Dy and <sup>171,172</sup>Yb Schiller . Moreover, we notice that the heat capacity of <sup>184</sup>W around $`T=`$0.5 MeV is larger than those of <sup>183</sup>W and <sup>185</sup>W. The SPA + RPA Rossignoli and SMMC Liu calculations exhibited the S shape of the heat capacity and this odd-even effect where the heat capacity of an odd-mass nucleus is smaller than that of the adjacent even-even nuclei. We can see also the deviations from the heat capacity of the back-shifted Fermi gas model, which is approximated by the Bethe formula $`C=2aT`$. In the SMMC calculation Liu , Liu and Alhassid identified a signature of the pairing transition in the heat capacity that is correlated with the reduction of the number of neutron pairs as the temperature increases. In their calculation, the pairing correlations are suppressed for even-even nuclei, but not for adjacent odd-mass nuclei.
Figure 3 (a) shows the thermal pairing gap extracted from the thermal odd-even mass difference defined by Eq. (LABEL:eq:5) as a function of temperature. We find a sudden decrease of the thermal odd-even mass difference curve around $`T`$= 0.5 MeV, which is interpreted as a rapid breaking of nucleon Cooper pairs and the suppression of pairing correlations. We can now regard an inflection point of the curve of $`\mathrm{\Delta }_n^{(3)}`$ in Fig. 3 (a) as a signature of pairing transition, and called it “transition temperature” in our previous paper Kaneko . To see more precise position of the inflection point, we differentiate $`\mathrm{\Delta }_n^{(3)}(^{184}\mathrm{W})`$ with respect to temperature $`T`$. It is very important to note that the thermal odd-even mass difference has the following identity:
$`{\displaystyle \frac{\mathrm{\Delta }_n^{(3)}(Z,N,T)}{T}}=(1)^N\{C(Z,N,T)`$
$`{\displaystyle \frac{1}{2}}[C(Z,N+1,T)+C(Z,N1,T)]\}.`$
(7)
This identity means that the odd-even difference in the heat capacities represents a variation of the pairing correlations depending on temperature. In Fig. 3 (b), we can see that the peak of the odd-even difference in the heat capacities is a signature of the pairing transition, and the thermal odd-even mass difference is a good indicator for the pairing correlations. The extrapolation of the level dencity to $`40`$ MeV affects the curve of $`\mathrm{\Delta }_n^{(3)}`$ at high temperature in Fig. 3. However, the effects do not change the sudden decrease of the pairing gap $`\mathrm{\Delta }_n^{(3)}`$ around $`T=0.5`$ MeV.
To describe the above characteristic behavior of the heat capacity and the pairing gap extracted from the measured level densities of <sup>183</sup>W, <sup>184</sup>W, and <sup>185</sup>W, we consider a monopole pairing Hamiltonian
$`H`$ $`=`$ $`{\displaystyle \underset{k}{}}\epsilon _k(c_k^{}c_k+c_{\overline{k}}^{}c_{\overline{k}})GP^{}P,`$ (8)
where $`\epsilon _k`$ are the single-particle energies and $`P`$ is the pairing operator $`P=_kc_{\overline{k}}c_k`$. By means of the SPA+RPA Rossignoli ; Attias based on the Hubbard-Stratonovich transformation Hubbard , the canonical partition function with the number parity projection $`P_\sigma =(1+\sigma \mathrm{e}^{i\pi N})/2`$ Tuominen ; Rossignoli is given by
$`Z_\mathrm{c}^\sigma `$ $`=`$ $`\mathrm{Tr}[P_\sigma \mathrm{e}^{H/T}]_{\mathrm{SPA}+\mathrm{RPA}}`$ (9)
$`=`$ $`{\displaystyle \frac{2}{GT}}{\displaystyle _0^{\mathrm{}}}\mathrm{\Delta }𝑑\mathrm{\Delta }\mathrm{e}^{\mathrm{\Delta }^2/GT}Z_\sigma C_{\mathrm{RPA}}.`$
where $`\sigma `$ means the even or odd number parity. Here
$`Z_\sigma ={\displaystyle \frac{1}{2}}{\displaystyle \underset{k}{}}\mathrm{e}^{\gamma _k/T}(1+\mathrm{e}^{\lambda _k/T})^2`$
$`[1+\sigma {\displaystyle \underset{k^{}}{}}\mathrm{tanh}^2(\lambda _k^{}/T)],`$
$`C_{\mathrm{RPA}}={\displaystyle \underset{k}{}}{\displaystyle \frac{\omega _k\mathrm{sinh}[\lambda _k/T]}{2\lambda _k\mathrm{sinh}[\omega _k/2T]}},`$ (10)
where $`\lambda _k=\sqrt{\epsilon _k^2+\mathrm{\Delta }^2}`$, $`\epsilon _k^{}=\epsilon _k\mu G/2`$, and $`\gamma _k=\epsilon _k\mu \lambda _k`$. The $`\omega _k`$ are the conventional thermal RPA energies. If the factor $`C_{RPA}`$ is neglected, the SPA partition function is obtained. Then the thermal energy can be calculated from $`E=\mathrm{ln}Z_\mathrm{c}^\sigma /(1/T)`$. In this calculation, we use the single-particle energies $`\epsilon _k`$ given by an axially deformed Woods-Saxon potential with spin-orbit interaction Cwoik . The Woods-Saxon parameters are chosen so as to fit the experimental single-particle enegies extracted from the energy levels of odd nucleus <sup>133</sup>Sn (<sup>132</sup>Sn core plus one neutron). The deformation takes into account effects of a quadrupole-quadrupole interaction in the mean-field approximation. The deformation parameter $`\beta =0.23`$ can be estimated from the experimental value $`B(E2)=119.3`$ W.u. in the even-even nucleus <sup>184</sup>W. The 50 doubly degenerate single-particle energies are taken by assuming <sup>132</sup>Sn core, and we fix the pairing force strength at $`G=20/A`$ so that the BCS pairing gap $`\mathrm{\Delta }_{BCS}`$ reproduce the experimental odd-even mass difference $`\mathrm{\Delta }`$ = 0.83 MeV for <sup>184</sup>W at zero temperature. Figure 4 (a) shows the calculated heat capacities for <sup>183</sup>W, <sup>184</sup>W, and <sup>185</sup>W. We can see that the characteristic behavior of the extracted heat capacities in Fig. 2 are well described. Then the neutron pairing energy calculated from $`GP^{}P=GT\mathrm{ln}Z_c^\sigma /G`$ is also shown in Fig. 4 (b). We notice that the calculated heat capacity of <sup>183</sup>W deviates from that of <sup>184</sup>W, in contrast with that of <sup>185</sup>W, which is different from the result of Fig. 2. However, for high temperature this deviation would approach to zero according to Eq. (7) because the derivative of pairing gap $`\frac{\mathrm{\Delta }_n^{(3)}}{T}`$ converges to zero when $`T\mathrm{}`$. The monopole pairing model qualitatively explains the reduction of the thermal odd-even mass difference in Fig. 3 (a). As pointed out in our previous paper Kaneko , the peaks of the S-shaped heat capacities is quite close to the critical temperature $`T_c=0.57\mathrm{\Delta }`$ in the BCS theory.
In conclusion, we have extracted the pairing gap of <sup>184</sup>W at finite temperature from the experimental level densities of <sup>183</sup>W, <sup>184</sup>W, and <sup>185</sup>W using thermal odd-even mass difference. The extracted pairing gap exhibits a similar behavior to that of in previous theoretical model predictions, that is, the pairing gap decreases from the value at zero temperature with increasing temperature. The calculations show that while there is no sharp phase transition, the pairing gap decreases with increasing temperature. In particluar, it decreases rapidly around the BCS critical temperature. Thus, we can demonstrate that the thermal odd-even mass difference is a good indicator for the pairing transition at finite temperature as well as usual one at zero temperature. In this paper, the monopole pairing model was used with the SPA+RPA to describe the heat capacity and the pairing gap, where the deformation effect was taken into account. For these quantities, however, the fluctuations Rossignoli to the contributions of the quadrupole-quadrupole interaction should be taken into account in more realistic calculations. Further investigations are in progress. We suggest that the pairing correlations can be estimated from the measured level densities of the triplet nuclei with neutron number $`N+1,N,`$ and $`N1`$. We hope for further experiments to extract the pairing gap at finite temperature.
One of the authors (K.K.) thanks Dr. A. M. Sukhovoj for information of the experimental data and inspiring discussion. |
warning/0507/astro-ph0507575.html | ar5iv | text | # Multi-frequency analysis of neutralino dark matter annihilations in the Coma cluster
## 1 Introduction
Most of the matter content of the universe is in form of dark matter, whose presence is indicated by several astrophysical evidences (e.g., gravitational lensing, galaxy rotation curves, galaxy clusters masses) but whose nature is still elusive. On the cosmological side, the most recent results of observational cosmology, i.e. WMAP vs. distant SN Ia, indicate that the matter content of the universe is $`\mathrm{\Omega }_mh^2=0.135_{0.009}^{+0.008}`$ with a baryon density of $`\mathrm{\Omega }_bh^2=0.0224\pm 0.0009`$ (Spergel et al. (2003)). The combination of the available data on large scale structures (Ly-$`\alpha `$ forest analysis of the SDSS, the SDSS galaxy clustering) with the latest SNe and with the 1-st year WMAP CMB anisotropies can improve the determination of the cosmological parameters (Seljak, et al. (2004)) and hence allow us to set a concordance cosmological model.
We refer, in this paper, to a flat $`\mathrm{\Lambda }`$CDM cosmology with parameters chosen according to the global best fitting results derived in Seljak, et al. (2004) (see their Table 1, third column): we assume, in fact, that the present matter energy density is $`\mathrm{\Omega }_m=0.281`$, that the Hubble constant in units of 100 km s<sup>-1</sup> Mpc<sup>-1</sup> is $`h=0.71`$, that the present mean energy density in baryons is $`\mathrm{\Omega }_b=0.0233/h^2`$, with the only other significant extra matter term in cold dark matter $`\mathrm{\Omega }_{CDM}=\mathrm{\Omega }_m\mathrm{\Omega }_b`$, that our Universe has a flat geometry and a cosmological constant $`\mathrm{\Lambda }`$, i.e. $`\mathrm{\Omega }_\mathrm{\Lambda }=1\mathrm{\Omega }_m`$, and, finally, that the primordial power spectrum is scale invariant and is normalized to the value $`\sigma _8=0.897`$. This choice sets our framework, but it is not actually crucial for any of the results presented in the paper, which can be easily rescaled in case of a re-assessment of best-fit values of the cosmological parameters, and in particular of the value of $`\mathrm{\Omega }_{CDM}`$ (the present concordance cosmological model, while widely used, has been also criticized and questioned in the light of still unexplored systematics, see, e.g., Myers et al. (2003); Sadat et al. (2005); Lieu & Mittaz (2005); Copi et al. (2003))
The Coma cluster has been the first astrophysical laboratory for dark matter (DM) since the analysis of F. Zwicky (Zwicky (1933)). In this respect, we can consider the Coma cluster as an astrophysical benchmark case-study for DM. Modern observations have led to an increasingly sophisticated exploration of the DM distribution in the universe, now confirmed to be a dominant component (relative to the baryonic material) over scales ranging from those of galaxy halos to that of the particle horizon.
The nature of DM is not yet known and several detection techniques have been used so far. Obviously, direct detection is the cleanest and most decisive discriminant (see e.g. Munoz (2003) for a review). However, it would be interesting if astronomical techniques were to reveal some of the fundamental properties of DM particles. In fact, if DM is supposed to consist of fundamental particles which are weakly interacting, then their own interaction will lead to a number of astrophysical signatures (e.g., high-energy gamma-rays, electrons and positrons, protons and neutrinos and hence by their emission/interaction properties) indicative of their nature and composition.
These facts provide the basic motivations for our study, which is aimed to: i) describe the multi-wavelength signals of the presence of DM through the emission features of the secondary products of neutralino annihilation. These signals are of non-thermal nature and cover the entire electro-magnetic spectrum, from radio to gamma-ray frequencies; ii) indicate the best frequency windows where it will be possible to catch physical indications for the nature of DM; iii) apply this analysis to the largest bound containers of DM in the universe, i.e. galaxy clusters. We shall focus here on the case of the Coma cluster, a particularly rich and suitable laboratory for which an extended observational database is at hand.
### 1.1 The fundamental physics framework
Several candidates have been proposed as fundamental DM constituents, ranging from axions to light, MeV DM, from KK particles, branons, primordial BHs, mirror matter to supersymmetric WIMPs (see, e.g., Baltz, (2004), Bertone et al. (2004), and Bergstrom (2000) for recent reviews). In this paper we will assume that the main DM constituent is the lightest neutralino of the minimal supersymmetric extension of the Standard Model (MSSM). Although no experimental evidence in favor of supersymmetry has shown up to date, several theoretical motivations indicate that the MSSM is one of the best bets for new physics beyond the Standard Model. Intriguingly enough, and contrary to the majority of other particle physics candidates for DM, supersymmetry can unambiguously manifest itself in future accelerator experiments. Furthermore, provided neutralinos are stable and are the lightest supersymmetric particles, next generation direct detection experiments feature good chances to explore most of the neutralino-proton scattering cross section range predicted by supersymmetry.
A long standing issue in phenomenological studies of low-energy supersymmetry is traced to the parameterization of the supersymmetry breaking terms (see Chung et al. (2003) for a recent review). In this respect, two somehow complementary attitudes have been pursued. On the one hand, one can appeal to a (set of) underlying high energy principles to constrain the form of supersymmetry breaking term, possibly at some high energy (often at a grand unification) scale (see e.g. Baer et al. (2000)). The low energy setup is then derived through the renormalization group evolution of the supersymmetry breaking parameters down to the electroweak scale. Alternatively, one can directly face the most general possible low energy realization of the MSSM, and try to figure out whether general properties of supersymmetry phenomenology can be derived (see e.g. Profumo and Yaguna (2004)).
In this paper we will resort to both approaches. We will show that the final state products of neutralino pair annihilations show relatively few spectral patterns, and that any supersymmetric configuration can be thought as an interpolation among the extreme cases we shall consider here. The huge number of free parameters of the general MSSM are therefore effectively decoupled, and the only relevant physical properties are the final state products of neutralino pair annihilations, and the mass of the neutralino itself. We will indicate this first strategy as a bottom-up approach (see Sect.4.1 for details).
Since most phenomenological studies have been so far based on GUT-motivated models, and a wealth of results on accelerator physics, direct and indirect detection has accumulated within these frameworks, we decided to work out here, as well, the astrophysical consequences, for the system under consideration, of a few benchmark models. The latter have been chosen among the minimal supergravity (mSUGRA) models indicated in Battaglia et al. (2003) with the criterion of exemplifying the widest range of possibilities within that particular theoretical setup (see Sect.3.2 for details).
### 1.2 The astrophysical framework
To make our study quantitative, we will compare the predictions of the above mentioned neutralino models with the observational set-up of the Coma cluster, which represents the largest available observational database for a galaxy cluster. The total mass of Coma found within $`10h^1`$ Mpc from its center is $`M_{<10h^1Mpc}1.65\times 10^{15}M_{}`$ (Geller et al. (1999)). The assumption of hydrostatic equilibrium of the thermal intra-cluster gas in Coma provides a complementary estimate of its total mass enclosed in the radius $`r`$. A value $`M1.85\times 10^{15}M_{}`$ within $`5h_{50}^1`$ Mpc from the cluster center has been obtained from X-ray data (Hughes (1989)).
X-ray observations of Coma also yield detailed information about the thermal electrons population. We know that the hot thermal electrons are at a temperature $`k_BT_e=8.2\pm 0.4`$ keV (Arnaud et al. (2001)) and have a central density $`n_0=(3.42\pm 0.047)h_{70}^{1/2}\times 10^3`$ cm<sup>-3</sup>, with a spatial distribution fitted by a $`\beta `$-model, $`n(r)=n_0(1+r^2/r_c^2)^{3\beta /2}`$, with parameters $`r_c=10.5^{}\pm 0.6^{}`$ and $`\beta 0.75`$ (Briel et al. (1992)). Assuming spherical symmetry and the previous parameter values, the optical depth of the thermal gas in Coma is $`\tau _{th}5.54\times 10^3`$ and the pressure due to the thermal electron population is $`P_{th}2.8010^2`$ keV cm<sup>-3</sup>. The hot intra-cluster gas produces also a thermal SZ effect (Sunyaev & Zel’dovich (1972, 1980); see Birkinshaw (1999) for a general review) which has been observed over a wide frequency range, from $`32`$ to $`245`$ GHz (see DePetris et al. (2003) and references therein).
Beyond the presence of DM and thermal gas, Coma also shows hints for the presence of relativistic particles in its atmosphere. The main evidence for the presence of a non-thermal population of relativistic electrons comes from the observation of the diffuse radio halo at frequencies $`\nu _r30MHz5GHz`$ (Deiss et al. (1997); Thierbach et al. (2003)). The radio halo spectrum can be fitted by a power-law spectrum $`J_\nu \nu ^{1.35}`$ in the range $`30`$ MHz-$`1.4`$ GHz with a further steepening of the spectrum at higher radio frequencies. The radio halo of Coma has an extension of $`R_h0.9h_{70}^1`$ Mpc, and its surface brightness is quite flat in the inner 20 arcmin with a radial decline at larger angular distances (e.g., Colafrancesco et al. (2005)).
Other diffuse non-thermal emissions have been reported for Coma (as well as for a few other clusters) in the extreme UV (EUV) and in the hard X-ray (HXR) energy bands. The Coma flux observed in the $`65245`$ eV band (Lieu et al. (1996)) is $`36\%`$ above the expected flux from the thermal bremsstrahlung emission of the $`k_BT8.2`$ keV IC gas (Ensslin et al. (1998)) and it can be modeled with a power-law spectrum with an approximately constant slope $`1.75`$, in different spatial regions (Lieu et al. (1999), Ensslin et al. (1999), Bowyer et al. (2004)). The EUV excess in Coma has been unambiguously detected and it does not depend much on the data analysis procedure. The integrated flux in the energy band $`0.130.18`$ keV is $`F_{EUV}(4.1\pm 0.4)10^{12}`$ erg cm<sup>-2</sup>s<sup>-1</sup> (Bowyer et al. (2004)). According to the most recent analysis of the EUVE data (Bowyer et al. (2004)), the EUV excess seems to be spatially concentrated in the inner region ($`\theta <\text{ }1520`$ arcmin) of Coma (see also Bonamente et al. (2003)). The nature of this excess is not definitely determined since both thermal and non-thermal models are able to reproduce the observed EUV flux. However, the analysis of Bowyer et al. (2004) seems to favour a non-thermal origin of the EUV excess in Coma generated by an additional population of secondary electrons. A soft X-ray (SXR) excess (in the energy range $`0.10.245`$ keV) has been also detected in the outer region ($`20^{}<\theta <90^{}`$) of Coma (Bonamente et al. (2003); Finoguenov et al. (2003)). The spectral features of this SXR excess seem to be more consistent with a thermal nature of the SXR emission, while a non-thermal model is not able to reproduce accurately the SXR data (e.g., Bonamente et al. (2003)). The SXR emission from the outskirts of Coma has been fitted in a scenario in which the thermal gas at $`k_BT_e0.2`$ keV with $`0.1`$ solar abundance (see, e.g. Finoguenov et al. (2003) who identify the warm gas with a WHIM component) resides in the low-density filaments predicted to form around clusters as a result of the evolution of the large-scale web-like structure of the universe (see Bonamente et al. (2003)). It has been noticed, however, that the WHIM component cannot reproduce, by itself, the Coma SXR excess because it would produce a SXR emission by far lower (see Mittaz et al. (2004)), and thus one is forced to assume a large amount of warm gas in the outskirts of Coma. Thus, it seems that the available EUV and SXR data indicate (at least) two different electron populations: a non-thermal one, likely yielding the centrally concentrated EUV excess and a thermal (likely warm) one, providing the peripherically located SXR excess. In this paper, we will compare our models with the EUV excess only, which is intimately related to Coma being spatially concentrated towards the inner region of the cluster.
There is also evidence of a hard X-ray (HXR) emission observed towards the direction of Coma with the BeppoSAX-PDS (Fusco-Femiano et al. (1999), Fusco-Femiano et al. (2004)) and with the ROSSI-XTE experiments (Rephaeli et al. (1999)). Both these measurements indicate an excess over the thermal emission which amounts to $`F_{(2080)keV}=(1.5\pm 0.5)10^{11}`$ erg s<sup>-1</sup> cm<sup>-2</sup> (Fusco-Femiano et al. (2004)). It must be mentioned, for the sake of completeness, that the HXR excess of Coma is still controversial (see Rossetti and Molendi (2004)), but, for the aim of our discussion, it could at worst be regarded as an upper limit. The nature of the HXR emission of Coma is not yet fully understood.
Finally, a gamma-ray upper limit of $`F(>100MeV)3.2\times 10^8`$ pho cm<sup>-2</sup> has been derived for Coma from EGRET observations (Sreekumar et al. (1996), Reimer et al. (2004)).
The current evidence for the radio-halo emission features of Coma has been interpreted as synchrotron emission from a population of primary relativistic electrons which are subject to a continuous re-acceleration process supposedly triggered by merging shock and/or intracluster turbulence (e.g., Brunetti et al. (2004)). The EUV and HXR emission excesses are currently interpreted as Inverse Compton scattering (ICS) emission from either primary or secondary electrons. Alternative modeling has been proposed in terms of suprathermal electron bremsstrahlung emission for the HXR emission of Coma (see Ensslin et al. (1999); Kempner & Sarazin (2000), see also Petrosian (2001) for a critical discussion) and in this case the EUV emission should be produced either by a different relativistic electron population or by a warm thermal population both concentrated towards the cluster center. Lastly, models in which the EUV and HXR emission can be reproduced by synchrotron emission from the interaction of Ultra High Energy cosmic rays and/or photons (Timokhin et al. (2003); Inoue et al. (2005)) have also been presented. The situation is far from being completely clear and several problems still stand on both the observational and theoretical sides of the issue.
Since DM is abundantly present in Coma and relativistic particles are among the main annihilation products, we explore here the effect of DM annihilation on the multi-frequency spectral energy distribution (SED) of Coma. The plan of the paper is the following. We discuss in Sect. 2 the DM halo models for the Coma cluster, the set of best fitting parameters for the DM distribution, and the role of sub-halos. The annihilation features of neutralinos and the main annihilation products are discussed in Sect. 3. The multi-frequency signals of DM annihilation are presented and discussed in details in Sect. 4 and 5, while the details of the transport and diffusion properties of the secondary particles are described in the Appendix A, together with the derivation of the equilibrium spectrum of relativistic particles in Coma. The conclusions of our analysis and the outline for forthcoming astrophysical searches for DM signals in galaxy clusters are presented in the final Sect.6.
## 2 A $`\mathrm{\Lambda }`$CDM model for the Coma cluster
To describe the DM halo profile of the Coma cluster we refer, as a general setup, to the $`\mathrm{\Lambda }`$CDM model for structure formation, implementing results of galaxy cluster formation obtained from N-body simulations. Free parameters are fitted against the available dynamical information and are compared to the predictions of this scheme. Substructures will play a major role when we will discuss the predictions for signals of DM annihilations. In this respect, the picture derived from simulations is less clean and, hence, we will describe in details our set of assumptions.
### 2.1 The dark matter halo profile
To describe the DM halo profile of the Coma cluster we consider the limit in which the mean DM distribution in Coma can be regarded as spherically symmetric and represented by the parametric radial density profile:
$$\rho (r)=\rho ^{}g(r/a).$$
(1)
Two schemes are adopted to choose the function $`g(x)`$ introduced here. In the first one, we assume that $`g(x)`$ can be directly inferred as the function setting the universal shape of DM halos found in numerical N-body simulations of hierarchical clustering. We are assuming, hence, that the DM profile is essentially unaltered from the stage preceding the baryon collapse, which is – strictly speaking – the picture provided by the simulations for the present-day cluster morphology. A few forms for the universal DM profile have been proposed in the literature: we implement here the non-singular form (which we label as N04 profile) extrapolated by Navarro et al. (2004):
$$g_{N04}(x)=\mathrm{exp}[2/\alpha (x^\alpha 1)]\mathrm{with}\alpha 0.17,$$
(2)
and the shape with a mild singularity towards its center proposed by Diemand et al. (2005) (labeled here as D05 profile):
$$g_{D05}(x)=\frac{1}{x^\gamma (1+x)^{3\gamma }}\mathrm{with}\gamma 1.2.$$
(3)
The other extreme scheme is a picture in which the baryon infall induces a large transfer of angular momentum between the luminous and the dark components of the cosmic structure, with significant modification of the shape of the DM profile in its inner region. According to a recent model (El-Zant et al. (2001)), baryons might sink in the central part of DM halos after getting clumped into dense gas clouds, with the halo density profile in the final configuration found to be described by a profile (labeled here as B profile) with a large core radius (see, e.g., Burkert (1995)):
$$g_B(x)=\frac{1}{(1+x)(1+x^2)}.$$
(4)
Once the shape of the DM profile is chosen, the radial density profile in Eq. (1) is fully specified by two parameters: the length-scale $`a`$ and the normalization parameter $`\rho ^{}`$. It is, however, useful to describe the density profile model by other two parameters, i.e., its virial mass $`M_{vir}`$ and concentration parameter $`c_{vir}`$. For the latter parameter, we adopt here the definition by Bullock et al. (2001). We introduce the virial radius $`R_{vir}`$ of a halo of mass $`M_{vir}`$ as the radius within which the mean density of the halo is equal to the virial overdensity $`\mathrm{\Delta }_{vir}`$ times the mean background density $`\overline{\rho }=\mathrm{\Omega }_m\rho _c`$:
$$M_{vir}\frac{4\pi }{3}\mathrm{\Delta }_{vir}\overline{\rho }R_{vir}^3.$$
(5)
We assume here that the virial overdensity can be approximated by the expression (see Bryan & Norman (1998)), appropriate for a flat cosmology,
$$\mathrm{\Delta }_{vir}\frac{(18\pi ^2+82x39x^2)}{1x},$$
(6)
with $`x\mathrm{\Omega }_m(z)1`$. In our cosmological setup we find at $`z=0`$, $`\mathrm{\Delta }_{vir}343`$ (we refer to Colafrancesco et al. (1994), Colafrancesco et al. (1997) for a general derivation of the virial overdensity in different cosmological models). The concentration parameter is then defined as
$$c_{vir}=\frac{R_{vir}}{r_2}\frac{R_{vir}}{x_2a},$$
(7)
with $`r_2`$ the radius at which the effective logarithmic slope of the profile is $`2`$. We find that $`x_2=1`$ for the N04 profile (see Eq. 2), $`x_2=2\gamma `$ for D05 profile (see Eq. 3), and $`x_21.52`$ for the Burkert profile (see Eq. 4).
Since the first numerical results with large statistics became available (Navarro et al. (1997)), it has been realized that, at any given redshift, there is a strong correlation between $`c_{vir}`$ and $`M_{vir}`$, with larger concentrations found in lighter halos. This trend may be intuitively explained by the fact that mean overdensities in halos should be correlated with the mean background densities at the time of collapse, and in the hierarchical structure formation model small objects form first, when the Universe was indeed denser. The correlation between $`c_{vir}`$ and $`M_{vir}`$ is relevant in our context at two levels: i) when discussing the mean density profile of Coma and, ii) when including substructures. Hence, we will review this relevant issue here and we will apply it to the present case of Coma. Bullock et al. (2001) proposed a model to describe this correlation, improving on the toy model originally outlined in Navarro et al. (1997). A collapse redshift $`z_c`$ is assigned, on average, to each halo of mass $`M`$ at the epoch $`z`$ through the relation $`M_{}(z_c)FM`$. Here it is postulated that a fixed fraction $`F`$ of $`M`$ (following Wechsler et al. (2001) we choose $`F=0.015`$) corresponds to the typical collapsing mass $`M_{}`$, as defined implicitly by $`\sigma \left(M_{}(z)\right)=\delta _{sc}(z)`$, with $`\delta _{sc}`$ being the critical overdensity required for collapse in the spherical model and $`\sigma (M)`$ being the present-day rms density fluctuation in spheres containing a mean mass $`M`$ (see, e.g., Peebles (1980)). An expression for $`\delta _{sc}`$ is given, e.g., in Eke et al. (1996). The rms fluctuation $`\sigma (M)`$ is related to the fluctuation power spectrum $`P(k)`$ (see e.g. Peebles (1993)) by
$$\sigma ^2(M)d^3k\stackrel{~}{W}^2(kR)P(k),$$
(8)
where $`\stackrel{~}{W}`$ is the top-hat window function on the scale $`R^3=3M/4\pi \overline{\rho }`$ with $`\overline{\rho }`$ the mean (proper) matter density, i.e. $`\overline{\rho }=\mathrm{\Omega }_m\rho _c`$ with $`\rho _c`$ the critical density. The power spectrum $`P(k)`$ is parametrized as $`P(k)k^nT^2(k)`$ in terms of the primordial power-spectrum shape $`k^n`$ and of the transfer function $`T^2(k)`$ associated to the specific DM scenario. We fix the primordial spectral index $`n=1`$ and we take the transfer function $`T^2(k)`$ given by Bardeen et al. (1986) for an adiabatic CDM model, with the shape parameter modified to include baryonic matter according to the prescription in, e.g. Peacock (1999) (see their eqs.15.84 and 15.85) and introducing a multiplicative exponential cutoff at large $`k`$ corresponding to the free-streaming scale for WIMPs (Hofmann et al. 2001, Chen et al. 2001, Green et al. 2005, Diemand et al. (2005)). The spectrum $`P(k)`$ is normalized to the value $`\sigma _8=0.897`$ as was quoted above.
The toy model of Bullock et al. (2001) prescribes a one to one correspondence between the density of the Universe at the collapse redshift $`z_c`$ of the DM halo and a characteristic density of the halo at the redshift $`z`$; it follows that, on average, the concentration parameter is given by
$$c_{vir}(M,z)=K\frac{1+z_c}{1+z}=\frac{c_{vir}(M,z=0)}{(1+z)},$$
(9)
with $`K`$ being a constant (i.e. independent of $`M`$ and cosmology) to be fitted to the results of the N-body simulations. We plot in Fig. 1 the dependence of $`c_{vir}`$ on the halo mass $`M`$, at $`z=0`$, according to the toy model of Bullock et al. (2001) as extrapolated down to the free-streaming mass scale for DM halos made of WIMPs, i.e. around $`10^6M_{}`$ (see Hofmann et al. 2001, Chen et al. 2001, Green et al. 2005, Diemand et al. (2005)). The predictions are compared to the results of a few sets of N-body simulations: we use “data” points and relative error bars from Bullock et al. (2001) (representing a binning in mass of results for a large sample of simulated halos; in each mass bin, the marker and the error bars correspond, respectively, to the peak and the 68% width in the $`c_{vir}`$ distribution) to determine the parameter $`K`$. The same value will be used to infer the mean $`c_{vir}`$ predicted in our cosmological setup. Other “datasets” refer actually to different values of $`\sigma _8`$ and different redshifts $`z`$ ($`z=26`$ for the two minihalos fitted in Fig. 2 of Diemand et al. (2005) and for the upper bound in the range up to $`10M_{}`$ quoted in the same paper; $`z=3`$ for the sample from Colin et al. (2004)) and have been extrapolated, consistently with our prescriptions, to $`z=0`$ and $`\sigma _8=1`$. Since small objects tend to collapse all at the same redshift, the dependence on mass of the concentration parameters flattens at small masses; the mean asymptotic value we find is slightly larger than the typical values found in Diemand et al. (2005), but it is still consistent with that analysis.
An alternative toy-model to describe the relation between $`c_{vir}`$ and $`M`$ has been discussed by Eke, Navarro and Steinmetz (Eke et al. (2001), hereafter ENS model). The relation they propose has a similar scaling in $`z`$, but with a different definition of the collapse redshift $`z_c`$ and a milder dependence of $`c_{vir}`$ on $`M`$. In our notation, they define $`z_c`$ through the equation
$$D(z_c)\sigma _{\mathrm{eff}}(M_p)=\frac{1}{C_\sigma }$$
(10)
where $`D(z)`$ represents the linear theory growth factor, and $`\sigma _{\mathrm{eff}}`$ is an ‘effective’ amplitude of the power spectrum on scale $`M`$:
$$\sigma _{\mathrm{eff}}(M)=\sigma (M)\left(\frac{d\mathrm{ln}(\sigma )}{d\mathrm{ln}(M)}(M)\right)=\frac{d\sigma }{dM}M$$
(11)
which modulates $`\sigma (M)`$ and makes $`z_c`$ dependent on both the amplitude and on the shape of the power spectrum, rather than just on the amplitude, as in the toy model of Bullock et al. (2001). Finally, in Eq. (10), $`M_p`$ is assumed to be the mass of the halo contained within the radius at which the circular velocity reaches its maximum, while $`C_\sigma `$ is a free parameter (independent of $`M`$ and cosmology) which we will fit again to the “data” set in Bullock et al. (2001). With such a definition of $`z_c`$ it follows that, on average, $`c_{vir}`$ can be expressed as:
$$c_{vir}(M,z)=\left(\frac{\mathrm{\Delta }_{vir}(z_c)\mathrm{\Omega }_M(z)}{\mathrm{\Delta }_{vir}(z)\mathrm{\Omega }_M(z_c)}\right)^{1/3}\frac{1+z_c}{1+z}.$$
(12)
As shown in Fig. 1, the dependence of $`c_{vir}`$ on $`M`$ given by Eq.(12) above is weaker than that obtained in the Bullock et al. (2001) toy-model, with a significant mismatch in the extrapolation already with respect to the sample from Colin et al. (2004) and an even larger mismatch in the low mass end. Moreover, the extrapolation breaks down when the logarithmic derivative of the $`\sigma (M)`$ becomes very small, in the regime when $`P(k)`$ scales as $`k^3`$. Note also that predictions in this model are rather sensitive to the specific spectrum $`P(k)`$ assumed (in particular the form in the public release of the ENS numerical code gives slightly larger values of $`c_{vir}`$ in its low mass end, around a value $`c_{vir}40`$ (we checked that implementing our fitting function for the power spectrum, we recover our trend).
### 2.2 Fitting the halo parameters of Coma
For a given shape of the halo profile we make a fit of the parameters $`M_{vir}`$ and $`c_{vir}`$ against the available dynamical constraints for Coma. We consider two bounds on the total mass of the cluster at large radii, as inferred with techniques largely insensitive to the details of the mass profile in its inner region. In Geller et al. (1999), a total mass
$$M(r<10\mathrm{h}^1\mathrm{Mpc})=(1.65\pm 0.41)\mathrm{\hspace{0.17em}10}^{15}\mathrm{h}^1M_{}$$
(13)
is derived mapping the caustics in redshift space of galaxies infalling in Coma on nearly radial orbits. Several authors derived mass budgets for Coma using optical data and applying the virial theorem, or using X-ray data and assuming hydrostatic equilibrium. We consider the bound derived by Hughes (1989), cross-correlating such techniques:
$$M(r<5\mathrm{h}_{50}^1\mathrm{Mpc})=(1.85\pm 0.25)\mathrm{\hspace{0.17em}10}^{15}\mathrm{h}_{50}^1M_{},$$
(14)
where $`h_{50}`$ is the Hubble constant in units of 50 km s<sup>-1</sup> Mpc<sup>-1</sup>.
In our discussion some information on the inner shape of the mass profile in Coma is also important: we implement here the constraint that can derived by studying the velocity moments of a given tracer population in the cluster. As the most reliable observable quantity one can consider the projection along the line of sight of the radial velocity dispersion of the population; under the assumption of spherical symmetry and without bulk rotation, this is related to the total mass profile $`M(r)`$ by the expression (Binney & Mamon (1982); Lokas & Mamon (2003)):
$`\sigma _{\mathrm{los}}^2(R)`$ $`=`$ $`{\displaystyle \frac{2G}{I(R)}}{\displaystyle _R^{\mathrm{}}}dr^{}\nu (r^{})M(r^{})(r^{})^{2\beta 2}`$ (15)
$`\times {\displaystyle _R^r^{}}\mathrm{d}r(1\beta {\displaystyle \frac{R^2}{r^2}}){\displaystyle \frac{r^{2\beta +1}}{\sqrt{r^2R^2}}},`$
where $`\nu (r)`$ is the density profile of the tracer population and $`I(R)`$ represents its surface density at the projected radius $`R`$. In the derivation of Eq. (15), a constant-over-radius anisotropy parameter $`\beta `$ defined as
$$\beta 1\frac{\sigma _\theta ^2(r)}{\sigma _r^2(r)},$$
(16)
has been assumed with $`\sigma _r^2`$ and $`\sigma _\theta ^2`$ being, respectively, the radial and tangential velocity dispersion ($`\beta =1`$ denotes the case of purely radial orbits, $`\beta =0`$ that of system with isotropic velocity dispersion, while $`\beta \mathrm{}`$ labels circular orbits). Following Lokas & Mamon (2003), we take as tracer population that of the E-S0 galaxies, whose line of sight velocity dispersion has been mapped, according to Gaussian distribution, in nine radial bins from $`4^{}`$ out to $`190^{}`$ (see Fig. 3 in Lokas & Mamon (2003)), and whose density profile can be described by the fitting function:
$$\nu (r)\frac{1}{(r/r_\mathrm{S})(1+r/r_\mathrm{S})^2},$$
(17)
with $`r_\mathrm{S}=7.^{}05`$. Constraints to the DM profile are obtained through its contribution to $`M(r)`$, in which we include the terms due to spiral and E-S0 galaxies (each one with the appropriate density profile normalized to the observed luminosity through an appropriate mass-to-light ratio), and the gas component (as inferred from the X-ray surface brightness distribution) whose number density profile can be described by the fitting function:
$$n(r)=n_0\left[1+\left(\frac{r}{r_\mathrm{c}}\right)^2\right]^{1.5b},$$
(18)
with $`n_0=3.42\times 10^3`$ cm<sup>-3</sup>, $`r_\mathrm{c}=10.^{}5`$ and $`b=0.75`$ (Briel et al. (1992)).
To compare a model with such datasets, we build a reduced $`\chi ^2`$-like variable of the form:
$`\chi _r^2`$ $`=`$ $`{\displaystyle \frac{1}{2}}[{\displaystyle \frac{1}{2}}{\displaystyle \underset{i=1}{\overset{2}{}}}{\displaystyle \frac{\left(M(r<r_i)M_i\right)^2}{(\mathrm{\Delta }M_i)^2}}`$ (19)
$`+{\displaystyle \frac{1}{9}}{\displaystyle \underset{j=1}{\overset{9}{}}}{\displaystyle \frac{\left(\sigma _{\mathrm{los}}(R_j)\sigma _{\mathrm{los}}^j\right)^2}{(\mathrm{\Delta }\sigma _{\mathrm{los}}^j)^2}}]`$
where the index $`i`$ in the first sum runs over the constraints given in Eqs. (13) and (14), while, in the second sum, we include the nine radial bins over which the line of sight velocity dispersion of E-S0 galaxies and its standard deviation has been estimated. Weight factors have been added to give the same statistical weight to each of the two classes of constraints, see, e.g., Dehnen & Binney (1998) where an analogous procedure has been adopted.
Nonetheless, we have derived in Fig. 2 the 1 $`\sigma `$, 2 $`\sigma `$ and 3 $`\sigma `$ contours in the $`(M_{vir},c_{vir})`$ plane for the Navarro et al. halo profile (Eq. (2) and for the Diemand et al. halo profile (Eq. 3). In Fig. 3 we show the analogous contours for the Burkert profile (Eq. (4)). In all these cases we have performed the fit of the line-of-sight radial velocity dispersion of E-S0 galaxies assuming that this system has an isotropic velocity dispersion, i.e. we have taken $`\beta =0`$. Best fitting values are found at $`M_{vir}0.910^{15}M_{}h^1`$ and $`c_{vir}10`$ (that we consider, hence, as reference values in the following analysis), not too far from the mean value expected from models sketching the correlation between these two parameters in the $`\mathrm{\Lambda }`$CDM picture. We show in Figs. 2 and 3 the predictions of such correlation in the models of Bullock et al. (solid line) and of Eke et al. (2001) (dashed line).
### 2.3 Substructures in the Coma cluster
Since the astrophysical signals produced by WIMP pair annihilation scale with the square of the WIMP density, any local overdensity does play a role (see e.g. Bergstrom et al. (1998) and references therein). To discuss substructures in the Coma cluster, analogously to the general picture introduced above for DM halos, we label a subhalo through its virial mass $`M_s`$ and its concentration parameter $`c_s`$ (or equivalently a typical density and length scale, $`\rho _s^{}`$ and $`a_s`$). The subhalo profile shape is considered here to be spherical and of the same form as for the parent halo. Finally, as for the mean DM density profile, the distribution of subhalos in Coma is taken to be spherically symmetric. The subhalo number density probability distribution can then be fully specified through $`M_s`$, $`c_s`$ and the radial coordinate for the subhalo position $`r`$. To our purposes, it is sufficient to consider the simplified case when the dependence on these three parameters can be factorized, i.e.:
$$\frac{dn_s}{dr^3dM_sdc_s}=p_s(r)\frac{dn_s}{dM_s}(M_s)𝒫_s(c_s).$$
(20)
Here we have introduced a subhalo mass function, independent of radius, which is assumed to be of the form:
$$\frac{dn_s}{dM_s}=\frac{A(M_{vir})}{M_s^{1.9}}\mathrm{exp}\left[\left(\frac{M_s}{M_{cut}}\right)^{2/3}\right],$$
(21)
Diemand et al. (2005) where $`M_{cut}`$ is the free streaming cutoff mass (Hofmann et al. 2001, Chen et al. 2001, Green et al. 2005, Diemand et al. (2005)), while the normalization $`A(M_{vir})`$ is derived imposing that the total mass in subhalos is a fraction $`f_s`$ of the total virial mass $`M_{vir}`$ of the parent halo, i.e.
$$_{M_{cut}}^{M_{vir}}𝑑M_s\frac{dn_s}{dM_s}M_s=f_sM_{vir}.$$
(22)
According to Diemand et al. (2005), $`f_s`$ is about 50% for a Milky Way size halo, and we will assume that the same holds for Coma. The quantity $`𝒫_s(c_s)`$ is a log-normal distribution in concentration parameters around a mean value set by the substructure mass; the trend linking the mean $`c_s`$ to $`M_s`$ is expected to be analogous to that sketched above for parent halos with the Bullock et al. or ENS toy models, except that, on average, substructures collapsed in higher density environments and suffered tidal stripping. Both of these effects go in the direction of driving larger concentrations, as observed in the numerical simulation of Bullock et al. (2001), where it is shown that, on average and for $`M510^{11}M_{}`$ objects, the concentration parameter in subhalos is found to be a factor of $`1.5`$ larger than for halos. We make here the simplified ansazt:
$$c_s(M_s)=F_sc_{vir}(M_{vir})\mathrm{with}M_s=M_{vir},$$
(23)
where, for simplicity, we assume that the enhancement factor $`F_s`$ does not depend on $`M_s`$. Following again Bullock et al. (2001), the $`1\sigma `$ deviation $`\mathrm{\Delta }(\mathrm{log}_{10}c_s)`$ around the mean in the log-normal distribution $`𝒫_s(c_s)`$, is assumed to be independent of $`M_s`$ and of cosmology, and to be, numerically, $`\mathrm{\Delta }(\mathrm{log}_{10}c_s)=0.14`$.
Finally, we have to specify the spatial distribution of substructures within the cluster. Numerical simulations, tracing tidal stripping, find radial distributions which are significantly less concentrated than that of the smooth DM component. This radial bias is introduced here assuming that:
$$p_s(r)g(r/a^{}),$$
(24)
with $`g`$ being the same functional form introduced above for the parent halo, but with $`a^{}`$ much larger than the length scale $`a`$ found for Coma. Following Nagai & Kravtsov (2005), we fix $`a^{}/a7`$. Since the fraction $`f_s`$ of DM in subhalos refers to structures within the virial radius, the normalization of $`p_s(r)`$ follows from the requirement
$$4\pi _0^{R_{vir}}r^2p_s(r)=1.$$
(25)
## 3 Neutralino annihilations in Coma
### 3.1 Statistical properties
Having set the reference particle physics framework and specified the distribution of DM particles, we can now introduce the source function from neutralino pair annihilations. For any stable particle species $`i`$, generated promptly in the annihilation or produced in the decay and fragmentation processes of the annihilation yields, the source function $`Q_i(r,E)`$ gives the number of particles per unit time, energy and volume element produced locally in space:
$$Q_i(r,E)=\sigma v_0\underset{f}{}\frac{dN_i^f}{dE}(E)B_f𝒩_{\mathrm{pairs}}(r),$$
(26)
where $`\sigma v_0`$ is the neutralino annihilation rate at zero temperature. The sum is over all kinematically allowed annihilation final states $`f`$, each with a branching ratio $`B_f`$ and a spectral distribution $`dN_i^f/dE`$, and $`𝒩_{\mathrm{pairs}}(r)`$ is the number density of neutralino pairs at a given radius $`r`$ (i.e., the number of DM particles pairs per volume element squared). The particle physics framework sets the quantity $`\sigma v_0`$ and the list of $`B_f`$. Since the neutralino is a Majorana fermion light fermion final states are suppressed, while – depending on mass and composition – the dominant channels are either those with heavy fermions or those with gauge and Higgs bosons. The spectral functions $`dN_i^f/dE`$ are inferred from the results of MonteCarlo codes, namely the Pythia (Sjöstrand 1994, (1995)) 6.154, as included in the DarkSUSY package (Gondolo et al. (2004)). Finally, $`𝒩_{\mathrm{pairs}}(r)`$ is obtained by summing the contribution from the smooth DM component, which we write here as the difference between the cumulative profile and the term that at a given radius is bound in subhalos, and the contributions from each subhalo, in the limit of unresolved substructures and in view of the fact that we consider only spherically averaged observables:
$`𝒩_{\mathrm{pairs}}(r)`$ $`=`$ $`[{\displaystyle \frac{\left(\rho ^{}g(r/a)f_sM_{vir}p_s(r)\right)^2}{2M_\chi ^2}}`$ (27)
$`+p_s(r){\displaystyle 𝑑M_s\frac{dn_s}{dM_s}𝑑c_s^{}𝒫_s\left(c_s^{}(M_s)\right)}`$
$`\times {\displaystyle }d^3r_s{\displaystyle \frac{\left(\rho _s^{}g(r_s/a_s)\right)^2}{2M_\chi ^2}}].`$
This quantity can be rewritten in the more compact form:
$`𝒩_{\mathrm{pairs}}(r)`$ $`=`$ $`{\displaystyle \frac{\overline{\rho }^2}{2M_\chi ^2}}[{\displaystyle \frac{\left(\rho ^{}g(r/a)f_s\stackrel{~}{\rho }_sg(r/a^{})\right)^2}{\overline{\rho }^2}}`$ (28)
$`+f_s\mathrm{\Delta }^2{\displaystyle \frac{\stackrel{~}{\rho }_sg(r/a^{})}{\overline{\rho }}}],`$
where we have normalized densities to the present-day mean matter density in the Universe $`\overline{\rho }`$, and we have defined the quantity:
$`f_s\mathrm{\Delta }^2`$ $``$ $`{\displaystyle \frac{𝑑M_s\frac{dn_s}{dM_s}M_s\mathrm{\Delta }_{M_s}^2(M_s)}{M_{vir}}}`$ (29)
$`=`$ $`f_s{\displaystyle \frac{𝑑M_s\frac{dn_s}{dM_s}M_s\mathrm{\Delta }_{M_s}^2(M_s)}{𝑑M_s\frac{dn_s}{dM_s}M_s}},`$ (30)
with
$$\mathrm{\Delta }_{M_s}^2(M_s)\frac{\mathrm{\Delta }_{vir}(z)}{3}𝑑c_s^{}𝒫_s\left(c_s^{}\right)\frac{I_2(c_s^{}x_2)}{\left[I_1(c_s^{}x_2)\right]^2}(c_s^{}x_2)^3$$
(31)
and
$$I_n(x)=_0^x𝑑yy^2\left[g(y)\right]^n.$$
(32)
Such definitions are useful since $`\mathrm{\Delta }_{M_s}^2`$ gives the average enhancement in the source due to a subhalo of mass $`M_s`$, while $`\mathrm{\Delta }^2`$ is the sum over all such contributions weighted over the subhalo mass function times mass. Finally, in Eq. (28) we have also introduced the quantity:
$$\stackrel{~}{\rho }_s\frac{M_{vir}}{4\pi (a^{})^3I_1\left(R_{vir}/a^{}\right)}.$$
(33)
In the limit in which the radial distribution of substructures traces the DM profile, i.e. $`a^{}=a`$, $`\stackrel{~}{\rho }_s`$ becomes equal to the halo normalization parameter $`\rho ^{}`$.
We show in Fig. 4 the scaling of the average enhancement $`\mathrm{\Delta }_{M_s}^2`$ in the source function versus the subhalo mass $`M_s`$. We have considered the three halo models introduced in the previous Section, i.e. the N04, D05 and Burkert profiles, for the two toy models describing the scaling of concentration parameter with mass, i.e. the Bullock et al. and the ENS schemes, as well as two sample values for the ratio $`F_s`$ between the average concentration parameter in subhalos and that in halos of equal mass. In each setup, going to smaller and smaller values of $`M_s`$, the average enhancement $`\mathrm{\Delta }_{M_s}^2`$ increases and then flattens out at the mass scale below which all structures tend to collapse at the same epoch, and hence have equal concentration parameter.
In Fig. 5 we show the scaling of the weighted enhancement $`\mathrm{\Delta }^2`$ in the source function due to subhalos versus the ratio between concentration parameter in subhalos to concentration parameter in halos at equal mass $`F_s`$; we give results for the usual set of halo profiles considered in our approach. Analogously to the enhancement for a fixed mass shown in the previous plot, $`\mathrm{\Delta }^2`$ is very sensitive to the scaling of the concentration parameter and hence we find a sharp dependence of $`\mathrm{\Delta }^2`$ on $`F_s`$. The fractional contribution per logarithmic interval in subhalo mass $`M_s`$ to $`\mathrm{\Delta }^2`$ is also shown in Fig. 5 for four sample cases. Note that, although the factorization in the probability distribution for clumps in the radial coordinate and mass (plus the assumption that $`F_s`$ does not depend on mass) are a crude approximation, what we actually need in our discussion is $`F_s`$ and the radial distribution for subhalos at the peak of the distribution shown in Fig. 5: unfortunately we cannot read out this from numerical simulations.
Fig. 6 shows the number density of neutralino pairs (we set here the neutralino mass to $`M_\chi =100`$ GeV) as a function of the distance from the center of Coma for the three representative halo profiles introduced here, i.e. the N04, D05 and Burkert profile in their best fit model, and a sample configuration for the subhalo parameters. For the D05 and N04 profiles, the central enhancement increases the integrated source function by a factor $`6`$ with respect to the Burkert profile, but this takes place on such a small angular scale that from the observational point of view it is like adding a point source at the center of the cluster. The enhancement of the annihilation signals from subhalos comes instead from large radii. This means that the enhancement from subhalos largely influences the results when the neutralino source is extended. This is the case of galaxy clusters, and more specifically of the Coma cluster which is our target in this paper.
### 3.2 Source functions spectral properties: generalities and supersymmetric benchmarks
The spectral properties of secondary products of DM annihilations depend only, prior to diffusion and energy losses, on the DM particle mass $`M_\chi `$ and on the branching ratio $`BR(\chi \chi f)`$ for the final state $`f`$ in the DM pair-annihilation. The DM particle physics model further sets the magnitude of the thermally averaged pair annihilation cross section times the relative DM particles velocity, $`\sigma v_0`$ at $`T=0`$.
The range of neutralino masses and pair annihilation cross sections in the most general supersymmetric DM setup is extremely wide. Neutralinos as light as few GeV (see Bottino et al. (2003)) and as heavy as hundreds of TeV (see Profumo (2005)) can account for the observed CDM density through thermal production mechanisms, and essentially no constraints apply in the case of non-thermally produced neutralinos.
Turning to the viable range of neutralino pair annihilation cross sections, coannihilation processes do not allow us to set any lower bound, while on purely theoretical grounds a general upper limit on $`\sigma v_0<\text{ }10^{22}\left(\frac{M_\chi }{\mathrm{TeV}}\right)^2\mathrm{cm}^3/\mathrm{s}`$ has been recently set (Profumo (2005)). The only general argument which ties the relic abundance of a WIMP with its pair annihilation cross section is given by the naive relation
$$\mathrm{\Omega }_\chi h^2\frac{3\times 10^{27}\mathrm{cm}^3/\mathrm{s}}{\sigma v_0}$$
(34)
(see Jungman et al. (1995), Eq.3.4), which points at a fiducial value for $`\sigma v_03\times 10^{26}\mathrm{cm}^3/\mathrm{s}`$ for our choice of cosmological parameters. The above mentioned relation can be, however, badly violated in the general MSSM, or even within minimal setups, such as the minimal supergravity scenario (see Profumo (2005)).
Since third generation leptons and quarks Yukawa couplings are always much larger than those of the first two generations, and being the neutralino a Majorana fermion, the largest $`BR(\chi \chi f)`$ for annihilations into a fermion-antifermion pair are in most cases<sup>1</sup><sup>1</sup>1Models with non-universal Higgs masses at the GUT scale can give instances of exceptions to this generic spectral pattern, featuring light first and second generation sfermions (see e.g. Baer et al. 2005b ). into the third generation final states $`b\overline{b}`$, $`t\overline{t}`$ and $`\tau ^+\tau ^{}`$. In the context of supersymmetry, if the supersymmetric partners of the above mentioned fermions are not significantly different in mass, the $`\tau ^+\tau ^{}`$ branching ratio will be suppressed, with respect to the $`b\overline{b}`$ branching ratio by a color factor equal to 1/3, plus a possible further Yukawa coupling suppression, since the two final states share the same $`SU(2)`$ quantum number assignment. Further, the fragmentation functions of third generation quarks are very similar, and give rise to what we will dub in the following as a “soft spectrum”. A second possibility, when kinematically allowed, is the pair annihilation into massive gauge bosons<sup>2</sup><sup>2</sup>2The direct annihilation into photons is loop suppressed in supersymmetric models (see e.g. Bergstrom & Snellman (1988) and Bergstrom & Ullio (1997))., $`W^+W^{}`$ and $`Z_0Z_0`$. Again, the fragmentation functions for these two final states are mostly indistinguishable, and will be indicated as giving a “hard spectrum”. The occurrence of a non-negligible branching fraction into $`\tau ^+\tau ^{}`$ or into light quarks will generally give raise to intermediate spectra between the ”hard” and ”soft” case.
Fig.7 shows the spectral shape of the electron source function in the case of the three sample final states $`b\overline{b}`$, $`\tau ^+\tau ^{}`$ and $`W^+W^{}`$ for $`M_\chi =100`$ GeV, and clarifies the previous discussion. In what follows we will therefore employ sample DM configurations making use of either soft ($`b\overline{b}`$) or hard ($`W^+W^{}`$) spectra, keeping in mind that other possibilities would likely fall in between these two extrema.
In order to make a more stringent contact with supersymmetry phenomenology, we will however also resort to realistic benchmark SUSY models: by this we mean thoroughly defined SUSY setups which are fully consistent with accelerator and other phenomenological constraints, and which give a neutralino thermal relic abundance exactly matching the central cosmologically observed value. To this extent, we refer to the so-called minimal supergravity model (Goldberg 1983; Ellis et al. 1983, (1984)), perhaps one of the better studied paradigms of low-energy supersymmetry, which enables, moreover, a cross-comparison with numerous dedicated studies, ranging from colliders (Baer et al. (2003)) to DM searches (Edsjo et al. 2004, Baer et al. (2004)).
The assumptions of universality in the gaugino and in the scalar (masses and trilinear couplings) sectors remarkably reduce, in this model, the number of free parameters of the general soft SUSY breaking Lagrangian (Chung et al. (2003)) down to four continuous parameters ($`m_0,M_{1/2},A_0,\mathrm{tan}\beta `$) plus one sign ($`\mathrm{sign}(\mu )`$). The mSUGRA parameter space producing a sufficiently low thermal neutralino relic abundance $`\mathrm{\Omega }_\chi h^2`$ has been shown to be constrained to a handful of “regions” featuring effective $`\mathrm{\Omega }_\chi h^2`$ suppression mechanisms (Ellis et al. (2003)). The latter are coannihilations of the neutralino with the next-to-lightest SUSY particle (“Coannihilation” region), rapid annihilations through $`s`$ channel Higgs exchanges (“Funnel” region), the occurrence of light enough neutralino and sfermions masses (“Bulk” region) and the presence of a non-negligible bino-higgsino mixing (“Focus Point” region).
With the idea of allowing a direct comparison with the existing research work in a wealth of complementary fields, we restrict ourselves to the “updated post-WMAP benchmarks for supersymmetry” proposed and studied by Battaglia et al. (2003). All of those setups are tuned so as to feature a neutralino thermal relic density giving exactly the central WMAP-estimated CDM density<sup>3</sup><sup>3</sup>3We adjusted here the values of $`m_0`$ given in Battaglia et al. (2003) in order to fulfill this requirement making use of the latest Isajet v.7.72 release and of the DarkSUSYpackage (Edsjo et al. (2003), see Table 1).. As a preliminary step, we computed the electrons, neutrinos, gamma-rays and protons source spectra for all the 13 $`𝐀^{}`$-$`𝐌^{}`$ models. Remarkably enough, although the SUSY particle spectrum is rather homogeneous throughout the mSUGRA parameter space, the resulting spectra exhibit at least three qualitatively different shapes, according to the dominant final state in neutralino pair annihilation processes. In particular, in the Bulk and Funnel regions the dominant final state is into $`b\overline{b}`$, and, with a sub-dominant variable contribution, $`\tau ^+\tau ^{}`$. The latter channel is instead dominant, for kinematic reasons, in the stau Coannihilation region. Finally, a third, and last, possibility is a dominant gauge bosons final state, which is the case along the Focus Point region. In this respect, in the effort to reproduce all of the mentioned spectral modes, and to reflect every cosmologically viable mSUGRA region, we focused on the four models indicated in Table 1, a subset of the benchmarks of Battaglia et al. (2003) (to which we refer the reader for further details).
We collect in Table 2 the branching ratios for the final states of neutralino pair annihilations. In the last column of this table we also provide the thermally-averaged pair annihilation cross section times the relative velocity, at $`T=0`$, $`\sigma v_0`$. Table 2 is an accurate guideline to interpret the resulting source spectra for the four benchmarks under consideration here, which are shown in Figs. 8 and 9. Fig.8 shows in particular the differential electron (left) and photon (right) yields per neutralino annihilation multiplied by $`\sigma v_0`$, i.e. the source function $`Q(r,E)`$ divided by the number density of neutralino pairs $`𝒩_{\mathrm{pairs}}(r)`$ as a function of the particles’ kinetic energy. As mentioned above, the Bulk and Funnel cases are very similar between each other, though in the latter case one has a heavier spectrum and a larger value of $`\sigma v_0`$. Fig.9 shows the same quantity for neutrinos and protons.
The products of the neutralino annihilation which are more relevant to our discussion are secondary electrons and pions. The secondary particles produced by neutralino annihilation are subject to various physical mechanisms: i) decay (which is especially fast for pions and muons); ii) energy losses which can be suffered by stable particles, like electrons and positrons; iii) spatial diffusion of these relativistic particles in the atmosphere of the cluster. Gamma-rays produced by neutral pion decay, $`\pi ^0\gamma \gamma `$, generate most of the continuum spectrum at energies $`E>\text{ }1`$ GeV and this emission is directly radiated since the $`\pi ^0\gamma \gamma `$ e.m. decay is very fast. This gamma-ray emission is dominant at high energies, $`>\text{ }0.30.5`$ of the neutralino mass, but needs to be complemented by other two emission mechanisms which produce gamma-rays at similar or slightly lower energies: these are the ICS and the bremsstrahlung emission by secondary electrons. We will discuss the full gamma-ray emission of Coma induced by DM annihilation in Sect. 4 below. Secondary electrons are produced through various prompt generation mechanisms and by the decay of charged pions (see, e.g., Colafrancesco & Mele (2001)). In fact, charged pions decay through $`\pi ^\pm \mu ^\pm \nu _\mu (\overline{\nu }_\mu )`$, with $`\mu ^\pm e^\pm +\overline{\nu }_\mu (\nu _\mu )+\nu _e(\overline{\nu }_e)`$ and produce $`e^\pm `$, muons and neutrinos. Electrons and positrons are produced abundantly by neutralino annihilation (see Fig. 8, left) and are subject to spatial diffusion and energy losses. Both spatial diffusion and energy losses contribute to determine the evolution of the source spectrum into the equilibrium spectrum of these particles, i.e. the quantity which will be used to determine the overall multi-wavelength emission induced by DM annihilation. The secondary electrons eventually produce radiation by synchrotron in the magnetized atmosphere of Coma, Inverse Compton Scattering of CMB (and other background) photons and bremsstrahlung with protons and ions in the atmosphere of the Coma cluster (see, e.g., Colafrancesco & Mele (2001) and Colafrancesco (2003, 2006) for a review). These secondary particles also produce heating of the intra-cluster gas by Coulomb collisions with the intra-cluster gas particles and SZ effect (see, e.g. Colafrancesco (2003), Colafrancesco (2006)). Other fundamental particles which might have astrophysical relevance are also produced in DM annihilation. Protons are produced in a smaller quantity with respect to $`e^\pm `$ (see Fig. 9, right), but do not loose energy appreciably during their lifetime while they can diffuse and be stored in the cluster atmosphere. These particles can, in principle, produce heating of the intra-cluster gas and $`pp`$ collisions providing, again, a source of secondary particles (pions, neutrinos, $`e^\pm `$, muons, …) in complete analogy with the secondary particle production by neutralino annihilation. Neutrinos are also produced in the process of neutralino annihilation (see Fig. 9, left) and propagate with almost no interaction with the matter of the cluster. However, the resulting flux from Coma is found to be unobservable by current experiments.
To summarize, the secondary products of neutralino annihilation which have the most relevant astrophysical impact onto the multi-frequency spectral energy distribution of DM halos are neutral pions and secondary electrons.
## 4 Neutralino-induced signals
A complete description of the emission features induced by DM must take, consistently, into account the diffusion and energy-loss properties of these secondary particles. These mechanisms are taken into account in the following diffusion equation (i.e. neglecting convection and re-acceleration effects):
$`{\displaystyle \frac{}{t}}{\displaystyle \frac{dn_e}{dE}}`$ $`=`$ $`\left[D(E,x){\displaystyle \frac{dn_e}{dE}}\right]+{\displaystyle \frac{}{E}}\left[b(E,x){\displaystyle \frac{dn_e}{dE}}\right]`$ (35)
$`+Q_e(E,x),`$
where $`dn_e/dE`$ is the equilibrium spectrum, $`D(E,x)`$ is the diffusion coefficient, $`b(E,x)`$ is the energy loss term and $`Q_e(E,x)`$ is the source function. The analytical solution of this equation for the case of the DM source function is derived in the Appendix A.
In the limit in which electrons and positrons lose energy on a timescale much shorter than the timescale for spatial diffusion, i.e. the regime which applies to the case of galaxy clusters, the first term on the r.h.s. of Eq. (64) can be neglected, and the expression for equilibrium number density becomes:
$$\left(\frac{dn_e}{dE}\right)_{nsd}(r,E)=\frac{1}{b\left(E\right)}_E^{M_\chi }𝑑E^{}Q_e(r,E^{}),$$
(36)
(see the Appendix A for a general discussion of the role of spatial diffusion and of the regimes in which it is relevant).
The derivation of the full solution of the diffusion equation (Eq. 64) and the effects of diffusion and energy losses described in the Appendix A, set us in the position to discuss the multi-frequency emission produced by the DM (neutralino) component of the Coma cluster. We will present the overall DM-induced spectral energy distribution (hereafter SED) from low to high observing frequencies.
We describe here our reference setup for the numerical calculations. Our reference halo setup is the N04 profile and other parameters/choice of extrapolation schemes as in Fig. 6. We consider the predictions of two particle models, one with a branching ratio equal to 1 in $`b\overline{b}`$, i.e. a channel with a soft production spectrum, and the second one with a branching ratio equal to 1 into $`W^+W^{}`$, i.e. a channel with hard spectrum. Since we have previously shown that diffusion is not relevant in a Coma-like cluster of galaxy, we neglect, in our numerical calculations, the spatial diffusion for electrons and positrons: this is the limit in which the radial dependence and frequency dependence can be factorized in the expression for the emissivity.
### 4.1 Radio emission
At radio frequencies, the DM-induced emission is dominated by the synchrotron radiation of the relativistic secondary electrons and positrons of energy $`E=\gamma m_ec^2`$, living in a magnetic field $`B\left(r\right)`$ and a background plasma with thermal electron density $`n\left(r\right)`$, and in the limit of frequency $`\nu `$ of the emitted photons much larger than the non-relativistic gyro-frequency $`\nu _0=eB/\left(2\pi mc\right)2.8B_\mu `$ Hz and the plasma frequency $`\nu _p=8980\left(n\left(r\right)/1cm^3\right)^{1/2}`$ Hz. Averaging over the directions of emission, the spontaneously emitted synchrotron power at the frequency $`\nu `$ is given by (Longair (1994)):
$$P_{\mathrm{synch}}(\nu ,E,r)=_0^\pi 𝑑\theta \frac{\mathrm{sin}\theta }{2}2\pi \sqrt{3}r_0mc\nu _0\mathrm{sin}\theta F\left(x/\mathrm{sin}\theta \right),$$
(37)
where we have introduced the classical electron radius $`r_0=e^2/\left(mc^2\right)=2.8210^{13}`$ cm, and we have defined the quantities $`x`$ and $`F`$ as:
$$x\frac{2\nu }{3\nu _0\gamma ^2}\left[1+\left(\frac{\gamma \nu _p}{\nu }\right)^2\right]^{3/2},$$
(38)
and
$$F\left(t\right)t_t^{\mathrm{}}𝑑zK_{5/3}\left(z\right)1.25t^{1/3}\mathrm{exp}\left(t\right)\left[648+t^2\right]^{1/12}.$$
(39)
Folding the synchrotron power with the spectral distribution of the equilibrium number density of electrons and positrons, we get the local emissivity at the frequency $`\nu `$:
$$j_{\mathrm{synch}}(\nu ,r)=_{m_e}^{M_\chi }𝑑E\left(\frac{dn_e^{}}{dE}+\frac{dn_{e^+}}{dE}\right)P_{\mathrm{synch}}(\nu ,E,r).$$
(40)
This is the basic quantity we need in order to compare our predictions with the available data. In particular, we will compare our predictions with measurements of the integrated (over the whole Coma radio halo size) flux density spectrum:
$$S_{\mathrm{synch}}\left(\nu \right)=d^3r\frac{j_{\mathrm{synch}}(\nu ,r)}{4\pi D_{\mathrm{Coma}}^2},$$
(41)
where $`D_{\mathrm{Coma}}`$ is the luminosity distance of Coma, and with the azimuthally averaged surface brightness distribution at a given frequency and within a beam of angular size $`\mathrm{\Delta }\mathrm{\Omega }`$ (PSF):
$$I_{\mathrm{synch}}(\nu ,\mathrm{\Theta },\mathrm{\Delta }\mathrm{\Omega })=_{\mathrm{\Delta }\mathrm{\Omega }}𝑑\mathrm{\Omega }_{l.o.s.}𝑑l\frac{j_{\mathrm{synch}}(\nu ,l)}{4\pi },$$
(42)
where the integral is performed along the line of sight (l.o.s.) $`l`$, within a cone of size $`\mathrm{\Delta }\mathrm{\Omega }`$ centered in a direction forming an angle $`\mathrm{\Theta }`$ with the direction of the Coma center.
We started from the full dataset on the radio flux density spectrum (Thierbach et al. (2003)) and minimized the fit with respect to the WIMP mass (with the bound $`M_\chi 10`$ GeV for the $`b\overline{b}`$ case, and mass above threshold for the $`W^+W^{}`$ case), the strength of the magnetic field (with the bound $`B_\mu 1\mu G`$) and the annihilation rate $`\sigma v_0`$. The spectrum predicted by two models with the lowest values of $`\chi _r^2`$ are shown in Fig. 10. In both cases the best fit corresponds to the lowest neutralino mass allowed, since this is the configuration in which the fall-off of the flux density at the highest observed frequency tends to be better reproduced. For the same reason, the fit in the case of a soft spectrum is favored with respect to the one with a hard spectrum (we have checked that in case of $`\tau ^+\tau ^{}`$ again does not give a bend-over in the spectrum where needed). The values of the annihilation rates required by the fit are fairly large: $`\sigma v_0=4.710^{25}`$ cm<sup>3</sup> s<sup>-1</sup> for $`b\overline{b}`$ case, and about one order of magnitude smaller, $`\sigma v_0=8.810^{26}`$ cm<sup>3</sup> s<sup>-1</sup>, for the $`W^+W^{}`$ case, despite the heavier neutralino mass, since the best fit values correspond to different values of the magnetic field of about $`1.2\mu G`$ and $`8\mu G`$, respectively.
In Fig. 11 we compare the radio-halo brightness data of Deiss et al. (1997) with the surface brightness distribution $`I_{synch}\left(r\right)`$ predicted at $`\nu =1.4`$ GHz, within a beam equal to the detector angular resolution (HPBW of $`9.^{}35`$), for the best fit model with $`M_\chi =40`$ GeV. In the left panel we plot the predicted surface brightness considering the case of a uniform magnetic field equal to $`1.2\mu G`$, showing explicitly in this case that the assumption we made of neglecting spatial diffusion for electrons and positrons is indeed justified, since the results obtained including or neglecting spatial diffusion essentially coincide. The radial brightness we derive in this case does not match the shape of the radio halo indicated by the data. However, it is easy to derive a phenomenological setup with a magnetic field $`B\left(r\right)`$ varying with radius in which a much better fit can be obtained, while leaving unchanged the total radio flux density $`S_{synch.}\left(\nu \right)`$. We show in the right panel of Fig.11 the predictions for $`I_{synch}\left(r\right)`$ considering a radial dependence of the magnetic field of the form:
$$B\left(r\right)=B_0\left(1+\frac{r}{r_{c1}}\right)^2\left[1+\left(\frac{r}{r_{c2}}\right)^2\right]^\beta ,$$
which is observationally driven by the available information on the Faraday rotation measures (RM) for Coma (see Fig.16). Such $`B\left(r\right)`$ profile starts at a slightly smaller value in the center of the Coma, rises at a first intermediate scale $`r_{c1}`$ and then drops rather rapidly at the scale $`r_{c2}`$. The basic information we provide here is that a radial dependence of the magnetic field like the previous one is required in DM annihilation models to reproduce the radio-halo surface brightness distribution. The specific case displayed is for best-fit values $`B_0=0.55`$ $`\mu `$G , $`\beta =2.7`$, $`r_{c1}=3^{}`$ $`r_{c2}=17.^{}5`$, and it provides an excellent fit to the surface brightness radial profile (see Fig.11). In that figure we also plot separately the contributions to the surface brightness due to the smooth DM component (essentially a point-like source in case of this rather poor angular resolution) and the term due to subhalos (which extends instead to larger radii). It is interesting to note that the surface brightness profile can only be fitted by considering the extended sub-halo distribution which renders the DM profile of Coma more extended than the smooth, centrally peaked component. This means that any peaked and smooth DM profile is unable to fit this observable for Coma.
A decrease of $`B\left(r\right)`$ at large radii is expected by general considerations of the structure of radio-halos in clusters and, more specifically, for Coma (see Colafrancesco et al. (2005)) and it is also predicted by numerical simulations (see, e.g., Dolag et al. (2002)): thus it seems quite natural and motivated. At small radii, the mild central dip of $`B\left(r\right)`$ predicted by the previous formula is what is phenomenologically required by the specific DM model we worked out in our paper. Finally, we notice that our specific phenomenological model for the spatial distribution of $`B\left(r\right)`$ is able to reproduce the spatial distribution of Faraday rotation measures (RMs) observed in Coma (see Kim et al. (1990)), as shown in Fig.12.
It is evident that models in which $`B`$ is either constant or decreases monotonically towards large radii seem to be difficult to reconcile with the available RM data. The RM data at $`\theta <\text{ }20`$ arcmin seem to favour, indeed, a model for $`B\left(r\right)`$ with a slight rise at intermediate angular scales followed by a decrease at large scales, like the one we adopt here to fit the radio-halo surface brightness of Coma. In this respect, it seems that our choice for $`B\left(r\right)`$ is, at least, an observationally driven result.
The synchrotron signal produced by the annihilation of DM depends, given the fundamental physics and astrophysics framework, on two relevant quantities: the annihilation rate and the magnetic field. Thus, it is interesting to find the best-fitting region of the $`\sigma v_0B`$ plane which is consistent with the available dataset for Coma.
Since the data on the radio flux density spectrum (Thierbach et al. (2003)) is a compilation of measurements performed with different instruments (possibly with different systematics), it is difficult to decide a cut on the $`\chi _r^2`$ value which defines an acceptable fit. In Fig. 13 we plot sample isolevel curves for $`\chi _r^2`$, spotting the shape of the minima of $`\chi _r^2`$ , in the plane $`B\sigma v_0`$, for the two sample annihilation channels and a few sample values of the WIMP mass (note that values labeling isolevel curves are sensibly different in the two panels). In Fig. 14 we show the analogous $`\chi _r^2`$ isolevels in the WIMP mass – annihilation rate plane, and taking at each point the minimum $`\chi _r^2`$ while varying the magnetic field strength between $`1\mu G`$ and $`20\mu G`$: the curves converge to a maximal value enforced by the lower limit of $`1\mu G`$, and the upper value does not enter in defining isolevel curve shapes.
In order to assess whether the outlined radio-data preferred regions are or are not compatible with supersymmetric DM models, we proceed to a random scan of the SUSY parameter space, in the bottom-up approach which we outline below. We relax all universality assumptions, and fully scan the low-energy scale MSSM, imposing phenomenological as well as cosmological constraints on the randomly generated models<sup>4</sup><sup>4</sup>4We scan all the SUSY parameters linearly over the indicated range.. We take values of $`\mathrm{tan}\beta `$, the ratio between the vacuum expectation values of the two Higgs doublets, between 1 and 60. The parameters entering the neutralino mass matrix are generated in the range $`1\mathrm{GeV}<m_1,m_2,\mu <1\mathrm{TeV}`$, and we define $`m_{\mathrm{LSP}}\mathrm{min}(m_1,m_2,\mu )`$. To avoid flavor changing effects in the first two lightest quark generations, we assume that the soft-breaking masses in the first two generations squark sector are degenerate, i.e. we assume $`m_{\stackrel{~}{Q},\stackrel{~}{U},\stackrel{~}{D}}^{\left(1\right)}=m_{\stackrel{~}{Q},\stackrel{~}{U},\stackrel{~}{D}}^{\left(2\right)}`$. The scalar masses are scanned over the range
$$m_{\mathrm{LSP}}<m_{\stackrel{~}{Q}}^{(1,3)},m_{\stackrel{~}{U}}^{(1,3)},m_{\stackrel{~}{U}}^{(1,3)},m_{\stackrel{~}{L}}^{(1,2,3)},m_{\stackrel{~}{E}}^{(1,2,3)},m_A<2.5\mathrm{TeV}$$
(43)
The trilinear couplings are sampled in the range
$`3\mathrm{max}(m_{\stackrel{~}{Q}}^{\left(i\right)},m_{\stackrel{~}{U}}^{\left(i\right)})<`$ $`A_U^{\left(i\right)}`$ $`<3\mathrm{max}(m_{\stackrel{~}{Q}}^{\left(i\right)},m_{\stackrel{~}{U}}^{\left(i\right)})`$ (44)
$`3\mathrm{max}(m_{\stackrel{~}{Q}}^{\left(i\right)},m_{\stackrel{~}{D}}^{\left(i\right)})<`$ $`A_D^{\left(i\right)}`$ $`<3\mathrm{max}(m_{\stackrel{~}{Q}}^{\left(i\right)},m_{\stackrel{~}{D}}^{\left(i\right)})`$ (45)
$`3\mathrm{max}(m_{\stackrel{~}{L}}^{\left(i\right)},m_{\stackrel{~}{E}}^{\left(i\right)})<`$ $`A_E^{\left(i\right)}`$ $`<3\mathrm{max}(m_{\stackrel{~}{L}}^{\left(i\right)},m_{\stackrel{~}{E}}^{\left(i\right)})`$ (46)
Finally, we take the gluino mass in the range $`200\mathrm{GeV}<m_{\stackrel{~}{g}}<3\mathrm{TeV}`$. The mass ranges for squarks and gluino have been chosen following qualitative criteria (Baer et al. (2003); Battaglia et al. (2003)), so that all viable models generated should be “visible” at the LHC.
We exclude models giving a relic abundance of neutralinos exceeding $`\mathrm{\Omega }_\chi h^2>0.13`$. Further, we impose the various colliders mass limits on charginos, gluinos, squarks and sleptons, as well as on the Higgs masses<sup>5</sup><sup>5</sup>5Since we do not impose any gaugino unification relation, we do not impose any constraint from collider searches on the neutralino sector.. Moreover, we also require the BR($`bs\gamma `$) and all electroweak precision observables to be consistent with the theoretical and experimental state-of-the-art (Eidelman et al. (2004)).
We classify the models according to the branching ratios of the neutralino pair-annihilations final states, according to the following criteria: we consider a model having a hard spectrum if
$$\mathrm{BR}\left(W^+W^{}\right)+\mathrm{BR}\left(ZZ\right)>0.8;$$
(47)
a soft spectrum is instead attributed to models satisfying the condition
$$\underset{i=1}{\overset{6}{}}\left[\mathrm{BR}\left(q_i\overline{q}_i\right)+\mathrm{BR}\left(q_i\overline{q}_ig\right)\right]+\mathrm{BR}\left(gg\right)>0.8.$$
(48)
We show, in Fig.15 a scatter plots of the viable SUSY configurations, indicating with filled green circles those thermally producing a neutralino relic abundance within the 2-$`\sigma `$ WMAP range, and with red circles those producing a relic abundance below the WMAP range (whose relic abundance can however be cosmologically enhanced, in the context of quintessential or Brans-Dicke cosmologies, or which can be non-thermally produced, as to make up all of the observed CDM (see Murakami & Wells 2001 and other refs. ). The low $`\chi ^2`$ ranges of $`\sigma v_0B`$ and $`\sigma v_0M_\chi `$ values indicated in Figs. 13 and 14 are therefore shown to be actually populated by a number of viable SUSY models.
### 4.2 From the UV to the gamma-ray band
Inverse Compton (IC) scatterings of relativistic electrons and positrons on target cosmic microwave background (CMB) photons give rise to a spectrum of photons stretching from below the extreme ultra-violet up to the soft gamma-ray band, peaking in the soft X-ray energy band. Let $`E=\gamma m_ec^2`$ be the energy of electrons and positrons, $`ϵ`$ that of the target photons and $`E_\gamma `$ the energy of the scattered photon. The Inverse Compton power is obtained by folding the differential number density of target photons with the IC scattering cross section:
$$P_{\mathrm{IC}}(E_\gamma ,E)=cE_\gamma 𝑑ϵn\left(ϵ\right)\sigma (E_\gamma ,ϵ,E)$$
(49)
where $`n\left(ϵ\right)`$ is the black body spectrum of the $`2.73K`$ CMB photons, while $`\sigma (E_\gamma ,ϵ,E)`$ is given by the Klein-Nishina formula:
$$\sigma (E_\gamma ,ϵ,E)=\frac{3\sigma _T}{4ϵ\gamma ^2}G(q,\mathrm{\Gamma }_e)$$
(50)
where $`\sigma _T`$ is the Thomson cross section and
$$G(q,\mathrm{\Gamma }_e)\left[2q\mathrm{ln}q+\left(1+2q\right)\left(1q\right)+\frac{\left(\mathrm{\Gamma }_eq\right)^2\left(1q\right)}{2\left(1+\mathrm{\Gamma }_eq\right)}\right]$$
(51)
with
$$\mathrm{\Gamma }_e=4ϵ\gamma /\left(m_ec^2\right)q=E_\gamma /\left[\mathrm{\Gamma }_e\left(\gamma m_ec^2E_\gamma \right)\right].$$
(52)
Folding the IC power with the spectral distribution of the equilibrium number density of electrons and positrons, we get the local emissivity of IC photons of energy $`E_\gamma `$:
$$j_{\mathrm{IC}}(E_\gamma ,r)=𝑑E\left(\frac{dn_e^{}}{dE}+\frac{dn_{e^+}}{dE}\right)P_{\mathrm{IC}}(E_\gamma ,E)$$
(53)
which we use to estimate the integrated flux density spectrum:
$$S_{\mathrm{IC}}\left(E_\gamma \right)=d^3r\frac{j_{\mathrm{IC}}(E_\gamma ,r)}{4\pi D_{\mathrm{Coma}}^2}.$$
(54)
In Eq. (49) and Eq. (53) the limits of integration over $`ϵ`$ and $`E_\gamma `$ are set from the kinematics of the IC scattering which restricts $`q`$ in the range $`1/\left(4\gamma ^2\right)q1`$.
The last relevant contribution to the photon emission of Coma due to relativistic electrons and positrons is the process of non-thermal bremsstrahlung, i.e. the emission of gamma-ray photons in the deflection of the charged particles by the electrostatic potential of intra-cluster gas. Labeling with $`E=\gamma m_ec^2`$ the energy of electrons and positrons, and with $`E_\gamma `$ the energy of the emitted photons, the local non-thermal bremsstrahlung power is given by:
$$P_\mathrm{B}(E_\gamma ,E,r)=cE_\gamma \underset{j}{}n_j\left(r\right)\sigma _j(E_\gamma ,E),$$
(55)
with the sum including all species $`j`$ in the intra-cluster medium, each with number density $`n_j\left(r\right)`$ and relative production cross section:
$$\sigma _j(E_\gamma ,E)=\frac{3\alpha \sigma _T}{8\pi E_\gamma }\left[\left(1+\left(1E_\gamma /E\right)^2\right)\varphi _1\frac{2}{3}\left(1E_\gamma /E\right)\varphi _2\right]$$
(56)
where $`\alpha `$ is the fine structure constant, $`\varphi _1`$ and $`\varphi _2`$ two energy dependent scattering functions which depend on the species $`j`$ (see Longair (1994) for details). The emissivity $`j_\mathrm{B}(E_\gamma ,r)`$ is obtained by folding the power over the equilibrium electron/positron number density, i.e. the analogous of Eq. (53), while the integrated flux density $`S_\mathrm{B}\left(E_\gamma \right)`$ is obtained by summing over all relevant sources as in Eq. (54). We apply this scheme to Coma implementing the gas density profile in Eq. (18) by including atomic and molecular hydrogen and correcting for the helium component.
As we have already mentioned, a hard gamma-ray component arises also from prompt emission in WIMP pair annihilations, either in loop suppressed two-body final states giving monochromatic photons, or through the production and prompt decay of neutral pions giving gamma-rays with continuous spectrum. Since photons propagate on straight lines (or actually geodesics), the gamma-ray flux due to prompt emission is just obtained by summing over sources along the line of sight; we will consider terms integrated over volume
$$F_\gamma \left(E_\gamma \right)=d^3r\frac{Q_\gamma (E_\gamma ,r)}{4\pi D_{\mathrm{Coma}}^2}.$$
(57)
### 4.3 The multi-frequency SED of Coma
We show in Fig. 16 the multi-frequency SED produced by WIMP annihilation in the two models used to fit the radio halo spectrum of Coma, as shown in Fig. 10.
The model with $`M_\chi =40`$ GeV provides the better fit to the radio halo data because the relative equilibrium electron spectrum is steeper and shows also the high-$`\nu `$ bending which fits the most recent data (Thierbach et al. (2003)). The IC and bremsstrahlung branches of the SED are closely related to the synchrotron branch (since they depend on the same particle population) and their intensity ratio depends basically on the value of the adopted magnetic field. The relatively high value $`B=1.2`$ $`\mu `$G indicated by the best fit to the radio data implies a rather low intensity of the IC and bremsstrahlung emission, well below the EUV and hard X-ray data for Coma. Nonetheless, the gamma-ray emission due to $`\pi ^0\gamma \gamma `$ decay predicted by this model could be detectable with the GLAST-LAT detector, even though it is well below the EGRET upper limit.
The detectability of the multi-frequency SED worsens in the model with $`M_\chi =81`$ GeV, where the flatness of the equilibrium electron spectrum cannot provide an acceptable fit to the radio data. Moreover, the adopted value of the magnetic field $`B=8`$ $`\mu `$G implies a very low intensity of the IC, bremsstrahlung and $`\pi ^0\gamma \gamma `$ emission, which should be not detectable by the next generation HXR and gamma-ray experiments.
The energetic electrons and positrons produced by WIMP annihilation have other interesting astrophysical effects among which we will discuss specifically in the following the Sunyaev-Zel’dovich (hereafter SZ) effect produced by DM annihilation and the heating of the intracluster gas produced by Coulomb collisions.
### 4.4 SZ effect
The energetic electrons and positrons produced by WIMP annihilation interact with the CMB photons and up-scatter them to higher frequencies producing a peculiar SZ effect (as originally realized by Colafrancesco (2004)) with specific spectral and spatial features.
The generalized expression for the SZ effect which is valid in the Thomson limit for a generic electron population in the relativistic limit and includes also the effects of multiple scatterings and the combination with other electron population in the cluster atmospheres has been derived by Colafrancesco et al. (2003). This approach is the one that should be properly used to calculate the specific SZ<sub>DM</sub> effect induced by the secondary electrons produced by WIMP annihilation. Here we do not repeat the description of the analytical technique and we refer to the general analysis described in Colafrancesco et al. (2003). According to these results, the DM induced spectral distortion is
$$\mathrm{\Delta }I_{\mathrm{DM}}\left(x\right)=2\frac{\left(k_\mathrm{B}T_0\right)^3}{\left(hc\right)^2}y_{\mathrm{DM}}\stackrel{~}{g}\left(x\right),$$
(58)
where $`T_0`$ is the CMB temperature and the Comptonization parameter $`y_{\mathrm{DM}}`$ is given by
$$y_{\mathrm{DM}}=\frac{\sigma _T}{m_\mathrm{e}c^2}P_{\mathrm{DM}}𝑑\mathrm{},$$
(59)
in terms of the pressure $`P_{\mathrm{DM}}`$ contributed by the secondary electrons produced by neutralino annihilation. The quantity $`y_{\mathrm{DM}}\sigma v_0n_\chi ^2`$ and scales as $`\sigma v_oM_\chi ^2`$, providing an increasing pressure $`P_{\mathrm{DM}}`$ and optical depth $`\tau _{\mathrm{DM}}=\sigma _T𝑑\mathrm{}n_\mathrm{e}`$ for decreasing values of the neutralino mass $`M_\chi `$. The function $`\stackrel{~}{g}\left(x\right)`$, with $`xh\nu /k_\mathrm{B}T_0`$, can be written as
$$\stackrel{~}{g}\left(x\right)=\frac{m_\mathrm{e}c^2}{k_\mathrm{B}T_\mathrm{e}}\left\{\frac{1}{\tau }\left[_{\mathrm{}}^+\mathrm{}i_0\left(xe^s\right)P\left(s\right)𝑑si_0\left(x\right)\right]\right\}$$
(60)
in terms of the photon redistribution function $`P\left(s\right)`$ and of $`i_0\left(x\right)=2\left(k_\mathrm{B}T_0\right)^3/\left(hc\right)^2x^3/\left(e^x1\right)`$, where we defined the quantity
$$k_\mathrm{B}T_\mathrm{e}\frac{\sigma _\mathrm{T}}{\tau }P𝑑\mathrm{}=\frac{P𝑑\mathrm{}}{n_\mathrm{e}𝑑\mathrm{}}=_0^{\mathrm{}}𝑑pf_\mathrm{e}\left(p\right)\frac{1}{3}pv\left(p\right)m_\mathrm{e}c$$
(61)
(see Colafrancesco et al. (2003); Colafrancesco (2004)), which is the analogous of the average temperature for a thermal population (for a thermal electron distribution $`k_\mathrm{B}T_\mathrm{e}=k_\mathrm{B}T_\mathrm{e}`$ obtains, in fact). The photon redistribution function $`P\left(s\right)=𝑑pf_\mathrm{e}\left(p\right)P_\mathrm{s}(s;p)`$ with $`s=\mathrm{ln}\left(\nu ^{}/\nu \right)`$, in terms of the CMB photon frequency increase factor $`\nu ^{}/\nu =\frac{4}{3}\gamma ^2\frac{1}{3}`$, depends on the electron momentum ($`p`$) distribution, $`f_\mathrm{e}\left(p\right)`$, produced by WIMP annihilation.
We show in Fig.17 the frequency dependence of the CMB temperature change,
$$\frac{\mathrm{\Delta }T}{T_0}=\frac{\left(e^x1\right)^2}{x^4e^x}\frac{\mathrm{\Delta }I}{I_0},$$
(62)
as produced by the DM-induced SZ effect in the two best fit WIMP models here considered, compared to the temperature change due to the thermal SZ effect produced by the intracluster gas. The most recent analysis of the thermal SZ effect in Coma (DePetris et al. (2003)) provides an estimate of the optical depth of the thermal intracluster gas $`\tau _{th}=4.910^3`$ which best fits the data. The model with $`M_\chi =40`$ GeV provides a detectable SZ<sub>DM</sub> effect which has a quite different spectral shape with respect to the thermal SZ effect: it yields a temperature decrement at all the microwave frequencies, $`<\text{ }600`$ GHz, where the thermal SZ effect is observed and produces a temperature increase only at very high frequencies $`>600`$ GHz. This behavior is produced by the large frequency shift of CMB photons induced by the relativistic secondary electrons generated by the WIMP annihilation. As a consequence, the zero of the SZ<sub>DM</sub> effect is effectively removed from the microwave range and shifted to a quite high frequency $`600`$ GHz with respect to the zero of the thermal SZ effect, a result which allows one, in principle, to estimate directly the pressure of the electron populations and hence to derive constraints on the WIMP model (see Colafrancesco (2004)).
The presence of a substantial SZ<sub>DM</sub> effect is likely to dominate the overall SZ signal at frequencies $`x>\text{ }3.84.5`$ providing a negative total SZ effect. It is, however, necessary to stress that in such frequency range there are other possible contributions to the SZ effect, like the kinematic effect and the non-thermal effect which could provide additional biases (see, e.g., Colafrancesco et al. (2003)). Nonetheless, the peculiar spectral shape of the $`SZ_{\mathrm{DM}}`$ effect is quite different from that of the kinematic SZ effect and of the thermal SZ effect and this result allows us to disentangle it from the overall SZ signal. An appropriate multi-frequency analysis of the overall SZ effect based on observations performed on a wide spectral range (from the radio to the sub-mm region) is required to separate the various SZ contributions and to provide an estimate of the DM induced SZ effect. In fact, simultaneous SZ observations at low frequencies $`30`$ GHz (where there is the largest temperature decrement due to SZ<sub>DM</sub>), at $`150`$ GHz (where the SZ<sub>DM</sub> deepens the minimum in $`\mathrm{\Delta }I/I`$ with respect to the dominant thermal SZ effect), at $`220`$ GHz (where the SZ<sub>DM</sub> dominates the overall SZ effect and produces a negative signal instead of the expected $``$ null signal) and at $`>\text{ }250`$ GHz (where the still negative SZ<sub>DM</sub> decreases the overall SZ effect with respect to the dominant thermal SZ effect) coupled with X-ray observations which determine the gas distribution within the cluster (and hence the associated dominant thermal SZ effect) can separate the SZ<sub>DM</sub> from the overall SZ signal, and consequently, set constraints on the WIMP model.
The WIMP model with $`M_\chi =40`$ GeV produces a temperature decrement which is of the order of $``$ 40 to 15 $`\mu `$K for SZ observations in the frequency range $``$ 30 to 150 GHz (see Fig.17). These signals are still within the actual uncertainties of the available SZ data for Coma and are below the current SZ sensitivity of WMAP (see, e.g., Bennet et al. (2003) and the results of the analysis of the WMAP SZ signals from a sample of nearby clusters performed by Lieu et al. (2005)). Nonetheless, such SZ signals could be detectable with higher sensitivity experiments. The high sensitivity planned for the future SZ experiments can provide much stringent limits to the additional SZ effect induced by DM annihilation. In this context, the next coming sensitive bolometer arrays (e.g., APEX), interferometric arrays (e.g., ALMA) and the PLANCK-HFI experiment, or the planned OLIMPO balloon-borne experiment, have enough sensitivity to probe the contributions of various SZ effects in the frequency range $`\nu 30250`$ GHz, provided that accurate cross-calibration at different frequencies can be obtained. The illustrative comparison (see Fig.17) between the model predictions and the sensitivity of the PLANCK LFI and HFI detectors at the optimal observing frequencies ($`\nu =31.5`$ and $`53`$ GHz for the LFI detector and $`\nu =143`$ and $`217`$ GHz for the HFI detector) show that the study of the SZ effect produced by DM annihilation is actually feasible with the next generation SZ experiments. We show in Fig.18 the expected ratio between the DM-induced SZ effect and the thermal SZ effect for the two WIMP models here considered. It is evident that while the model with $`M_\chi =40`$ GeV provides a detectable signal which is a sensitive fraction of the thermal SZ effect at $`\nu <250`$ GHz, the SZ signal provided by the model with $`M_\chi =81`$ GeV is by far too small to be detectable at any frequency.
The spectral properties shown by the SZ<sub>DM</sub> for neutralinos depends on the specific neutralino model as we have shown in Fig.17: in fact, the SZ effect is visible for a neutralino with $`M_\chi =40`$ GeV and not visible for a neutralino with $`M_\chi =81`$ GeV. Thus the detailed features of the SZ effect from DM annihilation depends strongly on the mass and composition of the DM particle, and - in turn - on the equilibrium spectrum of the secondary electrons. Each specific DM model predicts its own spectrum of secondary electrons and this influences the relative SZ effect. Models of DM which provide similar electron spectra will provide similar SZ effects.
### 4.5 Heating of the intracluster gas
Low energy secondary electrons produced by WIMP annihilation might heat the intracluster gas by Coulomb collisions since the Coulomb loss term dominates the energy losses at $`E<\text{ }200`$ MeV (see Fig. 26). The specific heating rate is given by
$$\frac{dE}{dtdV}=𝑑E\frac{dn_e}{dE}\left(\frac{dE}{dt}\right)_{Coul}$$
(63)
where $`\frac{dn_e}{dE}`$ is the equilibrium electron spectrum derived in Sect. A and the Coulomb loss rate is $`\left(dE/dt\right)_{Coul}=b_{Coul}^0n\left(1+\mathrm{log}\left(\gamma /n\right)/75\right)`$ where $`n`$ is the mean number density of thermal electrons in $`\mathrm{cm}^3`$ (see Eq. 18, the average over space gives about $`n\mathrm{1.3\hspace{0.33em}10}^3`$), $`\gamma E/m_e`$ and $`b_{Coul}^06.1310^{16}\mathrm{GeV}\mathrm{s}^1`$.
Fig.19 shows the specific heating rate of Coma as produced in the two WIMP models explored here. The non-singular N04 halo model adopted in our analysis does not provide a high specific heating rate at the cluster center, and thus one might expect an overall heating rate for Coma which is of order of $`10^{38}`$ erg/s ($`10^{36}`$ erg/s) for the WIMP model with $`M_\chi =40`$ GeV ($`M_\chi =81`$ GeV). We also notice that the region that mostly contributes to the overall heating of Coma is not located at the center of the cluster. This is again a consequence of the non-singular N04 DM profile which has been adopted. The diffusion of electrons in the innermost regions of Coma acts in the same direction and moves the maximum of the curves shown in the right panel of Fig.19 towards the outskirts of Coma, even in the case of a halo density profile which is steeper than the adopted one.
This implies, in conclusion, that WIMP annihilation cannot provide most of the heating of Coma, even in its innermost regions. Such a conclusion seems quite general and implies that non-singular DM halo models are not able to provide large quantities of heating at the center of galaxy clusters so to quench efficiently the cooling of the intracluster gas (with powers of $`10^{4345}`$ erg/s). Only very steep halo profiles (even steeper than the Moore profile) and with the possible adiabatic growth of a central matter concentration (e.g., a central BH) could provide sufficient power to quench locally (i.e. in the innermost regions) the intracluster gas cooling (see, e.g., Totani (2004)). However, we stress that the spatial diffusion of the secondary electrons in the innermost regions of galaxy clusters should flatten the specific heating rate in the vicinity of the DM spike and thus decrease substantially the heating efficiency by Coulomb collisions. In conclusion, we believe that the possibility to solve the cooling flow problem of galaxy clusters by WIMP annihilation is still an open problem.
## 5 Discussion
WIMP annihilation in galaxy cluster is an efficient mechanism to produce relativistic electrons and high-energy particles which are able, in turn, to produce a wide SED extended over more than 18 orders of magnitude in frequency, from radio to gamma-rays. We discuss here the predictions of two specific models which embrace a vast range of possibilities.
The $`b\overline{b}`$ model with $`M_\chi =40`$ GeV and annihilation cross section $`\sigma v_0=4.710^{25}cm^3s^1`$ provides a reasonable fit to the radio data (both the total spectrum and the surface brightness radial distribution) with a magnetic field whose mean value is $`B1.2\mu `$G. We remind here that the quite high value of $`\sigma v_0`$ is well inside the range of neutralino masses and annihilation cross-sections provided by the most general supersymmetric DM setup (see our discussion in Sect.3.2). Table 3 provides an illustrative scheme of the radiation mechanisms, of the particle energies and of the fluxes predicted by this best-fit WIMP model for a wide range of the physical conditions in the cluster atmosphere.
For the best-fit values of $`M_\chi =40`$ GeV and $`\sigma v_0=4.710^{25}cm^3s^1`$ this model yields EUV and HXR fluxes which are more than one order of magnitude fainter than the Coma data. The gamma-ray flux produced by this model is dominated by the continuum $`\pi ^0\gamma \gamma `$ component and it is a factor $`5`$ lower than the EGRET upper limit of Coma at its peak frequency (see Fig. 16, left panel). Such gamma-ray flux could be, nonetheless, detectable by the GLAST–LAT detector (we will discuss more specifically the detectability of the gamma-ray WIMP annihilation signals from galaxy clusters in a dedicated, forthcoming paper (Colafrancesco, Profumo & Ullio 2006b). The rather low neutralino mass $`M_\chi =40`$ GeV of this model makes it rather difficult to be testable by Cherenkov gamma-ray detectors operating at higher threshold energies.
Increasing the neutralino mass does not provide a good fit of the radio-halo spectrum (see Fig. 16, right panel) and yields, in addition, extremely faint EUV, HXR and gamma-ray fluxes, which turn out to be undetectable even by GLAST and/or by the next coming high-energy experiments.
It is possible to recover the EUV and HXR data on Coma with a IC flux by secondary electrons by increasing the annihilation cross-sections by a factor $`10^2`$ (i.e., up to values $`\sigma v_0710^{23}cm^3s^1`$) in the best-fit $`b\overline{b}`$ soft WIMP model (at fixed $`M_\chi =40`$ GeV). However, in such a case both the radio-halo flux and the hard gamma-ray flux at $`1`$ GeV as produced by $`\pi ^0`$ decay should increase by the same factor leading to a problematic picture: in fact, while the radio-halo data would imply lower values of the magnetic field $`B0.1`$ $`\mu `$G which might still be allowed by the data, the $`\pi ^0\gamma \gamma `$ gamma-ray flux at $`E>100`$ MeV should exceed the EGRET limit on Coma. This option is therefore excluded by the available data.
Alternatively, it would be possible to fit the EUV and HXR spectra of Coma with the adopted value of $`\sigma v_0710^{23}cm^3s^1`$ for the $`b\overline{b}`$ model with $`M_\chi =40`$ GeV, in the case we sensibly lower the mean magnetic field. Values of the average magnetic field $`<\text{ }0.2\mu `$G are required to fit the HXR flux of Coma under the constraint to fit at the same time the radio-halo spectrum (see Fig.20, left panel), consistently with the general description of the ratio between the synchrotron and IC emission powers in Coma (see, e.g., Colafrancesco et al. (2005), Reimer et al. (2004)). Magnetic fields as low as $`0.15\mu `$G can fit both the HXR and the EUV fluxes of Coma. However, also in this case the $`\pi ^0\gamma \gamma `$ gamma-ray flux predicted by the same model at $`E>100`$ MeV exceeds the EGRET limit on Coma, rendering untenable this alternative. Actually, the EGRET upper limit on Coma set a strong constraint on the combination of values $`B`$ and $`\sigma v_0`$ (see Fig.20, right panel) so that magnetic field larger than $`>\text{ }0.3`$ $`\mu `$G are required for the parameter setup of the $`b\overline{b}`$ model with $`M_\chi =40`$ GeV. Fig.20 shows the upper limits on the value of $`\sigma v`$ as a function of the assumed value of the mean magnetic field of Coma. According to these results, it is impossible to fit all the available data on Coma for a consistent choice of the DM model and of the cluster magnetic field. The EUV and HXR data in particular require extreme conditions, i.e. low values of the magnetic field and/or high values of the annihilation cross section, which violate the EGRET gamma-ray limit. Thus, realistic DM models that are consistent with the radio and gamma-ray constraints predict IC emission which falls short of fitting the EUV and HXR data of Coma.
An appealing property of the WIMP model worked out here is that it can reproduce both the spatial distribution of the radio-halo surface brightness of Coma and, in principle, also the spatial profile of the EUV emission (see, e.g., Bowyer et al. (2004)) which seems more concentrated than the radio-halo surface brightness. As for the radio-halo surface brightness profile, it seems necessary for this WIMP model \- due to the shape of the DM halo profile - to invoke a radial distribution of the magnetic field with a mild decrease towards the Coma center to counterbalance the centrally peaked DM profile, and with an exponential cutoff at large radii to counterbalance the effect of the subhalo distribution. We notice here that such a specific $`B\left(r\right)`$ spatial distribution is - interestingly enough - able to reproduce the radial distribution of the RMs in Coma (see Sect.5.1). While the (Synchrotron) radio surface brightness depends strongly on the magnetic field radial profile, the (ICS) EUV radial profile only depends on the DM halo profile and on the secondary electron properties and is, hence, more concentrated (see Fig. 10 for an example). Thus, the radial distribution of the EUV emission could be reasonably reproduced by the WIMP model which best fits the radio data but with a very low value of the magnetic field of order of $`<\text{ }0.15`$ $`\mu `$G. We already noticed, however, that models with values of the average magnetic field in Coma which are $`<\text{ }0.3\mu `$G produce a gamma-ray flux which exceeds the EGRET upper limit of Coma (see Fig. 20), rendering these models untenable.
We summarize all the constraints on the neutralino models set by the magnetic field and by the annihilation cross-section in Figs. 20 and 21. The available data set constraints on the WIMP annihilation rate. Figs.22 and 23 show the upper limits on $`\sigma v_0`$ as a function of the assumed value for the mean magnetic field in Coma. The EGRET limit proves to be the more constraining at the moment with respect to the HXR and EUV data. These limits are able to test directly the annihilation rate since they are independent of the magnetic field value. Nonetheless, the combination of the gamma-ray and/or HXR constraints with the radio constraints will be able to determine the full setup of the relevant quantities whose combination is able to fit the overall Coma SED. In this context, it is clear that the possible GLAST observations of Coma, combined with the radio data, will increase by far the constraints in the $`\sigma v_0B`$ plane.
The results of our analysis also depend on the assumed DM halo density profile. Fig. 22 shows the scaling of fluxes with the assumptions on the halo model for Coma. We compare here, for the sake of illustration, the scalings of the N04 and of the Burkert model. Switching to one of the halo models displayed here is equivalent to shifting all values of $`\sigma v_0`$ plotted in the figures to $`\sigma v_0`$ divided by the scaling value shown here. This analysis allows us to compare correctly the results of the multi-frequency analysis we have presented in this paper in terms of substructure enhancement and halo density profile.
Table 4 also shows the typical values for the annihilation cross-section $`\sigma v_0`$ and the relative signals expected at different frequencies for the Coma best-fit model ($`b\overline{b}`$ neutralino model with $`M_\chi =40`$ GeV and $`B=1`$ $`\mu `$G) that we explored in this paper.
Large neutralino pair annihilation cross section will, in general, produce sizable signals also for other indirect detection techniques, including antimatter searches and gamma rays from the center of the Milky Way. Antimatter and gamma-ray fluxes also largely depend on the Milky Way dark-matter halo and on the specific neutralino model (e.g. through the antimatter yield per neutralino annihilation). Existing analysis (Baer and Profumo (2005); Profumo and Ullio (2004); Baer et al. 2005a ; Profumo (2005)) make it possible to draw some qualitative estimates of the cross-section, $`\sigma v_0`$, required to produce sensible signals at future DM search experiments. We provide in Table 4 order-of-magnitude estimates for the value of $`\sigma v_0`$ expected to give observable signals in space-based antimatter (AMS-02, Pamela, GAPS) and gamma-ray search experiments (GLAST), for two extreme choices of the galactic dark-matter halo: a cuspy profile (such as the N04 profile) and a cored profile (such as the Burkert profile). Different search techniques, such as direct detection, or neutrino flux detection from the core of the Sun induced by the annihilation of captured neutralinos, critically depend upon the scattering cross section of neutralinos off nucleons, and the resulting detection rates are therefore unrelated, in general, to the pair annihilation process discussed here.
We show in Fig. 23 the overall SED of Coma as expected from the four benchmark models described in the Sect.3.2. The predictions are shown for the best fit N04 profile and for our reference choice for subhalo parameters, and for a mean magnetic field of $`2\mu G`$. None of these benchmark model may provide a reasonable fit to the radio data. Notice, in addition, the quite dim multi-frequency SED predicted for Coma in these benchmark models. The largest fluxes are, not surprisingly, obtained for the model lying in the focus point region ($`E^{}`$). In that case, neutralinos mostly annihilate into gauge bosons, as can be inferred from the spectral shape, which closely resembles that in Fig. 21. We therefore conclude that the expectation of astrophysical signatures from neutralino DM annihilations in the Coma cluster (with natural assumptions on the dark halo profile, substructures and magnetic field of Coma) is not promising in the context of the commonly discussed minimal supergravity scenario.
It would be interesting to compare the predictions of the WIMP annihilation for Coma with the implication of the presence of another population of cosmic rays of different origin like that, often invoked, produced by acceleration processes in the atmosphere of Coma. Acceleration scenarios usually produce power-law spectra for the electrons which are primarily accelerated by shocks or turbulence and are, hence remarkably different from the source spectra produced by neutralino acceleration. Specifically, acceleration models do not exhibit a cut-off at the neutralino mass and do not produce the peculiar peaked $`\pi ^0`$ gamma-ray emission which remains a distinctive feature of neutralino DM models. A continuum $`\pi ^0`$ gamma-ray emission can be produced in secondary models where the electrons are produced by proton-proton collision in the cluster atmosphere. But even in this case the gamma-ray spectrum is likely to be resembled by a power-law shape which keeps memory of the original acceleration events for the hadrons. Thus, it will be possible to separate DM annihilation models from acceleration models based on multi-frequency observations of the hadronic and leptonic components of the cluster SED.
Finally, It should be noticed that the same problem with the consistent fitting of both the synchrotron and IC components to the radio and EUV/HXR data of Coma still remains in both DM and acceleration models, pointing to the fact that these spectral features, if real, have probably different physical origin.
## 6 Summary and conclusions
WIMP annihilations in galaxy cluster inevitably produce high-energy secondary particles which are able, in turn, to produce a wide SED extended over more than 18 orders of magnitude in frequency, from radio to gamma-rays.
A consistent analysis of the DM distribution and of its annihilation in the Coma cluster shows that WIMP annihilation is able to reproduce both the spectral and the spatial features of the Coma radio halo under reasonable assumptions for the structure of the intracluster magnetic field. The mild decrease of the magnetic field towards the Coma center, which reproduces the radial trend of the observed RM distribution in Coma, could be better tested with a larger dataset of Faraday rotation measures of background radio sources obtainable with the next generation sensitive radio telescopes (LOFAR, SKA), and with the help of numerical MHD simulations. Radio data are the main constraint, so far, to WIMP models.
The ICS emission produced by the same secondary electrons is able, in principle, to reproduce both the spectrum and the spatial distribution of the EUV emission observed in Coma, provided that a quite small average magnetic field $`B0.15`$ $`\mu `$G is assumed. Such low value of the B field is also able to make the radio data and the hard X-ray data of Coma consistent within a Synchrotron/IC model for their origins. However, such low magnetic field values in Coma produce an unacceptably large gamma-ray flux, which exceeds the EGRET upper limit. The gamma-ray constraints are thus the most stringent ones for the analysis of the astrophysical features of DM annihilations.
In conclusion, the viable models of WIMP annihilation which are consistent with the available data for Coma yield a nice fit to the radio data but produce relatively low intensity emission at EUV, X-ray and gamma-ray frequencies. The hadronic gamma-ray emission could be, nonetheless, detected by the GLAST-LAT detector. These models also produce negligible heating rates for the kind of non-singular halo profile we worked out in this paper. It is interesting that the best-fit ($`b\overline{b}`$) WIMP model with $`M_\chi =40`$ GeV predicts a detectable SZ effect (with a peculiar spectrum very different from that of the thermal SZ effect) at the level of $``$ 40 to 10 $`\mu `$K in the frequency range $`10200`$ GHz, which could be observable with the next generation high-sensitivity bolometric arrays, space and balloon-borne microwave experiments, like PLANCK, OLIMPO, APEX, ALMA.
The observational ”panorama” offered by the next coming radio, SZ, and gamma-ray astronomical experiments might produce further constraints on the viable SUSY model for Coma and for other large-scale cosmic structures. Direct DM detection experiments have already explored large regions of the most optimistic SUSY models, and the planned increase in sensitivity of the next-generation experiments will probably be able to explore even the core of the SUSY models. In this context, we have shown that indirect DM detection proves to be not only complementary, but also hardly competitive, especially when a full multi-frequency approach is chosen. When combined with future accelerator results, such multi-frequency astrophysical searches might greatly help us to unveil the elusive nature of dark matter.
###### Acknowledgements.
We thank the Referee for the useful comments and suggestions that allowed us to improve the presentation of our results. S.C. acknowledges support by PRIN-MIUR under contract No.2004027755$`\mathrm{\_}`$003.
## Appendix A A solution to the diffusion equation
To understand quantitatively the role of the various populations of secondary particles emitting in the Coma cluster, we have to describe in details their transport, diffusion and energy loss. We consider the following diffusion equation (i.e. neglecting convection and re-acceleration effects):
$`{\displaystyle \frac{}{t}}{\displaystyle \frac{dn_e}{dE}}`$ $`=`$ $`\left[D(E,x){\displaystyle \frac{dn_e}{dE}}\right]+{\displaystyle \frac{}{E}}\left[b(E,x){\displaystyle \frac{dn_e}{dE}}\right]`$ (64)
$`+Q_e(E,x).`$
We search for an analytic solution of the diffusion equation in the case of diffusion coefficient and energy loss term that do not depend on the spatial coordinates, i.e. we take:
$`D`$ $`=`$ $`D\left(E\right)`$ (65)
$`b`$ $`=`$ $`b\left(E\right)`$ (66)
and we implement a slight variant of the method introduced in Baltz & Edsjo (1998) and Baltz & Wai (2004).
Let us define the variable $`u`$ as:
$$b\left(E\right)\frac{dn_e}{dE}=\frac{dn_e}{du}$$
(67)
which yields
$$u=_E^{E_{\mathrm{max}}}\frac{dE^{}}{b\left(E^{}\right)}$$
(68)
Then, it follows that $`b\left(E\right)=E/\tau _{loss}`$ in terms of the time scale $`\tau _{loss}`$ for the energy loss of the relativistic particles, which, for $`E_{\mathrm{max}}=\mathrm{}`$, gives $`u=\tau `$.
The diffusion equation can be rewritten as
$$\left[\frac{}{t}+D\left(E\right)\mathrm{\Delta }\frac{}{u}\right]\frac{dn_e}{du}=b\left(E\right)Q_e(E,x).$$
(69)
We search for the Green function $`G`$ of the operator on the left-hand-side. Consider the equation for its 4-dimensional Fourier transform ($`t\omega `$, $`xk`$):
$`\left[i\omega +D\left(E\right)k^2{\displaystyle \frac{}{u}}\right]\stackrel{~}{G}`$ $`=`$ $`{\displaystyle \frac{1}{\left(2\pi \right)^2}}\mathrm{exp}\left[i\left(\omega t^{}+kx^{}\right)\right]`$ (70)
$`\delta \left(uu^{}\right),`$
which has the solution
$`\stackrel{~}{G}`$ $`=`$ $`{\displaystyle \frac{1}{\left(2\pi \right)^2}}\mathrm{exp}[i(\omega t^{}+kx^{})i\omega (uu^{})`$ (71)
$`k^2{\displaystyle _u^{}^u}d\stackrel{~}{u}D\left(\stackrel{~}{u}\right)].`$
Transforming back from the Fourier space we find:
$`G_{\mathrm{free}}`$ $`=`$ $`{\displaystyle \frac{1}{\left(4\pi \left(vv^{}\right)\right)^{3/2}}}\mathrm{exp}\left[{\displaystyle \frac{\left|xx^{}\right|^2}{4\left(vv^{}\right)}}\right]`$ (72)
$`\delta \left(\left(tt^{}\right)\left(uu^{}\right)\right)`$
where we defined $`dvD\left(u\right)du`$, i.e. $`v=_{u_{\mathrm{min}}}^u𝑑\stackrel{~}{u}D\left(\stackrel{~}{u}\right)`$. The suffix ’free’ refers to the fact that there are no boundary conditions yet. These are implemented with the image charges method. To apply this technique to galaxy clusters, we can consider the approximation of spherical symmetry with Green function vanishing at the radius $`r_h`$. Introducing the set of image charges $`(r_n,\theta _n,\varphi _n)=(\left(1\right)^nr+2nr_h,\theta ,\varphi )`$, one can verify that
$$G(r,Y)=\underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}\left(1\right)^nG_{\mathrm{free}}(r_n,Y)$$
(73)
fulfills such boundary condition (here $`Y`$ labels the other variables in the Green function). Moreover, we choose the reference frame in such way that we look at the signal along the $`z`$ polar axis ($`\mathrm{cos}\theta =1`$) so that $`\left|x^{}x_n\right|^2=\left(r^{}\right)^2+r_n^22\mathrm{cos}\theta ^{}r^{}r_n`$. If the source function does not depend on $`\theta ^{}`$ and $`\varphi ^{}`$, the integral on these two variables can be performed explicitly and we find
$`{\displaystyle \frac{dn_e}{dE}}`$ $`=`$ $`{\displaystyle \frac{1}{b\left(E\right)}}{\displaystyle _E^{M_\chi }}dE^{}{\displaystyle \frac{1}{\left[4\pi \left(vv^{}\right)\right]^{1/2}}}{\displaystyle \underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}}(1)^n{\displaystyle _0^{r_h}}dr^{}{\displaystyle \frac{r^{}}{r_n}}`$ (74)
$`\left[\mathrm{exp}\left({\displaystyle \frac{\left(r^{}r_n\right)^2}{4\left(vv^{}\right)}}\right)\mathrm{exp}\left({\displaystyle \frac{\left(r^{}+r_n\right)^2}{4\left(vv^{}\right)}}\right)\right]Q_e(r^{},E^{},t^{})`$
with $`t^{}=t\left(uu^{}\right)`$ (or no time dependence for stationary source). Note that $`E^{}>E`$ (energy is lost) and hence $`u^{}<u`$, $`v^{}<v`$ and $`t^{}<t`$.
### A.1 Stationary limit and role of spatial diffusion in Coma
In the limit of time-independence of the source and electron number density that has already reached equilibrium, Eq. (74) takes the form:
$$\frac{dn_e}{dE}(r,E)=\frac{1}{b\left(E\right)}_E^{M_\chi }𝑑E^{}\widehat{G}(r,vv^{})Q_e(r,E^{})$$
(75)
with
$`\widehat{G}(r,\mathrm{\Delta }v)`$ $`=`$ $`{\displaystyle \frac{1}{\left[4\pi \left(\mathrm{\Delta }v\right)\right]^{1/2}}}{\displaystyle \underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}}(1)^n{\displaystyle _0^{r_h}}dr^{}{\displaystyle \frac{r^{}}{r_n}}`$
$`\left[\mathrm{exp}\left({\displaystyle \frac{\left(r^{}r_n\right)^2}{4\mathrm{\Delta }v}}\right)\mathrm{exp}\left({\displaystyle \frac{\left(r^{}+r_n\right)^2}{4\mathrm{\Delta }v}}\right)\right]{\displaystyle \frac{n_\chi ^2\left(r^{}\right)}{n_\chi ^2\left(r\right)}}.`$
In the limit in which electrons and positrons lose energy on a timescale much shorter than the timescale for spatial diffusion, i.e. if the first term on the r.h.s. of Eq. (64) can be neglected, the expression for equilibrium number density becomes:
$$\left(\frac{dn_e}{dE}\right)_{nsd}(r,E)=\frac{1}{b\left(E\right)}_E^{M_\chi }𝑑E^{}Q_e(r,E^{}).$$
(77)
This is analogous to the form in Eq. (75), except for the factor $`\widehat{G}(r,vv^{})`$ in the integrand: it follows that the latter is the Green function term which we need to study to understand whether spatial diffusion is important or not.
Since we have encoded the dependence on the energy loss term and the diffusion coefficient in the definition of the variable $`v`$, we preliminarily study what range of $`\mathrm{\Delta }v`$ is relevant in the discussion. To do that, we need to specify $`D\left(E\right)`$ and $`b\left(E\right)`$. For the diffusion coefficient we assume the form:
$$D\left(E\right)=D_0\frac{d_B^{2/3}}{B_\mu ^{1/3}}\left(\frac{E}{1\mathrm{GeV}}\right)^{1/3},$$
(78)
(Colafrancesco & Blasi (1998); Blasi & Colafrancesco (1999)) where $`d_B`$ is the minimum scale of uniformity of the magnetic field in kpc (throughout the paper we assume $`d_B20`$ for Coma), $`B_\mu `$ is the average magnetic field in $`\mu `$G units, and $`D_0`$ some constant that we estimate as $`D_0=3.1\times 10^{28}\mathrm{cm}^2\mathrm{s}^1`$.
The energy loss term is the sum of effects due to Inverse Compton, synchrotron radiation, Coulomb losses and Bremsstrahlung:
$`b\left(E\right)`$ $`=`$ $`b_{IC}\left(E\right)+b_{syn}\left(E\right)+b_{Coul}\left(E\right)+b_{brem}\left(E\right)`$ (79)
$`=`$ $`b_{IC}^0\left({\displaystyle \frac{E}{1\mathrm{GeV}}}\right)^2+b_{syn}^0B_\mu ^2\left({\displaystyle \frac{E}{1\mathrm{GeV}}}\right)^2`$
$`+b_{Coul}^0n\left(1+\mathrm{log}\left(\gamma /n\right)/75\right)`$
$`+b_{brem}^0n\left(\mathrm{log}\left(\gamma /n\right)+0.36\right).`$
Here $`n`$ is the mean number density of thermal electrons in $`\mathrm{cm}^3`$ (see Eq. (18), the average over space gives about $`n\mathrm{1.3\hspace{0.33em}10}^3`$), $`\gamma E/m_e`$ and we find $`b_{IC}^00.25`$, $`b_{syn}^00.0254`$, $`b_{Coul}^06.13`$ and $`b_{brem}^01.51`$, all in units of $`10^{16}\mathrm{GeV}\mathrm{s}^1`$. For GeV electrons and positrons the Inverse Compton and synchrotron terms dominate (see Fig.26).
To get a feeling about what is the electron/positron energy range which will be of interest when considering the radio emissivity, we can resort to the ”monochromatic” approximation, with relativistic particles of a given energy $`E`$ radiating at a single frequency, namely the peak frequency:
$$\nu 0.29\frac{3}{2}\frac{eB}{2\pi m_ec}\left(4.7\mathrm{MHz}\right)\mathrm{B}_\mu \left(\frac{\mathrm{E}}{\mathrm{GeV}}\right)^2.$$
(80)
Since radio data on Coma extend down to about $`30\mathrm{M}\mathrm{H}\mathrm{z}`$, for magnetic fields not much larger than 10 $`\mu `$G, this translates into radiating particles with energies larger than about 1 GeV.
As a sample case, in Fig. 24 we consider a WIMP mass $`M_\chi =100`$ GeV and sketch the mapping between the energy $`E^{}(E,M_\chi )`$, with $`E`$ some reference energy after diffusion, and the square root of $`\mathrm{\Delta }v=vv^{}v\left(E\right)v\left(E^{}\right)`$, for a few values of $`E`$ and of the mean magnetic field $`B_\mu `$. We find as largest value $`\left(\mathrm{\Delta }v\right)^{1/2}35`$ kpc, corresponding to $`E=1`$ GeV and $`B_\mu =1`$ $`\mu `$G; the maximum value of $`\mathrm{\Delta }v`$ diminishes rapidly when increasing $`E`$ or $`B_\mu `$.
On the right-hand side of Fig. 24, we plot $`\widehat{G}`$ as a function of $`\left(\mathrm{\Delta }v\right)^{1/2}`$ for a few values of the radial coordinate $`r`$ and in case the DM halo of Coma is described by a N04 profile. In the very central part of the halo, there are significant departures of the value of $`\widehat{G}`$ from unity, on scales $`\left(\mathrm{\Delta }v\right)^{1/2}`$ at which, for the given radius $`r`$, the mean squared value of the DM profile is significantly different from the square of the value of the profile at $`r`$. Note, however, that this effect is confined in the innermost region of the cluster, corresponding to an angular size of $`12`$ arcmin. We then expect that taking into account spatial diffusion will modify only slightly the predictions for the radio surface brightness distribution from moderate to large radial distances in Coma.
In Fig. 25, we plot $`\widehat{G}`$ as a function of $`\left(\mathrm{\Delta }v\right)^{1/2}`$ for the same values of the radial coordinate $`r`$ as in Fig. 24 but now in case the DM halo of Coma is described by a Burkert profile. It is clear that departures from unity are essentially negligible even in the inner portion of the halo and, hence, spatial diffusion can be safely neglected for all practical purposes in this case.
To get a more physical insight on the reason why spatial diffusion can be neglected, it is useful to consider the following qualitative solution (see, e.g., Colafrancesco (2005)) for the average electron density
$$\frac{dn_e(E,r)}{dE}\left[Q_e(E,r)\tau _{loss}\right]\times \frac{V_s}{V_s+V_o}\times \frac{\tau _D}{\tau _D+\tau _{loss}}$$
(81)
which resumes the relevant aspects of the transport equation (Eq. 64). Here, $`V_sR_h^3`$ and $`V_o\lambda ^3\left(E\right)`$ are the volumes occupied by the DM source and the one occupied by the diffusing electrons which travel a distance $`\lambda \left(E\right)\left[D\left(E\right)\tau _{loss}\left(E\right)\right]^{1/2}`$ before loosing much of their initial energy. The relevant time scales in Eq. (64) are the diffusion time-scale, $`\tau _DR_h^2/D\left(E\right)`$, and the energy loss time-scale $`\tau _{loss}=E/b_e\left(E\right)`$, where $`D`$ is again the diffusion coefficient for which we can assume the generic scaling $`D\left(E\right)=\stackrel{~}{D}_0\left(E/E_0\right)^\gamma B^\gamma `$, and $`b\left(E\right)`$ the electron energy loss per unit time at energy $`E`$.
For $`E>E_{}=\left(\stackrel{~}{D}_0E_0/R_h^2b_0B_\mu ^\gamma \right)^{1/\left(1\gamma \right)}`$ (for simplicity we have kept leading terms only, implementing $`b\left(E\right)b_0\left(B_\mu \right)\left(E/GeV\right)^2+b_{Coul}`$), the condition $`\tau _D>\tau _{loss}`$ (and consistently $`\lambda \left(E\right)<R_h`$) holds, the diffusion is not relevant and the solution of Eq. (64) is $`dn_e/dEQ_e(E,r)\tau _{loss}`$ and shows an energy spectrum $`Q\left(E\right)E^1`$. This situation ($`\lambda \left(E\right)<R_h`$, $`\tau _D>\tau _{loss}`$) applies to the regime of galaxy clusters which we discuss here for the specific case of Coma, as one can see from Fig. 26.
For $`E<E_{}`$, the condition $`\tau _D<\tau _{loss}`$ (and consistently $`\lambda \left(E\right)>R_h`$) holds, the diffusion is relevant and the solution of Eq. (64) is $`dn_e/dE\left[Q_e(E,r)\tau _D\right]\times \left(V_s/V_o\right)`$ and shows an energy spectrum $`Q\left(E\right)E^{\left(25\gamma \right)/2}`$ which is flatter or equal to the previous case for reasonable values $`\gamma =1/31`$. This last situation ($`\lambda \left(E\right)>R_h`$, $`\tau _D<\tau _{loss}`$) applies to the regime of dwarf galaxies and we will discuss this case more specifically elsewhere (Colafrancesco, Profumo & Ullio (2006)).
Fig.27 shows the energy shape of the electron equilibrium spectra derived in our approach for a ($`b\overline{b}`$) model with $`M_\chi =40`$ GeV and for a $`W^+W^{}`$ model with $`M_\chi =81`$ GeV. The astrophysical predictions of these two models will be extensively discussed in the following.
We notice that the energy losses in the diffusion equation erase almost completely the details of the electron source spectra (see Fig. 7). The equilibrium spectra are generally characterized by three different regions: i) a low-energy plateau at $`E<\text{ }0.1`$ GeV with a constant value of $`dn_e/dE`$ which remains almost constant down to the electron rest-mass energy; ii) an almost power-law branch at $`0.1M_\chi <\text{ }E<\text{ }0.5M_\chi `$ which is steeper in the softer $`b\overline{b}`$ annihilation final state with respect to the hard spectrum due to a $`W^+W^{}`$ channel; and iii) a sharp cut-off at the energy corresponding to the neutralino mass which marks the natural maximum energy of the secondary electron spectra. We will show in the next Sect.4 how these three branches of the electron equilibrium spectra will provide observable features in the multi-frequency spectrum of Coma and can, consequently, be used to constrain the neutralino model. |
warning/0507/quant-ph0507160.html | ar5iv | text | # SUSY transformation of the Green function and a trace formula
## 1 Introduction
At present there is a growing interest in the study of different properties transformations induced by supersymmetry (SUSY) in Quantum Mechanics. Recently a special issue of Journal of Physics A (see vol. 37, No 43, 2004) was devoted to research work in this subject. Despite of the growing number of papers in this field many questions still remain open and require further study. In particular, the authors are aware of only one paper devoted to the study SUSY transformations at the level of Green functions. For the case of a transformation deleting the ground state of the initial Hamiltonian, Sukumar has studied an integral relation between the Green functions for SUSY partners and has formulated conditions leading to the vanishing of some matrix elements of a Hamiltonian and related this property to a hidden supersymmetry of the system. Transformation of Green functions is not explicitly discussed in that paper. Moreover, we have found that formula (28) of relating integrals over Green functions for SUSY partners may need to be corrected if a continuous spectrum is present.
In this paper we give a simple formula for the Green function of the SUSY partner Hamiltonian both for confining and for scattering potentials and generalize results of the paper to the case where the continuous spectrum is present. As an application of this general formula we consider the case of the Schrödinger equation with a scattering potential defined both on the whole real axis and on a half line when the Schrödinger equation is reduced to a singular Sturm-Liouville problem. Regular Sturm-Liouville problem is considered in a separate publication .
## 2 Green function of the Schrödinger equation
In this section we cite some properties of the Green function of the one-dimensional Schrödinger equation for a spectral problem on the whole real line (see e.g. ) which are useful.
We consider the Schrödinger equation
$$\left(h_0E\right)\psi =0h_0=d^2/dx^2+V_0\left(x\right)x(a,b)$$
(1)
supplemented by the boundary conditions $`\psi \left(a\right)=\psi \left(b\right)=0`$. We will concentrate mostly on two cases; these are the whole real line $`a=\mathrm{}`$ and $`b=\mathrm{}`$ and the half line $`a=0`$ and $`b=\mathrm{}`$. We assume that the spectral set $`\text{spec}h_0`$ of this problem consists of $`M`$ discrete points with the possibilities $`M=0`$ or $`M=\mathrm{}`$ and possibly a continuum part filling the positive semiaxis.
The definition of the Green function used by different authors may differ by a constant factor. We use a definition of the Green function represented as the kernel of the operator $`\left(h_0E\right)^1`$ as an operator defined in the corresponding Hilbert space. It is well defined for all $`E\text{spec}h_0`$. It has two different but equivalent representations. The first representation is obtained with the help of two real solutions of equation (1) with a fixed value of the parameter $`E\text{spec}h_0`$, $`f_{l0}`$ and $`f_{r0}`$ (“left” and “right” solutions), satisfying zero boundary conditions: $`f_{l0}\left(a\right)=0`$, $`f_{r0}\left(b\right)=0`$. Since they correspond to the same $`E`$ their Wronskian $`W_0=W(f_{r0},f_{l0})`$ does not depend on $`x`$ and is a function of $`E`$ only, and the Green function is
$`G_0(x,y,E)=f_{l0}(x,E)f_{r0}(y,E)/W_0xy`$ (2)
$`G_0(y,x,E)=G_0(x,y,E).`$ (3)
These formulae are clearly equivalent to
$$G_0(x,y,E)=\left[f_{l0}(x,E)f_{r0}(y,E)\mathrm{\Theta }\left(yx\right)+f_{l0}(y,E)f_{r0}(x,E)\mathrm{\Theta }\left(xy\right)\right]/W_0$$
(4)
where $`\mathrm{\Theta }`$ is the Heaviside step function.
If the operator $`h_0`$ is essentially self-adjoint the set of its discrete spectrum (if present) eigenfunctions $`\left\{\psi _n\right\}`$, $`n=0,1,\mathrm{},M`$, $`\psi _n|\psi _m=\delta _{nm}`$ together with the continuous spectrum eigenfunctions (also if present) $`\psi _k`$, $`E=k^2>0`$, $`\psi _k|\psi _k^{}=\delta \left(kk^{}\right)`$, $`\psi _n|\psi _k=0`$ is complete in the Hilbert space
$$\underset{n=0}{\overset{M}{}}\psi _n\left(x\right)\psi _n^{}\left(y\right)+𝑑k\psi _k\left(x\right)\psi _k^{}\left(y\right)=\delta \left(xy\right)$$
and the second representation of the Green function may be found in terms of this set as follows:
$$G_0(x,y,E)=\underset{n=0}{\overset{M}{}}\frac{\psi _n\left(x\right)\psi _n^{}\left(y\right)}{E_nE}+\frac{\psi _k\left(x\right)\psi _k^{}\left(y\right)}{k^2E}𝑑k.$$
(5)
For the spectral problem on the whole real axis the continuous spectrum is two-fold degenerate and the integrals over $`k`$ run from minus infinity to infinity and for the problem on a half line they run from zero to infinity.
## 3 SUSY transformation of the Green function
It is well-known (see e.g. ) that there exist three kinds of SUSY transformations:
(i) deleting the ground state level of $`h_0`$
(ii) creating a new ground state level and
(iii) purely isospectral transformation.
In all cases the partner Hamiltonian $`h_1=d^2/dx^2+V_1`$ for $`h_0`$ is defined by the potential
$$V_1\left(x\right)=V_0\left(x\right)2w^{}\left(x\right)w\left(x\right)=\left[\mathrm{log}u\left(x\right)\right]^{}$$
(6)
where $`u`$ is a real solution to the initial equation $`\left(h_0\alpha \right)u=0`$ with $`\alpha `$ known as the factorization constant. We adopt the notation that a derivative with respect to $`x`$ is denoted by the prime symbol. To provide a nonsingular potential difference $`\alpha `$ should be less than or equal to the ground state energy of $`h_0`$ if it has a discrete spectrum or lower than the continuum threshold otherwise. The functions $`\phi _n=L\psi _n`$, $`n=1,2,\mathrm{},M`$ describe (unnormalized) bound states and $`\phi _E=L\psi _E`$ correspond to (unnormalized) scattering states of $`h_1`$. Here
$$L=d/dx+w\left(x\right)$$
(7)
is the transformation operator (intertwiner) satisfying $`Lh_0=h_1L`$ and $`Lu=0`$. The normalization constants are easily obtained with the help of the factorization property $`L^+L=h_0\alpha `$ where $`L^+=d/dx+w\left(x\right)`$. The functions
$$\chi _n=\left(E_n\alpha \right)^{1/2}L\psi _n\chi _E=\left(E\alpha \right)^{1/2}L\psi _E$$
(8)
form an orthonormal set.
###### Theorem 1
Let $`G_0(x,y,E)`$ be the Green function for $`h_0`$. Then for all three cases enumerated above the Green function for $`h_1`$ is
$$G_1(x,y,E)=\frac{1}{E\alpha }\left[L_xL_yG_0(x,y,E)\delta \left(xy\right)\right].$$
(9)
In case (ii) it has a simple pole at $`E=\alpha `$. In cases (i) and (iii) it is regular at $`E=\alpha `$ and can be calculated as follows:
$$G_1(x,y,\alpha )=\left[L_xL_y\frac{G_0(x,y,E)}{E}\right]_{E=\alpha }.$$
(10)
Here $`L_x`$ is the operator given in (7) and $`L_y`$ is the same operator where $`x`$ is replaced by $`y`$.
Proof. In case (i) $`u=\psi _0`$ and the set $`\left\{\chi _E,\chi _n,n=1,2,\mathrm{},M\right\}`$ is complete. Therefore
$$G_1(x,y,E)=\underset{n=1}{\overset{M}{}}\frac{\chi _n\left(x\right)\chi _n^{}\left(y\right)}{E_nE}+\frac{\chi _k\left(x\right)\chi _k^{}\left(y\right)}{k^2E}𝑑k.$$
(11)
Now we replace $`\chi `$ using (8) which yields
$$\begin{array}{c}G_1(x,y,E)=\hfill \\ \frac{1}{\alpha E}L_xL_y(_{n=1}^M\left[\frac{1}{E_n\alpha }\frac{1}{E_nE}\right]\psi _n\left(x\right)\psi _n^{}\left(y\right)+𝑑k\left[\frac{1}{k^2\alpha }\frac{1}{k^2E}\right]\psi _k\left(x\right)\psi _k^{}\left(y\right)).\hfill \end{array}$$
The statement for $`E\alpha `$ follows from here if in the first sum and in the first integral we express $`L\psi `$ in terms of $`\chi `$, make use of the completeness condition for the set $`\chi `$ and formula (5) for $`G_0`$. The fact that here the sum starts from $`n=1`$ and in (5) it starts from $`n=0`$ cannot cause any problems since $`L\psi _0=0`$. For $`E=\alpha `$ formula (11) can be written in the form
$$G_1(x,y,\alpha )=\left[\frac{}{E}L_xL_y\left(\underset{n=0}{\overset{M}{}}\frac{\psi _n\left(x\right)\psi _n^{}\left(y\right)}{E_nE}+\frac{\psi _k\left(x\right)\psi _k^{}\left(y\right)}{k^2E}𝑑k\right)\right]_{E=\alpha }$$
from which (10) follows in this case.
In case (ii) let $`\chi _\alpha 1/u`$ be the normalized ground state function of $`h_1`$ corresponding to the new discrete level $`E=\alpha `$. Then
$$G_1(x,y,E)=\underset{n=0}{\overset{M}{}}\frac{\chi _n\left(x\right)\chi _n^{}\left(y\right)}{E_nE}+\frac{\chi _\alpha \left(x\right)\chi _\alpha ^{}\left(y\right)}{\alpha E}+\frac{\chi _k\left(x\right)\chi _k^{}\left(y\right)}{k^2E}𝑑k.$$
(12)
Now the use of exactly the same transformations as in case (i) reduces (12) to (9).
In case (iii) we start from the same formula (11) with the only difference that the sum now starts from $`n=0`$ and following the same line of reasoning as in the earlier cases we get formula (9). It is interesting to notice the intermediate result
$$G_1(x,y,E)=\frac{1}{\alpha E}L_xL_y\left[G_0(x,y,\alpha )G_0(x,y,E)\right]$$
(13)
which makes clear how formula (10) arises for this case by taking the limit $`E\alpha `$. The fact that in case (ii) the function (9) has a simple pole at $`E=\alpha `$ is a consequence of the equivalence between (12) and (9). $`\mathrm{}`$
###### Corollary 1
In terms of the special solutions $`f_{l0}`$ and $`f_{r0}`$ of the Schrödinger equation for $`h_0`$ the Green function $`G_1`$ for all three cases listed above may be expressed as follows:
$`G_1(x,y,E)={\displaystyle \frac{1}{\left(E\alpha \right)W_0}}[\mathrm{\Theta }(yx)L_xf_{l0}(x,E)L_yf_{r0}(y,E)`$
$`+\mathrm{\Theta }(xy)L_yf_{l0}(y,E)L_xf_{r0}(x,E)].`$ (14)
In case (ii) this function has a simple pole at $`E=\alpha `$. In cases (i) and (iii) it is regular at $`E=\alpha `$ and can be calculated as follows:
$`G_1(x,y,\alpha )`$
$`=\left[{\displaystyle \frac{}{E}}{\displaystyle \frac{\mathrm{\Theta }\left(yx\right)L_xf_{l0}(x,E)L_yf_{r0}(y,E)+\mathrm{\Theta }\left(xy\right)L_yf_{l0}(y,E)L_xf_{r0}(x,E)}{W_0}}\right]_{E=\alpha }`$ (15)
To prove these formulae we substitute $`G_0`$ as given in (4) into (9) and (10). Taking the derivative of the theta functions in (4) gives rise to the Dirac delta function which cancels out the delta function present in (9). Formula (1) is clearly valid since the SUSY transformations necessarily preserve the boundary conditions for all $`E`$ except perhaps for $`E=\alpha `$. This implies that $`f_{l1}=Lf_{l0}`$ vanishes at $`x=a`$ and $`f_{r1}=Lf_{r0}`$ vanishes at $`x=b`$. The denominator in (1) is just the Wronskian of $`f_{r1}`$ and $`f_{l1}`$ and may be given in the form $`W(f_{r1},f_{l1})=\left(E\alpha \right)W(f_{r0},f_{l0})=\left(E\alpha \right)W_0`$. $`\mathrm{}`$
## 4 Trace formulae
The trace of the Green function defined as $`_a^bG(x,x,E)𝑑x`$ is usually divergent if the system has a continuous spectrum. It is remarkable that the trace of the difference $`G_0(x,x,E)G_1(x,x,E)`$ is a finite quantity which may or may not be equal to zero. In some cases this fact may be explained by another remarkable property. It may happen that the difference of infinite normalizations (they diverge as $`\delta \left(xy\right)`$ when $`yx`$) of the continuous spectrum eigenfunctions of the two SUSY partners is a finite quantity.
###### Theorem 2
Let $`f_{l0}(x,E)`$ and $`f_{r0}(x,E)`$ be solutions of the Schrödinger equation for $`h_0`$ satisfying the zero boundary conditions at the left and right bound of the interval $`(a,b)`$ respectively, and
$$f_{l1}(x,E)=Lf_{l0}(x,E)f_{r1}(x,E)=Lf_{r0}(x,E)$$
(16)
be similar solutions for $`h_1`$ related with $`h_0`$ by a SUSY transformation with $`\alpha `$ being the factorization constant. Let $`W_0`$ be the Wronskian of $`f_{r0}`$ and $`f_{l0}`$, $`W_0=W(f_{r0},f_{l0})`$. Then
$$_a^b\left[G_0(x,x,E)G_1(x,x,E)\right]𝑑x=\frac{Q\left(E\right)}{W_0\left(E\alpha \right)}$$
(17)
where $`Q(E)`$ can be calculated by one of the following formulae:
$`Q\left(E\right)=\left(f_{r0}f_{l1}\right)_{x=b}\left(f_{r0}f_{l1}\right)_{x=a}=\left(f_{l0}f_{r1}\right)_{x=b}\left(f_{l0}f_{r1}\right)_{x=a}`$ (18)
$`=W_0+\left(f_{l0}f_{r1}\right)_{x=b}\left(f_{r0}f_{l1}\right)_{x=a}=W_0+\left(f_{r0}f_{l1}\right)_{x=b}\left(f_{l0}f_{r1}\right)_{x=a}.`$ (19)
Proof. From Corollary 1 it follows that $`G_1(x,x,E)=\frac{1}{W_0\left(E\alpha \right)}Lf_{l0}\left(x\right)Lf_{r0}\left(x\right)`$. While integrating this expression over the interval $`(a,b)`$ one can transfer the derivative present in $`L`$ either from $`f_{l0}`$ to $`f_{r0}`$ or from $`f_{r0}`$ to $`f_{l0}`$ which leads to one of the following integrands $`f_{l0}\left(x\right)L^+Lf_{r0}\left(x\right)`$ or $`f_{r0}\left(x\right)L^+Lf_{l0}\left(x\right)`$. In both cases the factorization property may be used to reduce the integrand to $`\left(E\alpha \right)f_{l0}\left(x\right)f_{r0}\left(x\right)`$. Thus we arrive at the relation
$$_a^bG_1(x,x)𝑑x=\frac{1}{W_0}_a^bf_{l0}\left(x\right)f_{r0}\left(x\right)𝑑x\frac{Q\left(E\right)}{W_0\left(E\alpha \right)}$$
(20)
where $`Q\left(E\right)`$ is given by (18). To prove (19) it is sufficient to notice that
$$f_{l0}(x,E)f_{r1}(x,E)f_{r0}(x,E)f_{l1}(x,E)=W_0$$
(21)
which is a consequence of (16). The identification of the integrand on the right hand side of (20) as $`W_0G_0(x,x)`$ then leads to the result given in (17). $`\mathrm{}`$
Using the first of equalities (18) one can rewrite (17) as follows:
$$_a^b\left[G_0(x,x,E)G_1(x,x,E)\right]𝑑x=\frac{1}{\alpha E}+\frac{\left(f_{l0}f_{r1}\right)_{x=b}\left(f_{r0}f_{l1}\right)_{x=a}}{W_0\left(E\alpha \right)}$$
(22)
Now for the case (i) where $`\alpha =E_0`$ if we compare this result with the corresponding difference which can be obtained directly from the expressions for $`G_0`$ given by (5) and for $`G_1`$ given by (11) the following feature may be noted: the first term on the right hand side of (22) arises from the contribution to Green functions from the discrete spectra and the second term, which as we show below may be different of zero, is due to the presence of the continuous spectra. Just this contribution was neglected in . So, Theorem 2 presents a generalization of the result obtained in to the case where a continuous spectrum may be present. As an application of this theorem we are going to consider two particular cases of scattering potentials defined both on the whole real line and on a semiaxis.
###### Corollary 2
If $`h_0`$ is a scattering Hamiltonian with the potential $`V_0`$ satisfying for the spectral problem on the whole line the condition
$$_{\mathrm{}}^{\mathrm{}}\left(1+\left|x\right|\right)\left|V_0\left(x\right)\right|𝑑x<\mathrm{}$$
then for $`E\alpha `$, $`\text{Im}\sqrt{E}>0`$ the following equality
$$_{\mathrm{}}^{\mathrm{}}\left[G_0(x,x,E)G_1(x,x,E)\right]𝑑x=\frac{\delta }{\kappa ^2+ia\kappa }\frac{\delta }{\kappa ^2+a^2}$$
(23)
holds, where $`E=\kappa ^2`$, $`\alpha =a^2`$; $`\delta =1`$ for the case (i), $`\delta =1`$ for the case (ii) and $`\delta =0`$ for the case (iii).
Proof. The statement readily follows from the fact that any scattering potential has a pair of solutions (Jost solutions, see e.g. ) with the following asymptotics at the right infinity
$$f_{l,r}(x,E)e^{i\kappa x}E=\kappa ^2\text{Im}\kappa >0x\mathrm{}$$
and similar asymptotics at the left infinity and the use of an appropriate part of equalities (18) and (19). The Wronskian $`W_0`$ for Jost solutions can easily be calculated, $`W_0=2i\kappa `$. $`\mathrm{}`$
So, we see that despite the fact that for both $`h_0`$ and $`h_1`$ the continuous spectrum eigenfunctions are normalized to the Dirac delta function, (i.e.) that in both cases they have equal infinite norms, the difference of these infinities is a finite non-zero quantity in cases (i) and (ii) and it is zero in case (iii).
For instance in case (ii) the following equality arises:
$$_{\mathrm{}}^{\mathrm{}}\frac{P\left(k\right)dk}{k^2E}=R\left(E\right)R\left(E\right)=\frac{1}{\kappa ^2+ia\kappa }E=\kappa ^2\alpha =a^2$$
(24)
where
$$P\left(k\right)=_{\mathrm{}}^{\mathrm{}}\left[\left|\psi _k\left(x\right)\right|^2\left|\chi _k\left(x\right)\right|^2\right]𝑑x.$$
(25)
Equation (24) may be reduced to the Stieltjes transform and the function $`P`$ may be found by the Stieltjes inversion formula (see e.g. ). To establish this we first notice that the integral on the left hand side of (24) is different from zero only if $`P\left(k\right)`$ is an even function which we assume to be the case. Therefore it can be considered only for positive $`k`$s and we can let $`k^2=\lambda `$. So, (24) takes the form
$$_{\mathrm{}}^{\mathrm{}}\frac{d\rho \left(\lambda \right)}{\lambda E}=R\left(E\right)$$
where the measure $`\rho \left(\lambda \right)`$ is continuous for $`\lambda >0`$, $`d\rho \left(\lambda \right)=\frac{1}{\sqrt{\lambda }}P\left(\lambda \right)d\lambda `$ and such that for negative $`\lambda `$s the integral is zero. Now the Stieltjes inversion formula yields
$$\frac{P\left(\lambda \right)}{\sqrt{\lambda }}=\frac{\text{sign}\tau }{2\pi i}\underset{\tau 0}{lim}\left[R\left(E\right)R\left(\overline{E}\right)\right]E=\lambda +i\tau $$
where the bar over $`E`$ denotes the complex conjugate to $`E`$. Note that because of the condition $`\text{Im}\sqrt{E}>0`$ the square root of $`E`$ has different signs for $`E`$ in the upper and lower halves of the complex $`E`$-plane. Therefore the function $`R\left(E\right)`$ has a cut along the real axis and the jump across this cut defines the function $`P\left(\lambda \right)`$. After a simple calculation one gets
$$P\left(\lambda \right)=a\pi ^1\left(\lambda ^2+a^2\right)^1.$$
(26)
It must be noted that in the present case the interchange of the integrals over the space variable $`x`$ taken in the difference of (5) and (11) at $`y=x`$ with the integral over the momentum $`k`$ is justified.
###### Corollary 3
If $`h_0`$ is a scattering Hamiltonian with the potential $`V_0`$ for the spectral problem on a half line satisfying the condition
$$_0^{\mathrm{}}x\left|V_0\left(x\right)\right|𝑑x<\mathrm{}$$
then there exist only two kinds of SUSY transformations keeping the zero boundary condition at the origin. If $`h_0`$ has the discrete spectrum its ground state $`\psi _0`$ may be deleted, ($`u=\psi _0`$, case (i)) and there is a possibility to keep the spectrum unchanged (case (iii)). The last possibility may be realized with $`u(x)=f_l(x,E)`$, $`E<0`$ where $`f_l`$ is such that $`f_l(0,E)=0`$. In this case the following trace formula is valid:
$$_0^{\mathrm{}}\left[G_0(x,x,E)G_1(x,x,E)\right]𝑑x=\frac{\delta _2}{2\left(\kappa ^2ia\kappa \right)}\frac{\delta _1}{\kappa ^2+a^2}$$
(27)
where $`E=\kappa ^2`$ and $`\alpha =a^2`$; for the case (i) $`\delta _1=\delta _2=1`$ and for the case (iii) $`\delta _1=0`$, $`\delta _2=1`$.
Proof.. The proof is based on the fact that for such potentials the left solution goes to zero like $`x`$ when $`x0`$ and the right solution has the asymptotics $`f_{r0}\mathrm{exp}\left(i\kappa x\right)`$, $`E=\kappa ^2`$, $`\text{Im}k>0`$ (see e.g. ). It follows from here that when $`E`$ is not a spectral point the left solution has a growing asymptotics at infinity, $`f_{0l}\frac{W_0}{2i\kappa }\mathrm{exp}\left(i\kappa x\right)`$, $`x\mathrm{}`$, where $`W_0`$ is the Wronskian of $`f_{r0}`$ and $`f_{l0}`$. It is not possible to create a new bound state in this case since the Schrödinger equation with such a potential has no solutions going to infinity as $`x`$ approaches the origin. $`\mathrm{}`$
Example 1. Free motion on the line, $`V_0\left(x\right)=0`$, $`x`$
The Green function is
$$G_0(x,y,E)=\frac{i}{2\kappa }e^{i\kappa \left|xy\right|}\text{Im}\kappa >0E=\kappa ^2.$$
The choice $`u=\mathrm{cosh}\left(ax\right)`$, $`\alpha =a^2`$ leads to the one soliton potential $`V_1=2a^2\text{sech}^2\left(ax\right)`$ with the Green function
$$G_1(x,y,E)=if_\kappa \left(x\right)f_\kappa \left(y\right)/\left[2\kappa \left(\kappa ^2+a^2\right)\right]xy$$
where $`f_\kappa \left(x\right)=\mathrm{exp}\left(i\kappa x\right)\left(i\kappa +a\mathrm{tanh}ax\right)`$, which clearly has a pole at the ground state energy $`E=a^2`$. The residue at the pole is $`\phi _0\left(x\right)\phi _0\left(y\right)`$ where $`\phi _0\left(x\right)=\sqrt{a/2}\text{sech}\left(ax\right)`$ which is just the ground state of the one-soliton potential.
Continuous spectrum eigenfunctions of $`h_0`$, $`\psi _k\left(x\right)=1/\sqrt{2\pi }\mathrm{exp}\left(ikx\right)`$ are transformed into continuous spectrum eigenfunctions for $`h_1`$, $`\xi _k\left(x\right)=\left[ik+\mathrm{tanh}\left(ax\right)\right]\mathrm{exp}\left(ikx\right)/\sqrt{2\pi \left(k^2+a^2\right)}`$. The direct calculation of the function $`P\left(k\right)`$ given in (25) gives exactly the result (26).
Example 2. Free motion on a half-line with zero angular momentum, $`V_0\left(x\right)=0`$, $`x^+`$.
The Green function is
$$G_0(x,y,E)=\frac{1}{\kappa }\mathrm{sin}\left(\kappa x\right)\mathrm{exp}\left(i\kappa y\right)E=\kappa ^2\text{Im}\kappa >0xy.$$
It must be noted that the use of the transformation function $`u=\mathrm{cosh}\left(ax\right)`$ gives the same one-soliton potential as in the previous example which when considered on a half-line has only a continuous spectrum but its continuous spectrum eigenfunctions cannot be obtained by applying the transformation operator $`L`$ to the free particle eigenfunctions since it does not preserve the zero boundary condition at the origin. Another choice $`u=\mathrm{sinh}\left(ax\right)`$ creates the potential $`V_1=2a^2\text{csch}^2\left(ax\right)`$ which is singular at the origin and has only a continuous spectrum and which may be obtained with the help of the SUSY transformation from the free particle Hamiltonian. Its Green function is
$$G_1(x,y,E)=\frac{ie^{i\kappa y}}{\kappa \left(\kappa ^2+a^2\right)}\left[\kappa +ia\text{cth}\left(ay\right)\right]\left[\kappa \mathrm{cos}\left(\kappa x\right)a\text{cth}\left(ax\right)\mathrm{sin}\left(\kappa x\right)\right]$$
$$E=\kappa ^2\text{Im}\kappa >0.$$
It is not difficult to see that despite the presence of the denominator, $`G_1`$ is regular for all $`\kappa 0`$ including the point $`\kappa =ia`$ and discontinuous along the positive real axis in the complex $`E`$-plane. The jump across this cut is proportional to the product of continuous spectrum eigenfunctions of $`h_1`$ which are given by
$$\xi _k=\sqrt{\frac{2}{\pi \left(k^2+a^2\right)}}\left[k\mathrm{cos}\left(kx\right)+a\text{cth}\left(ax\right)\mathrm{sin}\left(kx\right)\right].$$
In contrast to the previous example the difference $`\psi _k^2\xi _k^2`$ now oscillates when $`x\mathrm{}`$, the Riemann integral of this difference over the space variable is divergent and the interchange of the integrals over $`k`$ and $`x`$ is impossible. Nevertheless, if one assumes that the improper integral over $`x`$ is a limit of a proper integral one can write
$$\underset{A\mathrm{}}{lim}_0^{\mathrm{}}\frac{P(k,A)dk}{k^2E}=R\left(E\right)R\left(E\right)=\frac{1}{2\left(ia\kappa \kappa ^2\right)}$$
(28)
$`\text{where }E=\kappa ^2,\alpha =a^2\text{ and }P(k,A)={\displaystyle _0^A}\left[\psi _k^2\left(x\right)\chi _k^2\left(x\right)\right]𝑑x.`$
In our example the function $`P(k,A)`$ is given by
$$P(K,A)=\frac{2a\text{cth}\left(aA\right)\mathrm{sin}^2\left(kA\right)k\mathrm{sin}\left(2Ak\right)}{\pi \left(k^2+a^2\right)}$$
which when substituted into the left hand side of (28) gives exactly the function $`R\left(E\right)`$. So this example shows that in contrast to the previous example where the difference of normalisations is a finite quantity, here this value is undetermined. Nevertheless, the contribution to the trace of the difference $`G_0G_1`$ from the continuous spectra of $`h_0`$ and $`h_1`$ is well defined.
## 5 Conclusions
In this paper we have studied the relation between the Green functions corresponding to two Hamiltonians which are SUSY partners. We have shown that it is possible to establish a relation between the traces of the Green functions for the two partner Hamiltonians for the cases of deletion of the ground state, the addition of a new ground state and when the two Hamiltonians are isospectral. The formulae derived in this paper are valid for the general case of Hamiltonians having both discrete and continuous spectra. Our results show that when a continuous spectrum is present, each of the traces of the Green functions for the SUSY partners may diverge but the difference between the traces can be finite. We have illustrated our results by considering the case of the free motion on the full line and the case of the free motion of a particle with zero angular momentum on the half-line.
Finally we would like to note that the difference of the traces of the Green functions of the two SUSY partner Hamiltonians appears as the trace (actually super-trace) of the Green function of the supersymmetric Schrödinger equation (supersymmetric Green function). Thus, our results reveal the possibility of divergence of the component traces of the supersymmetric Green function while its super-trace remains finite.
## Acknowledgments
The work of BFS is partially supported by the President Grant of Russia 1743.2003.2 and the Spanish MCYT and European FEDER grant BFM2002-03773.
## References |
warning/0507/cond-mat0507190.html | ar5iv | text | # Quasi One-Dimensional Bosons in Three-dimensional Traps: From Strong Coupling to Weak Coupling Regime
\[
## Abstract
We analyze a recent experiment on a Tonks-Girardeau gas of <sup>87</sup>Rb atoms (T. Kinoshita, T. Wenger, and D.S. Weiss, Science 305, 1125 (2004)). We find that the experimental data are compatible with the one-dimensional theory of Lieb, Seiringer and Yngvason (Phys. Rev. Lett. 91, 150401 (2003)) but are better described by a theory that takes into account variations in the transverse width of the atomic cloud. By using this theory we investigate also the free axial expansion of the <sup>87</sup>Rb gas in different regimes: Tonks-Girardeau gas, one-dimensional Bose-Einstein condensate and three-dimensional Bose-Einstein condensate.
\]
Two experimental groups have reported the observation of the one-dimensional (1D) Tonks-Girardeau (TG) gas with ultracold <sup>87</sup>Rb atoms in highly elongated traps. A rigorous theoretical analysis of the ground-state properties of a uniform 1D Bose gas, including the beyond-mean-field TG regime of impenetrable bosons, was performed by Lieb and Liniger (LL) forty years ago . Recently, motivated by the experimental achievements, an extension of the LL theory for finite and inhomogeneous 1D Bose gases under longitudinal confinement has been proposed on the basis the local density approximation (LDA) . Öhberg and Santos have suggested that the LDA is improved by including a gradient term that represent additional kinetic energy associated with the inhomogeneity of the gas. This conjecture has been rigorously proved by Lieb, Seiringer and Yngvason . More recently we have introduced a variational approach, called generalized Lieb-Liniger theory (GLLT), which reduces to the Lieb-Seiringer-Yngvason theory (LSYT) in the 1D regime and, in addition, gives an accurate description of the crossover from the 1D regime to 3D regime.
In this Brief Report we apply the GLLT to analyze the experimental results of Kinoshita, Wenger and Weiss . Contrary to the data of Paredes et al. which are deeply in the TG regime, the data of Ref. cover different quantum-dimensional regimes and we show that the experimental atomic cloud is better described by the GLLT than the LSYT. The GLLT is then used to determine the axial free expansion of the <sup>87</sup>Rb gas. We predict a self-similar expansion whose growth is strictly related to the quantum-dimensional regime of the initial configuration.
In the LSYT the longitudinal density $`\rho (z)`$ of a zero-temperature Bose gas is obtained by minimizing the following energy functional
$$E_{LSY}[\rho ]=_{\mathrm{}}^+\mathrm{}\left\{\frac{\mathrm{}^2}{2m}\left[(_z\sqrt{\rho })^2+\rho ^3e(\frac{g}{\rho })\right]+V\rho \right\}𝑑z,$$
(1)
where $`g=2a_s/a_{}^2`$ is the interaction parameter, with $`a_s`$ the s-wave scattering length and $`a_{}=(\mathrm{}/(m\omega _{})^{1/2}`$ the characteristic length of the transverse harmonic potential with frequency $`\omega _{}`$, $`V(z)`$ is the longitudinal external potential and $`e(x)`$ is the Lieb-Liniger function, which is defined as the solution of a Fredholm equation and it is such that $`e(x)x4/(3\pi )x^{3/2}`$ for $`x1`$ and $`e(x)(\pi ^2/3)(x/(x+2))^2`$ for $`x1`$ .
As previously stressed, the LSYT is valid in the pure 1D regime, where the transverse width $`R_{}`$ of the Bose gas is frozen and equal to the harmonic length $`a_{}`$. In our GLLT the transverse properties of the Bose gas are taken into account by considering the adimensional width $`\sigma =R_{}/a_{}`$ as a variational parameter, i.e. $`\sigma =\sigma (z)`$. The theory gives the following energy functional
$$E_{GLL}[\rho ,\sigma ]=_{\mathrm{}}^+\mathrm{}\{\frac{\mathrm{}^2}{2m}[(_z\sqrt{\rho })^2+\rho ^3e(\frac{g}{\rho \sigma ^2})]$$
$$+V\rho +\frac{\mathrm{}\omega _{}}{2}(\frac{1}{\sigma ^2}+\sigma ^22)\rho \}dz,$$
(2)
where $`\mathrm{}\omega _{}(\sigma ^2+\sigma ^2)\rho /2`$ is the transverse energy density of the Bose gas. Note that in the low density region where $`\sigma 1`$ (1D regime) the functional $`E_{GLL}`$ reduces to $`E_{LSY}`$. Instead, at higher densities (3D regime) where $`e(g/(\rho \sigma ^2))g/(\rho \sigma ^2)`$, the functional $`E_{GLL}`$ gives the energy functional of the nonpolynomial Schrödinger equation (NPSE) , an effective 1D differential equation we have derived from the 3D Gross-Pitaevskii equation to describe Bose-Einstein condensates (BECs) under transverse harmonic confinement.
Taking into account the normalization condition of the longitudinal density $`\rho (z)`$ to the total number $`N`$ of atoms, the minimization of the energy functional $`E_{GLL}`$ with respect to $`\rho (z)`$ gives the equation
$$\frac{\mathrm{}^2}{2m}\left[\sqrt{\rho }_z^2\sqrt{\rho }+3\rho ^3e(\frac{g}{\rho \sigma ^2})\frac{g\rho ^2}{\sigma ^2}e^{}(\frac{g}{\rho \sigma ^2})\right]$$
$$+V\rho +\frac{\mathrm{}\omega _{}}{2}(\frac{1}{\sigma ^2}+\sigma ^22)\rho =\overline{\mu }\rho ,$$
(3)
where $`\overline{\mu }`$ is the chemical potential fixed by the normalization condition. The minimization of the energy functional $`E_{GLL}`$ with respect to $`\sigma (z)`$ gives instead the equation
$$\sigma ^4=1+a_{}^2g\rho e^{}(\frac{g}{\rho \sigma ^2}).$$
(4)
This implicit equation must be solved numerically but analytical results can be obtained in limiting cases. In particular, under the condition $`a_sa_{}`$ , one gets $`\sigma \sqrt{a_{}}(\rho g)^{1/4}`$ for $`\rho 1/a_s`$ (3D BEC regime) and $`\sigma 1`$ for $`\rho 1/a_s`$ (1D regime). Note that the 1D regime contains two subregimes: the 1D BEC regime for $`a_s/a_{}^2\rho 1/a_s`$, and the TG regime, where $`e(g/(\rho \sigma ^2))\pi ^2/3`$, for $`\rho a_s/a_{}^2`$.
In the experiment of Kinoshita, Wenger and Weiss an ensemble of about 6400 parallel 1D traps has been created by means of a 2D optical lattice, that strongly confines atoms in 1D tubes. The traps differ only in the number of <sup>87</sup>Rb atoms each contains. As discussed in , each trap can be modelled by an anisotropic harmonic potential, where the axial frequency is $`\omega _z=2\pi \times 27.5`$ Hz and the transverse frequency $`\omega _{}`$ is tuned by changing the energy depth $`U_0`$ of the confining optical lattice. The maximum transverse frequency is $`2\pi \times 70.7`$ kHz; it follows that the maximum trap anisotropy is $`\lambda _{max}=2570`$, where $`\lambda =\omega _{}/\omega _z`$. Kinoshita et al. have worked with $`r=\lambda /\lambda _{max}`$ ranging between $`r=1`$ and $`r=10^1`$, a regime where the Bose gas is quasi-1D.
By adding an imaginary time term at right side of Eq. (3) and solving Eq. (3) and Eq. (4) with a finite-difference Crank-Nicolson predictor-corrector scheme , we calculate the density profile $`\rho (z)`$ of the Bose gas using the trap parameters of Ref. . In Eq. (3) we set $`V(z)=m\omega _z^2z^2/2`$. The results are shown in Fig. 1, where we compare LSYT with GLLT, plotting also their Thomas-Fermi (TF) approximations which neglect the gradient term in Eq. (1) and Eq. (2). Figure 1 shows that, for the trap anisotropy $`r=1`$, the TF approximation is remarkably accurate but there are instead quantitative differences between LSYT and GLLT due to the fact that $`\sigma (\rho (z))`$ is not strictly equal to one. The differences between LSYT and GLLT increase by reducing the anisotropy $`r`$ of the trap, and also the TF approximation worsens. In fact, by reducing $`r`$ one induces a crossover from the 1D regime to the 3D regime, whose description cannot be accounted for by LSYT that is purely 1D . We have verified that the differences between LSYT and GLLT are instead reduced by taking larger values of the anisotropy ratio $`r`$.
In Fig. 2 we plot the experimental data of the root mean square (RMS) length of the atomic cloud obtained by Kinoshita, Wenger and Weiss . This RMS axial length has been derived as the average over several clouds which have different numbers of atoms. In Fig. 2 the RMS full length is plotted versus $`U_0/E_{rec}=87r^2`$, where $`U_0`$ is the energy depth of the optical potential and $`E_{rec}`$ is the recoil energy of the gas. Kinoshita et al. have compared their data with LSYT in the TF approximation . By performing the averaging procedure of Ref. on our LSYT results, we obtain the same theoretical curve shown in Fig. 4 of Ref. and plot it (dashed line) in our Fig. 2. With the same averaging procedure we obtain also the GLLT curve (dotted line). Figure 2 shows that our GLLT fits better than LSYT the experimental data. As discussed in , the experimental data for $`U_0/E_{rec}<20`$ are strongly affected by the tunneling between adjacent tubes and the theoretical analysis based on a single tube overestimates the axial width of the Bose cloud.
Our GLLT is a simple and useful tool to investigate also dynamical properties of a Bose gas under transverse harmonic confinement. The dynamics of the Bose gas can be described by means of a complex classical field $`\mathrm{\Phi }(z,t)`$ that satisfies the equation $`i\mathrm{}_t\mathrm{\Phi }=\frac{\delta }{\delta \mathrm{\Phi }^{}}E_{GLL}[|\mathrm{\Phi }|^2,\sigma ]`$, where $`|\mathrm{\Phi }(z,t)|^2=\rho (z,t)`$. This time-dependent nonlinear Schrödinger equation can be explicitly written as
$$i\mathrm{}_t\mathrm{\Phi }=\left(\frac{\mathrm{}^2}{2m}_z^2+V+\mu [|\mathrm{\Phi }|^2]\right)\mathrm{\Phi },$$
(5)
where $`\mu [\rho ]`$ is the bulk chemical potential, given by
$$\mu [\rho ]=\frac{\mathrm{}^2}{2m}\left[3\rho ^2e(\frac{g}{\rho \sigma ^2})\frac{g\rho }{\sigma ^2}e^{}(\frac{g}{\rho \sigma ^2})\right]+\frac{\mathrm{}\omega _{}}{2}(\frac{1}{\sigma ^2}+\sigma ^22).$$
(6)
The Eq. (5) with Eq. (6) must be solved self-consistently with Eq. (4). It is important to observe that for $`\rho a_s/a_{}^2`$ (TG regime) one has $`\sigma 1`$ and $`\mu \mathrm{}^2\pi ^2\rho ^2/(2m)`$ and Eq. (5) reduces the the time-dependent nonlinear Schrödinger equation (NLSE) introduced by Kolomieski et al. . For $`\rho 1/a_s`$ (1D regime or TG regime) still we have $`\sigma 1`$ but $`\mu 3\mathrm{}^2\rho ^2e(g/\rho )/(2m)\mathrm{}^2g\rho e^{}(g/\rho )/(2m)`$ and Eq. (5) becomes equal to the time-dependent NLSE proposed by Öhberg and Santos , which reduces to that of Kolomieski et al. in the TG regime. For $`\rho a_s/a_{}^2`$ (1D regime or 3D regime) the Eq. (5) gives $`\sigma (1+ga_{}^2\rho )^{1/4}`$, the Eq. (6) reads $`\mu \mathrm{}^2g\rho /(m\sigma ^2)+\mathrm{}\omega _{}(\sigma ^2+\sigma ^22)/2`$ and Eq. (5) becomes the time-dependent NPSE . The NPSE coincides with the equation of Öhberg and Santos in the 1D BEC regime, where $`a_s/a_{}^2\rho 1/a_s`$, but it accurately describes also the 3D BEC regime, where $`\rho 1/a_s`$ and $`\sigma \sqrt{a_{}}(g\rho )^{1/4}`$.
The time-dependent (TD) GLLT given by Eq. (5) must be carefully employed. In fact, Girardeau and Wright have shown, studying the interference of two Bosonic clouds, that the time-dependent NPSE of Kolomieski et al. overestimates the coherence of the Tonks-Girardeau gas. However, the TD GLLT can be safely used to calculate collective properties of the Bose gas. In Ref. we have shown that the time-dependent GLLT gives the expected values for the axial breathing mode $`\mathrm{\Omega }_z`$ of the Bose cloud in a harmonic confinement with $`V(z)=m\omega _z^2z^2/2`$. In particular, we have found $`\mathrm{\Omega }_z=2\mathrm{\Omega }_z`$ in the TG regime, $`\mathrm{\Omega }_z=\sqrt{3}\omega _z`$ in the 1D BEC regime, and $`\mathrm{\Omega }_z=\sqrt{5/2}\omega _z`$ in the 3D BEC regime.
Here we analyze the free axial expansion of the Bose gas by means of the time-dependent GLLT. We consider the expansion of the Bose gas when the axial harmonic confinement is removed ($`\omega _z=0`$) while the radial one is kept fixed. By setting $`\mathrm{\Phi }(z,t)=\sqrt{\rho (z,t)}\mathrm{exp}(iS(z,t)/\mathrm{})`$, the Eq. (5) is equivalent to the two hydrodynamics equations $`_t\rho +_z\left(\rho v\right)=0`$ and $`m_tv+_z[\mathrm{}^2/(2m\sqrt{\rho })_z^2\sqrt{\rho }+mv^2/2+V+\mu ]=0`$ of a 1D viscousless fluid with density field $`\rho (z,t)`$ and velocity field $`v(z,t)=(\mathrm{}/m)_zS(z,t)`$. If the initial axial width of the cloud is larger than the healing length $`\mathrm{}/\sqrt{2m\mu }`$ then one can safely neglect the quantum pressure (QP) term $`\mathrm{}^2/(2m\sqrt{\rho })_z^2\sqrt{\rho }`$ in the second hydrodynamics equation. Note that the bulk chemical potential $`\mu `$ of Eq. (6) scales as $`\rho ^\gamma `$ in the three relevant regimes: $`\gamma =2`$ in the TG regime, $`\gamma =1`$ in the 1D BEC regime, and $`\gamma =1/2`$ in the 3D BEC regime. It is straightforward then to prove that in these regimes the cloud density decreases during the time evolution following the self-similar solution $`\rho (z,t)=\rho (z/b(t),t=0)/b(t)`$, where the adimensional axial width $`b(t)`$ satisfies the equation
$$\ddot{b}=\frac{\omega _z^2}{b^{\gamma +1}}.$$
(7)
The solution of this equation with initial conditions $`b(0)=1`$ and $`\dot{b}(0)=0`$ can be obtained by quadratures. For large $`t`$ the solution is $`b(t)=\sqrt{2/\gamma }\omega _zt`$. In general, $`\mu (\rho )`$ is not a power law and during the expansion the functional dependence of $`\mu (\rho )`$ changes throughout the whole cloud: the expansion is no more truly self-similar. However, by evaluating the dynamics at the center of the cloud ($`z=0`$) where the initial density is $`\rho _0`$, from the two hydrodynamics equations without the QP term one finds
$$\frac{1}{2}\dot{b}^2=\frac{\omega _z^2\left(\mu (\rho _0)\mu (\rho _0/b)\right)}{\rho _0\frac{\mu }{\rho }(\rho _0)}.$$
(8)
For large $`t`$ the solution of this equation is $`b(t)=\sqrt{2/\overline{\gamma }}\omega _zt`$, where naturally appears the effective polytropic index
$$\overline{\gamma }=\frac{\rho _0}{\mu (\rho _0)}\frac{\mu }{\rho }(\rho _0),$$
(9)
that is the logarithmic derivative of the chemical potential $`\mu `$. In order to verify these analytical results we simulate the free axial expansion of the Bose gas by numerically solving Eq. (5) self-consistently with Eqs. (4,6). In our simulation we employ a real-time Crank-Nicholson method .
In Fig. 3 we plot the time evolution of the normalized axial width $`W(t)`$ of the Bose gas with $`N=50`$ atoms and initial anisotropy of the harmonic trap given by $`r=1`$ ($`\omega _{}/\omega _z=2570`$). After a transient the width $`W(t)`$ grows linearly in time, as clearly shown by the lower panel of Fig. 3, where we plot the width velocity $`dW/dt`$. In Fig. 3 we plot also $`W(t)`$ and $`dW/dt`$ calculated with two alternative methods: by using Eq. (8) and by using Eq. (7) with $`\gamma `$ given by Eq. (9), in both cases taking into account Eqs. (4,6).
Figure 3 shows that the approximation based on the polytropic index is reliable; note that in all our tests the relative error in the slope of the asymptotic linear growth is always within $`5\%`$.
In Fig. 4 we plot the slope $`\alpha =\sqrt{2/\overline{\gamma }}`$ as a function of the initial density $`\rho _0`$ at the center of the cloud. $`\alpha `$ smoothly changes from $`1`$ to $`\sqrt{2}`$ and than to $`2`$ in the transition from the TG regime to the 1D BEC regime and then to the 3D BEC regime. Figure 4 shows that for $`r1`$ in addition to the two plateaus corresponding to $`\alpha =2`$ at high densities (3D BEC regime) and to $`\alpha =1`$ at low densities (TG regime), an additional plateau appears with $`\alpha =\sqrt{2}`$ at intermediate densities (1D BEC regime).
In conclusion, we have analyzed a recent experiment on dilute Bosons in the Tonks-Girardeau regime by using a beyond-mean-field theory that takes into account transverse variations of the atomic cloud. Remarkably, our theory shows a better agreement with the experimental results than the simple 1D approach proposed by various authors in the past years. On the basis of this theory, we have predicted a quasi self-similar expansion of the gas when the axial confinement is removed, showing that asymptotic linear growth of the axial width strongly depends on the quantum-dimensional regime of the Bosonic cloud.
L.S. thanks D.S. Weiss for enlightening e-suggestions. |
warning/0507/math0507609.html | ar5iv | text | # Weyl-Heisenberg Frame Wavelets with Basic Supports
## 1 Introduction
Let $``$ be a separable complex Hilbert space. Let $`B()`$ denote the algebra of all bounded linear operators on $``$. Let $``$ denote the set of natural numbers, and $``$ be the set of all integers. A collection of elements$`\{x_j:jȷ\}`$ in $``$ is called a frame (of $``$) if there exist constants $`A`$ and $`B`$, $`0<AB<\mathrm{}`$, such that
$$Af^2\underset{jȷ}{}|f,x_j|^2Bf^2$$
(1.1)
for all $`f`$. The supremum of all such numbers $`A`$ and the infimum of all such numbers $`B`$ are called the frame bounds of the frame and are denoted by $`A_0`$ and $`B_0`$ respectively. A frame is called a tight frame if $`A_0=B_0`$ and is called a normalized tight frame if $`A_0=B_0=1`$. Any orthonormal basis in a Hilbert space is a normalized tight frame. However a normalized tight frame is not necessarily an orthonormal basis. Frames can be regarded as the generalizations of orthogonal bases of Hilbert spaces. The concept of frames was introduced a long time ago () and have received much attention recently due to the development and study of wavelet theory . Among those widely studied lately are the Weyl-Heisenberg frames (also called Gabor frames). Let $`a`$, $`b`$ be two fixed positive constants and let $`T_a`$, $`M_b`$ be the translation operator by $`a`$ and modulation operator by $`b`$ respectively, i.e., $`T_af(t)=f(ta)`$ and $`M_bf(t)=e^{ibt}f(t)`$ for any $`fL^2()`$. For a fixed $`gL^2()`$, we say that $`(g,a,b)`$ generates a Weyl-Heisenberg frame if $`\{M_{mb}T_{na}g\}_{m,n}`$ forms a frame for $`L^2()`$. We also say that the function $`g`$ is a mother Weyl-Heisenberg frame wavelet for $`(a,b)`$ in this case. Furthermore, a measurable set $`E`$ is called a Weyl-Heisenberg frame set for $`(a,b)`$ if the function $`g=\chi _E`$ generates a Weyl-Heisenberg frame for $`L^2()`$ under modulates by $`b`$ and translates by $`a`$, i.e., $`\{e^{imbt}g(tna)\}_{m,n}`$ is a frame for $`L^2()`$. Characterizing the mother Weyl-Heisenberg frame wavelets in general is a difficult and open question. In fact, even in the special case of $`a=2\pi `$ and $`b=1`$, this question remains unsolved. There are many works related to this subject, for more information please refer to .
In 2002, Casazza and Kalton proved the following theorem :
###### Theorem 1.1.
For fixed integers $`n_1<n_2<\mathrm{}<n_k`$, the set
$$E=_{j=1}^k\left([0,2\pi )+2\pi n_j\right)$$
(1.2)
is a Weyl-Heisenberg frame set for $`(2\pi ,1)`$ if and only if the polynomial $`p(z)=_{j=1}^kz^{n_j}`$ has no unit roots.
This means that characterizing a Weyl-Heisenberg frame set $`E`$ with the special form (1.2) is equivalent to the following problem, which was proposed by Littlewood in 1968 :
###### Problem 1.2.
Classify the integer sets $`\{n_1<n_2<\mathrm{}<n_k\}`$ such that the polynomial $`p(z)=_{1jk}z^{n_j}`$ does not have any unit roots.
While it is unfortunate that Theorem 1.1 fails to provide a definite answer to the characterization problem of Weyl-Heisenberg frame sets since Problem 1.2 is also an open question (see and the references therein), it does reveal a deep connection between the two seemingly irrelevant subjects. In this paper, we will further investigate Weyl-Heisenberg frames defined by set theoretical functions and continuous functions with restricted domain. In Section 2, we will introduce some concepts, definitions, preliminary lemmas. In Section 3, we will state and prove our main theorems. Some examples are given in Section 4.
## 2 Definitions and Preliminary Lemmas
###### Definition 2.1.
The Zak transform of a function $`fL^2()`$ is defined as
$$Zf(t,w)=\frac{1}{\sqrt{2\pi }}\underset{n}{}f(t+2\pi n)e^{inw},t,w[0,2\pi ).$$
(2.1)
The above definition of Zak transform is slightly different from that given in . We have the following lemma.
###### Lemma 2.2.
The Zak transform is a unitary map from $`L^2()`$ onto $`L^2(Q)`$, where $`Q=[0,2\pi )\times [0,2\pi )`$.
One can easily check that the Zak transform maps
###### Lemma 2.3.
$$Z(M_mT_{2n\pi }g)(t,w)=e^{i(mt+nw)}Zg(t,w),gL^2().$$
(2.2)
###### Proof.
Since $`M_mT_{2n\pi }g(x)=e^{imx}g(x2n\pi )`$,
$`Z(M_mT_{2n\pi }g)(t,w)`$
$`=`$ $`{\displaystyle \frac{1}{\sqrt{2\pi }}}{\displaystyle \underset{\mathrm{}}{}}e^{im(t+2\mathrm{}\pi )}g\left(t+2(\mathrm{}n)\pi \right)e^{i\mathrm{}w}`$
$`=`$ $`{\displaystyle \frac{1}{\sqrt{2\pi }}}{\displaystyle \underset{\mathrm{}}{}}e^{imt}g\left(t+2(\mathrm{}n)\pi \right)e^{i\mathrm{}w}`$
$`=`$ $`{\displaystyle \frac{1}{\sqrt{2\pi }}}{\displaystyle \underset{\mathrm{}}{}}e^{i(mt+nw)}g\left(t+2(\mathrm{}n)\pi \right)e^{i\mathrm{}winw}`$
$`=`$ $`{\displaystyle \frac{1}{\sqrt{2\pi }}}e^{i(mt+nw)}{\displaystyle \underset{\mathrm{}}{}}g\left(t+2(\mathrm{}n)\pi \right)e^{i(\mathrm{}n)w}`$
$`=`$ $`{\displaystyle \frac{1}{\sqrt{2\pi }}}e^{i(mt+nw)}{\displaystyle \underset{k}{}}g(t+2k\pi )e^{ikw}`$
$`=`$ $`e^{i(mt+nw)}{\displaystyle \frac{1}{\sqrt{2\pi }}}{\displaystyle \underset{k}{}}g(t+2k\pi )e^{ikw}`$
$`=`$ $`e^{i(mt+nw)}(Zg)(t,w).`$
###### Lemma 2.4.
Let $`E`$ be a measurable subset of $`[0,2\pi )`$, $`F=_n(E+2n\pi )`$ and $`gL^2(F)`$. The following statements are equivalent:
(1) $`(M_mT_{2n\pi }g)_{m,n}`$ is a frame for $`L^2(F)`$ with frame bounds $`A_0`$, $`B_0`$.
(2) $`0<A_0=essinf_{(t,w)E\times [0,2\pi )}|Zg(t,w)|^2esssup_{(t,w)E\times [0,2\pi )}|Zg(t,w)|^2=B_0<\mathrm{}.`$
Let $`E=_{k=1}^mA_k`$, where each $`A_k=[a_k,b_k]`$ is a closed interval and $`A_iA_j=\mathrm{}`$ if $`ij`$. Let $`\tau _{2\pi }(x):[0,2\pi )`$ be the function defined by $`\tau _{2\pi }(x)=x2\pi [\frac{x}{2\pi }]`$, where $`[.]`$ is the integer function. Let us arrange the numbers
$$0,\tau _{2\pi }(a_1),\tau _{2\pi }(b_1),\mathrm{},\tau _{2\pi }(a_m),\tau _{2\pi }(b_m),2\pi $$
in the ascending order and write the resulting numbers as $`0=t_0<t_1<\mathrm{}<t_{j_0}=2\pi `$. Then for each $`0jj_01`$ and each $`k`$, we have either $`(t_j,t_{j+1})\tau _{2\pi }(A_k)`$ or $`(t_j,t_{j+1})\tau _{2\pi }(A_k)=\mathrm{}`$. Based on this observation we have:
###### Lemma 2.5.
Let $`\{A_i\}_{i=1}^m`$ be a sequence of finite and non-overlapping intervals and $`E=_{i=1}^mA_i`$, then there exists a finite sequence of disjoint intervals $`\{E_i\}_{i=1}^k`$ with $`E_i[0,2\pi )`$, and an integer sequence $`\{n_{ij}\}_{j=1}^{j_i}`$ for each $`i`$, such that
$$E=\underset{i=1}{\overset{k}{}}F_i,\mathrm{where}\mathrm{F}_\mathrm{i}=\underset{\mathrm{j}=1}{\overset{\mathrm{j}_\mathrm{i}}{}}(\mathrm{E}_\mathrm{i}+2\pi \mathrm{n}_{\mathrm{ij}}).$$
We will call the set $`E`$ defined in the above lemma a basic support set, which is just a finite disjoint union of finite intervals. The sequence $`\{E_i\}_{i=1}^k`$ associated with the set $`E`$ will be called the 2$`\pi `$-translation generators of $`E`$. Notice that $`_{j=1}^{j_i}(E_i+2\pi n_{ij})`$ is simply the pre-image of the function $`\tau _{2\pi }`$ restricted to $`E`$. We will call the sequence $`\{n_{ij}\}_{j=1}^{j_i}`$ the step-widths of the corresponding generator $`E_i`$.
###### Definition 2.6.
Let $`gL^2()`$ be a continuous function and $`E`$ be a basic support set. For each $`\xi E`$, the sequence $`\{g(\xi +2\pi n_{ij})\}_{j=1}^{j_i}`$ is called a characteristic chain of the function $`g`$ associated with the set $`E`$, where $`\xi E_i`$ and $`E_i`$, $`\{n_{ij}\}`$ are as defined in Lemma 2.5.
If $`_{j=0}^ka_{n_j}z^{n_j}=0`$ has zeros on the unit circle $`𝕋`$, then the coefficient sequence $`\{a_{n_j}\}_{j=0}^k`$ is called a root sequence of $`\{n_j\}`$. For instance, $`\{2,1,1\}`$ is a root sequence for $`\{0,1,2\}`$ since the polynomial $`2z+z^2`$ has a unit root. But $`\{2,1,1\}`$ is not a root sequence for $`\{0,1,3\}`$ since the polynomial $`2z+z^3`$ does not have a unit root. Furthermore, the sequence $`\{n_j\}`$ may contain negative integers as well, in which case $`_{j=0}^ka_{n_j}z^{n_j}`$ is a Laurent polynomial, but our results will still hold for such a polynomial.
## 3 Main Results and their Proofs
We first outline the main results obtained in this paper below. The first theorem generalizes the result of to step like functions.
###### Theorem 3.1.
Let $`n_0<n_1<n_2<\mathrm{}<n_k`$ be $`k+1`$ fixed integers, $`a_0`$, $`a_1`$, $`a_2`$, $`\mathrm{}`$, $`a_k`$ be $`k+1`$ given complex numbers and let $`F_j=[0,2\pi )+2\pi n_j`$ ($`j=0`$, $`1`$, $`2`$, $`\mathrm{}`$, $`k`$), $`g=_{j=0}^ka_j\chi _{F_j}`$. The following statements are equivalent:
(1) $`g`$ is a mother Weyl-Heisenberg frame wavelet for $`(2\pi ,1)`$ with frame bounds $`A_0`$, $`B_0`$.
(2) $`2\pi A_0=inf_{|z|=1}|_{j=0}^ka_jz^{n_j}|^2`$ and $`2\pi B_0=sup_{|z|=1}|_{j=0}^ka_jz^{n_j}|^2`$.
(3) For every measurable set $`E[0,2\pi )`$ of positive measure, let $`E_j=E+2n_j\pi `$, $`j=0`$, $`1`$, $`2`$, …, $`k`$, the function $`g_E=_{j=0}^ka_j\chi _{E_j}`$ is a mother Weyl-Heisenberg frame wavelet of $`L^2(\mathrm{\Omega })`$ for $`(2\pi ,1)`$ with frame bounds $`AA_0`$, $`BB_0`$, where $`\mathrm{\Omega }=_n(E+2n\pi )`$ and $`A_0`$, $`B=B_0`$ are the frame bounds for $`g_E`$ with $`E=[0,2\pi )`$.
###### Theorem 3.2.
Let $`E[0,2\pi )`$ be a measurable set of positive measure and $`n_0<n_1<n_2<\mathrm{}<n_k`$ be $`k+1`$ given integers. Let $`F=_{j=0}^k(E+2\pi n_j)`$, $`\mathrm{\Omega }=_n(E+2\pi n)`$. Then for any continuous function $`gL^2()`$, $`(g\chi _F,2\pi ,1)`$ generates a Weyl-Heisenberg frame of $`L^2(\mathrm{\Omega })`$ if and only if for any $`\xi \overline{E}`$, $`\{g(\xi +2\pi n_j)\}`$ is not a root sequence for $`\{n_j\}`$ where $`\overline{E}`$ is the closure of $`E`$. In particular, if $`E=[0,2\pi )`$, then $`(g\chi _F,2\pi ,1)`$ generates a Weyl-Heisenberg frame of $`L^2()`$ if and only if for any $`\xi [0,2\pi ]`$, $`\{g(\xi +2\pi n_j)\}`$ is not a root sequence for $`\{n_j\}`$.
Theorem 3.2 further generalizes Theorem 3.1 to continuous functions.
###### Theorem 3.3.
Let $`E`$ be a basic support set with generator and step-width pairs $`\{(E_i,\{n_{ij}\}_{j=0}^{j_i})\}_{i=0}^k`$, $`\mathrm{\Omega }=_n(E+2n\pi )`$ and $`gL^2()`$ be a continuous function, then $`(g\chi _E,2\pi ,1)`$ generates a Weyl-Heisenberg frame for $`L^2(\mathrm{\Omega })`$ with frame bounds $`A_0`$ and $`B_0`$ if and only if
$$2\pi A_0=\underset{0ik}{\mathrm{min}}\underset{\xi \overline{E},z𝕋}{inf}|\underset{j=0}{\overset{j_i}{}}g(\xi +2\pi n_{ij})z^{n_{ij}}|^2$$
and
$$2\pi B_0=\underset{0ik}{\mathrm{max}}\underset{\xi \overline{E},z𝕋}{sup}|\underset{j=0}{\overset{j_i}{}}g(\xi +2\pi n_{ij})z^{n_{ij}}|^2.$$
In other word, $`g_E=g\chi _E`$ is a mother Weyl-Heisenberg frame wavelet of $`L^2(\mathrm{\Omega })`$ for $`(2\pi ,1)`$ if and only if no characteristic chains of $`g`$ associated with $`\overline{E}`$ are root sequences (of the corresponding step-width sequences).
Notice that the special case $`g=1`$ of Theorem 3.3 will relate the characterization of a Weyl-Heisenberg frame set within the basic support sets (which is more general than the sets considered in ) to the classification of corresponding polynomials with unit roots. The result in the following corollary to Theorem 3.3 is immediate, which can also be obtained from the definition of mother Weyl-Heisenberg frames directly without difficulty.
###### Corollary 3.4.
If $`gL^2()`$ is a continuous function with a zero characteristic chain associated with a closed basic support set $`E`$, then $`g\chi _E`$ is not a mother Weyl-Heisenberg frame wavelet of $`L^2(\mathrm{\Omega })`$ for $`(2\pi ,1)`$, where $`\mathrm{\Omega }=_n(E+2n\pi )`$.
Proof of Theorem 3.1. For all $`x,y[0,2\pi )`$, we have
$`Zg(x,y)={\displaystyle \frac{1}{\sqrt{2\pi }}}{\displaystyle \underset{n}{}}g(x+2n\pi )e^{iny}`$
$`=`$ $`{\displaystyle \frac{1}{\sqrt{2\pi }}}{\displaystyle \underset{n}{}}{\displaystyle \underset{j=0}{\overset{k}{}}}a_j\chi _{F_j}(x+2n\pi )e^{iny}`$
$`=`$ $`{\displaystyle \frac{1}{\sqrt{2\pi }}}\chi _{[0,2\pi )}(x){\displaystyle \underset{j=0}{\overset{k}{}}}a_je^{in_jy}.`$
Thus
$`\mathrm{ess}\underset{(x,y)[0,2\pi )^2}{inf}|Zg(x,y)|^2`$
$`=`$ $`{\displaystyle \frac{1}{2\pi }}\mathrm{ess}\underset{y[0,2\pi )}{inf}|{\displaystyle \underset{j=0}{\overset{k}{}}}a_je^{in_jy}|^2`$
$`=`$ $`{\displaystyle \frac{1}{2\pi }}\mathrm{ess}\underset{z𝕋}{inf}|{\displaystyle \underset{j=0}{\overset{k}{}}}a_jz^{n_j}|^2`$
and
$`\mathrm{ess}\underset{(x,y)[0,2\pi )^2}{sup}|Zg(x,y)|^2`$
$`=`$ $`{\displaystyle \frac{1}{2\pi }}\mathrm{ess}\underset{y[0,2\pi )}{sup}|{\displaystyle \underset{j=0}{\overset{k}{}}}a_je^{in_jy}|^2`$
$`=`$ $`{\displaystyle \frac{1}{2\pi }}\mathrm{ess}\underset{z𝕋}{sup}|{\displaystyle \underset{j=0}{\overset{k}{}}}a_jz^{n_j}|^2,`$
where $`𝕋`$ is the unit circle. By Lemma 2.4, it follows that $`A_0`$ and $`B_0`$ are the frame bounds of $`g`$ if and only if $`2\pi A_0=inf_{|z|=1}|_{j=0}^ka_jz^{n_j}|^2`$ and $`2\pi B_0=sup_{|z|=1}|_{j=0}^ka_jz^{n_j}|^2`$. This proves the equivalence of (1) and (2). The special case of $`E=[0,2\pi )`$ in (3) implies (1). By (1), $`(g,2\pi ,1)`$ generates a Weyl-Heisenberg frame for $`L^2()`$ with frame bounds $`A`$, $`B`$. Let $`P`$ be the orthogonal projection of
$$L^2()L^2(\mathrm{\Omega })$$
defined by
$$Pf=f|_\mathrm{\Omega }=f\chi _\mathrm{\Omega }.$$
Then
$`P(E_mT_{2n\pi }g)=P\left(e^{imt}g(t2n\pi )\right)`$
$`=`$ $`\left(e^{imt}g(t2n\pi )\right)\chi _\mathrm{\Omega }`$
$`=`$ $`e^{imt}g_E(t2n\pi )=E_mT_{2n\pi }g_E.`$
So $`g_E=_{j=0}^ka_j\chi _{E_j}`$ is a mother Weyl-Heisenberg frame for $`L^2(\mathrm{\Omega })`$ with frame bounds $`AA_0`$, $`BB_0`$.
Proof of Theorem 3.2. Since $`g`$ is continuous on $``$, $`g_F=g\chi _F`$ is continuous on $`F`$ and it follows that $`Zg_F(x,y)`$ is continuous on $`E\times [0,2\pi )`$. So there exist $`0<AB`$ such that
$$A=\mathrm{ess}\underset{(x,y)E\times [0,2\pi )}{inf}|Zg(x,y)|^2\mathrm{ess}\underset{(x,y)E\times [0,2\pi )}{sup}|Zg(x,y)|^2=B$$
if and only if
$$0<A=\underset{xE,z𝕋}{inf}|\frac{1}{\sqrt{2\pi }}\underset{j=0}{\overset{k}{}}g(x+2\pi n_j)z^{n_j}|^2$$
and
$$B=\underset{xE,z𝕋}{sup}|\frac{1}{\sqrt{2\pi }}\underset{j=0}{\overset{k}{}}g(x+2\pi n_j)z^{n_j}|^2$$
by Lemma 2.4. The result of the Theorem then follows since $`A`$ and $`B`$ are actually attained on the set $`\overline{E}\times 𝕋`$ when $`g`$ is continuous. ∎
Proof of Theorem 3.3. By the definition of basic support sets, we have $`E=_{i=1}^kF_i`$, where
$$F_i=\underset{j=1}{\overset{j_i}{}}(E_i+2n_{ij}\pi )$$
and the $`E_i`$’s are disjoint intervals in $`[0,2\pi )`$. Let $`M_i=_n(E_i+2n\pi )`$, then $`M_iM_j=\varphi foranyij`$,
$$\mathrm{\Omega }=\underset{n}{}(E+2n\pi )=\underset{i=1}{\overset{k}{}}M_i$$
and
$$L^2(\mathrm{\Omega })=\underset{i=1}{\overset{k}{}}L^2(M_i).$$
It is easy to see that $`g`$ is a mother Weyl-Heisenberg frame wavelet of $`L^2(\mathrm{\Omega })`$ if and only if for each $`i`$, $`g\chi _{F_i}`$ is a mother Weyl-Heisenberg frame wavelet of $`L^2(M_i)`$. By Theorem 3.2, $`g\chi _{F_i}`$ is a mother Weyl-Heisenberg frame wavelet of $`L^2(M_i)`$ if and only if no characteristic chains of $`g`$ associated with $`\overline{F_i}`$ are root sequences (of their corresponding step-width sequences). We leave it to our reader to check that
$$A_0=\frac{1}{2\pi }\underset{0ik}{\mathrm{min}}\underset{\xi \overline{E},z𝕋}{inf}|\underset{j=0}{\overset{j_i}{}}g(\xi +2\pi n_{ij})z^{n_{ij}}|^2$$
and
$$B_0=\frac{1}{2\pi }\underset{0ik}{\mathrm{max}}\underset{\xi \overline{E},z𝕋}{sup}|\underset{j=0}{\overset{j_i}{}}g(\xi +2\pi n_{ij})z^{n_{ij}}|^2$$
are the frame bounds of $`g\chi _E`$ in case that $`A_0>0`$ (hence $`g\chi _E`$ is a mother Weyl-Heisenberg frame wavelet of $`L^2(\mathrm{\Omega })`$). ∎
## 4 Examples
###### Example 4.1.
The set $`[3\pi ,7\pi )`$ is a Weyl-Heisenberg frame set: We have $`E_1=[0,\pi )`$, $`E_2=[\pi ,2\pi )`$, $`n_{11}=2`$, $`n_{12}=3`$, $`n_{21}=1`$, $`n_{22}=2`$ since
$$[3\pi ,7\pi )=\left(E_1+4\pi \right)\left(E_1+6\pi \right)\left(E_2+2\pi \right)\left(E_2+4\pi \right).$$
Furthermore, the corresponding polynomials $`2+3z`$ and $`1+2z`$ have no unit roots.
###### Example 4.2.
The set
$$(\frac{5}{2}\pi ,\frac{7}{2}\pi ](4\pi ,\frac{11}{2}\pi ]=\left((0,\frac{1}{2}\pi ]+4\pi \right)\left((\frac{1}{2}\pi ,\frac{3}{2}\pi ]+2\pi \right)\left((\frac{1}{2}\pi ,\frac{3}{2}\pi ]+4\pi \right)$$
is a WH-frame set since $`E_1=(0,\frac{1}{2}\pi ]`$, $`E_2=(\frac{1}{2}\pi ,\frac{3}{2}\pi ]`$, $`n_{11}=2`$, $`n_{21}=1`$, $`n_{22}=2`$ and the polynomials $`2`$ and $`1+2z`$ have no unit roots.
###### Example 4.3.
On the unit circle of complex plane,
$$|4+3z+2z^3|=|2+3z^2+4z^3|.$$
The roots of $`p(z)=4+3z+2z^3`$ are: $`r_1=0.4398+1.4423i`$, $`r_2=0.43981.4423i`$, $`r_3=0.8796`$. So $`p(z)`$ doesn’t have unit zeros on the complex plane. It follows that the set functions
$$g_1(\xi )=4\chi _{[0,2\pi )}+3\chi _{[2\pi ,4\pi )}+2\chi _{[6\pi ,8\pi )}$$
and
$$g_2(\xi )=2\chi _{[0,2\pi )}+3\chi _{[4\pi ,6\pi )}+4\chi _{[6\pi ,8\pi )}$$
are both mother Weyl-Heisenberg frame wavelets for $`(2\pi ,1)`$ with the same frame bounds.
###### Example 4.4.
Similarly, using the fact that $`|12z+3z^35z^5|=|5+3z^22z^4+z^5|`$ on the unit circle and that $`p(z)=12z+3z^35z^5`$ doesn’t have unit zeros, we see that the following set functions
$$g_3(\xi )=\chi _{[0,2\pi )}2\chi _{[2\pi ,4\pi )}+3\chi _{[6\pi ,8\pi )}5\chi _{[10\pi ,12\pi )}$$
and
$$g_4(\xi )=5\chi _{[0,2\pi )}+3\chi _{[4\pi ,6\pi )}2\chi _{[8\pi ,10\pi )}+\chi _{[10\pi ,12\pi )}$$
are both mother Weyl-Heisenberg frame wavelets for $`(2\pi ,1)`$ with the same frame bounds.
###### Example 4.5.
The function $`f(t)=\mathrm{sin}t`$ is not a mother Weyl-Heisenberg frame wavelet of $`L^2()`$ for $`(2\pi ,1)`$ on any basic support set $`E`$ (in fact any measurable set) since for any set $`E`$ such that $`=_n(E+2n\pi )`$, $`\overline{E}`$ must contain a subsequence of $`\{k\pi \}_k`$. Consequently, the function $`\mathrm{sin}t`$ will always have a zero characteristic chain associated with the set $`\overline{E}`$. However, if $`E`$ is a WH-frame set such that $`\overline{E}`$ is disjoint from $`\{k\pi \}_k`$, then $`\mathrm{sin}t\chi _E`$ is a mother WH-frame wavelet of $`L^2(\mathrm{\Omega })`$ for $`(2\pi ,1)`$ where $`\mathrm{\Omega }=_n(E+2n\pi )`$ since any characteristic chain of $`\mathrm{sin}t`$ associated with $`\overline{E}`$ is a constant sequence. This example can be generalized to other $`2\pi `$-periodic functions. In particular, given any continuous $`2\pi `$-periodic function $`g(t)`$ that is bounded away from $`0`$ and a WH-frame set $`E`$, $`g\chi _E`$ is a mother WH-frame wavelet of $`L^2()`$ for $`(2\pi ,1)`$. For instance, the function
$$g(t)=\{\begin{array}{cc}|\mathrm{sin}(t)|\hfill & \mathrm{if}t[\frac{\pi }{6},\frac{5\pi }{6})+k\pi ,k,\hfill \\ \frac{1}{2}\hfill & \mathrm{otherwise}.\hfill \end{array}$$
is a mother Weyl-Heisenberg frame wavelet of $`L^2()`$ for $`(2\pi ,1)`$ when it is restricted to the set $`E=[3\pi ,7\pi )`$. See Figure 1 below.
###### Example 4.6.
Show that the following function $`g(t)`$ (see Figure 2) is a mother Weyl-Heisenberg frame wavelet of $`L^2()`$ for $`(2\pi ,1)`$ when it is restricted to the set $`E=[0,2\pi )[4\pi ,6\pi )[8\pi ,10\pi )`$. Notice that the set $`E`$ is not a WH-frame set.
The function $`g(t)`$ is defined piecewisely by
$$g(t)=\{\begin{array}{cc}0\hfill & t<0,\hfill \\ \mathrm{sin}(2t)/2\hfill & t[0,2\pi ),\hfill \\ \mathrm{sin}(\frac{16}{7}(t2\pi ))\hfill & t[2\pi ,\frac{15}{4}\pi ),\hfill \\ 2(\mathrm{sin}t+\mathrm{cos}t),\hfill & t[\frac{15}{4}\pi ,\frac{27}{4}\pi ),\hfill \\ 4\mathrm{sin}(\frac{6}{5}(t\frac{27}{4}\pi )),\hfill & t[\frac{27}{4}\pi ,8\pi ),\hfill \\ 4\hfill & t8\pi .\hfill \end{array}$$
In this case, $`E_1=[0,2\pi )`$ and $`n_{11}=0`$, $`n_{12}=2`$, $`n_{13}=4`$ since $`E=E_1(E_1+22\pi )(E_1+42\pi )`$. So for any given $`tE_1`$, the corresponding characteristic chain of $`g(t)`$ with respect to $`E`$ is $`\{\mathrm{sin}(2t)/2,2(\mathrm{sin}t+\mathrm{cos}t),4\}`$. It is not a root sequence of $`\{0,2,4\}`$ since $`\mathrm{sin}(2t)/2+2(\mathrm{sin}t+\mathrm{cos}t)z^2+4z^4=(\mathrm{sin}t+2z^2)(\mathrm{cos}t+2z^2)`$ apparently have no unit zeros for any given $`t`$. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.